| Summary: | 博士 === 國立交通大學 === 電信工程研究所 === 104 === Wireless patient monitoring has been of recent interest to academic and industrial circles with the goal of ubiquitous healthcare services. The purpose of this dissertation is to investigate three important aspects of a mobile ECG signal monitoring system: data compression, robust data transmission, and biometric identification. In this study, we present a novel means of exploiting the distributed source coding (DSC) in low-complexity
encoding of ECG signals. We first convert the ECG data compression to an equivalent channel coding problem and exploit the convolutional code for practical DSC construction. Performance is further enhanced by the use of a correlation channel that more precisely characterizes the statistical dependencies of ECG signals. Also proposed is a modified BCJR algorithm which performs symbol decoding of binary convolutional codes to better exploit the source’s a priori information. A complete setup system for online ambulatory ECG signal monitoring via mobile cellular networks is also presented. As a further step toward increased robustness against transmission errors, we also investigate the noisy-channel DSC problem for ECG data compression in conjunction with variable-length codes (VLCs) and channel codes. Using the concept of extrinsic information transfer (EXIT) from Turbo codes, we present a symbol-level iterative source-channel decoding (ISCD) algorithm for reliable transmission of variable-length encoded ECG data. Firstly, an improved source a posteriori probability (APP) decoding approach is proposed for packetized variable-length codes. Also proposed is a recursive implementation based on a three-dimensional (3-D) joint trellis for symbol decoding of binary convolutional codes. APP channel decoding on this joint trellis is realized by modification of the BCJR algorithm and adaptation to the non-stationary VLC trellis. The proposed symbol-level ISCD algorithm allows the receiver
to exploit the source residual redundancy as well as the channel code redundancy to the fullest extent as it avoids the conventional symbol-to-bit probability conversion problem between the two constituent decoders. Another important issue to address is the demand for improved security and privacy in wireless telecardiology applications. To this end, we propose a novel ECG biometric system which performs person identification
using content-based image retrieval (CBIR) techniques. To proceed with this, 1-D ECG signals are converted to 2-D images and afterwards part of the JPEG2000 encoding process is applied. Features relating to ECG morphology are then computed directly from the DWT coefficients and applied for indexing person identity by texture content in an enrollment database.
|
|---|
