Capacitors have been replaced about 4 years ago. It does work properly, the oscillation is small and inaudible. It is being somehow kept at bay with unorthodox capacitor sizes, to push phase shift below the low frequency cutoff of the transformer. The thing is, I would like to make it work better, it seems to me like this design is one band aid on top of another.
Yes, this is measured with a Simpson 260 set to AC volts. It is flat to about 10kHz. I usually use it for measuring amplifier power output.i verified it by measuring the voltage at the input and the output. To make sure it is not the meter acting weird.
Looking at the schematic, it does appear to be a marginal low frequency stability margin issue due to C6-R8, and C8-R13 and its opposite C9-R15, and the Valve Ra-OPT primary inductance.
Changing loop gain (ie. C5) at low frequency will impact the level of feedback down at the frequency band of instability.
This is typical of such amplifiers, going back to at least the Williamson. The often used technique to significantly improve LF stability margin was to change say C6-R8 to include a shelf filter, so that a substantial chunk of LF feedback was rolled off at a higher frequency, such that once frequency got down to sub 10Hz, the margin was much better as the shelf network phase shift had waxed and waned by then.
The variability in OPT inductance is significant as signal level is increased across the winding, which moves the Ra-Lpp corner frequency around, and the roll off is not a simple single order.
The easiest way I have observed that is to use your scope in X-Y mode, and have an oscillator that goes down to at least 1Hz, and slowly transition down and up through the frequency zone of concern to see the phase difference change, and likely see the level of peakiness of the gain change - both with and without feedback applied (although its influence diminishes the further down you go).
One means to alleviate the issue is to filter low frequency incoming signal content.
Changing loop gain (ie. C5) at low frequency will impact the level of feedback down at the frequency band of instability.
This is typical of such amplifiers, going back to at least the Williamson. The often used technique to significantly improve LF stability margin was to change say C6-R8 to include a shelf filter, so that a substantial chunk of LF feedback was rolled off at a higher frequency, such that once frequency got down to sub 10Hz, the margin was much better as the shelf network phase shift had waxed and waned by then.
The variability in OPT inductance is significant as signal level is increased across the winding, which moves the Ra-Lpp corner frequency around, and the roll off is not a simple single order.
The easiest way I have observed that is to use your scope in X-Y mode, and have an oscillator that goes down to at least 1Hz, and slowly transition down and up through the frequency zone of concern to see the phase difference change, and likely see the level of peakiness of the gain change - both with and without feedback applied (although its influence diminishes the further down you go).
One means to alleviate the issue is to filter low frequency incoming signal content.
Last edited:
Try using a 200k control grid (leak) resistor on the EF86. And increase the capacitance 20X of the cap that drives it, so that your low frequency oscillator will get through.
If the transient of taking away the large low frequency signal causes large negative feedback voltage to the EF86 cathode, you may be drawing grid current on the EF86 (will not be very good with a 4.7 meg resistor and trying to settle back to grid leak bias level).
Just one more thing to try, I am just saying.
If the transient of taking away the large low frequency signal causes large negative feedback voltage to the EF86 cathode, you may be drawing grid current on the EF86 (will not be very good with a 4.7 meg resistor and trying to settle back to grid leak bias level).
Just one more thing to try, I am just saying.
This is a fantastic idea. Now, the challenge is to figure out first of all what kind of filter is appropriate, and then design it such that it doesn't spoil the damping factor too bad at the high end of the shelf filter. I'm going to have a look for examples of this done before, I'm curious how it looks. I assume just a small value capacitor in parallel with a resistor, in series with a coupling capacitor.
Small coupling capacitors alone don't work this is for sure.
Limiting bandwidth of the input signal is of course viable, but if I can cure the root cause, I'd be happier knowing it.
Good idea! Got to fire all guns at once on this one, I dont see the issue getting better without trying everything.
It may be the same issue in the recent thread Occasional low frequency problem, and references for shelf filter design were linked in post #28, and its use in Williamson is briefly touched on in link in post #22.
Do you have a scope with XY plot ?
My soundcard setup with REW can identify response down to 2Hz, and an AWA Williamson with Red Line AF8 (OPT made for Williamson circa 1948) shows a low frequency peak that changes with signal level, with max peak of +6.2dB at 5Hz at 0.5W. An X-Y scope shows the amplitude swelling up as the frequency zone is approached from either direction.
