The bulk of DFE research has been devoted to proposal and performance study of various modifications of the basic DFE. In this tradition, a soft decision feedback equalization scheme has been proposed in this thesis, where the hard-limiter of a conventional DFE is replaced with a soft decision device. The intention of this modification is mitigation of error propagation. The mean squared error (MSE) optimized soft decision device is derived here using various simplifying assumptions. Simulation results demonstrating performance improvements are provided. Further, a bit error rate (BER) optimized soft decision device is derived for a BPSK modulation scheme. The joint optimization of the soft DFE decision device and the DFE taps is discussed for the special case of a source with Gaussian distribution.
Performance evaluation of the soft DFE based on a multi-level quantization approximation is discussed. The error dynamics of the system are modeled using a Markov chain and the corresponding state occupancy probabilities are computed. MSE and BER of the soft DFE are calculated from these probabilities. A few sub-optimal implementations, namely the multi-level quantization scheme and the piece-wise linear approximation, of the soft DFE decision device are considered. Simulation results on performance comparison of soft DFE with the conventional DFE, for measured microwave and broadband wireless channels, are provided.
In practical communication systems, error correction coding is a critical component needed to achieve reduced bit error rates. Most coding schemes are designed to correct random errors and perform poorly in the presence of ISI. This has motivated research on simple but effective joint equalization and decoding strategies. Combining DFE with error correction coding is a challenging problem due to the presence of error propagation. The bursty nature of DFE errors causes performance loss in the decoder.
A turbo DFE strategy, based on the turbo-equalization principle, has been proposed in this thesis. In this scheme equalization and decoding are performed iteratively. The more reliable decisions at the output of the decoder are fed back as inputs to the feedback filter of the DFE, thereby reducing the probability of occurrence of primary errors and reducing the likelihood of error propagation. Simulation results are provided for a few examples.