Blind and semi-blind equalization using hidden Markov models and clustering techniques. (Q1575758)

A novel blind channel equalizer is proposed which is suitable both for linear and nonlinear channels. The proposed equalizer consists of the following three steps: (a) identification of the clusters formed by the received data samples, via an unsupervised learning clustering technique, (b) labeling of the identified clusters by using a hidden Markov modeling of the process and (c) channel equalization by means of a cluster-based sequence equalizer. The performance of the equalizer is investigated for a variety of channels (minimum/non-minimum phase, linear/nonlinear channels). The case of channel order mismatch is studied. A semi-blind clustering-based equalizer, where a short sequence of known data is used for training, is also considered.