提出了一種新的基于遞歸神經(jīng)網(wǎng)絡(luò)的快速收斂盲均衡算法。設(shè)計(jì)中采用觀測(cè)信號(hào)的四階統(tǒng)計(jì)量構(gòu)造代價(jià)函數(shù)，簡(jiǎn)化了系統(tǒng)的復(fù)雜度；利用實(shí)時(shí)遞歸學(xué)習(xí)算法對(duì)系統(tǒng)參數(shù)進(jìn)行動(dòng)態(tài)調(diào)節(jié)。該算法具有鎮(zhèn)定性，其收斂性能不會(huì)受到失真信道的影響，適用于均衡衰落性嚴(yán)重的信道。實(shí)驗(yàn)仿真結(jié)果表明對(duì)具有頻率選擇性衰落的非線(xiàn)性信道，該算法在收斂速度和對(duì)抗碼間串?dāng)_方面都具有良好的性能。

A novel fast convergence blind equalization algorithm based on recurrent neural network is proposed. Four-order statistics of the observation signals are used to calculate the cost function in order to simplify the complexity of the equalization system. Real-time recursion training algorithm is used to dynamically adjust the system parameters. The blind equalization algorithm is “equanimous” and the characteristic of convergence is not influenced by distortion of channel, it is fit for equalizing deep attenuation channel. Simulation results show that the algorithm has good performance on convergence speed and compensating for inter-symbol interference created by multi-path within non-linear channel.