The transform encoder / decoder system utilizes carefully designed analysis / synthesis window pairs, adaptive transform coefficient quantization, and adaptive bit allocation to encode high quality music, including broadcast and compact disc applications, at low bits (128 kilobits per second).


The analysis / synthesis window pairs are designed to achieve sufficient conversion frequency selectivity to improve coder performance for exploiting psychoacoustic masking effects. The synthesis window of the decoder is characterized in that the product of its response and the product of the analysis window of the encoder generate a composite response that sums up to be integral for two adjacent overlapping sample blocks. Adjacent field domain signal sample blocks are superimposed and added to remove the effect of the analysis and synthesis windows. Techniques for deriving appropriate analysis / synthesis window pairs are provided. For an encoder, discrete transforms similar to critically sampled single sideband analysis synthesis systems generate frequency domain transform coefficients. The two transforms used are well stacked time domain aliasing (TDAC) and oddly stacked TDAC transforms.


The transform coefficients are non-uniformly quantized by assigning a plurality of fixed bits and a plurality of variable bits that are optimally determined according to psychoacoustic masking. Techniques for adaptive bit allocation and fixed bit allocation are described. Transmission of side information regarding adaptive bit allocation is unnecessary. Error code and guard data may be scattered across the formatted frame output from the encoder to reduce sensitivity to noise beasts.