Problem representation of dynamic resource allocation for flexible high throughput satellites Markus Guerster, Juan Jose Garau Luis, Edward Crawley, Bruce Cameron Massachusetts Institute of Technology 77 Massachusetts Ave 33-409 Cambridge, MA 02139 857-999-6103 {guerster, garau, crawley, bcameron}@mit.edu Abstract— Within the next years, flexible high-throughput (HT) The idea of providing broadband internet access through satellite with 100s-1000s of beams will be launched to provide space is not new. In the 90s, companies such as Iridium, broadband connectivity to a variety of customers on Earth. The Globalstar, and Orbcomm had similar ideas, but market user demand, especially in the mobility sector, is expected to uptake was poor [8-10]. However, over the past two decades, have large diurnal variations. To follow this dynamic demand behavior, many HT satellites will be equipped with flexible technological developments such as digital communication power and bandwidth capabilities. This flexibility comes with a payloads, advanced modulation, multi-beam antennas, and large number of adjustable parameters. Optimization of these advanced manufacturing are used by the new generation of parameters ensures that the demand can be met with the communication satellites. These advances lead to minimum required resources, resulting in an efficient utilization performance increases (in terms of data throughput) while at of assets on orbit. The challenge lies in the high dimensionality the same time reducing capital and operational costs. This of the problem, where manual resource allocation will quickly cost reductions could increase the competitiveness of satellite become impractical. This paper develops a representation of the broadband access relative to terrestrial broadband solutions. dynamic resource allocation problem and outlines an approach With the expected growth of demand, satellite broadband to solve the problem including different sets of algorithms such internet is expected to grow as well and might capture a larger as deterministic solvers and heuristic approaches from artificial intelligence. market share in the coming years [11]. This is especially true for the mobility sector (aircraft and ships), where terrestrial TABLE OF CONTENTS alternatives perform more poorly [12]. 1. INTRODUCTION ...................................................... 1 For current analog bent-pipe satellites, the spatial distribution 2. PROBLEM DECOMPOSITION .................................. 2 of capacity is fixed early in the design process based on 3. PERFECT VERSUS IMPERFECT INFORMATION ..... 3 demand predictions. If during the lifetime of the satellite, the demand behaves differently than predicted, the configuration 4. OPERATIONAL STATES OF THE DRM TOOL .......... 3 of the satellite cannot be changed to reflect the shift [13]. To 5. LIMITATIONS ......................................................... 6 avoid this lock-in, more new satellites use flexible 6. CONCLUSION AND OUTLOOK ............................... 6 communication payloads in the form of digital processors, ACKNOWLEDGMENT .................................................. 6 multi-beam antennas, and flexible amplifiers capable of REFERENCES ............................................................... 6 adapting to changing demand [14]. While these technologies are expected to provide great benefit, the operation of these BIOGRAPHY ................................................................. 7 satellites needs to be rethought. The degrees of freedom are orders of magnitude above those of current classical analog 1. INTRODUCTION bent-pipe satellites. Current resource allocation of power and The space industry is increasingly interested in providing bandwidth is mainly done manually. While this is still a broadband internet access through satellites. The Federal workable approach for satellites with limited degrees of Communications Comission (FCC) received 11 applications freedom, a new autonomous dynamic resource management from commerical companies for new high-throughput (HT) (DRM) needs to be developed to manage future satellites with non-geostationary satellites [1], especially large LEO significantly higher degrees of freedom. constellations, such as from Telesat [2], OneWeb [3] and SpaceX [4]. In addition to those newly proposed mega- Our objective with this paper is to present our first steps in constellations, established players in the communication developing a DRM for the future generation of highly flexible market expect to launch MEO and GEO satellites to HT satellites. specifically provide broadband connectivity. Viasat is The need for dynamic resource management is not new. planning to launch Viasat-3 to provide 100+ Mbps broadband Several authors addressed this issue by developing access by 2020 [5]. SES is launching a fleet of mPower optimization approaches [15-19]. Specifically, Sharma [20] spacecrafts to supplement their O3b MEO satellites within and Lagunas [21] developed an approach to monitor the the next years [6], as well as SES-17 – a HT GEO satellite spectrum and allocate bandwidth and exploit beamforming that will provide connectivity to America and the Atlantic capabilities. Ocean [7]. 