I'd urge you to use a dummy resistive load at first, rather than take out your speakers from oscillations that could happen until things are tuned in.
Yes good looking out, I always use a resistive load for initial testing. I turned on the solder iron and am crunching some numbers before I go ahead and change the circuit, I should have an update later tonight.
Don't assume anything, lowering the open loop gain does not mean better stability for lowering the closed loop gain. That's a misconception. Besides, lower open loop gain reduce slew rate.
Follow Tomchr's suggestion and read up opamp compensation first. You understand it, there is so many things you can play with to adjust slew rate and settling time etc. on top of stability. You understand how to tame an opamp, you'll understand what I mean.
Just remove all the compensation, remove the NFB and plot first. Post the graph and I am sure people here can help you.
It is good idea to use resistor load first. But opening the NFB should stop any oscillation. Ultimately you need the real speaker as a load as it's part of the forward gain.
Last edited:
A good quality audio amp should be unconditionally stable. The reason is that the effective loop gain can be reduced sometimes, for example under heavy overdrive. A sudden overload can kick off oscillations that then overdrive the amp by themselves, so it gets stuck in the unstable region. This is really Murphy's law: if it can oscillate it will.
If you hear any squeals or farting sounds as the HT dies away at turn-off, that's a smoking gun for conditional stability.
And no, open loop gain has nothing to do with slew rate. One is a small signal property, the other large signal.
If you hear any squeals or farting sounds as the HT dies away at turn-off, that's a smoking gun for conditional stability.
And no, open loop gain has nothing to do with slew rate. One is a small signal property, the other large signal.
Making it stable on a dummy load, is just the first step. It does not mean it is stable connected to real speakers, you should use the scope to check that too.
Lowering open loop gain certainly can increase stability. You need to know what to shoot for. The easy way to do this is to plot the closed loop gain and open loop gain on the same dB (log) graph. You need to calculate, estimate, or measure the open loop gain curve.
The whole idea is to have the open loop gain curve cross the closed loop gain curve at a 45 degree angle or less. A 45 degree angle intersection implies a slope of 6 dB an octave, which is within the margin of stability for the part of the curve that has loop gain. Any slope much steeper than around 45 degrees + implies instability. Once the phase of the output signal exceeds 120 degrees, we are approaching a slope of 12 dB an octave which would in fact result in positive feedback and of course instability.
The above concept in practice is approximate but will get you in the ball park. Once you try to draw a few of these plots you will realize why engineers are so fond of employing a Miller capacitor (aka Cdom), because it makes it very easy to achieve that coveted 45 degree angle intersection and still achieve design objectives.
This easy to use tool is courtesy of Walt Jung from one of his old school books; it's either "IC op-Amp Cookbook" or "Op-Amps for Audio Applications." The concept applies to any type of amplifier.
If the circuit is a single dominant pole roll off all the way, then you are right, it doesn't matter if you raise or lower the closed loop gain. BUT what is the chance of a single dominant pole? If it's only single dominant pole, there won't be this thread as the amp would not oscillate as drawn.
It was my observation before that if you don't have enough loop gain ( not open loop gain) overhead, slew rate decrease.
That's exactly what I was trying to say, using Bode plot method. You plot the open loop gain, then closed loop gain and make sure you have 1 pole 0dB crossover.
Yes, reduce open loop gain do not guaranty stability. In fact, for slew rate reason, I designed a closed loop feedback system with very high open loop gain, then I use 2 of the lead lag network plus a dominant pole to get a 60dB/dec roll off to bring it down really fast and then the 2 lead bend the graph back to a single pole rolloff at the point of 0dB crossover ( when the graph of closed loop gain intersect open loop gain). It worked beautifully. BUT if you lower the open loop gain ( same as increase the closed loop gain) without moving the poles and zeros, the cross over might intersect before the lead kicks in and you get oscillation.
You cannot make a blanket statement lowering the open loop gain ( raising the close loop gain) increase stability. It's all about the design. You design the open loop and close loop gain to tailor to your need.
Last edited:
That I have no idea, I am not experience in audiophile requirement. In guitar amp, people sometimes want it close to the edge of stability to get the sound. that's what people are talking about the famous Trainwreck amp.
Last edited:
The 5-20 did use a lot of nfb, but was still stable, with the intended OPT. (Nobody has mentioned that so far.) My own way of checking stability in practice is simply checking with a 4 - 5 kHz square wave (I am a little lazy to draw Bode plots!). One can observe quite nicely what effect the compensation has. And I agree: An amp should be stable up to open load - or at least with a permanent say 470 - 820 load resistor fitted just to keep some check on a no-load situation.
Since it was raised, I have a problem with using a triode as input stage; Miller-effect viz-a-viz an unspecified input impedance. As Andreas was the only person to mention so far: What about the (to my judgment) poor choice of a 12AX7 as a phase-inverter/driver? High impedances all over the place, Miller capacitance to input stage, etc.? With due respect to Mr Byrith for his alternative, I would much rather use a more suitable phase inverter like perhaps a 12AT7 or even lower µ like an ECC88 instead of going to a triode input stage. One has a better situation regarding output tube Miller-C, coping with the odd grid overdrive peak, etc.
