Noncoherent decision-feedback multiuser detection

The problem of noncoherent multiuser detection for multipulse modulation over the Gaussian multiuser channel is studied. It is assumed that neither the energies nor the carrier phases of the signals of any of the users are available at the receiver. Previously, a post-decorrelative generalized likelihood ratio test (GLRT) based detector was proposed as a solution to this problem. In this paper, we introduce the concept of noncoherent decision feedback (DF) that forms the basis for an improved solution to the same problem. The users are detected sequentially in some fixed order, and the decision for a particular user takes advantage of the reduction of uncertainty associated with the signal space resulting from decisions already made for previous users. In contrast to coherent DF, therefore, the feedforward and feedback transformations in the noncoherent case are themselves functions of the matched filter outputs. An efficient implementation of the noncoherent DF detector for M-ary modulation and ρ-dimensional signal space requires O(M_ρ) computational complexity per user. Upper and lower bounds on its symbol error probability are obtained. It is shown that significant improvements over the post-decorrelative GLRT-based detector are often possible. Sufficient conditions are obtained which guarantee, without ignoring error-propagation effects, that the high signal-to-noise performance of the DF strategy can be as good as its genie-aided version in which perfect past users´ decisions are used