Adaptive downsampling of traces with FPGAs

Abstract

Modern nuclear physics experiments produce large amounts of data. It has become custom to record and store signal traces, which can benefit from compression algorithms. General-purpose compression methods do not exploit the characteristics of the traces and are ill-suited for implementation on Field-Programmable Gate Arrays (FPGAs). We propose a lossy algorithm based on downsampling. Data reduction is achieved by replacing variable-length groups of samples with their average. The lengths of the groups are chosen according to their significance, which is determined based on the estimated signal noise. This results in an adaptive downsampling of the signal trace. The compression routine also assists subsequent analysis, by reducing the stored signal fluctuations due to noise. A detailed algorithm description is provided, accompanied by openly available implementations for both FPGAs and PCs (VHDL and C++).