DIYHiFi.org

Hi Markus, nice weather?

There's no battle, only healthy discussion.

I agree with you on what you said about the treble, it, is easily the biggest offender in a stock module. Coherency of that is easily improved, and it really needs it, but doens't take a new means of switching to do it either.

From what you're saying, your comparison only measured one class d, and of the supposed offending method of oscillation, while using other classes of amp as the standard. Obviously that does absolutely nothing to show if it is in fact the method of oscillation that's the problem, or something else, like a stock amp that's not up to a certain caliber that all the other components it was compared with were definately built to.

Yet you seem to confirm it wasn't the worst offender in all cases. Imagine how they'll compare once those minor limitations of the stock module are corrected for, hopefully you'll get the chance to listen then as well.

I'm not talking about a redesign here, I'm talking about the same amp, same system, with minor component variations to bring it up to that very same standard in quality all those other amps already enjoy, and then compare how the final degree in sound quality between them strikes your fancy.

Cheers

To conclude from such findings on the reason in topology is always alchemy.
Even if we do all the measurements we do not get the information about the reason in topology from this measurements.
If we want to learn to conlclude from differences in sound (or measurements) to a reason in topology, then we must act slightly more systematic. It is simple to define a reasonable procedure, but time consuming to follow it.
Means:
Step 1: Make two identical amps.
Step 2: Measure/listen and confirm that they both perform the same.
Step 3: Modify one amp.
Step 4: Measure/listen/compare.
Step 5: Conclude.

If we look to class D and find out if the self resonant modulation acts different from other modulation method, then we would need to build two amps with identical PSU, identical power stage and drivers, identical output filter.... and only vary the modulation method.

I think there was a guy around who did this in order to compare UCD vs. hysteresis modulation... can't remember his name...

_________________
Best overall performance is derived by the right trade off.

Hi AnalogSpiceman,

"The ideal UcD style amplifier only begins to do this as its output nears the rails. As to how noticeable or how much different or worse than normal clipping this may sound, I have no idea."

In a stock module, operating near max recommended voltage, I had noticed a very bad thinning in sound, increase in brightness, and overall compression, even a hefty change in pitch, started sounding like a chipmonks or minipops recording. It only occured during that last 20% before clipping, on any recording, and got much more prominent in conjuction with the volume being turned slowly up the rest of the way. So for that last 20% the volume acted like more of a pitch /compression control, and this effect was obvious enough for a few other people to have mentioned it.

Using a lower impedance bulk storage cap cured that completely, and left me with only the wicked crackle this type of amp makes with no further means of soft clipping.

Also in a similar amp, rated for twice the power, yet working off the same supply, say at two thirds its rated voltage, this same thinning, if there at all, was far less apparent even with similar highish impedance bulk storage caps still in use. For it to clip though is only that it runs out of voltage, the inductor isn't nearing the same levels of "DC" current content, so it isn't also experience that last 20% of saturation. Either way I changed that cap to a low ESR variety as well.

I think what I experienced was possibly what your simulations portrayed, made much worse by the higher impedance cap?

Changing to a lower impedance variety should be a simple enough fix if that's in fact all it takes as the difference of it being painfully audible or not. With the improved cap in place though, I can still play around in that last 20%, and not know it until it crackles.

Even the loss in clarity seen by partial clipping is pretty well overtaken by everything else rattling at those levels of insanity, and I can't for the life of me detect any issue caused by minor distortion increases at those levels.

BTW, insanly dynamic recordings never are played more than 3/4 volume. Faster transients on those driven into clipping would be well deafening and ruin any enjoyment almost instantly. Most material isn't all that dynamic though, and can be blasted higher without worry, but it's still pushing the limits of sanity.

So the only real argument I've seen about this is from pro audio guys saying they need amps that reliably work in that last 20% of their range 100% of the time, amps that need to be cleanly driven beyond clipping constantly, just because it's how they get used. Ok fine, but those PA amps are also rated to what, 10% THD or much worse at full power, not the 1% of the amp in question, but they don't mention that in their arguments either?

Cheers

I agree exactly, even a semi systematic approach is required in order to ascertain fact from fiction when such all encompassing statements are being made, yet all we've seen here so far has been the total opposite of that, including this recent listening session which didn't begin to compare apples to apples. Strange.

Analogspiceman's previous work has shown it to be a problem, when it is a problem, and anyone can judge for themselves how horrible it is in their own amps. His work has also shown it to be a complete non issue for any reasonable listening level, beyond which I dont' think one can reasonably argue the effect of those last few percent in THD.

Your proposed systematic approach may be used to confirm if maybe what's measurable distortion is also audible. In this case we already know there to be measurable distortion... is it audible? I dont' think so, and audibility will always be just a matter of opinion.

