Neural matched filter detection in synchronous code-division multiple-access (CDMA) systems

Abstract

In communication networks employing spread-spectrum (SS) code-division multiple-access (CDMA), the performance of single user matched filter receivers are highly degraded when the relative powers of the interfering signals are large (the near-far problem). Optimum multiuser (maximum likelihood) receivers have been suggested to overcome the near-far problem, but they are too complex to implement, specially when the number of users is high. In addition, such multiuser detectors lacks security, which is one of the main advantages of SS techniques, as all of the users messages are detected simultaneously. We introduce a new technique for separately demodulating the packets belonging to different users using feedforward artificial neural networks. A special neural network, the neural matched filter (NMF), is trained not only to detect the spreading code of the assigned user but also to suppress the multiple-access interference caused by the waveforms of other users. This technique is shown to be highly resistant to near-far effects. A comparative performance analysis of conventional, optimum multiuser and NMF single user receivers is carried out via Monte Carlo simulation. It is shown that the proposed single user NMF detectors have a better performance than the optimum receiver in severe near-far cases. This exalted performance is due to the implicit cancellation of cross-correlation between different users codes. The suggested technique is also not sensitive to the code selection and does not necessitates the use of sophisticated optimized code families. Our suggested NMF technique solves most of the security, complexity and near-field problems associated with CDMA systems.