Receiver Design Of Deep Learning Aided LDPC-BICM Systems

Taking advantage of a near Shannon limit performance and excellent error correction ca-pability,the low-density parity-check coded bit-interleaved coded modulation(LDPC-BICM)has emerged as a powerful air interface technique of wireless communication.Nevertheless,the LDPC-BICM suffers from several application scenarios,such as the impulsive interference of heterogeneous system,the nonlinear distortion of analog front and the co-channel interference of the multi-user.In the context of the BICM receivers,the demapper derives the mismatched extrinsic log likelihood ratio(LLR)values due to the above nonlinear issues,resulting in a serious degradation in performance.To address the challenges of diversified communication service and performance require-ment,the LDPC-BICM promises to ensure the transmission effectiveness and system reliability.Motivated by the artificial intelligence,we make the first effort to integrate the deep learning(DL)technologies into the BICM receiver for combining the generalization ability and the in-ferential capability.The DL-aided LDPC-BICM can achieve a significant improvement.In the context of the DL-aided LDPC-BICM,the main contributions include:1)We propose an enhanced Gaussian mixture model-based(GMM)LDPC-BICM receiver with blanking nonlinearity to mitigate the impulsive interference in the L-band Digital Aero-nautical Communication System Type1(L-DACS1).First,we investigate the threshold opti-mization for pulse blanking using the modified protograph based extrinsic information trans-fer(PEXIT)analysis for considering the influence of LDPC codes.Combined with the pulse blanking optimization via the PEXIT analysis,we propose a novel MAP demodulator based on the GMM and estimate the parameters with the expectation-maximization learning algorithm.Simulation results show the Gaussian mixture-based MAP demodulator can obtain the PEXIT thresholds that match the decoding curves well and provide the better BER performance in the interference-limit channel environment.2)We propose a DL approach for enhancing the LDPC-BICM receiver to mitigate the clip-ping distortion in the direct current-biased optical orthogonal frequency division multiplexing(DCO-OFDM)systems.We first develop a non-iterative neural network-aided BICM(NN-BICM)receiver,where the NN is trained with the loss function of cross-entropy to output the modified condition probability through the softmax activation function,thereby assisting in an LLR improvement.Then,we propose two iterative NN-BICM receivers for iterative demap-ping and decoding.The single iterative design feeds the soft decisions from the LDPC decoder back to the demapper only,while the joint iterative design feeds the soft decisions back to the demapper and NN jointly.We further investigate an efficient bit loading algorithm for DCO-OFDM systems employing the NN-BICM receiver.Both NN-BICM receivers and iterative schemes can obtain remarkable performance gains over the existing benchmarks.3)We propose a sequence model-aided LDPC-BICM receiver to combat the co-channel interference in the OFDM system.In the single antenna system,this intelligent LDPC-BICM receiver introduces a model-driven NN architecture for enhancing the soft demodulation in the presence of CCI distortion.By adopting the single and multiple subcarrier architecture,we pro-pose two candidate training modes and investigate the robustness analysis to demonstrate the generalization ability.By adopting the loss function of cross-entropy and the softmax activation function,Both the single and multiple subcarrier architecture exhibit a significant robustness on the improvement of LLR values by taking the interference power and CSI into consideration.In the multiple antenna system,this intelligent LDPC-BICM receiver resort to the bidirectional recurrent neural network(BRNN)for outputting the modified condition probability when con-sidering the spatial correlation into account.Simulation results show the superiority to the counterparts over the Rayleigh channel.