Bit-for-Bit Accuracy (Part 1 of 2)
July 10, 2007
The seemingly simple task of passing a bit-accurate signal from point A to point B is one of the most complex and difficult tasks when it comes to handling compressed digital audio. Carrying PCM audio is well defined by AES standards, and so is using the standard to carry non-audio data in place of PCM samples. The trouble is that the gear that implements these standards is generally either handling PCM audio or handling data. Expecting one type to work with the other type is the beginning of a long day.
PCM audio does not necessarily require bit-for-bit accuracy in the purest sense of the definition. If a slight gain change causes an offset of 1/1,000th of a dB, it will not be audible, but the path is not bit-for-bit accurate. Does this gain shift and loss of bit-for-bit accuracy then make much of a difference? Probably not. However, if audio were to pass through that path 3,000 times there would be a 3 dB gain change and this would be easily measured, observed and heard. Not likely to happen, but it could.
In the case of non-audio data being carried, that tiny offset of 1/1,000th of a dB is disastrous on the first pass. This is because the data has been modified, and will very likely appear scrambled and invalid to the gear expected to receive it. This is a common scenario for audio compression systems, which accept one or more PCM audio channels and output a compressed bitstream via an AES signal. The compressed data is carried in place of PCM audio data and must be conveyed perfectly intact or it will not be decodable. Many things other than simple level shifts can corrupt this data, and while it would not likely be audible for simple PCM, it will result in complete failure of compressed signals.