Noise predictive turbo systems

Iterative decoding has been widely studied for memoryless white Gaussian noise channels. For nonideal channels, e.g., correlated noise channels, iterative decoding combined with iterative noise estimation may improve the performance of the detector. The basic idea is to exploit the decoding results of the previous iteration to estimate the correlated noise so that better decoding results ran be obtained. Furthermore, these results may lead to better estimation and even better decoding results. There are many ways to exploit the decoding results of the previous iteration. In this paper, two noise prediction schemes, namely noise predictive turbo systems with soft feedback (NPTS/SF) and noise predictive turbo systems with hard feedback (NPTS/HF) are proposed, and their performance for a serially concatenated convolutional turbo system are investigated. Simulation results show that the noise can be iteratively whitened. For the systems studied, NPTS/SF exhibits less error propagation than NPTS/HF at very high recording density. Both NPTS schemes may provide lower error floors