Data and signal: are we progressing?

Apparently, an old “bits is bits” issue is repeating from time to time, so sometimes it looks like we did not move at all. So let me try to rephrase this problem, supposedly from year 2013 point of view.

The term “data” denotes ones and zeroes i.e. bits. Normally, we want reliable system, the one that can preserve data as such, so it remains unaltered when it is used or moved. Fortunately, as long as we talk about data rates required for audio, today’s technique easily preserves data integrity, and in fact even early CD had quite reliable data coding / reading / correction process, so even early CD system was already data perfect (“bit perfect”).

So usually bits IS bits – all zeroes are equal, and all ones are equal too. However, it is like that only until data remains in the digital domain. The harder part comes from our inability to hear this digital data directly, and once we have to do the things to convert it into the analog domain. And this is where basic data integrity requirements are getting insufficient.

Namely, once you start operating with digital data inside the hardware, data will be represented by some waveform. Zeroes will be “low” state (0 V), and ones will be “high” state (usually equal to the digital supply voltage). And it is known that the waveforms usually suffer of different kinds of inaccuracies and artifacts, such as noise, overshoot, ringing, finite slope, ground bounce, crosstalk, jitter, and alike, caused by hardware imperfections. It is normally considered that all these imperfections can not violate data integrity so long as error does not violate “low” and “high” state threshold requirements in voltage domain, or exceed one half of signal interval in time domain.

But as said, that is the easy, data integrity part. The harder part is that, since our converters operate in real time, these inaccuracies in the digital domain can, and will break thorough to the converter analog output.

For this purpose, you can think about converter being a black box with two sort of inputs:
(1) digital data (zeroes and ones) input, and
(2) triggering signals (clocks) inputs, which are necessary because, as said, conversion is real time event.

In other words, once D/A conversion takes its part, you don’t have anymore a 16 or 24 bit data written in the disc, and protocol that “knows” which 16 or 24 bits are one word, but you have a real time stream of bits, with other time signals (clocks) that tell converter the moments the single bits (zeroes and ones) arrive, the moments the word starts or finishes, and consequently the moment the output audio analog signal has to change its level. That is why this last clock signal, the one that triggers converter’s output, is essentially fully analog – its properties directly translate into the resulting audio output.

Among these “properties”, jitter or phase noise became the most known recently. Ideally, triggering clock signal would come in perfectly equal time intervals, otherwise its time uncertainty produces frequency modulation of audio signal, a.k.a. jitter. As said, in purely digital system any time uncertainty that stays below one half of time interval is acceptable, but in D/A conversion even picoseconds of such uncertainty can be audible. Normally, generation of this clock is a part of data reading process, and the way this clock is managed in (through) the system hugely determines its sonic qualities. But problems are not at all limited to jitter, as waveform of data input signal can also bring a classic noise into converter, and translate itself into its analog output.

In fact, the whole science lays behind such issues. Many phenomenons related to so called mixed signal circuits (digital + analog) became recently well understood and documented, and many problems are seriously taken into account in order to solve them, instead of denying their existence, or their audibility. Thus today we can differentiate between good and bad clocking schemes, and we know why and how we should design for low jitter clocks, or low noise or low RFI emission inside the units.

For us, sonic features are always what ultimately matters, but this also means that these issues are not only audible but also measurable. Even more, some recent findings in this domain opened the whole new path to understand relations between subjective perception and particular forms of distortion and noise.

Finally, notable part of improvements achieved in the digital audio in last two decades happened right here, in the area of designing for higher signal integrity criteria. So not that much for instance in the area of converter chips, which were during this time developed mostly for lower production costs. As a result, even if Philips D/A chips from 80’s are in many areas still unsurpassed, today it would be unimaginable for any serious design to repeat layout or clocking schemes as they were done in Philips CD players from 80’s (or majority of players from that time, for that matter).