Non-Linear Precoding and Equalisation for Broadband MIMO Channels

Multiple-input multiple-output (MIMO) technology promises significant capacity improvements in order to more efficiently utilise the radio frequency spectrum. To achieve its anticipated multiplexing gain as well as meet the requirements for high data rate services, proposed broadband systems are based on OFDM or similar block based techniques, which are afflicted by poor design freedom at low redundancy, and are known to suffer badly from co-channel interference (CCI) in the presence of synchronisation errors. Non-block based approaches are scarce and use mostly decision feedback equalisation (DFE) or V-BLAST approaches adopted for the broadband case, as well as Tomlinson-Harashima precoding (THP). These methods do not require a guard interval and can therefore potentially achieve a higher spectral efficiency. The drawback of these schemes is the large effort in determining the optimum detection order in both space and time, often motivating the adoption of suboptimal approaches. In this thesis, we focus on non-block based precoding and equalisation schemes aiming to achieve higher data throughputs with improved bit error ratio (BER) compared to existing approaches. In order to achieve this, a recently developed broadband singular value decomposition (BSVD) technique is applied to decouple a broadband MIMO channel into independent frequency selective single-input single-output (SISO) subchannels of ordered qualities, thereby cancelling CCI. Secondly, these dispersive broadband SISO subchannels are individually equalised using non-linear DFE or THP schemes with a variable ransmission rate that best matches the individual qualities of the respective subchannels, whereby the decision delay can be independently optimised for every subchannel. This method is benchmarked through simulations against a state-of-the-art broadband MIMO THP technique with optimised spatio-temporal ordering showing that improved BER performance can be achieved under the constraints of identical data throughput and transmit power. In order to maximise the data throughput of our proposed method or similar multichannel systems, adaptive bit and power loading schemes have been applied. A rate-optimal approach known as a greedy algorithm is considered, whereby optimality is guaranteed by considering an appropriate bit allocation cost function and then iteratively assigning one bit at a time to the least cost-expensive sub-channel. Constraining the transmit power budget and target BER of the overall transmission system, we propose a greedy power allocation (GPA) algorithm to optimise the achieved data throughput. While maximising data rate, the GPA algorithm can also save some unused power from the total transmit budget. This power is further utilised to enhance the mean BER w.r.t. the constrained target through two proposed power redistribution algorithms. It is well known that the GPA algorithm is computationally very expensive due to the iterative nature of the algorithm. In order to efficiently reduce the computational complexity of the GPA algorithm, suboptimal GPA schemes are proposed by considering a subchannel grouping concept. We show by numerical results that these schemes, while hardly sacrificing any performance compared to the original GPA algorithm, can significantly reduce the computational complexity by an order of magnitude.