You are on page 1of 6

A Novel Stochastic Geometrical Model for Wideband MIMO-V2V Channels

Intelligent Networks Center (WINC), Nile University, Giza, Egypt 12677 Email: ahmed.amr@nileu.edu.eg,aelkeyi@nileuniversity.edu.eg Department of Communications, The Military Technical College, Cairo, Egypt 11331. Email: mohasseb@ieee.org
Wireless

Ahmad ElMoslimany , Amr El-Keyi , Yahya Mohasseb

AbstractIn this paper, we present a novel wideband multiple-input multiple-output Vehicle-to-Vehicle channel model. The proposed channel model is derived using the geometrical elliptical scattering approach. In order to emulate the appearance and disappearance of scatterers (vehicles, terrain, roadside units, etc.) in the environment, we associate a persistence process with each physical scatterer in the model. The parameters of the proposed model can be tuned to represent a variety of vehicular environments. We also derive the temporal and spatial correlation functions of the channel coefcients. Channel measurements and numerical examples are presented to provide more insight into the model and the physical interpretation of its parameters. Index TermsMIMO-V2V channel modelling, geometrical elliptical scattering, nonstationary channels.

I. I NTRODUCTION Vehicle-to-Vehicle (V2V) channels experience high relative velocities between the transmitter and the receiver in addition to a dynamic ambient environment. This results in a rich multipath fading environment in which the rapid motion of scatterers leads to continuous variation in the power-delay prole. Due to these unique features, V2V channels do not lend themselves easily to standard channel models, e.g., [1]. Instead, statistical models are needed to represent the time varying nature of the channel impulse response (CIR). Classical statistical channel models typically use the Wide Sense Stationary Uncorrelated Scattering (WSSUS) assumption [2]. However, for V2V channels, this assumption is not valid for prolonged time intervals. In fact, V2V channels are statistically nonstationary because of the physical environment dynamics. This is mainly due to the motion of the transmitter, receiver, and signicant reectors/scatterers. For example, the presence of a large truck on the side of the transmitter or receiver can contribute to a multipath component for
This work was supported by General Motors, USA.

a relatively short duration (until the vehicle passes the truck). Furthermore, the omnidirectional antennas for the transmitter and receiver are at low elevations, and hence, over moderate spatial scales reectors/scatterers will appear and disappear. There are two approaches for handling the nonstationary nature of practical V2V channels. The rst approach is based on the concept of a local scattering function, developed in [3] and used, e.g., in [4], to estimate the time interval over which the WSSUS assumption is valid. The second approach is based on modelling the channel as a tapped delay line with the tap amplitudes following some probabilistic distribution and modulated by a birth/death (on/off) process [5]. The on/off process for each tap was modelled by a rst-order Markov chain with a prespecied state transition matrix in [6]. In this paper, we develop a novel model for wideband MIMO-V2V channels. The proposed channel model is derived using the geometrical elliptical scattering approach [7]. However in [7], scattering is assumed to occur uniformly on an ellipse. This assumption does not suit the V2V scenario where the low elevation of antennas precludes a such a rich scattering assumption. Here, we associate a persistence process with each physical scatterer in the environment. The persistence process models the existence and absence of the physical scatterers, and hence, we can capture the dynamics of the scatterers in V2V channels. Note that, in general each time delay path (TDP) is due to the contribution of multiple scatterers and not a single one. Hence, the proposed model is a more accurate representation of the vehicular environment than the approach in [6] which modulates the tap coefcient representing a TDP by a birth/death process [6]. We will present the derivation and the characteristics of this model that make it suitable for modelling a wide variety of vehicular environments. Also, we will analyze the temporal and spatial correlation functions of the channel coefcients and investigate