I know the project you speak of. We're unaware of the actual circuit itself, the optimizations and trade-off's either version of module have or haven't undergone, and the hysteresis method of oscillation is prone to its own anomalies...

What you could maybe do instead is just mix a clock in with the input of the UCD, but then you lose all advantage of it being a self oscillating amp, changing the nature of things completely. Still no direct comparison here, but maybe a useful one, you could see which is most offensive between a cheap clock based amp that doesn't account for additional delay, or a simple self oscillating one that's prone to minor non linearity in that last ~20% of range. I kind of doubt you'll come out of that thinking much less of the self oscillating version.

Cheers

Hi Chris,
I am of the opinion that a systematic approach could help mucg more than just confirming simulation results by a distorsion measuremt.
Modern measuremnts allow quite a detail insight. Besides THD we can measure IMD, TID, group delay, decays, step responses, complex output impedances over the entire audio range and much more.
Correlating such measurements with topologies and listening results would be a profesional base to design amplifiers with reasonable optimizations. Unfortunately nobody is willing to invest the time to do such fundamental examinations.
I am not so much in trouble with that. I am not a professional and just playing around for fun. And professionals do probably not spend the time and money on that, because the Audio market does NOT act according facts. The reason for this is simple. Human audio perception is highly subjective. As soon as the compared equipment is delivering at least a certain level of sound quality, even the more accurate sound might be rated worse. To reach this level, where the subjective impression becomes important or even dominant, is very simple with modern tech.
In addition to that most audio customers have no idea about physics, but love myths.
So the proper way to earn money is to combine a low cost design with the right story.

From commercial point of view more extensive detail examinations would be a silly thing to do.
In this field the know how is only growing because of some passionate design nerds out there.

Hm, this statement might call other readers to reply into this thread and I guess I must power up my defensive shields now.

_________________
Best overall performance is derived by the right trade off.

Hi, I had already asked about it here http://www.diyhifi.org/forums/viewtopic ... t=0#p25107 , and asked Bruno before too, his opinion was around -The UcD isn't euphonic amp, but neutral and transparent. Linear amps with similar sound does exist e.g. Bryston. I had tried to find nature of differences in the UcD and typical linear sound, my choice was Stax DA80M vs UcD400ad (coupling elcap shorted, Yageo replaced to the Elna Cerafine). We have results similar to John's, especially when Utopia with 6'' woofer used (>50hz -3db), UcD sound is too dry, bass fast, controlled, but at the Yello "Resistor" track (very low bass content <<50hz), sound was almost complete bassless - 6" woofer flays with 1" excursion in the silent, stax at this track had room filled bass. Sound stage of the stax was more complex, not so flat, with Z dimension. Reverberations through the stax was noticeable longer, vocals had more body. Also i had tried to hear simulated (.wav in/.wav out) UcD/Stax differences -no chance, the difference nature obviously came "under radar". Ok, next i had made by my EMU1212M two records (24bit ucd.wav vs stax.wav) directly from speakers at the typical loudness, and voila - differences between the two wav's is easy to hear (though not so easy like from amps directly)! So, ucd.wav vs original.wav had no or some little differences yet, but the stax.wav was completely other - more body, fatter bass, longer decays etc. I'm still not sure about nature of the difference, maybe thermal distortion of the Stax, maybe something other.

Hi Markus,

I agree with most of that, and we will not engage in the usual measurement vs what I hear, arguments of old. Suffit it to say I agree measurements are meaningful in terms of sound quality to a certain degree.

I'm just trying to keep the issue focused:

-It sounds less than expected
Granted..... why?
-Self oscillating is bad, and I know why
-doubtful, and proven unlikely at anything less than pretty crazy listening levels, where I've found it's still not really a problem.
-But it doesnt' sound as good as other classes of amps that it can't really compare to
- it still sounds less than expected
Seek new problem. Maybe a more obvious one to start with. Build a nice solide foundation before worrying if the top of the tower will hit the ground.

Ivan, Hi.

Hypex has a certain house sound, they all do. You've described it. Their house sound just changed and I have no idea what it will be like now, well, maybe some idea. It wont' be as hard, won't be as edgy, will be fuller with more body in detail, and lower more potent bass. If i really stretched my guess I'd say it will have one hell of a sparkle to it too.

Still, we should realize that Hypex house sound is not = to "self oscillating" sound. The design of the actual oscillator ensured this, other than a big "maybe" for that last 20% where they're a bit less than perfectly ideal.

I think changing two caps for what was are a boutique choice in terms of what you were hearing, with what you wanted to achieve could have been better executed.

