Beyond BAD:A Parallel Arbitration Framework for Low-Complexity Equalization

Since optimal equalization has complexity exponential in channel length and information rate, the suboptimal Decision Feedback Equalizer (DFE), whose complexity is linear in channel length, is a popular alternative for practical communication systems. The recently proposed BAD (Bidirectional Arbitrated DFE) algorithm provides substantial performance gains over the standard DFE by arbitrating between the outputs of a forward and reverse DFE which are run in parallel. The idea behind BAD is to generate two \suuciently diierent"candidate data sequences, and to arbitrate between them based on which sequence best explains the received data around the symbol being demodulated. In this paper, we demonstrate that a natural generalization of this methodology-arbitrating between multiple candidate data sequences generated in parallel with low-complexity equalizers-can further improve performance while still incurring complexity which is comparable to that of the DFE. We go beyond the BAD algorithm in two key respects, by providing methods for generating additional candidate data sequences that have low correlations among their error patterns, and by using improved arbitration mechanisms.