Wednesday 28 August 2013

Analog vs Digital

We are now over 30 years on from the launch of the Compact Disc and the digital audio revolution.  And still the argument rages on - is Digital better than Analog, or vice-versa?  The proponents of Digital can offer a seemingly inexhaustable list of solid technological arguments in favour of the superiority of that format.  Meanwhile, the proponents of Analog seem to have very little to offer in support of their position other than to point out that, well, it just sounds better.  Why, after all these years, has the argument not yet been resolved?

As someone with a foot in each camp I continue to look for answers.  I totally buy the technological arguments, yet the best that Digital has to offer does not yet to my ears sound as good as the best of Analog (although Light Harmonic's Da Vinci DAC sounds quite a lot better than my own dated analog rig).  The more I work with digital audio - most particularly with high-resolution digital audio - the more convinced I become that Digital holds the key to advancing the audio state-of-the-art.  There is absolutely no doubt in my mind that the best Digital totally out-performs the best Analog in every single quantifiable respect, mostly by a clear margin.  And yet…

The best amplifier technologies out there remain Analog (although the rapid rate of progress in Class-D digital amplifiers bears watching closely).  State-of-the-art analog amplifier design generally holds fast to a handful of core principles.  First, we reduce the number of components in the signal path as dramatically as possible.  For example, we reduce the number of gain stages to a minimum, we try to reduce negative feedback as much as possible, and we go to great lengths to eliminate capacitors from the signal path.  Second, we prefer to use discrete components over ICs.  This is largely because we can optimize both the specifications for, and the implementation of, every individual component, whereas in an IC we cannot even isolate any specific circuit element, let alone measure or optimize it.  IC designers are not shy when it comes to circuit complexity, whereas the best Analog designers inevitably come up with the simplest circuits.  Third, we focus 90% of our design efforts on the Power Supply.

But once you get the signal into the digital domain, whatever you do to it, however you choose to process it, it becomes possible to measure and/or quantify exactly the effect of your process with absolute precision.  The equivalent of a circuit element is an algorithm.  No matter how complex or simple the algorithm itself may be, we can totally and absolutely quantify its impact.  And that impact may be zero.  Of course, some of these analyses may be extremely challenging to perform - almost anything involving a DSD bitstream comes to mind - but the principle remains the same.  Digital data is absolute and finite.  It contains no more and no less information than it is known to contain.  And we can measure and observe every last bit of it.

This is very different from Analog.  We can never know or even characterize the sum total of the information contained within an actual analog signal.  The output of an amplifier is ephemeral - it exists only for that one instant and is then gone forever.  If we want to characterize the output of the amplifier we have to be able to recreate it on demand.  Also, the very act of measuring an analog signal perturbs it.  This is not important if the perturbation is swamped by the various inaccuracies or uncertainties (or even the noise) involved in the measurement itself.  But the fact remains inescapable.  There are things contained within an analog waveform that we can never hope to measure.

So, given that there is so much apparently in favour of Digital, how can Analog possibly sound better?  I suspect that a part of the answer may be down to something I have touched on in previous posts.  When an analog audio signal is digitized, the most common ADC technology employed is the Sigma-Delta Modulator.  While the output of a SDM is digital data stream, it is not in the PCM format.  So it is then passed through a conversion algorithm to produce PCM.  I have argued that this algorithm may have sonic consequences.  However, today I wish to shine the spotlight elsewhere.  I want to look more closely at the SDM itself.

Inside the SDM, the input analog signal first enters an integrator, which is basically a big capacitor that gets gradually charged up by the input signal.  The output of the integrator is then digitized by a coarse quantizer - often only a 1-bit quantizer - driven at an extremely high sampling rate (usually several MHz).  The output of the quantizer is then fed back into the input of the SDM where it is mixed with the input signal before feeding the integrator.  The output of the quantizer is also the output of the SDM.  What I have described would make for an uselessly noisy SDM bitstream.  To fix this, the SDM is "noise-shaped".  This is done by placing a low-pass filter before the input to the integrator.  This low-pass filter must be a high-order design in order to produce acceptable audio performance.  It is important to note that all of this is done in the analog domain, and is typically implemented on-chip as a single IC.

So, what we have is a SDM-based ADC, where the analog signal is passed through a multi-stage (high order) filter and an integrating capacitor.  All done inside an IC.  These are all things that Analog designers have for decades sought to eliminate from their very best designs.  In short, when we digitize a high quality Analog signal, we do so by feeding it into a circuit whose topology has a serious potential for degrading the sound.

I wonder whether anybody has set about designing an all-discrete, audiophile-grade, SDM ADC?  There is really no reason I can think of why it could not be done.  However, although consumer customers exist for cost-no-object DACs - seemingly at ANY price the industry is able to come up with - I don't know if there is even a single pro audio studio anywhere in the world which would pay five figure prices for an equivalent quality ADC.  It would be a mighty interesting product, though…