Iterative decoding with imperfect channel estimation for wireless systems

In this thesis, we propose a new iterative algorithm for turbo decoding with integrated channel estimation for Rayleigh flat-fading channels. The design of the proposed algorithm is based on a new turbo decoding metric for binary phase-shift keying (BPSK) signaling on fast fading Rayleigh channels with noisy channel estimates. The algorithm consists of a channel estimator that can reduce the error variance of the channel estimate iteratively by using the soft extrinsic information output from the turbo decoder. The extrinsic information generated from the turbo decoder has some a priori information of the transmitted data symbols which can be used to refine the channel estimate after each iteration of decoding. The refined channel estimate is then fed back to the turbo decoder for the next iteration of decoding. The resulting iteration between the channel estimator and the turbo decoder using the new metric, with intermediate exchange of soft channel-symbol information, yields very impressive results. Simulation results verify that the proposed algorithm can outperform conventional algorithms significantly and even its performance can approach that of a turbo decoding algorithm with perfect knowledge of the channel.;We also study the effect of signal-to-noise ratio (SNR) mismatch on the bit-error rate (BER) performance of turbo decoding algorithms. According to the previous results obtained by other researchers, turbo-decoding with the Max-Log-MAP decoder is independent of SNR for Rayleigh fading channels when the channel is assumed to be perfectly known to the receiver. Using such an assumption, the Rayleigh fading channel may then be viewed as an additive white Gaussian noise (AWGN) channel conditioned on the fading coefficient. Since in practical communication systems, the channel information is not available at the receiver, the channel must be estimated at the receiver. Because of the low signal to noise ratios typical of turbo coded systems, it is difficult to obtain perfect estimates of the fading coefficients. Thus, we show that the Max-Log-MAP decoder is sensitive to SNR mismatch over fading channels with noisy channel estimates. Also, it is shown that the Max-Log-MAP decoder is less sensitive to SNR mismatch than the MAP or Log-MAP decoder.;Finally, we propose a new SNR mismatch model to examine the sensitivity of turbo decoding algorithms to SNR mismatch for systems with an SNR estimator. This model is more practical than the conventional one. Under this model, we still can verify that the Max-Log-MAP decoder is sensitive to SNR mismatch over fading channels.;In most existing turbo decoding algorithms for fading channels, the received data symbols are multiplied by the complex conjugate of an estimate of the channel gain before turbo decoding without any mathematical justification. However, in this thesis, we prove that the result of this kind of preprocessing (or filtering) operation on the received signal to compensate the effect of fading distortion can be viewed as a method to obtain linear minimum-mean-square-error (MMSE) estimates of the transmitted data symbols.