Anda di halaman 1dari 5

Analysis of SINR for UMTS Rake Receiver-Smart Antenna Processing Using Two Different Modelling Approaches

Eindhoven

M. Jevrosimovi , D. Mati , L. Jorgueski , M.H.A.J. Herben and G. Brussaard c c s


University of Technology, PO Box 513, 5600MB Eindhoven, The Netherlands, e-mails: {m.jevrosimovic, m.h.a.j.herben, g.brussaard}@tue.nl TNO Telecom, PO Box 421, 2260AK Leidschendam, The Netherlands, e-mails: {d.matic, l.jorguseski}@telecom.tno.nl is valid for a micro-cell environment where the height of the base station antenna is much below the average rooftop level. The output is the complete composition of waves in terms of amplitude, delay, phase, angles of arrival at both the side of the mobile and the base station, for the given position of the mobile and that of the base station. The elements composing the wave will be referred to as the ray parameters further in the text. The WDCM model is based on the well-known Geometrically Based Single-Bounce Elliptical Model (GBSBEM) for micro-cells [5], which assumes a uniform distribution of scatterers within an ellipse with the mobile and the base station placed in the foci of the ellipse. The WDCM model additionally assumes the grouping of scatterers into clusters to account for more realistic scenarios where the dispersion of waves is not continuous in time and space. The output of the WDCM is also given in terms of the ray parameters. The model only accounts for a single scattering. In order to model the effect of increasing delay spread due to multiple reections in real scenarios, a parameter called the effective street width is introduced [6]. The paper is organized as follows. The signal model is given in Section II. The combined Rake receiver-smart antenna processing is presented in Section III. Simulation results based on the two models are given in Section IV, and the conclusions are drawn in Section V. II. S IGNAL MODEL The transmitted signal for the desired user is dened as: us (t) = bs (t) ss (t) (1)

Abstract In 3G mobile communication systems, smart antenna processing is seen as a powerful tool to combat directional interference. In this paper we derive the statistics of the signal-tointerference+noise ratio (SINR) achieved by the combined Rake receiver-smart antenna processing for the case where the thermal noise is the only source of interference, and for the case of another interfering user. The results are based on the output of two channel models with disjoint approaches: deterministic raytracing model Fipre and the stochastic Wideband Directional Channel Model (WDCM).

I. I NTRODUCTION The UMTS system level simulator will be an important platform for testing smart antenna adaptive algorithms in the power controlled environment. The dynamical power control is performed on the basis of the estimated instantaneous SINR at the receiver side, which is very much dependent on the signal processing technique, as well as the channel itself. The combined processing of the Rake receiver and smart antenna is common to receiver algorithms for uplink Wideband Code Division Multiple Access (WCDMA) channels, where multipath propagation and interference from other users is very pronounced [1], [2]. The effect of these algorithms on the instantaneous SINR on the link level has to be incorporated into the system level simulator using reliable channel models. The choice of the channel model is a difcult task. Most models accepted within 3GPP are still more based on the channel statistics than on the concrete site-specic information due to difculties in including more deterministic data in a exible way. In this paper we compare the performance achieved by the above mentioned processing scheme for the deterministic ray-tracing model Fipre, developed at the Eindhoven University of Technology (TU/e) [3], and the stochastic WDCM for UMTS, developed at the Technical University of Lisbon [4]. The comparison has been made on the basis of SINR spatial changes with the goal to put a basis of study to incorporate more deterministic data in the stochastic model. Explanations will be given for any possible mismatch. The ray-tracing model Fipre was developed for the microcell environment. It makes use of detailed information about buildings and vegetation in the given environment and accounts for transmission through buildings and scattering from trees in addition to standard propagation mechanisms like reections and diffractions, included in earlier ray-tracing versions. The model neglects over rooftop propagation, which

where bs (t) represents a bipolar information bearing signal (data bit, with Ts symbol duration), and ss (t) is the spreading waveform (pseudorandom sequence with Tc chip duration). The transmitted signal of the interfering user is similarly dened as: uq (t) = bq (t) sq (t) (2)

If we designate all signal and propagation characteristics of the desired user with the index s, and those of the interfering one as q the received signal at the base station is dened as

0-7803-7955-1/03/$17.00 (C) 2003 IEEE

follows:
Ks

r(t)

