Interference Estimation for Multi-Layer MU-MIMO Transmission in LTE-Advanced Systems

Transcription

1 Interference Estimation for Multi-Layer MU-MIMO Transmission in LTE-Advanced Systems Zijian Bai Biljana Badic Stanislaus Iwelski Rajarajan Balraj Tobias Scholand Guido Bruck Peter Jung University of Duisburg-Essen Oststrasse Duisburg Germany Intel Mobile Communications IMC) Düsseldorfer Landstrasse Duisburg Germany Abstract The aim of this paper is to get an insight of the interference estimation for multi-layer multi-user multiple-input and multiple-output ML-MU-MIMO) transmission for LTE-Advanced long term evolution) systems. Different interference-aware receivers have been investigated in ML-MU-MIMO the presence of co-layer intra-cell and inter-cell interferences. User-specific reference signal UE-RS) based channel estimation and various interference covariance estimation schemes in LTE-Advanced systems are investigated together analytically and numerically. Our investigation has shown the significant influence of the interference estimation schemes on the system performance at the link level. Furthermore the simulation results have shown the advanced MMSE receiver together UE-RS based interference estimation as the most robust receiver structure for the signal detection in ML-MU-MIMO transmission. Index Terms LTE-Advanced Multi-Layer MU-MIMO Interference Whitening Covariance Matrix Estimation I. ITRODUCTIO Multi-user Multiple-Input and Multiple-Output MU- MIMO) systems have become promising in the context of achieving high data rates required for long term evolution LTE) standards. MU-MIMO was firstly introduced in LTE Rel-8 up to four transmit antennas and two user equipments UE) each being assigned single layer [] [2]. By introducing user-specific reference signals UE-RS) in Rel-0 MU-MIMO transmission is extended up to four UEs four layers and maximum eight transmit antennas. Furthermore each UE may be scheduled dual layer transmission to support high throughput required by certain services. ence the MU-MIMO in Rel-0 is called multi-layer MU-MIMO ML-MU-MIMO). The main issue of signal detections in ML-MU-MIMO is the presence of different interference sources including co-layer from dual layer transmission) intra-cell from MU-MIMO transsmision) and inter-cell interferences. Similar to the study in Rel-8 MU-MIMO transmission [3] [4] interference-aware receivers are required in ML-MU-MIMO to eliminate the residual spatial interference in order to avoid the error floor in data detections. Typical receivers include interference rejection combiner IRC) minimum mean square error MMSE) [3] or fast maximum likelihood Fast-ML) [5] receivers. Since no explicit signals exist in Rel-0 LTE-Advanced systems to indicate ML-MU-MIMO the challenge in ML-MU-MIMO is to construct the knowledge of interference and to eliminate them by the interferece-aware receivers. Conventional methods of constructing interference knowledge are to detect the interference covariance matrix which can be used by IRC MMSE and Fast-ML receivers. Interference covariance matrix estimation schemes have been investigated in [6] [7]. The theoretical impact of channel and interference covariance estimation errors on the performance of IRC and MMSE receivers has been studied in [6] together the SIR distribution analysis and Gaussian noise based error model. Different procedures of the covariance estimation namely reference signal RS) based and user data based estimations have been summarized in [8] [7] and the references therein. In [9] the performance of using different estimations in LTE systems in a multi-cell interference scenario has been presented. Being different to the previous work in this paper we investigate the performance of receivers interference estimations in ML-MU-MIMO transmission. The intra-cell MU-MIMO interference in ML-MU-MIMO has strong impact on channel estimations and interference covariance estimations due to the non-orthogonal UE-RS in ML-MU-MIMO. This issue has not been addressed by previous studies and will be presented in our work the link level system performance in terms of throughput. By exploiting ML-MU-MIMO transmission different interference estimation schemes and interference-aware receivers we aim at finding out the most efficient and robust receiver structure to facilitate ML- MU-MIMO transmission various interference sources in LTE-Advanced systems. This paper is organized as follows. The system model is introduced in Section II together the receiver structures. In Section III different interference covariance matrix estimation schemes are discussed in addition to the channel estimation. Simulation results of the link level system performance are presented in Section IV. Section V concludes this paper. II. LTE SYSTEM OVERVIEW Consider subcarrier specific ML-MU-MIMO transmission in LTE-Advanced frequency division duplex FDD) systems multi-cell downlink scenario. It is assumed that all eodebs base stations) are mounted T transmit antennas and the target UE is equipped R receive antennas. The

