Genetic Design Method for Near-Optimal Training Sequences in Wideband Spatial Multiplexing Systems

Wireless communication systems that use multiple transmitting and receiving antennas have received a significant amount of attention. In these spatial multiplexing systems, by transmitting multiple co-channel signals simultaneously, the system capacity can be considerably increased especially in rich-scattering wireless channels. Over the last decade, many investigations into multiple-antenna systems have been reported. Signal processing in the receiver often requires channel state information (CSI) for spatial multiplexing systems. We can obtain CSI by sending known training sequences to the receiver. Since the residual channel estimation error seriously affects the error-rate performance, the design of good training sequences is important to realize highly spectrumefficient communications. This letter describes an efficient design method for superior training sequences that can obtain accurate CSI with a short training length. Several studies have been done on optimal training sequences for single antenna systems, as shown in examples [1], [2]. For multiple antenna systems, an analysis of the relationship between the achievable capacity and the training length is presented [3]. In [4], it is shown that optimal training sequences, which have perfect impulse-like auto-correlations and zero cross-correlations within a sliding window containing a number of the channel taps for estimation purposes, do not exist for all training lengths and constellation sizes. To construct optimal training sequences, an exhaustive search is required. Unfortunately, this is not practical due to the considerable amount of search space required, especially in spatial multiplexing systems