Blind maximum-likelihood equalization of CDMA signals via the EMV algorithm

We propose a blind maximum-likelihood (ML) equalization method for synchronous and asynchronous spread-spectrum code-division multiple access (CDMA) signals in intersymbol interference channels. Our method employs an expectation-maximization Viterbi algorithm (EMVA) capable of joint channel identification and signal estimation for transmission systems that admit a hidden Markov model description. By modelling the observed signal as a finite-state hidden Markov process, it is possible to use the Viterbi algorithm (VA) to execute the M-phase of the expectation-maximization iteration. When the parameter estimate converges to a stationary point, the best survivor of the VA is selected as the ML estimate of the transmitted sequence. Simulations show that the EMVA performs almost as well as the ML detector based on exact channel knowledge and that it is near-far resistant in the presence of large power disparities among users. We then devise an EMVA-based quasi-maximum-likelihood successive cancellation (QML-SC) method which reduces the trellis complexity by first ignoring weak users, equalizing the stronger users, subtracting off their contributions from the observed signal, equalizing and cancelling the weak users, and repeating the procedure until convergence. Though not near-far resistant, the QML-SC method tends to resist the near-far effect for a wide range of power disparities.