An optimal compression algorithm based on the principal components subspace projection transformation (PCT) was developed in Phase-I and extensively tested with HYDICE datacubes. The algorithm is designed to be implemented in hardware and operate with hyperspectral sensors for realtime data compression. Compression ratio is fully user controlled, and a ratio of at least 7:1 is achieved with PCT while reconstructed data fidelity is compatible with the sensor signal-to-noise ratio (SNR). Hence, as shown in the figure below, no practical data degradation is suffered. Much higher compression is possible with some reduction in the process SNR. An additional factor of 2 is obtainable with further compressed data encoding for a total compression ratio of 14.
The algorithm implementation with hyperspectral data is completely parallelizable using single program multiple data (SPMD) paradigm, it has been ported to run on a single and dual DSPs and all its basic arithmetic operations benchmarked on a TI C40 DSP. The execution timeline combined with a system model were used to establish DSP requirements.
A prototype system was designed for concept validation, test, and technology demonstration, based on the parallel multiple SHARC DSP architecture. As few as 3 to 5 present generation DSPís are needed to keep up with typical data streams (~100 MBits/sec). The Prototype is designed to operate with the TRWIS-III sensor as a model sensor, and during Phase-II the prototype hardware will be interfaced to the TRWIS-III. The prototype will serve as a basis for the design of an operational single board embedded systems.
The PCT was shown to be relatively insensitive to transmission bit errors. Telemetry errors in the transformation matrix can be detected at the receiving end and corrected in realtime. Bit errors in compressed image data only affect the specific pixel in which the error occurred.
Return to OKSI's homepageGo to TechExpo Homepage.