Sound Quality Vs. Measurements

Status
Not open for further replies.
Hello !
I have a curiosity.
Let's take, as an extreme, a single ended zero feedback tube amp
then select a speaker that sounds very good with the SET above mentioned.
Then take a solid state high feedback amp and use it with this speaker
What would it be the sound ? bland and clinical ?
I do not know really, but maybe it could be a worthwhile test
The comparison must be made with same speakers.
I think it is almost never the case
Kind regards,
gino


Sid Smith told everyone that they should add an 8R resistor to any amp then listen again if preferring SE valve amps . I am so lazy , I never did . I made friends with Sid when I said " Inside every good triode there is a good pentode trying to get out " . I won my argument after 2 hours at a diner in Long Island . Sid gave in and said the best triode he knew of was the EL 34 . Sid made a version of the Model 9 using 300B to satisfy a friend . Personally I preferred the EL34 version ( much tighter ) . Within reason a modernized Model 9 is the best amplifier I ever heard ( for accuracy ) . It should be said I hate euphonic amps although love it when barely obvious . The best amp of all perhaps is one made by Mr Furiama using 2A3 and what looks like an op amp in valves as the driver . It is SE. No hint of euphonic traits and crystal clear . Mostly I dislike valve amps , not least my own design .
 
ginetto61 said:
Let's take, as an extreme, a single ended zero feedback tube amp
then select a speaker that sounds very good with the SET above mentioned.
Then take a solid state high feedback amp and use it with this speaker
What would it be the sound ? bland and clinical ?
It would depend on what was meant by 'very good' with the SET amp.

If 'very good' means flat, well-controlled frequency response (despite the high amp output impedance i.e. low damping factor) then the speaker must have high mechanical damping. In this case running it with a typical SS amp will result in poor bass, as it will be overdamped.

If 'very good' means 'plenty of bass' (due to an underdamped resonance) then the speaker might sound fine when used with more appropriate DF from an SS amp. The SET-lover will wonder where the bass has gone, having become used to a bass peak which should not be there and is no longer there.

My hunch is that the latter is more likely, although both are possible.

nigel pearson said:
I was tempted to say degenerate the VAS a little . Arguments rage about that and as Douglas Self wrote to me it is the worst idea of all as it doesn't bring the expected benefits . I do it because it helps match stages better ( lifts the base voltage for one ) . Sound wise 0R = punch 47R = soft and spacious 16 R = just right .
In most amps, above the open-loop pole, the VAS acts as a transimpedance stage. Adding degeneration will degrade this, raising its output impedance and so bringing the possibility of more crossover distortion due to the non-linear input impedance of the output stage. It may be that what you were actually doing was adjusting the LF damping factor by reducing global NFB, until the amp and speaker combination matched whatever your ears decided was the right amount of bass resonance damping. In my view this is not the way to design hi-fi, although it could be fine for a guitar amp with build-in speaker.
 
... I prefer 33078 over NE5532 . The one thing I would like is a MC33077 single version with all the 5534 tweaks ( comp , offset ) .

One day, Nige, you'll discover the virtues of op amps such as AD 826 and AD 829, to name but two.

I had many a similar discussion with Graham Slee some years ago. He also stuck to The Great British Tradition of NE op amps. While it is indeed a good op amp, it was top flight like 30+ years ago.

People have learnt and worked for the last 30+ years, you know.

And, as I have said many times, I DON'T think it's the wild voltage slew rate of 200+ V/uS, it's their short settling time, well over 20 TIMES (sic!) shorter than in case of the NE family.
 
...

In most amps, above the open-loop pole, the VAS acts as a transimpedance stage. Adding degeneration will degrade this, raising its output impedance and so bringing the possibility of more crossover distortion due to the non-linear input impedance of the output stage. It may be that what you were actually doing was adjusting the LF damping factor by reducing global NFB, until the amp and speaker combination matched whatever your ears decided was the right amount of bass resonance damping. In my view this is not the way to design hi-fi, although it could be fine for a guitar amp with build-in speaker.

Surely, that depends on how much degeneration you throw in?

And, I suspect much more on whether that degenration is an afterthought, or was designed for?

