Abstract:
This study investigates machine-learning techniques aimed at enhancing turbo decoding in wireless communication. Traditional turbo decoders often struggle with challenges such as susceptibility to burst noise and high error rates at high Signal-to-Noise Ratios (SNRs). To tackle these issues, the study explores Sequence-to-Sequence attention models and Transformer architectures, adapting them for turbo decoding to potentially enhance accuracy and robustness across various channel noise conditions. The research includes foundational discussions on convolutional and turbo codes, simulations using the SOVA algorithm, reviews of neural networks in turbo decoding applications, and introduces the effective models TurboAttention and TurboTransformer. These models demonstrate promising results in terms of Bit Error Rate (BER) across a wide range of SNR values, with encouraging performance observed in hardware inference tests.
