The Neural Convolutional Decoders (NCD) have been used as an alternative for decoding convolutional codes. The motivation of the NCD is to reduce the hardware complexity of the conventional convolutional decoders and maintain the performance. To show the role of the nonlinearity of the neural networks, simulations are performed under the satellite channel that has the nonlinear distortion. We restrict our attention to the case of coherent QPSK modulation. In this case, the neural networks can learn the nonlinear distortion of the satellite channel including the filtering effects and the nonlinear effects of the traveling-wave tube (TWT) amplifiers. In result, the performance of the NCD with a simple structure is almost equal to that of the soft decision Viterbi decoder for the systematic code with 32-bit paths which need 128-bit memory storage. And, the performance difference between the NCD for the systematic code and the soft decision Viterbi decoder for the nonsystematic code is about 1.0 dB and nearly identical to that caused by the dissimilarity of the error-correcting ability in two codes. Therefore, if we could efficiently train the NCD for the nonsystematic codes, the NCD would perform as well as the soft decision Viterbi decoder for the nonsystematic code.
|Number of pages||6|
|Publication status||Published - 1995|
|Event||Proceedings of the 1995 IEEE International Conference on Neural Networks. Part 1 (of 6) - Perth, Aust|
Duration: 1995 Nov 27 → 1995 Dec 1
|Other||Proceedings of the 1995 IEEE International Conference on Neural Networks. Part 1 (of 6)|
|Period||95/11/27 → 95/12/1|
All Science Journal Classification (ASJC) codes