TSTCNN). Fig. 1 shows the BER of the frst user against the near-far-ratio (NFR = EJE,, i = 2,3,4,5) with the signal-to-noiseratio (SNR) of user 1 being futed at 8dB. It can be seen from Fig. 1 that the TSTCNN detector outperforms the HNN detector over the NFR range. When the NFR is low, the TCNN detector has poor performance. As the NFR increases, the performance of both the TCNN detector and the TSTCNN detector approach that of the optimum detector closely. In Fig. 2, the average BERs of these detectors are plotted against SNR with all users having the same power. From this Figure, it can be seen that the performance of the TSTCNN detector tracks that of the optimum detector, while the performances of the HNN and TCNN detectors are obviously interference limited with the increase in SNR. Conclusion: A time-varying scaling-parameter transiently chaotic neural network has been developed for multiuser detection in DS/ CDMA systems. Compared with the HNN detector and TCNN detector, it has much better performance. The performance of the TSTCNN detector can closely approximate that of the optimum detector.