Universal classification for wireless CDMA communications

Since channel characteristics of a practical code division multiple access (CDMA) system are usually unknown and difficult to model accurately, we focus on the design of non-parametric adaptive receivers with two general assumptions: First, we assume that one training sequence from each source is available; Second, the received continuous-valued sequences are quantized into a finite alphabet, and we assume that the quantized sequences can be modeled by discrete (interleaved) stationary Markov sources of certain finite order. We study and generalize the universal classification theory to develop both single-user and multi-user receivers for wireless CDMA communications based on these empirical observations. For the forward link (from base station to mobile station), we employ the type-based single-user receiver developed from universal classification for discrete stationary Markov sources. We describe an optimal quantizer design method for universal classification of continuous sources, and develop the type-based single-user receiver that achieves multiple access interference rejection through optimal quantization of empirical training data. Compared with detectors with the same knowledge, i.e., code and timing of the intended user only, our receiver yields smaller error probabilities in multi-user environments with non-Gaussian noise or weak Gaussian noise. We then develop the universal classification theory with multiple independent receptions from each source, and demonstrate that equal-gain combining of type-based statistics from each reception asymptotically achieves error probabilities that decay exponentially at larger rates than single-reception type-based detection. We apply this result to both diversity reception to combat multipath fading and soft-decision decoding of convolutional codes. We generalize the universal classification theory to discrete interleaved stationary Markov sources and show that error probabilities decay exponentially in the length of the test sequence asymptotically. Based on this theory, we develop the type-based multi-user receiver for the reverse link (from mobile station to base station). Given the knowledge of the codes of all users, but not the amplitudes or the channel noise model, we show that the type-based multi-user receiver yields asymptotically decaying error probability for each user. It yields reasonable performance degradation than the decorrelating detector in Gaussian noise, but achieves significant performance gain in non-Gaussian noise