Which brings me back to my initial statement - low global NFB will really work only if the amp was designed for it from scratch. Reducing it as an afterthought will usually produce a mish-mash sound, neither here, nor there. It will also wreak havoc on the amp's stability, methinks.

On the topic of local feedback (degeneration), it should also be used with as much prudence as global feedback. Ultimately, most amps will use about the same amount of overall feedback (or thereabouts, assuming nothing went seriously wrong), it's just a matter of how you disribute it among local and global.

This is something you have to have a firm stand on before you even start the simulator. In the course of your work, things might change a bit here and there, but generally, you need to stick to your initial idea and work at it.

This is why I usually say I believe in moderate global NFB, somwhere in the region of 20-26 dB approximately, but I wouldn't object a 1 dB more or less if it does the job.
 
That's right and Mr Self is right . I am very glad you said trans-impedance as some say voltage amplifier stage ( OK VAS and why not ) . However sometimes it works to corrupt the VAS a tad . I wrote back to Mr S and said in real life his current mirrors do not balance . I ventured to say 0.55 to 0.7 V is too close to the typical Vce of his current mirror . Adding a small bit of emitter resistance might do wonders . I did find a BCV 61 ready made current mirror would work better ( ideal ) . Some would look at it's modest voltage spec and reject it . Daft as the 0.7 V VAS clamping solves that problem . Next there is the effect of the VAS input impedance . Lets say the current is 8 mA and u = 100 , that is circa
Z in = 320 R . Whilst it is not always true it would be ideal to feed that from at least 100 R . In the Goodman's I gave as example the resistance was somewhere on a 2K5 pot . To be honest a bit of current mirror mismatch might help an amp ( more 2 nd harmonic albeit it a small amount ) . I was being pedantic to tell him that .

Some people put a series resistor into the VAS base . Often with no comp cap ! I can see why . It also looks nicer on a scope . The standard VAS usually looks like a triangle wave if sine-wave in ( VAS base ) . That is because it is working as I to V . I still think the triangle wave causes a ghost in the machine if below 1 : 1 ratio . One thing people get wrong is transistors are great current amps ( look at the curves ) . The modern transistor amp hardly does any voltage amplification ( and doesn't have to ) . Thus when carefully forced to be a current amp it works with near perfect linearity . Class AB throws a fly in the ointment . Also the VAS ( TIAS ? ) .

I think the LF factor is unlikely if taking Mr Self's graphs of that very thing . It looks as if there is a 40 dB surplus . He shows a way of degenerating the amp so as to have equal feedback over the audio band ( all from memory so forgive if I am wrong ) . As he says all very interesting but what is acheived ? Having said that it would be consistent with some of the qualities I heard . Certainly to loose HF correction by doing it is pure folly . 16 R seems to show no measured difference at HF ( strange but true ) . 47 R I didn't like . I don't seem to need any frequency compensation in my feedback loop so maybe what I loose one way I gain in another ? I also find excessive gain works rather well . I use a gain of 100 in one of my 100 W amps so as to feed in any source via a pot . A freind who insisted on using the amp with 100 dB sensitivity speakers had excessive hiss . Even that could be cured . I proposed to use 6 x 2SA 1085 for the input with bootstrapped input resistors and gain switch ( 2K ) . Even with a gain of 100 distortion was excellent ( even at 50 kHz ) .

I have looked at getting BCV 61 and BF 720/21 encapsulated with wires . A job for India ?
 
I believe in moderate global NFB, somwhere in the region of 20-26 dB approximately

its interesting how these low numbers of global feedback keep getting bandied about - this is still in the region where Baxandal's "harmonic multiplication" math shows some higher harmonics may be increased by the negative feedback

while >40 dB over audio bandwidth is quite practical today with RET or MOSFET output Q speed - the Linear Audio Putzeys article is unusually clear but the result that negative feedback should be maximized when used at all if you value accuracy is a standard feeback theory result

http://www.linearaudio.net/index.php?option=com_content&view=category&layout=blog&id=67&Itemid=89 (you have to cruise the site for the free sample of the article)

and at higher loop gains the higher order harmonic "hump" gets pushed into the noise floor - all visible harmonics are decreasing with order
 
Last edited:
Quite few years ago, I read a paper on amp design by somebody from Harman Kardon, don't remember who. Worse, I failed to keep that file for some reason other than sheer stupidity on my part.

