WIRELESS COMMUNICATIONS AND MOBILE COMPUTING Wirel. Commun. Mob. Comput. 2011; 11:829–842 Published online 30 October 2009 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/wcm.864
SPECIAL ISSUE PAPER New deterministic and stochastic simulation models for non-isotropic scattering mobile-to-mobile Rayleigh fading channels† Xiang Cheng1∗, Cheng-Xiang Wang1, David I. Laurenson2, Sana Salous3 and Athanasios V. Vasilakos4
1 Joint Research Institute for Signal and Image Processing, School of Engineering & Physical Sciences, Heriot-Watt University, Edinburgh, EH14 4AS, U.K. 2 Joint Research Institute for Signal and Image Processing, Institute for Digital Communications, University of Edinburgh, Edinburgh, EH9 3JL, U.K. 3 Center for Communication Systems, School of Engineering, University of Durham, Durham, DH1 3LE, U.K. 4 Department of Computer and Telecommunications Engineering, University of Western Macedonia, GR 50100 Kozani, Greece
ABSTRACT
For the practical simulation and performance evaluation of mobile-to-mobile (M2M) communication systems, it is desirable to develop accurate M2M channel simulation models for more realistic scenarios of non-isotropic scattering. In this paper, by using a ‘double-ring’ concept to describe M2M non-isotropic scattering environments, we propose new deterministic and stochastic sum-of-sinusoids (SoS) based simulation models. The proposed simulation models extensively consider the distributions of the angle of arrival (AoA) and the angle of departure (AoD), and thus provide a good approximation to the desired statistical properties of the reference model. Copyright © 2009 John Wiley & Sons, Ltd.
KEYWORDS mobile-to-mobile communications; sum-of-sinusoids channel simulator; Rayleigh fading; non-isotropic scattering
*Correspondence Xiang Cheng, Joint Research Institute for Signal and Image Processing, School of Engineering & Physical Sciences, Heriot-Watt University, Edinburgh, EH14 4AS, U.K. E-mail: [email protected]
1. INTRODUCTION Besides the modeling of a M2M channel and the investigation of its statistical properties, the development In recent years, mobile-to-mobile (M2M) communica- of accurate M2M channel simulation models also plays a tions have received increasing attention due to some new major role in the practical simulation and performance eval- applications, such as wireless mobile ad hoc networks uation of M2M systems. However, in the literature, most [1], relay-based cellular networks [2], and dedicated short channel simulation models for wireless communications range communications for intelligent transportation sys- were developed for conventional fixed-to-mobile (F2M) tems (IEEE 802.11p) [3]. Such M2M systems consider that cellular radio systems [11--17].As mentioned in Reference both the transmitter (Tx) and receiver (Rx) are in motion [18], simulation models for a F2M channel cannot be and are equipped with low elevation antennas. To success- readily used to simulate a M2M channel due to the dif- fully analyze and design such M2M systems, it is necessary ferences of their statistical properties. Wang and Cox [19] to have proper mathematical reference models for the were the first to propose a simulation model for isotropic underlying propagation channels. Many M2M mathemati- scattering M2M Rayleigh fading channels by using the cal reference models have been proposed for both isotropic line spectrum filtering method. Due to the application of scattering environments, e.g., the models in References [4] line spectrum filtering, the model has high complexity and [5], and more realistic environments of non-isotropic and severe performance deficiencies as pointed out in scattering, e.g., the models in References [6--10]. Reference [18]. Therefore, the sum-of-sinusoids (SoS)
† This paper was presented in part at VehiCom’09, co-located with IWCMC’09, Leipzig, Germany, 21–24 June 2009.
