Rfi Issue and Spectrum Sharing Paradigm for Future Satellite

RFI ISSUE AND SPECTRUM SHARING PARADIGM FOR FUTURE SATELLITE

COMMUNICATION AND RADIO ASTRONOMY SYSTEMS

Yucheng Dai

APPROVED BY SUPERVISORY COMMITTEE:

Dr. Hlaing Minn, Chair

Dr. John P. Fonseka

Dr. Andrea Fumagalli

Yucheng Dai

COMMUNICATION AND RADIO ASTRONOMY SYSTEMS

YUCHENG DAI, BS, MS

DISSERTATION

Presented to the Faculty of

The University of Texas at Dallas

in Partial Fulﬁllment

of the Requirements

for the Degree of

DOCTOR OF PHILOSOPHY IN

ELECTRICAL ENGINEERING

THE UNIVERSITY OF TEXAS AT DALLAS

August 2020 ACKNOWLEDGMENTS

I would like to thank my PhD adviser, Dr. Hlaing Minn, for his valuable advice, dedicated guidance, and unparalleled support that made this work possible. I would like to express my deepest appreciation to my committee members, Drs. Andrea Fumagalli, John P. Fonseka, and Murat Torlak.

I would like to thank my parents, for their guidance to the way to PhD, and for their support both spiritually and ﬁnancially. I also want to thank my friend and lab-mate, Dong Han, who shows an example of dedicated and helpful PhD student to me.

June 2020

v RFI ISSUE AND SPECTRUM SHARING PARADIGM FOR FUTURE SATELLITE

COMMUNICATION AND RADIO ASTRONOMY SYSTEMS

Yucheng Dai, PhD The University of Texas at Dallas, 2020

Supervising Professor: Dr. Hlaing Minn, Chair

Wireless services, which utilize radio spectrum resources, can be classified into two types: passive wireless services and active wireless services. Some passive wireless services (e.g., Radio Astronomy System (RAS)) are extremely vulnerable to the interference from active wireless services. In addition, the widely-distributed radio astronomical signals in the spectrum drive the radio astronomers seeking for opportunities to observe in the bands which have been assigned to active wireless systems for primary use. In view of the conflict between the RAS and other active wireless systems, this dissertation focuses on the spectrum sharing paradigm between the two sides which can bring benefits to both sides and thus achieve a harmonious utilization of the spectrum resource.

First, we investigate and develop a spectrum sharing paradigm between the RAS and the Geostationary Orbit (GSO) Satellite Communication Systems (SCSs). Via utilizing the idle RAS band(s), the proposed paradigm can bring more throughput/capacity to SCS side while reducing Radio Frequency Interference (RFI) and oﬀering more observation opportunities to RAS side.

Next, we propose a new paradigm where SCS and RAS are integrated into the Non-Geostatio- nary Orbit (NGSO) satellite system, thus eﬀectively creating large-scale telescopes in orbit. This integrated system not only avoids SCS’s RFI to RAS but also oﬀers more spectrum

vi access opportunities to both SCS and RAS. The proposed paradigm several additional advantages in terms of accessible spectrum bands, RAS observation performance, and SCS maximum mean supportable data rate as well as enabling coexistence and growths of both types of services.

vii TABLE OF CONTENTS

ACKNOWLEDGMENTS ...... v ABSTRACT ...... vi LIST OF FIGURES ...... x LIST OF TABLES ...... xiii CHAPTER 1 INTRODUCTION ...... 1 1.1 Background ...... 1 1.2 Outline and Contributions ...... 2 CHAPTER 2 A SPECTRUM SHARING PARADIGM FOR GSO SATELLITE SYS- TEM AND RADIO ASTRONOMY SYSTEM ...... 3 2.1 Introduction ...... 3 2.2 RFI Analysis for Ground Radio Telescopes under GSO satellites ...... 9 2.2.1 Interference from GSO Satellites to Ground Radio Telescopes . . . .9 2.2.2 Ground Telescope Model ...... 11 2.2.3 GSO Satellite System Model ...... 13 2.2.4 Unwanted Emission Power of the GSO Satellite’s Subband Transmission 15 2.2.5 RFI from Other Active Wireless Systems ...... 18 2.2.6 RFI and Sample Loss Rate of the Ground Radio Telescopes under GSO SCS ...... 20 2.3 Existing RFI Reduction methods as Benchmarks ...... 24 2.4 Proposed Spectrum Sharing Paradigm with Three RFI Reduction Methods . 31 2.4.1 RFI Determination for Ground Radio Telescopes in Diﬀerent Time Durations ...... 31 2.4.2 Reorganization of the Spectrum Resource to Minimize the Unwanted Emission Power of the GSO Satellites in the RAO Bands ...... 34 2.4.3 Rearranging the Subband Allocation of the GSO Satellite’s Beams to Minimize the RFI of the GSO Satellites ...... 40 2.4.4 Cell-based Beam Switch Approach to Suppress the RFI ...... 44 2.5 Numerical Results of RFI and Sample Loss Rate ...... 49 2.5.1 Eﬀects of the Parameters of the Three RFI Reduction Methods . . . 49

viii 2.5.2 RFI and Sample Loss Rate in Diﬀerent Scenarios ...... 51 2.6 Conclusions ...... 62 CHAPTER 3 IMPACTS OF LARGE-SCALE NGSO SATELLITES: RFI AND A NEW PARADIGM FOR SATELLITE COMMUNICATIONS AND RADIO ASTRONOMY SYSTEMS ...... 65 3.1 Introduction ...... 65 3.2 RFI Analysis for Ground Radio Telescopes under a Large-Scale LEO SCS . . 68 3.2.1 Interference Calculation ...... 68 3.2.2 Large-Scale LEO SCS Model: OneWeb ...... 70 3.2.3 Ground Telescopes Model ...... 72 3.2.4 Guardband and Emission Mask Based RFI Analysis ...... 73 3.2.5 RFI Analysis Based on OneWeb LEO Constellation ...... 76 3.3 Guardband, Transmission Muting and Sample Excision Based Solutions under Large-Scale LEO SCS ...... 81 3.4 A New Paradigm for NGSO SCS and RAS ...... 83 3.4.1 An Integrated NGSO SCS and RAS ...... 83 3.4.2 Observability of LEO versus Ground Telescopes ...... 86 3.4.3 Sensitivity of LEO versus Ground Telescopes ...... 92 3.5 Data Rate Analysis Based on a Shared RAS Band in the Proposed Paradigm 94 3.5.1 Gateway-Satellite Model Based Data Rate Analysis ...... 94 3.5.2 Communication System Maximum Mean Supportable Data Rate and RAO Data Rate Results ...... 98 3.6 RAO Data Transport Design ...... 101 3.6.1 Development of Data Transport ...... 101 3.6.2 Data Transport Performance Results ...... 103 3.7 Conclusions ...... 104

CHAPTER 4 CONCLUSION ...... 107 REFERENCES ...... 109 BIOGRAPHICAL SKETCH ...... 117 CURRICULUM VITAE

ix LIST OF FIGURES

2.1 The 58 existing ground observatories’ locations on the earth ...... 11 2.2 Downlink band allocation of the 3 groups of GSO satellites ...... 14 2.3 Cell locations of the GSO satellites with the number of spot beams which serve the cells ...... 14

2.4 Relative PSD attenuation versus different |f − fc|/Bsub values ...... 17 2.5 Unwanted emission power of the satellites’ transmission in subbands centered at different frequencies in different RAS bands ...... 18 2.6 RFI at the Effelsberg 100-m radio telescope in the RAS band I with different RAO modes and the corresponding beam directions in latitude and longitude . 21 2.7 RFI at the Effelsberg 100-m radio telescope in the RAS band I at different directions in different scenarios ...... 22 2.8 Average RFI at different telescopes in 200 random realizations ...... 24 2.9 Average sample loss rate at different telescopes in 200 random realizations . . . 25 2.10 Sample loss rate and average throughput ratio in case 1 with different guard band bandwidths ...... 29 2.11 Sample loss rate and average throughput ratio in case 2 with different muting thresholds ...... 30 2.12 Sample loss rate and average throughput ratio in case 3 with different muting thresholds ...... 30 2.13 An example of periods and subperiods in an RAO operation duration ...... 32 2.14 Potential spectrum resource allocation of the SCS in scenario 1 ...... 35 2.15 Average unwanted emission power versus minimum total SCS required bandwidth BSCS,0 under different Bsub,0 and Bsub,unit values in scenario 1 ...... 38

2.16 Average unwanted emission power in different scenarios with different BSCS,0 values ...... 41 2.17 Procedure diagram for rearranging the subband allocation at period n ...... 45 2.18 RFI generated by the spot beams from different satellites which serve a cell near the Effelsberg 100-m Radio Telescope ...... 45 2.19 Average RFI and sample loss rate of a realization in scenario 1 among the telescopes using the spectrum resource reorganization method with different BSCS,0 requirements ...... 50

x 2.20 Average RFI among the telescopes of a realization in scenario 1 using the subband allocation method or the beam switch method with different Nsp values . . . . . 51 2.21 Average RFI among the telescopes of a realization in scenario 1 using the beam switch method with different Nbeam,ex and Nswitch values ...... 52 2.22 Average RFI at different telescopes with different RFI reduction methods in scenario 1 ...... 54 2.23 Average sample loss rate versus average throughput ratio with different RFI reduction methods in scenario 1 ...... 55 2.24 Average RFI at different telescopes with different RFI reduction methods in scenarios 2 and 3 ...... 56 2.25 Average sample loss rate versus average throughput ratio with different RFI reduction methods in scenarios 2 and 3 ...... 57 2.26 Average RFI at different telescopes with different RFI reduction methods in scenario 4 ...... 58 2.27 Average sample loss rate versus average throughput ratio with different RFI reduction methods in scenario 4 ...... 59 2.28 Average RFI at different telescopes with different RFI reduction methods in scenario 5 ...... 60 2.29 Average sample loss rate versus average throughput ratio with different RFI reduction methods in scenario 5 ...... 61 2.30 Average RFI at different telescopes with/without the beam switch method in scenario 6 ...... 62 2.31 Average sample loss rate versus average throughput ratio in scenario 6 . . . . . 63

3.1 An illustrative scenario for angles θT and θR ...... 71 3.2 OneWeb LEO satellite constellation (+ denotes a satellite) ...... 72 3.3 Existing ground radio astronomy telescopes’ locations ...... 73

3.4 The required guardband bandwidth versus (θT, θR)...... 75 3.5 Instantaneous RFI EPFD at a ground telescope during 24 hours in the presence of LEO satellites ...... 77 3.6 Average RFI EPFD levels at diﬀerent ground telescopes during 24 hours in the presence of LEO satellites ...... 77 3.7 Average RFI level at diﬀerent azimuth and elevation angles of telescope 3 during 24 hours in the presence of LEO satellites ...... 79

xi 3.8 Average RFI level at different azimuth and elevation angles of telescope 36 during 24 hours in the presence of LEO satellites ...... 79 3.9 Ground telescopes VLBI observation performance with and without LEO satellites 81 3.10 Average RFI EPFD levels at different ground radio telescopes for the 4 considered cases during 24 hours ...... 83 3.11 Percentage of the beams that are turned off in case 3 and the instantaneous RAO sample loss rate in case 4 across time ...... 84 3.12 Average number of telescopes that can simultaneously observe a target versus target directions (from left to right are case 1 to case 5) ...... 89 3.13 Maximum baseline distance for different target directions (from left to right are case 1 to case 5) ...... 89 3.14 Sensitivities of the ground and LEO telescopes ...... 95 3.15 System topological graphs. (a) Local graph for data rate analysis with M = 4. (b) Topological graph for RAO data transport with L = 2 and Ns = Nd = 3 (In practice, Ns could be greater than Nd)...... 95 3.16 Aggregate RAO data rate and SCS maximum mean supportable data rate of the proposed integrated system for 3 band utilization cases ...... 100 3.17 RAO data rate versus the number of the selected gateways ...... 105 3.18 RAO data packet relaying cost ...... 105

xii LIST OF TABLES

2.1 Subscripts Used in Chapter 2 ...... 7 2.2 Notations Used in Chapter 2 ...... 7 2.3 GSO Satellite System Settings ...... 15 3.1 Notations Used in Chapter 3 ...... 69 3.2 LEO satellite settings ...... 72 3.3 Observability of the LEO and ground telescopes ...... 87 3.4 Parameters for the LEO SCS links ...... 100

xiii CHAPTER 1

INTRODUCTION

1.1 Background

Spectrum sharing has been proved to be an efficient strategy to improve the spectrum utilization [67]. Some efforts have been made for spectrum sharing between active wireless systems, including spectrum sharing in TV white space [25, 76], and ISM/unlicensed band [20, 3]. Yet, the spectrum sharing between active and passive wireless systems has not been fully investigated. Passive wireless systems, including radio astronomy system and earth exploration system, are designed to observe earth surface, solar systems and other cosmic objects. These systems play an important role in weather/drought prediction, sea-level/pollution monitoring, discoveries of pulsars, cosmic microwave background radiation recording, and solar flares and sunspots monitoring. Due to the economical and scientific significance of the passive wireless systems, many efforts have been made to protect them [45, 49, 46, 48, 43, 44]. Recently, with the increasing demand of spectrum resources in Ka band (26.5 – 40 GHz) and above from active wireless systems (e.g., ultra-dense LEO communication system and fifth-generation cellular communication system), the policy makers such as International Telecommunication Union (ITU) and Federal Communications Commission (FCC) are ac- tively seeking for the opportunities in relocating sub-6 GHz bands and millimeter wave bands to emerging active wireless services. On the other hand, due to the increasing amount of the active wireless systems and imperfection in their transmitters, the interference from those systems to the passive wireless systems is increasing. In addition, radio astronomy (as a rep- resentative of passive service) is assigned with several bands in Ku (12 - 18 GHz)/Ka band(s) and above, which could be attractive to cellular/satellite communication service providers. Therefore, it is logical and essential to develop a spectrum sharing paradigm between the two types of systems which can satisfy their demands respectively. We address this issue in this dissertation.

1 With the development of electronic instrument, compact-sized, light-weighted and power- eﬃcient space-based radio telescopes become feasible. Compared to ground radio telescopes, space radio telescopes can observe more bands (including the bands with total atmospheric opacity) and bear less interference from ground active wireless systems and atmospheric radiation [21]. A few projects [19, 12, 2] have been proposed to build dedicated space telescopes with limited size and mass. Diﬀerent from them, we propose an integrated satellite system for both SCS and RAS and address related technical issues.

1.2 Outline and Contributions

In the first part of the dissertation (Chapter 2), we propose a spectrum sharing paradigm for the GSO SCS and the RAS with three potential RFI reduction methods. Depending on the radio astronomical observation tasks of different observatories, an appropriate combination of RFI reduction methods can be chosen. Our analyses show that the proposed paradigm can guarantee most of the telescopes with sample loss rate lower than 2% in both protected and unprotected RAS bands, while the SCS can obtain up to 13.88% extra average downlink throughput. The second part (Chapter 3) of the dissertation proposes a new paradigm where SCS and RAS are integrated into a Non-Geostationary Orbit (NGSO) satellite system, thus effectively creating large-scale telescopes in orbit. This integrated system not only avoids SCS’s RFI to RAS but also offers more spectrum access opportunities to both SCS and RAS. Addition- ally, this paper addresses two related problems of the new paradigm, namely the spectrum resource allocation problem and the RAS data transport problem. Our performance evaluation illustrates the advantages of the proposed paradigm in terms of accessible spectrum bands, RAS observation performance, and SCS maximum mean supportable data rate as well as enabling coexistence and growths of both types of services.

2 CHAPTER 2

A SPECTRUM SHARING PARADIGM FOR GSO SATELLITE SYSTEM

AND RADIO ASTRONOMY SYSTEM 1

2.1 Introduction

Radio spectrum is a limited natural resource and its usage can be classified into two types – active wireless services which transmit and receive electromagnetic waves and passive wireless services which conduct signal reception only. Obviously, passive wireless services are prone to RFI from active wireless services and spectrum access conflict arises between the two types of services. A RAS is an example of important passive wireless services. RAS not only provides a description of the universe and its history, but also enables testing of the laws of fundamental physics, e.g., the General Theory of Relativity [54]. It is expanding from a phenomenological science to astro-physics and astro-chemistry for which the Radio Astronomical Observation (RAO) is intrinsically sensitivity-limited and interference-free environments are needed. Gen- erally, RAS wants to observe all available frequency bands ranging from 2 MHz to 1000 GHz to obtain as mush as possible the information of the universe [21]. Yet, ITU and FCC [24] only guarantee a few bands for RAS as primary use, which indicates that RAOs in others bands are unprotected and may face severe RFI from active wireless services [79, 73]. Satellite Communication System is one of the prominent active wireless systems with several benefits to the society as it offers voice, video, data, weather and navigation services to the users, especially for those not covered by ground communication networks. In addition, it plays a critical role in disasters and emergencies management [89], which would save many lives. Nevertheless, its global coverage nature, which is a unique advantage for global

1© 2019 IEEE. Reprinted, with permission, from Yucheng Dai, and Hlaing Minn, ”A Spectrum Sharing Paradigm for GSO Satellite System and Radio Astronomy System,” in IEEE Access, vol. 7, pp. 93952-93973, July 2019.

3 communication, inevitably causes a more prominent RFI issue to the RAS side than other active wireless systems would do. This chapter focuses on the spectrum access issue between

RAS and GSO SCS.

To capture weak cosmic signals which are much lower than the wireless communication signals generated by SCS, the ground telescopes are equipped with large size parabolic antennas and high-sensitivity receivers. The observatories are usually built in remote areas protected by radio quiet zones to prevent RFI from the ground wireless communication systems leaking into the main beam of the radio telescopes. In spite of this, with the growth of the human population and the broader use of the wireless communications, the RFI issue for the RAS becomes more thorny. Thus, RFI mitigation plays a more critical role. The existing RFI mitigation methods include digital filtering and sub-space filtering [18, 36, 35, 33], spatial filtering with array instruments [5, 38, 75, 28, 9, 37, 34], auxiliary antenna based RFI removal [50, 8, 16, 1], cognitive radio based approaches [6, 7, 4], and time-division spectrum sharing [57, 71, 70, 30, 29]. However, due to the relative locations of the ground telescopes and the satellites, the signals from the satellites may directly hit the dish surface of the telescope and consequently cause strong RFI to the RAS. And the existing RFI mitigation methods are not effective in such cases.

Several cases of GSO satellites’ RFI to RAS have been reported in the literature. For example, the Eﬀelsberg 100-meter Telescope suﬀered huge sample loss (up to 90%) in 10.6 –

10.7 GHz (a protected band allocated to RAS) since 1995 as a GSO satellite called ASTRA-

1D, started TV broadcasting with 10.714 GHz as its downlink at that time. In addition,

[58] indicates that the major source of the RFI in K band (18.0 – 26.5 GHz) at Very Large

Array in New Mexico is the GSO satellites in the Clarke Belt. Two frequency bands in K band with frequency range 22.21 – 22.5 GHz and 23.6 – 24.0 GHz have been allocated to

RAS for investigating water vapor and Ammonia in the space. However, the observation and investigation of another critical substance called Cyclopropenylidene, which has a ubiquitous

4 interstellar line at 18.343 GHz in K band [81], may face strong RFI from the GSO satellites as 18.3 – 18.8 GHz is assigned to the GSO satellites instead of RAS for space-to-earth link. According to the current spectrum allocation, among the three bands (22.21 – 22.5 GHz, 23.6 – 24.0 GHz, 18.3 – 18.8 GHz) mentioned above, active wireless communication systems need to avoid causing detrimental RFI to RAS only in the first two bands. Thus, the first two bands are protected and the last one is unprotected from the RAS’s perspective. Note that let alone unprotected bands, even protected bands for RAS often experience detrimental RFI as mentioned above. Considering the increasing conflict of the spectrum usage and RFI issue between the GSO SCS and the RAS, we propose a new spectrum sharing paradigm for the SCS and the RAS with the following characteristics/advantages: i) The RFI in the bands of interests for the RAS, whether being allocated to the RAS or not, can be reduced. ii) The GSO SCS can obtain more downlink bandwidth/throughput for transmitting data and hence will be willing to cooperate with the RAS and implement the RFI reduction methods. iii) The ground telescopes of different observatories, with some synchronization, offer some idle RAS bands to the SCS to afford the SCS’s downlink traffic. iv) The SCS will exploit the spare RAS bands together with an appropriate RFI reduction method to improve the RAO quality of the telescopes as well as to increase the SCS downlink bandwidth/throughput. Our main contributions in this chapter include the following:

1. We analyze the RFI impact of SCS on the ground telescopes in the unprotected RAS bands (which has not been studied in the literature) and compare it with that in the protected RAS bands.

2. We demonstrate how some existing RFI reduction methods can be used to reduce RFI in the unprotected RAS bands at the telescopes. We also analyze the corresponding RFI reduction capabilities and RAO sample loss rate for the RAS and the service/throughput loss for the SCS caused by the application of these methods.

5 3. We propose a new spectrum sharing paradigm where the ground RAS telescopes and the GSO SCS cooperate to achieve more observability for the RAS as well as more downlink bandwidth/throughput for the SCS. Based on this paradigm, we develop three methods to reduce the RFI from the GSO satellites to the ground telescopes, namely, i) reorganization of the spectrum resource to minimize the unwanted emission power of the GSO satellites in the RAO bands, ii) rearranging the subband allocation of the GSO satellite’s beams based on their unwanted emission power and bandwidth, and iii) cell-based beam switch approach to suppress the RFI of the GSO satellites.

4. We investigate the roles of the parameters in the three proposed RFI reduction methods on RFI reduction and RAO sample loss rate at the telescopes. Then, we evaluate the performance of these methods when jointly applied in different scenarios by means of Monte Carlo simulations and compare them with two recent existing RFI reduction methods. The results show that our proposed paradigm can lower the RAO sample loss rate of the telescopes in different RAS bands and offer more downlink bandwidth/throughput to the SCS in several scenarios.

The chapter is organized as follows. Section 2.2 introduces the GSO SCS and ground telescopes model, and analyzes the RFI level and sample loss rate of the ground telescopes caused by the GSO satellites’ downlink. Section 2.3 shows applications of some existing RFI reduction methods for RFI reduction in the unprotected RAS bands, and analyzes the sample loss rate of the ground telescopes and the service loss of the GSO satellites. Section 2.4 proposes three RFI reduction methods. Section 2.5 discusses about different choices of the parameters in the three RFI reduction methods and the corresponding effects on the RFI at the telescopes. Numerical results of the RFI and the sample loss rate of telescopes and the downlink average throughput ratio of the SCS with the proposed paradigm in different scenarios are analyzed and compared with the existing methods in the section. Section 2.6 concludes this chapter.

6 Table 2.1. Subscripts Used in Chapter 2 subscript Description d RAS band index i Ground telescope index j GSO satellite index k Beam index of a certain GSO satellite l Spectrum resource block index m GSO satellite group index n RAO operation duration period index p GSO satellite beam group index of a certain GSO satellite q RAO operation duration subperiod index of a certain period

Notations: The list of key subscripts and notations used in the chapter are shown in Table 2.1 and Table 2.2 respectively for easy access.