the effect of the model parameters on these functions. Channel measurements are presented that illustrate the non-stationary behavior of V2V channels. In addition, numerical simulations are provided to illustrate the spatial and temporal correlation characteristics of the proposed MIMO-V2V channel model. II. S CATTERING M ODEL Let us consider a transmitter and a receiver equipped with multiple antennas each where the number of transmit (receive) antennas is given by MT (MR ). The angle of the direction of the relative velocity between the receiver and transmitter is denoted by v . The proposed wideband MIMO-V2V channel model is derived from the geometric multi-elliptical scattering model shown in Fig. 1. In this model, we assume that the centers of the transmit and receive arrays are located on the two foci of M ellipses. The distance between the two focal points is 2f . The major axis half-length and minor axis half-length of the mth ellipse are denoted by am and bm , respectively. The scatterers (vehicles, trees, buildings, etc.) that contribute to the same TDP lie on the same ellipse, where each ellipse corresponds to a different time-differentiable path (TDP) (delay bin). (m ) The mth ellipse contains Nc slots where each slot contains a scatterer. The scattering from the mth ellipse contributes to the mth channel coefcient whose delay is m = mTs where Ts is the sampling interval of the baseband-equivalent transmitted signal. The scattering slots are distributed along the ellipses and their number and distribution depend on the physical and vehicular environment, e.g., terrain type, density of vehicles on the (m ) (m ) road, etc. We use T and R to respectively denote the angles of departure and arrival of a ray travelling from the transmitter to the receiver via a scatterer located (m ) (m ) on the mth ellipse. The angle T ( R ) is measured from the center of the transmit (receive) array. Note that the model can be simply extended to include a line-ofsight component that contributes to the rst TDP. We use a multiplicative birth/death process, zn,m [q ], to account for the appearance and disappearance of scatterers in each ellipse. However, we do not account for the drift of scatterers into a different delay bin. The process models the persistence of the scatterer where zn,m [q ] = 1 if a scatterer is present in the nth slot of the mth ellipse at the q th time instant and zn,m [q ] = 0 otherwise. We model zn,m [q ] using a rst-order Markov model that assumes that the presence of the scatterer in the slot depends only on the current state and not on how long the slot was occupied. Although this assumption

Fig. 1. Geometric elliptical scattering model for an MT MR MIMO channel with local scattering clusters lying on the ellipse.

might not be very accurate in practice, it simplies the analysis of the statistics of the channel coefcients and still allows us to model a wide range of vehicular environments. Note that a rst-order Markov process was also used to model the persistence of the coefcients of the tapped delay line representing the V2V channel in [6]. The state transition probabilities of the Markov chain reect the degree of nonstationarity of the environment. For example, as the velocity of the vehicles increases, the probability that the Markov chain will make a transition from 0 to 1 or from 1 to 0 will increase since the scatterers will appear and disappear more frequently. We can write the probability transition matrix of this Markov chain as ! (n,m) (n,m) 1 01 01 (n,m) = (1) (n,m) (n,m) 1 10 10 where the i, j th entry of the matrix (n,m) denotes the probability that the Markov chain will make a transition (n,m) to state j 1 given that it is in state i 1. Let i denote the steady state probability that the Markov chain will be in state i, which is also equal to the long run proportion of time that the Markov chain will be in state (n,m) (n,m) (n,m) i. Hence, 0 + 1 = 1, and 1 can be written as
1
(n,m) (n,m)

(n,m) 01 (n,m) (n,m) + 01 10

(2)

The ratio 01 / 10 determines the ratio between the long-run proportion of time that the scatterers will be present in the scattering slots to that in which they will be absent. For example in a dense urban environment where vehicles move regularly and rarely change their relative position with respect to each other, the ratio (n,m) (n,m) / 10 will be relatively high. 01

(n,m)

