The Effects of Biomedical Signal Compression in Wireless Ambulatory Healthcare
Higgins, Garreth
Higgins, Garreth
Loading...
Publication Date
2013-10-30
Keywords
Type
Thesis
Downloads
Citation
Abstract
This thesis seeks to contribute to the field of ambulatory healthcare by examining methodologies to minimise the size of bioelectric signals, whilst preserving the quality of the diagnostic information contained within them. It begins by examining two compression algorithms using JPEG2000 and SPIHT based approaches. At low levels of fidelity loss, it was found that these algorithms could compress EEG data up to a Compression Ratio (CR) of 9. This level of fidelity loss was found to have little impact on the diagnostic information in the signals. Higher levels of compression were then tested, employing an automated seizure detection algorithm to analyse the loss in seizure detection levels. It was found that high levels of seizure detection performance were maintained with CRs of up to 90. An alternative approach to SPIHT-based EEG compression is presented whereby the level of quantisation is used to control the level of fidelity loss and SPIHT is employed as an entropy encoder. This approach was found to achieve substantial benefits in compression gains, achieving a CR of 100. It was also observed that this approach preserved the energy envelope of the signal more faithfully than other approaches. The final portion of this thesis focuses on protecting SPIHT compressed ECG signals from the impact of bit errors. The importance of the location of the error is first examined and it is found that the earlier an error occurs in a signal, the larger the impact on the reconstructed signal. This result is extended to determine the percentage of the compressed bit stream that needs to be protected to preserve signal quality at two operating points: "good" and "very good". This was found to correspond to 12.5% and 50% respectively. Finally, a methodology to provide this level of protection at Bit Error Ratios (BERs) of up to 10(-2) is presented.
Funder
Publisher
Publisher DOI
Rights
Attribution-NonCommercial-NoDerivs 3.0 Ireland