Monte Carlo and reduced complexity methods for signal detection in wireless communication systems

We present a number of new algorithms for Monte Carlo (MC) sampling in wireless communications systems. These techniques address three inter-related problems that often arise in frequency selective systems: identifying an unknown signal constellation, detection of signals in nonlinear channels, and reducing the computational complexity of optimal Maximum A Posteriori (MAP) methods.; To achieve these goals a blind method of MC sampling, based on the Gibbs sampler, is developed for constellation classification. This method uses a super constellation to define the full conditional probability of a constellation. This algorithm is shown to require few observations for accurate classification, and to be robust to different constellations types. In addition, it is shown to be superior to a decoupled equalizer followed by an accepted classification algorithm. To improve the computational complexity introduced by a large super constellation, a multistage method is employed. This method reduces the size of the super constellation by removing low probability constellations at designated points in the algorithm. This method is shown to have comparable performance to s single stage algorithm.; A second MC algorithm, based on Gibbs sampling, is introduced to address the problem of a nonlinear channel. Given a system model, with a known nonlinearity subsequent to the frequency selective channel, this algorithm successfully estimates the model parameters at the receiver. Two methods of sampling from the channel are introduced. The first is a direct sampling method which uses the a priori distribution of the channel to generate a posteriori samples. The second method improves on the direct sampling method by using a weighted feedback parameter. This method effectively speeds the convergence of the algorithm. Simulation results demonstrate that the nonlinear MC algorithm achieves the optimal MAP performance of the same linear channel.; High computational complexity is a major weakness of trellis based Gibbs sampling methods when faced with long channel responses. A reduced complexity data sampling method is developed in order to reduce the complexity of the Gibbs sampling algorithm. By using fewer paths to trace through the trellis, a reduction in the number of calculations is achieved. In doing so, this method approaches the optimal MAP performance with a significant reduction in computational complexity. Simulations are performed for both short and long channels. It is shown that this algorithm outperforms an FFE-DFE pair having perfect channel knowledge.