2 received subcarrier specific signal vector at the target UE can be represented by r = G T d T + G I d I + G Ci d Ci + n ) L I = g Ti d T + g Ii d Ii + G Ci d Ci + n where G T = T P T C R L and G I = T P I C R I are the effective and intra-cell interference channel matrix between the serving eodeb and the target UE physical channel matrix T and the corresponding applied precoding matrix P T P I. Both P T and P I are generated at eodeb based on the UE feedback channel information. G Ci is the channel matrix for the inter-cell interference from the i-th interference cell. L [ 2] I [ 2 3] and I Cell are the total number of layers to the target UE layers to the intra-cell interference UE and the total number of interference cells respectively. d T d I and d Ci represent the signal vectors for the target UE and interferences and they are mutually independent unit transmission energy per symbol namely E d d ) = I. n C 0 0 I) represents the AWG vector. A. IRC Receiver in ML-MU-MIMO The IRC receiver in [3] utilizes the knowledge on the covariance matrix of the total interference plus noise and prewhiten the received signals being followed by a maximum ratio combiner MRC). The ideal IRC receiver for the subcarrier specific MIMO system ) in ML-MU-MIMO transmission can be represented by R II = G I G I M = G T R II 2) I Cell + G Ci G Ci + 0I. 3) The IRC maximizes the post signal to noise plus interference ratio SIR) and outperforms the classic MRC. B. Advanced MMSE Receiver in ML-MU-MIMO As the dual layer transmission for the target UE is scheduled namely L = 2 co-layer interference occurs in addition in the received signals. In this case the minimum mean square error MMSE) is desired to detect multi layer target signals. Being different from the conventional single-user MIMO MMSE receiver the advanced MMSE receiver in this paper consider all interference sources. With the full knowledge of channel matrices G T G I and G Ci the advanced MMSE receiver for all signals d T d I and d Ci can be constructed like M MMSE = G all G all G all + 0 I) 4) G all = [G T G I G C G C2... G CI Cell ]. 5) owever the decentralized MMSE receiver can be derived from 4) for the signals in the j-th target layer j { 2} as G T G T + G I G I m MMSEj = g I Cell Tj + G Ci G Ci + 0 I 6) Follow the conversion and matrix inverse lemma used in [0] 6) can be further simplified as m MMSEj = g Tj L i j g Ti g + G Ti I G I G Ci G Ci + 0I +. 7) Extending the results in [0] for dual target layer transmission yields the general decentralized MMSE receiver. One can proof that such general decentralized MMSE receiver consists of step-wise MMSE filters namely the pre-whitening filter on the intra-/inter-cell interference and the MMSE filter on the co-layer interference. With the definition of M W = R 2 II. 8) the general decentralized MMSE receiver is given by M MMSE = I + G I M W M W T) G G I M W M W ) 9) = I + G T R II G T G T R II which is equivalent to 4) and thus has the same performance. C. Fast-ML Receiver in ML-MU-MIMO The Fast-ML receiver is proposed in [5] for the signal detection in single user MIMO transmission for example two transmission layers. It can be extended for the signal detection in ML-MU-MIMO transmission where interference and noise are pre-whitened pre-whitening filter at first. The Fast-ML receiver will then search all possible transmit symbols for d T based on filtered signals and get the d T2 for each hypothesis d T as d T2 = slice ) ) ) g T2 R II g T2 g T2 R II r g d T T where slice represents the hard decision symbol detection. All these obtained symbol pairs dt d ) T2 construct the decision set for the soft bits generation for the channel decoding operations. III. COVARIACE MATRIX ESTIMAITO FOR ITERFERECE ELIMIATIO It is clearly presented that the pre-whitening filter or equivalently the covariance matrix 3) is the key point for all above interference-aware receivers. It is used to eliminate the intra-cell and inter-cell interference in ML-MU-MIMO

