Clustering-based blind maximum likelihood sequence detection for GSM and TDMA systems

A novel blind maximum likelihood sequence detector (MLSD) for GSM and TDMA based systems is proposed. The baseband data at the receiver are partitioned into clusters that are identified using a new class of unsupervised clustering algorithms known as K-Harmonic Means (KHM/sub p/). The KHM/sub p/ algorithms arc insensitive to the initialization of the cluster centers owing to a built-in boosting function, and thus provide reliable estimates of the cluster centers. The identified cluster representatives are then mapped to the corresponding combinations of input symbols using a discrete hidden Markov model formulation of the channel states and the mapping is used to compute the branch metrics in a cluster-based MLSD to perform signal detection. The proposed detector avoids any explicit channel modeling and training overhead and its performance is evaluated for the GSM systems.