Efficient health data compression on mobile devices

Abstract

There is an increase rise in the usage of mobile health sensors in wearable devices and smartphones. These embedded systems have tight limits on storage, computation power, network connectivity and battery usage making it important to ensure efficient storage/ communication of sensor readings to centralized node/ server. Frequency Transform or Entropy encoding schemes such as arithmetic or Huffman coding can be used for compression, but they incur high computational cost in some scenarios or are oblivious to the higher level redundancies in signal. To this end, we used the property of periodicity in these naturally occurring signals such as heart rate or gait measurements to design a simple low cost scheme for data compression. First, a modified Chi-square periodogram metric is used to adaptively determine the exact time-varying periodicity of the signal. Next, the time-series signal is folded into Frames of length equal to a pre-determined period value. We have successfully tested the scheme for good compression performance in ECG, motion accelerometer data and Parkinson patients samples, leading to 8-14X compression in large sample sizes (6-8K samples) and 2-3X in small sample sizes (200 samples). The proposed scheme can be used stand-alone or as pre-processing step for existing techniques in literature.