Improving efficiency of wireless sensor networks through lightweight in-memory compression

Data compression is an enabling technique to many applications, such as data center storage, multimedia streaming, and lightweight computing platforms, amongst others. These special-purpose compression approaches typically achieve incomparable compression ratios as core features of the application and dataset are leveraged to reduce duplication of identical or similar features in the original data. Sensor data in systems such as wireless sensor networks, often includes a variety of data types. However, within a particular data or sensor type, typically the data has a low dynamic range which can be leveraged to increase its compressibility. In this paper we present a memory and network system co-design approach that stores data using in-place lightweight compressed pages in memory, and utilizes this compressed data to send shortened blocks over a wireless point to point network. Additionally, we propose a technique named source-aware layout reorganization (SALR) to improve the compressibility of the sensor data, using either software- or hardware-based approaches. We demonstrate that our proposed lightweight compression approach in hardware with SALR, while achieving a slightly lower compression ratio to traditional software compression, can outperform software compression in wireless communication by 7.3% for relatively slow Bluetooth links and 65.4% with faster WiFi-Direct links.