SAIC Receiver Algorithms for VAMOS Downlink Transmission Mahesh Gupta Vutukuri∗ , Rakesh Malladi∗ , Kiran Kuchi†, R David Koilpillai∗ ∗ Department

of Electrical Engineering, Indian Institute of Technology Madras of Excellence in Wireless Technology, India Email: {maheshgupta444,malladirakesh,kkuchi}@gmail.com † Center

Abstract—Voice services over Adaptive Multiuser channel in One Slot (VAMOS) performance in the presence of GMSK interferer is presented. Widely-Linear (WL) MMSE filtering is used to cancel the co-channel GMSK interferer while performing α−QPSK detection. In this paper, two ways of estimating the Sub Channel Power Imbalance Ratio (SCPIR) are discussed. Due to the colouration of the resultant impairment after filtering, WL metric is proposed which incorporates error covariance between I/Q components in the metric computation. WL RSSE is used as the equalizer. Simulation results using the proposed receiver architecture show significant performance important over conventional VAMOS receivers.

I. I NTRODUCTION GSM/GPRS (Global System for Mobile communications/ General Packet Radio Services) is one of the most widely deployed cellular communication system in the world. The rapid growth of GSM subscriber base has led to an aggressive network expansion. Over the years, GSM has been constantly evolving and the capacity and transmission quality have been improving. Enhanced Data rates for GSM Evolution (EDGE, also known as EGPRS) system employs 8-PSK modulation to improve the capacity. Evolved EDGE which uses higher order modulations like 16-QAM, 32-QAM and turbo codes to increase the capacity, has been standardized. Opting for tighter frequency reuse offers another avenue of increasing the network capacity, however it increases the co-channel interference (CCI). Given the dense deployments, strong CCI is almost unavoidable in current networks. Hence mitigating the effects of CCI is of prime importance. Some of the techniques developed to reduce the effects of CCI are proposed in [1]. Cancellation of CCI is extremely important to improve the network capacity. CCI affects performance both in the downlink and the uplink. In the uplink, the base stations typically use multiple antennas for effective cancellation of interference. However in the downlink, due to cost, design and power consumption factors, having multiple antennas at the mobile station (MS) is difficult. Hence the receiver has to cancel the interference from the signal received via a single antenna [1] and this poses a challenging problem. To this end, substantial research work has been done on Single Antenna Interference Cancellation (SAIC) algorithms and several algorithms have been proposed in the literature [2]–[9]. Systems using SAIC algorithms in the down link The funding for this project was obtained in part from Nokia/Renesas R&D.

have been developed and up to 50% improvement in capacity have been reported in field trials [10]. SAIC algorithms for 8PSK and 16-QAM imply the necessity for complex equalizers. As a consequence Voice Services over Adaptive Multiuser channel in One Slot (VAMOS) (or Multiple Users Reusing One Slot, (MUROS)) is considered in 3GPP [11]. In VAMOS, the capacity can be doubled by deliberately pairing two users in the same time slot on the same frequency resource within a cell. In order to enable sufficiently good user separation, the Orthogonal Sub Channels (OSC) concept has been proposed in [11] for the downlink of VAMOS as signalling scheme among other approaches. In [12], the detection of a VAMOS user in the presence of co-channel interferer is considered where in the intra-channel user is treated as an interferer. Here a Mono Interference Cancellation (MIC) algorithm is used to cancel the interference. Alternatively since both users signals are coming from a single base station, one can employ joint detection instead of treating the intra-channel user as an interferer. Here we propose widely-linear filter based SAIC algorithm that cancels the GMSK modulated co-channel interferer while detecting both users within a cell, which is equivalent to α− QPSK detection in the presence of a GMSK interferer. In [12], the proposed receiver architecture needs to be properly switched between the algorithms according to the given environment. Here we propose a single receiver architecture which can be used in the presence of a GMSK interferer and as well as in its absence. The complex baseband received signal in VAMOS downlink is improper because α−QPSK, like BPSK is not a circularly symmetric constellation. So, widely-linear processing gives better performance than linear processing [13]. WL processing uses the signal and its conjugate [6] or alternatively the real and imaginary parts of the signal [14]. We consider the latter approach in this paper. The outline of the paper is as follows. In Section II, the system model under consideration is presented. In section III, channel estimation approaches are discussed. Section IV presents the SAIC algorithm and the WL equalizer with modified metric. Finally, simulation results are presented in section V.

