Sunday, November 29, 2015

NOSDAC: Cult or Cure ?

It was a very nice audio listening session.  The first ever where I'd invited audiophiles from the audio society to enter my main listening room and hear the magic.  All were favorably impressed, more or less, it seemed.  (You can never axiomatically interpret something like this, as if the system is Really Bad, you may not get any bad comments either.  It may only be where people actually feel threatened as you're nearing the pearly gates, but differently, and that can't be right.  So as they say among academics, the smaller the differences the greater the arguments.)

So from the most sage, the actual engineer with decades of professional experience actually making recordings, I got mostly (and actually somewhat surprising) approval.  What I had feared getting gored, the panel-to-subwoofer mating, as just got a not-so-quick touch-up days earlier, was pretty much accepted as successful.  But it was the panel sourced highs that got the barb.  "Harsh" or something like that was the word.  Despite all the praise, "I couldn't live with it" was the emphasis, which seemed out of place among all the other comments.

OK, I conceded.  I have indeed been aware of it and always struggle with it.  I deal with it mostly by moving my head into or out of the very central beam of the electrostatics.  You have to be sufficiently forward of the central beam of the electrostatics to get the smoothest highs, otherwise it sounds brittle.  Others later testified to what I described.  This near field speaker issue is real and needs further investigation.  I strongly believe this is the primary source of harshness in my system.  Later listeners confirmed the improvement from having head forward of the beam, and the criticality of this.

I didn't mention, but it may be clear to readers of this online diary, that much of my effort of the past 4 years has gone to addressing the harshness issue.  I recently bemoaned my trek in this direction as a loss to the then-alleged more important work of getting the bass decent.  But then I did some bass decency work just-in-time for the listening party.  But the kinds of things I've done were exactly replacing the Behringer DCX outputs, which aren't much praised by anyone (though objectophiles might say they were perfectly fine, having distortions well below audibility) with using a DEQ to drive an external DAC, currently the Audio GD DAC 19 which had inspired the engineer himself to obtain and now use the Audio GD Master 7.  Along with that change, the gain structure of my whole system changed, and now I run very close to 0dB with no attenuation, digital or otherwise, to compromise the resolution.  (I previously often ran with attenuation around 20dB, and despite my defending this with the claim that I have 24 bit resolution, I know it was suboptimal.)  So I've made major changes that I think can be defended on multiple objective grounds in attempts to eliminate the harshness, and they've been greatly beneficial.  Also, just using the Krell amplifier as compared to others I have, has a huge positive effect.

But my most sagacious guest had other ideas and didn't seem much interested in exploring the speaker/listener alignment possibilities during the listening session itself (though in much later comments, after the lamp had mysteriously become broken after someone tried to turn the lights off, he mentioned that speaker and listener orientation would be on the top of his list if I let him fine tune my system for a week).

Instead, he stepped into the measurements vs listening morass, and singled out the Krell amplifier as a potential part of the problem, effectively because of focus on achieving good measurements, and how many have moved to SET amplifiers, despite horrible measurements.

I look at the Krell FPB 300 as something very different from an amplifier designed to get "good measurements."  The very way it works is something very special in my view, with the output stage operating without loop feedback.  I believe this design is ideally suited for large electrostatics because it rejects electrical back EMF from the speaker, decoupling that from the actual amplification.

An amplifier designed simply to get "good measurements" could be made and used at far less cost than the Krell.

I have long experience with the Krell and a few other amplifiers, mostly the Aragon 8008 BB another very fine amplifier, I believe.  What amazes me is how much better the Krell sounds, and I believe this is precisely because of the special design of the Krell FPB, which isn't found on most other amplifiers, even Class A amplifiers, even most Krell amplifiers.  When I read about FPB amplifiers, the 300 and the 600, I just knew they were the ticket for electrostatic speakers which require vast VA swings, but accept vast EMF back from the speaker without compromising the signal integrity.  And everything in my experience has confirmed this.  They are the magic ticket for Acoustats anyway.  Now this is my experience, and my technical understanding, even if not acceptable to objectophiles.

Simply claiming that one might like the Krell because of good measurements demeaning.  That doesn't do justice to my sense about the Krell.  Actually, I suspect in a number of ways the Krell isn't the best measuring, and maybe not as good measuring as I would like.  If the amplifier has had to thermal limit down to the second stage of Class A operation, which is common, then peaks much above 100W are going to be increasingly distorted.  It is possible to get measurably significant distortion out of a Krell if one knows about how it works and can give it the right tests to show its weaknesses.

WRT SET's, I've never heard one that I liked, and have especially noted harsh sounding SETs at audio shows, sometimes besides very sweet sounding push pull tube amps.  There is every reason to believe that the high measured distortion is real and ultimately can do no good as it inter modulates with itself.  Many if not nearly all PhD audio scientists kind reject them (as do ALL contest for them...and amplifier with 10% distortion cannot possibly be virtuous, and there is certainly no measurable advantage in distortion at low levels, as might be claimed).  Most of my long time audiophile associates reject them too.

Would I call SET a cult?  Yes, although remarkably widespread.  The audio science behind SET is remarkably thin but the sound is preferred by many thousands of listeners.  I think in most cases the deficiencies only become overwhelming at loud levels because of the companion cult to super-high-efficiency speakers.  In this case we are not talking so much about claimed-to-be audible differences. SET's sound different, no question about that, can be easily distinguished from low distortion amplifiers once the SET's are operating with levels of distortion well above 1%.  You may well claim there is a difference, though, in SET Superiority.  SET cultists, of course, believe in the SET Superiority, which comes from the virtue of simplicity.  They can "hear" that among all the other qualities and limitations of real SET amplifiers, which they may admit to.  Yes it's harsh when cranked up, but played softly it has intimacy or whatever.

And along a similar vein, the sagacious advice was that I must try (and would likely love) the No Oversampling (NOS?  didn't that mean New Old Stock or do we just distinguish these things on context?) option of my Audio GD DAC.