=
i=1 Kq

i exp(ji )a(i )us (t i ) + j exp(jj )a(j )uq (t j ) + n(t)


j=1

where we refer to hs,l as spatial signature for the l-th resovable multipath (signal captured on the l-th Rake nger), which is dened as: hs,l = (3)
i

i exp(ji )a(i )Rs (i l )

(7)

where i , i , i , i , represent amplitude, phase, time delay, angle of arrival respectively for the i-th propagation path, K represents the total number of dominant propagation paths and n(t) denotes thermal noise. The antenna steering vector a() = [a1 (), a2 (), ..., aM ()] represents an M dimensional vector (assuming M antennas) with elements dened as: am () = exp(j 2 (m 1)d sin()) (4)

where Rs (.) is the autocorrelation of the pseudo-random sequence. The b0 term represents the data bit at the zeroth sampling time and nl represents the noise at the l-th Rake nger. According to MRC, the antenna weight vector is: ws,l = hs,l . The nal output of the Rake receiver is as follows:
L

Y = b0
l=1

|hs,l |2 +

nl (0)
l=1

(8)

where L denotes the number of Rake ngers. The desired signal is, therefore, given as: Us = Eb T s 2
L l=1

where d is the spacing between antenna array elements, and denotes the wavelength. III. R AKE RECEIVER - SMART ANTENNA PROCESSING For the case of at-fading channels, the optimum antenna weights converge to the well-known Wiener solution based on the perfect estimation of the spatial correlation matrix of interference and the antenna steering vector of the desired signal (optimum combining [7]). In a multipath frequency selective CDMA channel, optimum combining is performed on spatial signatures of the received signals. A spatial signature represents an overlay of multipath signals received by an M-dimensional antenna array, which fall into one delay bin determined by the chip length (260 ns for UMTS). In this way, each nger of the Rake receiver is performing optimum combining for the corresponding resolvable path. Antenna processing is thus based on the possibility to separate spatial signatures of the desired and interfering user, rather than on the basis of their angles of arrival [8], [9]. In case that total contribution of interfering signals can be modelled as spatially white noise, optimum combining resorts to maximal ratio combining (MRC) of antenna array signals for the desired user [7], [8]. A. MRC in the presence of thermal noise We rst analyze the case when the spatially white thermal noise is the only interfering source. The output of the l-th Rake nger Yl , after performing despreading and maximum ratio combining (MRC) is given as follows [10]: Yl =
l +Ts l

|hs,l |2

(9)

Here Eb represents the transmitted signal energy per bit. The variance of the noise term for the l-th Rake nger nl is as follows: T s N0 2 |hs,l |2 l = (10) 2 where N0 represents the spectral noise density. The total noise variance is: =
l=1 2 The total achieved signal-to-noise ratio (SNR) = Us / 2 is found to be given by: 2 L 2 l

(11)

Eb N0

L l=1

|hs,l |2

(12)

The term Eb /N0 can be rewritten as follows: Eb P Ts 1 P = N 1 T = N N0 N0 Tc c tot (13)

where P represents the transmitted power, and N = Ts /Tc is the length of the pseudo-random sequence, the so-called processing gain, determined by the particular service [11]. The total noise level Ntot for the system bandwidth of 1/Tc is given as Ntot = N0 /Tc . B. Optimum combining in the presence of an interfering user and thermal noise In the presence of one interfering user (directional interference), the signal at the the output of the l-th Rake nger consists of the desired signal Sl , the interference part Imai,l , and thermal noise part nl , as in: Yl = Sl + Imai,l + nl (14) Therefore, the instantaneous SINR of the l-th path, l , is given by: E{|Sl |2 } (15) l = E{|Imai,l |2 } + E{|nl |2 }

wH r(t)s (t l )dt s,l

(5)

where ws,l is the vector of M antenna weight values, which are dened for the l-th Rake nger. The H operator represents hermitian transpose, and complex conjugate. The output for the l-th Rake nger can be rewritten as: Yl = b0 wH hs,l + nl (0) s,l (6)

0-7803-7955-1/03/$17.00 (C) 2003 IEEE

The operation of averaging, denoted here as E, represents the short term averaging, i.e. averaging during the time the channel remains time-invariant (fading is being tracked) [8]. The average power of the desired signal, as shown in the previous section, is given by: |Sl |2 = Eb T s H |ws,l hs,l |2 2 (16)

