Sound Quality Vs. Measurements

[bcarso]

Originally Posted by dvv

Brad, I did take a look, but I won't pretend I understood everything perfectly. My reasoning is: if one has one transistor driving another and another, and we assume each transistor has a gain of only 50:1, by the end of that route the signal would have been multiplied (50x50x50) 125,000 times.

If not, then we are using some form of degeneration, or local feedback, along the way, no matter how exotic it may be.

If you are better acquainted with this principle, please correct me if I'm wrong.

I should have explained a bit, as the article is both more detailed and less to-the-point of the discussion.

Basically, the notion that is relevant: for a current amplifier in particular, the translinear circuits "predistort" the signal and compensate for the bipolar Vbe nonlinearity, without global or even local feedback.

The simple example is a two-transistor current mirror. Then use a larger emitter area for the output device, or multiple paralleled devices, and to a first approximation, the current gain scales with the areas. I'll get round to posting some examples.

Barrie Gilbert, who uses the approaches with great skill, reported results for multipliers and other variable gain stages with distortions in the vicinity of -80dB. This was in the 1960's, and as they say, he's kept it up

[nigel pearson]

Originally Posted by bcarso

Translinear circuit - Wikipedia, the free encyclopedia

Thanks for this link . I don't think I would have interpreted it this way had I stumbled upon it . I especially welcome the reference to a current mirror . Well beyond fascinating I would say . I told my girlfriend she is fascinating . I hope she never reads this thread . I am fairly confident she never will .

[AndrewT]

Originally Posted by bcarso

Translinear circuit - Wikipedia, the free encyclopedia

my brain just frazzled.

[nigel pearson]

Andrew it's the pretty pictures I like . I was reading Ultra Linear feedback the other day and there were pretty pictures in that . Suddenly the bleeding obvious jumped out of the pages . It is well worth a read . My brother refused to class UL as feedback . If anyone is so inclined I have a link .

ultra-linear

[DF96]

It is just using the fact that for a BJT over a wide range of currents, the output current is proportional to the exponential of Vbe. Turning it on its head, Vbe is proportional to the logarithm of the current. With logs and exps you can multiply and divide. The clever bit is losing the base current, which is an error term and unpredictable.

Going back to feedback:
Other things being equal, you can maximise linearity by maximising global feedback and minimising local feedback. This is because local feedback only linearises one stage (and possibly its immediate neighbours via impedance effects), while global feedback linearises the whole thing. There is even an argument for adding local positive feedback around the most linear stages to boost loop gain and help linearise the less linear stages.

Other things are not equal. Phase shifts limit the amount of global feedback you can add, as more than three rolloffs (typically, three stages) create scope for instability. Overload (voltage or current) in intermediate stages limit the size of error signal which can be handled. Common-mode distortion at the feedback summing point may limit the ratio between input and error signal. Hence there may be a sensible limit to feedback at the HF end in many circuits. Much more feedback can and is used at LF.

[nigel pearson]

I always say work towards a circuit that works well with global feedabck . Then all options are available . It also indicates a nice machine has be devised . I am not sure if MOS FET's allow more feedback ? It seems that way . I always found what I lost in MOS FET linearity I regained in being able to apply global feedback . I suspect overall that MOS FET's do provide the best solution if my assumption is correct . I have not bought enough of the very best bipolar transistors to say for certain .

I tried with great success the two pole VAS compensation described in Douglas Self's book . It looks like a recipe for disaster and has an unattractive hump in it's curve . It works very nicely , it offers scope for distortion reduction . It costs less than $1 per amp and < 1 hours work . The resistor can be removed to make comparisons . It should work for shop bought amps equally well . If lets say the VAS cap is 33 pF then approximately it is replaced with 39 pF and 330 pF with a 2K2 resistor to the rail . If the 2K2 is cut then it becomes 34.9 pF . There was some confusion about to the rail or ground . I think both are possible as is to the output . Please research before trying .

[dvv]

Originally Posted by DF96

It is just using the fact that for a BJT over a wide range of currents, the output current is proportional to the exponential of Vbe. Turning it on its head, Vbe is proportional to the logarithm of the current. With logs and exps you can multiply and divide. The clever bit is losing the base current, which is an error term and unpredictable.

Going back to feedback:
Other things being equal, you can maximise linearity by maximising global feedback and minimising local feedback. This is because local feedback only linearises one stage (and possibly its immediate neighbours via impedance effects), while global feedback linearises the whole thing. There is even an argument for adding local positive feedback around the most linear stages to boost loop gain and help linearise the less linear stages.

Other things are not equal. Phase shifts limit the amount of global feedback you can add, as more than three rolloffs (typically, three stages) create scope for instability. Overload (voltage or current) in intermediate stages limit the size of error signal which can be handled. Common-mode distortion at the feedback summing point may limit the ratio between input and error signal. Hence there may be a sensible limit to feedback at the HF end in many circuits. Much more feedback can and is used at LF.

Beggin' yer pardon, commodore, this seems to hark back to my hypothesis that the best OVERALL solution is to use a well judged balance between local and global NFB.

Too much of either will not end well for the sound.

Because of the variety of possible topologies, and the way one handles them, there is no one-fit-all formula. I shall be presumptious here and say that this where the art of HEARING rather than merely listening comes in, this is the part where the engineer meets the artist - in making the best possible balance between the two for any given circuit topology.

