BLDnet: Robust Learning-Based Detection for High-Order QAM With Nonlinear Distortion

The performances of wireless communication systems are strongly limited by the nonlinearities that exist in the transceiver. We first study the effect of nonlinearities of power amplifiers on high-order quadrature amplitude modulation (QAM) signals and then propose a bit-level demodulator network (BLDnet) to reduce the nonlinear interference. More specifically, the BLDnet can not only perform hard decisions but also provide the soft outputs for further processing in channel decoder. From the simulations, the BLDnet is observed to have a better performance than the conventional scheme in the Rapp model and the Saleh model. Compared to other detection schemes, the BLDnet has a comparatively low computation complexity without performance loss in the case of high-order modulation, such as 1024QAM.