BS

Trajectory

The interference part at the l-th Rake nger comes from all resolvable multipaths of the interfering signal (due to multipath, the orthogonality between pseudo-random sequences of users in the system is reduced). The average power of the interference part is given as [12]: |Imai,l |2 = C Eb Ts H |ws,l hq,n |2 2 n=1
L

Fig. 1. Building environment with LOS street trajectory; BS indicates the position of the base station
40

(17)
30 20 instantaneous SINR (dB) 10 0 10 20 30 40 20

1. 2. 3. 4.

where hq,n represents the spatial signature of the interfering user for the n-th resolvable path, dened analogous to (7). The last equation includes the averaged effect of partial crosscorrelations between the pseudo-random sequences of the desired and interfering user in the asynchronous channel, in the form of the correlation parameter C. The problem of using specic sequences is solved by assuming random binary 2 sequences. The correlation parameter is dened as C = 3N [13], [12], where N represents the pseudo-random sequence length. The SINR for the l-th resolvable path can be rewritten as: |wH hs,l |2 s,l l = 1 (18) L 0 |wH n=1 hq,n |2 + Nb |ws,l |2 s,l 3N E Optimum antenna weights for the l-th Rake nger, maximizing l according to the optimum combining Wiener solution, are dened as: ws,l = R1 hs,l in (19)

40

60 80 100 120 Distance between the Bs and Ms (m)

140

160

Fig. 2. Instantaneous SINR as a function of the distance between BS and MS obtained from Fipre 1. no interfering user, one Rake nger, one antenna 2. no interfering user, two Rake ngers, three antenna 3. with interfering user, one Rake nger, one antenna 4. with interfering user, two Rake ngers, three antenna

IV. S IMULATION RESULTS The calculation of SNR and SINR for both models was performed on the basis of Eq. 12 and Eq. 22 respectively. Spatial signatures for the desired and interfering user were computed on the basis of ray parameters, as dened by Eq. 7. The environment chosen as an input for the ray-tracing tool Fipre is the line-of-sight (LOS) street to allow for the fair comparison with the WDCM, which is actually valid only for the LOS scenario. The transmitted frequency was 1.9225 GHz, which represents the uplink UMTS central frequency. Instantaneous SINR was calculated for a large number of positions of the desired user mobile station (MS) on the LOS trajectory depicted in the Fig. 1. The base station (BS) was located at one end of the trajectory. The results obtained from Fipre represent the benchmark for the comparison with WDCM. Since WDCM is the stochastic model, it was impossible to reproduce exactly the same environment used for the Fipre deterministic model. The

where Rin represents the spatial correlation matrix of the total interference, dened as: Rin = (hq,equ )(hq,equ )H + N0 IM Eb (20)

Here IM represents an M M identity matrix. The composite inuence of L resolvable propagation paths of the interfering user to the l-th Rake nger, hq,equ is given as follows: 1 hq,equ = 3N
L

hq,l
l=1

(21)

Having dened the optimum weight values, we can derive the total SINR, , as follows:
L

=
l=1

|hH hq,equ |2 Eb s,l (|hs,l |2 N0 ) 2 N0 Eb + |hq,equ |

(22)

0-7803-7955-1/03/$17.00 (C) 2003 IEEE

1 0.9 0.8 Prob(SINR<Abscissa) 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 40 30 20 10 0 10 SINR (dB) 20 30 40 50 1. 2. 3. 4.
Prob(cos(v )<Abscissa)

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0.2 0.4 0.6 0.8 cos(v )
sq

1. 2.

sq

1.2

1.4

Fig. 3. Cumulative distribution function of SINR obtained from Fipre 1. no interfering user, one Rake nger, one antenna 2. no interfering user, two Rake ngers, three antenna 3. with interfering user, one Rake nger, one antenna 4. with interfering user, two Rake ngers, three antenna
1 0.9 0.8 Prob(SINR<Abscissa) 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 40 30 20 10 0 10 SINR (dB) 20 30 40 50 1. 2. 3. 4.