II. VAMOS S YSTEM M ODEL VAMOS uses Adaptive QPSK (AQPSK) or α-QPSK modulation in which each orthogonal sub-channel is assigned to a different user with in the cell. The Sub Channel Power

cell. w(n) denotes the complex valued white Gaussian noise of variance σn2 . The factor α determines the constellation points and the power difference between the two orthogonal users. We don’t assume the knowledge of α in the receiver. We assume the knowledge of training sequence of both orthogonal users as well as co-channel interferer for the estimation of channel vector. Fig.2 shows the proposed VAMOS receiver structure. III. C HANNEL E STIMATION

Fig. 1.

Here, we observe that both users signals, in principle, propagate through the same channel. If the received symbols corresponding to the time-aligned training sequences of both users are collected in a vector r, this vector can be expressed as

α-QPSK modulation in VAMOS

Imbalance Ratio (χ) is defined as ratio of power of 2nd (quadrature) user (sin2 (α)) to the power of 1st (in-phase) user (cos2 (α)). SCPIR = χ = 20 log10 (tan α) dB. Here we consider the discrete time model of VAMOS receiver which does α-QPSK detection in the presence of a dominant GMSK interferer.

r = cos(α) A1 h + sin(α) A2 h + B1 g + w

(2)

where A1 , A2 and B1 represent ( N − L ) x ( L + 1 ) Toeplitz convolution matrices corresponding to the training sequences of user 1, user 2 and the interferer respectively with T the training sequence length N , and h = [h[0] h[1] ... h[L]] , T and g = [g[0] g[1] ... g[L]] . For simplicity, the factor j in eqn (1) has been absorbed in A2 . A. Joint ML estimation

r[n] RC filtering Rxed signal

Im

Eqn (2) can be rewritten as p r = l S1 Θ + 1 − l 2 S2 Θ + S3 Θ + w (3)   where Θ = [h g]T , S1 = A1 0 and  l = cos(α),  S2 = A2 0 and S3 = 0 B The joint Maximum-likelihood (ML) estimates for Θ and α result from minimizing the L2 −norm of the error vector q ˆ − S3 Θ, ˆ ˆ − 1 − ˆl2 S2 Θ e = r − ˆl S1 Θ

Ree

Re

GMSK De−rotation

WL MMSE Filter

WL Equaliser

Joint Channel SCPIR, Estimation ML Estimation Blind Est... LS channel Est..

Fig. 2.

VAMOS receiver structure

In the considered scenario of VAMOS downlink transmission, the baseband received signal can be written as r [n] = cos(α)

L X

k=0

h(k) a1 (n − k) + j sin(α)

a2 (n − k) +

L X

k=0

L X

ˆ denote the estimated quantities. where ˆl , Θ Differentiating eH e w.r.t. l and Θ and setting the derivative to zero results in the following two conditions for ML estimates of Θ and l. 2la

h(k)

k=0

(1)

g(k) b1 (n − k) + w(n)

Here, the discrete-time channel impulse response h(k) of order L comprises the effects of GMSK modulation, the mobile channel from base station to desired user, receiver filtering and GMSK de-rotation at the receiver. We assume that the channel is constant within a transmission burst (time slot) and varies from burst to burst (block fading). a1 (k) and a2 (k) denote the GMSK transmit symbols of both users. b1 (k) denotes the GMSK transmit symbols of co-channel interferer and g(k) denotes the channel from the base station of neighbouring

=

Θ = where



   1 − 2l2 l b−c √ −d √ − f, (4) 1 − l2 1 − l2
−1 H PH P P r. (5)


H = ΘH SH 1 S1 − S2 S2 Θ   b = rH S1 Θ + ΘH S1 H r ,   c = ΘH S1 H S2 + S2 H S1 Θ,     S d = rH S2 Θ + ΘH S2 H r − ΘH S2 H S3 + SH 3 2 Θ ,   f = ΘH S1 H S3 + SH 3 S1 Θ, p P = l S1 + 1 − l 2 S2 + S3 a

Eqns (4) and (5) may be viewed as ML estimate of l given Θ, and ML estimate of Θ given l respectively. However, it does not seem possible to obtain a closed form solution to l, Θ from the coupled equations, one can solve them iteratively from any initial choice of l. B. Blind estimation The received signal during the training sequence is written as r = l A1 h +

p 1 − l2 A2 h + B1 g + w.

