The USC Andrew and Erna Viterbi School of Engineering USC Signal and Image Processing Institute USC Ming Hsieh Department of Electrical Engineering University of Southern California

Technical Report USC-SIPI-313

“A Precision Receiver for CDMA Communication Using Neural-Based Array Processors”

by Cheng-Hsiung Chen

August 1997

With rapid advances of deep-submicron semiconductor technology and progress in communication systems, our lives enter a new Multimedia Era: communicating in any place and at any time as evidenced by the availability personal communication systems. Among all possible technologies, Code Division Multiple Access (CDMA) has been receiving more and more attention and the market of wireless communication with CDMA is booming since the year 1990.

CDMA is a spread spectrum technology. All of the CDMA users share the same bandwidth and their communication channels are separated by means of pseudorandom codes. The universal frequency reuse is crucial to the high spectral efficiency. To maintain high quality and high spectral efficiency in CDMA systems, the power difference of received signals should be as small as possible. In satellite communication, the high-power and low-power transmitters co-exist. In land communication, some users may be near the base-station and some may be far away. A 60 dB or more signal power difference is quite possible. This is called near-far problem. In 1986, S. Verdu illustrated in the theoretical deprivation that the optimal near-far resistance detector could be achieved. However, no electronic implementation has been developed so far.

The one-dimensional compact neural network is very suitable for communication receivers in CDMA systems. Its architecture is based on a combination of the locally connected cellular neural network and the fully-connected Hopfield neural network. The compact neural network is a very efficient architecture for electronic implementation. It exhibits high degree of fault tolerance, high data throughput rate, and even low power consumption. By properly mapping the cost function of optimal near-far resistance detector onto the energy function of the compact neural network and applying the innovative hardware annealing technique, multiuser detector with optimized solution is achieved. Extensive computer simulation using MATLAB codes has been conducted. Satisfactory results are obtained. The neural network-based CDMA receiver design is a very convincing solution in future personal communication systems.

To download the report in PDF format click here: USC-SIPI-313.pdf (15.9Mb)