Effects of Non-Uniform Quantization on ECG acquired using Compressed Sensing

This paper analyzes the effects of quantization on Compressed Sensing (CS) measurements applied to Electrocardiogram (ECG) signals. Two methods of quantization are proposed in this paper: uniform and non-uniform. Reconstruction is performed using a dictionary based on the Mexican Hat wavelet. A distortion-based performance metric Percent Root-mean-squared Difference (PRD) will be monitored at various Compression Ratios (CR) to quantify the impact of quantization. The energy cost of transmission is also evaluated for different levels of quantization and compared, at certain PRD levels. The results demonstrate that non-uniform quantization outperforms the uniform approach and that employing non-uniform quantization improves implementation efficiency for applications with acceptable PRDs above 6.75%. Results show that utilizing non-uniform quantization can increase the CR from 9.8 to 14.1 for a PRD of 30%. Furthermore, this amounts to a 28.91% reduction in wireless transmission per frame from 37.7 µJ to 26.8 µJ considering Bluetooth Low Energy (BLE) as a target wireless communication protocol.