Copyright © 2009 John Wiley & Sons, Ltd. 829 New deterministic and stochastic simulation models X. Cheng et al. method was used in References [18] and [20] to develop The paper is structured as follows. Section 2 gives a better isotropic scattering M2M Rayleigh fading simulation brief description of the reference model for non-isotropic models. Generally speaking, the SoS method allows us to scattering M2M Rayleigh fading channels. In Section approximate the underlying random fading processes by 3, we propose two new SoS based simulation models the superposition of a finite number of properly weighted (deterministic and stochastic models). The performance harmonic functions. An SoS based simulation model can be evaluation of our models is presented in Section 4. Finally, either deterministic or stochastic in terms of the underlying conclusions are drawn in Section 5. parameters (gains, frequencies, and phases) [16]. For a deterministic model, all the parameters are fixed for all simulation trials. In contrast, a stochastic model has at 2. REFERENCE MODEL least one parameter (gains and/or frequencies) as a random variable which varies for each simulation trial. Therefore, Using Akki and Haber’s mathematical model [4] and the relevant statistical properties of a stochastic model considering the directions of movement of the Tx and vary for each simulation trial but converge to the desired Rx, we can express the complex fading envelope of ones when averaged over a sufficient number of trials. In our reference model, under a narrowband non-isotropic Reference [18], the authors used the ‘double-ring’ concept scattering M2M Rayleigh fading assumption, as to derive two SoS based simulation models (deterministic and stochastic models) for isotropic scattering M2M 1 h(t) = hi(t) + jhq(t) = lim √ channels. The stochastic model in Reference [18] was N→∞ N further improved in Reference [20] by choosing orthogonal N functions for in-phase and quadrature components of the jψ j πf t φn −γ + πf t φn −γ e n e [2 Tmax cos( T T ) 2 Rmax cos( R R)] complex fading envelope. More recently, in Reference n= [21] the Rayleigh M2M stochastic simulation model in 1 Reference [18] was extended to include a line-of-sight (1) component, i.e., for Ricean fading channels. However, it is worth noting that all the aforementioned simulation models where hi(t) and hq(t) are in-phase and quadrature compo- limit their applications to isotropic scattering environments. nents√ of the complex fading envelope h(t), respectively, The simulation models for M2M Rayleigh fading channels j = −1, N is the number of harmonic function represent- f f under a more realistic scenario of non-isotropic scattering ing propagation paths (effective scatterers), Tmax and Rmax are scarce in the current literature. are the maximum Doppler frequency due to the motion To the best of the authors’ knowledge, only one of the Tx and Rx, respectively. The Tx and Rx move in stochastic SoS based simulation model [22] has been directions determined by the angles of motion γT and γR, proposed for the simulation of non-isotropic scattering respectively. The random AoA and AoD of the nth path are n n M2M Rayleigh fading channels. However, this model only denoted by φR and φT , respectively, and ψn is the random considered the symmetrical property of the distributions phase uniformly distributed on [−π, π). It is assumed that n n of the angle of arrival (AoA) and angle of departure φR, φT , and ψn are mutually independent random variables. (AoD) and was designed based on the acceptance rejection Since the number N of effective scatterers in the algorithm (ARA). This leads to the ARA model having reference model h(t) tends to infinity, the discrete expres- (n) (n) a notable difficulty in reproducing the desired statistical sions of the AoA φR and AoD φT can be replaced properties of the reference model and a comparatively by the continuous expressions φR and φT , respectively. high