Effects of Lossy Compression on the Value of Information in a Low Powered Network

Low power, long range wireless sensor networks (WSMs) have found many applications, but are often limited to small packets and low data rates. To improve data throughput, lossy compression can be utilized. This however, induces distortion in the data which can harm the accuracy of the underlying applications. This can be especially dangerous in safety critical applications, where incorrect classifications can cause serious harm, but a lack of new updates can be just as dangerous. In this paper, we provided an example application, nodes transmitting images to a back-end classifier. The effects of distortion on classification accuracy were studied, and different metrics were examined to better predict the usefulness of the data transmitted. These metrics were used to develop models to predict the VoI for a given compression ratio (CR). With these models, we developed a reward function to provide a trade-off between timeliness of data and the precision of the data. We used this to develop algorithms which can adjust to changing channel conditions to provide the most useful information possible, in a range of network traffic conditions. In this way, data quality can degrade gently as channel traffic increases. This provided more value to back-end nodes than with other methods tested.