Table 2.2: Notations Used in Chapter 2

Notation Description

IGSO,i Index set of the GSO satellites which can be viewed by telescope i

Isub Index set of the SCS subbands

IRAS,j,d Index set of the telescopes which can view the satellite j and observe RAS band d

Isub,group,j,n Index set of the subband used by satellite j during period n

IBG,j,p Beam index set of the spot beams belonging to the beam group p

Icell,j,k Index set of the satellites and corresponding spot beams which serve the same cell as the satellite j’s kth spot beam

PUE,isub,d Unwanted emission power of subband isub in RAS band d

PUE,j,k Unwanted emission power of satellite j’s kth spot beam in the considered RAS bands

Continued on next page

7 Table 2.2 – Continued from previous page

Notation Description

PUE,j,k,n,q,d Unwanted emission power of satellite j’s kth spot beam in RAS band d in period n’s qth subperiod

PUE,j,p,n,q,d Unwanted emission power of satellite j’s pth spot beam group in RAS band d in period n’s qth subperiod

Bsub Subband bandwidth of a spot beam of the GSO satellites used for downlink

Bsub,l Subband bandwidth in spectrum block l

Bsub,0 Minimum required subband bandwidth

Bsub,unit Unit subband bandwidth in subband bandwidth design

Bl Bandwidth of spectrum block l

Bim,n Total bandwidth of subband group im,n

BSCS,0 Minimum total bandwidth required by the SCS

BSCS,used Total bandwidth used by the SCS

Bgroup,m Average total bandwidth assigned to satellite group m in an RAO operation duration

Nbeam Number of spot beams per satellite

Nbeam,ex Number of extra data streams a satellite can carry

Np Number of periods per RAO operation duration

Nsp Number of subperiods per period

Nsub,l Number of subbands in spectrum block l

EPFDi,j,k(t) RFI Equivalent Power Flux-Density generated by satellite j’s kth spot beam to telescope i at time t

EPFDi(T ) Average RFI Equivalent Power Flux-Density received by telescope i during an RAO integration interval T

Continued on next page

8 Table 2.2 – Continued from previous page

Notation Description

EPFDi,d(T ) Average RFI Equivalent Power Flux-Density received by telescope i during an RAO integration interval T in RAS band d caused by GSO SCS

EPFDex,i(T ) Average RFI Equivalent Power Flux-Density received by telescope i during an RAO integration interval T in RAS band d caused by other active wireless systems

EPFDj,k,n,q Sum of RFI Equivalent Power Flux-Densities generated by satellite j’s kth spot beam to the telescopes that can view satellite j during period n’s qth subperiod

normalized by the total RAO integration time

EPFDj,n,q Sum of RFI Equivalent Power Flux-Densities generated by satellite j’s spot beams to the telescopes that can view satellite j during period n’s qth subperiod nor-

malized by the total RAO integration time

EPFDexp,j,n,q Expected sum of RFI Equivalent Power Flux-Densities (under the proposed subband allocation method) generated by satellite j’s spot beams to the telescopes

that can view satellite j during period n’s qth subperiod normalized by the total

RAO integration time

EPFDA,j,k,n,q The sum of RFI Equivalent Power Flux-Densities (under the proposed beam switch method) generated by satellite j’s spot beams to the telescopes that can

view satellite j during period n’s qth subperiod normalized by the total RAO

integration time

2.2 RFI Analysis for Ground Radio Telescopes under GSO satellites

2.2.1 Interference from GSO Satellites to Ground Radio Telescopes

As mentioned in the ITU documents [45, 46, 47], the interference level of the GSO satellites can be evaluated in terms of the Equivalent Power Flux-Density (EPFD). The instantaneous

EPFD between telescope i and satellite j’s kth beam at time t can be calculated with the

9 following formula:

PUE,j,kGT,j,k(t)GR,i,j(t) EPFDi,j,k(t) = 2 (2.1) 4πdi,j(t)

where PUE,j,k is the unwanted emission power of satellite j’s kth beam at RAO band, GT,j,k(t) is the transmitting antenna gain of the GSO satellite j’s kth beam towards the direction of telescope i at time t, GR,i,j(t) is the receiving antenna gain of telescope i towards the direction of satellite j at time t, and di,j(t) is the distance between telescope i and satellite j at time t. Since the GSO satellites are relatively stationary to the earth surface, di,j and

2 GT,j,k are ﬁxed across time . Then, for telescope i, the average RFI EPFD it receives during a certain RAO task with time duration Tinte,i can be represented as

N 1 Z X Xbeam EPFDi(Tinte,i) = EPFDi,j,k(t)dt (2.2) Tinte,i Tinte,i j∈IGSO,i k=1

where IGSO,i is the index set of GSO satellites that can be viewed from telescope i (the

satellites with elevation angle larger than 0° in the view of telescope i) and Nbeam is the

number of beams that each GSO satellite uses for its downlink transmission. Due to the

shape of earth, not all GSO satellites are visible to a certain telescope. Substituting Eq. (2.1)

into Eq. (2.2) and separating the time-varying element, we have

Nbeam Z X X PUE,j,kGT,j,k 1 EPFDi(Tinte,i) = 2 · GR,i,j(t)dt. (2.3) 4πdi,j Tinte,i Tinte,i j∈IGSO,i k=1

Note that GR,i,j(t) depends on the RAO beam direction as well as the telescope-to-satellite

direction. If the RAO beam direction is ﬁxed at a particular RAO target during an interval,

then GR,i,j(t) is constant during that interval.

2In practice, the GSO satellites may move within a ‘box’ due to the eﬀect of the lunar/solar gravitation. Nevertheless, the size of the ‘box’ is limited (about 0.15° around nominal satellite position), and such devi- ation will not aﬀect our analysis. Furthermore, the GSO satellites will typically use their thrusters to keep their positions/orbits stable.

10 Figure 2.1. The 58 existing ground observatories’ locations on the earth

2.2.2 Ground Telescope Model

To evaluate the GSO SCS induced RFI and the corresponding improvement of our proposed new paradigm, we use the 58 existing ground observatories as our reference. The red dots in Fig. 2.1 show the locations of the referenced observatories. To simplify the RFI analysis process, we assume all of the observatories use their biggest single dish telescopes to conduct the RAO. In addition, the radio telescopes are considered to be capable of observing all the bands we discuss in the chapter and a minimum elevation angle of 10° for observation is required to avoid the RFI from ground wireless service leaking into the main beam of the telescopes and corrupting the observation.

To verify our proposed paradigm can deal with the ground telescopes in diﬀerent RAO modes, we consider three diﬀerent types of RAO modes, which are

• Target tracking mode: The radio telescope tracks a certain target when the target is

within its observation range (which mean the elevation angle of the target is larger

than 10°). As the RAO target is far away from the earth, the relative position of the

target can be viewed as static in the solar earth system. Thus, the telescope will adjust

its beam direction to focus on the target and cancel the eﬀect of the earth rotation

during RAO.

11 • Sky-mapping mode: The radio telescope scans across the sky to obtain the emission

information of diﬀerent directions. Here we assume that the telescope will slew its

beam in north-south direction periodically. With the rotation of the earth, the beam

will naturally scan the east-west direction. By this, the telescope can capture the

emission strength distribution of the sky.

• Fixed-direction mode: The radio telescope ﬁxes an observation direction in a period

of time. This RAO mode can be used for observation in dispersed targets or a small

area in sky.

In addition, in this chapter, we focus on three bands that arouse the interest from RAS side, which are

1. RAS band I: 18.28 – 18.36 GHz,

2. RAS band II: 22.21 – 22.5 GHz,

3. RAS band III: 23.6 – 24.0 GHz.

The 3 bands are used for searching cyclopropenylidene (C3H2), water vapour (H2O), and ammonia (NH3) in space respectively. The RAS bands II and III have been assigned to

RAS for its primary use [24] and therefore are well protected with less RFI from other active wireless services. On the other hand, the RAS band I is assigned to satellites for space-to- earth (downlink) transmission and consequently the telescopes may receive strong RFI while conducting RAO in this band. Hence, new RFI reduction methods are needed to enable

RAO in the RAS band I.

In addition, according to [45], the detrimental RFI threshold for the RAO depends on the speciﬁc RAO band’s bandwidth and center frequency, the minimum antenna noise temperature, the receiver noise temperature, the observation type (e.g. continuum observation, spectral-line observation or Very Long Baseline Interferometry (VLBI) observation), as well as the RAO integration time. Table 1 in [45] oﬀers the detrimental RFI thresholds for the

RAS bands II and III for continuum observation as the two bands are assigned to RAS for

12 its primary use, while the detrimental RFI threshold for the RAS band I is unmentioned. Therefore, we use the minimum antenna noise temperature and the receiver noise temperature for the RAS band 15.35 – 15.40 GHz as an example to determine the threshold for the RAS band I.

2.2.3 GSO Satellite System Model

Currently, about 700 functional GSO satellites are in the Clarke Belt and most of them are operated by individual operators with limited or no cooperation between each other. Their downlinks are in C, Ku, K, or/and Ka bands. We consider a future large-scale RAS-friendly GSO satellite communication system where the satellites can be controlled and operated by multiple operators but with tight cooperation. To illustrate the proposed approaches, we consider a GSO satellites model with 221 GSO satellites, using a proportion of K band (17.3 – 20.2 GHz) for downlink (space-to-earth) transmission3. Specifically, the GSO satellites are divided into 3 groups and each group of satellites uses a proportion of that band. Fig. 2.2 demonstrates the detailed band allocation of different groups of satellites4. As shown in the figure, each group of the satellites use one SCS band, which consists of 10 equal-bandwidth subbands. Each spot beam of the satellite uses one subband for downlink transmission. In addition, we assume each GSO satellite has 10 groups of spot beams and each beam group has 4 spot beams. In other words, each satellite is equipped with 40 spot beams to support the users in 40 different areas. The satellite will allocate each group of spot beams with one subband from its assigned band. In addition, we assume the GSO satellites are serving the users on the land area within latitude range 75°S – 75°N. The corresponding areas are divided into cells and each satellite

3In this chapter, we consider a fully cooperative GSO satellite system and leave the problem of the partially cooperative GSO satellite system for future investigation.

4In the rest of the chapter, we consider the band allocation in Fig. 2.2 as the original band allocation plan of the SCS. In addition, we name the SCS using the original band allocation as the ‘original SCS’.

13 SCS Band 1 for Satellite SCS Band 2 for Satellite SCS Band 3 for Satellite 17.3 GHz Group 1 18.3 GHz Group 2 19.3 GHz Group 3 20.2 GHz 1000 MHz 1000 MHz 900 MHz

1 1 1 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 0 0 0

100 MHz 100 MHz 90 MHz

Figure 2.2. Downlink band allocation of the 3 groups of GSO satellites

Figure 2.3. Cell locations of the GSO satellites with the number of spot beams which serve the cells

14 Table 2.3. GSO Satellite System Settings Parameter Value

Number of GSO satellite NGSO 221 Number of beams per satellite Nbeam 40 Dish size 3 meter Antenna model [41] 0.39° (at 18.32 GHz) Beamwidth 0.32° (at 22.355 GHz) 0.30° (at 23.8 GHz) 52.6 dBi (at 18.32 GHz) Boresight gain GT 54.3 dBi (at 22.355 GHz) 54.9 dBi (at 23.8 GHz) serves 40 cells which are able to view the specific satellite in line of sight. In this case, several spot beams from different satellites may share the same service area (cell), which offers the flexibility of using one spot beam to support the users of another spot beam without adjusting the pointing directions of the spot beams. Fig. 2.3 shows the cell locations where ‘x’ markers represent the centers of the cells and the color reflects the number of the spot beams which are serving this cell. Other detailed parameters of the GSO satellites we consider are shown in Table 2.3. In addition, to implement the beam switch method, we assume the users of the SCS are able to change their antenna pointing directions. This can be realized by a multi-focal parabolic antenna [13], a motor-driven antenna [56] or an antenna array [60].

2.2.4 Unwanted Emission Power of the GSO Satellite’s Subband Transmission

Since the major task of the radio telescopes is to obtain the emissions of natural celestial objects, the emissions from other objects like GSO satellites are generally considered as unwanted emissions which will cause interference to the cosmic signals collected by the radio telescopes. The unwanted emission power spectrum of the SCS downlink subbands is a key factor that shapes the RFI received by the ground telescopes. However, this power spectrum in practice is related to the modulation type, the RF ﬁlter, the transmission power and

15 other hardware settings. In this chapter, to be general, we assume the GSO SCS satellites

use the emission mask deﬁned in [22, 61] by FCC and US government. The Power Spectrum

Density (PSD) of the transmitted signals out of the assigned band is limited by the emission

mask with respect to the maximum PSD of the signals in the assigned band measured in a

reference bandwidth. Denote the bandwidth of a subband which is assigned for the downlink

transmission as Bsub, the center frequency of the assigned subband as fc, and the maximum

PSD of the signals measured in the reference bandwidth in this subband as PSDmax. Then,

the PSD of the unwanted emission PSDUE(f) at frequency f out of the subband should satisfy

SEM(f) PSDUE(f) ≤ PSDmax · 10 10 (2.4)

where

2 · |f − fc| Bsub SEM(f) = max{−40 · log10( ) − 8, −60}, |f − fc| ≥ . (2.5) Bsub 2

The unwanted emission power in the RAO band with fRAO,L and fRAO,U as the lower and upper edges must satisfy5

fRAO,U Z SEM(foﬀ ) PUE(fRAO,L, fRAO,U) ≤ PSDmax · 10 10 df. (2.6) fRAO,L

Here, we assume the SCS allows the satellites to generate maximum allowable unwanted

emission power and therefore we can pick the equality in Eq. (2.6). Since PSDmax is related to the speciﬁc power distribution of the signals in the assigned band, without loss of generality,

10 Watt we consider PSDmax = 100 MHz and keep this value for diﬀerent Bsub in this chapter. In

Bsub Bsub addition, we assume PSD(f) = PSDmax for fc − 2 ≤ f ≤ fc + 2 so that we can evaluate the interference power if the RAO band is overlapping with the downlink band

of the satellites. Fig. 2.4 shows the relative attenuation function SEM(·) versus diﬀerent

relative frequency diﬀerence (separation) |f − fc|/Bsub values. It can be noticed from the

5See Chapter 5.6.2 of [61] for details about the equation.

16 0

-10

-20

-30

-40

-50 Relative spectral attenuation (dB) -60

0 50 200 400 600 800 1000 1200 Frequency difference relative to the assgined bandwidth B (%) sub

Figure 2.4. Relative PSD attenuation versus diﬀerent |f − fc|/Bsub values

curve that for the same Bsub and same RAO band, a larger frequency separation between the RAO band and the assigned subband, in general, leads to a smaller PUE. However, if the frequency separation is large enough (e.g., exceeding 1000% relative to Bsub), the PUE will reach its lower bound and no further RFI reduction can be achieved with a larger frequency separation due to the ﬂat spectral mask ﬂoor in Eq.(2.5).

As mentioned before, we focus on 3 RAS bands which are I) 18.28 – 18.36 GHz, II)

22.21 – 22.5 GHz and, III) 23.4 – 24.0 GHz. For the subbands shown in Fig. 2.2, the

10 Watt corresponding unwanted emission powers with PSDmax = 100 MHz are shown in Fig. 2.5. The x-axis represents the location of these subbands on the frequency spectrum and the y-axis

denotes the unwanted emission power of these subbands in the three RAS bands. From the

ﬁgure we can see that for diﬀerent subbands, the unwanted emission power remains the same

in the RAS bands II and III, which is due to that the frequency separations (2.01 GHz and

3.4 GHz) between the SCS downlink bands and the RAS bands exceed 1000% of Bsub (100 MHz). On the other hand, the downlink bands and the RAS band I are close to each other

and consequently the unwanted emission power of the subbands varies a lot in the RAS

17 SCS Band 1 20 SCS Band 2 SCS Band 3 0

-20

-40 (dBW) in RAS band I UE

P -60 17 17.5 18 18.5 19 19.5 20 20.5

0 -10 -20 -30 -40 (dBW) in RAS band II

UE -50 P 17 17.5 18 18.5 19 19.5 20 20.5

0 -10 -20 -30 -40 (dBW) in RAS band III

UE -50

P 17 17.5 18 18.5 19 19.5 20 20.5 SCS downlink bands frequency (GHz)

Figure 2.5. Unwanted emission power of the satellites’ transmission in subbands centered at different frequencies in different RAS bands band I. Specifically, for the two subbands centered at 18.25 GHz and 18.35 GHz which are overlapping with the RAS band I, the corresponding unwanted emission power is extremely high compared to the unwanted emission power of other subbands. This imposes a major concern of RFI in the two overlapping subbands.

2.2.5 RFI from Other Active Wireless Systems

Besides the GSO satellite systems, other active wireless systems, including the Non-Geostatio- nary Orbit (NGSO) satellites, the ground-based/airborne RADAR, and the ground cellular communication systems, can also generate harmful RFI to the ground telescopes and there-

18 fore should be included in the RFI evaluation. Without losing generality, we assume the

average RFI generated by other active wireless systems at telescope i consists of two parts, which are a slowly-changing part EPFDex,S,i(t) and a rapidly-changing part EPFDex,R,i(t).

The ﬁrst part is generated by some stable wireless sources like the TV broadcasting tower, whose RFI EPFD can be viewed as a constant EPFDex,S,i within the RAO duration. The

second part is caused by time-varying bursty communications from all other active wireless

systems, for which we can invoke the central limit theorem and the law of large numbers to

assume identically distributed complex Gaussian interference in diﬀerent RAO bands and

hence model EPFDex,R,i(t) with an exponential distribution as

EPFDex,R,i(t) ∼ E[β], t ∈ Tinte,i, (2.7)

where E[β] denotes an exponential distributed variable with mean β. Then, the average RFI

at telescope i caused by other active wireless systems in an RAO time duration Tinte,i can

be represented as

1 Z EPFDex,i(Tinte,i) = EPFDex,S,i + EPFDex,R,i(t)dt Tinte,i Tinte,i

≈ EPFDex,S,i + β. (2.8)

In practice, EPFDex,i should be low enough to meet the RFI and sample loss rate requirement

2 by ITU. In this chapter, if not speciﬁed, we consider EPFDex,S,i = −175 dBW/m and

β = −175 dBW/m2 for all the telescopes. With the above model, we can incorporate the eﬀects of other active wireless systems on the RFI and the corresponding RAO sample loss rate at the telescopes.

19 2.2.6 RFI and Sample Loss Rate of the Ground Radio Telescopes under GSO

SCS

RFI in Diﬀerent RAO Modes and RAO Directions

To address the RFI issue of the ground telescopes, we ﬁrst pick the Eﬀelsberg 100-m Radio

Telescope at Germany as an example to show the RFI received by the ground telescope in different observation modes and observation directions. The following assumptions are made for the different observation modes. For the target tracking mode, we assume the target’s initial position in earth coordinate is 86°E in longitude and 0° in latitude and changes with the earth rotation. For the sky-mapping mode, we suppose the telescope slews in the north-south direction with minimum elevation angle 10°and a speed of 40°/hour. For the fixed-direction mode, we consider the telescope randomly picks a direction from 2066 potential directions, which are the sampled directions with elevation angles larger than 10°. In addition, the

RAO direction changes every 30 minutes, reﬂecting a periodic target switch manner of the telescope. Fig. 2.6 shows the instantaneous RFI EPFD in RAS band I received by the telescope with the 3 observation modes and the corresponding beam directions in latitude and longitude. From the ﬁgure we can see that:

• Due to the elevation angle requirement, the telescope can track a certain target only

in a certain time window.

• In the sky-mapping mode, the instantaneous RFI at the telescope is periodic as the

beam direction of the telescope slews.

• In the ﬁxed-direction mode, the instantaneous RFI at the telescope is ﬁxed (varied)

when the telescope’s beam direction is ﬁxed (varied), which is consistent with our

discussion for Eq. (2.3).

20 Target-tracking Mode -100 Sky-mapping Mode )

2 Fixed-direction Mode

-110

-120

-130 RFI EPFD (dBW/m

0 4 8 12 16 20 24 90 180

0 0

Target-tracking Mode Beam direction Latitude

-90 -180 Beam direction Longitude 0 4 8 12 16 20Sky-mapping 24Mode Fixed-direction Mode Time (hr)

Figure 2.6. RFI at the Eﬀelsberg 100-m radio telescope in the RAS band I with diﬀerent RAO modes and the corresponding beam directions in latitude and longitude

To exhibit a comprehensive image of the RFI at the Effelsberg telescope from different directions and the relationship between the RFI level and the GSO satellites’ locations, we collect the RFI results of the all potential directions in the fixed-direction mode and demonstrate them in Fig. 2.7. The blue stars denote the locations of the GSO satellites which can be viewed from the Effelsberg telescope and the blue lines indicate the line-of- sight directions from the telescope to the satellites. The dome on the earth surface shows the celestial sphere viewed by the Effelsberg telescope with colors indicating the RFI levels at the corresponding directions. From the figure we can see that RAOs in the directions of the GSO satellites experience stronger RFI, while the RFI in other directions are relatively low, which is consistent with the descriptions about the RFI of the GSO satellite in [45]. In addition, considering the RAO integration time as 30 minutes, the corresponding detrimental

RFI threshold in the RAS band I is −153.9 dBW/m2. The RFI levels of the directions shown

21 Figure 2.7. RFI at the Effelsberg 100-m radio telescope in the RAS band I at different directions in different scenarios

in the ﬁgure, whether with or without GSO satellites lie in, exceed this threshold. This fact

points out the diﬃculty of RAO in the RAS band I for the Eﬀelsberg telescope.

RFI and Sample Loss Rate in Diﬀerent RAS Bands at Diﬀerent Ground Radio

Telescopes

To show the RFI and the corresponding sample loss rate in the 3 RAS bands at diﬀerent

ground telescopes, we use Monte Carlo method with 200 realizations of the RAO of the

telescopes and average the RFI and the sample loss rate among diﬀerent realizations. In each

realization, the telescopes will randomly pick an RAO mode among the 3 aforementioned

modes and conduct RAO in an operation duration of 24 hours. Speciﬁcally, in the target

tracking mode, the latitude of the target’s initial position in the earth coordinate can be

chosen from the set {90°S, 75°S, 60°S, 45°S, 30°S, 15°S, 0°, 15°N, 30°N, 45°N, 60°N, 75°N,

90°N}. In addition, the RAO integration time for the target-tracking mode is assumed

22 to be the time duration when the target is available for the telescopes to observe (the

target elevation angle is larger than 10°). For the sky-mapping mode and the ﬁxed-direction mode, the RAO integration time are 4 hours and 30 minutes, respectively. Then, with the integration time and the corresponding threshold, we can obtain the sample loss rate by dividing the number of the samples with RFI higher than the threshold during the RAO with the total number of samples recorded in the RAO.

Fig. 2.8 shows the average RFI at diﬀerent telescopes in the 200 random realizations.

From the ﬁgure we can see that the RFI in the RAS bands II and III are relatively low as the SCS-induced unwanted emission powers in the two RAS bands are relatively low. On the other hand, as the RAS band I and the SCS downlink bands overlap with each other, the

RFI in the RAS band I at different telescopes is approximately 40 dB higher than the RFI in the other 2 protected RAS bands. This RFI difference is much larger than the differences in the detrimental RFI thresholds (−154 dBW/m2, −146 dBW/m2, and −147 dBW/m2 for

the RAS bands I, II, and III, respectively with the 2000 seconds RAO integration time). In

addition, we can see that the average RFIs of diﬀerent telescopes may have 30 dB diﬀerence.

The dominant factor, which contributes to the diﬀerence, is the locations of the telescopes.

The telescopes surrounded by sea (e.g., telescope 36 on the Big Island of Hawaii in the

Paciﬁc Ocean) may receive less RFI than the telescopes on the land (e.g., telescope 3 at

North Liberty in Iowa) as more communication cells and potentially more spot beams are

located around the telescopes on land (which can be observed from Fig. 2.3).

Next, we evaluate the sample loss rate based on the RAO integration time mentioned in

the previous sections and demonstrate the results in Fig. 2.9. We can see from the ﬁgure that

the sample loss rate of RAO in the RAS band I in the 3 diﬀerent modes are approximately

100%, which indicates that the RAO in the RAS band I is completely corrupted by the RFI

from the SCS downlink. On the other hand, the sample loss rates of the RAO in the RAS

bands II and III are much lower, and the RAOs in these two bands are barely aﬀected by

23 Figure 2.8. Average RFI at diﬀerent telescopes in 200 random realizations the RFI from the SCS’s downlink. Nevertheless, some telescopes still have sample loss rate larger than the acceptable sample loss rate of 2% deﬁned in [49].

2.3 Existing RFI Reduction methods as Benchmarks

To address the eﬀectiveness and advantages of the proposed RFI reduction methods over the existing ones, we consider two RFI reduction methods adopted by the newly-launched Iridium Next satellites [23], which are 1. In-band and out-of-band emission suppression method: This method limits the in-band and out-of-band emission of the downlink signals. First of all, the SCS satellites stop using RAO bands for downlink transmission to eliminate the in-band interference to the ground telescopes. In addition, two guard bands on the upper and lower side of the RAO bands will be arranged to reduce the unwanted emission of the downlink

24 Figure 2.9. Average sample loss rate at diﬀerent telescopes in 200 random realizations

band leaking to the RAO band. Secondly, the SCS satellites use specially-designed

ﬁlter (e.g., a SAW ﬁlter) to reduce the out-of-band emission (OOBE) from the satellite

downlink to the RAO bands. It is believed that the OOBE of the satellite downlink

can be suppressed by 10 dB [66] compared to the OOBE of existing in-orbit satellites6.

However, according to [40], this method may cause larger power consumption in data

transmission.

2. Beam muting method: This method mutes the beams which cause strong RFI to the

telescopes. To simplify the simulation, we assume the SCS can switch oﬀ the beams

6The exact OOBE in the RAO band is not mentioned in the Iridium Next documents. Thereby, we 0 model the eﬀect of this emission suppression method in the RAO band as an emission mask with SEM(f) = 2·|f−fc| max{−40 · log ( ) − 8 − ∆SEM, −60} where ∆SEM is the OOBE suppression achieved by the emission 10 Bsub suppression method.

