AtomBeam proposes to incorporate a significant modification to its Compaction product, called count sketch, to enable Compaction to encode LiDAR data from UAS as well as other data streams. Compaction in its current release is capable of encoding most typical UAS data, other than video. AtomBeams process entails building Codebooks from representative sample datasets using machine learning. Codebooks are loaded on a source and a destination, and with the Compaction runtime executable, they act as translation dictionaries, allowing greatly reduced datastreams to be transmitted losslessly (fully reversibly) in two-way communication. Typical runtime reduction of data size is 75% for machine/IoT data. The result is an average 4x expansion of effective bandwidth for most users. For most data types, samples of 1 10 MB of representative data are all that is required to build a Codebook. In an average of 10 minutes in processing time, Compaction builds 50 Codebooks using machine learning, seeking the most efficient data pattern length to achieve maximum Compaction in runtime. LiDAR, however, and some RADAR data are very high volume and sample datasets can be multiple gigabytes in size, which is impractical for Codebook generation without modification of the Compaction code. Count sketch is a sophisticated mathematical approach for dimensionality reduction that can make Codebook generation from very large datasets practical. With this added capability, Compaction should be capable of building Codebooks from any LiDAR or RADAR dataset, irrespective of its size. This would enable the full suite of the UAS datastream other than video (which eliminates patterns by using lossy compression) to be compacted, and therefore effectively realize a 4x improvement in effective bandwidth. This Phase 1 STTR will utilize the combined expertise of AtomBeams mathematical and scientific personnel with the expertise of members of UMKCs faculty, who are experts in LiDAR and UAS data communications.