My mind waffles about whether I should even try this.  If I call SET a Cult, then beyond any question NOSDAC is a Cult.  It's far smaller group of people who use this kind of DAC (though interestingly, a group that highly overlaps SET and especially openness-to-SET as could be assessed by including those who formerly-used-SET), the backing among objectophiles and PhD audio scientists equally or even more thin, etc.

Strangely, then, one advantage that NOSDAC has is a more technically defensible argument, one that can be shown With Measurements (ironically).  NOSDAC's preserve the timing information.   They are technically perfect with regards to timing.  You can just look at the oscilloscope photos.  All the pre-ringing, post-ringing, and similar effects go away.  Now, finally, what you see "looks" exactly like, or the closest visible approximation to, the signal going in.  (Apparently the timing aspect is a key part of the visible picture.)

The problem is that technically perfect timing is obscured by enormous mass of audible distortion.  Unlike SET's which may be sufficiently undistorted at low levels to actually suit many people by objectophile standards, NOSDAC's distort horribly at every amplitude level, in every way, etc.  It turns out those apparently small "visible" differences are actually huge, we don't judge amplitude distortion well on visible graphs of square waves.

And what do I mean horrible?  Like, say, 30% THD.

That number 30% I am taking from PeterSt, Netherlands based senior member of ComputerAudiophile, in this long threat started by NOSDAC'ers   He seems knowledgeable and respected.  He was once a NOSDAC owner, started to investigate how it worked, began to realize it was a horror, then devised his own minimum phase DAC based on 16x oversampling.

The ever reliable Archimago has an exploration of NOSDAC's with actual graphs from a forward thinking DAC by TEAC with multiple options, one of them being NOSDAC.

I might point out that many cult (and otherwise) respectable DAC designers, such as King Wa at Audio GD, prefer their own default oversampling settings.  That's where he put a lot of his intellectual energy, into getting right as-close-as-possible digital interpolation and filtering.  He also wants to show you what he has achieved by giving you the other options as well.  And it seems like many try the settings and end right back at King Wa's default.

Not to say that there aren't NOSDAC designers, in which category I would include Audio Note, who love their NOSDAC's.  This category has many cult famous companies, Metrum comes to mind, but by far Audio Note is the largest.  But of DAC designers overall, these are a tiny proportion.  In fact NOSDAC designers are what you might call anti-designers.  They "design" the oversampling by simply not including it.  And usually (following Audio Note) not including the filter either.  That's it, design over.  You don't have to be a very technical designer to be like that.  You don't have to have a PhD or even know math or audio science or anything at all.  You simply have to have a lot of Chutzpah, and if you know anything about the subject generally, combined with intense contrarianism.  But what if you had, one day, tried making a NOSDAC, and, voila, it sounded good by Listening?

So here we are back to the sagacious question: Listening vs Measurements.  Though perhaps my dislike for this drafting of the dichotomy comes from many particular issues with measurement per se.  By definition, a set of measurements cannot be complete, it's only a set.  There are infinite possible measurements, performed in infinitely many ways, as well as facing infinite inherent measuring limitations.  In audio, measurements have been fantastically exaggerated (such as Instantaneous Peak Power, occurring the moment the unit instantaneously self-destructs with man-made lightning), specified with incredible accuracy, or rounded up to nice numbers just to look like you don't care.  The best audio measurements usually come expressed by graphs rather than numbers, and the output of RAA is very nice, but still far from complete, and what manufacturer has given anything like the output of RAA?

You see the word Listening is being used here as if it The Ultimate Answer (because in a way, it must be), and Measurement as the thing we know which is often somewhat incomplete or misleading (and we have loads of other experiences with that in our expending enterprises), in a way to color this as if you must be a fool not to do what the questioner begs: "listen for yourself."

With a similar misdirection, I could redraft the question as Guessing vs Sound Engineering Principles.

While a seemingly simple argument can be made for the notion that you should just "listen for yourself" (and be fooled too, I would add) there are many fundamental problems with it as a tool of audio design and/or selection.  To wit:

1) Potentially biased (most likely, I would say, without thorough DBT)
2) Subject to future change (including the very next time)

Because of 1 and 2, the results of any one listening test have to be held in complete contempt, as not even having enough wrongness to be wrong, without either additional thorough testing (pref DBT), or reasonable objective story to back them up.

Further, the big problem with bias is that it feeds on itself.  Once you have developed the bias, it then becomes part of a very powerful bias to stick with it.

Now until researching this post I hadn't realized that NOSDAC's have a duller sound with proportionately less high frequencies.  I had known they produce more distortion, from aliasing, but had wrongly figured that distortion to add to HF making the sound brighter.  Instead what happens is the audible distortion is in the form of products which appear throughout the audio band giving it a kind of richness, not brightness per se.

So now that I know this, I could use the duller highs of a NOSDAC to balance out what may be the excessive brightness of the Acoustats.  I'm not sure if it would work like that, but it might seem to help in this way.  I'm tempted to try this simply for this reason.

But crucially any such "help" is truly a band-aid.  It wouldn't fix the underlying problem(s) whatever they are, just by adding a companion kind of distortion to cover up the original problem.

Certainly a better approach would be that rather than covering up problems, they should be sought out, and engineered out directly.

Of all the band-aid approaches, using something like NOSDAC to cancel out problems is the worst approach.  It's generally adding heaps of distortion, unrelated to the original signal altogether.  It's far worse than other band-aid approaches like using EQ to ameliorate room nodes.  Of course EQ isn't solving the actual problem (which is reflections in a bounded space), but it is using a linear means (which isn't itself non-linear distortion, nor produce more additional downstream non-linear products) to cancel out a major linear chunk of the original problem, and it is doing that directly by directly addressing the actual room mode frequencies.

Probably my NOSDAC loving friend would agree that NOSDAC's shouldn't be used as a palliative either, but rather that the problem is made to sound worse by being layered with the time distortions caused by oversampling.  If NOSDAC works as advocates claim, it should be used with the best systems as well as the worst, making the best systems even better.

And at this point, we can imagine a world of existing audio systems, many sounding fantastically good, and almost all of them, probably 99% or higher, NOT using NOSDAC's.  Ordinary OSDACs must be pretty good if they are used in most of the best audio systems.