Fig. 5. Cumulative distribution function of cos(vsi ) 1. average values with upper and lower boundaries obtained from WDCM 2. deterministic values obtained from Fipre

Fig. 4. Cumulative distribution function of average values with upper and lower boundaries for SINR obtained from WDCM 1. no interfering user, one Rake nger, one antenna 2. no interfering user, two Rake ngers, three antenna 3. with interfering user, one Rake nger, one antenna 4. with interfering user, two Rake ngers, three antenna

approximate reproduction of the same environment in the stochastic WDCM was done on the basis of matching the delay and angular spread. These parameters were chosen as a matching criteria being the two most inuential global channel characteristics on the Rake-antenna array processing. The number of Rake ngers chosen was L = 2, because the delay spread found for this LOS trajectory is considerably smaller than the UMTS chip length. Also, the inter-path or self-interference coming from multipath components of the same user was considered negligible due to the difference of around 20 dB between powers received on the two Rake ngers (which is not surprising with the low delay spread). The number of antennas was M = 3 (all antennas were assumed

omnidirectional with vertical polarization), with antenna separation of d = and antenna array orientation orthogonal to 2 the trajectory direction (LOS component direction). The main gain achieved by Rake receiver-antenna processing is expected to stem from the antenna processing only. Instantaneous SINR was calculated for the four processing cases. In the Fig. 2 we see results obtained from Fipre. The rst two cases correspond to results without an interfering user, one representing the reference case, assuming only one Rake nger and one receiving antenna, and the other assuming two Rake ngers and three antennas. The other two cases deal with the presence of an interfering user at the distance of 22.68 m from the BS (the beginning of the trajectory), also comprising the reference case, and 2-Rake ngers-3-antennas processing. The reference cases were calculated to allow the analysis of the gain achieved using Rake-antenna array processing. The cumulative distribution functions of SINR for the four processing cases from Fig. 2 are depicted in the Fig. 3. The curves corresponding to the no-interfering user case are placed in the higher range of SINR values. Tthe actual values of SINR are naturally higher without an interfering user present. From these two curves a gain of around 4.7 dB is visible, which corresponds to the theoretical results of the MRC case using three antenna (or 10 log M in the case of arbitrary number of antennas M .) We can also observe from the two curves placed in the lower range of SINR values, that the gain is somewhat higher when the interfering user is present due to interference suppression and deep fades compensation. It is important to mention that all SINR values in the graphs are normalized with respect to the processing gain (explained in previous section). The processing gain does not affect the SINR distribution. The results for the cumulative functions of SINR for the above mentioned four cases, obtained from the WDCM model, are shown in Fig. 4. Since WDCM is a stochastic model, it produces a variety of channel realizations using the same input

0-7803-7955-1/03/$17.00 (C) 2003 IEEE

parameters [6]. For each case obtained from Fipre we have produced corresponding three curves using WDCM, which represent average SINR values, upper and lower boundaries on the basis of 100 channel realizations. The rst thing that is obviously different compared to the ray-tracing model is an overestimation of the total signal power. In Fig. 4, the values are placed in the higher range than in the ray-tracing model for the reference case without an interfering user. The calibration was correctly reproduced, because the LOS component has the same power in both models for the same distance from the base station. It means that the rest of the power, dispersed by scatterers in the environment, is predicted higher in the WDCM than in this particular ray-tracing scenario. This may be due to the fact that WDCM model does not correctly reproduce shadowing effects caused by the scatterers. In Fig. 4, average values of SNR (without interference) for the reference case and Rake-antenna processing are spaced at around 4.7 dB, which is the nature of the MRC. At the same time, average values for SINR (with interference) for the reference case and Rake-antenna processing show a much higher gain than predicted by the ray-tracing model. It is difcult, however, to give any concrete values for the gain, due to large spread of channel realizations in WDCM for the same input parameters (indicated by upper and lower boundaries). Instead, it is better to look at the range of SINR values as a result of 2 Rake ngers-3 antenna processing with respect to the range of SNR values (without interfering user present). In the ray-tracing model, there is clear distinction between these two ranges (Fig. 3), whereas they nearly overlap in the WDCM. This means that a higher level of interference suppression is predicted by the WDCM (close to the case where only thermal noise is present). Further analysis was made to explain the underlying cause of such an effect. The level of interference suppression is highly dependent on the level of separation between spatial signatures of the desired user and interferer. In the presence of a dominant interferer, as in our case, Eq. (22) can be approximated to (also taking into consideration that only one Rake nger has signicant power, L=1) [8]: = Eb |hs,1 |2 sin2 (vsq ) N0 (23)

