How I tuned my push-pull feedback amp

While I’m waiting for my bad Edcor output transformer to be replaced, I thought I would spend some time re-tuning one of my other push-pull amps. I built this amp several years ago. The description of this amp is the following:

1. 6V6 output tubes, UL connected at 40% taps, fixed bias.
2. 6SN7 phase splitter/driver.
3. 12BH7 first stage, direct coupled to the phase splitter.
4. Hammond 1650F output transformers. (7600 ohms:8 ohms)
5. 5R4 GYS rectifier, C-R-C-L filtering putting 375 VDC on the plates of the output tubes.
6. 14 dB of global negative feedback, fed from the 8 ohm tap of the output transformer back to the cathode of the first gain stage.
7. The amp puts out about 12 watts per channel into an 8 ohm resistive load.

The parameters I varied for the final tuning were the feedback capacitor (C1) that is strapped across the feedback resistor, and the step network, which is a series combination of a resistor (R2) and a capacitor (C2) that are strapped across the plate resistor of the 12BH7. I did not vary the amount of feedback. That was previously set (by me) at 14 dB.

The basic procedure was to iterate values of C1, C2, and R2 until I got the desired shape of a 10 KHz square wave when scoped on a purely resistive 8 ohm speaker load. First iteration was to start with a set of values that I thought would get me pretty close to the square wave shape I wanted, then look at the reproduced square wave and adjust the passive components until I get the shape I want. When I get the shape I want, check high end frequency response. I want a flat response out to at least 20 KHz.

I found it quite easy to hone in on values that worked by using a 50K linear pot for R2, and once settling on a square wave shape that suited, and verifying that the frequency response of the amplifier was within bounds (not less than 20KHz), I then substituted the nearest E24 value resistor in place of the pot.

For this exercise I wanted to compare two sets of tuning parameters, I’ll call them the A set and the B set, and I wanted to be as unbiased as possible, in the end, picking either the A set or the B set, which ever sounded the best to me.

The A set consisted of passive component values that give a nicely cornered square wave with no overshoot, no ringing, and flat square wave tops. The passive component values selected for the A set were: C1=1300 pF, C2=330 pF, R2=7.5K. High end frequency response is flat just barely to 20 KHz. A 10 KHz square wave at 1 watt gives a shape as depicted here:




The B set of tuning parameters allow for a small amount of overshoot on the initial rise, then no more than one cycle of well damped ringing before flattening out. The passive component values selected for the B set were: C1=1000 pF, C2=220 pF, R2=10K. High end frequency response is flat to 32 KHz. A 10 KHz square wave at 1 watt gives a shape is depicted here:




I then let both sets of passive component values burn in (each) for 24 hours before doing any serious listening. I then evaluated the A set and the B set for another day each using a variety of program material.

In summary, both sets of tuning parameters sound “good.” But I can easily distinguish the differences. The A set (flat square wave) makes the amp sound a bit ragged, or choppy maybe at the highest frequencies. Cymbals, hi hats, snare drums, etc., had just a bit more edge, glare, and hardness to them. The amp just didn’t sound as smooth as with the B set of parameters.

The B set of tuning parameters (slight overshoot square wave), had a noticeably smoother high end. Cymbals, hi hats, snare drums, etc., had a more delicate, feathery, and extended sound. This difference was not exactly subtle. The amp was definitely easier to listen to and was not nearly as fatiguing after a few hours of listening using the B set of parameters as opposed to the A set.

So…that’s my story and I’m sticking to it. All I can say for sure is the slightly bumpy square wave sounded significantly better in this particular amp in my listening room. I cannot say that this is the end-all, be-all way to tune every generic push-pull feedback amp out there. But in my case, it was clearly easy to know which sounds better to me, in my listening room.

In one sense I am surprised by these results since both sets of tuning values produced results flat to 20 KHz, and therefore should sound identical to the limits of human hearing. But they did not sound identical, the B set (giving the slight overshoot square wave) sounding significantly better, and the distinctions between the A set and the B set were clearly obvious to my hearing.

I have my theories as to why the B set of tuning parameters sounded easily better. They have to do more with double-E theory as to what’s happening in the frequency spectrum with feedback amplifiers.

But, I will give nod to this probably being an unfair test across the board...there may be some sort of interaction with the acoustics in my room that favored the B set of parameters. On my next amp build, I will do a similar exercise, and pick the parameters that work best for that amplifier also. Who knows, I may pick the flat square wave for that amp as sounding better.

-Kevin

Interesting information.

Thanks for your insight and pictures.

Kevin -- Your experience mirrored mine from well over a couple of decades ago now. When my goal was a nice looking square wave, I could never achieve a high level of HF stability, and HF detailed always suffered in the listening experience IMHO. When I finally hit on the right combination of HF stability, distortion, and frequency response, the difference in HF presentation was immediately noticeable, and even startling how much clearer and well defined it sounded. It was then that I became aware of just how important a roll HF stability played in the overall presentation of sound from a global feedback amplifier design.