Williamson had an OPT with somewhat higher primary inductance, and along with KT66, that peak was down a few Hz.
Do you have a scope with XY plot ?
My soundcard setup with REW can identify response down to 2Hz, and an AWA Williamson with Red Line AF8 (OPT made for Williamson circa 1948) shows a low frequency peak that changes with signal level, with max peak of +6.2dB at 5Hz at 0.5W. An X-Y scope shows the amplitude swelling up as the frequency zone is approached from either direction.
Williamson had an OPT with somewhat higher primary inductance, and along with KT66, that peak was down a few Hz.
Was that original Williamson amp wired with the KT66s in Triode mode? That is a bit different than the Pentode wired amp we are talking about here. How do you drive an OPT's primary inductance? With low rp 'triodes', and moderate negative feedback? Or with high rp pentodes, with lots more negative feedback?
Last edited:
The referenced AWA Williamson used 6L6GB triode mode, so pretty close Ra to benchmark KT66.
Williamson just got away with using 6 to 7Hz CR corner frequencies, and requiring OPT primary inductance at least 90-100H at a low 5V excitation, which pushes the Ra-Lpp pole down around few Hz. And he also uses the first and second stage plate load decoupling to provide some compensating phase shift by locating those zeros below 1Hz. Even splitting and raising the coupling CR corner frequencies does not separate them enough to help in some cases it appears. Really good modern OPT's (eg. like Patrick Turner describes) can push Ra-Lpp even lower for better margin, but for most other OPT's it looks like a LF shelf network may be required to remove LF peaking.
Williamson just got away with using 6 to 7Hz CR corner frequencies, and requiring OPT primary inductance at least 90-100H at a low 5V excitation, which pushes the Ra-Lpp pole down around few Hz. And he also uses the first and second stage plate load decoupling to provide some compensating phase shift by locating those zeros below 1Hz. Even splitting and raising the coupling CR corner frequencies does not separate them enough to help in some cases it appears. Really good modern OPT's (eg. like Patrick Turner describes) can push Ra-Lpp even lower for better margin, but for most other OPT's it looks like a LF shelf network may be required to remove LF peaking.
You may need to make C6 even smaller. 0.022uF into 680k rolls off at 10Hz. You need to establish a dominant pole which is well removed from the other LF poles. You are stuck with one pole set by the OPT inductance. I am guessing that their relatively high coupling cap values were an attempt to use the OPT as the dominant LF pole. Then the relatively low cathode decoupler does a lead-lag which can aid stability. If you want to put a bigger cap in the cathode then you may need to completely re-engineer the LF loop gain.maxhifi said:
For example, put the dominant pole at the input to the second stage by making C6 small. This is the best place to do it because signal levels are low. Try 0.0047uF.
With C6 as 0.0047uF, amplifier remains unstable. The G1 resistor of the following stage is an enormous 680k, to keep the starved EF86 happy and making as much gain as possible. Only 1uF works here. When I had s dynaco st70, i tried a range of coupling capacitors, and nothing seemed to make it unstable. This amplifier is a whole other case!
My scope does do XY mode, I just need to come up with a low frequency generator somehow.
I checked the link for the shelf filter - this looks like a good idea, I wonder what the phase response of a shelf filter is. This looks like the low frequency analog of the capacitor in parallel with the feedback resistor, to make amplifiers stable at high frequencies.
Okay, I think I get it. C5 IS making a shelf filter, which reduces open loop gain and therefore negative feedback by 6dB below the corner frequency. This means that phase is preserved down where the opt rolls off, and the amplifier is (somewhat) stable. The coupling capacitors are all so high in value, to keep their phone shift from interacting with the phase shift of the shelf filter. (This is my theory) this explains why any attempt to increase their value has negative consequences.
I think I will try and see if bouncing remains with C5 removed. If so, I can try and substitute a smaller value for C5, and see if it helps with stability.
I also want to check the output impedance of the amplifier - maybe I could also reduce global feedback by about 3dB or so, to improve the stability of the amplifier, and not impact performance too bad.
My scope does do XY mode, I just need to come up with a low frequency generator somehow.
