Some Mathematics Behind Multi-Channel Prediction
Internal Technical Memo, Sept. 1994
Universität Hannover, Institut für Theoretische Nachrichtentechnik und Informationsverarbeitung
Multi-channel prediction is a technique to exploit redundancy between different channels of a multi-channel audio signal. Thus, the coding of these signals can be improved.
The basic idea is to calculate an estimate z^(n) for the signal z(n) in a channel z based on the signals x(n) and y(n) in the channels x and y. If desired, the prediction can also be based on only a single signal x(n).
When using prediction, instead of the signal z(n) only the predictor error signal
e(n) = z(n)-z^(n)
has to be transmitted. The parameters describing the applied predictor (predictor coefficients etc.) have to be transmitted as side-information.
In the decoder, x(n), y(n) and the transmitted predictor parameters are used to calculate z^(n) as in the encoder. Together with the transmitted predictor error signal e(n), the original signal $z(n)$ can be reconstructed.
Prediction can be applied independently for several channels z1, z2, ... if desired. If a subband-based coding scheme is used, prediction can (and should) be applied independently in the different subbands.