Sunday, June 18, 2017

Peterson Information Theory

(Named after me, of course.)

In Peterson Information Theory, the composition of a composite signal is the important thing, not what can, under some circumstances, be reliably detected later.   In other words, the information is what goes in.  A fairly intuitive notion, actually, until about 1900, when the importance of the observer became paramount in many fields.

So this is similar to how many subjectophiles believe that analog has infinite resolution, whereas digital has finite resolution.  However, that intuitive believe, in the full context of PIT becomes very qualified.

For one thing, at worst, properly implemented digital has incredibly fine resolution, about 100 times finer the the apparent quantization, for various reasons, which in the case of 24 bit digital is pretty mind boggling and probably adequate to be considered "infinite."

For another analog itself is limited at many levels, even if you discount "noise" as PIT tends to do.  Noise does not "obscure" simply because PIT doesn't involve itself with "obcuring," there is a kind of belief that in the infinitude of listenings, all the obscuring will be seen around in one way or another, freak unrepeatable occurances perhaps but important to the evolution of feeling.

One obvious level of limitation of the supposedly unlimited resolution is the quantum level, though that is quite far away.  But any reduction in voltage, say, does actually reduce information in that way, and it turns out, in others.

Many aspects of circuits we deal with actually do not involve fully continous phenomenon.  For example, both magentic tape and analog recordings do not slice at quantum levels, even at best they slice at molecular levels.  So now we are up quite aways from "infinity" or even the 10e-33 of quantum levels.  Infinite resolution there never is.

But I'm not claiming infinity, just a different way of scoring.

And so it matter that the DAC actually attempts to do 24 bits, or just fakes it.  No question about it, true PCM chips like 1704 "attempt" to do 24 bits in that there are resistors that get switched in and out to do that.  The attempt falls far short (perhaps 20 bits or so actually accuracy) because thermal effects and others cause everything not to be as perfectly calibrated as when the manufacturing was done.  But in PIT, that doesn't matter as much as the information not being there, for that lack of 24 bit performance is caused by an "obscuring," not by a lack of doing.

But all kinds of Sigma Delta systems never ever try to "do" high accuracy.  They fake it by doing low accuracy lots of times with feedback so they ultimately achieve something like high accuracy.  But that is what PIT views as not actually doing it, but faking it.

When things are so faked, they lack their attachment to reality, they become unreal in one way or another.  So Sigma Delta reproduction is cleaner than clean, cleaner than the real thing itself, because it is built on repeated self cleaning, rather than just doing things right, or as best as possible, in the first place, with maybe a little crud added here or there in the process.

When we demand that each layer of a system be truly crud free, we cannot achieve what we set out to do, the crud creeps back in some other way.  Best to create processes which create only very little crud in the first place.  So we have balanced circuits, cascodes, Class A operation, differential, etc.  Those are all low crud.  Feedback is only a way of taking errors in one domain and shifting them to another domain.  It may be needed in some cases, but best be low crud in the first place.

Anyway, that's my unscientific irrationality spelled out, and why I like no feedback (or non-global at least) and true PCM such as R2R type DACs.

That's why I wanted a Master 7 Singularity, not because I wanted to experiment with NOS or I2S.

The Master 7 has a highly sophisticated non-feedback direct coupled semiconductor circuitry, combined with dual differntial 1704's--a configuration which really helps pull out more of the true 24 bit "doings" going on in there.

Mind you, op amps are so small, feedback around them is probably ok, not unlike local degeneration. The only op amp that should be used is the best, OPA 211.

Tubes are mostly used as coloring agents, not to make things more transparent but to add tint or bloom.  Tubes can in some cases be done at the state of the art, but that is exceedingly rare.  I know of some almost unique tubes and designs up there.  Allen Wright's balanced RTP3D preamp looks pretty good, that's the least obscure, then there are some ultra rare high current microwave tubes, rarely done (but never to state of the art, which would include full balanced) an extremely tricky to work with,  and balanced OPA 211 with passive EQ between amps (no product that I know of does that).

Oversampling is not bad because that's not feedback, at most it's feed forward, or more technically, I see it as the ultimate end-procedure of the digital process, which helps enable the full potential dynamic range.  It's not endlessly overloading tail catching feedback.

Strangely 1 bit may be sort of ok (at DSD128 rates) because it is actually doing what it is doing, one bit is a perfect 1 bit.  Any multibits until full multibit is only an approximation, which must endlessly dither to serve the present physical world, and therefore strangulate the past with the present, rather than letting the past flow freely in.

*****

Thinking about the above, I wonder if I haven't become something like a Schitt salesman.  They are also in the anti-delta-sigma camp.  I don't think ANY true audio objectivists I know (about) would consider this anything but the usual subjectivist crap.  To a fully died Objectophile, all that matters is the (measured) distortion and noise, and getting there the cheapest way possible, which is delta sigma.  Delta sigma dacs became popular when they could approach and then ultimately exceed the performance (S/N, dynamic range, linearity, and distortion) of the R2R chips.  And that became the big commercial end of the Burr Brown PCM 1704, though it appears they were still making a few as late as 2010 or so still being under intense demand by high end companies.

There is a growing list of High End audio companies that have embraced the pro-PCM and anti-Delta-Sigma ideas like I have presented, including:

Schiit (their statement DACs are all PCM and they strongly defend this)

Audio GD (they had their feet in both waters, but are seem to have more fully embraced PCM solutions now, if nothing else to own the low cost cult market)

MSB (they make multichip PCM converters from high to super high prices)

CH Precision (with their megabuck products, you can have it either way)

At the end of the day, I won't disagree with the idea that speakers and room acoustics are more important.  If there is a there here, it probably isn't that big.

But somehow I've fallen into the trap, and I don't necessarily feel good that it has distracted me from the more important speaker adjustment and EQ work.

But it has also been a learning, it's a little bug to make you keep listening, and I've had lots of fun exploring my edge of the world.

At the same time, there is nothing really Objectively wrong about the PCM 1704 chip, it has adequate measured performance, anything that seems to have Even better performance (like the latest ESS chips for example) isn't really necessary.

It seems to me that once you attain "adequate" objective performance, you ought to be free to explore your own angles without too much sneering.

Many people feel that they ought to be able to explore their own angles no matter what, and measurements don't matter, etc.  This is where I do not go myself, but if other people want to go that way, it's their hobby.



No comments:

Post a Comment