In many communication applications, the dominant impairment is interference. This is especially true of wireless environments. There is typically self-interference due to multipath propagation. When system bandwidths are large enough, this takes the form of intersymbol interference. However, there is also interference from other users in the next work, as well as various forms of extra-network interference. The degree to which such interference can be mitigated through signal processing within the network strongly affects the overall capacity. Thus interference-cancellation algorithms are of considerable interest. However, traditional low-complexity linear interference cancellation techniques cannot exploit enough of the structure in the interference to be effective, while maximum-likelihood (ML) interference cancellation are the most effective, but have exponential complexity. We show that in fact, the performance of ML cancellation can be approached at high SNR with a complexity that is negligibly higher than the linear techniques.
In particular, we develop an efficient convergent iterative algorithm structure that alternates between generating increasingly reliable symbol decisions and increasingly reliable interference estimates. Combining such decoding structures with a precoding technique at the encoder we refer to as mode-interleaving extends their effectiveness to a particularly broad range of channels.