Adaptive prediction
In digital signal processing, adaptive prediction refers to an adaptive filter that uses a reference signal to make the best possible prediction for a given signal.
Prediction methods are also called decorrelators because they subtract the predictable signal components from an input value and thus significantly reduce the dynamics of the signal. This means that the signal can be transmitted on a transmission path with fewer bits than the original signal. On the receiving end, the decorrelation must be reversed.
The predictable signal components are described by prediction coefficients, which are also transmitted to the receiver. These coefficients are recalculated over and over again for small signal sections and thus "adapted" to the input signal.