III. C HANNEL I MPULSE R ESPONSE Let hkl [p, q ] denote the discrete-time basebandequivalent impulse response of the channel between the lth transmit antenna and the k th receive antenna where the index p is the delay index and q is the time index, i.e., M X (m ) hkl [p, q ] = hkl [q ] (p m) (3)
m=0

where (p m) is the Dirac delta function. According to the geometrical elliptical scattering model, the mth TDP is due to the contribution of the scatterers on the mth ellipse. Therefore, the mth channel coefcient can be written as (m) Z Nc X (m ) (n,m) (m) (m ) hkl [q ] = zn,m [q ] gkl ( R , q )d R (4)
n=0
(n,m) R

where gkl ( R , q ) is the contribution of the ray transmitted from the lth transmit antenna to k th receive element and scattered via the nth scatterer slot in the mth (m ) ellipse and received at an angle R at the receive array. Note that the scattering slot extends over the receive (n,m) (n,m) (m) angular interval R . We can write gkl ( R , q ) as (n,m) (m) (m ) (m ) gkl ( R , q ) = En,m ( R ) exp j n,m R (n,m) (m ) (m ) exp jK0 Dkl + j 2 FD R qTs (5) R where En,m ( R ) is the amplitude density function of the scattered wave from the nth scattering slot in the mth (m ) (m ) ellipse with respect to the angle R , n,m R is a random phase shift due to the scattering process, K0 = 2 / where is the wavelength of the RF propagating (m ) signal, and FD R is dened as
FD
(m ) R (m )

(n,m)

(m )

= fD cos

(m ) R

(6)

where fD is the maximum Doppler frequency. In (5), (n,m) (m) Dkl ( R ) is the total distance travelled by a ray emitted from lth transmit element to the k th receive element via a scatterer in the nth slot in the mth ellipse (m ) and received at an angle R . Note that we can write
Dkl
(n,m)

in the nth slot of the mth ellipse to the k th element in the receiver. For the sake of readability, in the remainder of this paper we will suppress the dependence (l,n,m) (m ) (n,k,m) (m ) (n,m) (m ) of DT , DR , gkl T R R ,q , (m ) (m ) (m ) FD and on the transmit and receive n,m R R (l,n,m) (n,k,m) (n,m) angles and denote them by DT , DR , gkl (q ), (m ) FD , and n,m , respectively. We assume that the aperture of the transmit and receive arrays is much less than the distance p am f for the M ellipses. Using the approximation 1 + x 1 + x/2 (l,n,m) (n,k,m) for x 1, we can approximate DT and DR as (l,n,m) (n,m) (m ) DT = DT dTl cos T Tl (8) (n,k,m) (n,m) (m ) DR = DR dRk cos R Rk (9) (n,m) (n,m) where DT DR is the distance from the center of the transmit (receive) array to the scatterer at the (m ) (m ) transmit angle T receive angle R in the nth slot of the mth ellipse, dTl (dRk ) is the distance between the lth element of the transmit array (k th element of the receive array) and the center of the transmit (receive) array, and Tl (Rk ) is the inclination angle of the lth transmit element (k th receive element). Note that (n,m) (m) (n,m) (m) DT ( T ) + DR ( R ) = 2am is constant for the mth ellipse and does not depend on the scatterer location (m ) on the ellipse. From Fig. 1, we can write T as 8 (m ) (m ) (m ) > f 0< R 0 > R > < (m ) (m ) (m ) (m ) (m ) = f R + < R 2 0 0 T > > > ( m ) ( m ) ( m ) :f + 2 2 < R 2 0 R (10) (m ) 2 m 1 where 0 = arctan( 2m ), m is the reciprocal of the eccentricity of the mth ellipse, i.e., m = am /f , and 0 1 (m ) 2 1 sin m R (m ) A f R =arctan @ (m ) 2 m + ( 2 + 1) cos m R (11) IV. S TATISTICAL P ROPERTIES OF THE C HANNEL C OEFFICIENTS In order to simplify the derivation of the correlation functions of the channel coefcients, we assume that 1) The channel coefcients that account for different TDPs are uncorrelated as they result from interactions from scatterers on different ellipses. 2) Scattering from

(m ) R )

= DT
(m )

(l,n,m)

(m ) T )

