Compression of multidimensional biomedical signals with spatial and temporal codebook-excited linear prediction

Area of Science:

• Biomedical Engineering
• Signal Processing
• Data Compression

Background:

• Biomedical signals often exhibit temporal and spatial redundancies, which can be exploited for efficient data compression.
• Existing coding techniques may not fully leverage these multidimensional redundancies, limiting compression performance.

Purpose of the Study:

• To propose and evaluate a model-based lossy coding technique for multidimensional biomedical signals.
• To exploit both temporal and spatial redundancies for enhanced compression performance.

Main Methods:

• The proposed method utilizes a codebook-excited linear prediction approach, modeling signals as filtered noise.
• Temporal redundancy is addressed by filter modeling, while spatial redundancy is exploited across related signals.
• Waveform reconstruction is achieved by quantizing the power spectrum of the signal and residual noise using an algebraic codebook.

Main Results:

• Testing on multichannel electromyography (EMG) and electroencephalography (EEG) signals demonstrated significant improvements.
• For EMG, an 89% compression ratio yielded an SNR gain of over 3.4 dB compared to independent coding.
• For EEG, the same compression ratio resulted in an average SNR gain of 2.4 dB.

Conclusions:

• The proposed coding technique effectively exploits both temporal and spatial redundancies in multidimensional biomedical signals.
• This method offers superior compression performance in terms of SNR for a given bit rate compared to previous schemes.