3 transmission. In addition a precise target channel estimation g j = 2 is necessary and determine the signal detection Tj performance. A. Channel Estimation based on UE-RS In LTE-Advanced systems the UE-RS is designed exactly for the purpose of channel estimation in ML-MU- MIMO. A hybrid time domain code-division multiplexing CDM achieved by orthogonal cover codes OCC) and frequency-division multiplexing FDM scheme is adopted for the design of UE-RS on different antenna ports []. There are totally 2 resource elements RE) used in two time-domain adjacent resource blocks RB) for UE-RS. To support ML-MU-MIMO transmission up to four UEs and four layers and keep the overhead of UE-RS small the scrambling ID SCID) is introduced in addition to the OCC. With two different SCID and two different OCC four layer transmission is provided on antenna port 7 and 8 the same frequency and time domain resources. The UE-RS in the given positions [] can be represented by x kpnscid = [ a knscid0 w p0... a knscid3 w p3] T k = 2 3; p = 7 8; n SCID = 0. 0) where [w p0... w p3 ] is the OCC sequence n SCID is the SCID [ a knscid0... a knscid3] is the pseudo random sequence and p is the antenna port indicator. The UE-RS from different antenna ports the same SCID are orthogonal. owever it is not the case for the UE- RS different SCIDs. In order to achieve a good quality of channel estimations at the target UE we use a two-step channel estimation scheme all available UE-RS symbols available inside the two adjacent RBs. Assume the received UE-RS in two RBs to be Y = g T a T7 X07 + g T2 a T8 X08 + g I a I7 X7 + g I2 a I8 X8 + Ñ ) Ñ = I Cell G Ci d Ci + n being the sum of inter-cell interference and thermal noise X nscidp = Diag [ x T pn SCID... x T 3pn SCID ]) n SCID = 0 p = 7 8 being the UE-RS and a Tp C 2 a Ip C 2 being the attenuation matrices given by the channel variance in frequency and time domain at different antenna ports. Diag ) is the operator which puts the arguments in diagonal positions of a new matrix. In the first step of channel estimation an orthogonal decoding for the target channel is carried out. With [ ] ) B x p0 B 2 x p0 ˇX 0p = Diag... [ ] B x 3p0 B 2 x 3p0 B = Diag [ 0 0]) B 2 = Diag [0 0 ]) p = 7 8 the decoded channel for the target layer from antenna port 7 is given by ˇY = 2 Y ˇX 07. After that a two-dimension MMSE based channel estimation filter W is applied so that the estimated target channel at antenna port 7 is ĝ T = ˇY vec W ). The operator vec ) turns the input matrix to a vector its all elements. The same procedure can be applied to estimate channel at the antenna port 8 which results to ĝ T2 = 2 Y ˇX 08 vec W ). The final estimated target channel matrix can be represented by Ĝ T = 2 g a ˇ T T7 X 07X 07 W + 2Ñ07 2 g a ˇ T2 T8 X 08X 08 W + 2Ñ08 Ñ 07 = g a T2 T8 X 08 + g I a I7 X7 ˇX07 + g I2 a I8 X8 + Ñ W Ñ 08 = g a T T7 X 07 + g I a I7 X7 ˇX08 + g I2 a I8 X8 + Ñ W. B. CRS Based Covariance Estimation 2) 3) To estimate the covariance matrix of the interference and noise 3) a CRS based estimation scheme has been presented in [8] and the references therein. The key idea is to use the CRS subcarriers to estimate the covariance matrix of inter-cell interference and noise. Assume the received CRS signals is y RSkl = h RSkl x RSkl + Ǧ Ci ď Ci + n 4) x RSkl being the deployed CRS in LTE-Advanced systems. The estimation of R II is given by ) R II = y RSkl ĥrskl x RSkl ) S 5) kl) S y RSkl ĥrskl x RSkl S being the set of RE for CRS transmission and ĥrskl being the estimated channel vector on the given CRS. Due to the spatial orthogonal CRS design the estimation of the covariance matrix including intra-cell MU interference is impossible in this scheme. Since arbitrary transmission modes may be used in interference cells Ǧ Ci might be unequal to G Ci. This means that the additional precoding matrix estimation is required to get the real interference channels. C. UE-RS Based Covariance Matrix Estimation Another covariance matrix estimation scheme in ML-MU- MIMO transmission is based on UE-RS. Consider having received UE-RS in ) and estimated target channel matrix 2) the estimated R II on UE-RS can be obtained by R II = 2 2 [ ] y i [ X07 ĜT [ y i [ X07 ĜT ] ]) [ X08 ] ]) [ X08 ] 6)

