Abstract:To improve the real-time performance of the system and reduce the computational complexity of Log-Likelihood Ratio (LLR) demodulation in traditional Orthogonal Frequency Division Multiplexing (OFDM) systems, a demodulation optimization method for OFDM optical communication systems based on Reservoir Computation (RC) is designed. This method normalizes the received signal and expands it with four-dimensional features before inputting it into a reservoir. Utilizing the nonlinear dynamic mapping characteristics of the reservoir, the received signal is mapped to a high-dimensional feature space, and then sparsely connected recursive neurons complete the state update. Since the reservoir parameters are fixed, only ridge regression is needed to train the output weights, avoiding the high complexity of backpropagation, thus enabling fast approximate calculation of LLR. Simulation results show that in a 16QAM modulated AWGN channel, the RC-LLR achieves a signal-to-noise ratio of 14.69dB at a bit error rate of 10-3, differing from the theoretically optimal scheme by only 0.12dB, with an average single-symbol processing time of 2.01μs. In an 80km fiber transmission scenario, its link budget reaches 27.90dB, an improvement of 3.57dB compared to the MMSE hard-decision scheme. The results confirm that the proposed method combines high accuracy, strong robustness, and excellent real-time performance, providing an optimized solution, low-latency demodulation in high-speed optical communication systems.