Bayesian detection algorithms for trellis coded modulation and code division multiple access

In this dissertation, recursive maximum a posteriori (MAP) symbol detectors are proposed for signal detection in the presence of intersymbol interference (ISI), multipath fading, and additive white Gaussian noise. Although the MAP symbol detectors are optimal for a priori known channels, the nonlinearity in these detectors precludes analytical evaluation of bit-error rate (BER). Thus special consideration is required to evaluate the BER performance and to speed up the simulation. These problems are overcome by using a novel importance sampling randomized bias technique (RBT). Assuming exact knowledge of the channels, MAP symbol detectors are extended to trellis coded modulation (TCM) and asynchronous code division multiple access (CDMA) systems. Importance sampling, based on the RBT, is then applied to accurately evaluate the BER performance of MAP symbol detectors.; We next consider quasi-synchronous CDMA (QS-CDMA) systems, where mobile users synchronize their transmissions to a common GPS-generated clock. The key advantage of QS-CDMA exploited here is that the need for code acquisition is eliminated. However, residual delay/channel estimation is still an important problem in QS-CDMA. For a time-invariant channel, we propose a new joint delay/channel estimation and data detection algorithm, based on the QR decomposition combined with the M-algorithm, (QRD-M). We use an exponentially windowed RLS algorithm as the delay/channel estimator. Constrained maximum likelihood (CML) data decisions are made using the QRD-M algorithm, or alternatively, the Lenstra-Lenstra-Lovasz (LLL) basis reduction algorithm.; The QRD-M algorithm is finally extended to the time-varying multipath QS-CDMA channel. Under the assumption that both user delays and channel coefficients are first-order Gauss-Markov processes, we use the extended Kalman filter (EKF) to track them. By separating data detection from the delay/channel estimation process, we are able to use only one EKF, such that we can significantly reduce the computational complexity as the number of users increases.