4 TABLE I DOWLIK LTE PARAMETERS Parameters Setting Carrier frequency 2 Gz Bandwidth / FFT Size 0 Mz / active subcarriers) tx rx Single target layer: 4 2 Dual target layers: 8 4 Antenna array at eodeb Signle target layer: ULA high and medium correlation Dual target layers: Cross-polarized low correlation Receiver type IRC Advanced MMSE Fast-ML Target channel estimation Ideal channel knowledge ICK) UE-RS based estimation UE-RS Est) MU-MIMO precoding codebook precoding wideband PMI UE Pairing PMI orthogonal pairing Channel models LTE Channel Model [2]: EVA-5 EPA-5 Inter-Cell interference One interfereing cell I/ = 0 db Fig QAM MMSEDMRS whiten ICK) 0.3 MMSEDMRS whiten UE-RS Est) MMSECRS whiten UE-RS Est) Performance in medium correlated EVA 4 2 single target layer Fig.. 6 QAM MMSEDMRS whiten ICK) MMSEDMRS whiten UE-RS Est) MMSECRS whiten UE-RS Est) Performance in high correlated EVA 4 2 single target layer y i being the i-th column of Y in ). When channels from target layers are equal in neighbour REs the OCCs keep orthogonal between antenna ports same SCID. Then the first term in 3) vanishes and the other terms become small due to the pseudo orthogonal between UE- RS different SCIDs. In this case 6) represents a good approximation of the ideal covariance matrix 3). In contrast to the CRS based covariance matrix estimation the UE-RS based covariance matrix estimation includes intra-cell MU interference source together inter-cell interference and noise. As mentioned in [7] this estimation scheme also avoids the mismatch of target channel matrix and residual crosscovariance coefficients between target signals and interference signals. IV. SIMULATIOS Following the discussed receivers and covariance matrix estimations numerical simulations are carried out a link level simulator. With these simulations the performance of combinations of receiver structures covariance estimations in ML-MU-MIMO have been investigated. Downlink parameters in LTE-Advanced systems are used in simulations and summarized in Table I. Simulation results of ML-MU-MIMO a single target layer and 3 additional interference layers in the target cell are depicted in Figure and Figure 2. Receiver performance is given in terms of fraction of maximum throughput. Figure shows the performance in a high spatial correlated scenario two modulation and code schemes MCS) 6-QAM /2 code rate and 64-QAM /2 code rate. The MMSE receiver is considered here as the receiver structure both discussed covariance estimation schemes. In the high spatial correlation scenario the correlation between UE channels is quite small so that the precoding operations carried out at the eodeb can separate the UE-specific signals very well leading to small residual intra-cell interferences namely small G I. ence no significant difference is observed between CRS and UE-RS based covariance matrix estimations. Since G I is small 5) and 6) are similar and yield same performance in both MCS. For a benchmark purpose performance of MMSE receiver ICK is plotted in Figure. About db difference at 90% throughput between ICK and UE-RS Est results are observed in both MCS which indicates good performance of the UE-RS channel estimation. Figure 2 shows the results same settings as in Figure but in a medium spatial correlation case. Due to the increased residual intra-cell interference caused by less spatial separation via UE pairing and precoding operations the complete system performance degrades compared results in Figure. The channel estimation is more noisy due to relative stronger residual MU interference in the medium correlation scenario. In Figure 2 CRS based covariance matrix degrades significantly due to the lack of estimating intra-cell interference where G I G I is not a small term any longer. UE-RS based covariance estimation suffers the noisy channel estimations as well. owever its performance is close to the case of ICK which indicates better accuracy of estimated covariance compared the CRS-based scheme. Figure 3 illustrates simulation results dual target layer transmission associated two interference layers in a low spatial correlation scenario. Both advanced MMSE and Fast-ML receiver are evaluated three different MCS QPSK /3 code rate 6-QAM /2 code rate and