25 immediately when its accumulated RFI EPFD at the ground telescopes exceeds a

certain threshold. Consequently, the users served by these beams will lose connection.

The two methods suﬀer some inevitable service loss. Applying the emission suppression method may force the satellite to use low transmitting power or stop transmitting in the adjacent bands of the RAO band and therefore shrinking the downlink capacity. Applying the beam muting method may cause service outage of users in the corresponding cells during the operation period. For instance, the emission suppression method will cause at least

2.76% downlink throughput loss for suspending the transmission in the 80 MHz RAO band, aiming to suppress the RFI in the RAS band I. Meanwhile, the service loss of the beam muting method can be measured in terms of the percentage of the beam which are muted partly or completely during an operation duration (which means the users in the cell served by these beams will have service outage).

To illustrate more comprehensive performance of the existing RFI reduction methods under diﬀerent settings, we consider three diﬀerent application cases, which are

1. Emission suppression method only with OOBE reduction ∆SEM = 10 dB and guard

band bandwidth {0, 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000} MHz. The 0 MHz

guard band bandwidth indicates the case where we only prevent the data transmission

in the RAO band.

2. Beam muting method only with muting threshold {−160, −164, −168, −172, −176,

− 180} dBW/m2.

3. A joint use of the two aforementioned methods with ∆SEM = 10 dB and beam muting

threshold {−160, −164, −168, −172, −176, −180} dBW/m2.

In evaluating the above cases, we choose the RAS band I as an example. In addition, we

2 2 consider EPFDex,S,i = −175 dBW/m and β = −175 dBW/m in the simulation to address the RFI from other active wireless systems.

26 To evaluate communication service loss caused by the RFI reduction methods, we focus on the downlink throughput of the GSO SCS7. Suppose the transmission spectral efficiencies of different downlink subbands are the same, the downlink throughput of the GSO SCS is determined by the available spectrum resource. In addition, since some of the spot beams are muted to reduce RFI, the average throughput of these beams will decrease consequently. To avoid being restricted to specific transmission spectrum efficiency and operation duration values, we define our performance metric as the ratio of the average throughput of the GSO SCS with RFI reduction methods over that without RFI reduction methods as

PNGSO PNbeam 0 0 j=1 k=1 B j,k · Tj,k RA = (2.9) PNGSO PNbeam j=1 k=1 Bj,k · Tj,k

0 where Bj,k and B j,k are the average downlink bandwidth of satellite j’s kth spot beam

0 without and with RFI reduction methods respectively. Tj,k and Tj,k are the beam operation time of satellite j’s kth spot beam without and with RFI reduction methods respectively in the operation duration. This ratio reﬂects the average downlink throughput loss (gain) of the GSO SCS after applying the RFI reduction methods when compared to the original GSO SCS without RFI reduction. The emission suppression method may need to insert guard bands between the RAO bands and the downlink bands of the satellites and thus Bj,k will

0 be changed to B j,k. The beam muting method may mute the spot beams with high RFI and

therefore reduce Tj,k. In addition, for the beam muting method specifically, the spot beams of the GSO satellites can be divided into 3 groups, which are 1) the spot beams which are not muted during the operation duration, 2) the spot beams which are muted during the operation duration for some time, and 3) the spot beams which are completely muted in the operation duration. As the operation duration is relatively long (e.g., 24 hours), the users of the beams in the first group will suffer a long-time service outage, which would make the beam muting method less attractive to the SCS users and operators.

7We assume there always exist the demand of data service.

27 Fig. 2.10 shows the average sample loss rate and average throughput ratio across different telescopes in case 1 (emission suppression method only) with diﬀerent guard band bandwidths8. The label ‘none’ in the ﬁgure represents the case where no RFI reduction method is applied and the downlink bands of the GSO satellite are overlapping with the

RAO band. In addition, as mentioned before, the ‘0 MHz’ label indicates the case where the

GSO satellites only stop using the RAO band as downlink without inserting any guardband between the GSO downlink bands and the RAO band. As we can see from the ﬁgure, the sample loss rate and the average throughput ratio decrease with a larger guard band bandwidth. Nevertheless, the sample loss rate is above 5% even when the guard band bandwidth is 1000 MHz, which indicates that the emission suppression method alone cannot reduce the

RFI-induced sample loss rate to a desired level (less than 2% per ITU-R). In addition, when the guard band bandwidth is larger than 600 MHz, the sample loss rate cannot be further reduced.

Fig. 2.11 shows the average sample loss rate in case 2 (beam muting only) across different muting thresholds. We show the percentage of the three groups of spot beams with respect to the total number of spot beams in Fig. 2.11 to demonstrate the service loss of the beam muting method. It can be observed from the figure that the beam muting method alone can lower the sample loss rate to 2% or less on the condition that around 15% of spot beams do not provide service and about 35% of spot beams may temporally shut down during the operation duration. Comparing Fig. 2.10 and Fig. 2.11, we can find that the emission suppression method can achieve less sample loss rate when the average throughput ratio is above around 88% and the corresponding guard band bandwidth is 300 MHz. Thus, we consider applying this guard band bandwidth in the 3rd case (joint emission suppression and beam muting).

8The guard band bandwidth includes both the upper and the lower guard bands.

28 100 Average throughput ratio 90 Sample loss rate 2% line 80

Percentage 40

0 None 0 100 200 300 400 500 600 700 800 900 1000 Guard band bandwidth (MHz)

Figure 2.10. Sample loss rate and average throughput ratio in case 1 with diﬀerent guard band bandwidths

To show the performance of jointly using the two RFI reduction methods, we plot the average sample loss rate and the average throughput ratio in case 3 across diﬀerent muting thresholds with ∆SEM = 10 dB and 300 MHz guard band bandwidth in Fig. 2.12. Com- paring Fig. 2.12 and Fig. 2.11, we can see that similar sample loss rate can be achieved with larger average throughput ratio when jointly applying the two RFI reduction methods.

Nevertheless, up to 16% of spot beams are required to stop providing service temporarily during the operation duration and about 12% average throughput ratio is lost if we want to achieve the sample loss rate less than 2%.

In general, the existing RFI reduction methods can lower the RFI and sample loss rate for RAS. But the corresponding service loss of SCS will be high which is unfavorable to the

GSO SCS. Therefore, we propose a spectrum sharing paradigm with three RFI reduction methods which can reduce RFI and sample loss rate for RAS as well as increase the downlink capacity for the SCS.

29 100

60 Completely muted 50 Partly muted Not affected

Percentage 40 Average throughput ratio Sample loss rate 2% line 30

0 None -160 -164 -168 -172 -176 -180 Muting threshold (dBW/m2)

Figure 2.11. Sample loss rate and average throughput ratio in case 2 with diﬀerent muting thresholds

100

80 Completely muted 70 Partly muted Not affected Average throughput ratio 60 Sample loss rate 2% line 50

Percentage 40

0 None -160 -164 -168 -172 -176 -180 Muting threshold (dBW/m2)

Figure 2.12. Sample loss rate and average throughput ratio in case 3 with diﬀerent muting thresholds

30 2.4 Proposed Spectrum Sharing Paradigm with Three RFI Reduction Methods

2.4.1 RFI Determination for Ground Radio Telescopes in Diﬀerent Time Du-

rations

From the analyses in the previous sections we can see that the RAO in the unprotected RAS band I is completely lost due to the SCS-induced RFI issue and the RAO in the protected

RAS band II and III still suﬀer some sample loss rate higher than the ITU-R deﬁned threshold of 2%. Considering the detrimental RFI in the three RAS bands and the demand of SCS for extra downlink bandwidth, we propose a spectrum sharing paradigm between the GSO

SCS and the RAS. Suppose each telescope only conducts RAO in one of the 3 RAS bands in one operation duration (e.g., 24 hours). The diﬀerent combinations of the RAO bands at all telescopes can be divided into 7 scenarios as follows:

1. RAO in RAS band I only

2. RAO in RAS band II only

3. RAO in RAS band III only

4. RAO in RAS band I and II only

5. RAO in RAS band I and III only

6. RAO in RAS band II and III only

7. RAO in all the 3 RAS bands.

Then, depending on the idle RAS band(s) oﬀered by the RAS, we can design the corresponding band allocation to diﬀerent satellite groups. To implement our proposed RFI reduction methods, we divide the whole operation duration (e.g., 24 hours) into Np equal periods, and divide each period into Nsp equal subperiods. For RFI and fairness concern, we design the band allocation plan for each group of satellites in a period and a subband allocation plan for each satellite in each subperiod. As mentioned in Section 2.2-C, the satellites in the same group use same SCS band for downlink transmission. To avoid causing in-band interference

31 RAO operation duration Period 1 Period 2 1 2 3 1 2 3 RAO integration time of telescope i Time Subperiod RAO integration time in subperiod 1 of period 1

Figure 2.13. An example of periods and subperiods in an RAO operation duration

between diﬀerent satellites, we design and assign the new SCS bands (groups of subbands

which are designed with our proposed RFI reduction method) to the satellite groups instead

of individual satellites. Similarly, the subbands in the SCS band which has been allocated

to a certain satellite are assigned to the beam groups (as mentioned in Section 2.2-C) in-

stead of individual beams of the satellite to avoid interference between diﬀerent beams of

the satellite. Although this design may reduce the ﬂexibility in bands/subbands allocation,

it on the other hand narrows down the search region of possible allocation and circumvents

the interference issue between diﬀerent satellites and their spot beams.

Fig. 2.13 shows an example of the telescope i observing in an operation duration with

Np = 2 and Nsp = 3. Due to the availability of the target, the telescope may have fractional integration time in some subperiods. In addition, diﬀerent telescopes may observe in diﬀerent

RAS bands9. Supposing the telescope i observes RAS band d in the considered operation duration, the average RFI EPFD at the telescope i generated by GSO satellites in Eq. (2.3) can be formulated as

Np Nsp Nbeam Z X X X X GT,j,k,d EPFDi,d(Tinte,i) = PUE,j,k,n,q,d · 2 GR,i,j,d(t)dt (2.10) 4πdi,jTinte,i Tinte,i,n,q n=1 q=1 j∈IGSO,i k=1

where PUE,j,k,n,q,d is the unwanted emission power of satellite j’s kth beam in RAS band d at

period n’s qth subperiod, Tinte,i,n,q is the integration time in period n’s qth subperiod, GT,j,k,d

9We assume their observation bands are not changed during an operation period.

32 is the corresponding satellite transmit antenna gain towards the telescope, and GR,i,j,d is the corresponding telescope’s receiving antenna gain. To further simplify Eq. (2.10), we deﬁne the equivalent gain factor at period n’s qth subperiod in RAS band d as R GT,j,k,d GR,i,j,d(t)dt Tinte,i,n,q Gi,j,k,d(Tinte,i,n,q) = 2 . (2.11) 4πdi,j · Tinte,i Then, Eq. (2.10) can be represented as

Np Nsp N X X X Xbeam EPFDi,d(Tinte,i) = PUE,j,k,n,q,d ·Gi,j,k,d(Tinte,i,n,q). (2.12)

n=1 q=1 j∈IGSO,i k=1 The average RFI EPFD at telescope i caused by all active wireless systems can be represented as

EPFDi,d,all = EPFDi,d(Tinte,i) + EPFDex,i(Tinte,i) (2.13)

and the RAO performance and sample loss rate of telescope i are related to this value. To assess the overall RFI eﬀect of active wireless systems on the RAS, we consider the sum of the average RFI at all telescopes in an operation duration which is deﬁned as

X X EPFDsum = EPFDi,d,all (2.14)

d∈D i∈IRAS,j,d where D is the RAO band index set which indicates the RAS band(s) that the telescopes observe in the operation duration and IRAS,j,d is the index set of telescopes that observe RAS band d. Our goal here is to minimize EPFD via minimizing P P EPFD (T ) sum d∈D i∈IRAS,j,d i,d inte,i as well as to guarantee a certain downlink bandwidth for diﬀerent groups of GSO satellites.

From Eq. (2.12), we can see that EPFDi,d(Tinte,i) is determined by the sum of the products of the unwanted emission powers and the equivalent gain factors of diﬀerent beams in diﬀerent subperiods. Based on this observation, we design three RFI reduction methods which aim to 1) lower unwanted emission power of the subbands via spectrum resource reorganization, 2) reduce equivalent antenna gain via beam switching, and 3) minimize the product of unwanted emission power and equivalent antenna gain via a subband allocation design.

33 2.4.2 Reorganization of the Spectrum Resource to Minimize the Unwanted Emission Power of the GSO Satellites in the RAO Bands

As we can see from Eq. (2.12), the RFI can be reduced via lowering the unwanted emission power of the subbands. For scenarios 1∼5 of Section 2.4.1, as one or two protected RAS band(s) are spare at RAS side, the SCS satellites can use this spectrum resource to reduce the RFI to the telescopes as well as to increase their downlink capacity. To ﬁt into the spectrum usage of GSO SCS, the overall available spectrum resource should be divided into 30 subbands. Then, the subbands are grouped into 3 subband groups and assigned to the satellite groups for use. Thus, the reorganization of the spectrum resource can be divided into 2 steps, which are 1) subbands design and grouping, and 2) subband groups assignment to the satellite groups. Subbands design and grouping: While designing the subbands of the SCS in the available spectrum resource, we should take the following issues into consideration:

• From Fig. 2.5, we can see that the PUE’s caused by the SCS downlink subbands which are overlapping with the RAO band are extremely high. Therefore, the SCS should not use the RAO band for its downlink transmission. • Although the bandwidth of the SCS subbands can take arbitrary value within the total available bandwidth, for simplicity we assume that the subband bandwidth should be

a multiple of a unit bandwidth Bsub,unit and this value can be viewed as the bandwidth of one channel or one resource block. • Given the RAO bands, the available spectrum resource will naturally be in a form of several discontinuous spectrum blocks. For example, in scenario 1, the available spectrum resource consists of 4 blocks which are 17.3 – 18.28 GHz, 18.36 – 20.2 GHz, 22.21 – 22.5 GHz, and 23.6 – 24 GHz. Fig. 2.14 shows a potential spectrum resource allocation of the SCS in scenario 1 based on the aforementioned 4 spectrum blocks. The ﬁgure also explains the relationship between the spectrum blocks and subband

34 RAO band

Spectrum Spectrum Spectrum block 1 Spectrum block 2 Subband group 1 block 3 block 4 Subband group 2 1 1 2 3 4 5 6 7 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 Subband group 3 9 10 8 9 10 35 0 Frequency 17.3 GHz 18.28 GHz 18.36 GHz 20.2 GHz 22.21 GHz 22.5 GHz 23.6 GHz 24 GHz

Figure 2.14. Potential spectrum resource allocation of the SCS in scenario 1 groups. As we can see, the spectrum blocks are the available spectrum resources which

are separated by unauthorized bands (the bands which are not allocated for the SCS)

or RAO bands. Considering the hardware complexity of the transceiver components

and the electromagnetic compatibility of the satellites, we assume that each subband

group can have the subbands from at most 2 diﬀerent spectrum blocks.

• The bandwidths of the subbands in the same block are designed to be the same. So,

each subband group may have subbands with at most 2 diﬀerent bandwidths which in

turn makes it easier to facilitate a fair or balanced distribution of bandwidth among

spot beams.

• A minimum bandwidth Bsub,0 of the subbands should be arranged to ensure that the

capacity/service performance of a single spot beam will not be degraded if compared

to the original band assignment in Fig. 2.2.

Consider a scenario with L discontinuous spectrum blocks10, where we use the block index l = {1, 2, ..., L} to mark the block with the center frequency from low to high and denote the number of subbands in block l as Nsub,l. To clearly distinguish subbands in diﬀerent blocks, we number the subband according to its center frequency from low to high as isub ∈ Isub, where Isub denotes the index set of the subbands. In addition, we denote the bandwidth of each subband in block l as Bsub,l, the total bandwidth of the subbands in block l as Bl and

the unwanted emission power in RAS band d caused by subband isub as PUE,isub,d. Then, we can formulate the spectrum resource reorganization problem as the one which minimizes the

10L depends on the RAO scenarios: L = 4 for scenario 1, L = 3 for scenarios 4 and 5, L = 2 for scenarios 2, 3, and 7, and L = 1 for scenario 7.

36 average unwanted emission power among the subbands as

1 X X arg min P (2.15) {Bsub,l,Nsub,l} UE,isub,d |Isub| isub∈Isub d∈D

s.t.Bsub,l ≥ Bsub,0 ∀l ∈ {1, 2, 3, ..., L},

mod (Bsub,l,Bsub,unit) = 0 ∀l ∈ {1, 2, 3, ..., L},

Nsub,l · Bsub,l ≤ Bl ∀l ∈ {1, 2, 3, ..., L},

L X Nsub,l = |Isub|, l=1 L X Nsub,l · Bsub,l ≥ BSCS,0, l=1

where |Isub| means the total number of subband indexes, which is equal to the number of

subbands we need to arrange for the SCS. BSCS,0 is the minimum total bandwidth required by the SCS. mod (A, B) means A modulo B. Generally, we group the subbands according to

their center frequencies to reduce hardware complexity and therefore we will have subbands

isub = 1, 2, ..., 10 as subband group 1, isub = 11, 12, ..., 20 as subband group 2 and isub = 21, 22, ..., 30 as subband group 3. However, for scenario 1 speciﬁcally, as the bandwidths of

the last spectrum blocks are too narrow (which can be found in Fig. 2.14) and not enough

to form a subband group, we decide to have the subbands in spectrum block 1 and 4 as a

subband group. Consequently, the number of subbands in the spectrum blocks should satisfy

Nsub,1 + Nsub,4 = 10 and Nsub,2 + Nsub,3 = 20.

To demonstrate the eﬀects of BSCS,0, Bsub,0, and Bsub,unit on the unwanted emission power 1 of the subbands, we plot the average unwanted emission power P P in isub∈Isub UE,isub,1 |Isub| the RAS band I versus the minimum total SCS required bandwidth BSCS,0 under diﬀerent

Bsub,0 and Bsub,unit values in scenario 1 in Fig. 2.15. From the ﬁgure we can see that the average unwanted emission power decreases as the BSCS,0 reduces in general. However, a

ﬂoor/limit can be reached for the case with Bsub,0= 100 MHz and BSCS,0 ≤ 3000 MHz and

37 -15 B = 80 MHz, B = 1 MHz sub,0 sub,unit B = 90 MHz, B = 1 MHz sub,0 sub,unit -20 B = 100 MHz, B = 1 MHz sub,0 sub,unit B = 90 MHz, B = 2 MHz sub,0 sub,unit B = 90 MHz, B = 5 MHz -25 sub,0 sub,unit

-37.5 -30 -38

-38.5 -35 -39

-39.5 -40 -40

Average unwanted emission power (dBW) -45

-50 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400 3500 B (MHz) SCS,0

Figure 2.15. Average unwanted emission power versus minimum total SCS required bandwidth BSCS,0 under diﬀerent Bsub,0 and Bsub,unit values in scenario 1

the cases with Bsub,0= 90 MHz and BSCS,0 ≤ 2700 MHz. The ﬂoors of the curves are owing

to that Bsub,0 substitutes BSCS,0 to limit the total bandwidth used by the SCS. In addition, comparing the 5 curves in the figure we can see that smaller Bsub,0 and Bsub,unit can offer more flexibility in reducing the average unwanted emission power in some cases. Nevertheless, the

improvement provided by smaller Bsub,0 and Bsub,unit is limited (e.g., less than 2 dB) and

insigniﬁcant compared to the eﬀect of BSCS,0. Therefore, in the rest of the chapter, if not

speciﬁed, we consider Bsub,unit = 1 MHz to obtain relatively low unwanted emission power

and choose Bsub,0= 90 MHz as the smallest subband bandwidth in the original subbands allocation plan is 90 MHz.

Based on the aforementioned Bsub,0 and Bsub,unit settings, we show in Fig. 2.16 the average unwanted emission powers in the corresponding RAS bands in the ﬁrst 5 scenarios of

Section 2.4.1 with diﬀerent BSCS,0 values. Due to the large frequency separation between the RAO band and the SCS downlink bands in scenarios 2 and 3, the average unwanted

38 emission power is relatively low and unchanged with the BSCS,0 values we use. Hence, the RFI at the telescopes will not increase if we allow SCS to use almost all of the idle RAS bands for its downlink transmission. Then, the total bandwidth which can be used by SCS

PL 11 can be BSCS,used = l=1 Nsub,l · Bsub,l = 3288 MHz in scenario 2 and BSCS,used = 3177

MHz in scenario 3. On the other hand, in scenarios 1, 4, and 5, a larger BSCS,0 causesa larger average unwanted emission power, and hence a trade-off should be made between the total bandwidth of the SCS and the average unwanted power. In addition, as the SCS owns a total of 2900 MHz bandwidth in the original band allocation plan, it would be in general unacceptable to SCS to arrange bandwidth less than this amount which serves as a lower limit for the trade-off range. Subband groups assignment: After obtaining the 3 subband groups, we need to assign them to 3 satellite groups as we discussed in the previous section. In the subband group assignment design, two factors should be taken into consideration, which are the fairness in the downlink bandwidth amounts of different satellite groups and the RFI generated by these satellites. Nevertheless, we regard the fairness among different satellite groups as the major concern in the subband group allocation design. To guarantee the balanced distribution of the downlink bandwidth among different groups of satellites, we apply different band assignments in different periods to ensure the average downlink bandwidth of a certain satellite group over BSCS,used to be around an expected value. Denote the subband group used by the satellite group m at the nth period by im,n (im,n ∈ {1, 2, 3}). The average bandwidth of the satellite group m in the operation duration is B = 1 PNp B group,m Np n=1 im,n

where Bim,n denotes the total bandwidth of subband group im,n. Then, the average ratio of

11 In scenario 2 speciﬁcally, if we allow SCS to have BSCS,used = 3250 MHz, the average unwanted emission power is around -45 dBW which remains the same for smaller BSCS,used (as shown in Fig. 2.16). However, if we allow SCS to have BSCS,used = 3288 MHz, the nearest subband (isub =30) to the RAS band II will have a little bit higher unwanted emission power than the lower bound of the unwanted emission power with the emission mask. Nevertheless, the increase in the average unwanted emission power is only 0.56% and can be mitigated with other RFI reduction methods we apply in this chapter. Therefore, we still allow SCS to use that much BSCS,used in scenario 2 to oﬀer extra downlink bandwidth to SCS as much as possible

39 the spectrum resource used by the satellite group m in an operation duration over the total

Bgroup,m used spectrum resource BSCS,used is rgroup(m) = . Naturally, we can form an average BSCS,used

ratio vector rgroup with the mth element rgroup(m). To fairly assign the subband groups to the satellite groups, the assignment problem can be formulated as

2 argim,n min||rgroup − rgroup,0|| (2.16)

s.t. im,n ∈ {1, 2, 3},

im1,n 6= im2,n, ∀m1, m2 = 1, 2, ..., M and m1 6= m2

where rgroup,0 is a target spectrum resource usage ratio vector we want to obtain via the

subband groups assignment. Here, we consider rgroup,0 to be the average spectrum resource usage ratio vector of SCS with the original band assignment mentioned in Section 2.2-C. With a limited number of periods in the operation duration, an exhaustive search can be applied to ﬁnd the solution of this problem. If there are multiple {im,n} combinations that can minimize the target function, we can pick the one which generates minimum average RFI in the operation duration among the telescopes.

2.4.3 Rearranging the Subband Allocation of the GSO Satellite’s Beams to Minimize the RFI of the GSO Satellites

From Section 2.2-D, we know that the unwanted emission powers in the RAO band caused by the different SCS downlink subbands can be fairly different (e.g., about 50 dB difference in the RAS band I in Fig. 2.5). This offers us some flexibility in reducing the RFI via carefully assigning the subbands to the spot beam groups. Based on Eq. (2.12) and Eq. (2.14), the average RFI EPFD generated by the satellite j in period n’s qth subperiod EPFDj,n,q can be represented as

N X X Xbeam EPFDj,n,q = PUE,j,k,n,q,d ·Gi,j,k,d(Tinte,i,n,q). (2.17)

d∈D i∈IRAS,j,d k=1

40 -15 Scenario 1 Scenario 2 Scenario 3 -20 Scenario 4 Scenario 4 RAS band I Scenario 4 RAS band II -25 Scenario 5 Scenario 5 RAS band I Scenario 5 RAS band III

-30

-35

-40

Average unwanted emission power (dBW) -45

-50 2700 2800 2900 3000 3100 3200 3300 3400 3500 B (MHz) SCS,0

Figure 2.16. Average unwanted emission power in diﬀerent scenarios with diﬀerent BSCS,0 values

As mentioned in Section 2.2-C, the spot beams of a satellite are divided into 10 groups and

the beams in the same group use the same subband. Then, Eq. (2.17) can be formed as

10 X X X X EPFDj,n,q = [PUE,j,p,n,q,d · Gi,j,k,d(Tinte,i,n,q)] (2.18)

p=1 d∈D k∈IBG,j,p i∈IRAS,j,d where PUE,j,p,n,q,d denotes the unwanted emission power of the subband which is assigned to the beam group p in RAS band d in period n’s qth subperiod. IBG,j,p is the beam index set of the beams belonging to the beam group p. As mentioned in Section 2.4-B, each group of satellites will be assigned with a group of subbands for downlink in a period. If the satellite is assigned with the subbands which share same subband bandwidth, the fairness among diﬀerent beam groups of the satellite can be achieved with arbitrary subband allocation and therefore we can focus on minimizing the RFI generated by the satellite in the individual

41 subperiod. Otherwise, the fairness issue among diﬀerent beam groups regarding the downlink bandwidth needs to be considered and consequently the subband allocation of the satellite in the subperiods which belong to the same period may aﬀect each other. Thus, we decompose the RFI minimization problem and minimize the RFI at the telescopes in an RAO operation duration via minimizing the RFI of individual satellites in individual periods.