Experience helps a lot here. For example, I know that when I make a cascode with a gain factor of 6.8:1, if using a lower pair FET (2SK170) with a total bias of 3.65 mA, or if using a pair of low noise BC 550B bipolars with a total bias of 2.2-2.4 mA, my input stage is done. Having made it at least 59 times over the last 10 years, I have had ample opportunity to make sure it really is if not THE best, than at least the next absolute best.

I know what my predriver-driver-output stage will look like, again, I have tried it several times in the past and it does work like a charm indeed. One 5W BF trannie driving MJE 50W drivers, driving 4 pairs of MJL 200W devices (or 3 pairs of MJ 21195/21196 250W devices), you can hardly fail (of course you can, but you really have to be dummy to do so badly).

Which leaves you with the VAS stage. If you only assume that this too will be a cascode, then your work is greatly simplified. Basically, you are left with a few dilemmas, for example should you use a buffer stage between the input and the VAS, should you use dual supply lines, shoud the VAS line be regulated or not, and if yes, which regulator, etc.

Personally, I'm not so much turned on by the fact that my distortion levels decrease somewhat if I throw in a buffer stage between the input and VAS, I could even say that the difference was not worth the time and trouble, but by the fact that it completes the 100% cycle of input being totally cut off from the output - which is good news if you happen to be driving an evil load. Much the same for the predriver in the current gain stage.

Obviously, this complicates the entire thing. Transistors start turning up everywhere for this or that. The crude part of the DIY gang, the transistors counters (people who believe that the less transistors there are, the better the sound, but deny that the best sound is produced by zero transistors ), are appaled at the semiconductor count. They fail to realize that many, or even most, of the power amps considered to be the best in the world, are extremely complex. For example, the Krell FPB series has no less than 118 transistors in the VAS stage only (common MPS series all of them), Levinson is not quite so complex but is not far behind, etc, and the man they swear on, Nelson Pass, may have created the Zen and its family, but in his Threshold commecial series, he uses like 60 output devices alone.

And it's a LOVELY point of argument between Nige and me.

[nigel pearson]

I think what DF 96 is saying is for the ideal amp global feedback always wins . To me it does as it has the low impedance output advantage ( the free lunch ) . It is also the only way as far as I know to tackle crossover distortion ( if I am wrong so be it ) ? He also says why ideal is limited by stability etc .

I do believe that op amps tell us an ideal amplifier is possible . After all a power amplifier is an op amp . As much as I know I will regret saying this I do believe if cables can affect sound an op amp at unity gain might be less degrading . I don't want to open the cable debate except to say I have come to beleive it is connectors if anything .

Sometimes technically op amps perform less well at a gain of 3 compared with unity . My ears thought so also . With a power amp I doubt we would easily be able to do such a test . Happy to be wrong . Very likely a unity gain power amp with good stability would sound superb .

I have never tried one of my power amps at unity gain . I must try . I have tried at very low gain and liked what I got ( 7 ? ) .

You may have gathered I often say things which if the soldiering iron was in use I wouldn't do . I say it to know what others think and why . I never dreamed to get the link on current mirrors . Worth it . I feel a bit embarrassed it didn't occur to me that is how they work .

Honda proved many years ago how to make ideal engines . We none do and they know why . Close to ideal is fine . Practical is even better . Inexpensive is better still . I think I am right in saying the Italians had V8 500 cc bikes in 1953 ? Norton challenged them with a single . I think the V8's were banned . In audio we can have what we like within reason . Funny thing is Norton didn't always loose to the exotic bikes . It was by refinement very fit for it's job . It probably took your teeth out doing it !

[jcx]

adequate bias is required to reduce crossover distortion - no amount of feedback can make the output not have a deadband if under biased - high loop gain feedback can only try to drive the output Q harder, faster through the deadband

local feedback loops can be "nested" around the output stage - which is often the highest distorting stage of an amp: Hawksford/Cordell "Error Correction", TMC, Cherry's Nested Differentiating Feedback Loops

the local gain can help smooth out output stage errors like gm doubling or beta droop - as long as they aren't too extreme and the gain never goes to zero

even the near universal use of unity gain emitter/source follower output Q stages is local output stage feedback

Cherry does point out some advantages can be had with common emitter/source high gain output stages - but in practice they aren't as popular except in the pro PA floating supply amps by QSC et al

basically Cherry's claim is that V gain in the output stage is a good thing for distortion reduction in ealier stages, and no worse or actually the same as using the output devices devices in follower mode in terms of output impedance and stability

the low drive voltage at the bases/gates of grounded emitter/source output stages requires only low voltage input/driver devices which enable faster transitors or even op amp drive

[Wavebourn]

I like nested feedbacks. For example, in Pyramid amps I have quite deep global feedback and quite low distortions (-80 dB of 2'nd order only on 40W), but corner frequency for compensation is about 15 kHz, quite high, despite the output transformer (10K P-P 100W from 20 Hz) is ringing on the frequency about 30 kHz.

Local feedbacks also greatly help to control behavior on transients (if made properly!).

Here is one more interesting amp with local only feedbacks (attached, from vintage magazine). Output stage has deep local feedback by voltage and input resistance of few kiloohms only. Driver has deep feedback by current and high output resistance. The overall result is quite remarkable.