The Study of Neural Network Based Nonlinear Channel Equalizers

• Main Authors: Wang, Kuang-Jen, 王光仁
• Other Authors: Chern, Shiunn-Jang
• Format: Others
• Language: en_US
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/60421208248656462230

碩士 === 國立中山大學 === 電機工程研究所 === 86 === 　　Intersymbol interference, noise and nonlinear distortion due to the finite bandwidth of the channel and nonlinearities presented in the channel respectvely mainly affect data transmission over realistic chamnnels. Nonlinear distortion may arise during aignal companding in telephone transmission, or in the application field where amplifiers oprate near the saturation point, such as setellite communication for example. These nonlinear impairments become the major limitation for high-speed data transmission. Since the performance of channel equalizer using traditional transversal filters are poor in term of optimal decision making. We have to resort to nonlinear equalizers. 　　The adaptive nonlinear equalizers based on radial basis function network have been adopted here for nonlinear channel equalization withourt any assumption of the type of nonlinear effect. There is a clear relationship between Bayesian solution and this network. According to the Bayesian minimum error probability criterion, we implement a near optimum equalizer through the RBF network. And by using the detected feedback signals, we are able to reduce the required computation units of this network efficiently. 　　The system performance of bit error rate is compared with the conventional decision feedback equalizer and wildly used multilayer perceptron network equalizer. As confirmed by the simulation results, the performance of the RBF equalizer is superior to any other competitors.