Audio decoding with frequency and complexity scalability

The paper describes a technique to design a complexity-scalable audio codec based on the partial reconstruction of the spectrum of a signal. The technique can be applied to any subband coding system based on cosine-modulated filterbanks. In the decoder, only a part of the subband is reconstructed, achieving an output signal with a lower sampling rate. The new synthesis filterbank, with a reduced number of channels and low computational cost, must be suitably designed. Hence, a computational saving is traded for lower quality in the reconstructed signal. The technique can be used to design low-complexity decoders compatible with the MPEG audio standard, i.e. able to decode only a part of an MPEG bit stream. In this way, a single MPEG-coded audio can be decoded by different receivers, i.e. standard and low-complexity ones. The problem of having fixed bit rate streams for all the decoders is discussed.