Unless all the OS DAC designers, virtually all academic audio scientists, and most system designers are wrong, there is not really any need to even try NOSDACs.  For most people, this is simply a waste of time.  There isn't enough evidence to think this is a worthwhile change.  Everyone doesn't need to try everything for themselves, especially when the results of listening tests done improperly, as most are, can be confounding and create superstitions.

If we did want to verify NOSDAC claims of the importance of audible distortions, we could devise a number of possible blind tests.  One is a simple bypass test.

I haven't done this test, but I find it very hard to believe that the NOSDAC path would sound more like the original overall than the OSDAC.  In my imagination, the NOSDAC sounds somewhat different, whereas the OSDAC sounds about the same.  This would conform to the most important (if imperfect) set of tests done on digital audio by Meyer and Moran.  Listeners could not reliably hear the effect of digital encoding and decoding added to an audio path, even when done ten times.

The only way this imaginary bypass test could be claimed as a win by NOSDAC folks would be the claim that it maintains something special, amidst all the other clearly audible differences, a claim that would not be very credible.

While what one does with audio reproduction for personal enjoyment is a personal matter, I continue to believe that listeners are best served by the most accurate reproduction.  Harry Pearson called such a thing The Absolute Sound, and I believe in that, though not perhaps in his methods for gauging and approaching it.

Accurate reproduction is the goal of virtually all of the audio press, the audio engineering establishment, academic audio scientists, and so on.

There is a major dispute between subjectophiles and objectophiles, or black hats and white hats in Peter Aczel's terminology (himself being a self described white hat, John Atkinson being a self described grey hat) on the essential methods of achieving accurate reproduction.

Sighted listening tests are the standard of subjectophiles, double blind tests and in some cases just measurements are the methodologies of objectophiles, with subjectophiles often claiming measurements are unimportant.

My belief is not in measurements, which are incomplete and tentative, but in principles, when you can find them.  And the question doesn't come down to listening or not but to something more fundamental: how do you assess truth?

Is it really necessary to tour Auschwitz to know that the holocaust happened?  I think not, and in fact some denialists have toured such places only to bolster their opinions.

Truth in our experience comes not so much from direct observation as received wisdom.  In most cases, it isn't necessary to go beyond that.

Exploring the edges, where contrarians lurk, can be an honorable activity, but it is not something everyone need do.

Now this is not to say that in all our interactions with people and objects we are constantly assessing truth in various ways.  But most of these assessments are based on thinking and reasoning about known (or presumed) facts.  Not on direct observations.

Even our direct observations aren't really observations.  As much as science knows about perception, it knows it to be a largely constructive process, that builds models, often first, then tests them against the experience, or constructs them based on a perceived reaction to experience.

So there is actually no "need" to do listening tests at all.  It is generally a good idea to do listening tests on speakers, but largely I have purchased most of the speakers I own on pure speculation, without an opportunity to pre-listen (and such pre-listening tests, as at dealers, I'm not sure how much validity I would attach to, the best testing is always in-home testing, which is also the most difficult and most rare).  And I think I've done pretty well, with the Revel M20's, Acoustat 1+1's, and Gallo A'Diva's I own.  The speakers I have in my now unused pile wasn't all that bad either.  Given that I've been in this game for 47 years, it's not surprising I have a large collection of what I might now consider mistakes, though I'd try to rationalize them as stepping stones.

One merely needs to find the experts, or the ideas, one actually believes in.  Most of this is not a process of direct observation, it is a process of reasoning.  At best it is a measuring of other's consistencies.  At worst it is an assessment of other's tribe.  I fear I may not be doing the best always.

But that's ultimately what it comes down to.  I believe generally in what the majority of academic audio scientists believe, roughly the mainstream of the editorial staff of the Journal of the Audio Engineering Society, who won't accept subjective testing only double blind in most cases.

That means, I'm likely to find little or no merit in NOSDAC's.  Even if they initially sound pleasant, or uniquely satisfied, I'd still rationalize that as a euphonic quality I'd ultimately get tired of.

I believe if ultimately NOSDAC's actually are the way to go (which is entirely contrary to the way things look now), eventually this will be discovered, and pretty much everyone will switch to them.  There is no essential need for me to be on the cutting edge of such new things.  Not if I don't want to. Not if I think it's going to be useless.

Ultimately I don't think it's worth doing most audio listening tests.  There are infinitely many subjectophile "tweaks" one could try to make better sound.  There are endless cables one could try, and cable stands, cable separators, and on and on.  There's no respectable evidence (JAES is the standard, or the objectophiles at Hydrogen Audio) that such things make any difference.  As such, the best that could be reasonably believed about them is that they make small improvements.  How many small improvements does one need to bother with?

Exactly none by some accounts.  All by others.  I apply a little attention to everything, with more hoping to focus on the most important than doing so.  I excuse myself: this is my playtime, I'm just trying to have fun, and the important work has more difficult and boring parts.  And realistically there are indeed problems with exploring the most important kind of changes (or perfections!): loudspeakers.

One should always go after the biggest possible improvements, or at least the low hanging fruit.  In almost all audio systems, that means the loudspeakers (and the room, EQ, absorbers, etc).  Very little attention need be paid to the electronics and uncompressed digital sources because they are already very highly perfected.  And I do hew to that at least some of the time.  I pretty much ignore audiophile tweaks except as fashion accessories--I like them for that.

If loudspeakers are the biggest factor, as objectophiles and I believe, loudspeakers should be what almost all of our tests are about.

But what about the recording engineer?  It is commonplace that even experts aren't experts about anything in their own precise field, let alone others.  It seems to me he has keener observation about many things than most, but that does not prove his all his observations, ideas, or beliefs are correct when I find others with ideas I find more believable.

Given what I am, I can't simply accept any expert's opinion exactly.  And I believe that even many people with good powers of observation and refined expertise can still be profoundly wrong about some if not many things.

So I decide which ideas of which experts to accept, and which to reject.