I have always fancied a pentode input stage as more desirable (the extra noise because of the extra grid present is academic in power amplifiers). The screen grid bypass can further serve as a measure to regulate l.f. phase response for l.f. stability where needed with its phase angle returning to zero - not to further expound here (OT?). I would agree that satisfactory low distortion should be possible with some 20 - 23 dB global nfb. (The Williamson achieved 0,06% distortion with 20 dB of nfb.) My own experience with several designs using input pentode - ECC88 phase inverter - 6L6GC output pair in UL has been highly satisfactory and docile, with open-loop frequency response flat over the audio range. (Again, of course, the OPT must be up to its task regarding its own specs.)
Since it was raised, I have a problem with using a triode as input stage; Miller-effect viz-a-viz an unspecified input impedance. As Andreas was the only person to mention so far: What about the (to my judgment) poor choice of a 12AX7 as a phase-inverter/driver? High impedances all over the place, Miller capacitance to input stage, etc.? With due respect to Mr Byrith for his alternative, I would much rather use a more suitable phase inverter like perhaps a 12AT7 or even lower µ like an ECC88 instead of going to a triode input stage. One has a better situation regarding output tube Miller-C, coping with the odd grid overdrive peak, etc.
I have always fancied a pentode input stage as more desirable (the extra noise because of the extra grid present is academic in power amplifiers). The screen grid bypass can further serve as a measure to regulate l.f. phase response for l.f. stability where needed with its phase angle returning to zero - not to further expound here (OT?). I would agree that satisfactory low distortion should be possible with some 20 - 23 dB global nfb. (The Williamson achieved 0,06% distortion with 20 dB of nfb.) My own experience with several designs using input pentode - ECC88 phase inverter - 6L6GC output pair in UL has been highly satisfactory and docile, with open-loop frequency response flat over the audio range. (Again, of course, the OPT must be up to its task regarding its own specs.)
To use 20db of feedback and still get the same input sensitivity I needed an open loop gain of 537. First I removed the EF86 cathode bypass cap and it got me down to 875. Next I triode strapped the EF86 and it went down to 534. With the same value feedback resistor (5.6k) I get a closed loop gain of 52, which is just about exactly where I started for a closed loop gain but now I am only using 20db of feedback
Both lead and lag compensation is disabled I plan to hook it up to a speaker (everything is stable open and closed loop) first in open loop and plot gain vs frequency and then in closed loop same thing.
One more thing, the phase inverter is wired correctly but grid number 1 is 10v less then grid 2 Both plate voltages are 313 and cathode is at 101, but grid 1 is 88v and grid 2 is 98v. There is a 1Meg resistor going from grid 2 to grid 1, and grid 1 has a .22uf cap to ground. Is this normal?
Hopefully tonight after some dinner I will plot the charts for open/closed loop frequency to gain.
Both lead and lag compensation is disabled I plan to hook it up to a speaker (everything is stable open and closed loop) first in open loop and plot gain vs frequency and then in closed loop same thing. One more thing, the phase inverter is wired correctly but grid number 1 is 10v less then grid 2
Both plate voltages are 313 and cathode is at 101, but grid 1 is 88v and grid 2 is 98v. There is a 1Meg resistor going from grid 2 to grid 1, and grid 1 has a .22uf cap to ground. Is this normal?Hopefully tonight after some dinner I will plot the charts for open/closed loop frequency to gain.
Do the plate voltages exist while your meter is on grid 1?
As I see it you are drawing grid 1 voltage down with your meter when connected there. It is a high impedance point and even a 10M resistance meter will draw the voltage down. The fact that the plate voltages are equal shows that the grid voltages must be equal. The common cathode at 101V shows that all is in order.
As I see it you are drawing grid 1 voltage down with your meter when connected there. It is a high impedance point and even a 10M resistance meter will draw the voltage down. The fact that the plate voltages are equal shows that the grid voltages must be equal. The common cathode at 101V shows that all is in order.
Last edited:
It's OK. I hope I'm communicating the concept so others can understand it.
I had a hard time with this for a long time when I started out. I never read a simple and practical explanation of it until the early 80s when I got my hands on a couple of Walt Jung's books. (I read all the old Howard Sams books from the 60s and 70s at least 100 times and I still have them. ) In practice this can still be tricky and I go to great pains to avoid as many poles as possible in my designs.
Transformer coupled amplifiers always look terrible when evaluated like this but you can still come out with very practical designs, due in part to the typical spectral content and transient nature of a typical music program. For example, my best tube amp produced 40 watts before clipping at 1 kHz, but only 12.5 watts at 20 kHz - and this was perfectly practical and in fact better than virtually all tube amps from the era (70s). Let's face it - a 20 kHz tone driven with 12.5 watts isn't going to accomplish anything but drive your dog crazy. There is virtually no program material that would ever require this power.
Reasonably flat frequency response, acceptable damping, and reasonable power bandwidth are about all you can ask of an amplifier that has an output transformer. And yet, many wonderful tube amps have been built and will be built. You have to work with what you have and parse it to your advantage.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.