Sound Quality Vs. Measurements

[tvrgeek]

Originally Posted by Pano

Funny about the Quad 405. It and the 303 were part of one of my first amp A/B tests. Up to that point I really thought amps didn't make much difference. I was amazed to hear much difference there was between amps that basically "measured the same".
The 405 and 303 aren't much different, tho.

What is even funnier, is there are Chinese ripoff kits of these amps. You would think they would pick one to rip off that was up to modern standards.

[Wavebourn]

Originally Posted by jcx

why is it that anyone could expect “simple” answers to such an ill posed question as (amplifier) measurements vs sound?

Because people want simple answers. Like, without deep knowledge of what, how, and why we design, to be able to judge results of our designs objectively. While what we are trying to achieve, to better fool subjective imagination as if sounds are live.

[vanderpol4]

Ok here is simple enough test, take for example the spectrum of a trumpet playing g
and run it through the stereo amplifier in question, subtract the original from each channel and compare the resultant G(w) acceptable to G(w) unacceptable for different amplifiers . The metric to compare the difference spectra may not be intially obvious but I would suggest a power summation split above and below 2khz may be enough. Or even more interesting would a listening test of the measured spectral distortion. Also interesting would be if phase angle P(w) between channel distortion was at all correlated to subjective dislike.

[ThorstenL]

Hi,

Originally Posted by tvrgeek

In this specific case, it does not require exotic or expensive design as a mere Rotel 840 or 951 passes her subjective yes/no test, where a lot of others costing quite a bit more do not.

Very interesting.

The Rotel 840 design is essentially a variation of the Ottala/Lohstrom design, with heavy degeneration in each voltage amplifying stage. Preamp uses NE5532, but at low gain and low levels (where they are not as bad as can be). They even omitted the speaker relay and output inductor...

At least this would argue in favour of your wife liking amplifiers more build along the lines of say a John Curl or Charles Hansen with low or no feedback, compared to the kind of designs promoted much in this thread (maximise global looped feedback).

I'd probably try to modify the heck out of it (as the basics are excellent) rather than buying an expensive Amp...

It used to be a hobby of mine to hotrod cheap gear with good basics into stuff I would drag around and beat up very expensive gear with. The look on the sales peoples face when the cheap consumer stuff I dragged in "to compare to their good gear to see if it was worth upgrading" made their notionally 10 Times as expensive gear bad in direct comparisons. Only one dealer ever caught onto my game, despite the changed RCA's and Binding posts on the outside being clear indications of "Beware of Hotrod..."

Ciao T

[ThorstenL]

Hi,

Originally Posted by tvrgeek

a mere Rotel 840 or 951 passes her subjective yes/no test

The RB951 poweramp is even more interesting as it contains many design features one would find generic. The two or three things "out of the ordinary" are 22dB degeneration in the VAS Stage and the VAS output loaded so that the VAS Gain is padded to around 56dB, the input stage has also lowish gain but lacks any degeneration!

The frequency compensation is also not quite the generic way.

The parallels I can see between both Rotel Amplifiers is that they have relatively low open loop gain and try to linearise the VAS, plus, they do not use Miller Compensation.

Take your pick from the above. My personal take would be to have a look at the Miller compensation, as that is about as problematic as it is common... It would also explain why D. Self's carefully optimised and degenerated "blameless" amplifiers sound NDFG to these ears...

Ciao T

[DF96]

The GedLee metric is interesting (see link in post 550). It measures deviation from linear gain, with an emphasis on nonlinearity around the centre/zero crossing. As presented in the paper it seems to lack a scaling factor for overall gain, but I assume adding that would be a trivial change. The cos^2 term does the centre weighting. The (D^2 F)^2 term is the non-linearity, and will not distinguish between odd and even order because of the overall squaring.

I suspect that the main reason this metric seems better than THD/IMD is because of the centre weighting. It will pick up nonlinearity mostly at low signal levels, whereas THD picks up mostly high level problems as any polynomial nonlinearity will have low distortion at low signal levels. A plot of THD vs signal level would be more revealing. THD of a music sample is almost meaningless, as it has to be an average over time and amplitude.

