MPEG Layer 2 Audio Coding (MUSICAM)
The MPEG layer 2 algorithm is the preferred algorithm for European DTV and includes a number of simple enhancements of layer 1 (or PASC). Layer 2 was originally adopted as the transmission coding standard for the European digital radio project (Digital Audio Broadcasting or DAB) where it was termed MUSICAM. The full range of bit rates for each layer is supported, as are all three sampling frequencies, 32, 44.1, and 48 kHz. Note that MPEG decoders are always backward compatible, that is, a layer 2 decoder can decode layer 1 or layer 2 bit streams; however, a layer 2 decoder cannot decode a layer 3-encoded stream.
MPEG layer 2 coding improves compression performance by coding data in larger groups. The layer 2 encoder forms frames of 3 by 12 by 32 = 1152 samples per audio channel. Whereas layer 1 codes data in single groups of 12 samples for each subband, layer 2 codes data in three groups of 12 samples for each subband. The encoder encodes with a unique scale factor for each group of 12 samples only if necessary to avoid audible distortion.
The encoder shares scale factor values between two or all three groups when the values of the scale factors are sufficiently close or when the encoder anticipates that temporal noise masking will hide the consequent distortion. The layer 2 algorithm also improves performance over layer 1 by representing the bit allocation, the scale factor values, and the quantized samples with a more efficient code. Layer 2 coding also added 5.1 multichannel capability. This was done in a scaleable way so as to be compatible with layer 1 audio.
MPEG layers 1 and 2 contain a number of engineering compromises. The most severe concerns the 32 constant-width subbands which do not reflect accurately the equivalent filters in the human hearing system (the critical bands). Specifically, the bandwidth is too wide for the lower frequencies so the number of quantizer bits cannot be specifically tuned for the noise sensitivity within each critical band. Furthermore, the filters have insufficient Q so that signal at a single frequency can affect two adjacent filter bank outputs. Another limitation concerns the time frequency–time domain transformations achieved with the wave filter. These are not transparent so, even without quantization, the inverse transformation would not perfectly recover the original input signal.