Patent US 4,375,013

A digit rate reducing system for PCM signals comprises a predictor stage, an automatic gain control stage, and a quantizer stage for adaptive quantizing. The predictor stage includes a predictor and a subtractor for subtracting from a sample x.sub.n of the PCM input signal a predicted signal x.sub.pn derived from preceding samples. The subtractor provides a difference signal d.sub.n representing prediction error to the automatic gain control (AGC) stage. A divider in the AGC stage divides the difference signal d.sub.n by m, providing a divided signal e.sub.n to the quantizer stage. In this last stage, a quantizer encodes the divided signal e.sub.n into a signal Y.sub.n of reduced rate. The predictor uses a first adder which receives the predicted signal x.sub.pn and a reconstituted signal d.sub.n derived from the reduced rate signal Y.sub.n ; a plurality of shift registers and associated multipliers; and a second adder connected to the outputs of the multiplier. Shift registers T.sub.1 . . . T.sub.N are coupled to the output of the first adder, and provide a series of delayed signals to multipliers M.sub.1 . . . M.sub.N for multiplying by respective coefficients a.sub.i (n) derived from the preceding samples. At least one of the predictor coefficients a.sub.i (n) is applied to the quantizer, whereby the type of coding is adapted to the statistical properties (ie, the conditional probability distribution curve) of the PCM input signal.