Abstract: Superior performance, better tracking, and stability of an adaptive predictor are realized by employing a so-called high order (eighth) prediction filter and by updating the filter coefficients in accordance with an algorithm based on the sign of a reconstructed signal sample (s.sub.q (k)) multiplied by a quantized version (e.sub.q (k)) of a differential error signal normalized by an expected magnitude of the reconstructed signal sample.

Abstract: Accumulation of distortion possibly resulting from multiple digital code conversions is prevented by including in an ADPCM-to-PCM decoder (100, FIG. 1 or 600, FIG. 6) an arrangement (119, FIG. 1 or 601, FIG. 6) for making a trial transmission of the resulting PCM sample (.PSI.(k)) to obtain (via 209, FIG. 2) a composite noise value (.alpha.(k)) for the PCM sample (.PSI.(k)). The composite noise value includes noise components resulting from both quantizing the code word and truncating digital signals used in the decoder and subsequent coder. The composite noise value is evaluated (via 211) to determine whether adjustment of the PCM sample value (.PSI.(k)) is necessary.In one embodiment (FIG. 1, 100), a reconstructed linear version (Y.sub.r (k)) of the PCM sample is adjusted (via 216, 218, 220 or 221) when the composite noise value (.alpha.(k)) is not within the ADPCM quantizer interval for the ADPCM sample (I.sub.n (k)) being converted to PCM.