V. C ONCLUSION The comparison between deterministic ray-tracing Fipre and stochastic WDCM was based on the SINR analysis of the Rake receiver-smart antenna processing. The basic motivation was to analyze the accuracy of the current stochastic models with respect to system performance prediction by comparison with models that include more deterministic data. There is a mismatch between the deterministic Fipre and stochastic WDCM model for the particular LOS street chosen for the analysis, which is caused by the large difference in predicted level of spatial separation between users. ACKNOWLEDGMENTS This research work has been funded by TNO Telecom within the B4 project. The authors would also like to thank professor Luis M. Correia from the Technical University of Lisbon for providing us with the simulation tool for the WDCM model, developed within IST. Further thanks to J. Gil, F. Cardoso and L. Ferreira, for useful discussions and help about the use of the model. R EFERENCES
[1] P. Rooyen, M. Lotter, and D. Wyk, Space-Time Processing for CDMA Mobile Communications. Kluwer Academic Publishers, 2001. [2] D. Mottier, L. Brunel, P. Pinho, A. Silva, L. Goncalves, A. Morgado, P. Marques, M. Pesce, D. Dahlhaus, J. Gil, J. Mendez, L. Correia, P. Hertach, L. Herault, and F. Bucci, Final Report of the Different Algorithms in terms of Link Performance, IST Asilum Public Deliverable D4.4, 2002. [3] Y. Jong, Measurement and modelling of radiowave propagation in urban microcells, Ph.D. dissertation, Eindhoven University of Technology, May 2001. [4] M. Markes and L. Correia, A Wideband Directional Channel Model for UMTS Micro-cells, 12th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, vol. 1, no. 1, pp. B122 B126, 2001. [5] J. Liberty and T. Rappaport, Smart Antennas for Wireless Communications. Prentice Hall PTR, 1999. [6] M. Marques, A Wideband Directional Model for Micro-Cells in UMTS. MSc thesis, Technical University of Lisbon. [7] L. Godara, Application of Antenna Arrays to Mobile Communications, Part II: Beam-forming and Angle-of-Arrival Considerations, IEEE Proceedings, vol. 85, no. 8, pp. 11951245, 1997. [8] J. Hammerschmidt, Adaptive Space and Space-Time Signal Processing for High-Rate Mobile Data Receivers, Ph.D. dissertation, TUM/LIS, June 2000. [9] T. Pham, Statistical Behaviour and Performance of Adaptive Antennas in Multipath Environments, IEEE Trans. on microwave theory and techniques, vol. 47, no. 6, pp. 727731, 1999. [10] K. Pedersen and P. Mogensen, Performance Comparison of VectorRake Receivers Using Different Combining Schemes and Antenna Topologies, IEEE Trans. Commun., vol. 1, pp. 233237, 1999. [11] H. Holma and A. Toskala, WCDMA for UMTS: radio access for third generation mobile communications. Wiley, 2001. [12] G. Efthymoglou, V. Aalo, and H. Helmken, Performance Analysis of Coherent DS-CDMA Systems in a Nakagami Fading Channel with Arbitrary Parameters, IEEE Transactions on Vehicular Technology, vol. 46, no. 2, pp. 289297, 1997. [13] M. Pursley, Performance Evaluation for Phase-Coded Spread-Spectrum Multiple-Access Communication-Part I: System Analysis, IEEE Trans. Commun., vol. 25, no. 8, pp. 795799, 1977.

Here the parameter vsq is an indicator of the separation between the spatial signature of the desired user hs,1 and that of the interferer hq,equ . It is dened from the equality: cos(vsq ) = |hH hq,equ | s,1 |hs,1 ||hq,equ | (24)

In Figure 5, we see that the difference for this parameter between the two models is very large. For big values of vsq (small cos(vsq ) when sin(vsq ) approaches unity), the Eq. (23) approaches Eq. (12) (MRC, no interferer case) for one active Rake nger, L=1. The explanation of the near-overlap between the interference and no-interference case in WDCM is due to the fact that predicted cos(vsq ) is considerably smaller than what is obtained by the ray-tracing model.

0-7803-7955-1/03/$17.00 (C) 2003 IEEE

Anda mungkin juga menyukai