复合调制AM广播接收机的DL-SAE码元判决法

Deep learning-sparse autoencoder for symbol discrimination in composite AM broadcasting receiver

调幅波段数字广播系统采用AM-MPPSK(amplitude modulation—m-ary phase position shift keying)复合调制有众多优势。针对AM-MPPSK复合调制系统的数字解调部分,提出了基于DL-SAE(deep learning-sparse autoencoder)的码元判决法。利用深度学习中的稀疏自编码网络从接收信号样本中抽取出包含码间干扰在内的信号特征后,利用深度学习对稀疏自编码网络进行训练,使其能够在较强码间干扰环境下对接收信号样本进行分类,从而实现对于AM-MPPSK复合调制系统的抗码间干扰的解调。提出的码元判决法可将解调误码率从传统匹配滤波判决和幅度判决算法的10-2量级降低到10-4量级,比传统的判决法有着更为优异的解调性能和更强的适应能力。

Many advantages can be achieved by using AM-MPPSK hybrid modulation in digital AM broadcasting system. Based on the AM-MPPSK hybrid modulation system, a DL-SAE （deep learning-sparse autoencoder） symbol discrimination algorithm for the digital demodulation section is proposed. The sparse autoencoder network extracts the features including inter symbol interference （ISI） from the received signal samples, and then been trained by using deep learning algorithms. The well trained SAE network can achieve the satisfied classification of the received signal samples in the situations where there are much ISI, and realize the anti-ISI demodulation for the AM-MPPSK hybrid modulation system. The BER of the DL-SAE symbol discrimination algorithm can be reduced to 10-4 from 10-2 of the matched filtering discrimination algorithm and the integral discrimination algorithm. The proposed algorithm has more excellent detection performance and stronger adaptive ability than that of the traditional symbol discrimination algorithms.