Radio Resources Management For The Downlink Of Large-scale Distributted Antenna Systems

Large-scale distributed antenna system (LDAS) is a new communication architec-ture that connects tens or hundreds of distributed remote antenna units (RAU) to a centralunit, via wide-bandwidth and low-loss cabling technology, for a centralized signal pro-cessing. Compared with the conventional collocated multiple antenna systems, LDASprovides efficient utilization of space, frequency and power resources, thus spreading thetransmission power more evenly in the coverage area, reducing the probability of shad-owing and providing a macroscopic spatial diversity gain. In this paper, we focus on theradio resource management for the downlink transmission of LDAS, taking advantage ofLDAS by designing high performance and low complexity radio resource managementalgorithms for the two different transmission mode respectively.For the downlink unicast transmission, LDAS employs multi-user multiple-inputmultiple-output(MU-MIMO)transmissiontoexploitthemassivespacedegreeoffreedomby serving multiple users simultaneously. MU-MIMO requires channel state informationat transmitter (CSIT) and multi-user precoding to transmit different data streams to mul-tiple users simultaneously in the same frequency. A fundamental question arised hereinis spatial scheduling, i.e., determining the optimal subset of users which optimizes prese-lected objective function, such as the sum rate, for given precoding scheme and CSIT. Inthis paper, we have designed several user selection algorithms with high performance andlowcomplexityforMU-MIIMOwithzero-forcingbeamforming(ZFBF)linearprecoding.Firstly, targeted on increasing the sum rate performance, we proposed a new user se-lection algorithm that approaches global optimum sum rate with low complexity by solv-ing two main flaws that exist in traditional algorithms, i.e., the ‘redundant users’ problemand the ‘local optimum’ problem. The new algorithm is called greedy user selection withswap (GUSS). GUSS initializes with an empty set, and then updates the selected user setiteratively with ‘add’,‘delete’ and ‘one-for-one swap’ operations until the sum rate drops.Since GUSS eliminating all the redundant users through ‘delete’ operation and escapingfrom the local optima through ‘one-for-one swap’ operation, it provides significant highersum rate than zero-forcing with selection (ZFS) algorithm, which is the second best userselection algorithm in literature so far with sum rate performance only lower than exhaus-tive search. Supposing the number of RAUs in LDAS is M and the number of users is K,theLDASsupportsequaltoorlessthanM userssimultaneouslywhentheZFBFlinearpre-codingisutilizedforMU-MIMOtransmission. Simulationresultsindicatethatonaverage GUSS achieves99.3percent of the sum rate upper bound that is achieved by exhaustivesea(rch, over the range of transmit signal-to-noise ratios considered with the complexity ofO(a+12)KM3). The parameter a is the number of iterations of ‘one-for-one swap’ thatis utilized by GUSS for given channel condition. It is a small number that is influencedby the number of users K, the number of RAUs M and the total transmission power P.To reduce the complexity of GUSS, we introduced a new parameter, the effective channelvector (ECV), for ZFBF precoding. The ECVνiof user i is the orthogonal component ofhithatisorthogonaltothesubspacespannedbythechannelsofalltheotherselectedusers.ThelowcomplexityECVupdatingschemesfor‘add’,‘delete’and‘one-for-oneswap’up-dated user set are derived from both algebra and geometry perspectives. The ECV-basedeffective-channel-gain λ updating method reduces the computation complexity greatly,with only42.9%the complexity of λ updating method that proposed in literature. Thenew ECV-basedλ updating method reduces the complexity of GUSS greatly, and it is alsoa useful component to build more delicate low complexity user selection algorithms forZFBF precoding.Secondly, targeted on decreasing the computation complexity, we proposed severaluser selection algorithms that have low complexity and provide sum rates equal to orhigher than that of ZFS algorithm for K≤M and K> M, respectively. Three decre-mental user selection algorithms are designed for K≤M: decremental user selection(DUS) algorithm, decremental user selection with add (DUSA) algorithm and delete theminimum lambda (DML) algorithm. The three algorithms all initialize by selecting allusers and then deleting one user per iteration. DUS deletes a user to provide the highestsum rate increment in each step; DUSA retrieves the ‘mis-deleted user’ based on DUS;DML deletes the user with the minimum effective-channel-gain and increases the sumrate. They all provide sum rates equal to or higher than the sum rate of ZFS algorithmwith lower complexity than ZFS. Especially, DML achieves a complexity of O(MK2),whichissignificantlylowerthancomplexityofallZFS,DUSAandDUS,whichallhaveacomplexity of O(MK3). For K> M, an ECV-based ZFS (eZFS) algorithm is designedby applying the ECV-based effective-channel-gain updating method in ZFS algorithm.The capacity based incremental eZFS achieves exactly the same sum rate performance asZFS with a lower complexity of O(12KM3). Simulation results indicate that whenK, M,P are large enough and K M, the complexity ratio of eZFS over ZFS approaches37.For the downlink broadcast transmission, LDAS employs orthogonal space-time codingto exploit space diversity gain and increase the transmission reliability. However, the or- der of space-time coding matrix nothat can be utilized is much smaller than the number ofRAUsM atLDAS,whichisintheorderoftensorhundreds. Itisimpracticaltogenerateahundredth order STBC because of the large implementation complexity, coding/decodingdelay, overhead of padding bits in physical layer protocol, and the stringent requirementthat all the channel coefficients need to be constant over a much longer time to ensure thedesired micro-diversity gain. On the condition of no<M, it is a challenge for LDASto assign noorthogonal streams to M RAUs on the target of minimize the averaged bit-error-rate (BER) across the coverage area. It is influenced by the following three factors:1. the topology of RAU deployment on the coverage area;2. the transmitting power allo-cation for each RAU;3. the order of space-time coding and the RAU grouping schemesin assigning each space-time coding streams. The first two factors decide the distributionof total transmission power, and all the three factors decide the the distribution of eachorthogonal stream’s transmission power. Thus, we need to optimize all the three tightlycoupled factors jointly. According to our analysis, the precondition to minimize the aver-age BER is that all noorthogonal streams share the total received signal power as equallyas possible in the whole coverage area. We further proposed two heuristic RAU group-ing rules, which correspond to balancing the transmission power between and inside eachorthogonal stream, separately. The scheme that satisfies the two heuristic rules is called“stream-balancedgroupingscheme”. OntheconditionofuniformlydistributedRAUsandevenly distributed transmission power, stream-balanced grouping schemes are derived forRAUs placed in infinitely long straight line for all no, for RAUs placed in hexagonal lat-tice with n2o=p+q2+pq, and for RAUs in square lattice with no=p2+q2. Simulationresults verify that receivers achieve minimum average BER under the proposed groupingscheme. We have also shown that the average BER decreases with both noand the RAUdensity when the stream-balanced grouping scheme is used.