Anyway, the gentleman had a clean and clear point. Initially, let your input stage work with a degenerated gain factor of 20 dB, but make sure it's working with better than 1 mA of bias current per trannie and into lower collector resistors (do not go above 2k for a power amp, preferably 1.2, 1.3 or 1.5 k). Assume all is well and go on to complete the VAS and the rest of it.

When testing, you cannot avoid some comensation here and there stability's sake, but the final adjustment of overall gain should be done only by degenerating the input stage. It should not have a gain factor of less than 14 dB and no more than 20 dB for best overall results.

About the only thing he didn't mention was how high up can you move with bias current for the input stage - OK, no less than 1 mA/trannie, but how high up: 1.1, 1.3, 1.5 mA, how far?

I took his advice and am to this day grateful for it.
 
One thing that occurs to me . Would an amp with an adequate feedback factor cope better with a complex load than one with " excessive " ? My instinct it to think it would . I would expect that amp to have better stability ? How many of us are brave enough to use complex loads when testing ? I' m not . I don't mean 8R 2 uF .

Excessive = where laws of mechanics says not required . Too much of hard fact in electronics would not be so in mechanics . Damping factors of 3 are thought to be effective in mechanics ( WW circa 1955 ) .
 
its interesting how these low numbers of global feedbak keep getting bandied about - this is still in the region where Baxandal's "harmonic multiplication" math shows some higher harmonics may be increased by the negative feedback

while >40 dB over audio bandwidth is quite practical today with RET or MOSFET output Q speed - the Linear Audio Putzeys article is unusually clear but the result that negative feedback should be maximized when used at all if you value accuracy is a standard feeback theory result

Putzeys Bruno (you have to cruise the site for the free sample of the article)

and at higher loop gains the higher order harmonic "hump" gets pushed into the noise floor - all visible harmonics are decreasing with order

I can't speak for others, but I do not "bandy" these number just for the hell of it.

I own two amps, one using 17 and the other 12 dB of global NFB, both by HK, and I find them to sound better than most of the others I get to hear.

The few units I built myself also sounded better with lower global NFB, but then, they were designed rigth from the bat to need less global NFB.
 
Cherry, Cordell took on "interface distortion" speculation in JAES articles - there is no "problem" with feeback at audio frequency and output load within the design's drive capability

what can happen is that feedback gain stages "wind up" while the output is clipping as the feedback tries to push the output harder to correct the error it sees

then on recovery from clipping there is delay - "rail sticking" as the feedback gain stages have to slew back to normal operating range

standard RCA/Lin dominant pole amps with unity gain output seem to be acceptible, higher order "2-pole" compensation can give oscillation burst or ringing on clipping recovery
clipping recovery can be improved with various clamping arrangements, diodes sometimes seen across diff pair collectors, current limit in VAS/TIS


if you have enough power, limit the input signal before the amp then even higher order compensation can be used if the amp itself will never clip
 
I always like to talk about op amps that can pretend to be expensive . As I always build prototypes with chip holders I see no caution is required . And guess what .....
If only that was true of power amps .

I have an answer to my question and provided it yesterday . A bit of hiss does wonders . I have always thought digital errors and crossover distrotion to be similar ( digital is worse , no idea why ) . Now might a bit of pseudo random noise do wonders for an amp ?
 
nigel pearson said:
16 R seems to show no measured difference at HF ( strange but true ) .
No, not strange - what should be expected. Moderate amounts of VAS degeneration will simply reduce LF loop gain, thus raising the open-loop bandwidth and increasing LF distortion. No effect at HF, as gain there will be set by the Miller capacitor.

Emitter degeneration is good for the input LTP and current mirror. Bad for VAS, unless the amp is completely redesigned so the VAS becomes a true voltage amplifier followed by a buffer to drive the output stage. Then, of course the input stage would have to ditch the current mirror (or at least add a resistor in parallel) so it too becomes a voltage amplifier. You can't tinker with a design and expect to improve it, if it is already well-balanced. You have to adopt a different architecture.
 