Given that some of their transfer functions were non-monotonic it is not surprising that THD/IMD did not do too well. You can come up with nasty transfer functions which create cancellation of harmonics so measure well and sound bad, but how close to real life is this? Below clipping, how common are non-monotonic amplifiers? This is like the mathematicians' love of pathological functions to exhibit an arcane point.

GedLee does not take account of any memory effects, such as the thermal effects we were discussing earlier, althouigh it might spot their results. In real life GedLee might be difficult to measure, as it involves dividing the output by input then finding derivatives - prone to noise, interference and bandwidth limiting. So an interesting idea, but probably of less use in practice than in theory.

[ThorstenL]

Hi,

Originally Posted by DF96

Below clipping, how common are non-monotonic amplifiers?

Very common.

Ciao T

[a.wayne]

Originally Posted by ThorstenL

Hi,



The RB951 poweramp is even more interesting as it contains many design features one would find generic. The two or three things "out of the ordinary" are 22dB degeneration in the VAS Stage and the VAS output loaded so that the VAS Gain is padded to around 56dB, the input stage has also lowish gain but lacks any degeneration!

The frequency compensation is also not quite the generic way.

The parallels I can see between both Rotel Amplifiers is that they have relatively low open loop gain and try to linearise the VAS, plus, they do not use Miller Compensation.

Take your pick from the above. My personal take would be to have a look at the Miller compensation, as that is about as problematic as it is common... It would also explain why D. Self's carefully optimised and degenerated "blameless" amplifiers sound NDFG to these ears...

Ciao T

Hello T,

Audio is a bit strange, one man's euphoria is another's anathema, could you give your definition of good sound? What makes one design better than another, what's your intention when designing, do you go for accuracy ( faithful to the input signal) low tim, thd , bandwidth, et al or do you go for a specific sound and adjust accordingly ...?

Regards,

[ThorstenL]

Hi,

Originally Posted by a.wayne

Audio is a bit strange, one man's euphoria is another's anathema, could you give your definition of good sound?

Yes. It is sound I like. I like sound that seems to blurr the boundary between illusion and reality. It is NOT about Bass or treble or midrange or space/sound stage.

I have for example heard some british solid state systems (not Naim) with crazy French speakers that did the "blurring of boundaries" superbly well, though they for me at least induced unacceptable colourations on some music, despite sounding somehow a lot like live and they had a soundstage as flat as a chinese pancake (the one you get with the imitation peking duck they serve in gwailo chinese chop shops).

I can tell what it does not relate to (at least not directly). It is not distortion levels in the Amp or elsewhere, it is not frequency response aberation.

In addition I'm a little like tvrgeeks wife, there are some qualities in reproduced music that I find make want to almost run from the room screaming. I find this quality physcially uncomfortable. I have heard this objectionable quality with a lot of systems, but the epitome must be the system a UK Distributor of expensive US High End used to bring each and every London HiFi Show in the late 90's and early oughties.

The system comprised of amplification and and sources from one company and speakers from another, both High US High End. The sound was so physically unpleasant that I really could not stand being in this suite for more than a few minutes. The culprit BTW I concluded where the amplifiers as I heard the speakers with other electronics and then this quality was lacking, while I heard the electronics also with other Speakers (big ESL's FWIW) and while this quality was not equally obvious and objectionable in this second system, it was there. For the record, the electronics have an impeccable measured performance.

Originally Posted by a.wayne

What makes one design better than another, what's your intention when designing, do you go for accuracy ( faithful to the input signal) low tim, thd , bandwidth, et al or do you go for a specific sound and adjust accordingly ...?

Given that the input signal is not really faithful to the performance and given it is going to be reproduced through transducers which are very unfaithful to their input signal, I can see little point pursuing "being faithful to the input signal" at least in terms of THD and such very far. However given my background I dislike most very obvious colourations. So no, I do not go for a specific sound, but rather the absence of such.

You might want to read the review in March Stereophile by Ard D of one of my designs. I would say his review captures very well what I am trying to achieve, or perhaps have achieved.

Ciao T

[DF96]

Originally Posted by ThorstenL

Very common.

Could you expand on this comment? Even bad crossover distortion will not necessarily cause non-monotonicity. I explicitly excluded clipping behaviour, as there it could be common.