Multi-Channel Nonlinearity Mitigation Using Neural-Network-Based Feedback Cancellation with Channel Decision Passing Algorithm

This paper investigates neural network (NN) based feedback cancellation in combination with the channel decision passing (CDP) algorithm for the mitigation of multi-channel receiver nonlinearities. The CDP algorithm iteratively assesses output symbols from each channel, utilizing them in successive channels to mitigate nonlinearities within a feedback loop until convergence across all channels is achieved. While prior work relied on a polynomial-based model to estimate the harmonic distortions and intermodulation products introduced by the nonlinearities of the analog front end (AFE), this paper employs trained neural networks to not only avoid the need for an exhaustive search for the inband nonlinearities expression required by the polynomial-based model but also to achieve improved cancellation accuracy. Simulation results on a 3- channel receiver demonstrate that the NN-based CDP algorithm achieves improved communications signal-to-noise ratio (SNR) in comparison with conventional cancellation techniques including multi-channel decisions feedback cancellation (MCDFC) and transversal feedforward NN-based topologies. The bit-error-rate (BER) performance comparison is made between the NN-based and the polynomial-based CDP algorithms, which demonstrates that the NN-based CDP algorithm achieves comparable bit- error-rate (BER) performance to the polynomial-based model counterpart without significantly increasing complexity for moderate AFE nonlinearities. Moreover, it offers enhanced BER performance for pronounced AFE nonlinearities but with a considerable increase in complexity.