Let us ﬁrst formulate and solve the RFI minimization problem for the subperiods in a period without considering the fairness issue (e.g., the subbands in the considered subband group share the same bandwidth) and then demonstrate a simple method to obtain a suboptimal solution under fairness requirement. Assume during period n’s qth subperiod, the satellite j uses the subband group with subband index set Isub,group,j,n ⊆ Isub and denote the subband which is used by the beam group p on satellite j in period n’s qth subperiod as isub,j,n,p,q. Then, the RFI minimization problem for the satellite j in the considered subperiod can be formulated as

arg min EPFD (2.19) isub,j,n,p,q j,n,q

s.t. isub,j,n,p,q ∈ Isub,group,j,n, ∀p ∈ {1, 2, ..., 10}

isub,j,n,p1,q 6= isub,j,n,p2,q, ∀p1 6= p2 ∈ {1, 2, ..., 10}.

We can use Rearrangement Inequality12 to solve this problem for |D|=1. For |D| ≥ 2 case, an exhaustive search with size 10! can be applied to ﬁnd the subband allocation that minimizes the RFI for each subperiod.

The solutions in the previous paragraph are valid only for the case when the satellites are using the subbands with the same bandwidth. From the previous sections we know that a subband group may contain subbands with two diﬀerent bandwidths. In this case, the

12 The Rearrangement Inequality states that xny1 + ... + x1yn ≤ xσ(1)y1 + ... + xσ(n)yn ≤ x1y1 + ... + xnyn for every choice of real numbers x1 ≤ ... ≤ xn and y1 ≤ ... ≤ yn and every permutation xσ(1), ..., xσ(n). It oﬀers the upper and lower bound of the sum of the pairwise products of two real number sequences [32].

42 fairness in downlink bandwidth between the users of diﬀerent spot beams (cells) should be

guaranteed. Suppose the subbands in the subband groups Isub,group,j,n can be divided into two groups according to the bandwidth, which are NA subbands with bandwidth Bsub,A and 10−

13 NA subbands with bandwidth Bsub,B. For simplicity in design, we allocate Nsp,A subperiods

to the subbands with bandwidth Bsub,A and Nsp − Nsp,A subperiods to the subbands with

bandwidth Bsub,B for each group of spot beams. Then, the average subband bandwidth ¯ used by the spot beams will be Bsub = [Nsp,ABsub,A + (Nsp − Nsp,A)Bsub,B]/Nsp. We want to ¯ choose a pair of Nsp and Nsp,A values to have Bsub = [NA · Bsub,A + (10 − NA) · Bsub,B]/10. A

convenient solution of Nsp and Nsp,A is Nsp = 10 and Nsp,A = NA. In the rest of the chapter, we adopt this solution to avoid increasing the complexity of the subbands allocation design

in a period. In this situation, isub,j,n,p,q will be aﬀected by the subband allocations in other

subperiods within the same period. An exhaustive search with (10!)Nsp possible solutions is

computationally prohibitive. Therefore, we propose a method to ﬁnd a suboptimal solution

with less computation complexity. Deﬁne the expected RFI of the period n’s qth subperiod

of the satellite j as

10 X 1 X X X X EPFD = [( P ) · ( G (T ))] exp,j,n,q 10 UE,isub,d i,j,k,d inte,i,n,q d∈D isub∈Isub,group,j,n p=1 k∈IBG,j,p i∈IRAS,j,d (2.20)

where PUE,isub,d denotes the unwanted emission power of the subband with index isub in the RAS band d. We can view 1 P P as the average unwanted emission 10 isub∈Isub,group,j,n UE,isub,d power in the RAS band d caused by diﬀerent SCS downlink subbands which are assigned to satellite j in period n and (P10 P P G (T )) as the total equivalent p=1 k∈IBG,j,p i∈IRAS,j,d i,j,k,d inte,i,n,q gain factor of RAS band d in the period n’s qth subperiod. Then, we sort the subperiods

0 in period n according to EPFDexp,j,n,q from high to low. Denote q ∈ {1, 2, ..., Nsp} as the subperiod in period n with q0th largest expected RFI of satellite j. Then, we design the

13From Section 2.2-C we know that the number of subbands in a subband groups is 10.

43 0 subband allocation isub,j,n,p,q0 , p ∈ {1, 2, ..., 10} at the period n and the subperiod q , one by one from the subperiod with large expected RFI to the one with low expected RFI (e.g., q0 from 1 to Nsp) for satellite j. After obtaining the subband allocation of one subperiod, we check the number of subperiods with two diﬀerent subband bandwidths that each spot beam group is assigned with and eliminate some subband allocations in the next subperiod which cannot meet the fairness requirement. For example, after acquiring subband allocations in several subperiods, suppose the spot beam group p has already been assigned with Nsp,A subperiods with subband bandwidth Bsub,A. Then, when designing the subband allocation of the remaining subperiods, we will ﬁnd the subband allocation which minimizes EPFDj,n,q0 under the constraint that the spot beam group p is allocated a subband with bandwidth

Bsub,B. The detailed procedure is shown in Fig. 2.17. With this method, we obtain a suboptimal subband allocation during a period in the operation duration which meets the

fairness requirement.

2.4.4 Cell-based Beam Switch Approach to Suppress the RFI

As mentioned in Section 2.2-C, spot beams from diﬀerent satellites may point to the same

cell. Here we show an example of spot beams which serve a cell near the Eﬀelsberg 100-m

Radio Telescope and corresponding instantaneous RFI generated by the spot beams during

an RAO task in a target-tracking mode in Fig. 2.18. From the ﬁgure we can see that,

depending on the relative positions of the satellites, the SCS spot beams’ directions, and

the RAO beam (target) direction, the RFI generated by diﬀerent spot beams can have

approximately 30 dB diﬀerence and the times of the peaks of the instantaneous RFI EPFD

are different. The huge RFI differences between different spot beams enable us to reduce

the RFI at the telescopes via enabling a spot beam of one satellite to support the users of

another satellite in the same cell. To simplify its application in practice, we assume that

each spot beam can support users of at most one other spot beam of another satellite and

44 Calculate expected RFI of the subperiods in period n.

Sort subperiods based on the expected RFI from high to low

Yes End p’ > Nsp ?

No Design subband alocation of p’th subperiod based on the p’ = p’+1 available subband allocation set to minimize the RFI in the subperiod

Check the fairness requirement for beam groups and update the available subband allocation set.

Figure 2.17. Procedure diagram for rearranging the subband allocation at period n

-150 Satellite 61 Satellite 66 -160 Satellite 118 Satellite 132 Satellite 157 Satellite 163 -170 ) 2

-180

EPFD (dBW/m -190

-200

-210 0 100 200 300 400 500 600 700 Time (min)

Figure 2.18. RFI generated by the spot beams from diﬀerent satellites which serve a cell near the Eﬀelsberg 100-m Radio Telescope

45 the corresponding spot beam can stop transmission and be muted during the subperiod to

reduce the RFI at the telescopes. We name the spot beam with extra users as the supporter

beam and the spot beam of another satellite which is supported by the supporter beam as the

muted beam. The supporter beam can generate at most two data streams simultaneously on

two diﬀerent subbands and therefore can take place of the muted beam of another satellite

which serves the same cell as the supporter beam. In addition, we assume the subband

which conveys the data stream from the supporter beam will be the same as the one used

by the muted beam and therefore the users of the muted beam only need to adjust their

pointing direction to receive the signals from the supporter beam. To avoid the interference

between the two data streams on the supporter beam, the satellite with the supporter beam

and the satellite with the muted beam should belong to diﬀerent satellite groups so that the

subbands used by the supporter and the muted beams are diﬀerent. In addition, considering

the capacity of other links (e.g, the satellite to gateway link which oﬀers the Internet access),

each satellite is assumed to be able to support at most Nbeam + Nbeam,ex data streams with

Nbeam,ex denoting the extra number of data streams that can be supported by the satellite. If there is no limitation on the total number of data streams one satellite can support,

we can ﬁnd the optimal solution of the beam switch, cell by cell, via an exhaustive search.

Otherwise, we propose a suboptimal beam switch algorithm that meets the spot beam lim-

itation per satellite. Based on Eq. (2.17), the RFI generated by satellite j’s kth spot beam

in period n’s qth subperiod without the beam switch is

X X EPFDj,k,n,q = PUE,j,k,n,q,d ·Gi,j,k,d(Tinte,i,n,q). (2.21)

d∈D i∈IRAS,j,d

0 0 Deﬁne a satellite and beam index set Icell,j,k where the (j , k ) element belongs to Icell,j,k if the satellite j0’s k0th spot beam serves the same cell as the satellite j’s kth spot beam and the satellites j and j0 are not in the same satellite group. Then, we can ﬁnd an alternative

46 RFI for the satellite j’s kth beam EPFDA,j,k,n,q such that   EPFDj,k,n,q, Icell,j,k = ∅   EPFDA,j,k,n,q =   P P  min [PUE,j,k,n,q,d · Gi,j0,k,d(Tinte,i,n,q)], otherwize.  0 0 d∈D i∈IRAS,j0,d {j ,k }∈Icell,j,k (2.22)

Next, we can construct two NGSO × Nbeam RFI matrices R0 and RA with R0(j, k) =

EPFDj,k,n,q and RA(j, k) = EPFDA,j,k,n,q. Then, the algorithm for searching the beam switch plan for period n and subperiod q is described in algorithm 1 shown in the next page.

There, line 4 indicates that in each iteration, we pick the beam switch pair which makes

the most RFI reduction. In line 8, we update the beam index set Icell,j,k to avoid for other spot beams choosing the beam {j0, k0} as an alternative beam as this beam has supported the users of the beam {j, k}. Line 3 is the termination condition of the iteration by which the iteration will stop if no further RFI reduction can be achieved via the beam switch.

An alternative termination condition can be the total number of beam switches, which also limits the number of users that need to slew their antenna directions.

In addition, from the previous section we know that the subband allocation is related to the equivalent gain factors, which will be changed if the beam switch method is applied.

On the other hand, the beam switch plan is also aﬀected by PUE,j,k,n,q,d. For example, for multiple spot beams which choose the same spot beam as an alternative, the one with higher

PUE,j,k,n,q,d may gain some priority in designing the beam switch plan as more RFI reduction can be achieved. Since the two methods are dependent on each other, we offer a simple way to use them jointly. We can first design the subband allocation plan without considering the beam switch and fairness among spot beams, which can be viewed as an optimal plan in the subband allocation. Based on the subband allocation plan, we use the aforementioned algorithm to find the beam switch plan. Then, with the beam switch plan obtained, we

47 Algorithm 1 Algorithm for the beam switch plan Input: Unwanted emission power PUE,j,k,n,q,d; Equivalent gain factors Gi,j,k,d(Tinte,i,n,q); Satellite and beam index set Icell,j,k; Maximum number of data streams per satellite Nbeam + Nbeam,ex. Output: Beam switch plan matrix for subperiod q of period n: Bswitch,n,q Initialization 1: Find the RFI matrices R0 and RA 2: Initialize the elements in the number of data streams vector Ndata,re as Nbeam LOOP Process 3: while max(R0 − RA) > 0 do 4: Find {j, k} = arg{j,k} max (R0(j, k) − RA(j, k)) 5: Find {j0, k0} that leads to EPFDA,j,k,n,q 6: Record the beam switch {j, k} → {j0, k0} in Bswitch,n,q 7: EPFDj,k,n,q = EPFDA,j,k,n,q, Ndata,re(j0) = Ndata,re(j0) + 1, Ndata,re(j) = Ndata,re(j) − 1

8: Update Icell,j,k for those {j, k} with {i0, j0} as an alternative beam and update EPFDA,j,k,n,q and RA accordingly. 9: Check Ndata,re and update Icell,j,k, EPFDA,j,k,n,q and RA for Ndata,re(j) = Nbeam + Nbeam,ex 10: end while 11: return Bswitch,n,q update the equivalent gain factors for the spot beams and re-design the subband allocation plan with the fairness concern.

Note that RAO tasks at the observatories are scheduled well in advance and the GSO satellites and radio telescopes are stationary with respect to each other. Thus, in implementing the proposed methods, dynamic coordination between SCS and RAS is not needed within each RAO schedule time frame (which is several orders of magnitude longer than the communication side time frame). The SCS just needs to obtain detailed information of RAO tasks for each schedule time frame in advance from the RAS and then execute the proposed RFI reduction methods for each RAO schedule time frame without needing further coordination. Executing the proposed approaches is similar to conducting existing (adap-

48 tive) resource allocation schemes of cellular systems but generally at a slower rate due to the

relatively stationary environment of GSO SCS. Thus, it is within practical feasibility and

cost. As will be shown in the next section, the proposed approaches successfully address

the critical issue of friendly coexistence between GSO SCS and RAS and furthermore yield

mutual beneﬁts to both sides. These advantages clearly outweigh the cost of implementation.

2.5 Numerical Results of RFI and Sample Loss Rate

2.5.1 Eﬀects of the Parameters of the Three RFI Reduction Methods

First of all, to demonstrate the eﬀect of BSCS,0 in the reorganization of the spectrum resource method on RFI and sample loss rate, we consider a realization in scenario 1 and plot the average RFI and the sample loss rate at the telescopes with diﬀerent values of BSCS,0 in

Fig. 2.19. Comparing Fig. 2.16 and Fig. 2.19, we can see that with a smaller BSCS,0 required by the SCS, a lower average unwanted emission power can be achieved which generally leads to lower average RFI and lower sample loss rate. Nevertheless, a sample loss rate of about

12% is still suffered by the RAS side even if BSCS,0 is 2700 MHz in this specific realization. Secondly, we compare the performances due to the different numbers of subperiods in the subband allocation method and the beam switch method with the corresponding average

RFI among the telescopes in a fixed period time. As we consider the subband allocation in Fig. 2.2 without application of the reorganization of spectrum resource, the subband bandwidth is the same in different bands and therefore the fairness issue is out of concern in the subband allocation design in this case. We pick a realization in scenario 1 and plot the RFI with different values of Nsp in Fig. 2.20. As we can see from the figure, with a larger Nsp, the two methods can reduce more RFI at the telescopes. Nevertheless, the

2 diﬀerence in average RFI between Nsp = 2 and Nsp = 20 is approximately 0.22 dBW/m with the subband allocation method and 0.5 dBW/m2 with the beam switch method, which is

49 -125 100

90 -130 80

-135 70 ) 2

60 -140 50 -145 40 RFI EPFD (dBW/m Sample loss rate (%) -150 30

20 -155 10

-160 0 2700 2800 2900 3000 3100 3200 3300 3400 3500 B (MHz) SCS,0

Figure 2.19. Average RFI and sample loss rate of a realization in scenario 1 among the telescopes using the spectrum resource reorganization method with diﬀerent BSCS,0 requirements relatively limited, indicating robustness of these approaches with diﬀerent parameter settings.

Note that for a larger Nsp, the users may switch from one subband to another or re-focus the beam direction frequently, which may need extra expense on synchronization. Thus, in the rest of the chapter, we consider Np = 4 and Nsp = 10 to ensure a good RFI reduction performance and to simplify the process for subband allocation in a period and maintaining fairness between diﬀerent beam groups (as mentioned in Section 2.4-C). The corresponding time duration of the subperiod Tsp is 36 min. Thirdly, we investigate the eﬀects of the number of extra data streams and the number of beam switch operations conducted in each subperiod for the beam switch method. For this, we terminate the algorithm 1 when the number of beam switches reaches a certain threshold Nswitch and thereby limit the total number of beam switches. Fig. 2.21 shows the

50 )

2 -126.6 Subband allocation method -126.65

-126.7

-126.75

-126.8 RFI EPFD (dBW/m 2 4 6 8 10 12 14 16 18 20 N sp

) -125.8 2 Beam switch method -126

-126.2

RFI EPFD (dBW/m -126.4 2 4 6 8 10 12 14 16 18 20 N sp

Figure 2.20. Average RFI among the telescopes of a realization in scenario 1 using the subband allocation method or the beam switch method with different Nsp values average RFI received by the telescopes in a realization in scenario 1 using the beam switch method with different Nswitch and Nbeam,ex values. As we can see from the figure, larger

Nswitch and Nbeam,ex can lead to lower average RFI at the telescopes. In addition, Nswitch has stronger impact on RFI than Nbeam,ex as the RFI difference between different Nswitch values spans several dBs while the one for different Nbeam,ex values only varies within 1 dB.

Thus, in the rest of the chapter, we consider Nbeam,ex = 4 while not limiting Nswitch in the performance evaluation of the RFI and sample loss rate in diﬀerent scenarios with the beam switch method.

2.5.2 RFI and Sample Loss Rate in Diﬀerent Scenarios

As we can see from Section 2.4, the subband allocation method and the beam switch method are dependent on the equivalent gain factors, which are related to the speciﬁc RAO task

51 -119 N = 1 beam,ex N = 2 beam,ex -120 N = 3 beam,ex N = 4 beam,ex N = 10 -121 beam,ex ) 2 -125.3 -122 -125.4

-123 -125.5 RFI EPFD (dBW/m -124 -125.6 Unlimited

-125

-126 1 10 50 100 500 Unlimited N switch

Figure 2.21. Average RFI among the telescopes of a realization in scenario 1 using the beam switch method with diﬀerent Nbeam,ex and Nswitch values

of the telescopes (e.g., diﬀerent RAO task types mentioned in Section 2.2-B). To obtain a comprehensive performance of the RFI and sample loss rate at the telescopes with the proposed methods, we evaluate the performance of the proposed RFI reduction methods with 100 random realizations of RAO tasks for the telescopes with the detailed settings mentioned in Section 2.2-F. To highlight the eﬀectiveness of the proposed methods, we also demonstrate the performance of jointly using the existing emission suppression and beam muting methods [40] (we name them as “reference methods”). As mentioned before, we let

Np = 4 and Nsp = 10 for the subband allocation method and the beam switch method. In

2 2 addition, we consider EPFDex,S,i = −175 dBW/m and β = −175 dBW/m in the simulation to address the RFI from other active wireless systems.

For scenario 1, we consider BSCS,0 = 2800, 2900, and 3000 MHz to demonstrate the proposed methods with different RFI reduction performance. In addition, the performance of the reference methods with guardband bandwidth 300 MHz and different beam muting thresholds are shown for comparison. Fig. 2.22 shows the average RFI at different telescopes with different RFI reduction methods in scenario 1 among 100 random realizations. As we

52 can see from the ﬁgure, the proposed methods can lower the average RFI by about 50 dB.

The average RFI of different telescopes range from -170 dBW/m2 to -160 dBW/m2, which is around the detrimental RFI threshold for 24-hour observation in the RAS band I (-162.3 dBW/m2). Although similar RFI reduction can also be achieved by the reference methods, the average throughput ratios of the two are different. Fig. 2.23 demonstrates the average sample loss rate versus average throughput ratio for the different RFI reduction methods in scenario 1. As we can see from the figure, both the proposed methods and the reference methods can lower the average sample loss rate to less than 2%, which complies with the

ITU-R requirement in [49]. However, the proposed methods can achieve approximately 15% more throughput than the reference ones with similar sample loss rate. Furthermore, recall that the average throughput ratio represents the SCS throughput advantage of the proposed cooperative spectrum sharing paradigm over the existing SCS without spectrum sharing with RAS. From the ﬁgure, with a proper setting (e.g., BSCS,0 = 3000 MHz), our proposed paradigm can achieve 4.5% more average throughput than the original SCS in scenario 1, which is an attractive beneﬁt to the SCS side.

Scenarios 2 and 3 share the similarity in that the RAO is conducted in one of the protected

RAS bands and the remaining RAS band is exploited by the satellite system to improve the downlink throughput and reduce the RFI to the telescopes. In addition, as the downlink subbands are relatively far from the RAO band, the unwanted emission powers in the RAO band caused by the SCS downlink subbands, as shown in Fig. 2.5, have already reached the lower bound with the aforementioned emission mask and they have no diﬀerence between each other. Since the subbands share the same unwanted emission power in the considered

RAS bands (which can be observed from Fig. 2.16), the subband allocation method will not bring any further RFI reduction and therefore is only used for maintaining the fairness of bandwidth among the spot beam groups. Similarly, the guard band insertion in the emission suppression method is not necessary due to the huge frequency separation between the

53 -100

-110

-120 ) 2

-130 No RFI reduction Proposed methods with B =3000 MHz SCS,0 -140 Reference methods with muting threshold -173 dBW/m2 RFI EPFD (dBW/m -150

-160

-170 0 10 20 30 40 50 Telescope index

Figure 2.22. Average RFI at different telescopes with different RFI reduction methods in scenario 1 downlink bands and the RAO band. Fig. 2.24 shows the average RFI at different telescopes in scenarios 2 and 3. As we can see from the figure, the RFI of the case without any RFI reduction method are already relatively low for these scenarios. However, our proposed RFI reduction methods can still improve the quality of the RAO in these two scenarios with beam switching. From the figure we can see that the average RFI at different telescopes of the proposed paradigm in the two scenarios range from -170 dBW/m2 to -160 dBW/m2, which is below the detrimental RFI threshold of 24-hour observation in the RAS band II (-154.3 dBW/m2) and band III (-154.7 dBW/m2). Fig. 2.25 demonstrates the corresponding average sample loss rate versus the average throughput ratio in scenarios 2 and 3. The figure indicates that the proposed methods can offer around 13% and 9% more throughput than the reference ones in scenarios 2 and 3 respectively, given similar RAO sample loss rate. In addition, as discussed in the footnote

54 1.6

Proposed methods -172 dBW/m2 Reference methods 1.4

1.2 B = 3000 MHz SCS,0 2 1 -173 dBW/m

B = 2900 MHz SCS,0 0.8 B = 2800 MHz SCS,0 2

Sample loss rate (%) -174 dBW/m 0.6

-175 dBW/m2 0.4

-176 dBW/m2 0.2 85 90 95 100 105 Average throughput ratio (%)

Figure 2.23. Average sample loss rate versus average throughput ratio with diﬀerent RFI reduction methods in scenario 1

in Section 2.4-B, applying BSCS,0 = 3288 MHz may increase the RFI in scenario 2. However, as can be observed from Fig. 2.25, the RFI increase due to applying BSCS,0 = 3288 MHz has negligible impact on the sample loss rate (compared to the case with BSCS,0 = 3250 MHz, which achieves lowest unwanted emission power). This allows us to fully exploit the idle RAS bands without causing detrimental sample loss using our proposed methods in scenario 2. In general, the proposed paradigm can achieve 13.88% and 9.6% more average throughput than the original SCS in scenarios 2 and 3 respectively without violating the sample loss rate requirement. Scenarios 4 and 5 can be viewed as a special case of scenario 2 or 3 where some of the telescopes need to observe the RAS band I instead of band II or III. Fig. 2.26 and Fig. 2.28 show the average RFI at diﬀerent telescopes in scenarios 4 and 5 while Fig. 2.27 and Fig. 2.29 demonstrate the average sample loss rate versus the achievable throughput ratio in scenarios

55 Scenario 2 -130 No RFI reduction )

2 Proposed methods with B =3288 MHz SCS,0 -140 2 Reference methods with muting threshold -164 dBW/m

-150

-160 RFI EPFD (dBW/m -170 0 10 20 30 40 50 Telescope index Scenario 3 -130 No RFI reduction )

2 Proposed methods with B =3177 MHz SCS,0 -140 Reference methods with muting threshold -164 dBW/m2

-150

-160 RFI EPFD (dBW/m -170 0 10 20 30 40 50 Telescope index

Figure 2.24. Average RFI at diﬀerent telescopes with diﬀerent RFI reduction methods in scenarios 2 and 3

4 and 5. We pick Bsub,0 = 86 MHz speciﬁcally for BSCS,0 = 2600 MHz to avoid the situation where Bsub,0 limits the unwanted emission power reduction using the spectrum resource reorganization method. Comparing Fig. 2.23, Fig. 2.27, and Fig. 2.29, we can see that the telescopes in the proposed paradigm suﬀer higher sample loss rate of the observation in the

RAS band I in scenarios 4 and 5 than in scenario 1 given the same BSCS,0 (e.g., BSCS,0 =

2900 MHz ). The diﬀerences in the sample loss rates of the three scenarios are due to the relatively small amount of available spectrum resource (idle RAS band(s)) in scenarios 4 and

5 to accommodate the SCS downlink traﬃc and to reduce the unwanted emission power of

56 Scenario 2 0.35 Proposed methods -160 dBW/m2 Reference methods 0.3 -161 dBW/m2 0.25 -162 dBW/m2 B = 3288 MHz 0.2 SCS,0 -163 dBW/m2 0.15 2 B = 3250 MHz Sample loss rate (%) -164 dBW/m SCS,0 0.1 98 100 102 104 106 108 110 112 114 Average throughput ratio (%)

Scenario 3 0.6 2 Proposed methods -160 dBW/m Reference methods 0.5 -161 dBW/m2 0.4 -162 dBW/m2 B = 3177 MHz 0.3 -163 dBW/m2 SCS,0 -164 dBW/m2 0.2 Sample loss rate (%) 0.1 98 100 102 104 106 108 110 Average throughput ratio (%)

Figure 2.25. Average sample loss rate versus average throughput ratio with diﬀerent RFI reduction methods in scenarios 2 and 3 the downlink subbands. The results indicate that the proposed paradigm needs to suﬀer 5% throughput loss in scenario 4 and even more throughput loss (10%) in scenario 5 to lower the average sample loss rate below 2%. Nevertheless, Fig. 2.27 and Fig. 2.29 point out that our proposed methods can still achieve similar average sample loss rate with more SCS downlink throughput than the reference methods.

