What is Dolby AC-2 audio compression?

The Dolby AC-2 and AC-3 audio compression standards deal with compression of combined audio and video signals, and so synchronisation between the audio and the video information is important. The audio compression depends on matching the representation of the audio data to the sensitivity of the human ear, and is based upon a 'psychoacoustic' model of the human ear. Since the human ear responds to frequencies rather than to time domain signals, much of the processing is done by manipulating the data in the frequency domain.

The AC-2 algorithm encodes each audio channel entirely separately. AC-3 takes advantage of redundancies that may exist between channels in multitrack audio, and so achieves higher compression ratios.

The AC-2 algorithm takes audio samples and buffers them in blocks. Each block is multiplied by a window function before being transformed into the frequency domain. The frequency values are grouped into subbands that approximate the critical bands of human hearing. These frequency values are represented in floating point notation (an exponent and mantissa). The exponents collectively provide an estimate of the spectral envelope for the audio block. The mantissas are quantised, according to a quantisation scheme which is worked out from the exponents. So the quantisation of the mantissas depends on the spectral envelope of the block. The process is called a 'forward adaptive quantiser'. The final stage multiplexes the quantised frequency value mantissa bits with the exponent bits. The audio block size - and so the resolution - is varied, depending on the characteristics of the signal within the block. The analysis interval for AC-2 is 10 ms. During this time the algorithm determines the optimal block length to achieve the most compression. The model for this determination is based on a psychoacoustic model of human hearing. If the signals are continuous or unchanging, a long block size (512 or 1024 samples) is chosen. If the signals are transitory then a shorter block size (128 samples) is used. A 'block size controller' preprocesses each block of input samples and supplies a block identification number. While the block size controller is preprocessing the information a 64 sample delay is introduced. So the block size controller can anticipate coming events.

Note: Dolby's hardware design specification is based on a Motorola DSP56001 DSP processor with a 24 bit word length.