This repository exists purely to analyze the effectiveness of various compression algorithms on payloads captured from Meshtastic by @anotherandrew in early 2025.
Some compromises were made to reduce the scope of this analysis, such as a simply not compressing anything that caused errors in a given compressor. Additionally, the training data for Huffman and aritmetic codecs could be be improved significantly. In both cases, training data was based off payloads from the text messages.
Huffman and arithmetic encoding were the only encodings that beat out uncompressed text in terms of mean size in bytes for text messages.
Arithmetic encoding was the only compression method that beat or matched uncompressed text in every test.
Aritmetic encoding
It might make sense to have separate code/frequency tables for each PortNum.