978-1-5386-6854-2/19/$31.00 © 2019 IEEE 1 However, little work has been done to develop a descriptive • RF power-level adjustment per beam representation of the requirements for such a dynamic resource management (DRM) tool. The cited literature • Bandwidth adjustment per beam addresses technical sub-problems of the DRM. With this paper, our specific objective is to provide a more holistic • Beamforming and pointing (not considered in this system view. We will decompose the DRM into its main problem description) functionalities (Section 2), describe the behavior for perfect and imperfect information (Section 3), followed by a To operate satellites with those high degrees of freedom, discussion of the different states and configuration of the additional functional blocks are required, and the DRM (Section 4). We discuss the limitations of the chosen combination of these blocks results in the DRM architecture. representation in Section 5 and conclude the paper in Section Each main functionality is depicted in its own block in Figure 6 with an outlook on algorithms that are worth exploring for 1 and described below. the functional blocks within the DRM. Virtual models and constraints (VMC) to simulate the 2. PROBLEM DECOMPOSITION physical network The main task of the DRM tool is to control satellites in an The virtual models and constraints (VMC) are a virtual copy optimal way, so resources are used efficiently, and the data of the physical network which simulates the telemetry and rates provided to users can be maximized. To do so, the telecommands data streams. Both, the VMC and physical current operational infrastructure needs to be supplemented network connect to either an offline simulation or real-time by additional functional blocks. operation instance of the Real Time Engine (RTE), respectively. Loosely speaking, the current infrastructure consists mainly of the satellite physical network, which provides telemetry to Demand estimator to simulate the real demand a human operator and is controlled by telecommands. On a high-level, we decompose the physical network system into The demand estimator is continuously trained with real or 1…N satellites with gateways and user terminals for a variety simulated data sets to predict a pool of potential users of use cases. Specifically, we assume that future satellites throughout time that is a simulation of the real demand the have built-in flexibility in one or more of the following areas operational system might see. that make traditional manual operation impractical: Stage 1: Offline Stage 2: Real-time operational Development, and simulation, and what-if scenario analysis Deployment, and real-time operation of assets VMC (Virtual models & constraints) Physical network Model outputs Model inputs Telemetry Telecommand Offline simulation: Online, real operation: RTE (Real Time Engine, per satellite) RTE (Real Time Engine, per satellite) Mirrored vary: power, bandwidth copies vary: power, bandwidth constraint by: set of users, telemetry constraint by: set of users, telemetry to minimize: total power, total bandwidth to minimize: total power, total bandwidth n feasibility, performance set of users 1 Offline planning: Pool of CPO (Capacity Planning Optimizer, fleet) Optimal set of users Demand estimator potential (signing SLAs), users vary: set of users Guiding Trained with training demand data sets constraint by: power, bandwidth to maximize: revenues Figure 1: Architecture of the dynamic resource allocation (DRM). Decomposing the optimization functionality into capacity planning optimizer (CPO) and an offline and real-time instance of the real time engine (RTE) 2 Real Time Engine (RTE) to control the physical network and • Once the optimal set of users is found, the CPO passes VMC on the corresponding optimal plan (guidance) to the The RTE is responsible for assigning optimal power and online RTE instance. bandwidth levels to satisfy all demand using the minimum amount of resources possible and considering all the • The online RTE operates the system based on this constraints introduced by the physical network or VMC, optimal plan. The online RTE does not need to deviate respectively. The RTE is continuously working in cycles from the optimal plan since the demand predictions are (real-time operating system).