Low-Complexity Compression for Sensory Systems

Abstract

This paper presents a low-complexity mixed-domain data compression solution suitable for resource-constrained wireless sensory systems. Data compression reduces the transmission bandwidth of sensor nodes helping them to save energy and extend their operation time. In the proposed compression solution, the sensor signal is decorrelated in the analog domain and converted to digital using a compressing analogto-digital converter. The compressing converter is based on a cyclic converter architecture and is able to jointly perform the functions of quantization, signal conversion and entropy coding in a single circuit. Since data compression is performed as the sensor signal is acquired, there is no need to use a microprocessor or a dedicated circuit to compress the signal reducing the computational requirements of a sensor node. As a proof of concept the proposed data compression scheme was implemented using programmable hardware and employed to acquire and compress physiological and speech signals.

Variable length integer codes based on radix number system conversion

“…Huffman codes [2] and arithmetic codes [3] are the most widely used statistical codes for data compression. The main disadvantage of statistical codes is that two passes are required over the data to produce codes/codewords and this two‐pass approach is not efficient for low computational devices such as sensory systems [4] and storage devices [5], where the data needs to be read, compressed and written within one pass over input data. In contrast to statistical codes/coding methods, universal codes/coding methods do not require probability values of input symbols.…”