I checked the link for the shelf filter - this looks like a good idea, I wonder what the phase response of a shelf filter is. This looks like the low frequency analog of the capacitor in parallel with the feedback resistor, to make amplifiers stable at high frequencies.
Okay, I think I get it. C5 IS making a shelf filter, which reduces open loop gain and therefore negative feedback by 6dB below the corner frequency. This means that phase is preserved down where the opt rolls off, and the amplifier is (somewhat) stable. The coupling capacitors are all so high in value, to keep their phone shift from interacting with the phase shift of the shelf filter. (This is my theory) this explains why any attempt to increase their value has negative consequences.
I think I will try and see if bouncing remains with C5 removed. If so, I can try and substitute a smaller value for C5, and see if it helps with stability.
I also want to check the output impedance of the amplifier - maybe I could also reduce global feedback by about 3dB or so, to improve the stability of the amplifier, and not impact performance too bad.
If your amp suffers from conditional stability then reducing the loop gain by reducing the feedback could push it into instability! This counter-intuitive result is a feature of conditional stability.
Alternatively, try an even smaller value for C6.
Alternatively, try an even smaller value for C6.
Bingo: C6 = 0.0022 makes it almost stable, but C6 = 0.001 makes it absolutely stable, and response is still good.
No bounce.. it's gone Square wave still has a small bit of ringing, but I better put this thing back in my system before I kill it with too much soldering on the delicate old boards.
Check for LF frequency response and distortion in the 20-200Hz region. With C6 as small as that you need to develop a lot more signal voltage at the previous anode.
More fiddling... 2.2M in parallel with 0.001uf, in series with 0.22uf C6. The shelf increases feedback below 200Hz but permits stability. Values are obtained empirically, not theoretically.
Headroom at output of V1 appears to not be an issue, distortion at 20Hz is dominated by transformer saturation, and is similar to before modifications. LF damping factor may suffer a bit, but I prefer to have it stable. Listening test is next.
Headroom at output of V1 appears to not be an issue, distortion at 20Hz is dominated by transformer saturation, and is similar to before modifications. LF damping factor may suffer a bit, but I prefer to have it stable. Listening test is next.
If the amplifier is oscillating then there must be some positive feedback somewhere.
Either through the power rails or some other feedback path.
Either through the power rails or some other feedback path.
I'm going to call this fixed - it sounds great now, although it has a little bit of hum, possibly from a dying rectifier diode. The stability is much better though, I can't provoke it to oscillate at all with the shelf filter in place.
Thanks very much for all the responses, it was extremely helpful.
Thanks very much for all the responses, it was extremely helpful.
The problem is too much LF phase shift in what is otherwise a negative feedback situation. Thus the feedback path is the intentional one from OPT secondary to input stage cathode.nigelwright7557 said:
Thank you very much for clarifying this for me - the solution works great. Perhaps it's the placebo effect, but I think it sounds better with the marginal stability issue fixed.
An amp on the verge of instability is likely to have bumps in its frequency response and even bigger bumps in its phase response. Even when the bumps are below the audio spectrum they can still have an effect because the envelope/syllabic rate of music comes into play; at the very least you could get bias shifts and hence gain or distortion shifts triggered by the shape of the music dynamic range. Hence it is best to steer clear of instability.
I think in the old days we used to call this motorboating normaly caused by insuffcient power supply decoupling(input stages) It also sounds as if stability was marginal(never good) overload recovery is very important for good performance I recomend you tune the amp until it is unconditionly stable and let static THD take a backseat. Good luck
One more thing I discovered many years ago for best performance each stage in an amp should be star grounded. Years later Douglas Self in his wonderfull series on audio amp design in ww put facts and figures to the improvement this can make. well worth it.
One more thing I discovered many years ago for best performance each stage in an amp should be star grounded. Years later Douglas Self in his wonderfull series on audio amp design in ww put facts and figures to the improvement this can make. well worth it.
Can someone help me understand this concept? I’m thinking with a high pass step filter in place right before C6, with corner at ~200 Hz, that the forward gain of the amp would be lowered below 200 Hz and therefore if gain is lower, amount of applied feedback would also be lower at frequencies below 200 Hz. Is this not the case? Can someone ‘splain this to me?