A Novel SINR Estimation Scheme for WCDMA Receivers

1 A Novel SINR Estimation Scheme for WCDMA Receivers Venkateswara Rao M 1 R. David Koilpillai 2 1 Flextronics Software Systems, Bangalore 2 Department of Electrical Engineering, IIT Madras, Chennai - 36. Venkateswara.Manepalli@flextronicssoftware.com koilpillai@ee.iitm.ac.in ABSTRACT In the downlink, CDMA systems utilize channelization codes to separate users and scrambling codes to distinguish different cells and sectors. In this paper, we present a Signal to Interference plus Noise Ratio (SINR) estimation scheme based on exploiting both the scrambling codes and channelization codes for Wideband CDMA (WCDMA) systems. This scheme is shown to be significantly more accurate than SINR estimation scheme using only the channelization codes. The algorithm is also extended for the SINR estimation for each path in a time-dispersive multipath fading scenario. I. INTRODUCTION Based on various advantages such as high spectral efficiency, single cell-frequency reuse, and receiver performance enhancement by RAKE combining, Direct Sequence Code Division Multiple access (DS-CDMA) has been used as the multiple access technique for 3G cellular systems such as Wideband CDMA (WCDMA) [1] and CDMA 2000. Some of the well-known features of Wideband CDMA system are variable and flexible data rates for different users and services with the use of Orthogonal Variable Spreading Factor (OVSF) code [1], [2]. This requires accurate, real time estimation of channel Signal to Interference plus Noise Ratio (SINR) to adapt data rates effectively for different users and services. A conventional RAKE receiver combines the multipath components so as to get diversity gain in the effective SINR after RAKE combining. There are two strategies available for assigning RAKE fingers to the multipath components [3]. The first strategy assigns the paths with the largest instantaneous amplitudes to the available fingers. The second strategy assigns the paths with the largest average powers to the available fingers. In both cases, the interference contaminating the desired signal is ignored, and hence these conventional strategies suffer from a lower probability of detection of multipath components at low SINR where the actual delays are masked in the interference [3]. Thus a better assignment strategy should consider the SINR in each path as the assignment metric. To this end, a novel per-path SINR estimation technique is required at the receiver. Several techniques are introduced in literature for SINR estimation, the basic idea of SINR estimation schemes is to separate the information component of the signal and the interference and noise part from the received signal. In [4], SIR estimation scheme based on exploiting the channelization codes to extract the information component of the signal from the received signal to estimate the interference and noise part is proposed. This paper exploits both the channelization codes and scrambling codes to improve the performance of SINR estimation. In this method, the SINR of each multipath component can be efficiently estimated. This paper is organized as follows. The system model is given in section II. In section III and IV, we present the improved SINR estimation algorithm and per-path SINR estimation algorithm. Simulation results are given in section V. Section VI concludes the paper. II. SYSTEM MODEL This section models the Wideband CDMA downlink signal in a frequency-selective Rayleigh fading channel. Here, we represent the chip-level based signal model, which will be used for the chip-based SINR estimation. A. Wideband CDMA Channel Model Consider a CDMA communication system with K users. The signature waveform of the k th user during m th symbol period is denoted as q k (t) for k =1,..., K. where q k (t) = N 1 j=0 s k [j]ψ(t jt c ), 0 t<t (1) s k [mn + n] =c k [n]d[n + mn], 0 n N 1 (2) where N is the spreading factor, c k [j] =±1,j =0,..., N 1 is the channelization code assigned to the k th user, d[j] is the cell or sector dependent complex scrambling code having much longer repetition period than the symbol duration, s k [n] is the combination of channelization and scrambling codes, ψ(t) the pulse shaping waveform, and T c (= T/N) the chip duration. Assume each user receives signal through L multipath components. Let h i,j (t) and τ i,j represent the channel response and the propagation delay of j th path of the i th user, respectively. The received signal at the receiver can be denoted as