computational complexity. Furthermore, it is worth Note that since h(t) describes a non-isotropic scattering noting that accurate deterministic simulation models for M2M Rayleigh fading channel, the AoA φR and AoD φT non-isotropic scattering M2M Rayleigh fading channels exhibit non-uniform distributions. In the literature, many are not available in the current literature. different non-uniform distributions have been proposed to To fill the above gap, in this paper, based on the ‘double- characterize the AoA φR and AoD φR, such as Gaussian, ring’ concept in Reference [18], we first propose a new wrapped Gaussian, and cardioid PDFs [23]. In this paper, M2M deterministic SoS based simulation model. By allow- the von Mises PDF [24] is used, which can approximate ing at least one parameter (frequencies and/or gains) to be all the aforementioned PDFs and match well the measured a random variable, our deterministic model can be further data in References [7] and [24]. The von Mises PDF is modified to be a stochastic model. It is worth noting that defined as f (φ) exp[k cos(φ − µ)]/[2πI0(k)], where the proposed simulation models incorporate the probability φ ∈ [−π, π), I0(·) is the zeroth-order modified Bessel density functions (PDFs) of the AoA and AoD, and thus can function of the first kind, µ ∈ [−π, π) accounts for the approximate the desired statistical properties of the refer- mean value of the angle φ, and k (k ≥ 0) is a real-valued ence model for any non-isotropic scattering M2M Rayleigh parameter that controls the angle spread of the angle φ.For fading channel. Moreover, compared to the ARA stochastic k = 0 (isotropic scattering), the von Mises PDF reduces model in Reference [22], our stochastic model presents bet- to the uniform distribution, while for k>0 (non-isotropic ter approximation to the desired properties of the reference scattering), the von Mises PDF approximates different model with an even smaller number of harmonic functions. distributions based on the values of k [24]. Applying
830 Wirel. Commun. Mob. Comput. 2011; 11:829–842 © 2009 John Wiley & Sons, Ltd. DOI: 10.1002/wcm X. Cheng et al. New deterministic and stochastic simulation models
the von Mises PDF to the AoA φR and AoD φT ,we To model non-isotropic scattering environments, the obtain f (φR) exp[kR cos(φR − µR)]/[2πI0(kR)] and effective scatterers are assumed to be non-uniformly f (φT ) exp[kT cos(φT − µT )]/[2πI0(kT )], respectively. distributed. The Tx and Rx move with speeds υT and υR The autocorrelation function (ACF) of the reference in directions determined by the angles of motion γT and model h(t) can be expressed as [10] γR, respectively. Based on the ‘double-ring’ model, we can then design a new deterministic simulation model ∗ ρhh(τ) = E h (t) h (t − τ) as 1 1/2 = I A2 + B2 0 T T h˜ (t) = h˜ i (t) + jh˜ q (t) (4) I0(kT )I0(kR) 1/2 2 2 Ni,Mi × I0 AR + BR (2) ˜ 1 hi (t) = √ cos ψ˜ nimi NiMi with ni,mi=1 + πf t φ˜ mi − γ 2 Tmax cos( T T ) AT = kT µT + j πτfT γT cos( ) 2 max cos( ) (3a) ni + 2πf R t cos(φ˜ R − γR) (5) B = k µ + j πτf γ max T T sin( T ) 2 Tmax sin( T ) (3b) A = k µ + j πτf γ N q,Mq R R cos( R) 2 Rmax cos( R) (3c) 1 h˜ q (t) = sin ψ˜ nqmq BR = kR sin(µR) + j2πτfR sin(γR) (3d) NqMq max nq,mq=1 m ∗ + πf t φ˜ q − γ where τ is the time separation, (·) denotes the complex 2 Tmax cos( T T ) conjugate operation, and E[·] is the statistical expectation nq + 2πf R t cos(φ˜ R − γR) (6) operator. max
