Time and Frequency Domain Compression Techniques for ECG signal

• Main Authors: LIN, I-FENG, 林宜鋒
• Other Authors: See-May Phoong
• Format: Others
• Language: en_US
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/78182849588254023816

碩士 === 國立臺灣大學 === 電信工程學研究所 === 89 === One of the main techniques for diagnosing heart disease is based on the electrocardiogram (ECG). The monitoring ECG device is usually required of ECG information in long-time. If efficient compression methods are employed, memory requirement can be reduced significantly. Many algorithms for ECG compression have been proposed. ECG compression has also been divided into three functional groups – direct methods, parameter extraction methods and transform methods. Thought direct methods are very simple, their PRD performance is in general not very good, especially at low bit rate compression. In this thesis, we focus on parameter extraction and transform methods. They have a good PRD performance at low bit rate compression. The most effective parameter extraction method is based on linear prediction. It belongs to the class of the time-domain compression. The short-term with lossless long-term predictor and short-term predictor with first-order prediction have good performances for normal and abnormal ECG signals. The frequency domain compression technique is better than the time domain compression techniques. The DCT, DWT, and SPIHT coding are studied in this thesis. The DWT with SPIHT coding has a good performance. The ECG signal is the periodical signal, so we apply the long-term predictor before the transform coder. The DWT with long-term predictor has a superior performance. The two-dimension frequency domain compression technique is also studied in this thesis. The 1-D ECG signals are first cut into several beats and then aligned into a 2-D array. By doing so, the 2-D data compression technique can be applied. For normal ECG signal, the 2-D method is better than 1-D method. For abnormal ECG signal, the 1-D method is better than 2-D.