Lossless compression of waveform data using multiple-pass adaptive filtering

Describes a novel approach to losslessly compressing one-dimensional waveform data using predictive filters in multiple passes. The authors are concerned with methods suitable for remote sensing. It is important to find ways in which this data can be compressed for storage and transmission purposes. In many cases, it is desired that this data be preserved without loss. This lossless compression is commonly performed in an algorithm with two stages. The first stage decorrelates the input data and then compression is achieved in the second stage where an entropy coder performs symbol coding. In their algorithm, integer valued input data is decorrelated with the floating point coefficients of predictive filter, but steps are taken to integerize the filter output such that a similar filter on the receiver side may reproduce the original data exactly. This research takes the two-stage scheme a step further by answering the question, "is the predictor as efficient in decorrelation as possible?" They subject the input data sequence and subsequent predictor output sequences to multiple passes of a predictor in an attempt to further decorrelate the data prior to the symbol coder. They show that improved lossless compression can be achieved using these multiple passes. They provide some comparative lossless compression results using several coding schemes and show that the use of multiple pass predictive filtering provides improved compression