That said, two caps and the shorting of the coupling electrolytics doesn't automatically get you out from under the grip of the house sound, it takes just a little more.


Cheers

Hi Chris,
please be aware that I do not support Johns phase shift theory. As described in earlier posting I do not agree, because the frequency of the self oscillating amps shifts up and down two times within one period of the sound signal...
In the same posting I am also wondering if such frequency shift does cause an error all. So I am absolutely not of the opinion that such an modulation would be poor. Personally I am moving away from self oscillating, because of frequency beating and lock ups. Both theoretic issues did NOT appear during listening to the UCD.
So I am not sticking to the modulation-method-sound-issue at all. Also it is almost impossible for me to seek for a why some other amps pleased my ears more than the UCD and why I simply disliked the Mark Levinson. For this I would need to measure/examine/listen probably 100 different amp evolution steps to come from the most pleasant linear amp via class D including UCD back again to a linear Mark Levinson... No chance I already need years to design one proper class D amp (OK, my linear MOSFET Rookie and BJT Rookie were designed within a few months, but they are just super simple linear amps with low power...).
My only foundation that I have is to design only amps which I do fully understand and then verify/make sure that every single stage of it is behaving like I would expect it to behave. Fortunately this approach does deliver reasonable results up to now. Of course I am also trying to learn some correlation between sound and design, but it is impossible for me to do this in a scientific way. ...time is money, somehow I managed to lack of both...

_________________
Best overall performance is derived by the right trade off.

...one may call me Mr. Slow Brain...
From this thread I do understand that A-Spice has also determined some theoretic issue of the frequency shifts at high levels. So let's assume the frequency shift can be an issue.
And now I am going to promote my step response examination again.
May be the UCD step response is depending much on the magnitude of the step?
Linear amps usually do show more or less indentical output wave form shapes for various levels of input steps as long as the input stages do not clip.
May be the self oscilating topologies have a systematical change of step response as function of the step magnitude?
May be the more traditional triangle comparator modulator behave more similar to the linear topologies?
...of course I am just suspecting, but for me it is looking promising to dig into that direction...

A-Spice:
...if you have a working simulation...

John:
It should be easy to measure the step response of the UCDs at various step magnitudes...

_________________
Best overall performance is derived by the right trade off.

In scientific terms, a good hypothesis is one which is falsifiable, i.e. it makes testable predictions. Successful predictions (ie that are subsequently confirmed experimentally) add to its credibility, while a single false prediction shoots the entire hypothesis out of the water immediately.
The "shift in loop phase with instantaneous modulation index" hypothesis is a good one, because it makes such a prediction, namely frequency-dependent THD. The effect should manifest itself even using a single-tone THD test and would show up as a frequency-dependent distortion, more precisely rising at a minimum of 6dB per octave across the full frequency range (from very low frequencies upward). In practical terms, the distortion would be predominantly 3rd harmonic. Mathematically there is no way in which such a modulation dependent phase shift could exist without it already becoming visible on a simple steady-sine THD test in this way. Amplitude-dependent overshoot would also become obvious, but it would not be a sufficient indicator. After all if the instantaneous loop gain is modulation-index dependent, we would also see a difference in overshoot. UcD loop gain is known to be modulation-index dependent. Step response plots bear this out. It is the cost of simplicity.

Thus, a THD versus frequency plot is all we need to verify the hypothesis, and it is the best test.

Since self-oscillating class D amps are closed loop by definition, the easiest target for such a test would be an amplifier known to have constant loop gain over a wide frequency range. A THD vs frequency sweep is done for frequencies up to 1/3rd of this range.

If the hypothesis holds, no self-oscillating amplifier should exist producing a flat THD graph at elevated signal levels.
Conversely, if at least one self-oscillating amplifier is found having such a flat THD vs frequency response, the hypothesis is falsified.

A very short meta-study readily turns up papers describing self-oscillating amplifiers exhibiting exactly constant THD versus frequency for all harmonics residing in the frequency range over which loop gain is constant.

I should stress that only the hypothesis is disproved, not the observation that gave rise to its formulation. It may well be that John has heard a good sounding fixed-frequency amplifier against a less-good-sounding self-oscillating one. Frankly I have no doubt at all concerning whether he heard what he heard (i.e. the nature of the difference, not necessarily the subjective preference). Finding the exact cause, however, requires strict control of variables. This means building two amplifiers which are exactly identical except for the variable under scrutiny. In this case, the variable is the control method. All other things should be made equal:
-Loop gain
-Signal path leading up to the modulator
-Feedback takeoff point (before/after filter) etc
-Power stage
-All parts not explicitly in the modulator circuit

Clearly this means building a "test mule" amplifier that allows precisely the isolation of this one variable, and will show whether it can be attributed with certain sonic characteristics. The antithesis of a controlled trial, of course, would consist of obtaining amplifiers from two sources using different control strategies and subsequently attributing the observed sonic differences to the single most salient difference between the two.