It's also important to realize that most (but not all) high quality OPTs of typical power capability will display the leading edge rise even as a stand alone device when directly driven from a signal generator -- in spite of an otherwise flat response curve it will typically display within its specified frequency range. As a product of the transformer's natural inductance, the hump then is much more a "quirk" of display presentation, rather than a measurable quirk in actual response. In fact, the overall flatness of the wave top after the hump is of far greater importance than the hump itself is. The biggest concern with the hump is not to absolutely eliminate it, but to ensure it is not excessive when the proper mix of stability, distortion, and response has been achieved, in conjunction with fast damping and a flat wave top.

Did you make any checks of stability by checking the waveforms presented with no load and cap only load conditions on the output? They are equally as important as those displayed across a purely resistive load as well!

Dave

My guess, is that the first tuning (rolloff at ~20KHz), had appreciably more phase shift in the audible regions (5K and up) than the second tuning (~32KHz rolloff).

That's a half octave more bandwidth, at the very least. That could make a significant difference in the "timing" of things- I'd expect, to use subjective terms, for the second to have less "smearing" of HF transients (less group delay)...

Regards,
Gordon.

dcgillespie said:

When I finally hit on the right combination of HF stability, distortion, and frequency response, the difference in HF presentation was immediately noticeable, and even startling how much clearer and well defined it sounded. It was then that I became aware of just how important a roll HF stability played in the overall presentation of sound from a global feedback amplifier design.

Click to expand...

Excellent point! There is a number of articles which describes correlation between stability of amplifier and final perception of sound quality.

One of the best: "Stability of the amplifier and its correlation to natural sound sonic signature" - A.Vitushkin, B.Telesnin, Radio 1980 Nr 7 (in Russian). Its all about solid state amps, but the same rules apply to tubes.

Linux -- Thank-you for providing that reference! Is there any English version of it available? I've not read much about this topic and would enjoy reading it if possible.

Dave

LinuxGuru said:

Excellent point! There is a number of articles which describes correlation between stability of amplifier and final perception of sound quality.

One of the best: "Stability of the amplifier and its correlation to natural sound sonic signature"

Click to expand...

Yes!, Interpreting a scopes subtle variations can and is agonizingly subjective
in correlation to what our hearing (in training) is trying to perceive.

Are those output transformers have ultralinear taps and are you using them?

Stability measurement

Interesting thread! Dave Gillespie - would you be willing to describe your stability test?

Thanks,
Dan

dcgillespie said:

Linux -- Thank-you for providing that reference! Is there any English version of it available? I've not read much about this topic and would enjoy reading it if possible.
Dave

Click to expand...

Unfortunately, this Radio magazine published only in Russian since 1924. I can e-mail this article, you will need to OCR it and put through Google translate or something better.

Authors researched phenomena of harsh, dry sonic signature of solid state amplifiers with several NFB loops, which supposedly have superior technical specification (at least on the paper). By means of using several smart tricks, they found a common culprit which was quite hard to detect - oscillation under certain circumstances, which could be as small as several mV, and at frequency as high as several MHz. Such tiny scale oscillation often can't be detected by monitoring output waveform, it is required to match both signal by amplitude and phase, and then feed difference to oscilloscope to look for a traits.

The same rule of correlation between stability and quality of sound seem to be applicable to tube amplifiers. I wouldn't pay attention to this matter until I drove vacuum tube DUT into small oscillation under heavy overload at 20 - 30 Hz.

Other article which cover this subject in some degree:

Amplifier Response Stability & All That, John Moyle, 1958 (Radio, Television, and Hobbies), available at http://www.tubebooks.org/vintage_data.htm

An Ultra-Linear Amplifier, D.Hafler, I.Keroes, Audio Engineering 1951/11
http://www.aikenamps.com/UL.pdf

Some time ago I didn't paid any attention to square wave test of vacuum tube amplifiers (they are not supposed to do so after all), but studying this material reversed my point of view.

PS. Few days ago found that even LTSpice may be good to detect stability problems in tube circuits.

I use the pot trick whenever I'm dialing in a pentode driver. It makes a tedious job quick and easy. Great post.

Linux -- Thank you very much for the additional info. I certainly have the Hafler/Keroes article in my library, but will definitely research the Moyle article!

Dan -- Back in the day when most tube equipment was designed, there really was no defined standard for much of anything regarding stability, except for a generally accepted understanding that emerged stating that good HF stability was achieved if there was a 10db margin achieved between the feedback level employed, and the feedback level that would cause sustained instability -- which was a pitiful standard to say the least. Never the less, that is why Eico and others would specify a "stability margin" as a value in so many DB.

Before that, there was nothing but a huge emphasis on frequency response numbers that went to the moon -- along with low distortion as well. That was great until everybody started complaining about blown out tweeters, which forced the stability issue front and center. But still, even by the end of the tube era, there was no standard developed establishing the relationship between HF distortion, HF response, and HF stability. Hafler was the one who came in and really turned the industry on its ear, producing amplifiers that generally achieved a good mix of the three parameters.

