A proposed improved code-division multiple-access (CDMA) detector would achieve high resistance to jamming of weak ("far") received signals by strong ("near") ones. The problem of such jamming is called the "near/far problem," and resistance to such jamming is called "near/far resistance." The near/far problem is a major technical obstacle in the operation of CDMA radio-communication systems; for example, a system in which a base station must communicate simultaneously with near and far mobile units.

In a CDMA system, the binary signal from each source is multiplied, before transmission, by a distinct pseudonoise-code or other signature waveform. In a conventional CDMA detector, the signal received from each of K sources is isolated from the other K-1 signals by use of one of Kmatched filters that function in parallel. In each matched filter, the total received signal is multiplied by the signature waveform for the source in question, and the resulting product is integrated in time over a symbol period. A simple decision device then chooses the received bit (+1 or -1) on the basis of the sign of the output of the matched filter.

A Compact Neural Network would process the outputs of matched filters in such a way as to obtain an optimal solution to the near/far problem.

The point of departure for the design of the improved CDMA detector was the observation that the near/far resistance of a CDMA detector could be maximized (in the sense that the probability of error in detected bits could be minimized) by minimizing a quadratic objective function. Given a tentative decision as to the signals received from the various sources and as to the bits that the signals represent during a symbol period, one suitable quadratic function is equivalent to an estimate of noise energy in the received signal.

The problem of minimizing an objective function amounts to an optimization problem. Thus, the near/far problem in CDMA reception can be converted to an optimization problem. A CDMA detector that achieves near/far resistance by solving this optimization problem is called an "optimal multiuser detector" (OMD).

In the improved CDMA detector, the outputs of the matched filters would be fed to a parallel-processing, compact neural network (see figure) in which the synaptic weights would be based on the K × K matrix of cross-correlations of signature waveforms. The neural network would implement a gradient-descent algorithm in an effort to minimize an energy function that would be modified by addition of a constraint energy. A simulated-annealing technique called "hardware annealing" would be employed to escape from local minima of the energy function. In the particular type of hardware annealing contemplated for this detector, the gain of each neuron would be continuously increased from a minimum to a maximum value. The combination of the constraint energy and the hardware annealing could significantly improve detection performance. A prototype very-large-scale integrated (VLSI) circuit version of the detector has been designed.

This work was done by Wai-Chi Fang of Caltech and Bing J. Sheu and Theodore W. Berger of USC for NASA's Jet Propulsion Laboratory. In accordance with Public Law 96-517, the contractor has elected to retain title to this invention. Inquiries concerning rights for its commercial use should be addressed to

Technology Reporting Office
Mail Stop 122-116
4800 Oak Grove Drive
Pasadena, CA 91109
(818) 354-2240

Refer to NPO-20205

This Brief includes a Technical Support Package (TSP).
Document cover
Parallel-processing CDMA detector with neural network

(reference NPO20205) is currently available for download from the TSP library.

Don't have an account? Sign up here.