Although the sample loss rates in the RAS bands II and III can still be reduced with the proposed paradigm in scenario 4 and 5, a more eﬃcient choice for the telescopes is to work on the same RAS band together and switch from one to another in diﬀerent operation durations.

57 RAS band I -100 ) 2 -120

No RFI reduction -140 Proposed methods with B =2750 MHz SCS,0 Reference methods with muting threshold -172 dBW/m2 -160 RFI EPFD (dBW/m -180 0 10 20 30 40 50 Telescope index RAS band II -130 ) 2 -140

-150

-160 RFI EPFD (dBW/m -170 0 10 20 30 40 50 Telescope index

Figure 2.26. Average RFI at diﬀerent telescopes with diﬀerent RFI reduction methods in scenario 4

In this way, the whole system is operating in scenario 1, 2, or 3, and low sample loss rate and high system throughput can be achieved simultaneously. Nevertheless, scenarios 4 and

5 are not meaningless as in some cases the telescopes need to observe diﬀerent RAS bands and the proposed RFI reduction methods can enable the telescopes to obtain meaningful

RAO in the RAS band I at the cost of some SCS throughput loss.

Scenario 6 is an example of the existing RAO practice of the telescopes where they have primary use in the RAS bands II and III and can make observation at any time as they want. As mentioned in the analysis of scenarios 4 and 5, the unwanted emission powers

58 RAS band I 12 Proposed methods Reference methods 10 2 -160 dBW/m 2% line 8 -163 dBW/m2

6 -166 dBW/m2 B = 2900 MHz 4 -169 dBW/m2 SCS,0 2 B = 2750 MHz 2 -172 dBW/m B = 2600 MHz SCS,0

Sample loss rate (%) SCS,0 0 86 88 90 92 94 96 98 100 102 Average throughput ratio (%)

RAS band II 0.2 Proposed methods Reference methods -160 dBW/m2 0.15 -163 dBW/m2

0.1 2 -166 dBW/m B = 2900 MHz SCS,0 -169 dBW/m2 0.05 2 B = 2600 MHz -172 dBW/m SCS,0 B = 2750 MHz SCS,0 Sample loss rate (%) 0 86 88 90 92 94 96 98 100 102 Average throughput ratio (%)

Figure 2.27. Average sample loss rate versus average throughput ratio with diﬀerent RFI reduction methods in scenario 4 caused by the subbands with the original allocation plan in the exiting paradigm are the same and already reach the lower bound. Therefore, the subband allocation method is not able to further reduce the RFI. In addition, as we forbid using RAO bands for downlink transmission (which would cause unavoidable strong RFI in the overlapping RAO band(s)), there will be no idle band from RAS side to apply the spectrum resource reorganization method. In this case, the beam switch method brings major improvement on the RAO quality. Fig. 2.30 shows the average RFI at diﬀerent telescopes in scenario 6 and Fig. 2.31 demonstrates the average sample loss rate versus the average throughput ratio in scenario 6.

59 RAS band I -100 ) 2 -120

-140

-160 RFI EPFD (dBW/m -180 0 10 20 30 40 50 Telescope index

RAS band III -130 No RFI reduction )

2 Proposed methods with B =2600 MHz SCS,0 -140 Reference methods with muting threshold -172 dBW/m2

-150

-160 RFI EPFD (dBW/m -170 0 10 20 30 40 50 Telescope index

Figure 2.28. Average RFI at diﬀerent telescopes with diﬀerent RFI reduction methods in scenario 5

From Fig. 2.30 we can see, as we explained before, the average RFIs in the two RAO bands are relatively low even without applying any RFI reduction methods. Nevertheless, the proposed method can still lower the RFI. In addition, Fig. 2.31 indicates that the proposed methods maintain same service capacity/throughput while oﬀering RFI and sample loss rate reduction. On the other hand, the reference methods inevitably bring some service loss to

SCS side when improving the RAO quality of the telescopes.

Scenario 7 is a tricky scenario where neither RAS nor SCS can obtain beneﬁts. To reduce the RFI and the sample loss rate for telescopes which make RAO in the RAS band I, the

60 RAS band I 30 Proposed methods 25 Reference methods 2% line 20 -160 dBW/m2 15 -163 dBW/m2 B = 2750 MHz -166 dBW/m2 SCS,0 10 B = 2900 MHz -169 dBW/m2 SCS,0 -172 dBW/m2 5 Sample loss rate (%) B = 2600 MHz 0 SCS,0 86 88 90 92 94 96 98 100 102 Average throughput ratio (%) RAS band III 0.6 Proposed methods Reference methods

0.4 -160 dBW/m2 -163 dBW/m2 B = 2750 MHz -166 dBW/m2 SCS,0 0.2 B = 2900 MHz -169 dBW/m2 SCS,0 -172 dBW/m2

Sample loss rate (%) B = 2600 MHz SCS,0 0 86 88 90 92 94 96 98 100 102 Average throughput ratio (%)

Figure 2.29. Average sample loss rate versus average throughput ratio with diﬀerent RFI reduction methods in scenario 5

SCS needs to spare some bandwidth to form a guardband between the RAS band I and its downlink band. However, the two other RAS bands cannot be used for SCS’s downlink transmission as some of the telescopes are conducting RAO in these bands. Under this dilemma, either the SCS needs to sacriﬁce some of its downlink bandwidth to meet the RFI and sample loss rate requirement of the RAS side or the RAS has to bear severe sample loss in the RAS band I. In consequence, the SCS and RAS are not encouraged to work in this scenario to avoid potential loss to both sides.

61 RAS band II -130 No RFI reduction )

2 Proposed methods -140 Reference methods with muting threshold -164 dBW/m2

-150

-160 RFI EPFD (dBW/m

-170 0 10 20 30 40 50 Telescope index

RAS band III -130 ) 2 -140

-150

-160 RFI EPFD (dBW/m

-170 0 10 20 30 40 50 Telescope index

Figure 2.30. Average RFI at diﬀerent telescopes with/without the beam switch method in scenario 6

2.6 Conclusions

In this chapter, we investigated the eﬀects of the GSO satellites on the RFI and sample loss rate of the RAO using three bands of interests for RAS as an example. The analyses showed that the RFI of the GSO satellites in the unprotected RAS band (e.g. 18.28 – 18.36

GHz) could be tens of dB above the detrimental RFI threshold of the RAO and hence could disrupt the RAO of the ground telescopes. To overcome such spectrum access conﬂict, we proposed a spectrum sharing paradigm for GSO SCS and RAS, together with three RFI reduction methods. These methods were formulated as i) reorganization of the spectrum

62 RAS band II Proposed methods 0.3 Reference methods -160 dBW/m2 0.25 -161 dBW/m2 0.2 -164 dBW/m2 -162 dBW/m2 0.15 2 Sample loss rate (%) -163 dBW/m 0.1 99.8 99.82 99.84 99.86 99.88 99.9 99.92 99.94 99.96 99.98 100 Average throughput ratio (%)

10-3 RAS band III 8 Proposed methods 7 Reference methods

6 2 5 -160 dBW/m 2 -164 dBW/m 2 4 -161 dBW/m 2 3 -162 dBW/m Sample loss rate (%) -163 dBW/m2 2 99.8 99.82 99.84 99.86 99.88 99.9 99.92 99.94 99.96 99.98 100 Average throughput ratio (%)

Figure 2.31. Average sample loss rate versus average throughput ratio in scenario 6 resource to minimize the unwanted emission of the GSO satellites in the RAO bands, ii) rearranging the subband allocation of the GSO satellite’s beams based on their unwanted emission powers and bandwidths, and iii) cell-based beam switch approach to suppress the RFI of the GSO satellites. We investigated the effects of the parameter settings in the three RFI reduction methods which showed performance robustness of different parameter settings as well as some tradeoff between RAO quality and SCS downlink capacity. Next, our performance evaluation results conveyed that applying the three proposed methods jointly yielded larger SCS throughput performance gain over the existing RFI reduction methods. Furthermore, the results illustrated the benefits of the proposed paradigm over the existing

63 paradigm in terms of its capabilities in enabling the RAO in the unprotected RAS band (18.28 – 18.36 GHz), improving the RAO quality in the protected RAS bands (22.21 – 22.5 and 23.6 – 24 GHz), and increasing the SCS downlink capacity/throughput. Overall, the proposed approaches enable a friendly coexistence between GSO SCS and RAS with mutual beneﬁts which are unavailable in the existing approaches.

64 CHAPTER 3

IMPACTS OF LARGE-SCALE NGSO SATELLITES: RFI AND A NEW

PARADIGM FOR SATELLITE COMMUNICATIONS AND RADIO

ASTRONOMY SYSTEMS 1

3.1 Introduction

Non-geostationary orbit (NGSO) satellite communication systems (SCSs), namely low earth orbit (LEO) and medium earth orbit (MEO) systems, have been investigated for decades. However, the unsuccessful commercial applications of the former NGSO systems launched decades ago have reduced further eﬀort to promote such systems for years [55]. Recently, due to the increasing demand for ubiquitous high-speed and low-latency Internet connections as well as the rapid development of low-cost commercial spacecraft launching [64, 78], the space industry is planning to launch thousands of NGSO satellites. For instance, companies such as OneWeb and SpaceX are proposing to launch thousands of LEO and MEO satellites [31, 27]. These future NGSO satellites will form a tremendous space backhaul network via inter-satellite links (ISLs) [77] as well as a ubiquitous global wireless access network. Radio astronomy provides a description of the universe and enables testing of laws of fundamental physics, e.g., General Theory of Relativity [73]. It is expanding from a phenomenological science to astro-physics and astro-chemistry for which the observations are intrinsically sensitivity-limited and interference-free environments are needed. Similarly, advances in radio astronomy require more and more radio astronomical observations (RAO) outside the frequency bands allocated to radio astronomy system (RAS) [21]. However, the prospects of large-scale NGSO SCSs cast a distressing RFI situation to RAS. The ground-based RAS uses highly sensitive receivers to observe very weak signals from cos-

1© 2019 IEEE. Reprinted, with permission, from Yucheng Dai, Dong Han, and Hlaing Minn, ”Impacts of Large-Scale NGSO Satellites: RFI and A New Paradigm for Satellite Communications and Radio Astronomy Systems,” in IEEE Transactions on Communications, vol. 67, pp. 7840 - 7855, July 2019.

65 mic sources within a wide frequency range. Out-of-band spectrum sidelobes from satellite transmitters, which are negligible to other communication systems, could substantially disrupt RAO. Furthermore, due to inherent nonlinearity of some transmitter components as well as device imperfection, unintended/unexpected RFI from satellites to RAS can occur.

Although some eﬀorts have been made to mitigate the RFI from active wireless services

(including satellite communication) to the ground RAS, e.g., setting up Radio Quiet Zones

(RQZs) [82, 83], blanking and excision [86, 10, 15, 63, 90, 74, 26], beamforming and spatial

ﬁltering approach [84, 17, 72, 9, 75, 11, 52, 59, 53], auxiliary antenna based RFI removal

[50, 8, 16, 1] and time-division sharing [57, 71, 70, 30, 29, 4], unfortunately, their applicabil- ity to the large-scale NGSO systems is very limited. As large-scale NGSO systems plan to cover most of the earth surface ubiquitously, radio observatories on earth cannot hide from

NGSO satellites’ potential RFI. As an example, we can recall the Iridium satellite system with 66 LEO satellites launched in 1998. Even though several attempts were made to avoid

RFI to RAS, in practice RAO data were corrupted by Iridium’s RFI as conﬁrmed in the new measurements conducted in 2010 [80].

In facing potential strong RFI from the large-scale NGSO SCSs, space-based radio telescopes are attractive solutions as they may have higher orbit than the NGSO SCSs and therefore receive less RFI than the ground telescopes. In addition, the space-based telescope like HALCA [39] or Spektr-R [51] can form a Very Long Baseline Interferometry (VLBI) with ground telescopes to increase RAO performance. However, due to the cost and other issues, the number of the space-based radio telescopes is very limited and the overall performance of the existing space based radio telescopes is not compatible with the ground telescopes.

Motivated by both the critical conﬂict between the next generation NGSO SCS and

RAS and the higher performance demands of RAS, we propose a new paradigm which over- comes the issues of the existing paradigm and oﬀers several additional advantages. The new paradigm changes NGSO SCS into an integrated NGSO satellite communication and radio

66 astronomy system (SCRAS) where satellites provide both RAO and communication services.

The direct beneﬁts are that RAS gains more RAO opportunities and performance enhance- ments (in terms of sensitivity through combining as in [30] and [29] and resolution through

VLBI) and SCS obtains higher throughput and new services or business opportunities. The proposed approach oﬀers a new infrastructure and paradigm at the side of data acquisition from radio astronomical objects. It is in synergy with the recent development of virtual astronomy observatory (VAO) [85] which is at the data processing side, oﬀering a large scale electronic integration of radio astronomy data and tools for radio astronomers.

This chapter’s major contributions are summarized below.

We analyze the RFI at ground radio telescopes caused by a large-scale NGSO SCS

and investigate the required guardband bandwidth to keep RFI below the acceptable

continuum observation threshold based on the emission mask requirement of National

Telecommunications and Information Administration (NTIA). Then, we evaluate time

and location dependent RFI caused by the OneWeb LEO system. Next, we assess

the maximum baseline distance for VLBI observation and the number of telescopes

that can observe the same target below the RFI threshold where both metrics are

time-varying.

We evaluate performance of RFI mitigation approaches such as guardband insertion,

transmission muting, and sample excision, in the presence of large-scale OneWeb LEO

SCS. Their RFI suppression performances, limitations, and costs in terms of SCS

service degradation and RAO sample loss are assessed.

We introduce a new paradigm for NGSO SCS and RAS by means of an integrated

NGSO SCS and RAS, which not only eliminates the RFI from the devices operating

below the NGSO but also oﬀers additional advantages for both NGSO SCS and RAS.

We investigate RAO performance of the proposed paradigm in terms of the observable

bands without RFI concern, the average number of telescopes that can simultaneously

67 observe a target, the maximum baseline distance for various target directions, and the

observation sensitivity.

As the bands originally allocated to RAS can be released to the integrated SCS and

RAS system, we also address spectrum access and resource allocation in these bands

and conduct corresponding data rate analysis for both SCS data and RAO data.

Since the proposed paradigm conducts RAO in space, we develop a design of RAO

data transport from satellites to ground stations, and evaluate its performance.

The chapter is organized as follows. Section 3.2 introduces the LEO satellite system and ground telescopes model, and analyzes the RFI level at ground telescopes caused by the LEO satellites’ downlink. Section 3.3 proposes three alternative RFI reduction methods and points out that these methods cause service degradation to LEO SCS or data loss to RAS. Section 3.4 presents a new paradigm for LEO SCS and RAS and discusses its observability improvement for RAS. Section 3.5 analyzes the data rate improvement for

SCS in the proposed paradigm. Section 3.6 addresses RAO data transport issue. Finally,

Section 3.7 concludes this chapter. Key notations used in the chapter are shown in Table 3.1.

3.2 RFI Analysis for Ground Radio Telescopes under a Large-Scale LEO SCS

3.2.1 Interference Calculation

As satellite communication is one of the major sources of RFI, the International Telecom- munication Union Radiocommunication Sector (ITU-R) has already provided several rec- ommendations about this issue. The ITU-R document [45] oﬀers a method to determine whether RFI is detrimental or not and some bands that should be protected from RFI. The

ITU-R document [47] provides a method to calculate the RFI between NGSO satellites and radio telescopes based on the average Equivalent Power Flux-Density (EPFD). The instantaneous EPFD between telescope i and satellite j’s kth beam at time t can be calculated

68 Table 3.1. Notations Used in Chapter 3 Notation Description Instantaneous RFI EPFD from satellite j’s kth beam to ground EPFDi,j,k(t) telescope i at time t Average RFI EPFD received by telescope i during RAO integration EPFDi(T ) time T Transmitting antenna gain of satellite j’s kth beam to ground GT,j,k(t) telescope i at time t in the RAO band Receiving antenna gain of telescope i to satellite j at time t in the GR,i,j(t) RAO band Unwanted emission power of satellite j’s kth beam in the RAO PUE,j,k band

di,j(t) Distance between satellite j and telescope i at time t

PSDUE(f) Power spectrum density of the unwanted emission at frequency f

pLEO,beam Transmitting power of a spot beam of a LEO satellite Propagation delay induced fractional phase diﬀerence between ∆φj,j (t) ref telescope j and jref at time t Propagation delay induced integer sample index diﬀerence τj,j (t) ref between telescope j and jref

∆fj(t) Doppler shift of the RAO signal received by satellite j at time t

Pout,n Outage probability of link n

Pout,req,n Required/target outage probability of link n

λn Mean of instantaneous traﬃc of link n

Maximum value that λn can take without violating the outage Λn,max requirement of link n ˆ Minimum capacity in the SG link that needs to be assigned to CSG,i(Pout,req,SG) SCS to meet the outage probability requirement Pout,req,SG with the following formula:

PUE,j,kGT,j,k(t)GR,i,j(t) EPFDi,j,k(t) = 2 (3.1) 4πdi,j(t) where PUE,j,k is the unwanted emission power of satellite j’s kth beam in the RAO band,

GT,j,k(t) is the transmitting antenna gain of the NGSO satellite j’s kth beam towards the direction of telescope i at time t in the RAO band, GR,i,j(t) is the receiving antenna gain of telescope i towards the direction of satellite j at time t in the RAO band, and di,j(t) is the distance between telescope i and satellite j at time t. Since GT,j,k(t) and GR,i,j(t) are determined

69 by the relative positions of the satellite and the telescope, we have GT,j,k(t) = GT,j(θT,i,j,k(t))

and GR,i,j(t) = GR,i(θR,i,j(t)) where θT,i,j,k(t) is the angle between the boresight of the transmitting beam k and the direction from satellite j to telescope i at time t and θR,i,j(t) is the angle between the RAO direction and the direction from telescope i to satellite j at time t.

Fig. 3.1 demonstrates a scenario of satellite and telescope we consider in the RFI calculation with θT and θR. Then, for a certain RAO task conducted by telescope i, the average RFI at telescope i 2 during the integration time Tint can be represented as

N 1 Z t0+Tint X Xbeam EPFDi(Tint) = EPFDi,j,k(t) dt (3.2) Tint t0 j∈INGSO(t) k=1

where t0 is the beginning time of the RAO, INGSO(t) is the index set of NGSO satellites that

can be viewed from telescope i at time t and Nbeam is the number of beams that each NGSO satellite uses for its downlink transmission. Due to the shape of the earth, not all LEO

satellites are visible to a certain telescope. It is commonly assumed that only the visible

satellites would cause RFI to RAO. Besides, in practice the integration time Tint can be 15 min, 1 hr, 2 hrs, 5 hrs, 10 hrs or other duration depending on the visibility of the RAO target

and the required level of signal to noise ratio. Thus, we need to adjust the detrimental RFI

threshold with respect to the integration time of each RAO task.

3.2.2 Large-Scale LEO SCS Model: OneWeb

Although many companies propose their individual plans to build sky networks via a large

number of LEO and MEO satellites, only a few of them (including OneWeb) have so far

obtained the permission from Federal Communication Commission (FCC). In this chapter,

we use the constellation of OneWeb as our reference LEO satellites model. As mentioned in

2The accumulated RFI at the telescope is a more appropriate metric than the RFI generated by a satellite as it determines RAO performance.

70 RAO target NGSO satellite

θT NGSO beam Telescope direction θR pointing direction

NGSO system user Ground telescope

Figure 3.1. An illustrative scenario for angles θT and θR

[65], there will be in total 720 LEO satellites running on circular orbits at 1200 km altitude. The satellites operate on 18 different orbital planes with 10 degree longitude spacing between two planes and each orbital plane has 40 LEO satellites. Fig. 3.2 shows a snap-shot of the OneWeb LEO satellite constellation. The red + symbol represents a LEO satellite and the green line connection between satellites indicates the path of the orbital plane. Each LEO satellite has 16 identical spot beams with fixed directions for communications with users. According to the description in [65], the spot beams should be highly elliptical to provide enough geographic coverage. However, as no detailed information is revealed in [65], we consider using a classical parabolic antenna model from [41] to simulate the downlink transmission of OneWeb LEO satellites. According to [65], the OneWeb user terminals will be equipped with mechanically steered parabolic reflectors and/or low-cost phased array designs with ability to track the on-the-move LEO satellites. The satellites will allow the users to switch from one spot beam to another, providing seamless network connection in continuous movement. Similar idea can be found in [88, 91]. In addition, as there are much

71 Figure 3.2. OneWeb LEO satellite constellation (+ denotes a satellite)

Table 3.2. LEO satellite settings Parameter Value Orbital period 6565 sec Bandwidth per beam 250 MHz Tx power per beam pLEO,beam 7 Watt Beamwidth (at 10.65 GHz) 10.1° Boresight gain GT (at 10.65 GHz) 24.4 dBi Downlink band 10.7 – 12.7 GHz Total downlink bandwidth per satellite 2 GHz Frequency reuse factor 8 fewer users on the ocean than users on land and the radio telescopes are located on land, we assume that the RFI eﬀect of the beams pointing on the ocean is negligible. Table 3.2 shows other settings of the LEO satellites we consider in the chapter, including the band assignment.

3.2.3 Ground Telescopes Model

In addition to the LEO satellites model, the ground radio astronomy telescopes model is another key factor in the performance evaluation. In this chapter, we consider 58 existing

72 Figure 3.3. Existing ground radio astronomy telescopes’ locations observatories around the world as our reference ground radio astronomy telescopes model and assume all of them have the capability to observe the bands discussed in the chapter. The red dots in Fig. 3.3 show the locations of these radio telescopes. We note that the distribution of ground radio astronomy telescopes is not even as more telescopes are located at the northern hemisphere (mostly in north America and west Europe). The unbalanced distribution of radio telescopes may cause some limitation for certain target directions. For simplicity, we consider the telescopes can observe the target with 10° minimum elevation angle to ensure that no detrimental ground interference leak into the telescopes. In addition, we assume that the ground telescopes have capability to observe both in daytime and nighttime as they can have large refrigeration and calibration systems to eliminate the eﬀect of the solar illumination. Furthermore, as suggested by the ITU-R in [47], we consider the antenna model in [42] as the antenna model of the ground telescopes.