(6)

When the knowledge of l is available,

 √ (7) 1 − l2 A2 B Θ + w = S Θ + w.   Let Φθθ = E ΘΘH ∼ = I2(L+1) be the second order statistics of the channel. The pdf of r given l is given by r=



lA1 +

1 1 ×e fr/l (r) = M × π det(Φrr/l )

  r −rH Φ−1 rr /l

(8)

H

where Φrr/l = S Φθθ S + η IM . The ML estimate of l can be obtained by maximizing the ln(fr/l (r)) ∴ ˆl = arg max {−rH Φ−1 rr/l r − ln(det(Φrr/l ))} l

=⇒ ˆl = arg min { rH Φ−1 rr/l r + ln(det(Φrr/l ))} l

ˆ = Θ

H

S S


−1

SH r.

Here, a one-dimensional search is required which is less complex than the iterative solution of two equations, and the resultant performance is almost the same. When implementing a single dimensional search, the MS can use the knowledge of α in previous frame and can get α of the current frame with out much computation. So, essentially the blind technique involves a one-dimensional search in the first burst of communication and it can be updated to design low complexity receivers. IV. SAIC

L X

k=0

ALGORITHM

h(k) [cos(α) a1 (n − k) + j sin(α) a2 (n − k)] +

L X

k=0

=⇒ rn =

L X

k=0

where

As shown in Fig.2, the vector impairment signal can be suppressed using a 2-D MMSE DFE-feed forward filter. We denote the Fn as matrix 2-D prefilter that is used. This matrix prefilter is designed to shorten the length of channel as well to suppress the interference (should maximize the SINR at the output of prefilter). MMSE based prefilter can be used to solve the problem. For example, [15], [16] explore the channelshortening prefilters for MIMO channels. We use prefilter designed in [15], but it gives a coloured impairment at the output of filter. So we replace the metric calculation in the conventional equalizer with a Widely-Linear (WL) metric, which takes care of error covariance as follows. When signal rn is passed through the 2-D filter (F), the resultant output can be written as Vn =

Nb X

Bk an−k + en ,

(10)

k=0

where en denotes 2×1 impairment vector and Bn is the resultant matrix channel obtained by 2-D filtering. We use Identity Tap Constraint (ITC) as defined in [15] to get monic, causal, minimum-phase resultant matrix channel Bn . Once the resultant matrix channel Bn is known, Reduced State Sequence Estimator (RSSE) [17] can be executed with modified metric (WL-RSSE) using postulates of sub-states (or symbols). In WL-RSSE, we use Ungerboeck set partitioning [18] principles to increase the intra-subset distance. The branch metric computation is given by M

=

e = ˆn V

=

eH R−1 ee e, ˆ n, Vn − V Nb X

˜n−k , Bk a

(11) (12) (13)

k=0

Once the estimates of α and Θ are obtained, Eqn.(1) can be written as follows. r [n] =



   hI (k) −hQ (k) gI (k) , Gk = hQ (k) hI (k) gQ (k)     cos(α) a1 (k) wI (n) ak = , wn = , bk = b1 (k). sin(α) a2 (k) wQ (n) Hk =

Hk an−k +

g(k) b1 (n − k) + w(n)

L X

Gk bn−k + wn ,

k=0

|

{z

impairment vector qn

}

(9)

