On The Belief Propagation Detection Of Vector Channel And Its Mean Field Theory Interpratation

CDMA and MIMO are key technologies in wireless communication systems. Their detection algorithms are closed related to the performance of the systems. And the signal detection in vector channel is a unifed view of the multiuser detection in CDMA systems and MIMO detection.A novel detection algorithm for vector channel called belief propagation (BP) detection is introduced in the thesis. It is derived from its counterpart in probability inference. The algorithm is applied in the detection of vector channel, and we take multiuser detectors of DS-CDMA systems as examples to show the algorithm. If hard decision is made in the end of iterations, it's hard BP that has the same mathematical form of multistage detection algorithm. In order to achieve convergence, we propose asynchronous BP. A proof shows asynchronous BP must converge to the local maximum after a finite number of iterations. And a detector based on Ant Colony Optimization (ACO) is discussed to get the global maximum of likelihood function with high probability. We cover not only the synchronous scenario but also the asynchronous scenario of CDMA systems.Then we focus on soft BP detection that indicates the belief of binary random variables by probability density function rather than the sign of each one. Even though more complex than hard BP, it is also an NP algorithm with quite good performance.We demonstrate the essence of soft BP detection is Mean Field Theory (MFT), which is popular in statistical mechanics. A new re-interpretation of BP detection by information geometry is given to illustrate this idea. Soft BP detector replaces an e-flat submanifold with a m-flat submanifold, and then represents the posteriori probability distribution by its m-projection to a factorable submanifold.In the end, BP detection is used as a suboptimal algorithm in another vector channel model-V-BLAST systems. Its error performance is better than Order Sorted Interference Cancellation (OSIC), which is shown by simulation.