r(t) = A k b k q k (t)h k,l (t τ k,l )+η(t) (3) k=1 l=1 where A 2 k is the k th user s transmitted signal power, b k ɛ {±1 ± j}/ 2 the symbol transmitted by the k th user, and η(t) the zero mean complex Gaussian noise with power spectral density σ 2. B. The Receiver The received signal r(t) is passed through a matched filter which is a square root raised cosine filter with roll off factor α =0.22 in WCDMA system. Also, we assume that the path delays are multiples of T c, and are known at the receiver. Then the received signal after filtering and sampling can be modeled as a discrete signal as follows y[n] = k=1 l=1 A k b k s k [n d k,l ]h k,l [n d k,l ]+η[n] (4) where y[n] =y[nt c ], d k,l = τ k,l /T c is the normalized path delay, η[n] is the Gaussian noise with zero mean and variance σ 2. Assume the channel condition does not vary significantly over the duration N + D chips, where D is the maximum multipath delay in multiples of T c and N>(D +1). Then we can write the received signal in the vector form y = b k S k h k + η (5) k=1 where y = [y[mn],y[mn +1],..., y[mn + N 1]] T is a observation vector of N chips, h k is the channel response vector of (D +1) chips of the k th user, and η = [η[mn],η[mn +1],..., η[mn + N 1]] T the noise vector. S k is a N (D +1) matrices formed by s k [n] which is a combination of channelization and scrambling codes of the k th user that can be viewed as a time-varying channelization code can be defined as, s k [mn]... s k [mn D] s k [mn +1]... s k [mn +1 D]..... s k [mn + N 1]... s k [mn + N 1 D] III. TOTAL SINR ESTIMATION ALGORITHM For user k, S k is known both at the transmitter and the receiver. From (5) we have y = b k S k h k + b i S i h i + η (6) i k Let w k denote the last two terms of (6), w k = b i S i h i + η (7) i k Note that w k is equal to the Interference plus Noise (I+N). To calculate the power of w k, we first find the left null space of S k, denoted by N(S T k ), let s ɛ N(ST k ) be a vector. Because s T y =0+s T w k, (8) project w k, onto the whole vector space of N(S T k ). Let N(S T k ) be spanned by the orthonormal basis {e 1, e 2,..., e P } where P = N rank(s k ) [5], then we have since E[ e p, w k 2 ]= = i=1 j=1 i=1 e p,i e p,je[w k,i wk,j], e p,i 2 σ(i+n) 2, = Pσ 2 (I+N), (9) e p, y = e p,b k S k h k + w k, = e p,b k S k h k + e p, w k, =0+ e p, w k, = e p, w k. (10) the (I + N) can be estimated by ˆσ 2 I+N = 1 P e p, w k 2 = 1 P and the SINR is given by SINR total = E[ 1 P 1 e p, y 2 (11) N E[ y 2 ] P e p, y 2 ] 1 (12) IV. PER-PATH SINR ESTIMATION ALGORITHM Assuming the multipath delays are known at the receiver. For user k the l th received multipath signal can be separated as y[n] = A k b k h k,l [n d k,l ]s k [n d k,l ] + A k b k h k,j [n d k,j ]s k [n d k,j ] + A i b i h i,j [n d i,j ]s i [n d i,j ] + η[n] (13) and the l th path is synchronized to d k,l, so the synchronized received signal for l th path y l [n] =y[n+d k,l ] can be expressed as y l [n] = A k b k h k,l [n]s k [n] + A k b k h k,j [n d k,j + d k,l ]s k [n d k,j + d k,l ] + A i b i h i,j [n d i,j + d k,l ]s i [n d i,j + d k,l ] + η[n] (14) 2