How do I do this?  Mostly by my own reason.  Very little based on my own observations, but I have performed DBT's on myself and others, and that experience continuously shapes my ideas.  Most subjectivists haven't done such things, no matter how many sighted tests they have performed.  Basically what even the most minimal participation in DBT shows is that hearing is unbelievably biased and unreliable at qualitative testing, and audiophiles can't prove they hear the things they claim to. This is because we've evolved to correlate, to ferret out connections even if we get a lot of false ones, not comparators that can reliably compare things over time, and time is the ether of sound.

Even for audiophiles the ultimate sense isn't hearing, it's thinking, and especially determining what to believe and what is important.  And examining the thinking of others is not a distraction, it may be the best place to start.

Monday, November 23, 2015

System tuning requires iterative process

For several years now, I had not used my GenRad Oscillator because I thought it was broken.  I finally got around to removing the MDA 0.25A fuses, and found they measured about 5 ohms.  Then I looked up the cold resistance for such fuses, and it's rated as 9 ohms.  So by the book these fuses could still be good.  I put them back into the oscillator, and, It Worked!  Sometimes things are just like that.

This is the kind of oscillator best used for speaker system tuning because it has a big full turning logarithmic frequency control which sweeps the decade 20Hz-200Hz, and going a bit lower than 20Hz actually.  So it's easy to do useful frequency sweeps at any speed, and the resolution is immense.  I have other oscillators, such as a B&K with a linear control, and it's nearly useless for speaker tuning.

Now I had really intended to do measurements, and possibly parametric EQ calculations, by Room Eq Wizard.  I did my first measurements with REW in January this year.  And I haven't done any since, because there is some complication in setting up the computer, interface, microphone, and stand.  And it could be argued that is far better than my very old fashioned approach of sweeping an oscillator by hand and adjusting EQ's to fit by ear.

But just because I got the oscillator working, which was a little miracle, I went ahead and checked out the oscillator.  I wasn't planning to spend a whole day, just a few minutes.  But I ended up only finishing the tunings (they're only finished when you can write them down and get them recorded and measured without being forced, simply forced by one's own ego, to change them again) about 20 hours later, with 8 hours sleep and breakfast in the middle.  Now in the process, I did listen to a lot of music also, which then sometimes motivated me to make more changes.

So like I said, many times, including after the first hour or so, I tried to write down the adjustments, and move on.  But just going over them, I couldn't leave them alone.  I couldn't believe they were done correctly, I must have been mistaking the Q as I often do, not fully remembering the correspondences to bandwidths.

So that's how it went, just changing them, sometimes just because I thought they must be wrong.  Often making the changes, at first, or in the middle, I didn't bother to verify.  But then towards the end it became clear that verifying is the key step.  And not just at any one frequency, but sweeping up and down at various speeds.  And that's because the correction process is far more complex than you might think.  It's far more complex than most automatic systems can actually handle.

That's because the speaker and room system has a lot of possible states.  And those states, the position of compressions and rarefactions in the air at any one time, and the positions of everything in the room, and so on, are very very complex, even if ultimately "linear" it is far more complex than any single set of measurements can reveal, and more complex than a single deconvolution can address.  (And the simplification in most FFT's of what is going on in the nearly absurd.  You absolutely cannot do a tuning based on 1/6 octave FFT, though that can be an additional useful test and verification.  But imagine a problem with 10,000 variables, and you're modeling it with 30.)

It's very much like this, you have a particular peak, so you make an exactly equivalent notch to notch it out.  But it doesn't notch it out.  You're left with at least half as much as you started with, most likely, and possibly slightly shifted in frequency and q, or split into two separate frequencies and Q's.

So the only way a correction can be done is iteratively, not just in the mathematics but in the measurements.  You make one measurement, do one set of fixes, make another measurement, make another set of fixes to the fixes, and so on.

Since automatic systems generally just do one measurement and pray for the best, they can't possibly be as good as obsessive kid with his oscillator, sweeping up and down until all the bumps are gone, at least if he had infinite time.

Unfortunately, I was so unsystematic in everything, I can't assuredly say that my work here is truly great, though it's a vast improvement over the way things were before.  I really should do R.E.W measurements along with the oscillator sweeps, and so on.  But for now I am going to let it go, because I've already invested too much time in this for now, and it's actually pretty good too.  FWIW, here's the smoothed 1/6 octave measurements from Analyzer app on my iPhone:

That might look rough, for one you can only blame me for about 120 Hz and below, and it was far rougher as well as more complex at first.  The flat shelf from about 40Hz to 100Hz would be several huge peaks, falling off at the bottom for no good reason, and hugely bulging at 100Hz to sound like mid fi.

In the process of doing this, I discovered that my disc players, not often used, had channels reversed.  I think you can't blame the audiophile with so many different things that can go wrong, and constantly experimenting, for getting lots of things wrong.

Oscillator system tunings, November 2015

LR24 at 80Hz using two 2nd order Q.707 filters in combination
Right Only, 120 Hz, 1/2 octave, -4dB

32Hz +1.0dB Q2.8
40Hz -4dB Q6.3
45Hz -6.4dB Q4.5
57Hz -3dB Q4.5
69Hz -3.5dB Q7.9
96Hz -8.4dB Q6.3

32Hz +1.5dB Q3.2
40Hz -7dB Q7.9
46Hz -9dB Q4.0
51Hz -4dB Q7.0
71Hz -8dB Q7.1
LR48 (oops!?  I thought I changed that to LR24 to be symmetrical with panels)
-10.6dB (both)

Super Tweeters
Set down 6dB after monumental bulge on iPhone analyzer, still about 4dB bulge, believed correct for wider dispersion pattern in random noise.

Saturday, November 21, 2015

I love digital volume controls

Digital volume controls are one of the things Most Misunderstood by subjectivist audiophiles (and I'm one of those partly--but one of the rare ones who have come to embrace digital volume controls).

I was where many audiophiles are now in 1979.  That was when I began surgery on my Marantz 2270 trying different opamps to replace the line amp board in that unit, and finally going to a "passive preamp" consisting of a 32 position digital switched attenuator (same as used in G.A.S. Thedra...I had friends at G.A.S, and I installed that in a ModuLine box that had been used as the original prototype of the design which was ultimately sold to Dennessen as the Dennessen Preamp...designed by other friends of mine with my help...but I fried the prototype Dennessen board I inherited by connecting the unterminated 9V battery wires I ended up with nothing more than a passive attenuator box that I upgraded to the G.A.S. stepped pot, which IIRC was 25K).

