Study On Applications Of Neural Network To Adaptive Equalization

Adaptive Equalizer is a key part of modern data communication system. In this dissertation, we focus on neural network adaptive equalization technique in the presence of non-linear distortion, additive noise and channel burst-interference. The main research work and results are as follows: In Chapter 1, the background and content of this dissertation are introduced. A review of the evolution of adaptive equalization technology is given, and the applications of neural networks for adaptive equalization are also presented. In Chapter 2, equalization is described as a space mapping in the geometry space with the view of mathematics, which establishes the base for the application of Neural Networks. The recurrent neural network equalizer (RNNE) is proposed in Chapter 3. The realization of the equalizer serial and parallel output is presented. RNNE is compared with traditional linear equalizer, and its characters are comprehensively analyzed. In Chapter 4, two decision feedback recurrent neural network equalizer (DFRNNE) are proposed which put the traditional decision feedback structure for linear channels equalization skillfully into the recurrent neural networks, and tow learning algorithms are approached for adjusting the learning step adaptively. In Chapter 5, a complex recurrent neural network adaptive equalizer is investigated and a modified Decision Feedback Complex Recurrent Neural Network Equalizer (DFCRNNE), is proposed. Based on DFCRNNE, a modified complex RTRL training algorithm is presented. In Chapter 6, the characters of decision feedback recurrent neural network equalizer in the presence burst-interference are analyzed, two anti-burst-interference algorithms are proposed. In Chapter 7, Decision feedback BP network equalizer (DFBPE) is discussed. Two common structures of DFBPE, single input & multi-output and multi-input & multi- output, are devised. DFBPE is compared with traditional BP network equalizer, and its characters are comprehensively analyzed. Moreover, characters of DFBPE and DFRNNE are compared systematically. In Chapter 8, Active functions of DFBPE output and equalization algorithm are modified for the adaptive equalization of multi-level modulation signal. Finally in Chapter 9, the research results of neural network adaptive equalization are applied to adaptive noise cancellor.