Neural Network Based Lossless Data Compression Schemes For Telemetry Data
Data compression deals with removal of redundancy, reducing bandwidth and storage requirements, and thus lowering transmission and storage costs of data acquired from remote sensors. Some telemetry data are sensitive to inaccuracies and reguire lossless compression techniques to enable exact reconst...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2000
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Data compression deals with removal of redundancy, reducing bandwidth and storage requirements, and thus lowering transmission and storage costs of data acquired from remote sensors. Some telemetry data are sensitive to inaccuracies and reguire lossless compression techniques to enable exact reconstruction of the data at the receiver. One technology that has been successfully applied in a wide range of lossy compression applications is neural networks, a massively parallel system with pattern recognition capabilities. This report describes simulation results of implementing this technology for lossless compression of telemetry data. The general approach taken is to modify a two-stage scheme consisting of a predictor-encoder combinations in the past. Adaptations to enhance the scheme for the neural implementation are introduced. A variety of neural network models from a selection of different network types, including feedforward, recurrent and radial basis configurations are tested. Various schemes are proposed and implemented to determine suitable strategies for training the network. Characteristic features of the models, transmission issues and other practical considerations are taken into account to determine optimised configuration and implementation of the schemes. Variations of the two-stage scheme are introduced for both the traditional and neural schemes. Significant observations and results of the strengths and weaknesses observed amongst over 50 implementations simulated with the telemetry data in the report reveal interesting results for existing and proposed neural schemes. The distributed processing characteristic of neural models implies speed and error tolerance. Simulation results, combined with estimations are used for comparisons of these characteristics with the existing schemes. The findings and shortcomings of the research may be beneficial for future work, such as in the hardware implementations of dedicated neural chips for lossless compression. |
|---|