where h˜ i(t) and h˜ q(t) are in-phase and quadrature compo- 3. NEW SIMULATION MODELS nents of the complex fading envelope h˜(t), respectively, Ni/q is the number of effective scatterers located on the ring In this section, based on the reference model introduced in around the Rx, Mi/q is the number of effective scatterers ni/q Section 2, we propose the corresponding deterministic and located on the ring around the Tx, the AoA φ˜ R and the mi/q stochastic SoS based simulation models. AoD φ˜ T are discrete realizations of the random variables ˜ φR and φT , respectively, the phases ψni/qmi/q are the single outcomes of the random phases ψn in Equation (1). It 3.1. New deterministic simulation model ni/q mi/q ˜ is assumed that φ˜ R , φ˜ T , and ψni/qmi/q are mutually (ni/q) (mi/q) independent. Note that the AoA φ˜ and the AoD φ˜ Before proposing a new deterministic simulation model, R T remain constant for different simulation trials due to the which needs only one simulation trial to obtain the desired deterministic nature of the proposed simulation model. statistical properties, we would like to first introduce It is worth stressing that the single summation in the the ‘double-ring’ concept [18] to model non-isotropic reference model (1) is replaced by a double summation due scattering M2M scenarios. In such a case, the ‘double-ring’ to the double reflection incorporated by the ‘double-ring’ model defines two rings of effective scatterers, one around model. However, the channel essentially remains the same the Tx and the other around the Rx, as shown in Figure 1. for the following two reasons: (1) the assumption of the independence between the AoA and AoD is used in both the reference model (1) and the deterministic simulation model (4); and (2) for both the reference model (1) and the deter- ministic simulation model (4), each path will be subject to a υ T υ different Doppler shift due to the motion of both the Tx and R γ T Rx. From the above proposed deterministic simulation T R γ X X R model, it is obvious that the key issue in designing a M2M deterministic simulation model is to find the sets {φ˜ ni/q }Ni/q {φ˜ mi/q }Mi/q of AoAs R ni/q=1 and AoDs T mi/q=1 that make the simulation model reproduce the desired statistical properties of the reference model as faithfully as possible with reasonable complexity, i.e., with a finite number of Ni/q and Mi/q. Under the condition of non-isotropic Figure 1. Mobile-to-mobile scattering environment. scattering environments, the PDFs of the AoA φR and AoD
Wirel. Commun. Mob. Comput. 2011; 11:829–842 © 2009 John Wiley & Sons, Ltd. 831 DOI: 10.1002/wcm New deterministic and stochastic simulation models X. Cheng et al.
φ {φ˜ ni/q }Ni/q become T should be used to design the sets of AoAs R ni/q=1 mi/q Mi/q and AoDs {φ˜ T }m = that guarantee the uniqueness of the n − / i/q 1 n −1 1 4 (n) φ˜ R = FR (8a) sine and cosine functions related to the AoA φR and AoD N (n) φT in the reference model (1). This means that the design m − / of the sets of AoAs and AoDs for the reference model h(t) m −1 1 4 φ˜ T = FT (8b) in Equation (1) should meet the following two conditions: M n ni/q i/q (1) cos(φ˜ R − γR) = cos(φ˜ R − γR), ni/q = ni/q; and (2) m Note that the value of 1/4 used in Equations (8a) and mi/q i/q cos(φ˜ T − γT ) = cos(φ˜ T − γT ), mi/q = mi/q. (8b) guarantees that the designed sets of AoAs and Considering the non-isotropic scattering M2M environ- AoDs can meet the aforementioned two conditions n (n) m m ments, we now design the sets of AoAs and AoDs of our (φ˜ R =−φ˜ R + 2γR, n = n and φ˜ T =−φ˜ T + 2γT , deterministic simulation model. Via extensive investigation m = m) for Case I. The proof of the above statement is of the PDFs of the AoA φR and AoD φT , we find that unlike omitted here since one can easily prove it by following the isotropic scattering M2M environments [18], it is difficult procedure provided in Reference [25]. The performance of to obtain sets of AoAs and AoDs that meet the two condi- this design will be validated in Section 4. tions for any non-isotropic scattering M2M environments. Under the condition of Case II, the cross-correlation Therefore, we divide the non-isotropic scattering M2M between the in-phase component hi(t) and quadrature environment into the following three categories in terms of component hq(t) of the complex fading envelope h(t) the mean AoA µR and mean AoD µT (i.e., the PDFs of the in Equation (1) is equal to zero. Therefore, by setting