Data Compaction with Dynamic Multiple-sector Encoding

Yazdan Aghaghiri and Massoud Pedram Department of Electrical Engineering-Systems University of Southern California Los Angeles, CA Abstract In this paper, we introduce Dynamic Multiple-sector Encoding (DME), which is an irredundant encoding technique that can effectively exploit locality in different kinds of data streams. The basic idea is to have a set of registers called sector heads that are spread freely all over the source word space. There are three important facts about these sector heads. First, based on the position of these sector heads in the source word space, this space is partitioned into a number of non-intersecting sectors. Each sector is represented by its sector head, which will be used for encoding the source words that lie in that sector. Second, for any given set of sector heads and its corresponding sectorization, for almost any given source word, if we find the closest sector head, it is going to be the sector head of the sector in which the source word resides. And finally, sector heads are continuously updated with the source word values. In this way, sector heads track source words, i.e., sectorization is adaptively changed based on latest source words. As a result of the above facts, almost any source word is encoded with respect to a nearby previously-referenced source word. This technique can be targeted to either compaction of data streams or reduction of interpattern transitions for streams. In the experimental results, we show the effectiveness of DME when it is applied to three different streams namely, data address streams, image files and mixed data streams that represents sensor data communicated in a wireless sensor network. DME achieves an average of 68% transition reduction for data addresses, an average of 23% compaction for image files, and up to 19% additional lossless compression for sensor data.