ECG data compression by modelling of subbands

Some novel approaches to the compression of ECG signals, based on filter banks and transforms used in conjunction with parametric signal modelling are proposed. Quite recently the use of ARMA models for representing the transform coefficients obtained by applying large size discrete cosine transforms (DCT) to ECG signals has been examined (see B. Madhukar and I.S.N. Murthy, Computers and Biomedical Research, vol.26, p.310-17, 1993). The authors extend and generalize the ideas put forward by Madhukar and Murthy by: (1) Allowing the algorithm to work on arbitrary ECG signals-thus not making any assumptions about the suitability of the ECG signal to be split into so called beat-synchronous frames; (2) By viewing the DCT as a member of the class of perfect reconstruction filter banks the authors examine alternative transforms/filter banks, and finally; (3) The authors propose a scheme for selecting an optimum filter bank/transform in the proposed coding scheme. It is concluded, somewhat surprisingly, that the DCT outperforms the more common filter bank choices known from the subband coding literature. However, the authors find that optimizing the transform, rather than just selecting the DCT, gives rise to improved coding performance.<>