In this paper, we propose a low-complexity equalizer whose performance approaches that of the optimum maximumlikelihood (ML) estimator, in ultra-wideband (UWB) multipleinput multiple-output (MIMO) channels. Although the ML estimation offers the excellent performance, it is generally of high complexity due to the large size of searching space. Through a theoretical search-space analysis, we reveal that local searching algorithms may approximate the ML estimator when a large number of diversity branches is available. Based on the analysis, we develop a novel equalizer with simple bit-flipping for severely frequency-selective fading in UWB-MIMO channels.
I. INTRODUCTIONThe recent demand of ubiquitous networking has necessitated breakthroughs realizing high-speed radio communications. One promising technique is termed multiple-input multiple-output (MIMO), where multiple antennas are used at both transmitter and receiver. A lot of theoretical and experimental studies have revealed that the MIMO systems can significantly improve the channel capacity [1][2][3]. For the near future, the signal bandwidth will likely increase to several GHz to accommodate higher data-rate transmissions. For instance, ultra-wideband (UWB) techniques have drawn much attention especially for body area networks. Future systems envision boosting the capacity even further by incorporating UWB into MIMO. In the UWB-MIMO systems, we require sophisticated equalization techniques to deal with dispersive channels.In frequency-selective MIMO channels, an equalizer employing the maximum-likelihood sequence estimation (MLSE) offers an optimum performance. However, it is of high complexity in general due to the large number of trellis states, which increases exponentially according to the channel memory length. Among low-complexity sub-optimum equalizers, frequency-domain (FD) equalization has received recent attention due to its reasonable tradeoff of performance and complexity. Although the FD equalizer based on the minimum mean-square error (MMSE) may outperform that based on the zero-forcing (ZF), it is still significantly inferior to the optimum MLSE equalizer in general.We start off by analyzing a search space of the ML estimator in order to derive an insight for complexity reductions. The high complexity of the ML estimation comes from the fact that it needs to search for the global optimum over all the possible candidates in the very large search space. By focusing on the behavior of undesired local optima in the search space, we discover that the probability of hitting local optima decreases exponentially with the number of diversity branches. This