This was a topic I studied for decades with my own tests and development work, ultimately developing my own set of standards for what good HF response, distortion, and stability all looked like in a mix where excellent performance was achieved in all three areas. This is because all three areas are very intertwined: Circuit remedies that affect one area positively can have a direct negative effect on the others, and visa versa.

The tests I use are likely very similar to those used in the very last days of vacuum tube circuit design, which basically amount to (for HF stability) testing the amplifier with a 10 kHz square wave, under all manner of different input and output loading conditions. But as they say, the devil is always in the details: Meaning, the test itself is not the trick; but interpreting the results is. Therefore, it's really not so much what the display is produced with a non-inductive resistive load, but that display coupled with the displays of the other loading conditions -- which then is coupled with the performance co-achieved with response and distortion. Ultimately and unfortunately, I have found that achieving the optimum performance from all three of these parameters is much like designing a good transformer: It's as much art as it is science.

What I can do that might help, is when I get done with my Fisher X-1000 project I am currently working on, I can take a typical amplifier, and use it to produce pics of what good performance (regarding HF stability) looks like under various loading conditions, and what some deceiving resistive load only shots look like that imply good stability, but in fact produce very poor stability -- or good stability, but at much too heavy a penalty to response or distortion. This last scenario is exactly what I have experienced with the Fisher project.

It really is an involved topic for those who develop their own designs, but it can have a very significant impact on the perceived sonic performance that the test gear says should be produced.

Dave

dcgillespie said:

...
What I can do that might help, is when I get done with my Fisher X-1000 project I am currently working on, I can take a typical amplifier, and use it to produce pics of what good performance (regarding HF stability) looks like under various loading conditions, and what some deceiving resistive load only shots look like that imply good stability, but in fact produce very poor stability -- or good stability, but at much too heavy a penalty to response or distortion. This last scenario is exactly what I have experienced with the Fisher project.

It really is an involved topic for those who develop their own designs, but it can have a very significant impact on the perceived sonic performance that the test gear says should be produced.

Dave

Click to expand...

That would be so awesome Dave! Very much looking forward to your pics.

John

dcgillespie said:

Did you make any checks of stability by checking the waveforms presented with no load and cap only load conditions on the output? They are equally as important as those displayed across a purely resistive load as well!

Click to expand...

So far I have done two kinds of stability tests: a large cap strapped across an 8 ohm purely resistive load. The amp appears stable at any size cap. Biggest film cap I've got is a Kimber 16 uF, and the amp appears stable at any frequency.

A cap only test (no resistive load). First I tried a 0.1uF cap, and a sine wave swept clear up to 500MHz doesn't appear to show any oscillation. However, when upping the cap size to 2uF cap, it does appear to oscillate at around 300KHz.

audiodon said:

Are those output transformers have ultralinear taps and are you using them?

Click to expand...

These output transformers have ultralinear taps at 40%, which I am using in this amplifier design.

dcgillespie said:

It really is an involved topic for those who develop their own designs, but it can have a very significant impact on the perceived sonic performance that the test gear says should be produced.

Click to expand...

Indeed, as I am discovering.


Also, I tried scoping the transformer directly by removing the power tubes and connecting the square wave generator directly to the secondary, while connecting the scope probes directly to each of the plate leads. I was hoping to get a square-wave looking result on the scope....what I got instead was a jumbled mess that I could not decipher on the scope screen. It looked like three or four really undamped square waves all together on the screen. The center tap is connected to the power supply cap (I don't want to unsolder anything), so maybe that is partially to explain for the strange scope reading.

However, in the future, it does seem like a good test to "scope the transformer" directly before putting it in the circuit, so that you have an idea of what a 10KHz square wave looks like reproduced through the device, so you have an idea of how to tune the amp to match that shape.

Hi Kevin -- It may be too, that your generator simply cannot drive the low DCR of the secondary winding. You might try driving the primary instead and viewing the secondary.

Dave

Kevin,

What is the arrangement of C1, C2 and R2? Why would you use a different feedback amount for the two tests? I'm kind of shocked at the amount of cap value you had to use to achieve the 1st and 2nd examples...I think that is the culprit in your results.


EDIT:
I think after re-reading your post a few times I've kind of figured out what you are doing. So I take it the feedback resistor off the 8 ohm tap value (R1 ?) did stay the same. You changed the bypass cap value across that resistor from 1000pF to 1300pF (those are huge values). You changed your plate resistor step network resistor value and the cap value..

Is there a schematic for this amplifier. If you are interested email me it and I will reply with some changes. I think your preferred results can easily be improved upon by changing much of the feedback arrangement. Less nasty loop feedback and more local. The local done in a couple entirely different ways.

I personally always start from a zero feedback base line and work my way to what I want for an end result. In my experience the least loop feedback you can get away with and still have reasonable response, low distortion nets the best sound. I think both above example are using way to much cap values to achieve the square waves. I also have no experience with your form of step network. Not that its not used out there its just not what I do.

Craig, I'll sketch up the audio circuit and PM it to you. The audio circuit is quite similar to the KT120 amp that we worked on earlier this year on this forum, minus the better negative bias supply and output tube balancing mechanism.

Positron, good info thanks. And thank you for the kind words!