˜ are the hypotheses of the sub-state sequence (symbol where a sequence) in trellis based equalizers and are previously detected symbols in DFE with B0 = 02×2 . Here the Gaussian assumption of resultant impairment is made. Eqn.(11) shows the difference between WL metric to that of conventional metric in RSSE equalizers. Presence of cochannel interference in the system leads to an Ree , which is not white. The off-diagonal elements are significant as the interference power increases, and these elements indicate the correlation between real and imaginary parts of the impairment which need to be taken into account of while computing the branch metric. V. S IMULATION R ESULTS In this section, we evaluate the performance of VAMOS in GMSK interference with algorithms discussed above. In all

DL MTS−1, TU3iFH

−1

BER of user 1

10

χ=−10 χ=−4 χ=0 χ=4 χ=10

DL MTS−1, TU3iFH

−1

10

BER of user 1

the simulations, the radio channel for the desired signals and interferer is assumed to be a Typical Urban channel with a vehicular speed of 3 kph, i.e., a TU3 channel. We assumed the frequency of operation is 900 MHz, and implemented independent fading for each burst to simulate the effect of ideal Frequency Hopping (iFH). The entire receiver uses 2x oversampling of the received signal. All the equations derived above can be applied even when the receiver does oversampling. We augmented both polyphase channel outputs into a single vector and applied all the equations. We have used a single dominant interferer model with the interferer passing through an independent TU3 fading channel throughout the analysis. Perfect synchronization between desired signal and interferer is assumed. Received signal is passed through a raised cosine filter of roll off factor 0.5. To simulate interference limited system, the noise level is set very low, SNR=30 dB. The performance has been presented in terms of uncoded BER (raw BER) as a function of average signal to interference power ratio. In Fig.3, the raw BER of user-1 versus C / I is presented. Here SCPIR is estimated using ML estimation technique described in Section III-A. The iterative solving of coupled equations is started with α ˆ = π3 and 10 iterations are carried out before converging to final estimates of α ˆ

χ = −10 χ=−4 χ =0 χ=4 χ=10

−2

10

−3

10

−4

−2

0

2

4

6

8

10

C / I [dB] Fig. 4. Raw BER of user-1 versus C / I (C1 + C2 = C) for different blind estimated SCPIR values, WL RSSE with [ 2 2 ] set partitioning.

MMSE filtering is applied to the received signal and WLRSSE is used as equalizer. In the absence of GMSK interferer, MLSE is the most favourable detection scheme. But in the case of external interferer, suppression of CCI has to be carried out and its resultant impairment covariance need to be taken care in the metric computation. Even when there is no interferer, we can still use this receiver architecture without any switching between different architectures because WL MMSE DFE prefilter gives monic, causal, minimum phase channel and the performance loss with WL-RSSE will be minimal w.r.t WL-MLSE. VII. ACKNOWLEDGEMENTS

−2

10

The authors would like to thank Dr. Lars Christensen, Renesas R&D, Denmark and Dr. Jorma Lilleberg, Renesas R&D, Finland for the encouragement and guidance provided for this research work. R EFERENCES

−3

10

−4

−2

0

2 4 C / I [dB]

6

8

10

Fig. 3. Raw BER of user-1 versus C / I (C1 + C2 = C) for different ML estimated SCPIR values, WL RSSE with [ 2 2 ] set partitioning.

In Fig.4, the raw BER of user-1 versus C / I is presented. Here SCPIR is estimated using blind estimation technique described in section III-B. A one dimensional search is implemented using “golden section search ” technique. As χ increases, power of user-1 decreases and hence the performance decreases. We can see that raw BER even in the worst case meets the requirement suggested in [12]. We can see from Figs.3,4, that the loss in performance is negligible. So blind estimation of SCPIR can be employed in the receiver. VI. C ONCLUSION Algorithms for SCPIR estimation and detection based on new WL metric in VAMOS downlink are presented. WL

