A generalized formula for the decision error probability is derivedfor an M-ary PAM system. Then, by applying the gradient method to thisgeneralized formula, a GMSER algorithm is developed to minimize thesymbol error rate for linear equalization. Then a SGMSER algorithm isobtained by further simplification of GMSER to reduce thecomputational complexity in GMSER. Comparison is then made between theSGMSER algorithm, the MMSE criterion, and an AMBER algorithm asproposed by Yeh and Barry. The results show that the minimum symbolerror rate criterion outperforms the minimum mean-squared errorcriterion. Then, between AMBER and SGMSER, we find that the AMBERoffers less computational complexity, while the SGMSER offers fasterconvergence rate and lower error rate for higher order signaling.
Proceedings of 2002 IEEE International Symposium on Intelligent Signal Processing and Communication Systems，頁448-452