Decision feedback sequence estimation for unwhitened ISI and multiuser CDMA channels

Decision feedback sequence estimation (DFSE), which is a reduced-complexity alternative to maximum likelihood sequence estimation performs very well for whitened (minimum-phase) inter-symbol interference (ISI) channels. However, it may not be feasible to compute the whitening filter for many time-varying channels like the multiuser asynchronous direct-sequence CDMA channel. Moreover, the performance of DFSE is not very good for many non-minimum-phase whitened ISI channels and for channels for which it is desirable to perform bidirectional equalization as in the GSM system. In these scenarios, it is beneficial to use the Ungerboeck formulation for sequence estimation which operates directly on the discrete-time unwhitened statistic obtained from conventional matched filtering. In this paper, we show that conventional decision feedback sequence estimation in the Ungerboeck form is interference limited. An error probability analysis leads us to a modified unwhitened decision feedback sequence estimator (MUDFSE). Simulation results indicate that the modified algorithm used on unwhitened ISI channels with low to moderate dispersion, provides similar performance/complexity tradeoffs as the DFSE used on the corresponding whitened minimum-phase channels. The algorithm is especially attractive for multiuser detection for asynchronous DS-CDMA channels with long spreading codes, where it obtains sizable gains over the (unwhitened) M-algorithm