This was of concern to me. It raised the question:
“Are all our recordings of inferior quality? Is there anything one can do to prevent signal deterioration?”
Stav´s thesis: analogue has better resolution than digital in the range that matters.
Stav draws an analogy for the resolution of a signal over its dynamic range by showing a picture of a skyscraper.
For analogue signals recorded onto tape, the bottom of the building is blurred (noise), and also the top (peak saturation) (Fig.1). The part, where most listening occurs (the dynamic middle part), is crisp.
The picture for the digital analogy has the top sharp and becomes increasingly blurred towards the bottom. (Fig.2)
He says that he perceives smaller signals on digital less clear than on an analogue version. He quotes support by colleagues, who think the same.
I wrote to Stav, and he answered (I have to paraphrase here since this stems from private conversation):
Postulate: A signal recorded at low volume is represented by a lower number of bits than the same exact signal recorded at a higher volume. It thus has less resolution available. If you increase its volume, the steps that represent the signal become coarser, like a picture becomes pixilated when zooming in.
I was confused over this, since I have just watched “Digital Show And Tell” by Monty Montgomery from Xiph.org, who shows that there is no difference between an analogue signal and a digital signal, if properly handled. Also, there is no steps, and there is actually no square pixels in an image. (I wrote to him, but no answer at the time of writing.)
I waited a few days and wrote to Bob Katz, whom we all know from his book “Mastering Audio”.
For me, he is the Guru for digital recordings.
He answered promptly, and here is his answer (paraphrased again to respect privacy)
This is all covered well in the book. Digital audio, if properly dithered, behaves exactly like analogue There is no loss of resolution down to the very smallest signals, which drown in noise.
What Stavrou talks about is a poor implementation of digital audio.
A good implementation thereof can be expected to behave better than analogue, since the noise floor is typically lower than that of analogue circuitry.
I went and read the relevant chapters in his book again.
He says the following in short:
When a signal is digitized, i.e. converted from analogue to digital, some sort of dither has to be added to minimize digitization artefacts. This is a small random noise signal, mostly frequency shaped, which dramatically reduces digitization noise and distortion.
(This all totally coincides with the message of Monty´s video. I recommend watching it).
While the noise signal itself is inaudible, the artefacts might not be, particularly if they add up during boosting processors.
This process of dithering has to be done during any conversion where the word-length is changed, for example, during A/D conversion or during down sampling from 24 bits to 16 bits. (For any input hardware we may have, we have to rely on the manufacturer that they have done it right. Adding noise afterwards is futile...)
It certainly has to be done whenever rendering a project or whenever printing a file (bouncing).
Within the DAW, we have 32 bit float values, so this should theoretically be unproblematic.
However, it is not guaranteed that any plug-in works with 32 bit float values internally. It might use fixed word-length computation internally, bigger word-length and/or oversampling, which means it must truncate at the output. Now for this, we are at the mercy of the programmer. Again, there is no use whatsoever to dither afterwards.
But herein lies a danger. In previous threads it has been stated by the Mixcraft team that you virtually cannot overload the DAW, as long as you stay in the box. You can literally boost a signal by a ridiculous amount and then attenuate it again by the same amount and hear no difference.
While this may be true for the DAW itself, there is no guarantee for the plug-ins to behave the same. Be aware of that and stay cautious.
It has also been mentioned by the Mixcraft staff in a thread on dithering:
Mixcraft doesn't include dithering because the creators of Mixcraft feel it isn't necessary to obtain the highest quality audio. Dithering works by introducing random noise to the audio, which may or may not cover up any rendering artefacts There may not be any rendering artefacts in the first place and you'd just be adding random noise for nothing.
Now this is where I feel they are wrong. There will be rendering artefacts as the above sources have proven. While those may not be readily audible, they may be after some cumulative processing. As mentioned before, there is nothing we can do any more to a signal that has been converted by our input hardware. There is nothing we can do for a plug-in processor if it does word-length manipulation. However, any signal leaving the DAW should be dithered properly.
So, back to Stav and his skyscraper.
To sum up Bob Katz´ words, improperly converted signals and noise summing up over the plug-in queue´s amplification processes can put a veil over small signals, and that is what Michael Stavrou probably encounters. So Stav may be correct in his perception, but not in the statement that this applies to digital audio in general.
In a nutshell
So what can be concluded for us project studio heads from all that:
* Digital is no worse than analogue. It is most likely better for S/N ratio.
* There is no need for hot tracking levels like in analogue days. Stay conservative in recording levels. Within sane boundaries, there is no deterioration of signal quality and S/N ratio to be expected.
* Dither whenever you are truncating. This has also been recommended here, but now we know why.
* Do not rely on the DAW´s 32 bit float overload proof capability. The DAW might be safe, but not necessarily the plug-ins you use.