in vector notation this can be expressed as y l = b k s k h k,l + b k s k h k,j + b i s i h i,j + η (15) where y l = [y l [mn],y l [mn +1],..., y l [mn + N 1]] T is an observation vector of N chips, h i,j is the channel response vector of N chips of the j th path of i th user, and η = [η[mn],η[mn +1],..., η[mn + N 1]] T the noise vector. s k =[s k [mn],s k [mn +1],..., s k [mn + N 1]] T is a N 1 vector formed by the combination of channelization and scrambling codes of the k th user. Let w l denote the last three terms of above equation, i.e., the Interference plus Noise (I + N) l terms present in the l th path of k th user. w k,l = b k s k h k,j + b i s i h i,j + η (16) To calculate the power of w k,l, we first find the left null space of s k, denoted by N(s T k ), let s l ɛ N(s T k ) be a vector. Because s T l y =0+sT l w k,l, (17) project w k,l, onto the whole vector space of N(s T k ). Let N(s T k ) be spanned by the orthonormal basis {e l1, e l2,..., e lp } where P = N 1 [5], then repeating the equations (9),(10),(11), we get the total interference plus noise in the l th path of k th user σ(i+n) 2 l. Then the effective SINR at the l th path of k th user can be given as SINR l = E[ 1 P 1 N E[ y 2 ] P e lp, y 2 ] 1 (18) Per-path SINR estimation described above can also be done using the null space of channelization codes (OVSF) after descrambling the l th multipath component of the received signal y[n] delayed by d k,l, i.e., y[n + d k,l ] with scrambling sequence d[n]. But, the imperfect descrambling operation for smaller spreading factor (N) and fading may result in a bias in SINR estimates. Hence, when N is large null space of channelization codes can be used for per-path SINR estimation to reduce complexity, as the null space is not required to compute for every symbol, and if N is small, null space of the combination of channelization codes and scrambling codes can be used for better estimates. Figs. 3 and 4 shows the performance of per-path SINR estimator when the null space of channelization (OVSF) codes is used against null space of combination of channelization codes and scrambling codes. When spreading factor N is small there may be problems in computing the null space of matrices S k. In such cases the summation of SINR in individual paths can be taken as overall SINR, i.e., SINR total SINR l. (19) l This can also be used when N is high, since, in that case per-path SINR estimation can be done using the null space of channelization codes and reduce the complexity. V. NUMERICAL RESULTS Computer simulations were conducted to evaluate the performance of estimators in frequency selective multipath fading scenario. In our simulation, we considered multiple users with power sharing in the downlink as per the 3GPP WCDMA system specifications [2]. In this simulation we considered 17 users (K = 17) and a pilot channel, the power of the user we are interested is allocated 35% of the total power, pilot channel is allocated 10% of the total power, the other 16 users are modelled as Orthogonal Channel Noise Simulator (OCNS) [6], and is given the rest of the power. The user of interest is assigned a spreading factor of 16 (= N). As per the 3GPP specifications the power control frequency in WCDMA is 1500Hz, i.e., once per slot, so our estimator must be converging within a slot interval with an acceptable amount of error for power adjustment and the estimator should be a causal filter, in order to minimize delay in power adjustment. Fig. 1 shows the average percentage error of the SINR estimator at different mobile speeds. In this simulation there are four (L = 4) multipath components in the channel and the noise level is kept at 20 db below the total power level. We observe that the estimator converges with a reasonable amount of error in a slot interval. Average percentage error (%) 80 70 60 50 40 30 20 10 SF = 128 L = 4, K = 17 + pilot N = 16, V = 120 km/h N = 16, V = 3 km/h N = 128, V = 120 km/h 0 0 50 100 150 200 250 300 350 Symbols Fig. 1. Average percentage error performance of total SINR total estimator for varying velocities and spread factors Fig. 2 shows the performance of the per-path SINR estimator compared with ideal SINR values in which the multipaths are 3dB one below the other, i.e., 0, -3, -6,,-9 (db). From the