Blind deconvolution homomorphic analysis of abnormalities in ECG signals
Madiha Sabry-Rizk · 1995
This paper presents two different approaches to the problem of modelling and characterisation of the complex signal patterns observable in the electrocardiogram (ECG), with a view to open new signal domains for diagnostic purposes besides the time and frequency domains. The first approach is based on homomorphic prediction and involves the deconvolution of a non-minimum phase ECG signal into its minimum phase and maximum phase components through cepstral filtering and linear prediction of each component separately. The second approach is based on blind deconvolution and is primarily directed at utilising the Higher Order Statistics (HOS) of the ECG signal portrayed as a plant driven by a non-gaussian independent and identically distributed (iid) process and identifying both non-minimum phase and magnitude. Again, linear prediction may be employed but it is redundant to include inverse filtering (equalisation) as it serves no purpose from a pattern recognition view point. The paper presents simulated results and shows that it is feasible in future to develop higher order statistics algorithms based on a syntactic approach with grammar inference for each class pattern in different domains. Some results obtained from real data are also given.