3.2.4 Guardband and Emission Mask Based RFI Analysis

From Eq. (3.1), we can see that the instantaneous RFI EPFD level is related to the relative positions of the LEO satellites and the ground telescopes. To get more insight, let us consider a simpliﬁed model where one LEO satellite is at the zenith direction of a ground radio telescope and it has only one spot beam for downlink transmission. Suppose the radio

73 telescope, as mentioned in [45], conducts a continuum observation in the 100 MHz bandwidth centered at 10.65 GHz and the LEO satellite uses a 250 MHz bandwidth of downlink near the RAO band. We first assume that the LEO satellite obeys the current unwanted emission requirements defined by FCC and NTIA [61, 22] and we will find the required guardband bandwidth between the RAO band and the satellite downlink band which satisfies the RFI threshold in Table 1 in [45]. The emission mask defines the maximum allowable emission power of the transmitter at frequency f. In this chapter, we use an emission mask from NTIA [61], which limits the Power Spectrum Density (PSD) of the emission of inband signal at f based on the frequency offset foff = |fc − f|, where fc is the central frequency of the assigned band. Then, for any f out of the assigned band, the PSD of the unwanted emission

PSDUE(f) should satisfy

SEM(foﬀ ) 10 PSDUE(f) ≤ psdmax · 10 (3.3)

BA where foﬀ ≥ 2 , BA is the bandwidth of the assigned band,

2foﬀ SEM(foﬀ ) = max{−40 · log10( ) − 8, −60}, (3.4) BA

and PSDmax is the maximum PSD of the satellite signals in the assigned band measured in

a reference bandwidth [61]. Since PSDmax is related to the speciﬁc power distribution of the

pLEO,beam signals in the assigned band, without loss of generality, we consider PSDmax = in BA this chapter where pLEO,beam is the transmit power per a LEO satellite beam. In addition, we assume that the LEO SCS will generate the maximum allowable unwanted emission, and the unwanted emission power PUE in the RAO band is

fRAO,U Z SEM(foﬀ ) PUE(fRAO,L, fRAO,U) = PSDmax · 10 10 df (3.5) fRAO,L

where fRAO,L and fRAO,U are the lower and upper edges of the RAO band, respectively.

Eq. (3.5) also indicates that PUE depends on the frequency separation between the SCS downlink band and the RAO band.

74 Figure 3.4. The required guardband bandwidth versus (θT, θR)

From Eq. (3.1), we can see that for a given θT and θR pair, we can ﬁnd a corresponding

PUE that makes the RFI EPFD meet the RFI requirement in Table 1 in [45]. One way to

achieve this PUE is to insert a guardband between the RAO band and the downlink band of the LEO satellite. Fig. 3.4 shows the relationship between (θT, θR) and the required guardband bandwidth. From the ﬁgure we can see that the required guardband bandwidth ranges from 150 MHz to 2375 MHz. From Eq. (3.3) and Eq. (3.4), we know that the minimum

−6 value of PSD mask of the satellite downlink signals is PSDmax · 10 when foﬀ /BA ≥ 1000%.

Given BA = 250 MHz and pLEO,beam = 7 Watt, inserting a guardband with bandwidth of

−6 2375 MHz or larger yields a minimum PUE of 2.8 × 10 Watt. Therefore, 2375 MHz can be viewed as the maximum effective guardband bandwidth as no lower unwanted emission power can be achieved via adopting a larger guardband bandwidth due to the flat emission mask floor. Consequently, there are some θT and θR pairs (e.g., θT = θR = 0°) which make

GT · GR too large that even the minimum PUE cannot lower the RFI below the detrimental RFI threshold. However, since the LEO satellites are moving fast, their relative positions with reference to a ground telescope will change from time to time and thus the instantaneous

75 EPFD will not always be such high. On the other hand, from Eq. (3.1) we can see that the

lower bound of the required guardband bandwidth is related to the minimum value of GT and

GR when the distance d is ﬁxed. Fig. 3.4 indicates that the minimum required guardband bandwidth is 150 MHz for the considered d = 1200 km.

3.2.5 RFI Analysis Based on OneWeb LEO Constellation

In the previous section, we analyze the eﬀects of θT and θR angle pairs on RFI assuming the distance between the LEO satellite and the radio telescope is ﬁxed. However, since the

LEO satellites are moving fast in the space (e.g., the OneWeb satellites have an angular velocity of 3.03°/min), we evaluate the average of the instantaneous RFI EPFD under this practical scenario [47]. In this section, we consider a model that the ground radio telescopes are tracking a specific target in the far field, which can be viewed as fixed in the solar coordinate. Due to the blockage of the earth and the minimum elevation angle requirement, not all radio telescopes can observe the target at the same time. In addition, owing to the self-rotation of the earth, the ground telescopes may have their own certain time window to observe the target during a day, which is determined by their locations on the earth and the target direction. The RFI at the ground telescopes in the simulation comes from the downlink of the LEO satellites, which is, as mentioned in the previous sections, a band centered at 11.7 GHz with 2 GHz bandwidth and the ground telescopes are observing in the band 10.6 – 10.7 GHz.

Fig. 3.5 shows the instantaneous RFI EPFD at a ground radio telescope along the observation time with the target direction at latitude 0° and longitude 180° in the earth coordinate when the RAO starts. From the ﬁgure we can see that the instantaneous RFI, although varies from time to time, has a fundamental period of approximately 2.7 min, which is the time interval between two successive LEO satellites in the same orbit that would ﬂy across the main direction of the radio telescope. In addition, the envelope of the

76 ) -80 2

-100

-120

-140

RFI EPFD (dBW/m -160 0 1 2 3 4 5 6 7 8 9 10 Simulation time (hr)

) -135 2

-140

-145

RFI EPFD (dBW/m -150 0 20 40 60 80 100 120 Simulation time (min)

Figure 3.5. Instantaneous RFI EPFD at a ground telescope during 24 hours in the presence of LEO satellites

-110 10.8 Average Interference all Threshold Observing time 10.6 -120

10.4

) -130 2 10.2

-140 10

9.8

RFI epfd (dBW/m -150 Observing time (hr)

9.6

-160 9.4

-170 9.2 0 10 20 30 40 50 Ground telescope index

Figure 3.6. Average RFI EPFD levels at diﬀerent ground telescopes during 24 hours in the presence of LEO satellites

77 RFI would rise and fall as the RAO direction traverses the LEO orbital planes due to the earth’s self-rotation. Fig. 3.6 shows the average RFI levels of different ground radio telescopes with their corresponding RFI thresholds, which are determined by their respective observation time durations. It can be concluded from the figure that none of the ground telescopes are able to observe that certain target since the corresponding RFI are above the thresholds. In other words, the ground telescopes permanently lose the chance to observe this target in the presence of LEO satellites. For different ground telescopes, the average RFI EPFD ranges from -144 to -110 dBW/m2, which has about 35 dB difference. Multiple factors may contribute to this difference, among which the dominant one is that the spot beams on ocean use much less transmitting power and thus cause negligible RFI to the ground telescopes. Consequently, the ground telescopes near or surrounded by the sea receive less RFI than the ones located inland. In addition to the RFI at the telescopes when observing a certain target, we also numer- ically evaluate the RFI at certain telescopes with different azimuth and elevation angles of their own locations to show that the RFI from the LEO satellites affect almost all directions. Here we pick telescope 3 and 36 as our examples. Since telescope 3 is at North Liberty in Iowa and surrounded by land while telescope 36 is on the Big Island of Hawaii in the Pacific Ocean, the two are good representatives of the telescopes which face high level and low level of the RFI from the LEO satellites, respectively. Fig. 3.7 and Fig. 3.8 show the average RFI of the two telescopes during 24 hours. As we can see from the figures, the RFI peaks are usually located at directions with high elevation angles (e.g., > 60°). Generally speak- ing, telescope 3 receives stronger RFI than telescope 36 in most directions. Both telescopes have average RFI EPFD larger than −160 dBW/m2, which is the ITU-R recommended RFI threshold for the observed band we consider with the 2000 seconds (sec) observation time. The aforementioned analyses are based on continuum observation’s requirements. Let us consider another possible situation where the ground telescopes can form a network and

78 Figure 3.7. Average RFI level at diﬀerent azimuth and elevation angles of telescope 3 during 24 hours in the presence of LEO satellites

Figure 3.8. Average RFI level at diﬀerent azimuth and elevation angles of telescope 36 during 24 hours in the presence of LEO satellites

79 conduct VLBI observation. Since the VLBI observation has greater immunity to RFI, the threshold of VLBI observation is much looser than that for continuum observation. For the speciﬁc RAO band we consider in the previous sections, the threshold of VLBI observation

(−113 dBW/m2) is 47 dB higher than the threshold of continuum observation assuming 2000 sec observation time [45]. Besides the RFI, another key metric that aﬀects the quality of the VLBI observation is the maximum baseline distance, which is deﬁned as the maximum distance of any two radio telescopes that are observing a certain target at the same time.

To evaluate the performance of VLBI observation of the ground telescopes and the eﬀect of the RFI, we plot the maximum baseline and the number of telescopes versus the observation time for the cases with and without RFI from the LEO satellites in Fig. 3.9. Here we consider three diﬀerent cases, which are 1) the ground telescopes are completely RFI free in the RAO band (100 MHz centered at 10.65 GHz), 2) the ground telescopes have RFI from the LEO satellite downlink band (10.7 – 12.7 GHz) which is adjacent to the RAO band, and

3) the ground telescopes have RFI from the LEO satellite downlink band (10.6 – 12.7 GHz) which is in the RAO band. In this case, the downlink subband bandwidth of each spot beam is 262.5 MHz.

From Fig. 3.9, we can see that the maximum baseline distance of the ground telescope is not aﬀected by the RFI even when the LEO satellites are using the RAO band as downlink.

The number of telescopes that can observe the target is slightly affected by the RFI from the LEO satellites in the case 2 and 3, which are marked with green cross and purple circle respectively. But this degradation (0.011% and 0.178% sample loss in case 2 and 3) is insignificant in terms of the whole RAO process. The negligible degradation is owing to the higher detrimental RFI threshold for VLBI observation, which reflects immunity of VLBI observation against RFI. Another observation is that the distribution of the ground telescopes on earth surface is not even, and the number of ground telescopes and their maximum distance vary a lot during the RAO period. This variation may affect the performance of

80 104 1.2

1.1

1 Case 1 0.9 Case 2 Case 3 0.8 0 4 8 12 16 20 24

Maximum baseline distance (km) Simulation time (hr) 40 Case 1 Case 2 30 Case 3

Numer of ground telescopes 0 4 8 12 16 20 24 Simulation time (hr)

Figure 3.9. Ground telescopes VLBI observation performance with and without LEO satellites

VLBI observation as during some of the time the number of telescopes that can observe is quite low (e.g., < 15 telescopes) and the corresponding maximum baseline distance is relatively short (e.g., < 9000 km).

3.3 Guardband, Transmission Muting and Sample Excision Based Solutions

under Large-Scale LEO SCS

From the previous section, we can see that the LEO satellites downlink transmission in adjacent bands of RAO will cause strong RFI in continuum observation. One potential solution is to temporally shut down the spot beams that may cause high RFI EPFD (e.g., larger than

-180 dBW/m2, which is 20 dB below the threshold in Table 1 in [45]). Here we assume that the LEO satellite system knows a priori the RAO plan of the ground radio telescopes (which is typically scheduled with much time in advance) and based on the locations and the obser-

81 vation direction of radio telescopes along with the orbital tracks of the LEO satellites, the system operator can determine the potential detrimental spot beams in advance. Another option is that instead of using all the assigned bandwidth for downlink transmission, the LEO satellite system will spare some bandwidth to be the guardband in between the RAO band and the satellite downlink band to reduce the RFI experienced at the telescopes. In addition, we can also let the ground telescopes drop the samples with high RFI to reduce to average RFI EPFD levels. To compare the effects of the three methods, we consider the following 4 different cases: 1) No RFI reduction: No method is applied for RFI reduction. It is used as a reference. 2) Guardband approach: It inserts a 400 MHz additional guardband between the RAO band and the LEO SCS downlink band. Then, the subband of one beam is 200 MHz. 3) Transmission muting approach: It turns off the beams if they generate instantaneous RFI EPFD at any of the ground telescopes higher than the threshold -180 dBW/m2. 4) Sample excision approach: The ground telescopes drop the RAO samples with total instantaneous RFI EPFD above the threshold -150 dBW/m2. Fig. 3.10 shows the average RFI EPFD levels at different ground radio telescopes observing the same target as we use in the previous section for the four considered cases. From the figure we can see that, although the three aforementioned methods effectively reduce some RFI (approximately 18 dB – 25 dB for the guardband approach, 35 dB – 50 dB for the transmission muting approach and 10 dB – 15 dB for the sample excision method), there are still some of ground telescopes with average RFI EPFD levels higher than the threshold even in case 3. Meanwhile, the transmission muting approach causes temporary communication service outage for some satellite users at some time, the guardband insertion approach leads to approximately 20% capacity loss in downlink, and the sample excision approach causes severe sample loss to the ground telescopes. The percentage of the LEO satellites’ beams which are shut down by the transmission muting approach during observation and the instantaneous RAO sample loss rate of the

82 -110 Case 1 Case 2 -120 Case 3 Case 4 -130 Threshold

-140

-150

-160

RFI EPFD (dBW/m2) -170

-180

-190

-200 0 10 20 30 40 50 Ground telescope index

Figure 3.10. Average RFI EPFD levels at diﬀerent ground radio telescopes for the 4 considered cases during 24 hours sample excision approach are shown in Fig. 3.11. From the ﬁgure we can see that at least

10% of spot beams are turned oﬀ during 24 hours and the corresponding users which are covered by these beams experience temporary connection loss. On the other hand, the ground telescopes may lose most of the RAO samples when the sample excision approach is applied during 24 hours and the overall RAO sample loss rate is 94.9%. In brief, these approaches are insuﬃcient to handle the RFI issue of a large-scale NGSO SCS.

3.4 A New Paradigm for NGSO SCS and RAS

3.4.1 An Integrated NGSO SCS and RAS

Since the three aforementioned methods cause unpleasant and inevitable service loss of the

LEO SCS or sample loss of the RAO, a more eﬃcient approach is needed to avoid RFI at telescopes for RAS and maintain communication service quality for SCS. For this, we

83 30 Case 3 25

Beam off (%) 10

5 0 4 8 12 16 20 24 Simulation time (hr)

100 Case 4

80 Data loss (%)

70 0 4 8 12 16 20 24 Simulation time (hr)

Figure 3.11. Percentage of the beams that are turned oﬀ in case 3 and the instantaneous RAO sample loss rate in case 4 across time propose a new paradigm in the form of an integrated NGSO satellite communication and radio astronomy system.

In the proposed paradigm, the communication satellites will be equipped with additional antennas and receivers to make RAO in addition to their main communication services. The zone for active communication services is towards the earth from the satellites while the one for RAO is from the satellites outwards the earth. Hence, the antennas for communication and RAO can be mounted at opposite sides of the satellite to each other. The satellites can use the RAO spectrum also in their active communication services as the spatial zones for the two services are non-interfering. Similarly, RAO can be made in the bands allocated for active wireless services. In other words, the communication satellites in the proposed paradigm now take the role of radio telescopes on earth for RAO in exchange for their

84 spectrum uses of the RAS spectrum for active communication systems. Satellites need to make RAO at a mutually agreed data rate and forward their RAO data through their earth- station gateways to RAS.

This innovation will beneﬁt the NGSO SCS as follows:

The bands in which NGSO systems can make suﬃcient RAO can be reused for active

wireless services, thus oﬀering more spectrum access opportunities for SCS.

For the above bands, SCS will no longer need to implement RFI-avoiding mechanisms.

SCS systems can obtain new services/business opportunities for additional RAO be-

yond their obligation.

The proposed paradigm oﬀers RAS the following beneﬁts:

RAO from the satellites has signal strength gain due to the removal of atmospheric

attenuation and weather impact (e.g., the space-based telescopes are free from at-

mospheric absorption which is especially severe in infrared, ultraviolet, 23 GHz, and

60 GHz bands and therefore are suitable to conduct photon detection and contin-

uum/spectral line observation in these bands).

The bands allocated for active wireless services which typically do not yield meaningful

RAO at the ground telescopes (e.g., 10.7 – 12.7 GHz) can now be observed for RAS

measurements.

RFI from consumer electronic equipment and wireless systems, which are diﬃcult to

prevent from happening in practice, would not aﬀect the RAO of the satellites.

Due to large-scale NGSO systems, large-scale RAS telescope arrays infeasible with

ground telescope systems can be realized.

Large-scale NGSO satellites provide more RAO time than ground-based radio obser-

vatories.

The proposed large-scale NGSO RAS can be combined with the existing ground RAS

to yield a more capable RAS while avoiding conﬂicts with active wireless systems.

85 The following section will present more detailed RAO performance of the proposed

paradigm.

3.4.2 Observability of LEO versus Ground Telescopes

We assume that the LEO telescopes can observe within 60° from the zenith direction of the

LEO satellites to avoid the RFI from earth surface and inter-satellite links. Furthermore, as mentioned in [69], the space based telescopes cannot make RAO (under cost constraint) if the sun illuminates the dish surface. Thus, we assume that the LEO telescope can observe when the sun is at least 90 degree from the zenith direction of the satellite. With this requirement, nearly half of the LEO telescopes cannot make RAO at each time instant due to the sun illumination. In addition, though we focus on the RAS bands near the satellites downlink in the previous sections, the LEO telescopes can observe not only in these bands but also in any other bands if they are equipped with corresponding receivers and if there are no RFI from the higher altitude SCSs. Speciﬁcally, since the LEO telescopes are above the atmosphere, they are very suitable for RAO in the bands with high atmospheric absorption (e.g., around

22 or 63 GHz) or with higher weather impact (e.g., > 11 GHz ) where the ground telescopes fail.

In Table 3.3, we summarize five different types of bands and corresponding observability of ground and LEO telescopes with continuum and VLBI observation. The check-mark means the effect of RFI is negligible compared to the detrimental RFI threshold. We can see that except the bands used by SCSs with higher altitude than the proposed SCRAS, our proposed paradigm encounters less RFI than the ground telescopes and therefore gains more observability.

To evaluate the observability of the LEO telescopes versus the ground telescopes in VLBI observation, we focus on two key performance metrics which are the number of telescopes that can observe the same target simultaneously and the maximum baseline distance between

86 Table 3.3. Observability of the LEO and ground telescopes

Ground telescope Band type/ Ground telescope LEO telescope LEO telescope continuum description VLBI observation continuum observation VLBI observation observation Bands within or adjacent Detrimental RFI RFI with limited to the downlink of blocks RAO in almost RAO data loss XX the integrated SCRAS all directions Bands within or adjacent to the Detrimental RFI downlink of other large scale RFI with limited blocks RAO in almost SCSs with lower altitude than RAO data loss2 XX all directions2 the integrated SCRAS 87 Bands within or adjacent to the Without RQZ, RFI Without RQZ, RFI ground wireless communication, can cause potential can cause potential RADAR system or other active XX RAO data loss RAO data loss wireless systems Bands within or adjacent to the Detrimental RFI Detrimental RFI downlink of other large scale RFI with limited RFI with limited blocks RAO in almost blocks RAO in almost SCSs with higher altitude than RAO data loss2 RAO data loss2 all directions2 all directions2 the integrated SCRAS Opacity of the Opacity of the Bands with high atmospheric atmosphere atmosphere absorption or high weather impact XX blocks RAO blocks RAO 2 Assuming no RFI reduction methods are applied. those telescopes. To show the observability of the ground and LEO telescopes at different target directions, we first choose a reference direction in the earth coordinate, which is the opposite direction of the sun. As the time in simulation is relatively short with respect to the orbital period of the earth, we can assume the reference direction is fixed in the coordinate of the sun and represent other directions with relative latitude and longitude. For simplicity, we assume the date is equinox and the daytime and nighttime are of approximately equal duration all over the planet for all simulations except one example at winter solstice, which aims to show the performance variation of the LEO telescopes. In this section, we compare

5 potential VLBI observation cases, which are:

1. The LEO telescopes conduct VLBI observation at equinox.

2. The LEO telescopes conduct VLBI observation at winter solstice. Here we assume the

same reference direction as in the previous case for comparison purpose.

3. The ground telescopes form a huge VLBI network and conduct VLBI observation at

equinox. Its performance can be viewed as an upper bound of the ground telescopes.

4. The ground telescopes in Very Long Baseline Array (VLBA) conduct VLBI observation

at equinox. The VLBA is a VLBI network with telescopes located in USA.

5. The ground telescopes in European VLBI Network (EVN) conduct VLBI observation

at equinox. The EVN is a VLBI network with telescopes located in Europe and Asia.

Fig. 3.12 compares the average numbers of ground and LEO radio telescopes that can observe the same target simultaneously at several directions in cases 1, 2, 3, 4, and 5. The

ﬁgure indicates the following.

The plot of the number of the LEO telescopes forms a saddle-shaped distribution

and the minimum number of LEO telescopes appears at the directions with relative

longitudes ±180°, (e.g., the direction of the sun) where the LEO telescopes cannot

observe. The relative latitudes of the directions with the minimum number of the LEO

88 Figure 3.12. Average number of telescopes that can simultaneously observe a target versus target directions (from left to right are case 1 to case 5)

Figure 3.13. Maximum baseline distance for diﬀerent target directions (from left to right are case 1 to case 5)

telescopes are related to the subsolar point and therefore vary with different times of a year. Comparing the first two subfigures, we can see that when it is winter solstice, the astronomical polar night at the north polar region helps the LEO telescopes gain more observability in the north polar directions while at the same time the midnight sun at the south polar region decreases the number of the LEO telescopes that can conduct RAO. However, there are still at least 40 LEO telescopes that can observe the south polar directions simultaneously. Comparing subfigures 1, 3, 4, and 5, we can see that in most of the directions, there are more LEO telescopes than the ground telescopes that can observe. In addition, as most

89 of the ground telescopes are located at the northern hemisphere, their observability

is more in the north (positive relative latitude) than in the south (negative relative

latitude).

Fig. 3.13 compares the maximum baseline distance of diﬀerent observation directions

averaged across time for the 5 cases. From the ﬁgure, we can observe the following.

For the LEO and ground telescopes, the larger number of telescopes that can conduct

observation simultaneously leads to the larger maximum baseline distance in the same

direction. Nevertheless, comparing the maximum baseline distance of case 1 and 2

and the corresponding numbers in Fig. 3.12, we can see that the number of telescopes

that can observe in the south polar direction (−90° relative latitude) in case 1 is

approximately 2 times of that in case 2, while the maximum baseline distance of the

same direction in case 1 is only 10% larger than the one in case 2, which means the

relationship between the number of satellites and the maximum baseline distance is

non-linear.

The ﬁrst three subﬁgures indicate that the proposed LEO telescopes can achieve similar

maximum baseline distance as the upper bound of the ground telescope VLBI network

in most directions at diﬀerent times of the year except those that are aﬀected by the

sun.

The last two subﬁgures reveal the poor performance of VLBA and EVN in terms of the

maximum baseline distance in observing the south polar directions. The two existing

VLBI networks lack of available telescopes in the south hemisphere of the earth and

therefore lose some observability in those directions.

As we analyze in the previous sections, under the current ITU-R RFI threshold guide- line the eﬀect of the RFI from the LEO satellites to the ground telescopes is negligible for

VLBI observation even if the satellites are using the RAO band as downlink. Under this circumstance, our proposed LEO telescopes can cooperate with current VLBI networks to

90 improve the observation performance of both sides. For example, the LEO telescopes help

the ground telescopes to improve their poor performance in the south hemisphere while the

latter help the former cover the direction of the sun. Another notable aspect of the VLBI

observation of the LEO telescopes is the timing synchronization. The LEO telescopes in the

proposed paradigm will send the raw RAO data with time stamp to the ground gateways and

further data synthesis and processing will be done at the ground data center. An accurate

and reliable clock/time stamp can be established by using a ﬁne-tuned internal clock (e.g.,

an atomic clock) or external clock (e.g., the GPS signals) or jointly using the two types of

clocks. Similarly, the on-board clocks are synchronized before conducting RAO to ensure

the accuracy of the time stamp. In addition, since the LEO telescopes are moving fast in

the space, the Doppler eﬀect of the astronomical signals needs to be considered. As the

orbits of the LEO telescopes are known (as can be measured [68, 14] ) in advance, the cor-

responding Doppler shift of the observed signals can be determined based on the telescopes’

movements and the RAO target direction and therefore can be canceled in data processing.

