Analysis and comparison of three techniques for the compression of digital instrumentation data

When instrumentation signals are acquired and recorded digitally it is desirable to minimize the data storage space required. This thesis analyzes and compares three methods which were thought to be promising for compressing bandlimited smoothly varying instrumentation data with stationary characteristics over some block length to minimize the quantity of digital data which must be recorded.

The first method considered adaptively varies the sample rate by factors of two using real-time fast Fourier transforms (FFTs) to determine the minimum rate that captures the equivalent spectral components of the signal. To test the procedure, representative simulated and test signals were adaptively sampled and then restored to their original sample rate to examine the amount of data compression achieved and the accuracy of the restoration. The restoration accuracy is examined as a function of the threshold used in the frequency domain to define the portion of the spectrum with significant information content.

The second method records portions of the FFTs of the original digital signal. The portion of each FFT that contains spectral components above a threshold which defines the part of the spectrum with significant information content is stored. Signals were recorded in this way to examine the data compression achieved and were restored to the time domain to determine the accuracy of the restoration as a function of the frequency domain threshold.

The third method estimates the current value of the signal by a linear prediction from previous signal samples. The difference between this estimate and the actual sample is then coded with a Huffman coding scheme and stored to achieve storage reduction. The decoded difference signal can then be added to the predicted signal to reproduce the input signal with no error within the accuracy of the analog-to-digital converter.

All three data compression methods were compared with respect to the accuracy of reconstruction and the amount of compression attained. It was found that the method of linear prediction gives the best overall result with storage space reduction of about 2:1 on representative test signals. An important factor of merit for the third method is its ability to perfectly reproduce a signal with effectively no error within the accuracy of the analog-to-digital converter.