Folding: a method for semantic encoding of error-tolerant data

Abstract

Underwater acoustic sensor networks are characterized by both low link data rates, and very low data generation rates of the sensors. In this regime, Shannon capacity results, which presume long channel codes and an infinitely long information bitstream, are not directly applicable. Further, for scientific data collection, distortion and errors are tolerable at the semantic layer. For this regime, we formulate the problem of sending k successive source symbols using n successive modulation intervals, where n > k. We introduce "folding" as a technique to map a k-dimensional source manifold into the n-dimensional modulation space, in order to minimize the average energy consumption per source symbol. We use an Archimedes' spiral, ahelix, and Fermat's spiral, as good foldings for low-dimensional mappings, and compute the energy consumption per source symbol under each.