Performance of the distributed KLT and its approximate implementation

The Karhunen-Loève Transform (KLT) is an important processing tool in data compression. Recently, a distributed version of the KLT was introduced together with an iterative algorithm for its implementation. Then, a recursive greedy algorithm for its approximate implementation was presented. Performance evaluations of both algorithms in the literature have been based on a few toy examples, which is not enough to show the intricacies of the distributed KLT. In this paper, a multi-terminal Markov-chain model is presented and used to evaluate the KLT algorithms using a large number of randomly generated covariance matrices. One of the findings is that the recursive greedy algorithm does not perform as well as predicted, and thus a variation of this algorithm is proposed which outperforms the original one. Finally, a promising application for both approximate algorithms is considered when the optimal distribution of the transmitted vectors among terminals is to be obtained.