| Resumo: | This paper describes a lossy ECG signal coder with an adaptive predictive coding scheme initially proposed for speech coders. The predictors include linear predictive coding that takes advantage of the correlation between consecutive samples and long-term predictor that takes advantage of the signal quasi-periodicity. The prediction residue, with less dynamic range and therefore able to be encoded with less bits than the original, is transmitted sample by sample. The prediction coefficients and the amplitude of the residue are transmitted once for each heartbeat, with a negligible number of bits compared to the total bit rate. The long-term predictor is shown to obtain reliable performance when the heart rate does not change rapidly. Linear predictive coding, on the contrary, is more reliable and presents better prediction gain. The best developed coder uses double prediction and with 45% compression ratio allows a prediction gain of 24.8 dB.
|
|---|