_________________
The Guru Principle: "Everything sufficiently true is incomprehensible. Therefore, anything sufficiently incomprehensible must be true."

Here is the difference plot:

And here is the schematic:

Here is a plot of the outputs from the two amplifiers near the rail. Notice that the self oscillation frequency of the UcD style amplifier has decreased, yet its output still seems to be well centered about the linear equivalent. To make a fair interpretation of the prior difference / distortion plot, one must keep in mind that the "distortion" signal is filtered through a 32kHz 2-pole low pass network. The simulated UcD style amplifier is remarkably well-behaved. -- a.s.

Hi Bruno,

Good to meet you here,

The listening tests have only been between a Phase Shift Controlled Self Oscillating Class D and standard AB amplifiers.

Yes, it would be very interesting to compare with a Fixed Frequency Class D, but as you point out there would still be many other variables.

I have listen to Fixed Frequency Class D, but not in the current system and location, so there are no controls in place, however the effects I heard with the Phase Shift Controlled Self Oscillating Class D where not apparent with the Fixed Carrier based designs, but again it really needs to be confirm under controlled conditions.

I would like to add some more observations from our latest listening session (with Marcus):-

Throughout the session I was sitting off axis to the speakers, so I was not affected so much by the disturbances of the sounds stage, but concentrated more on the overall sound quality – and I can confer with Jockos observations about the lack of detail / audio quality in the “Mid Band”, although I would say upper Midband – lower treble. (The Bass was always completely wrong, but I don’t believe Jocko’s Maggie’s can go deep in a standard size room).

It’s apparent that a Phase Shift Controlled Self Oscillating Class D can sound good with simple music - when the music is not complex and low level such as single vocals & single instruments etc, but once the music becomes more complex and or louder the audio quality completely fails. As Marcus has said, its start to make “noise” rather then detail - the detail is transferred to a strange “Grainy” sound, the effect is rather like an overly compressed MP3 recording, OK on simple music, but rapidly deteriorates on anything more complex. You can hear the vocals and instruments playing, but in no way focused or with any fine detail.

Also, the imaging is very strange, to obtain any form of “centre” sound stage, i.e. an image between the speakers, you must be in the EXACT center of the speakers, move your head even a fraction of an inch, and you can only hear imaging from L or R (and beyond L / R) – but very almost nothing in the centre, where as other amplifiers can form the centre image over a very much wider listening position.

I have no problem to accept that my theory based on the inherent modulating “Group delay” of the Phase Shift Controlled modulation technique is not to blame, but how to explain the extremely acute effects on sound staging and imaging – I’ve never experienced this effect before…. but then I’ve also never listened to a Phase Shift Controlled Self Oscillating Class D amplifier in my own reference system….

So two effects which I cannot say if they are related, the extreme loss of detail and audio quality on louder / more complex music (and anything with Bass Content), and the “between the speakers” imaging problem…

How to investigate / discover the cause?

John

_________________
Hi-Fi??? "I may not be a psychiatrist but I know insanity when I see it - what ever happened to the normal old pervert who was content to sniff bicycle seats and collect women's shoes anyway?"

I am hesitant to wander in the business end of things but...........

Most of "our" customers also have Maggies.........very forgiving in some areas, and incapable of exposing certain flaws. Couple that with a tendency to listen to traditional Jazz, and you can understand why "we" have had success using both UcD and "Ice Box" modules.

I personally have heard both types on conventional cone-driver speakers. The results are generally favourable. No mention of soundstage and/or imaging. Bass quality is usually rated as good as what they presently use. A few thought it had too much......one thought it was thin.

However, the midrange "problem" seems most prevelant on female vocals. (Hey we all know that audiophiles are geeks without a sex life, and.......well.......they vicariously get sexual release through some music.) Any perceived shortcomings in soundstage are not lateral imaging, but more the "presence". More of the front-to-back spacing being wrong. Not too little, but somehow just wrong. To most listeners, the (female) vocalist just doesn't seem to occupy the right place in space. (Don't ask me to translate..........)

This anomaly exists on both UcD and "Ice Box" modules. Both are self-oscillating. But use lots of industrial grade parts. Easy to fix in one, not so easy in the other.

No amount of futzing on "our" part seems to do diddly-squat to the midrange anomaly. Yes, changing parts does help some other areas, but not this one.

Jocko