Cross-correlation for Dynamic Predictive Sampling Signals in Biomedical Signal Acquisition

This paper presents a novel method to calculate the cross-correlation of signals using dynamic predictive sampling. The primary application is to save computing overhead for wearable devices that monitor biomedical signals. Compared to the conventional Nyquist-rate data acquisition systems, dynamic predictive sampling saves data throughput by only recording the digital value of the turning points in the analog waveform. However, it generates a non-uniform sampled sequence with both amplitude and timestamp data. To calculate the cross-correlation of the recorded waveform, we propose an area-based method that utilizes the Jaccard Index instead of using interpolations. Based on our simulation result, compared to the conventional cross-correlation that uses interpolation for nonuniformly sampled signals, the proposed Jaccard Index method saves computing resources for dynamic predicted sampling sequences and provides better linearity.