A NOVEL DECENTRALIZED MIMO-OFDM UPLINK DETECTION SCHEME. Andreas Ahrens, Xinning Wei, Tobias Weber, Shiyang Deng

Transcription

1 A NOVEL DECENTRALIZED MIMO-OFDM ULINK DETECTION SCHEME Andreas Ahrens, Xinning Wei, Tobias Weber, Shiyang Deng University of Rostock Institte of Commnications ABSTRACT Decentralized interference cancellation in MIMO-OFDM orthogonal freqency division mltiplexing) systems can be considered as a promising approach in next generation wireless systems. Considering the entirety of the antennas of all mobile terminals at one end and the antennas of the access points s) at the other end of the commnication link, state of the art interference cancellation is based on a central signal processing nit, e. g. a central nit CU), where joint detection can be applied in the plink UL) and joint transmission in the downlink DL), respectively. Unfortnately sch setps reqire cost-intensive optical fibers or point-to-point radio links in order to deliver all the reqired information to the CU. Therefore decentralized, cost-efficient soltions, which by-pass the CU, are of common interest. In this contribtion a novel decentralized plink detection scheme for MIMO- OFDM systems is presented and evalated nder real channel conditions.. INTRODUCTION Mltiple-Inpt Mltiple-Otpt MIMO) systems are capable of increasing the achievable capacity and integrity of wireless systems and hence, they may be expected to form an integral part of next generation wireless systems. Classic infrastrctrebased wireless networks, sch as celllar systems or wireless Local Area Networks LANs) have attracted a lot of research and have reached a state of matrity. By contrast, despite decades of research, the family of networks operating withot an infrastrctre-based network, sch as ad hoc wireless networks, reqire sbstantial frther research. Celllar systems constitte a specific example of an infrastrctre-based network, where s distribted over a given geographic area, provide access for mobile terminals with the aid of a central signal processing nit. However sch soltions are costintensive and soltions are of common interest that by-pass the central signal processing nit. The athors are indebted to the Detsche Forschngsgemeinschaft DFG) for sponsoring this work in the framework of the TakeOFDM project. Frthermore they grateflly appreciate the fritfl exchange of ideas with their project partners rof. Klein from Technical University of Darmstadt and rof. Rohling from Technical University of Hambrg-Harbrg. MIMO-OFDM systems can be considered as a promising techniqe in Next-Generation Wireless Systems based on their ability to establish a reliable, cost-efficient data commnication []. Different proposals for MIMO-OFDM systems are known in the literatre, where state of the art interference cancellation is based on joint signal processing in the CU [2 5]. Unfortnately, sch setps reqire cost-intensive optical fibers or point-to-point radio links in order to deliver all the reqired information to the CU. Therefore soltions which by-pass the CU are of common interest [6]. In this contribtion a novel decentralized plink detection scheme, based on a distribted signal processing at the s, is proposed. A prereqisite for this kind of signal processing are efficient commnication links between neighboring s. The remaining parts of this contribtion are organized as follows: Section 2 introdces the system model and state of the art interference cancellation schemes are briefly reviewed. The novel decentralized plink detection scheme is introdced in section 3, whereas in section 4 the obtained reslts are presented and discssed. Finally, section 5 provides some conclding remarks. 2. SYSTEM MODEL AND STATE OF THE ART INTERFERENCE CANCELLATION In the following, a mltiser MIMO-OFDM system is considered and the time-discrete eqivalent low-pass representation of signals is chosen. Conseqently, signals are represented by complex vectors and matrices, which are printed in bold face. In the investigated scenario K A s are at fixed locations and K M s are simltaneosly active. In general, the nmber of s, simltaneosly spported on each sbcarrier withot interference from each other is limited by the nmber of antennas sed at the s when Zero-Forcing ZF) detection is applied. This limitation can simply be abolished by sing more antennas, at both, the transmitter and receiver sides in order to increase the available degrees of freedom [7]. Frthermore, the combination with other mltiple access techniqes, sch as TDMA time division mltiple access) or FDMA freqency division mltiple access) is feasible.

