PPG Signal Compression Based on Sparse Dictionary Learning

This paper presents an efficient method exploiting compressed sensing for monitoring photoplethysmography (PPG) signals. The method starts with the acquisition of PPG signals at the Nyquist rate, followed by pulse detection and systolic peak alignment. Subsequently, a deterministic block binary diagonal matrix is applied to PPG signals to compress before storage or transmission. Finally, the compressed data are reconstructed by means of sparse dictionary learning and the reconstructed frames are merged to generate the original PPG signals. Specifically, an over-complete dictionary is built on the discrete cosine transform, together with two optimized dictionaries, i.e., dictionaries reduced by column dimension. The experimental comparison reveals that the optimized dictionaries enable faster signal reconstruction than an over-complete dictionary, while maintaining the same reconstruction performance, assessed in terms of percentage of root-mean-squared difference.