Bit-rate reduction for high-quality audio signals using floating-point wave digital filters

A source coding scheme for the low-bit-rate transmission of high-quality audio signals (16-bit quantizing resolution) is discussed. Encoding with approximately 3 bits per sample is possible without perceptible impairment of the signal quality. A subband coder with 27 bands is obtained by a cascade of digital filters implemented with floating-point arithmetic; the bandwidths are chosen according to the psychoacoustical `critical bands´ of the human ear, and the signals in the subbands are quantized with variable bit allocation depending on the masking threshold. If conventional digital filters are used in the arrangement, large parasitic oscillations and forced-response instabilities may occur. In the case of an implementation using lattice wave digital filters with a bireciprocal characteristic function, no precautions need to be taken to avoid instabilities