Lossless Compression of Time-Series Data Based on Increasing Average of Neighboring Signals

摘要

Golomb-Rice encoding is a well-known compression algorithm for sensor data. When time-series data change drastically with large amplitudes, as found in a pulse signal, the code length based on Golomb-Rice coding becomes long. In order to shorten the code length, the amplitude of the signal is decreased by calculating the differential signal between a raw signal and a similar signal. In this paper, we develop a lossless compression method for time-series data such as sensor data. In traditional methods, finding the past signal from which a differential signal with low amplitude can be generated is the main topic. However, if there are no past signals that can be used to sufficiently reduce the amplitude of the differential signal, the data compression procedure has little effect. In our approach, a signal that decreases the energy of a pulse signal or increases the energy of the neighboring signal of a pulse signal is used to generate differential signals. In order to select an effective signal, we propose a method for detecting reference signals based on the cumulative distribution features of the time-series data. Experiments confirm that the proposed method can generate codes whose length is shortened. The code length was decreased to 97% on average and to as little as 81% in comparison with the traditional method.