To explain further, denote the satellite location vector of satellite j at time t as Lj(t), the unit target direction vector as D(t) and the movement vector of satellite j as Vj(t). The inter-angle between the target direction and the zenith direction of the satellite i is given

D(t)·Lj (t) as θj(t) = arccos( h ) where h is the height of the satellite referred to the earth cen-

ter. Assuming the maximum oﬀ-axis observation angle of the satellite-based telescope is θ0, the index set of the telescopes that can observe the target at time t can be represented as j ∈ IT(t) such that θj(t) ≤ θ0. The movement (speed) of satellite j in the target direction is ∆Vj(t) = D(t) · Vj(t). Given the sampling frequency fs, the kth sampled signal on satellite j at time t can be represented as sj[k] , sj(t = t0 + k/fs). Suppose the center fre-

quency of the RAO band as fRAO. Then the corresponding Doppler shift of satellite j’s kth

∆Vj (t=t0+k/fs) sample is ∆fj[k] = c fRAO where c is the speed of the light. The Doppler com- √ pensated baseband RAO signal can be represented as s0 [k] = s [k] exp(− −1 2π∆f [k] k ). j j j fs

91 After canceling the Doppler shift, the data processing center will synchronize the RAO

data from diﬀerent telescopes. The time delay for satellite j with reference to the center

h of earth is ∆Tj(t) = − c sin θj(t) where the minus sign means the time when the signal of the target reaches the telescope is earlier than the time when it reaches the earth center

(hypothetically). Then, the propagation delay induced fractional phase diﬀerence between telescope j and j is ∆φ (t) = 2πf · mod(∆T (t) − ∆T (t), 1 ) and the propa- ref j,jref RAO jref j fs

gation delay induced integer RAO sample index diﬀerence between telescope j and jref is

τj,jref (t) = b(∆Tj(t) − ∆Tjref (t))/fsc where j, jref ∈ IT(t). For the given satellite network,

{∆φj,jref (t), τj,jref (t)} can be determined before conducting RAO. Then, the synchronization

0 for the Doppler compensated baseband RAO signal sj[k] of the satellite j can be performed √ 0 at the ground RAO data processing center as {exp(− −1 ∆φj,jref (t))sj[k − τj,jref (t)]} where

3 t = t0 + k/fs.

3.4.3 Sensitivity of LEO versus Ground Telescopes

The sensitivity of the telescope reﬂects the lowest level of astronomical signals that can be detected by the telescope. To compare the RAO performance of the proposed system with the existing ground telescopes, we analyze the sensitivity performance of the proposed LEO telescopes and the ground telescopes. Based on [62], the sensitivity of a single dish telescope can be represented as

2kTsys ∆Ssingle = √ (3.6) Ae TintBRAO

where BRAO is the RAO band bandwidth, k is the Boltzmann constant, Tsys is the system

noise temperature of the telescope, and Ae is the eﬀective area of the telescope in the RAO

band. Ae can be represented as Ae = Aphy · ηeﬀ where Aphy is the physical aperture of

the parabolic antenna and ηeﬀ is the aperture eﬃciency of the antenna in the considered

3The effects of the local oscillator induced phase offset on the VLBI measurements can also be identified and compensated, for example, by a typical calibration phase based on known target objects.

92 RAO band. For the telescope array with Na identical telescopes (telescopes with identical hardware and levels of system noise), the sensitivity of the telescope array can be represented as

2kTsys ∆Sarray = p (3.7) Ae Na(Na − 1)TintBRAO Speciﬁcally, as the ground telescopes may face the RFI from the NGSO satellites’ downlink, the corresponding degradation should be considered. Therefore, we can reﬁne Eq. (3.6) to incorporate the RFI from the NGSO satellites as

0 2kTsys(1 + κ) ∆Ssingle = √ (3.8) Ae TintBRAO where κ reﬂects the ratio between the RFI power and the system noise power. Here we consider a noise-like RFI which cannot be split from the desired astronomical signals. As mention in [45], the RFI should not introduce an error of 10% in measurement. In other words, the κ should be less than 10% to avoid corrupting the RAO data. However, from the analysis in Section 3.2.5, we can see that the instantaneous RFI level generated by the OneWeb NGSO system will be 15 db – 50 dB higher than the detrimental RFI level, which means that the κ can be up to 10000 (50dB higher than 10%). Under this condition, the RFI becomes the major source that severely limits the sensitivity of the ground telescopes. To compare the sensitivity of the two types of telescopes, we choose ground telescopes with 25m (meter) (e.g., the VLBA telescope in Owens Valley, California) and 100m (e.g., the Green Bank telescope in Green Bank, West Virginia) dish sizes as the benchmarks to address the sensitivity advantages of the proposed LEO telescope array. The Tsys of the ground telescope in 10.6 – 10.7 GHz RAO band is considered to be 35 Kelvin (K) [45] as the ground telescope can use cryocooler to lower the system noise temperature. On the other hand, depending on the solar illumination as well as the cooling component(s) on the satellite (e.g., passive and/or active cooling component(s)), the system temperature of the LEO telescopes can be diﬀerent. Therefore, we pick {35, 85, 135} K [45] as the alternative

93 system temperatures for the LEO telescopes. Note that the LEO telescope conduct RAO during nighttime and the temperature of the components can be as low as 70 K [87]. As the OneWeb satellites have limited size, the dish size of the LEO telescopes can not be too large. A conservative estimation of the dish size of the LEO telescopes is 3 meter. The aperture efficiency is assumed to be 0.15 [69] for both types of telescopes. Then, we can obtain the sensitivity of the proposed LEO telescope array as a function of the number of the telescopes in the array which are conducting the RAO to the same target simultaneously. The corresponding results are shown in Fig. 3.14. In addition, we show the sensitivity of the ground telescope with aforementioned dish sizes and levels of RFI from the LEO satellites in the figure. From the figure we can see that larger Na can help the LEO telescopes to reduce the sensitivity level. Note that lower sensitivity level means the telescope can detect signal with lower power, which indicates better observation performance. Given enough number of LEO telescopes conducting RAO simultaneously (e.g., Na > 120) and Tsys ≤ 85 K, the proposed LEO telescopes array have lower sensitivity level than the ground telescope has with 25m dish size even if no RFI is assumed at the ground telescope. However, due to the large difference of the effective area between the 100m ground telescope and the proposed LEO telescopes, the sensitivity level of the proposed LEO telescopes is higher than that of 100m ground telescope assuming no RFI at the telescope. Nevertheless, from the analysis in the previous section we can see that large-scale NGSO system will inevitably generate strong RFI to the ground telescopes and under such condition the proposed system can provide better sensitivity performance than the ground system as can be observed in Fig. 3.14.

3.5 Data Rate Analysis Based on a Shared RAS Band in the Proposed Paradigm

3.5.1 Gateway-Satellite Model Based Data Rate Analysis

As mentioned in the previous sections, the LEO SCS may use the bands which are assigned to RAS while it provides RAS a network of LEO telescopes. To evaluate how much more

94 104 Space telesocpes with T = 35 K sys Space telesocpes with T = 85 K sys 3 Space telesocpes with T = 135 K 10 sys Ground telescope with d = 25 m and κ = 10000

2 10 Ground telescope with d = 100 m and κ = 10000

Ground telescope with d = 25 m and κ = 100 1 10 Ground telescope with d = 100 m and κ = 100 Sensitivity (mJy) 100

Ground telescope with d = 25 m and κ = 0 -1 10 Ground telescope with d = 100 m and κ = 0

10-2 0 50 100 150 N a

Figure 3.14. Sensitivities of the ground and LEO telescopes

Satellite Source satellite Gateway Destination Satellite Gateway

(a) (b) Figure 3.15. System topological graphs. (a) Local graph for data rate analysis with M = 4. (b) Topological graph for RAO data transport with L = 2 and Ns = Nd = 3 (In practice, Ns could be greater than Nd). data rate the new RAS bands can bring to the SCS, we consider a system model based on

[65] which captures the essence of the data transmission in the SCS. Instead of considering all gateways and satellites in the SCS, we start analyzing the maximum supportable data rate of a certain gateway-satellite chain.

95 From [65], we can see that a gateway can directly connect to one specific LEO satellite via one antenna and other adjacent M − 1 LEO satellites connect the gateway via this satellite, which means the directly connecting satellite serves as a relay for other satellites. An example of the topological graph is shown in Fig. 3.15(a) to illustrate the connectivities we consider in this section. Then, for the directly connecting satellite, there are 4 major links which are: satellite to gateway (SG) link, gateway to satellite (GS) link, satellite to user (SU) link and user to satellite (US) link. On the other hand, the remaining M − 1 LEO satellites in the gateway-satellite chain only have their own SU and US links. We regard the traffics from multiple users within one satellite coverage as an aggregate traffic so that the SU broadcast link and the US multiple access link are simplified to point-to-point links. In addition, as the LEO telescopes need to send the observation data to the data processing center through the gateways, we also need to take this RAO data into account and evaluate the overall data rate of the aforementioned SCS model. We assume a fixed data rate RRAS/M is reserved for RAO data downlink transmission for each satellite which results in an aggregate RAO data rate of RRAS in the SG link. Besides, the traffic (in terms of packets per second) at the same satellite obeys a Poisson distribution and each link has its own packet size. For a link n, the total capacity Cn can be represented as

ηnGn Cn = Bn, n ∈ N (3.9) βn where ηn, Gn, βn, and Bn are the spectrum eﬃciency, multiplexing gain, frequency reuse factor, and assigned bandwidth of link n and N = {SU, US, SG, GS}. Then, we can deﬁne the outage probability of link n as

Pout,n = P (rn > Cn), n ∈ N (3.10)

where rn is the instantaneous data rate, which can be represented as rn = ρnxn with ρn and xn being the packet size and instantaneous traﬃc (packets/sec) of link n. Then, the data

96 rates of the 4 links are given by

rn,i = ρnxi, n ∈ {SU, US}, i = 1, ..., M, M (3.11) X rm = ρm xi, m ∈ {SG, GS}. i=1

Then, denoting the mean of xn as λn, to meet the required outage probability Pout,n, we can find a maximum mean supportable data rate (MMSDR) Rn as Rn = ρnΛn,max, n ∈ N , where Λn,max = max λn such that Pout,n ≤ Pout,req,n. Specifically, for the SG link, as the SCS will provide RAO data transmission service to RAS side, a part of the data rate will be reserved for RAS data downlink transmission. Thus, we have ΛSG,max = max λSG such that P (rSG > CSG − RRAS) ≤ Pout,req,SG. Assuming the average traffic ratio between the user

ρSU downlink and uplink is ζ = , we have RSU = ζRUS and RGS = ζRSG where the second ρUS equation can be obtained from Eq. (3.11). Then, due to the cascaded nature of the links between users and gateways, we will have the maximum mean supportable data rate TGSU for the cascaded gateway-satellite-user (GSU) link, and TUSG for the cascaded user-satellite- gateway (USG) link as TGSU = min(RGS,MRSU) and TUSG = min(RSG,MRUS). After that, we can have the overall MMSDR of the SCS as the sum of TGSU and TUSG. From [65] we can see that the OneWeb LEO satellites use 4 diﬀerent and discontinuous bands for the 4 diﬀerent links. In this chapter, we consider the SCS may exploit the shared

RAS band in two potential modes: Time division Multiplexing (TDM) mode and Frequency

Division Multiplexing (FDM) mode. In TDM mode, the SCS will let the four different links use different subframes at different times and each link can use the whole band during its own subframes. On the other hand, in FDM mode, each of the four links will use a sub-band of the RAS band and transmit information independently. Suppose link n uses αn proportion of the shared RAS band (in TDM mode, the αn can be viewed as the ratio of the number of subframes that are assigned to this link over the total number of subframes per frame), we

97 can represent the new channel capacity of link n as

˜ ηnGn Cn = (Bn + αn∆B), n ∈ N (3.12) βn

where ∆B is the bandwidth of the shared RAS band. Note that in TDM mode, αn can be adjusted according to the required RAS data rate due to the ﬂexibility in subframe

assignment while in FDM mode, αn is fixed due to inflexibility/ infeasibility of filtering between different links. Then, the data rate maximization problem of the system can be

represented as

max TGSU + TUSG, (3.13) {αn: n∈N } X s.t.RSU = ζRUS,RGS = ζRSG, αn + α0 = 1 n∈N where α0 is the proportion of the shared RAS band that is assigned for guard band/period or other purposes and thus cannot be used for data transmission.

3.5.2 Communication System Maximum Mean Supportable Data Rate and

RAO Data Rate Results

To evaluate the MMSDR of the integrated SCRAS, we consider 3 cases of band utilization in the proposed paradigm, which are i) the system uses the bands which are originally assigned to SCS only, ii) the system uses the original SCS bands and a shared RAS band in TDM mode, and iii) the system uses the original SCS bands and a shared RAS band in FDM mode. Note that when RAO data rate is 0, the performance of case 1 can be viewed as the performance of the original SCS. Table 3.4 shows the parameters of the 4 links we use in the performance evaluation, which is originated from [65]. We choose the RAS band in 10.6

– 10.7 GHz as the example shared RAS band. In addition, we consider each link uses 20

MHz subband and the total guardband is 20 MHz in the FDM mode. For TDM mode we

98 conﬁgure each frame with 100 subframes and each subframe has 1 ms duration. The guard

period in TDM mode is 11 ms and equivalent to 11 subframes.

Fig. 3.16 shows the relationship between the RAO data rate per gateway and SCS

MMSDR with different values of M and outage probability Pout in the different band utilization cases. As we can see from the figure, both of the spectrum sharing modes (case 2

and 3) can aﬀord more SCS data transmission than case 1 in general. In addition, due to

the resource allocation ﬂexibility, the TDM mode can achieve higher SCS MMSDR than the

FDD mode. Comparing the SCS MMSDRs achieved by diﬀerent modes, we can ﬁnd out

that SCS has approximately 1.1 Gbps more data rate in the TDM mode than in the original

allocation when M = 5 and 0.33 Gbps more data rate when M = 2. In other words, if the integrated SCRAS maintains the same MMSDR supported by the original SCS (case 1 with 0 RAO data rate), it can support approximately 0.4 Gbps RAO data rate when M = 5

and 3.8 Gbps when M = 2 with the new band from RAS. Since the extra bandwidth of

0.1 GHz is relatively small compared with the SCS’s original bandwidth of 6.9 GHz, the

MMSDR improvement over the original SCS is limited. Nevertheless, several suitable RAS bands including 15.35 – 15.4 GHz, 22.21 – 22.5 GHz and, 23.6 – 24 GHz are around the LEO satellite downlink bands and therefore greater improvement can be achieved if the RAS side also shares these bands.

Fig. 3.16 also indicates how the bottleneck of the local system MMSDRs is aﬀected by the aggregate RAO data rate and the number of satellites supported by the gateway. For

M = 5, the SG link is the bottleneck link of the local system. On the other hand, for

M = 2, the bottleneck link changes from the US link to the SG link when the RAO data rate increases from below 3.5 Gbps to above 3.5 Gbps, which leads to slope changes of the corresponding curves. The different bottleneck links for M = 2 and 5 with the same RAO data rate are caused by the different SCS traffics on the SG link, which is related to the different values of M. Moreover, with the same bottleneck link, the two groups of curves

99 Table 3.4. Parameters for the LEO SCS links Assigned Bands Bandwidth Spectrum Eﬃciency Multiplexing Frequency Reuse Link n 4 (GHz) (GHz) ηn (bits/s/Hz) Gain Gn Factor βn User downlink (SU) 10.7 – 12.7 2.0 1 16 8 User uplink 12.75 – 13.25, (US) 14.0 – 14.5 1.0 1 16 8 Gateway 17.8 – 18.6, downlink 18.8 – 19.3, 1.8 4 2 1 (SG) 19.7 – 20.2 Gateway 27.5 – 29.1, uplink (GS) 29.5 – 30.0 2.1 2 2 1 4 Here we use a conservative setting. In practice, the spectrum eﬃciency depends on several system settings such as modulation type and SNR.

22 P =10-5 Case 1 outage P =10-4 Case 1 20 outage P =10-3 Case 1 outage 18 P =10-5 Case 2 M = 5 outage P =10-4 Case 2 16 outage P =10-3 Case 2 outage P =10-5 Case 3 14 outage P =10-4 Case 3 M = 2 outage 12 P =10-3 Case 3 outage

6 SCS maximum mean supportable data rate (Gbps) 4 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Aggregate RAO data rate (Gbps)

Figure 3.16. Aggregate RAO data rate and SCS maximum mean supportable data rate of the proposed integrated system for 3 band utilization cases

(M = 2 and 5) overlap when RRAS > 3.5 Gbps. In this case, the RAO data occupies a large proportion of SG link capacity and the spectrum resource shared by RAS side is used only for increasing the capacity of the SG link.

100 3.6 RAO Data Transport Design

3.6.1 Development of Data Transport

Data acquisition and transport are the two critical parts of an RAO mission. For an RAO,

the suitable satellite positions on the orbital surface which meet the angle requirement be-

tween the target direction and the zenith direction of the LEO telescope can be represented

by a dome centered at the target direction C with arc radius Rob. Under this circumstance, the proposed SCRAS selects the L nearest gateways to the C for RAO data downlink trans-

mission. In addition, we assume Nd,l LEO satellites directly connect the lth selected gateway and the connections between the LEO satellites and the gateways are based on the nearest neighbor criterion. As mentioned in [65], a gateway can have 10 antennas (or more in some cases) and one antenna can establish a two-way link connection with one LEO satellite at a time. Therefore, for the performance evaluation in this section, we assume a gateway connects to at most the 10 nearest satellites above the minimum elevation angle and the satellite selects the nearest gateway to set up a two-way connection. Then, the total number of the gateway-connected satellites of the selected L gateways Nd can be represented PL as Nd = l=1 Nd,l such that Nd,l ≤ 10. A simple example of the connectivities among the involved satellites and gateways is shown in Fig. 3.15(b) to illustrate the considered problem. Then, with this model, we can analyze the relationship between the SCS traﬃc and the supportable RAS data rate and design the data transport strategy accordingly. Despite our analysis is based on a snap-shot of the whole RAO period, it can be extended to the whole RAO period by dividing the whole period into several fractional periods with ﬁxed satellite-gateway connections.

Assume the SG link traﬃc from SCS side of satellite i can be represented as a Poisson random variable xi (packets/second) with a mean value λSG,i where i = 1,...,Nd. To guarantee the SCS SG link data transmission within a required outage probability Pout,req,SG

101 and accomplishing the RAO data transmission, the aﬀordable RAO data rate of the ith satellite RRAS,i is

˜ ˆ RRAS,i = max{CSG,i − CSG,i(Pout,req,SG), 0} (3.14)

ˆ where CSG,i(Pout,req,SG) is the minimum capacity in the SG link that needs to be assigned to

SCS to meet the outage probability requirement Poutage,req,SG. Thus, the total supportable

PNd RAO data rate is RRAS = i=1 RRAS,i. Eq. (3.14) indicates that a larger RAO data rate can be accommodated with more gateways or at the cost of either higher SCS outage probability or smaller SCS average traﬃc using the same number of gateways.

After obtaining the total supportable RAO data rate of the selected gateways, another data transport problem is how to allocate the RAO data rate of the working LEO telescopes to the selected gateways. As the RAO data is transmitted from the working LEO telescopes to the Nd gateway-connected satellites via Inter-Satellite Links (ISLs), a primary concern of this procedure is the relaying cost of the data. Suppose the LEO telescope j in the

RAO region generates RAO data with date rate sj, and to make full use of the aggregate

PNs PNd supportable RAO data rate and avoid congestion, we have j=1 sj = RRAS = i=1 RRAS,i where Ns is the number of the working LEO telescopes in the RAO region. Then, we can design the RAO data transport based on the Ns working LEO telescopes as the sources and the Nd gateway-connected satellites as the destinations. Denoting the data flow (in terms of packets per second) from the jth source to the ith destination as fj,i, we aim to minimize the total relaying cost of the RAO data by optimizing the allocation of the flows between the sources and the destinations. However, for a certain source and destination pair, there could be multiple paths depending on the connection topology of the satellite network and hence the corresponding relaying costs may vary. For the performance evaluation in this section, we define the relaying cost of a source and a destination as the number of ISL hops the data flow passed through. For simplicity,

102 we consider no maximum rate constraint to the RAO data ﬂow in ISLs and therefore the

relaying cost from the jth source to the ith destination cj,i is determined by the path with the minimum number of ISL hops. Then, the minimum cost RAO data ﬂow allocation problem can be formulated as

N N Xs Xd min cj,ifj,i, (3.15) {0≤fj,i} j=1 i=1 N Xs s.t. fj,i = RRAS,i, i = 1,...,Nd, j=1 N Xd fj,i = sj, j = 1,...,Ns. i=1 This ﬂow allocation problem can be recognized as a linear programming problem and therefore can be solved with some existing software such as MATLAB.

3.6.2 Data Transport Performance Results

In the simulation, we assume the RAO region is centered at 45◦N 100◦W with an observation radius Rob = 3000 km and 35 working LEO telescopes are in the region for the speciﬁc snap- shot we consider. We apply the same FDM settings in the previous section at all L gateways and assume the bandwidth of the RAS band assigned to the SG link is 100 MHz. In addition,

the mean traﬃc λSG,i of the ith satellite is generated by a Poisson distribution with the mean of 1200 packets/s. The RAO data rate of the working LEO telescopes is assumed to be same

and fixed in the period we evaluate so that sj = RRAS/Ns, j = 1,...,Ns. The simulation results are based on the average of 100 realizations of the random locations of the gateways (where the minimum distance between any two gateways is 1029 km) and of 100 realizations of the Poisson distributed traffics. Fig. 3.17 shows the results of the aggregate supportable RAO data rate with different numbers of selected gateways under different outage probability requirements of SCS. For the same outage probability requirement, the aggregate supportable RAO data rate increases

103 with the number of selected gateways, as more SG links are available. On the other hand, for the same amount of selected gateways, the larger SCS outage probability leads to large RAO data rate. The results are consistent with the discussion of Eq. (3.14) in the previous subsection. Fig. 3.18 demonstrates the relationship between the average relaying cost per packet and the aggregate RAO data rate under different outage probability requirements of SCS. From the figure we can see that the relaying cost increases with the RAO data rate. This can be explained by that the growing RAS data rate requires more gateways at farther locations from the RAO region center to be involved and consequently it increases the average relaying cost. In addition, the slope of the curves in the figure varies as the RAO data rate increases, which is caused by the non-uniformly distributed locations of the selected gateways. As the gateways can only be placed on land, the shape of the land will affect the distribution of the locations of selected gateways and therefore results in a non-constant slope of the curves. Furthermore, for the RAO data rate, the higher SCS outage probability results in lower relaying cost per packet, which is due to that more RAO data rate can be accommodated to the gateways with lower relaying cost.

3.7 Conclusions

In this chapter, we investigated the RFI eﬀect of the emerging large-scale LEO satellite system (using OneWeb LEO satellite system as an example) on the ground radio telescopes. As the communication beams of the LEO satellites cover almost our entire planet, the RFI to the ground radio telescopes is inevitable. Our evaluation shows that the potential RFI can be tens of dB above the acceptable interference threshold of the continuum observation, corrupting the radio astronomical observations in the LEO satellites’ (adjacent) downlink bands. On the other hand, with inherent high immunity to the RFI, the VLBI observation can withstand the same level of RFI. To reduce the RFI from the emerging large-scale LEO

104 180 Outage probability = 10-5 Outage probability = 10-4 160 Outage probability = 10-3

140

120

100

80 RAS data rate (Gbps)

20 5 10 15 20 25 30 Number of selected gateways

Figure 3.17. RAO data rate versus the number of the selected gateways

8 Outage probability = 10-5 Outage probability = 10-4 7 Outage probability = 10-3

Cost per packet (hops/pkt) 2

0 20 40 60 80 100 120 140 160 180 RAS data rate (Gbps)

Figure 3.18. RAO data packet relaying cost

105 satellites, we apply three existing methods namely transmission muting, guardband insertion, and samples excision method. Our numerical evaluation shows that although these methods successfully reduce the average RFI levels for some of the ground telescopes, they can cause significant capacity loss to the LEO satellite system or severe sample loss to the ground telescopes. To address the large-scale LEO SCS’s RFI issue and guarantee the performances of both the SCS and the RAS, we proposed an integrated NGSO satellite communication and radio astronomy system where the NGSO satellites are configured as an infrastructure for both SCS and RAS. With the proposed paradigm, the RAS can make continuum observation in the LEO satellite downlink bands as well as other bands if they are equipped the corresponding receivers in these bands. In addition, the LEO telescopes can achieve larger maximum baseline distance and larger number of simultaneous RAO in most directions in VLBI observation compared to the existing ground telescope VLBI networks. Moreover, as the proposed paradigm causes negligible data loss to the ground telescopes, the two types of telescopes can work together to further improve the performance of the VLBI observation. The sensitivity analysis also shows the advantages of the proposed space telescopes over the existing ground telescopes. With the shared RAS band, our new paradigm also increases the maximum mean supportable data rate of the SCS. Furthermore, we also developed a minimum cost RAO data transport design. Our results show that the data rates can be traded off between SCS and RAS, and a larger RAS data can be transported from space to ground at the cost of larger numbers of inter-satellite hops and gateways. Overall, the performance results collaborate that the proposed paradigm offers mutual benefits to both SCS and RAS and facilitates growth of both services.