I could see no reason to do anything other than a passive attenuator for two decades.  (I built a tube-based phone stage for turntables.)  I certainly believed it was the cleanest approach.  But with my passive attenuator box, I couldn't easily adjust balance, and I never could get the imaging right on my modified speakers.  And I didn't think I could set balance any other way than by remote control.  And that locked up my imagination for about two decades.  I cursed my lack of good center image, but I never did anything about it.

I could have added balance control in various ways, it's not wonderful to do this without a line amplifier which provides buffering, but it can be done.  But balance really has to be adjusted from the perspective of the listening position.  Any kind of balance control that you have to adjust while standing up and away from the listening position won't work.  You have to pop up and down making little balance adjustments and then checking out how they sound.  This turns out to be impossible because each time you sit down the position may be slightly different making the required balance control adjustment different.  And remember you have to do this every time you change the volume level with most stereo pots.  The worst of all are preamps with separate analog level controls for each channel.  By the time you get the balance and level right for listening to the recording it will be sometime the next day.

Finally around 1998 I bought my dream, a preamp with remote volume and balance controls.  I also at the time thought it to be about the cleanest possible preamp, an Aragon 28k.  So while it wasn't passive, I figured the added distortion and noise were sufficiently low.  It has a low distortion discrete transistor amplifier (essentially a discrete differential op amp with feedback) powered by a discrete transistor regulator.   (Actually, it's very hard to beat the best chip op amps--if you really have the best.  It's not hard to beat the early 741 op amp.  It is extremely difficult to beat the best chip op amp made, an OPA 211.)

Anyway, other than the all discrete part, which is debatable, it is straight forward no-nonsense, just a good circuit with good parts (except perhaps that motorized pot...).

It has very low distortion, always at the residual of my measurements (the last one was 0.004% or better).  And it has remote control balance, and quality jacks as I had been using for quite awhile on my prototype unit.

Now I tried adjusting the balance on every recording.  Still easier said than done with my very wide angled speakers.  And I noticed right away that it varied a lot.  Finally I figured out the #1 variable.  The position of the volume control.

Something else I became aware of at that time.  I was usually only turning up the volume about 9 to 10 o'clock.  Barely cracked.  It occurred to me (1) that it had poorly tracking channel balance in that part of the control, and (2) what a waste!  internally the signal goes through a pot reducing the level 100 times, then gets re-amplified by the line amplifier back up 10 times.  Would have been much simpler just to have a passive attenuator...well, as I had been doing.

So now that I had balanced control (instead of the fixed attenuators) I really did have to use it as a correction all the time.

I became aware of the Classe CP-35 (with digital volume and balance) sometime around 2000.  I should have bought it at that lovely store.  But I didn't actually buy one until around 2005 and on eBay.  Not only did I love the perfect volume tracking (now discovering I didn't need to use it anyway, now having perfectly matched high end Revel M20 loudspeaker) I just alway thought it sounded good.

Quite contrary to my long standing technical belief in passive preamps, the CP-35 wasn't just liberation from having to reset balance with each volume adjustment, or the limited range or size of adjustments with stepped attenuators which I had used before.  It Sounded Better To Me.  This rarely happens with so much clarity.   All the passive attenuators, and even the re-amplified potentiometer of the Aragon 28K, sounded dark.  The CP-35 seemed to lighten the music in the same way the volume display lightened the front panel.  It brought the music more to life, and I decided I liked this.

And then in occurred to me I should not have been ignoring impedances, gain reductions, re-amplifications and so on.  Every time you have a big R followed by some C you have a RC network which is rolling off the highs.  All the rolloffs are additive, and we sense not so much the higher frequencies as the slower rise times when nearly any kind of rolloff is added to the rolloff there already is.  So figuring bandwidth of 20k being "good enough" is wrong.  A reasonable analysis suggests that every electronic component should have bandwidth to 200kHz or better, to achieve sufficiently flat time delay.

Anyway, never even having a big RC in the middle is good.  There is nothing magic about a potentiometer, pots are horrible, though some less horrible.  That's just for starters.  People somehow believe that pure digital gain reduction (which, btw, is not what most preamps simply with digital volume do, and even fairly ordinary home theater receivers have "direct" modes that bypass conversion of analog inputs to digital) looses resolution.  That may not be true, if reducing no more than 48dB when attenuating a 16 bit digital source in a 24 bit digital pathway, as SPDIF automatically is, particularly with sources that do attenuation, and most things with SPDIF input these days for sure.  At least not in digital domain and output.

But meanwhile, even a simple pair of resistors to attenuate introduces lots of losses, among them automatically being resolution.  No matter how you figure resolution, it is lost in the process of resistive attenuation.  Relative to noise, absolute volts, absolute quanta, every way you look at it resolution is lost when you reduce level with a passive pair of resistors, even a fixed pair (though other kinds, especially pots, add other problems).  Plus whatever distortion from the resistors themselves, which is always plenty btw.  But even if the resistors were perfect, information is inherently lost (though you could say, not enough lost to make any difference, but there we are back in compromise land again, not the world of perfection that no-information-loss implies).  And this is especially true if you take the most hard headed Shannon's Laws approach which considers noise a limiting factor.  In addition to adding noise, the resistor bridge reduces signal.  It's inherently a signal to noise ratio loser.  Plus it looses bandwidth because of the R being added to the signal path with must always be followed by some capacitance, the worst case being a long hi capacitance cable to the power amplifier.

(Some would argue the pot or resistors also reduce noise, and that is true.  They reduce the incoming noise by exactly the same proportion as the signal is reduced.  But there no increase in S/N from that. To that, the resistors will always add a little noise.  So overall, S/N is decreased, though possibly only by a small amount, depending on the series resistance in the pot being used.  It's worst, of course, with a large value pot.  Old timey tube equipment often used 250K and 500K pots.  Even if followed by gain circuitry, those have to be understood as information loss devices.  With something like 10K of overall resistance achieved with high quality resistors, as in a device like the Placette passive preamp, there is far less information loss.  But still there is loss.)