+ DR

(n,k,m)

(m ) R )

(7)

where DT ( T ) is the distance from the lth element in the transmitter to the scatterer in the nth slot (m ) of the mth ellipse at the angle T and analogously (n,k,m) (m) DR ( R ) is the distance from the same scatterer

(l,n,m)

different slots in the same ellipse is uncorrelated as each (m ) slot corresponds to a distinct scatterer. 3) En,m R = (m ) En,m is independent of R , i.e., the scatterers have a uniform radar cross section. Note that the scattering slots can be occupied by a vehicle, tree, building, etc. and their radar cross-sections are different and in general (m ) unavailable. 4) The scattering phase angles n,m R (m ) are independent for different n, m, and R and independent of the process zn,m [q ]. They are modelled as random variables uniformly distributed between 0 and 2 . The temporal correlation of the mth channel coefcient between the lth transmit antenna and the k th receive antenna at time instants q and q + p is given by
(m ) rkl [p, q ] =E
c nN X (m)

zn,m [q ]zn0 ,m [q +p]Skl

(n,m)

[q ]Skl

(n0 ,m)

n,n0 =0

o [ q + p]

Fig. 2.

Delay-temporal contour of the measured channel.

(12)

where E {} denotes the expectation operator which is taken over the random variables zn,m , zn0 ,m , n,m and n0 ,m , and
(n,m) Skl [q ] =

(n,m) R

gkl Z

(n,m)

(m ) (m ) R , q )d R

(13)
(m) R

Since both n,m [p, q ] and I ( R , p) do not depend on the index of the transmit or receive antenna, we can (m ) drop the dependence of rkl [p, q ] on the indices k and l and write the temporal correlation of the mth channel coefcient as
(m) Nc

(n,m)

= En,m e

(m) 2 FD qTs +n,m

(n,m) K0 Dkl

(14)

(m )

[p, q ] =

(n,m) R

where (14) follows from substituting with (5) into (13) and applying the third assumption.
(m) Nc X (m ) rkl [p, q ]= n,m n=0

[p, q ] En,m Skl

(n,m)

[q ] Skl

(n,m)

[ q + p] (15)

where n,m [p, q ] = Ezn,m {zn,m [q ] zn,m [q + p]} is the probability that the nth slot of the mth ellipsoid will be occupied by a scatterer at both time instances q and (q + (n,m) p). Given the probability 1 [q ] = P {zn,m [q ] = 1} and the probability transition matrix (n,m) , we can evaluate n,m [p, q ] using the Chapman-Kolmogorov equation as (n,m)p (n,m) [q ] (16) n,m [p, q ]=(2,2) 1 where (i,j ) is the i, j th entry of the p-step state tranp sition matrix (n,m) . Using the fourth assumption, we (n,m) can write the expectation in (15) as | En,m |2 I ( R , p) (n,m) where the integration I ( R , p) is given by Z (m) (n,m) (m ) I ( R , p) = ej 2FD pTs d R . (17)
(n,m) R

(18) The resulting expression for the temporal correlation in (18) shows that the channel coefcients are nonstationary as it depends on the index q through the process zn,m [q ]. Next, we evaluate the spatial correlation function as a function of the geometry of the transmit array Tx and receive array Rx . Let us consider the mth channel coefcient of two channels at the time instant q ; the rst from transmit antenna k to receive antenna l, and the second from transmit antenna k 0 to receive antenna l0 . The spatial correlation function of the mth coefcient of these two channels is given by
(m) rkl,k0 l0 (q, Tx , Rx )=E Nc n X
(m)

n=0

|En,m|2 (2,2) 1

(n,m)p (n,m)

[q ]I (

(n,m) R , p).

zn,m [q ]zn0 ,m [q ]Skl

(n,m)

[ q ] Sk 0 l 0

(n0 ,m)

(n,m)p

n,n0 =0

(19)

o [q ]

