Performance of Karhunen¿Loÿve and discrete cosine transforms for data having widely varying values of intersample correlation coefficient

The capabilities of various transforms in respect to their `energy compaction¿ properties are well known. Their other important property¿the ability to decorrelate a data sequence¿has largely been taken for granted. In the letter, previously known results are extended to the case in which the optimum transform for a given correlation coefficient is used to transform data for which that value may vary widely. It is demonstrated that, although the decorrelation behaviour of the transform is markedly affected in such a situation, its energy compaction is virtually unaffected by the choice of correlation coefficient for which the transform is optimised.