It is better, at least in principle, best to avoid gain reduction.  If gain is being increased overall, it's pure stupidity to reduce the amplitude level first, then apply a fixed amount of gain.  And yet that's the standard analog preamplifier design, all the classics to now pretty much I can think of.  Some do switch in varying amounts of gain during the attenuation, such as reducing gain by 6 and 12 dB when possible, thus, the worse case attenuation down to 18dB is only 6dB, and the average less.  But that's usually only done at the very high end, like the top Mark Levinson.

Another approach is to go variable gain all the way, sometimes varying it far negative.  That may be what some or all chips do.  Only practical limitations apply to this approach, in principle it could be perfect, though of course the gain reduction loses information as all analog gain reductions do, it could theoretically be the most perfect and therefore lose the least information.  I think this is what many gain reduction chips do, though some may combine resistive switching with that for performance or hype.

Anyway, I'm not quite sure what the Classe did, I think basically it used a audio attenuation chip, which were widely available in 2000 or even 1995.  Some of the better ones are quite good now (though becoming harder to find).  Mark Levinson in their first preamp with digital gain controls, the 38, from the mid 1990's, said they varied the reference level to a DAC, in effect making an analog amplifier with digital volume control, from some of industries most refined parts.  But are DAC's really the way to do this, and not custom engineered parts, engineered as volume control, but with very high quality?  I believe the top parts now are quite good, at least since the top now discontinued Burr Brown unit, which I think was available in 2006, then discontinued and reappeared in a Finnish part.

I concluded in previous article that any weakness in the Emotiva XSP-1 (Gen 2) in unbalanced outputs is probably related to my guess that it makes everything balanced then unbalanced.  Nothing to do with the digital volume per se, which is clearly balanced and shows no weakness--no distortion down to invisible above a SOTA noise floor--in full balanced mode.  Then a SOTA preamp has to have both balanced and unbalanced preamps in parallel, depending on nature of inputs and outputs, so as to go through as few conversions as necessary for a given kind of inputs and outputs.  I don't think emotive did that, not even sure about Levinson et al.

 Anyway, with all that irrelevant personal anecdote covered, I have to think deeply about digital volume controls themselves, and not just my background story...

First, there are two kinds of digital volume controls.  One type operates in the digital domain.  Almost always when this is done, the signal is already in digital form for some other reason, such as it was recorded and distributed that way.

When starting with 16 bit, and reducing less than 48dB (the equivalent of 16 bit), I have heard it said and I believe that no resolution is lost (with proper dithering, which turns out to be necessary in maintaining resolution) when outputting and inputting 24 bit, and spdif is automatically 24 bit, it maintains that many bits whether they are used or not.

So when my Tact Audio RCS 2.0 reduces level in 0.1dB steps, it computes the needed reduction and applies dither with 48bit resolution (or something like that, it's done in a DSP) then outputs 24 bit SPDIF, there is no loss to deep reduction, way below anything I do, especially now in the Living Room where I can use as little as no reduction at all thanks to low gain.  So I can sometimes run right at 0dB and even wish for a bit more (which Tact delivers, up to 6dB more, though I don't like the idea of using digital multiplication, but technically it should be fine as long as you don't get to digital clipping, which can be seen and heard).

So, repeat after me: analog volume inherently loses information, can't be avoided, digital volume loses no information down to volume reduction you never seriously use.  Plus digital volume is accurate, has no channel imbalance so no endless worry with each volume adjustment, resolution to imperceptible-difference 0.1dB available in some equipment I have, and above average resolution in others.  Plus it does not essentially induce bandwidth loss, another kind of information loss.  Distortion is possible, but can be far below that of most other active circuits.

Ahah, you could say there's one catch.  Your fair DAC's aren't truly perfect to 24 bit.  Ahah!  But the output of the best DAC's, in my opinion, are far superior to most if not all preamps.  DACs are the most perfected thing.  That's why I power my power amplifiers now that I can direct from DAC's (though it limits my level capabilities a tad, the 2.5V output of the Audio GD DAC doesn't come close to driving the Krell either to rated power or actual clipping (though, reasonably speaking, one would think it to be enough).

It turns out, in electronic circuits, the hardest thing to do extremely well is drive a load.  So it is that analog to digital converters are the most perfect of all things, per dollar spent.  Digital to analog converters are next, up to generally limited outputs of 2 volts into high impedance load.  Preamps nowadays put out 8V or more, and can drive 100 ohm loads if not 10 ohm loads.  That is great, but makes it a difficult thing to accomplish with low distortion.

So in my view, it makes sense, if you are starting with digital sources anyway, of doing the attenuation if any in digital also, and then only driving power amplifiers with the outputs of DAC's.  That's sort of what I've been doing since 2006 or so, but the "DAC" has really been the output of a Behringer 2496 DCX, which isn't terrible at 2V output but isn't great above 3.5V.  I ran the Tact through a DEQ then a DCX all in digital mode.  So digital never left 24 bit digital domain until the DACs.

But anyway, as I've sad before, most audiophile stuff that takes analog inputs and outputs (with the exception of a fully loaded Tact or something like it) and reduces volume does so never "sampling" the input into the digital domain.  There are fancy parts that do that with the equivalent of varying the gain of an amplifier, or precisely switching in resistive elements when needed too,  And that's probably the way to go if you do actually have analog inputs and want analog outputs.  I now eschew devices that don't give you at least 0.5dB resolution down to -30dB, and I think 0.1dB is best.  Most audiophile solutions based on using relays to switch fixed resistors, such as the early Krell preamps or the Placette passive or active preamps, don't have enough resolution in gain adjustment.  And it may well be, the chips actually outperform them, except in the audiophile imagination.  (Mind you, I've lusted over these relay and fixed resistor solutions for a long long time.  They've either been unaffordable, or I've decided they weren't the best way to go, which is where I am now.)

As for me, I started using digital crossovers and room correction in 2006 and believing in their most important advantages (such as time alignment, and room mode attenuation) I haven't looked back.  Digital gain is the way to go when you do that, when you're already in the digital domain, level adjustment can be done effectively without loss of information in most cases.

