Receiver Design for Massive MIMO

Abstract

Massive multiple-input-multiple-output (MM) is becoming a promising candidate for wireless communications. The idea behind MM is to use a very large number of antennas to increase throughput and energy efficiency by one or more orders of magnitude. In order to make MM feasible, many challenges remain. In the uplink a fundamental question is whether to deploy single massive arrays or to build a virtual array using cooperative base stations. Also, in such large arrays the signal processing involved in receiver combining is non-trivial. Therefore, low complexity receiver designs and deployment scenarios are essential aspects of MM and the thesis mainly focuses on these two areas. In the first part, we investigate three deployment scenarios: (i) a massive co-located array at the cell center; (ii) a massive array clustered at B discrete locations; and (iii) a massive distributed array with a uniform distribution of individual antennae. We also study the effect of propagation parameters, system size, correlation and channel estimation error. We demonstrate by analysis and simulation that in the absence of any system imperfections, a massive distributed array is preferable. However, an intermediate deployment such as a massive array clustered at a few discrete locations can be more practical to implement and more robust to imperfect channel state information. We then focus on the performance of the co-located scenario with different types of antenna array, uniform square and linear arrays. With MM, it may be the case that large numbers of antennas are closely packed to fit in some available space. Hence, channel correlations become important and therefore we investigate the space requirements of different array shapes. In particular, we evaluate the system performance of uniform square and linear arrays by using ergodic capacity and capacity outage. For a range of correlation models, we demonstrate that the uniform square array can yield similar performance to a uniform linear array while providing considerable space saving. In the second part of the thesis we focus on low complexity receiver designs. Due to the high dimension of MM systems there is a considerable interest in detection schemes with a better complexity-performance trade-off. We focus on linear receivers (zero forcing (ZF) and maximum ratio combining (MRC)) used in conjuction with a Vertical Bell Laboratories Layered Space Time (V-BLAST) structure. Our first results show that the performance of MRC V-BLAST approaches that of ZF V-BLAST under a range of imperfect CSI levels, different channel powers and different types of arrays as long as the channel correlations are not too high. Subsequently, we propose novel low complexity receiver designs which maintain the same performance as ZF or ZF V-BLAST. We show that the performance loss of MRC relative to ZF can be removed in certain situations through the use of V-BLAST. The low complexity ordering scheme based on the channel norm (C-V-BLAST) results in a V-BLAST scheme with MRC that has much less complexity than a single ZF linear combiner. An analysis of the SINR at each stage of the V-BLAST approach is also given to support the findings of the proposed technique. We also show that C-V-BLAST remains similar to ZF for more complex adaptive modulation systems and in the presence of channel estimation error, C-V-BLAST can be superior. These results are analytically justified and we derive an exhaustive search algorithm for power control (PC) to bound the potential gains of PC. Using this bound, we demonstrate that C-V-BLAST performs well without the need for additional PC. The final simplification is based on the idea of ordering users based on large scale fading information rather than instantaneous channel knowledge for a V-BLAST scheme with MRC (P-V-BLAST). An explicit closed form analysis for error probability for both co-located and distributed BSs is provided along with a number of novel performance metrics which are useful in designing MM systems. It is shown that the error performance of the distributed scenario can be well approximated by a modified version of a co-located scenario. Another potential advantage of P-V-BLAST is that the ordering can be obtained as soon as the link gains are available. Hence, it is possible that mean SINR values could be used for scheduling and other link control functions. These mean values are solely functions of the link gains and hence, scheduling, power adaptation, rate adaptation, etc. can all be performed more rapidly with P-V-BLAST. Hence, the P-V-BLAST structure may have further advantages beyond a lower complexity compared to C-V-BLAST.