[1] P. Hoeher, S. Badri-Hoeher, W. Xu, and C. Krakowski, “Single-antenna co-channel interference cancellation for tdma cellular radio systems,” Wireless Communications, IEEE, vol. 12, no. 2, pp. 30 – 37, april 2005. [2] H. Trigui and D. Slock, “Cochannel interference cancellation within the current GSM standard,” in Universal Personal Communications, 1998. ICUPC ’98. IEEE 1998 International Conference on, vol. 1, Oct 1998, pp. 511–515. [3] D. Hui and R. Ramesh, “Maximum likelihood sequence estimation in the presence of constant envelope interference [cellular radio networks],” in Vehicular Technology Conference, 2003. VTC 2003-Fall. 2003 IEEE 58th, vol. 2, Oct. 2003, pp. 1060 – 1064 Vol.2. [4] R. Meyer, W. Gerstacker, R. Schober, and J. Huber, “A single antenna interference cancellation algorithm for GSM,” in Vehicular Technology Conference, 2005. VTC 2005-Spring. 2005 IEEE 61st, vol. 2, May-June 2005, pp. 821 – 825 Vol. 2. [5] P. Chevalier and F. Pipon, “New insights into optimal widely linear array receivers for the demodulation of BPSK, MSK, and GMSK signals corrupted by noncircular interferences-application to SAIC,” Signal Processing, IEEE Transactions on, vol. 54, no. 3, pp. 870 – 883, March 2006. [6] J. Olivier and W. Kleynhans, “Single antenna interference cancellation for synchronised GSM networks using a widely linear receiver,” in IET Commun, vol. 1, 2007, pp. 131–136.

[7] P. A. Hoeher, S. B. Hoeher, and W. Xu, “Single antenna interference cancellation ( SAIC ) for cellular TDMA networks by means of decoupled linear filtering/nonlinear detection,” in Proc. 17th Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC),Helsinki,Finland, September 2006. [8] R. Meyer, W. H. Gerstacker, R. Schober, and J. B. Huber, “A single antenna interference cancellation algorithm for increased GSM capacity,” in IEEE Trans. Wireless Comm., vol. 5, 2006, pp. 1616–1621. [9] K. Kuchi and V. Prabhu, “Interference cancellation enhancement through generalized widely linear equalization in QAM systems,” Wireless Communications, IEEE Transactions on, vol. 8, no. 4, pp. 1585 –1590, april 2009. [10] A. Mostafa, R. Kobylinski, I. Kostanic, and M. Austin, “Single antenna interference cancellation (SAIC) for GSM networks,” in Vehicular Technology Conference, 2003. VTC 2003-Fall. 2003 IEEE 58th, vol. 2, Oct. 2003, pp. 1089 – 1093. [11] Nokia, “Voice capacity evolution with orthogonal subchannels,” in 3GPP TSG GERAN, GERAN #36, Vancouver, Canada, Nov.2007. [12] R. Meyer, W. Gerstacker, F. Obernosterer, M. Ruder, and R. Schober, “Efficient receivers for GSM MUROS downlink transmission,” in Personal, Indoor and Mobile Radio Communications, 2009 IEEE 20th International Symposium on, Sept. 2009, pp. 2399 –2403. [13] B. Picinbono and P. Chevalier, “Widely linear estimation with complex data,” Signal Processing, IEEE Transactions on, vol. 43, no. 8, pp. 2030 –2033, Aug 1995. [14] K. Kuchi and V. Prabhu, “Performance evaluation for widely linear demodulation of PAM/QAM signals in the presence of rayleigh fading and co-channel interference,” Communications, IEEE Transactions on, vol. 57, no. 1, pp. 183 –193, January 2009. [15] N. Al-Dhahir, “Fir channel-shortening equalizers for MIMO ISI channels,” Communications, IEEE Transactions on, vol. 49, no. 2, pp. 213 –218, Feb 2001. [16] A. Hafeez, R. Ramesh, and D. Hui, “Maximum SNR prefiltering for MIMO systems,” in Signal Processing Advances in Wireless Communications, 2005 IEEE 6th Workshop on, June 2005, pp. 186 – 190. [17] M. Eyuboglu and S. Qureshi, “Reduced-state sequence estimation with set partitioning and decision feedback,” Communications, IEEE Transactions on, vol. 36, no. 1, pp. 13 –20, jan 1988. [18] G. Ungerboeck, “Channel coding with multilevel/phase signals,” Information Theory, IEEE Transactions on, vol. 28, no. 1, pp. 55 – 67, Jan 1982.