Hopfield Neural Network Multiuser Detection for DS-UWB Systems

• Main Authors: GuoJun Wen, 溫國俊
• Other Authors: none
• Format: Others
• Language: en_US
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/61041055885713980036

碩士 === 國立中正大學 === 通訊工程研究所 === 95 === Recently, wireless communications are more momentous in daily life as a result of their fast development. With the increasing number of users and the request of transmission quality, wireless systems with high transmission rate increase gradually; therefore, many wireless techniques are developed, such as ultra-wide band (UWB) systems, IEEE 802.11a/b/g, worldwide interoperability for microwave access (WiMax), orthogonal frequency-division multiplexing (OFDM), multiple-input multiple-output OFDM (MIMO-OFDM). Among these techniques, UWB systems attract many attentions due to their high transmission rate for indoor environment, low implementation cost, low interference, low power consumption and precise positioning capability. The main idea of this thesis is to discuss the multiuser detections of direct sequence-UWB (DS-UWB) systems. The DS-UWB systems are mainly applied in the wireless transmission of indoor environment in which are many obstacles to cause serious multipath interference. In order to overcome the problems of multipath interference efficiently, we adopted the maximal ratio combining (MRC) Rake receiver. Besides, with the increasing number of users, the multiuser DS-UWB systems will cause serious multiuser interference. Consequently, we adopted matched filter to match bit information of each user. According to theoretical analyses and simulations, we know that the multiuser detector with maximum likelihood (ML) estimation can provide the best performance. On the other hand, its computational complexity will grow exponentially with the number of the users. The multiuser detectors that employ decorrelating and minimum mean square error (MMSE) can give the suboptimal performance; however, the structure of hardwire is hard to be implemented as a result of inverse matrix of their crosscorrelation matrix R. Hence, we adopted the Hopfield neural network (HNN) that can be approximated to maximum likelihood estimation for DS-UWB system. However, the conventional HNN algorithm can converge easily to partial minimum. In order to improve this problem, we proposed modified HNN based on simulated annealing (SA) method which can avoid local minimum and obtain global minimum. From simulation results, the HNN detector based on SA method can provide an attractive performance and be implemented easily on hardware in the multiuser DS-UWB systems.