AoA and AoD), and the angles of motion γR and γT and Ni = Nq and Mi = Mq in this case, we can directly use the then design the sets of AoAs and AoDs separately for these expression of our simulation model itself to guarantee the three cases: (1) Case I: the main transmitted and received aforementioned cross-correlation is equal to zero rather powers come from the same direction as or opposite than through the design of the AoAs and AoDs. This makes direction to the movements of the Tx and Rx, respectively, for a more efficient use of the number of harmonic func- ◦ i.e., |µT − γT |=|µR − γR|=0 or π; (2) Case II: the tions and thus results in better performance of the model. main transmitted and received powers come from the Following the similar parameter computation method directions that are perpendicular to those of the move- of Case I, we design the AoA and AoD of our model ments of the Tx and Rx, respectively, i.e., |µT − γT |= as ◦ |µR − γR|=90 ; and (3) Case III: different from Cases I n n − / n and II. i/q −1 i/q 1 2 i/q φ˜ R = FR , φ˜ R ∈ [−π, π) For Case I, the in-phase component hi(t) and quadrature Ni/q component hq(t) of the reference model h(t) in Equation ni/q = 1, 2,...,Ni/q (9a) (1) are correlated. Therefore, to meet this correlation, we set Ni = Nq = N and Mi = Mq = M and thus the AoA n m i/q i/q n m mi/q − mi/q − 1/2 mi/q φ˜ R and AoD φ˜ T can be replaced by the φ˜ and φ˜ , φ˜ = F 1 , φ˜ ∈ −π, π R T T T M T [ ) respectively. Inspired by the modified method of equal area i/q
(MMEA) proposed in Reference [25] for non-isotropic scat- mi/q = 1, 2,...,Mi/q (9b) tering F2M channels, we design the AoA and AOD of our model as Note that the value of 1/2 is applied in Equations (9a) and (9b) instead of the value of 1/4 in Equations (8a) and φ˜ n R (8b). The reason is that unlike Case I, for Case II it is n − 1/4 n n n = f (φ˜ R)dφ˜ R, φ˜ R ∈ [−π, π) difficult to design sets of AoAs and AoDs that meet the N φ˜ n−1 R two conditions. Although it is difficult to find one value n = 1, 2,...,N (7a) that can guarantee the designed sets of AoAs and AoDs have the best approximation to the two conditions. Based φ˜ m m − / T on simulations and motivated by the modified method of 1 4 = f φ˜ m dφ˜ m, φ˜ m ∈ −π, π ( T ) T T [ ) exact Doppler spread (MEDS) in Equations (37) and (38) M φ˜ m−1 T in Reference [18] for the simulation of isotropic scattering m = 1, 2,...,M (7b) M2M channels, we found that with the value of 1/2, the proposed simulation model presents better performance than the one with other values, e.g., 1/4. The performance φ˜ 0 =−π φ˜ 0 =−π f φ˜ n f φ˜ m where R and T , ( R) and ( T ) denote of this design will be validated in Section 4. φ˜ n φ˜ m the PDFs of the AoA R and the AoD T , respectively. The For Case III, since the in-phase and quadrature com- cumulative distribution functions (CDFs) of the AoA φ˜ R x n ponents of the reference model h(t) in Equation (1) are φ˜ F x = f φ˜ n dφ˜ and the AoD T are defined as R( ) −∞ ( R) R and N = N = N x m correlated (similarly to Case I) we set i q and F x = f φ˜ m dφ˜ ni/q T ( ) −∞ ( T ) T , respectively. If the inverse func- Mi = Mq = M as well and thus the AoA φ˜ R and AoD m −1 −1 i/q n m tions FR (·)ofFR(·) and FT (·)ofFT (·) exist, (7a) and (7b) φ˜ T can be replaced by the φ˜ R and φ˜ T , respectively.
832 Wirel. Commun. Mob. Comput. 2011; 11:829–842 © 2009 John Wiley & Sons, Ltd. DOI: 10.1002/wcm X. Cheng et al. New deterministic and stochastic simulation models
Following the similar parameter computation method of and frequencies to be random variables. Unlike the deter- Case I, we design the AoA and AoD of our model as ministic model, the properties of the stochastic model vary for each simulation trial, but will converge to the desired n − 1/2 ones when averaged over a sufficient number of simula- φ˜ n = F −1 , φ˜ n ∈ −π, π R R N R [ ) tion trials. A hat is used to distinguish this model from the deterministic one, thus n = 1, 2,...,N (10a)