Tabled Asymmetric Numeral Systems compression for the Laser Altimeters on ESA's BepiColombo and JUICE missions

Abstract

The Laser Altimeters BELA and GALA are en-route to Mercury and Ganymede. A short laser pulse between 10 and 48Hz is emitted and sampled upon return with a single-pixel high speed camera (APD) running with 200MHz (GALA) or 80MHz (BELA) to get accurate time-of-flight measurements and therefore range. An on-board FPGA selects 4 short windows of 64 12-bit samples from an ADC per shot as candidates and transmits these to the central processing unit for further inspection. Now a Gaussian fit takes place and only the fitted center position, pulse width and amplitude are transmitted back to Earth to keep data volume low. This enables high detection accuracy in the decimeter range despite the low sampling frequency.

Unfortunately, secondary science goals such as roughness and slope get by just transmitting fit information. Moreover, even higher accuracy can be achieved if we account for the non-Gaussian nature of laser pulses. As it became evident that examining raw data while maintaining a low data rate is crucial, implementing a compression method that our central processor can handle has become a priority.

When dealing with noisy sensor data, entropy coding emerges as the method of choice. To leverage the existing Gaussian fit, we subtracted it from the signal to decrease the remaining symbol count. The remaining samples could then be compressed by a factor of 3.5 using an implementation of the Tabled Asymmetric Numeral Systems (TANS) encoder. We provide a comprehensive comparison with Huffman coding, as well as detailed implementation specifics and performance metrics for our C version on both the Gaisler GR712RC in-flight CPU and Commercial Off-The-Shelf (COTS) hardware.