But I still need analog domain digital volume controls in limited applications, such as the preamp setting the gain for my Masterlink digital recorder...that's the #1 job my Emotiva XSP-1 does now and might do for a long time, because it's hard to beat it's balanced outputs, though it would be nicer to have the 0.1dB resolution, balance, and polarity of a Mark Levinson 38/38S/380/380S/32/320/etc, so if I ever get one of those it could replace the Emotiva.  In the end (until it died) that is what I had used the Classe CP 35 for, except I also ran the main outputs in parallel to the Masterlink and Sonos, I figured it to be important to send the same 2V output to both.  (I've subsequently found that generally to be unnecessary, particularly with the high gain output (!) of the Emotive phono stage (at Processor Output).

The #2 job of the Emotiva, though actually the #1 in regularity of use, is as a selector switch for which analog devices feeds Sonos, as well as the into the aforementioned preamp-to-Masterlink.  This is fine and perhaps good enough.  Though I haven't seen measurements of the Emotiva Processor Output, I'd guess they are significantly better looking then the Unbalanced Main Outputs, which has a tiny tiny bit of tiny distortion spikes poking up above the very low noise floor.  If it did have those same distortions, or worse, I wouldn't want to keep the Emotiva in this selector-switch role.  And just because of the uncertainty I imagine myself ultimately getting (if not the Levinson) simply an unbuffered high end selector switch.  Which I could make out of my old passive pre (but it's too ugly by my modern standards).  So anyway I imagine making or buying a nice unbalanced selector...I use the upgraded (Teflon Jack) version from dB Systems to select the inputs to the Lavry AD 10 which allows me to run high quality analog into my digital system.  And that wasn't expensive at all.  I did chose to re-work it with some dampening material inside the chassis.  But that would be one acceptable solution, and there are no doubt others (and why wouldn't I use a higher end one than the dB systems in the living room, and move the dB systems to the bedroom).

But as far as the phono stage itself, if I continue using the Emotiva in that role, there's no much advantage in bypassing the Emotive-as-selector because, essentially, I can't bypass it.  I can't take anything other than the Processor output as the input to that hypothetical switch box, and then run it to the box with more cables.  Right now, the output of the Processor Output, which I'd have to use anyway, goes through a short low cap cable to the Sonos Line Input.  It Could Not Be Better, assuming I were using the Emotiva Phono Stage.  The only improvement that an external selector could provide would be for the other inputs, and then at the slight expense of the Phono Stage.

And as far as eliminating the Emotiva Phono Stage, that's easier said than done.  I could easily modify the dB systems to have the optimal loading for my cartridge.  That's the easy part.  The hard part is that it doesn't have quite as much gain.  And then it would be uncomfortable to use at the same Sonos Line Input level setting.  The phono would always be too soft, and I'd have to boost the gain...with a line amplifier of some kind.  So then, is my conglomeration of phono stage and line amplifier still better than the Emotiva?  My initial impression is that the Emotiva may be overall better sounding than the dB, and significantly quieter as well as having more gain.  So really to replace the Emotiva I might be talking about something much higher end.  Probably the only way to get what I'd want here would be to make it myself, perhaps with OPA 211's.  Short of that, and until I make that, the Emotiva may be good enough.  High End phonos rapidly become very very expensive these days. and mostly not worth it in my opinion.  The Emotiva is probably as low noise and distortion as most of them, only lacking in qualities perhaps now not measured, if they even exist.  Though mind you, I'd love to have such if it wasn't too bad technically and sounded good, and there are a lot of lust worthy ones in the $20k range.  But most audiophile preamps likely have far more distortion to the outlets having sufficient gain to make LP and CD levels comparable, 2V peak.


A discussion of the legendary Burr Brown audio volume control and its successor.

A discussion of audiophile superstitions against digital level reduction (digital domain digital control is the only kind mentioned, somehow even these geeky people don't seem aware of digitally controlled analog devices--which is the vast majority of digitally controlled preamps--but otherwise their analysis is devastating to audiophile superstitions, basically everything lost by digital reduction is also lost by analog reduction--which then loses more, but analog-o-philes can't see that, but of course an analysis that entirely presumes Shannon's Laws, as I don't, but similar principles still apply indifferent to the belief or disbelief in unmeasurable resolution).

The successor to the Burr Brown audio control made by Finnish company.

Clueless megabuck high enders rant and rave.

Glossy gloss.  Still don't get that there are digitally controlled analog devices.

Sunday, November 15, 2015

Krell FPB 300 back, now with Emotiva phono stage

It was cool all day Sunday, and by late afternoon I started playing the Krell FPB 300.  All the harshess and glare disappeared, leaving beautiful music in 3D.  I should listen to the Krell always, but have problems with my air conditioning when I do.  I need to get remote thermostat in different room than living room, but contractor has been reluctant to do that with my 12 year old system.

The Krell is 1.14 dB louder than the Aragon 8008BB I use during hotter months, so I increased the bass level from -11.7dB to -10.7dB and the  super tweeters from -1 to 0dB.  Using new Keithley 2015 meter the first time, reading seemed more stable than with Fluke 8060.

Somehow the bass is more authoritative too.  But before I had switched to the Krell I had increased the left and right notches at 45Hz and 66Hz by another dB.  At 45Hz they're now -6 and -7.  Going back years, I had notched to -9 or even -11 at 45 Hz, but while it makes the room less boomy, the sweet spot begins to sound thin.  In most recent adjustments I reduced the notches to -5 and -6, but that now seems inadequate.  Playing Jessy J it needed this adjustment.  Funny as I recall I had decreased the notches while listening to Jessy J also.

All Sunday I was playing the turntable with the Emotiva XSP-1 preamp I set up late Saturday night after going to the Symphony.  I decided it sounded best from the Processor Output into my Sonos uncompressed Line In, and thence to the living room or wherever.  From the unbalanced main outputs into Sonos Line Input, it was very very transparent but also, I feared, slightly harsh.  Just one inconclusive A/B test was performed after I first thought I hear the harshness, but then I decided I didn't need the volume control for just feeding the Sonos Line In, so I went with the Processor Output, which is a more direct path anyway.  (The "slightly harsh" finding was at best ambiguous if not totally dubious, and most noticeable actually when taking signal from the Alesis Masterlink playing a hybrid RCA Living Presence CD, which was harsh sounding either way, just somewhat harsher when being played through the unbalanced Main Outputs than through the Processor Output, I thought.)

