Studies On Computational Intelligence And Its Application To Multiuser Detection

Code-Division Multiple-Access (CDMA) mobile communications systems are interference-limited systems. Multiple access interference (MAI) is the main interference in the communications systems. It is important that the MAI is suppressed so that the system performance and capacity are increased. An efficient method suppressed MAI is multiuser detection (MUD) which views the MAI as an useful resource and makes full use of the relationship between users to increase the detection performance. So the MUD is one of key techniques in CDMA communications systems.Computational intelligence (CI) is proposed in recent years as an important information processing method which is studied and applied widely all over the world. CI has explored the intelligent information processing in science and technology by modeling behaviors and mechanisms that underlie biologically intelligent organisms. CI has shown many advantages over conventional optimization algorithm.MUD problem can be viewed as a combinational optimization problem. The CI optimization strategy is applied to the search process of the MUD. In this dissertation, we have studied the MUD techniques based on the CI and made the numerical simulation, analyses and discussion. Moreover, a new training approach for the training algorithm of a fully connected recurrent neural network based on the digital filter theory is also proposed.The contents of each part in the dissertation are as follows:Chapter 1 first mainly introduces the review and research significance of the MUD techniques. The chapter then introduces the CI and the potential for applications of CI to MUD.Chapter 2 first introduces the equivalent model of CDMA communications systems. The synchronous and asynchronous model is described in detail. The chapter then introduces the MUD performance measure which includes bit error rate (BER) and near-far resistance. Moreover, the method and performance of the optimum multiuser detection are described. The method and classification of sub-optimum multiuser detection are also introduced. The chapter finally introduces the basic theory and method of CI, such as genetic algorithm and tabu search algorithm, etc.Chapter 3 proposes and discusses four MUD techniques based on neural network.They are : 1) MUD based on optimization neural network; 2) MUD based on Lagrange neural network; 3) MUD based on hysteretic Hopfield neural network; 4) MUD based on recurrent neural network. The chapter gives the results of numerical simulation and makes the discussion. The comparison and analyses of the four MUD techniques are made then from computational complexity and detection performance. The chapter finally proposes and discusses a MUD method based on modified recurrent neural network which uses the chaotic technique to produce good initial values of the neural network. The numerical simulation is also made.Chapter 4 proposes and discusses two MUD techniques based on hybrid optimization strategy. One is the MUD based on genetic algorithm and tabu search. The other is the MUD based on genetic algorithm and recurrent neural network. The optimization strategy and mechanism of the two hybrid methods are described. Moreover, the numerical simulation is made. The chapter then makes the comparison and analyses of the two hybrid methods from computational complexity and detection performance.Chapter 5 proposes a new training approach for the training algorithm of a fully connected recurrent neural network based on the digital filter theory. Each recurrent neuron is modeled by an IIR filter. The weights in the network are updated by optimizing IIR filter coefficients and optimization is based on the layer-by-layer optimizing procedure (LBLO) and the recurrent least squares (RLS) method. The performance of the proposed algorithm is demonstrated with application in complex communication channel equalization. Computer simulation results indicated that the proposed method provides fast convergence rate. This provides a new way to the fast training of complex valued recu...