Using the modelling assumptions and going through the same steps as in the derivation of the temporal correlation function, we can write
(m) rkl,k0 l0
( m) Nc

(q, Tx , Rx )=

n=0

(n,m)

n o (n,m) (n,m) [q ]En,m Skl [q ]Sk0 l0 [q ]

(20)

Fig. 3. Temporal correlation function of the channel versus time and delay

For simplicity, we evaluate the spatial correlation function at q = 0. Using equations (7) (9), we can write (n,m) Dkl as
Dkl Rk ) (21) (n,m) (m) Substituting with Dkl ( R ) in the expression for (n,m) Skl [0] in (14) and using the fourth assumption, we can write the spatial correlation function at q = 0 as
Z Nc X (m ) (n,m) (n,m) (n,m) (m ) rkl,k0 l0(Tx , Rx )= |En,m|2 1 Cll0 (Tx ) Kkk0 (Rx ) d R n=0 (n,m)
( m) R

zeros), the CIR can be sampled once every 25.6 sec which is an order of magnitude higher than the expected coherence time of the channel. The captured samples are stored temporarily in an on-chip buffer and then transferred to the SRAM. The capture cycle is repeated every 1 msec, and hence, the CIR can be measured over a temporal duration of 256 msec. The CIR is computed by correlating the received samples with the transmitted sequences. In the measurement experiment, we use two vehicles where the antennas are mounted on the rooftops of the vehicles with a separation of 2.5 cm. The transmitted power is given by 19 dBm. The experiment is performed in a suburban area in the campus of Nile University. The two vehicles move in opposite directions at a speed of 30 km/hr each. Fig. 2 shows the delay-temporal contour plot of the impulse response of the channel between the rst receive and the rst transmit antennas. The gure illustrates the non-stationary nature of the CIR as some coefcients appear and disappear repeatedly. VI. A N UMERICAL E XAMPLE In this section, we will present a numerical example to illustrate the temporal and spatial correlation characteristics of the proposed model. We consider a vehicular channel at fc = 5.85 GHz with bandwidth 10 MHz, and hence, the sampling frequency of the CIR is Ts = 100 nsec. The relative speed between the two vehicles is 100 km/hr and the inclination angle of the velocity vector is v = 0. We consider only one TDP corresponding to one ellipse. We set the lengths of the major and minor axes as 20 and 12, respectively. We assume that there are Nc = 2 scattering slots that extend over the angular (1) (2) intervals R = [98 , 108 ] and R = [110 , 235 ]. The transition probabilities of Markov chain associated with the two scattering slots are identical and its parameters are given by 01 = 0.005 and 10 = 0.001. Note that the selected transition probabilities indicate that the scatterers exist in each slot for 83.33% of the time. The parameters of the simulation correspond to a highway environment with high mobility. Fig. 3 shows the temporal correlation of the channel coefcient versus the delay = pTs and the time t = qTs . The initial (n ) probabilities 1 [q ] were selected as 0.5 for n = 1, 2. From the gure we can see that the channel coefcient is not stationary as the temporal correlation is a function of time. Next, we investigate the spatial correlation characteristics of the channel model. We consider a uniform linear transmit and receive array equipped with MT = MR = 8

(n,m)

= 2a

dTl cos(

(m ) T

Tl )

dRk cos(

(m ) R

(22)

where the above integration can be evaluated numerically (m ) (m ) using the relationship between T and R in (10) and
Cll0
(n,m)

( Tx ) = e

jK0 dTl cos(

(m) T (m) R

Tl ) dT 0 cos(
l k

(m) T ( m) R

Kkk0 (Rx ) = e

(n,m)

jK0 dRk cos(

T 0 )
l

Rk ) dR0 cos(

R 0

(23) (24)

V. E XPERIMENTAL VALIDATION We have measured the CIR of a 2 2 MIMO-V2V channel operating at a center frequency of 5.8 GHz. The channel sounder was implemented on Rice Universitys Wireless Open-Access Research Platform (WARP) boards V1.2. The transmitted signal from each antenna is a periodic 511-chip Gold sequence that provides a coding gain of 54 dB. The signals from the two transmit antennas are time-multiplexed to remove the mutual interference. The sequences are generated at a rate of 40 106 chip/sec where sampling is performed at the chip rate. Since the duration of one code cycle is 1024 chips (511 chip for the m-sequence followed by 513

Fig. 4. Spatial correlation between channel coefcients h11 and h22

Fig. 5.