106 CHAPTER 4

CONCLUSION

Satellite communication systems (SCSs) have been crucial components of the overall global communication infrastructure. The demands for next-generation ubiquitous global wireless access and the advances in low-cost commercial spacecraft launching have recently ignited substantial interests for developing large-scale SCSs. Their inherent global coverage is an in- disputable advantage for SCSs but a serious RFI threat for a radio astronomy system (RAS). Advances and expansions in SCSs and RASs need more radio spectrum without detrimental RFI for both. These situations lead to increasing conflicts in spectrum access, service growths, and coexistence between SCSs and RASs. This dissertation first addressed the RFI issue of the GSO satellites and proposed a spectrum sharing paradigm for GSO SCSs and RAS to solve the conflict between them with three potential RFI reduction methods. These methods were formulated as i) reorganizing of the spectrum resource to minimize the unwanted emission of the GSO satellites in the RAO bands, ii) rearranging the subband allocation of the GSO satellite’s beams based on their unwanted emission powers and bandwidths, and iii) cell-based beam switch approach to suppress the RFI of the GSO satellites. Our investigation shows that the proposed paradigm can guarantee most of the ground telescopes with sample loss rate lower than 2% in both protected and unprotected RAS bands and achieve more SCS average downlink throughput than the existing RFI reduction methods. Secondly, this dissertation proposed a new paradigm where SCS and RAS are integrated into the NGSO satellite system, thus effectively creating large-scale telescopes in orbit. This integrated system not only avoids NGSO SCS’s RFI to RAS but also offers more spectrum access opportunities to both SCS and RAS. The sensitivity analysis also shows the advantages of the proposed space telescopes over the existing ground telescopes. With the shared RAS band, our new paradigm also increases the maximum mean supportable data rate of the

107 SCS. Furthermore, we also developed a minimum cost RAO data transport design. Our results show that the proposed paradigm offers mutual benefits to both SCS and RAS and facilitates growth of both services. We pioneer the new paradigm with the integrated SCS and RAS large-scale satellite system. Comparing to the existing two individual systems, the integrated system can allow the space-based telescopes conduct continuum in or adjacent to NGSO satellites’ downlink bands while maintaining the accessibility of these bands for the ground telescopes when the ground telescopes are conducting VLBI observation. Besides, the NGSO communication satellites can avoid implementing RFI-reduction methods in these bands and reuse the RAS bands for downlink transmission to improve the overall throughput of the system. Furthermore, the LEO-based telescope can observe the bands with total atmospheric opacity as well as other bands which suffer ground interference if observed at ground radio telescopes. The proposed space telescopes, if formulated as an array for observation, can achieve similar or better sensitivity than the ground single-dish telescope in some cases. Under the proposed new paradigm, in addition to what we have investigated in this thesis, a few interesting topics may worth investigation and exploration in future. For instance, a more complicated protocol for SCS and RAO data transport can be proposed to meet more practical constraints (e.g., limited ISL capacity). Furthermore, as the SCS and RAO data may have different delay/cost requirements, the new protocol may exploit this aspect to achieve a balance between different types of services. In addition, the spectrum sharing methods in the first part can be further improved/ adjusted so that they can be applied in a more complicated system with LEO, MEO, and GSO communication satellites as well as ground and/or space-based radio telescopes. In such a way the overall throughput of the system can be increased while the RFI at ground and/or space telescopes can be limited at an acceptable level.

108 REFERENCES

[1] Abdelgelil, M. E. and H. Minn (2018). Impact of nonlinear RFI and countermeasure for radio astronomy receivers. IEEE Access 6, 11424–11438.

[2] Agasid, E., K. Ennico-Smith, and A. Rademacher (2013). Collapsible space telescope (cst) for nanosatellite imaging and observation.

[3] Al-Dulaimi, A., S. Al-Rubaye, Q. Ni, and E. Sousa (2015). 5G communications race: Pursuit of more capacity triggers LTE in unlicensed band. IEEE vehicular technology magazine 10 (1), 43–51.

[4] Ayodele, P. and F. Olabisi (2015). Interference protection of radio astronomy services using cognitive radio spectrum sharing models. In Proc. IEEE Networks and Commun. (EuCNC), pp. 86–90.

[5] Barnbaum, C. and R. F. Bradley (1998). A new approach to interference excision in radio astronomy: Real-time adaptive cancellation. The Astronomical J. 116 (5), 2598.

[6] Bentum, M., A.-J. Boonstra, and W. Baan (2010). The impact of cognitive radio on radio astronomy. In RFI mitigation Workshop, Volume 107, pp. 009. SISSA Medialab.

[7] Bentum, M. J., A. Boonstra, and W. Baan (2009). The coexistence of cognitive radio and radio astronomy. In Proc. IEEE/CVT 16th Annual Symp.

[8] Black, R. A., B. D. Jeﬀs, K. F. Warnick, G. Hellbourg, and A. Chippendale (2015). Multi-tier interference-cancelling array processing for the ASKAP radio telescope. In Proc. IEEE Signal Process. and Signal Process. Education Workshop (SP/SPE), pp. 261–266.

[9] Boonstra, A. and S. Van der Tol (2005). Spatial ﬁltering of interfering signals at the initial low frequency array (LOFAR) phased array test station. Radio science 40 (5).

[10] Boonstra, A.-J., A. Leshem, A.-J. van der Veen, A. Kokkeler, and G. Schoonderbeek (2000). The eﬀect of blanking of TDMA interference on radio-astronomical observations: experimental results. In Acoustics, Speech, and Signal Processing, 2000. ICASSP’00. Proceedings. 2000 IEEE International Conference on, Volume 6, pp. 3546–3549. IEEE.

[11] Bower, G. C. (2005). Radio frequency interference mitigation for detection of extended sources with an interferometer. Radio science 40 (5).

[12] Chung, S.-J. and F. Hadaegh (2011). Swarms of femtosats for synthetic aperture applications.

109 [13] Cluniat, C. and J.-J. Delmas (1994, May 3). Multifocal receiving antenna with a single aiming direction for several satellites. US Patent 5,309,167.

[14] Degnan, J. J. (1993). Millimeter accuracy satellite laser ranging: a review. Contributions of space geodesy to geodynamics: technology 25, 133–162.

[15] Dong, W., B. Jeﬀs, and J. Fisher (2005). Radar interference blanking in radio astronomy using a Kalman tracker. Radio science 40 (5).

[16] E. Abdelgelil, M. and H. Minn (2017). Non-linear interference cancellation for radio astronomy receivers with strong RFI. In Proc. IEEE Global Commun. Conf. (GLOBE- COM), pp. 1–6.

[17] Ellingson, S. W. (2003). Beamforming and interference canceling with very large wide- band arrays. IEEE Transactions on Antennas and Propagation 51 (6), 1338–1346.

[18] Ellingson, S. W. and G. A. Hampson (2002). A subspace-tracking approach to interference nulling for phased array-based radio telescopes. IEEE Trans. Antennas Propag. 50 (1), 25–30.

[19] Engelen, S., C. J. Verhoeven, and M. J. Bentum (2010). Olfar, a radio telescope based on nano-satellites in moon orbit.

[20] Etkin, R., A. Parekh, and D. Tse (2007). Spectrum sharing for unlicensed bands. IEEE Journal on selected areas in communications 25 (3), 517–528.

[21] European Science Foundation. Committee on Radio Astronomy Frequencies (2005). CRAF Handbook for Radio Astronomy, Third Edition. CRAF Secretariat, Netherlands Foundation for Research Astronomy.

[22] FCC (1996). Title 47 part 90.210 : Emission masks. https://www.gpo.gov/fdsys/ pkg/CFR-1996-title47-vol5/pdf/CFR-1996-title47-vol5-sec90-210.pdf. On- line; accessed 13 November 2019.

[23] FCC (2016, August). Before the Federal Communications Commission Washington, D.C. 20554. https://docs.fcc.gov/public/attachments/DA-16-875A1.pdf.

[24] FCC (2018, Jun.). FCC online table of frequency allocations. https://transition. fcc.gov/oet/spectrum/table/fcctable.pdf.

[25] Flores, A. B., R. E. Guerra, E. W. Knightly, P. Ecclesine, and S. Pandey (2013). IEEE 802.11 af: A standard for TV white space spectrum sharing. IEEE Communications Magazine 51 (10), 92–100.

110 [26] Ford, J. M. and K. D. Buch (2014). Rﬁ mitigation techniques in radio astronomy. In Geoscience and Remote Sensing Symposium (IGARSS), 2014 IEEE International, pp. 231–234.

[27] Foreman, V. L., A. Siddiqi, and O. De Weck (2017). Large satellite constellation orbital debris impacts: Case studies of OneWeb and SpaceX proposals. In AIAA SPACE and Astronautics Forum and Exposition, pp. 5200.

[28] Goris, M., A. Joseph, G. Hampson, and F. Smits (1999). Adaptive beamforming system for radio-frequency interference rejection. IEE Proceedings-Radar, Sonar and Naviga- tion 146 (2), 73–77.

[29] Han, D. and H. Minn (2017a). Performance analysis of distributed auxiliary radio telescopes under shared spectrum access paradigm and cooling power constraint. IEEE Access 5, 21709–21722.

[30] Han, D. and H. Minn (2017b). Performance analysis of distributed radio telescopes under shared spectrum access paradigm. In Proc. IEEE Global Commun. Conf. (GLOBE- COM), pp. 1–6.

[31] Hanson, W. (2015). A global Internet: The next four billion users. New Space 3 (3), 204–207.

[32] Hardy, G., J. Littlewood, and G. Polya (1952). Inequalities, Cambridge U.

[33] Hellbourg, G. (2015). Subspace smearing and interference mitigation with array radio telescopes. In Proc. IEEE Signal Process. and Signal Process. Education Workshop (SP/SPE), pp. 278–282.

[34] Hellbourg, G., K. Bannister, and A. HotarP (2016). Spatial ﬁltering experiment with the askap beta array. In Proc. Radio Freq. Interference (RFI), pp. 37–42.

[35] Hellbourg, G., A. Chippendale, M. J. Kesteven, and B. D. Jeﬀs (2014). Reference antenna-based subspace tracking for rﬁ mitigation in radio astronomy. In Proc. IEEE Signal and Inf. Process. (GlobalSIP), pp. 1286–1290.

[36] Hellbourg, G., T. Trainini, R. Weber, E. Moreau, C. Capdessus, and A. Boonstrd (2012). RFI subspace estimation techniques for new generation radio telescopes. In EUSIPCO, pp. 200–204.

[37] Hellbourg, G., R. Weber, K. Abed-Meraim, and A.-J. Boonstra (2014). RFI spatial processing at nancay observatory: Approaches and experiments. In Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP), pp. 5387–5391.

111 [38] Hellbourg, G., R. Weber, C. Capdessus, and A.-J. Boonstra (2012). Oblique projection beamforming for RFI mitigation in radio astronomy. In Proc. IEEE Statistical Signal Process. Workshop (SSP), pp. 93–96.

[39] Hirabayashi, H. et al. (2000). The VLBI space observatory programme and the radio- astronomical satellite HALCA. Publications of the Astronomical Society of Japan 52 (6), 955–965.

[40] IRIDIUM (2017, August). IRIDIUM NEXT ENGINEERING STATEMENT. http: //licensing.fcc.gov/myibfs/download.do?attachment_key=1031348.

[41] ITU-R (1997, Sept.). Satellite antenna radiation pattern for use as a design objective in the ﬁxed-satellite service employing geostationary satellites. Recommendation S.672-4, ITU.

[42] ITU-R (2001, Feb.). Reference FSS earth-station radiation patterns for use in interference assessment involving non-GSO satellites in frequency bands between 10.7 GHz and 30 GHz. Recommendation S.1428-1, ITU.

[43] ITU-R (2002, Feb.). Protection of radio astronomy measurements above 60 GHz from ground based interference. Recommendation RA.1272-1, ITU.

[44] ITU-R (2003a, Jun.). Preferred frequency bands for radio astronomical measurements. Recommendation RA.314-10, ITU.

[45] ITU-R (2003b, May). Protection criteria used for radio astronomical measurements. Recommendation RA.769-2, ITU.

[46] ITU-R (2006, May). Protection of the radio astronomy service from transmitters operating in adjacent bands. Recommendation RA.517-4, ITU.

[47] ITU-R (2007a, Oct.). Interference calculations between non-geostationary mobile- satellite service or radionavigation-satellite service systems and radio astronomy telescope sites. Recommendation M.1583-1, ITU.

[48] ITU-R (2007b, Jun.). Protection of the radio astronomy service in frequency bands shared with other services. Recommendation RA.1031-2, ITU.

[49] ITU-R (2015, Mar.). Levels of data loss to radio astronomy observations and percentage- of-time criteria resulting from degradation by interference for frequency bands allocated to the radio astronomy service on a primary basis. Recommendation RA.1513-2, ITU.

[50] Jeﬀs, B. D., L. Li, and K. F. Warnick (2005). Auxiliary antenna-assisted interference mitigation for radio astronomy arrays. IEEE Trans. Signal Process 53 (2), 439–451.

112 [51] Kardashev, N. et al. (2013). “RadioAstron”-A telescope with a size of 300 000 km: Main parameters and ﬁrst observational results. Astronomy Reports 57 (3), 153–194.

[52] Kesteven, M., G. Hobbs, R. Clement, B. Dawson, R. Manchester, and T. Uppal (2005). Adaptive ﬁlters revisited: Radio frequency interference mitigation in pulsar observations. Radio science 40 (5).

[53] Kocz, J., F. Briggs, and J. Reynolds (2010). Radio frequency interference removal through the application of spatial ﬁltering techniques on the parkes multibeam receiver. The Astronomical Journal 140 (6), 2086.

[54] Kramer, M., I. H. Stairs, R. Manchester, M. McLaughlin, A. Lyne, R. Ferdman, M. Bur- gay, D. Lorimer, A. Possenti, N. D’Amico, et al. (2006). Tests of general relativity from timing the double pulsar. Science 314 (5796), 97–102.

[55] Lim, J., R. Klein, and J. Thatcher (2005). Good technology, bad management: A case study of the satellite phone industry. J. of Inf. Technol. Manage. 16 (2), 48–55.

[56] Micklethwaite, I. W. (1986, December 2). Motorized antenna mount for satellite dish. US Patent 4,626,864.

[57] Minn, H., Y. R. Ramadan, and Y. Dai (2016). A new shared spectrum access paradigm between cellular wireless communications and radio astronomy. In Proc. IEEE Global Commun. Conf. (GLOBECOM), pp. 1–6.

[58] Momjian, E. (2011, July). RFI at K-Band (18-26.5 GHz). https://science.nrao. edu/facilities/vla/observing/RFI/K-Band.

[59] Nagel, J. R., K. F. Warnick, B. D. Jeﬀs, J. R. Fisher, and R. Bradley (2007). Exper- imental veriﬁcation of radio frequency interference mitigation with a focal plane array feed. Radio Science 42 (6).

[60] Natera, M. A. S., A. G. Aguilar, J. M. Cueva, J. M. Fern´andez,P. P. De La Torre, J. G.-G. Trujillo, R. M. Rodr´ıguez-Osorio,M. Sierra-Perez, L. D. H. Ariet, and M. S. Casta˜ner(2011). New antenna array architectures for satellite communications. In Advances in Satell. Commun. InTech.

[61] National Telecommunications and Information Administration (2014). Manual of reg- ulations and procedures for federal radio frequency management. Assistant Secretary of Commerce for Communications and Information.

[62] Netherlands Institute for Radio Astronomy (2018). Sensitivity of the LOFAR array.

[63] Niamsuwan, N., J. T. Johnson, and S. W. Ellingson (2005). Examination of a simple pulse-blanking technique for radio frequency interference mitigation. Radio Sci- ence 40 (5).

113 [64] Nizhnik, O. (2012). A low-cost launch assistance system for orbital launch vehicles. International Journal of Aerospace Engineering 2012.

[65] OneWeb (2013). OneWeb Non-Geostationary Satellite System Attachment A.

[66] Pat Tatum (2017, Jun.). IRIDIUM SATELLITE L-BAND TRANSMIS- SION MEASUREMENTS. https://www.cept.org/Documents/se-40/36939/ se40-17-info-07-iridium-next-measurements.

[67] Peha, J. M. (2009). Sharing spectrum through spectrum policy reform and cognitive radio. Proceedings of the IEEE 97 (4), 708–719.

[68] Preston, R., R. Ergas, H. Hinteregger, C. Knight, D. Robertson, I. Shapiro, A. Whitney, A. Rogers, and T. Clark (1972). Interferometric observations of an artiﬁcial satellite. Science 178 (4059), 407–409.

[69] RadioAstron Science and Technical Operations Group (2018, March). RADIOASTRON USER HANDBOOK. http://www.asc.rssi.ru/radioastron/documents/rauh/en/ rauh.pdf.

[70] Ramadan, Y. R., Y. Dai, H. Minn, and F. S. Rodrigues (2016). Spectrum sharing between WiFi and radio astronomy. In Radio Freq. Interference (RFI), pp. 90–95.

[71] Ramadan, Y. R., H. Minn, and Y. Dai (2017, Sept.). A new paradigm for spectrum sharing between cellular wireless communications and radio astronomy systems. IEEE Trans. Commun. 65 (9), 3985–3999.

[72] Raza, J., A.-J. Boonstra, and A.-J. Van der Veen (2002). Spatial ﬁltering of RF interference in radio astronomy. IEEE Signal Processing Letters 9 (2), 64–67.

[73] Rosenberg, M., P. Russo, G. Bladon, and L. L. Christensen (2014). Astronomy in everyday life. Communicating Astron. to the Public J. 14, 30–36.

[74] Roshi, D. A. and G. B. NRAO (2004). RFI mitigation/excision techniques. In Spectrum Management for Radio Astronomy, pp. 237.

[75] Sardarabadi, A. M., A.-J. Van Der Veen, and A.-J. Boonstra (2016). Spatial ﬁltering of RF interference in radio astronomy using a reference antenna array. IEEE Trans. Signal Process 64 (2), 432–447.

[76] Shellhammer, S. J., A. K. Sadek, and W. Zhang (2009). Technical challenges for cognitive radio in the TV white space spectrum. In 2009 Information Theory and Applications Workshop, pp. 323–333. IEEE.

114 [77] Siddique, U., H. Tabassum, E. Hossain, and D. I. Kim (2015). Wireless backhauling of 5G small cells: challenges and solution approaches. IEEE Wireless Communica- tions 22 (5), 22–31.

[78] Straub, J. (2012). Cubesats: A low-cost, very high-return space technology. In Proceed- ings of the 2012 Reinventing Space Conference.

[79] Sullivan, W. T. (2001). The cultural value of radio astronomy. In Proc. Int. Astronomical Union Symp., Volume 196, pp. 369–376. Cambridge University Press.

[80] Tallinn (2011, October). Impact of unwanted emissions of Iridium satellites on radioas- tronomy operations in the band 1610.6-1613.8 MHz, ECC REPORT 171.

[81] Thaddeus, P., J. Vrtilek, and C. Gottlieb (1985). Laboratory and astronomical identi- ﬁcation of cyclopropenylidene, C3H2. The Astrophysical J. 299, L63–L66.

[82] Umar, R., Z. Z. Abidin, Z. A. Ibrahim, Z. Rosli, and N. Noorazlan (2014). Selection of radio astronomical observation sites and its dependence on human generated RFI. Research in Astronomy and Astrophysics 14 (2), 241.

[83] Van Driel, W. (2009). Radio quiet, please!–protecting radio astronomy from interference. Proceedings of the International Astronomical Union 5 (S260), 457–464.

[84] Van Veen, B. D. and K. M. Buckley (1988). Beamforming: A versatile approach to spatial ﬁltering. IEEE assp magazine 5 (2), 4–24.

[85] VAO LLC (2020). US Virtual Astronomy Observatory. http://www.usvao.org/index. html.

[86] Weber, R., C. Faye, F. Biraud, and J. Dansou (1997). Spectral detector for interference time blanking using quantized correlator. Astronomy and Astrophysics Supplement Series 126 (1), 161–167.

[87] Wertz, J. and W. J. Larson (1999). Space Mission Analysis and Design, Space Tech- nology Library. Microcosm Press and Kluwer Academic Publishers, El Segundo, CA, USA,.

[88] Wilcoxson, D., B. Sleight, D. Buchman, and R. VanderMeulen (2005). Ku-band satcom on-the move network. In Proc. IEEE Mil. Commun. Conf.(MILCOM), pp. 231–237.

[89] Xu, Y., X. Chen, and L. Ma (2010). LBS based disaster and emergency management. In Proc. IEEE Int. Conf. on Geoinformatics, pp. 1–5.

[90] Zhang, Q., Y. Zheng, S. G. Wilson, J. R. Fisher, and R. Bradley (2005). Excision of distance measuring equipment interference from radio astronomy signals. The astronomical journal 129 (6), 2933.

115 [91] Zhao, J., F. Gao, Q. Wu, S. Jin, Y. Wu, and W. Jia (2018). Beam tracking for uav mounted satcom on-the-move with massive antenna array. IEEE J. Sel. Areas Com- mun. 36 (2), 363–375.

116 BIOGRAPHICAL SKETCH

Yucheng Dai received his BS in Electrical Engineering from Nanjing University of Science and Technology, Nanjing, China in 2014. He received his MS in Electrical Engineering from Northwestern University, IL, USA in 2015. He is currently pursuing his PhD degree in Electrical Engineering from The University of Texas at Dallas, Richardson, TX, USA. From 2016 to 2020, he was a research/teaching assistant at The University of Texas at Dallas, Richardson, TX. USA. His research works are in the areas of spectrum access and sharing, interference analysis and suppression, satellite communication, MIMO beamforming design and realization, and millimeter-wave communications.

117 CURRICULUM VITAE

Yucheng Dai May 2020

Contact Information: Department of Electrical Engineering Email: [email protected] The University of Texas at Dallas 800 W. Campbell Rd. Richardson, TX 75080-3021, U.S.A. Educational History: B.S., Electrical Engineering, Nanjing University of Science and Technology, Nanjing, China, 2014 M.S., Electrical Engineering, Northwestern University, IL, USA, 2015 Ph.D., Electrical Engineering, The University of Texas at Dallas, TX,USA, 2020

Employment History: Research/Teaching Assistant, University of Texas at Dallas, TX, USA, January 2016 – May 2020

Professional Memberships: Institute of Electrical and Electronics Engineers (IEEE), 2017–present PUBLICATIONS Journals 1. Yucheng Dai, and Hlaing Minn, “A Spectrum Sharing Paradigm for GSO Satellite System and Radio Astronomy System,”IEEE Access, vol. 7, pp. 93952-93973, July 2019.

2. Yucheng Dai, Dong Han, and Hlaing Minn, “Impacts of Large-Scale NGSO Satel- lites: RFI and A New Paradigm for Satellite Communications and Radio Astronomy Systems,”IEEE Transactions on Communications, vol. 67, no. 11, pp. 7840-7855, July 2019.

3. Yahia R. Ramadan, Hlaing Minn, and Yucheng Dai, “A New Paradigm for Spec- trum Sharing between Cellular Wireless Communications and Radio Astronomy Sys- tems,”IEEE Transactions on Communications, vol. 65, no. 9, pp. 3985-3999, Sept. 2017. Conferences 1. Hlaing Minn, Dong Han, and Yucheng Dai, “A New Paradigm for Non-Geostationary Satellite Communications and Radio Astronomy System in IEEE International Con- ference on Communications (ICC) 2018, Kansas City, May 2018.

2. Yucheng Dai, Yahia R. Ramadan, Hlaing Minn, Jiu Xiong, Jin Liu, and Alan Gath- erer, “A New Receiver Architecture for MIMO Beam-Forming Applications,” in IEEE Global Communications Conference (Globecom) 2017, Singapore, December 2017.

3. Yahia R. Ramadan, Yucheng Dai, Hlaing Minn, and Fabiano S. Rodrigues, “Spec- trum Sharing Between WiFi and Radio Astronomy,” in RFI 2016 Conference, Socorro, NM, USA, October 2016.

4. Hlaing Minn, Yahia R. Ramadan, and Yucheng Dai, “A New Shared Spectrum Ac- cess Paradigm between Cellular Wireless Communications and Radio Astronomy,” in IEEE Global Communications Conference (Globecom) 2016, Washington DC, USA, December 2016.

5. Jiu Xiong, Jin Liu, Yucheng Dai, and Hlaing Minn, “An Open-loop 28 GHz 16-phase Clock Generator in 28nm CMOS,” in IEEE 60th International Midwest Symposium on Circuits and Systems 2017, Boston, MA, USA, August 2017.