results in Fig. 2 we can note that the estimated values in an average are close to the ideal values and also we can see the 3dB difference in each of the paths. Figs. 3 and 4 shows the performance of per-path SINR estimation when the null space of channelization codes and null space of combination of channelization codes and scrambling codes are used when N is 16 and 128. From the results it is clearly seen that When N is large (128) both estimates are same and this is not the case when N is 16. Fig. 5 compares the overall estimated SINR and the summation of estimated SINR in all paths. From the results it 3

0 5 N = 16, L = 4, K = 17 + pilot, V = 120km/h Path 1 Ideal SINR Estimated SINR 5 N = 128, L = 4, K = 17 + pilot, V = 120 Km/h RMP (db) = 0 3 6 9 Path 1 Null space of combined codes Null space of OVSF codes 15 15 25 25 30 5 0 5 10 15 20 25 30 35 40 35 5 0 5 10 15 4 6 8 Fig. 2. Performance of per-path SINR l estimator N = 16, L = 4, K = 17 + pilot, V = 120 Km/h Path 1 Null space of combined codes Null space of OVSF codes Fig. 4. Performance of per-path SINR l estimator when null space of channelization codes and the null space of combination of channelization and scrambling codes are used when N is 128 2 4 6 N = 16, 128, L = 4, K = 17 + pilot, V = 120 Km/h N = 16 N = 128 12 14 16 18 [Sum of per path SINR in L paths] 8 12 14 16 22 18 24 5 0 5 10 15 Fig. 3. Performance of per-path SINR l estimator when null space of channelization codes and the null space of combination of channelization and scrambling codes are used when N is 16 22 22 18 16 14 12 8 6 4 [Overall SINR estimated] Fig. 5. Comparison of overall SINR and the summation of estimated SINR in all paths can be observed that both the estimates follow linearly as the spreading factor (N) increases. VI. CONCLUSION In this paper we addressed the total SINR total estimation and per-path SINR l estimation problem for Wideband CDMA receivers. We proposed a SINR estimation technique that exploits both scrambling codes and the channelization codes. This method has also been extended to per-path SINR estimation. This method is shown to perform better than the one using only the channelization codes for SINR estimation, through computer simulations. When the spreading factor is high, a method of reducing the complexity for overall and per-path SINR estimation is also suggested. As the only requirement for this algorithm is the knowledge of the channelization codes and scrambling codes, this SINR estimation method can be used both in uplink and downlink. REFERENCES [1] H. Holma and A. Toskala, WCDMA for UMTS - Radio Access for Third Generation Mobile Communications, John wiley and Sons, Second Edition, 2002. [2] 3GPP, Spreading and Modulation(FDD), 3GPP Technical Specification TS 25.213 V5.5.0, Dec 2000. [3] Mohamed Abou-Khousa, Ali Ghrayeb, and Mohamed El-Tarhuni Signal-to-Interference Ratio Estimation in CDMA Systems, Canadian Conference on Electrical and Computer Engineering, Vol.3, pp.1333-1337, May 2004. [4] Li-Chun Wang and Chih-Wei Wang, A Near Real-Time Signal to Interference Ratio Estimation Technique in A Frequency-Selective Multipath Fading Channel for the WCDMA System, Vehicular Technology Conference, Vol.2, pp.752-756, Oct 2001. [5] Gilbert Strang, Linear Algebra and its Applications, Academic Press. [6] 3GPP, UE Radio Transmission and Reception (FDD), 3GPP Technical Specification TS 25.101 V5.5.0, Dec 2002. [7] K. Balachandran, S. R. Kadaba, and S. Nanda, Channel Quality Estimation and Rate Adaptation for Cellular Mobile Radio, IEEE J. Select. Areas commun, Vol.17, pp.1244-1256, July 1999. 4

[8] M. Turkboylari and G. L. Stuber, An Efficient Algorithm for Estimating the Signal-to-Interference Ratio in TDMA Cellular Systems, IEEE Trans. on Commun, Vol.4, pp.728-731, June 1998. [9] Carmela Cozzo and Gregory E. Bottomley, DS-CDMA SIR Estimation With Bias Removal, IEEE Wireless Communications and Networks Conference (WCNC), Vol.1, pp.239-243, March 2005. 5