2 Sfrag replacements De to the application of OFDM a sbcarrierwise modelling is feasible [2]. The data symbols of the s, which are transmitted simltaneosly over the same sbcarrier, can be stacked in a vector and reslt in d = d ), d2),, dkm) ) T. ) Throghot this contribtion, SK phase shift keying) modlation is assmed. OFDM leads to a non-freqency selective channel per sbcarrier, described by a complex channel coefficient. Extending these considerations to a mltiser MIMO-OFDM system, the following sbcarrier specific system matrix can be obtained H = H,)... H,KM).. H KA,)... H KA,KM), 2) where the vale H ka,km) denotes the sbcarrier specific channel transfer fnction between the k A -th and the k M -th in case of a single antenna per [8]. The channel transfer fnctions are characterized by a path loss model with a path loss exponent of 4. and Rayleigh fast fading. In a centralized system, the signals are received by the s and collected at the CU. Let the received vector e be expressed as e = e ), e2) ) T,, eka). 3) Then, the sbcarrierwise representation leads to e = H d + n, 4) where the noise vector n is defined as follows n = n ), n2) ) T,, nka). 5) The noise n is assmed to be white with a variance of σ 2 /2 for both the real and imaginary parts. The interference between the different data streams, which is introdced by the non-diagonal channel matrix H, reqires appropriate data detection strategies. In a centralized system, interference cancellation is based on a central signal processing nit, e. g. a CU, where joint detection JD) can be applied in the UL and joint transmission JT) in the DL, respectively [9, ]. oplar techniqes are based on IC or SIC parallel or serial interference cancellation), which have attracted a lot of attention within the last years [2, 3, ]. The general strctre of a IC scheme in the UL is shown in Fig.. The matched filter MF) incldes several correlators to match the corresponding channel coefficients and can be defined as follows )) V f = diag H T H H T. 6) z e r d p) ˆd p) V f - qantization V r Fig.. UL interference cancellation ˆd p ) ) The expression diag H T H retrns a diagonal matrix ) with the elements of the sqare matrix H T H on its diagonal. The MF delivers a biased estimation vector r = V f e = V f H d + V f n. 7) Rearranging eqation 7) leads to r = d + V f H I) d + V f n, 8) with I describing the identity matrix. Analyzing 8) it is obvios that the remaining interferences can be removed by a matrix V r, which has to be defined as follows Finally, the vector d p) is given by V r = V f H I. 9) d p) = r V r ˆd p ). ) The hard or soft decision of d p) reslts in ˆd p) and can be sed in the next stage to otperform the crrent detection. Nonetheless, it is worth mentioning that the proposed strctre can also be applied withot qantization. Assming convergence and an infinite nmber of iterations p, i. e. asymptotically, eqation ) corresponds to the ZF soltion and reslts in d ) = ) H T H H T e. ) 3. NOVEL ULINK DETECTION SCHEMES 3.. State of the Art Interference Cancellation Assming SK modlation, the normalization contained in 6) simplifies and the matrix V f reslts in V f = H T. 2) For the matrix V r defined in 9) the following soltion can be fond V r = V f H diag V f H ). 3)

3 Using 2) and 3), the estimate of the k M -th ser signal in the p-th iteration, described by the k M -th element of the data vector d, reslts in d k M) p) = K A k A = H k A,k M ) ek A) K M H k A,k) k= k k M Sfrag ˆd p replacements ). 4) In the first detection stage, i. e., p =, only a coarse estimate can be achieved de to ˆd ) =, i. e., no knowledge abot the interferers is available. In this case, the detection of the k M -th ser signal takes the interferences from the other s into accont. In the following stages, this inflence can be approximately eliminated from the received signal e ka) ˆd at the k A -th sing detection reslts p ) from the preceding stage. In general, each contribtes to the detection of the k M -th ser signal. The complex conjgate mltiplication by H ka,km) in 4) allows a coherent smmation of the detection reslts for the k M -th ser signal. The hard or soft decision of d p) reslts in ˆd p) and can be sed in the next stage to otperform the crrent detection. In general the above described algorithms are performed at a central processing nit [2, 3] Decentralized Interference Cancellation The drawback of a centralized system, sing a CU, can be avoided as a main featre of the proposed scheme. As shown in 4), at least theoretically, all s contribte to the detection of the k M -th ser signal. In a decentralized system, the matched filter estimate r = k A= H ka,km), 5) mst be separated into its specific contribtions and leads to with r = r ka,km) k A= r ka,km), 6) = H ka,km). 