Deep Learning-Aided Hybrid HARQ System with CRC-Based Feedback for MIMO-OFDM in Fading Channels

This paper presents a new cross-layer simulation framework which combines deep learning (DL) approaches at the physical (PHY) layer and hybrid automatic repeat request (HARQ) techniques at the medium access control (MAC) layer for effective wireless communication. The system is structured on a MIMO-OFDM configuration with 8 transmit and 8 receive antennas, QPSK modulation, and 64 subcarriers. In order to simulate actual communication scenarios, different fading channel models such as AWGN, Rayleigh, and Rician are added along with other degrading factors like channel estimation errors and changes in signal-to-noise ratio (SNR) levels. At the PHY layer, a very deep neural network (DNN) was used to improve the output of a conventional MMSE equalizer. The DNN learns to remove the noise and distortion created by the wireless channel as well as the imperfect channel state information (CSI) in the decoding process. This provides additional improvement to the decoded signal beyond that which can be achieved from normal signal processing. At the MAC layer, there is a CRC-based HARQ mechanism. A cyclic redundancy check (CRC) is applied to each data packet, and the system uses the result to trigger retransmissions, up to a maximum of three attempts. This imitation is of practical feedback mechanisms in current wireless protocols like LTE and 5G. Extensive simulations are created in order to estimate system performance based on bit error rate (BER), symbol error rate (SER), training loss, latency, and decision accuracy (ROC curves). The proposed framework realizes notable improvements in reliability and decoding accuracy, which manifest the success of combining DL-based PHY enhancements and intelligent MAC-layer feedback control.