5 QPSK CR /3 6 QAM MMSEDMRS whiten) fastmlddmrs whiten) QPSK CR /3 6 QAM MMSECRS whiten) fastmldcrs whiten) Fig. 3. Performance in low correlated EPA 8 4 dual target layer 64-QAM /2 code rate. Simulation results show that CRS covariance estimation scheme faces strong performance degradation and error floor. With 6-QAM and 64-QAM MCS no more than 30% throughput can be achieved. In contrast receivers UE-RS based covariance estimation schemes perform significantly better. In ML-MU-MIMO transmission both target layers are strongly disturbed by the intra-cell MU interfernce in a low spatial correlation scenario. Therefore prewhitening both interference layers is essential for the further co-layer spatial multiplexing signal detections. Since the UE-RS scheme provides more complete and precious covariance than the CRS scheme it has the better performance. There is about to 2 db difference at 90% throughput between advanced MMSE and Fast-ML receivers UE-RS covariance estimation scheme. With the CRS-based scheme Fast-ML receiver is slightly worse than advanced MMSE receiver. This is caused by the corrupted modelling of noise plus interference after whitening filters in 5). V. COCLUSIO In this paper the ML-MU-MIMO transmission in downlink LTE-Advanced systems has been investigated. Different interference-aware receiver structures and interference-noise covariance matrices have been discussed and their performance are analysed and illustrated. The results have shown that the covariance matrix estimation is the key part for the ML-MU- MIMO transmission for any type of interference-aware receivers. The CRS based covariance matrix estimation scheme suffers performance degradation due to estimation inaccuracy in low and medium spatial correlation scenarios whereas UE- RS scheme yields stable performance in different channel conditions. Additionally the advanced MMSE receiver performs close to the Fast-ML receiver lower complexity. Based on our investigation results the advanced MMSE receiver together UE-RS covariance matrix estimation is the most robust and efficient solution for ML-MU-MIMO transmission. the scope of the SAMURAI project which has been partially funded by the European Commission under FP7. REFERECES [] 3GPPTSG-RAWG Evolved Universal Terrestrial Radio Access E- UTRA); Physical layer procedures 3GPP Sophia Antipolis Technical Specification v8.8.0 Sep [2] Evolved Universal Terrestrial Radio Access E-UTRA); Physical layer procedures 3GPP Sophia Antipolis Technical Specification v0.2.0 Sep. 20. [3] Z. Bai B. Badic S. Iwelski T. Scholand R. Balraj G.. Bruck and P. Jung On the receiver performance in MU-MIMO transmission in LTE in Proceeding of the Seventh International Conference on Wireless and Mobile Communications ICWMC 20 Jun. 20 pp [4] J. Duplicy B. Badic R. Balraj R. Ghaffar P. orvath F. Kaltenberger R. Knopp I. Z. Kovacs. T. guyen D. Tandur and G. Vivier MU- MIMO in LTE Systems EURASIP Journal on Wireless Communications and etworking vol. 20 pp [5] Y. Lomnitz and D. Andelman Efficient maximum likelihood detector for MIMO systems small number of streams Electronics Letters vol. 43 no. 22 pp. 2 Oct [6] L. Thiele M. Schellmann T. Wirth and V. Jungnickel On the value of synchronous downlink MIMO-OFDMA systems linear equalizers in Wireless Communication Systems ISWCS 08. IEEE International Symposium on Oct pp [7] TT-DOCOMO Tp for enhanced performance requirements for lte ue si TT DOCOMO Approval R Oct [8] Reference receiver structure for interference mitigation on enhanced performance requirement for LTE UE TT DOCOMO Approval R4 523 Oct. 20. [9] Y. Ohwatari. Miki T. Asai T. Abe and. Taoka Performance of advanced receiver employing interference rejection combining to suppress inter-cell interference in LTE-Advanced downlink in Vehicular Technology Conference VTC Fall) 20 IEEE Sep. 20 pp. 7. [0] Z. Bai B. Badic S. Iwelski T. Scholand R. Balraj G. Bruck and P. Jung On the equivalence of mmse and irc receiver in mu-mimo systems vol. 5 no. 2 pp [] 3GPPTSG-RAWG Physical channels and mapping of transport channels onto physical channels FDD) 3GPP Sophia Antipolis Technical Specification 25.2 v.0.0 Dec. 20. [2] Evolved Universal Terrestrial Radio Access E-UTRA); User Equipment UE) radio transmission and reception 3GPP Sophia Antipolis Technical Specification 36.0 v0.3.0 Oct. 20. ACKOWLEDGMET The authors wish to gratefully acknowledge the support of their colleagues at Intel Mobile Communications and at the Department of Communication Technologies of University of Duisburg-Essen. Parts of this work are carried out in