7) The weighting sing H ka,km) has to be performed at each k A for each ser signal k M. Separating d p) into its specific ser contribtions leads to K M p) = r k A,k M ) H k A,k M ) H k A,k) ˆd p ). k= k k M 8) From the specific matched filter estimates for the k M -th ser signal r ka,km) the interferences introdced by the other s have to be removed. s s s s significant s relevant interferers estimating Fig. 2. Relationship between significant s and relevant interferers in an exemplarily considered scenario ˆd ) The novelty of the proposed decentralized UL detection scheme reslts from the point that only local channel state information is necessary to estimate the specific ser contribtions. As shown in 8), the processing of the specific data estimates d p) at the k A -th reqires only ka,km) local channel state information, e. g., H ka,km) for k M =, 2,..., K M, which is available at the k A -th. Taking the networking between the s into accont, an improved data estimate can be obtained based on an exchange of specific ser reslts d p) = k A= d ka,km) p), 9) which describes the novelty of the proposed decentralized system concept. Based on 7) a coherent sperposition of the different signal parts stemming from the same mobile is possible. From a practical point of view 8) can be simplified taking only dominant interferers into accont as it is exemplarily highlighted in Fig. 2. In this case, eqation 8) is simplified to p) = H k A,k M ) ek A) relevant interferers k H k A,k) ˆd p ). 2) Frthermore only significant signal parts have to be taken into accont Fig. 2). This means that not all ser signals have to be considered at each. Only few dominant neighboring s need to be inclded when doing the matched filtering. Eqation 9) reslts in d k M) p) = significant neighboring) s k A p). 2) Eqations 2) and 2) show the theoretical basis of the decentralized plink detection scheme for mltipoint-to-mltipoint OFDM systems. The proposed scheme reqires only local channel state information at the s. In comparison to a centralized system, here no channel state information has to

4 .8 Fig. 3. Investigated scenario consisting of 2 cells, 2 s and 2 s Sfrag replacements fll, MF individal, MF partial, p = partial, p = Fig. 4. UL CCDF of the BER distribtion b withot qantization and one antenna per be exchanged between the s. This will redce the complexity of the proposed algorithm significantly. Only preliminary detection reslts have to be exchanged between neighboring) s. 4. RESULTS In order to assess the performance of the proposed distribted system architectre with partial cooperation, a scenario consisting of 2 cells, 2 s and 2 s is considered as shown in Fig. 3. One is randomly located in each cell with niform distribtion. The channel transfer fnctions between the s and s are characterized by a path loss model with a path loss exponent of 4. and Rayleigh fast fading [2]. Frthermore, QSK modlation is chosen. In order to evalate the BER characteristic properly, the CCDF complementary cmlative distribtion fnction) is sed, whereby a prely interference limited system is considered, i. e., the thermal noise is ignored. The simlation reslts are shown in Fig. 4, 5 and 6, respectively. Applying fll cooperation, MF describes the first iteration of IC. The k M -th ser signal, described by the k M -th element of the data vector d, reslts in d p) = k A= Sfrag replacements H ka,km). 