Angular power spectra of the simulated MIMO channel.

elements each and their tilt angles are given by T = R = 0. Fig. 4 shows the spatial correlation function between two channel coefcients h11 (t, ) and h22 (t, ) versus the normalized transmit antenna elements spacing T / and the normalized receive antenna elements spacing R / . We can see from this gure that the channel coefcients become almost uncorrelated when the spacing between the antenna elements of the transmit and receive arrays is in the order of 3 . Hence, spatial diversity gains can be harvested with an antenna array with inter-element spacing in the order of 15 cm that can be easily mounted on vehicles and roadside units. Next, we evaluate the joint direction-ofdeparture/direction-of-arrival (DoD/DoA) angular power spectrum (APS) of the channel coefcients generated according to the model. We consider the same parameters used in the last example except that we have a horizontal circular arrays at the transmitter and receiver sides whose elements are separated by 0.75 . The APS is calculated using the Capon beamformer from the sample spatial covariance matrix of the generated channel coefcients [8]. Fig. 5 shows the APS of the MIMO channel. We can see from this gure that the highest values of the APS are in the angular directions of the two scattering slots where the (1) rst slot extends over the DODs T = [14.5 , 17.5 ] (1) and DOAs R = [98 , 108 ] and the second slot (2) extends over the DODs T = [18 , 336 ] and DOAs (2) R = [110 , 235 ]. VII. C ONCLUSION We have derived a wideband MIMO-V2V channel model using the geometrical elliptical scattering ap-

proach that captures some of the inherent characteristics of vehicular environments. The persistence of the scatterers in the channel was modelled by a birth/death process. We have analyzed the statistical properties of the channel and derived closed-form formulas for the temporal and spatial correlation functions. Experimental measurements were presented that illustrate the nonstationary behavior of the channel. Our future work includes improving the spatial resolution and capture duration of the measurement platform to be able to extract the model parameters corresponding to different V2V environments. R EFERENCES
[1] J. D. Parsons, The Mobile Radio Propagation Channel, 2nd Edition. Wiley, 2000. [2] P. Bello, Characterization of randomly time-variant linear channels, IEEE Trans. Commun. Syst., vol. 11, pp. 360393, Dec. 1963. [3] G. Matz, On non-WSSUS wireless fading channels, IEEE Trans. Wireless Commun., vol. 4, pp. 24652478, 2005. [4] A. Paier et al., Non-WSSUS vehicular channel characterization in highway and urban scenarios at 5.2 GHz using the local scattering function, in Proc. Int. Workshop Smart Antennas, Feb. 2008, pp. 915. [5] Q. Wu, D. Matolak, and I. Sen, 5-ghz-band vehicle-to-vehicle channels: Models for multiple values of channel bandwidth, IEEE Transactions on Vehicular Technology, vol. 59, no. 5, pp. 26202625, 2010. [6] D. W. Matolak, Channel Modeling for Vehicle-To-Vehicle Communications, IEEE Commun. Mag., vol. 46, pp. 7683, May 2008. [7] M. Patzold and B. Hogstad, A wideband mimo channel model derived from the geometric elliptical scattering model, in 3rd International Symposium on Wireless Communication Systems (ISWCS), 2006, pp. 138143. [8] M. Ozcelik, N. Czink, and E. Bonek, What makes a good MIMO channel model, in Proc. Vehicul. Techn. Conf., Jun. 2005, pp. 156160.

You might also like