For a single user scheme with frequency dependant channel, a generalized maximum likelihood sequence estimation (MLSE) algorithm was demonstrated to improve the performance over the single antenna method. It achieved almost perfect combining using blind joint data and channel estimation by means of combining of metrics from two antenna diversity. The dominant cause of errors was due to the presence of two trellises with almost identical metrics and these trellises were related being the same sequence delayed by plus or minus one sample period. This was especially prone to happen when the estimated model order was not the true order. Another difficulty with this algorithm was the reduction in the number of choices for trellises entering a state as the trellises tended to converge to the best one; this provided a larger exhaustive search at the beginning of the algorithm than later when fading affected the channel.
PDF file, 1.160 MByteshttps://www.ece.unb.ca/petersen/pubs/theses/students/Wi03/