22) In comparison to fll cooperation, individal MF describes the operation at the corresponding for the in the same cell, i.e., it corresponds to a conventional celllar system with no cooperation among cells. The BER CCDFs show a speriority of the individal MF compared to the fll MF. The reason for this behavior can be jstified by the nconsidered interferences from the other s. artial cooperation reqires a cooperation between the s in order to bypass the CU. Mostly, not all s are involved in fll, MF individal, MF partial, p = partial, p = Fig. 5. UL CCDF of the BER distribtion b withot qantization and two antennas per the specific data detection process [3]. Ignoring fast fading, the sqares of the channel transfer fnction amplitdes between the s and s decay with a power of 4. of the distances. This implies that the received energy at an is mainly contribted by the s close to the. Therefore we can conclde that only a few s have to exchange preliminary specific data detection reslts and therefore partial cooperation reqires only a cooperation between neighboring s with significant specific channels. The BER distribtions with partial cooperation are depicted in Fig. 4, 5 and 6, whereby the nmber of considered s per was limited to two. From the theoretical point of view it can be conclded, that nder certain circmstances, e. g., the received power at the stemming from different s is in the same range,

5 g replacements fll, MF individal, MF partial, p = and MC-CDMA for Broadband Mlti-ser Commnications WLANs and Broadcasting, Wiley, West Sssex, 23. [3]. Vandenameele, L. Van Der erre, and M. G. E. Engels, Space Division Mltiple Access for Wireless Local Area Networks, Klwer, Boston, 2. [4] T. Weber, A. Sklavos, Y. Li, and M. Weckerle, The Air Interface Concept JOINT for Beyond 3G Mobile Radio Networks, in 5th International Conference on Wireless Commnications, Jly 23, vol., pp Fig. 6. UL CCDF of the BER distribtion b withot qantization and three antennas per no reliable estimation can be performed. In order to overcome this limitation, the degrees of freedom shold be frther increased, e. g., the s can be eqipped with more than a single antenna or a more advanced mlti-ser detection is necessary [2, 4]. Therefore we have assmed that each is eqipped with two or three antennas. The arising performance improvements can be seen in Fig. 5 and 6. For a high nmber of iterations partial cooperation is able to achieve a good compromise between performance and complexity. 5. CONCLUSIONS A novel decentralized plink detection scheme for mltiser MIMO-OFDM systems was presented and evalated nder real channel conditions. A prereqisite for the proposed distribted signal processing are efficient commnication links between neighboring s. These cold be based on wires, optical fibers or point-to-point radio links. Frthermore, or proposed architectre reqires no central nit and only local commnication and local channel state information are reqired. The reslts have shown that decentralized, cost-efficient soltions, which by-pass the CU, are possible and seem to be a promising approach in next generation wireless systems. 6. REFERENCES [] H. Sampath, S. Talwar, J. Tellado, V. Erceg, and A. alraj, A Forth-Generation MIMO-OFDM Broadband Wireless System: Design, erformance, and Field Trial Reslts, IEEE Commnications Magazine, vol. 4, no. 9, pp , September 22. [2] L. Hanzo, M. Münster, B. J. Choi, and T. Keller, OFDM [5]. Zhang, X. Tao, J. Zhang, Y. Wang, L. Li, and Y. Wang, A Vision from the Ftre: Beyond 3G TDD, IEEE Commnications Magazine, vol. 43, no., pp , Janary 25. [6] T. Weber, A. Ahrens, and S. Deng, Decentralized Interference Cancellation in Mobile Radio Networks, in IEEE Wireless Commnications and Networking Conference WCNC), Hong Kong, March 27. [7] D. Tse and. Viswanath, Fndamentals of Wireless Commnication, University ress, Cambridge, 25. [8] M. Merer,. W. Baier, T. Weber, Y. L, and A. apathanassio, Joint Transmission: An Advantageos Downlink Concept for CDMA Mobile Radio Systems sing Time Division Dplexing, Electronics Letters, vol. 36, no., pp. 9 9, May 2. [9] T. Weber, I. Maniatis, A. Sklavos, and Y. Li, Joint Transmission and Detection Integrated Network JOINT), a Generic roposal for Beyond 3G Systems, in International Conference on Telecommnications, Beijing, Jne 22, vol. 3, pp [] T. Weber and M. Merer, Optimm Joint Transmission: otentials and Dalities, in International Symposim on Wireless ersonal Mltimedia Commnications WMC), Yokoska, Oct. 23, vol., pp [] J. G. Andrews, Interference Cancellation for Celllar Systems: A Contemporary Overview, IEEE Wireless Commnications, vol. 2, no. 2, pp. 9 29, April 25. [2] B. Sklar, Digital Commnications Fndamentals and Applications, rentice Hall, Upper Saddle River, New Jersey, 2 edition, 2. [3] S. Deng, T. Weber, and A. Ahrens, Interference Statistics in the Downlink of Service Area based Mobile Radio Networks., in th International OFDM-Workshop, Hambrg, Agst 26, pp [4] S. Verdú, Mltiser Detection, University ress, Cambridge, 998.