Excellent point about the VAS cap , I have never asked myself about the factors of magnitude . Usually stability is the question . Doubtless the stability requirement robs the amp of many dB's of feedback potenial . I must research this a bit more . I believe my points about how trans-conductance is arrived at and current mirror balance is a valid reason to use a bit of lift on the VAS . I doubt the LF degradation could be detected using ears if only 16R ? I would question the validity of making the amp into a true voltage amp . I am sure that wasn't the point . The HC Lin circuit has much to recommend it . Big thanks about 16R verses VAS cap , I should have seen that . I must push on until it does appear in measurements . I did the same with bootstraps once and had a few surprises , things books don't say or at best only hint at . Doubtless I should be able to calculate or simulate it . I prefer to build it .
 
No, not strange - what should be expected. Moderate amounts of VAS degeneration will simply reduce LF loop gain, thus raising the open-loop bandwidth and increasing LF distortion. No effect at HF, as gain there will be set by the Miller capacitor.

Emitter degeneration is good for the input LTP and current mirror. Bad for VAS, unless the amp is completely redesigned so the VAS becomes a true voltage amplifier followed by a buffer to drive the output stage. Then, of course the input stage would have to ditch the current mirror (or at least add a resistor in parallel) so it too becomes a voltage amplifier. You can't tinker with a design and expect to improve it, if it is already well-balanced. You have to adopt a different architecture.

Exactly what happens with the VAS.

The only workable way out is to use cascodes stages with well paired transistors, an upper pass transistor which can handle reasonable currents (by small signal transistor standards, that is) and a lower transistor which has very low capacitance values all around.

This involves both knowing which two will do very well together (which is a matter of experience and much experimenting) and how much leftover Miller capacitance will be required to be small yet not compromise stability. SOME Miller compenstation will always be required, but it's hardly the same thing whether that's say 10 or 47 pF. God forbid more.

Another point comes to play - is the VAS actively buffered from the input stage or not. Nota at all the same thing.

Lastly, sa Dave pointed out quite correctly, there's also the question of one's output stage. If it's a simple driver-power affair, or pre driver-driver-power three stage deal determines how much current do you have to have from the VAS to make it all work together. A two stage output will require more VAS current, a three stage less, thus offloading the VAS.

As I said, all these are things you have to have cleared in your head before you even sit down to work at it. If you, for example, decide to add a predriver stage half way through, you are in fact going back to the beginning.
 
My 16 R is used with 27 pF Cdom and 8 mA . The only reason I brought it up is I don't consider it local feedback because as far as I know it isn't with this trans-impedance stage . This is confusing as it resembles a degeneration resistor used with voltage amps . Anyway , that aside I think it can tweak the circuit for no apparent loss . I liked the suggestion elsewhere that hiss will mask distrotion when there is enough ( and why not ? ) . When CD players were dithered the distortion figures improved . When they got very good at it there was no penalty elsewhere . Surely crossover distortion resembles the same problem ?

1 Hz I think is the bandwidth of a primitive op amp ( a true joke often repeated ) ? Still good to 20 kHz when used at realistic gain .

Wolfram Demonstrations Project
 
nigel pearson said:
The only reason I brought it up is I don't consider it local feedback because as far as I know it isn't with this trans-impedance stage . This is confusing as it resembles a degeneration resistor used with voltage amps . Anyway , that aside I think it can tweak the circuit for no apparent loss .
It is local feedback, but only at low frequencies where Cdom does not do anything. How much feedback it applies at LF depends on how the stage is driven. The usual current-mirror in the LTP collectors will provide a high impedance drive so the resistor won't do much except to raise the VAS input impedance, which could raise LTP distortion by a tiny amount. A simpler resistor drive from the LTP will be affected more, but these are not seen so often nowadays.

If the 16R doesn't do anything then you won't hear any change! The bare transconductance of the VAS BJT will drop from about 0.3A/V to 0.05A/V, which will affect things even when Cdom is active. Careful simulation may show exactly what happens.
 
Status
Not open for further replies.