Performance of Generalized Diversity Combining System for Coded Transmissions

Diversity combining is an important technique for combating fading in wireless communications. Generalized diversity combining selects a subset of diversity branches and combines them coherently. It has been established that when the transmission is uncoded, the optimal generalized diversity combining in the sense of minimizing bit error rate is to select branches that offer the maximum sum log-likelihood ratio (LLR). In the coded case, such diversity combining is not optimal any more. We present in this paper an upper bound result on the bit error rate performance of coded transmissions over general memoryless channels, and apply the analysis to the diversity selection problem. The resulting semi-analytical method for evaluating performance of coded diversity combining schemes offers insight on the difference between the performances of the coded and uncoded systems. It is shown via a numerical example that a scheme that selects individually "strong" branches (in the sense of large LLR) can out-perform a scheme that selects branches that are jointly "strong"