Signal Compression by Linear Prediction the Oracle Gang's Elec301 Project

Contents [Home] The Problem The ARMA Model Autocorrelation What we did The Data Conclusions A Word on Decoding Acknowledgements Who we are

Why Signal Prediction?    Signal prediction is about more than predicting the next value of the stock market. It has applications in many different fields. For example, it can be used for compression, especially in the case of speech signals.    Speech signals have a high degree of periodicity, making them easier to predict. Basic signal prediction takes past and current values of the signal and weights them with different coefficients to estimate the next value of the signal. The difference between the original signal and the predicted signal is called the prediction error. Transmitting only the error and the coefficients would allow an inverse predictor at the other end of a transmission line to reconstruct the signal without losing information.    With any decent predictor, the magnitude or range of values of the error will be much less than that of the signal. This is where the compression comes in. Signals are sent in digital form to reduce the effect of noise. Analog to digital conversion uses quantization levels to determine the degree of digital resolution and the number of bits needed to transmit each sample of the signal. If the prediction noise has a smaller magnitude than the original signal, for the same resolution fewer bits would be required, making the transmitted signal smaller and therefore faster to send than the original signal.

Copyright (c) 2000 by the Oracle Gang.