In this paper, we consider design aspects of adaptive trellisbased blind sequence estimators, which are derived from the joint ML data/channel estimation problem. The number of states in the trellis is considered as a design parameter, which provides a trade-off between the performance and complexity. It is shown that for uncoded data the data sequence can only be determined up to an unknown shift index. This shift ambiguity can be resolved by taking an outer channel encoder into account. The average magnitude of the soft outputs from the corresponding channel decoder is exploited to identify the shift index. For frequency-hopping systems over fading channels, a double serially concatenated scheme is proposed, where the inner code is applied to combat the shift ambiguity and the outer code provides time-diversity in conjunction with an inter-burst interleaver. Moreover, the selection of inner/outer channel codes is also considered.

Abstract not found