Kernel-stopped parallel turbo decoding for HomePlug Green PHY systems

In this paper, we propose a kernel-stopped technique based on parallel turbo decoding. In order to reduce redundant computations of convergent information, reliable symbol-level or bit-level a posteriori values in each window are detected with a threshold. By using these criteria, computations of backward metrics and a posteriori information are eliminated in later iterations. Furthermore, computations of convergent decoding kernels are terminated when all the decoding windows within the kernel are convergent. Based on the simulations using HomePlug Green PHY standards, the proposed technique reduces number of decoding windows within turbo decoding by up to 70% at SNR 2.0dB with a slight coding gain degradation.