Meanwhile I felt the main balanced outputs were just fine, wonderful actually, and I continue to use those for driving the balanced inputs to the Masterlink digital recorder, which was really the only reason I needed a "preamp" (instead of mere source switcher).  I made a quick test recording and it was perfectly transparent.

This must be my imagination, because published measurements show the XSP-1 to have remarkably low distortion through both outputs, including the unbalanced outputs.  But I just had this feeling, and anyway it can hardly be argued that the processor outputs would be worse--they do have less processing after all.  While using the Processor Outputs I also run Direct (no Tone Amplifier) and sometimes also Proc On (no signal running in output section) to get the best sound of all and the Processor Outputs.  But I need to do Proc Off to get output to the balanced main outputs, when doing a Masterlink recording.)  So it meets all my requirements if not as I had intended, by using processor output to drive Sonos and balanced outputs to drive Masterlink.  I found Sonos works OK with Nakamichi and XSP-1 phono straight into Sonos line input at max level, sometimes boosted one for low level recordings when max straight-thru gain isn't loud enough.  The XSP-1 MC phono, loading my Dynavector 17D3 to 100 ohms, is giving me the best sound I've ever had with the Mitsubishi LT-30 table, and perhaps the best sound ever.  The 100 ohm loading soaks up all the remaining resonances, some of which still had a speed instability sound.  Now the sound is rock solid, perfect pace, and sounds clear and less sibilant, more fully balanced.  Under optimal circumstances, such as a heavier arm with electromagnetic damping, the Dynavector might shine better just above 100 ohms, say 150 or 200.  Unfortunately the Emotiva gives me only 3 choices, 100, 500, and 1k, but 100 works fine under present circumstances anyway, possibly even better than the oft recommended 200 ohms because of too light arm and too little arm damping.  I'm loving the phono preamp, possibly the best part of the whole deal.  Previously my dB systems preamp had a fixed 1k ohm load, I think.  Possibly that preamp could be improved with loading and better caps, and the Classe repaired and juiced up (how about relay inputs straight to tape output, no buffering, pure).  Then I might have a pairing that would beat the XSP-1.  But right now the XSP-1 works and is working for me.  Even the processor outputs are at least muted if not buffered, and the muting works without relays, as does the volume control.  I wonder if all relays used for these tasks wouldn't be better.  That's my audiophilia nervosa.

Archimago did detailed measurements of the Emotiva through balanced and unbalanced outputs.  The technical quality of the balanced outputs is stunningly superb.  The unbalanced outputs are just a tad worse, definitely not as good, there are lots of tiny spikes rising from the noise floor, but still should be good enough for audio purposes.  Too bad he didn't also check out the processor output I am using, which I believe is better.

But those little spikes in the Unbalanced output are a bit worrisome, and especially if you think they might be induced by chassis EMF and RFI.  Another little tweak I did was to pull up on the shielded portion of the Ethernet cable so that the unshielded last-foot jumper I use does not come close to touching the Emotiva chassis.  I used a velcro tie to hold the unshielded jumper cable several inches above the Emotiva.

John Johnson of Secrets of Home Theater and High Fidelity tested the XSP-1, and the measurements are stunning, but note that he tested ONLY balanced input to balanced output.

I chose the XSP-1 precisely because of the precise digital volume control, which because of typically audiophile prejudices tend NOT to be found on high end preamps.  Other than the Emotiva, the only others I can think of are Classe and Mark Levinson at far higher prices.  And though I often scan ebay and Audiogon for Levinson preamps in the 38/38S/380/380S series, those are quite old products now and even when they were new many audiophiles didn't like them.

While I was at first inclined to be suspicious of the volume control on the Emotiva, I've moved on to another fear.  I think it may be asking too much of such a preamp to have both balanced and unbalanced outputs.  Likely the unbalanced output is taken from the balanced path in the end.  So if you are running an unbalanced input to unbalanced outputs, you are going through these conversions:

unbalanced to balanced
balanced volume control
balanced line amplifier
balanced to unbalanced

So if you are really intending this for purely unbalanced use, you are getting two extra conversions which likely do no good.

A cost no-object way to do this correctly would be essentially to have both balanced and unbalanced preamplifiers working in the same box.  Then unbalanced would stay unbalanced all the way through, and likewise for the balanced.

Now I loved the sound of my Classe CP-35 which was in some ways similar to the Emotiva.  The CP-35 had digital volume with balanced and unbalanced inputs and outputs.  But here's the difference.  I measured approximately 0.05% distortion with the CP-35, which is much higher than the Emotiva, but when looking at the spectrum it was all noise, not actually distortion.  There were no little spikes poking up through the noise floor, just a relatively high noise floor for a line amplifier.  As long as you don't actually hear the noise, noise is much more pleasant than distortion because it seems to be uncorrelated to the original signal.  We are correlation detectors par excellence (or run amok).

My current plan is to do my line source switching with an external switch box which will feed both Sonos Line Input and Emotiva.  Then just use the Emotiva to drive the balanced inputs of my Masterlink.  Balanced output is what Emotiva does best (and pretty much state of the art).  The unbalanced output should be good enough, but if you are perfectionistic or obsessive, "should be good enough" never is.

Sunday, November 8, 2015

iPhone audio output quality

It looks quite good, so long as you avoid 15 ohm loading.  At higher impedances, it looks to be lower distortion that the audio generator built in to my Keithley 2015 Generator/Analyzer.  iPhone is reported to have (iPhone 6) THD of 0.0012.  The Keithley generator has THD of 0.02%, while the Keithley Analyzer has THD of 0.004%.  I could measure the superiority of the iPhone to the Keithley Generator on the Keithley Analyzer.

Other performance is excellent too, except crosstalk is a bit high (still better than any FM tuner).