I seem to be the victim of all kinds of weird problems with amps. My latest one is that when I use a constant current cathode biasing system based on LM317s as opposed to the usual cathode bias resistor / cap combo, the amp develops a low level oscillation, who's severity changes depending on which output tap I use, 4 ohms being the worst and progressively better from 8 to 16 ohms.
It's at a low level - in the fraction of a volt range, if aggravated it doesn't get much beyond that as the output transformer saturates at such a low frequency. On the 4 ohm setting, it tries to develop into a full power oscillation, whereas on the others it settles at a maximum. With a woofer connected you can push it in time with the oscillation and make it worse, until the movement is visible on the cone, which then sustains itself until the amp is turned off.
With the input valve pulled, there is no oscillation, it certainly seems to be a feedback issue.
My theory is that the resistance of a normal cathode resistor acts as some kind of damping for an LF resonance or phase shift around the OPT, and the (near) infinite impedance of the CCS down around DC exacerbates this.
When I get around to it (hopefully tomorrow) I will try smaller coupling capacitors to the output valves to roll off excessive LF gain (it's a mullard topology so the first stage is DC coupled) and HOPEFULLY this will solve it, although I'm not getting my hopes up
Does anyone have any insight into this before I go chopping things up?
It's at a low level - in the fraction of a volt range, if aggravated it doesn't get much beyond that as the output transformer saturates at such a low frequency. On the 4 ohm setting, it tries to develop into a full power oscillation, whereas on the others it settles at a maximum. With a woofer connected you can push it in time with the oscillation and make it worse, until the movement is visible on the cone, which then sustains itself until the amp is turned off.
With the input valve pulled, there is no oscillation, it certainly seems to be a feedback issue.
My theory is that the resistance of a normal cathode resistor acts as some kind of damping for an LF resonance or phase shift around the OPT, and the (near) infinite impedance of the CCS down around DC exacerbates this.
When I get around to it (hopefully tomorrow) I will try smaller coupling capacitors to the output valves to roll off excessive LF gain (it's a mullard topology so the first stage is DC coupled) and HOPEFULLY this will solve it, although I'm not getting my hopes up
Does anyone have any insight into this before I go chopping things up?
Is your house built over a cemetery? 🙂
By any chance have you reversed the phase of the output transformer in such a way that the feedback is now positive?
Does you amp have a choke loaded rectifier?
The schemo of your amp would definitely help me to help you.
The schemo of your amp would definitely help me to help you.
Too many coupling capacitors. Williamson topology makes an especially excellent phase shift oscillator.
Might I suggest DC coupling? ;D
Tim
Might I suggest DC coupling? ;D
Tim
Link removed at member's request.The PSU is a simple CLC afair, I will post the circuit
It's a Mullard topology 😉
I'm 99.9% sure the feedback is correct, the amp works fine with resistor based cathode bias
I think so. By itself powered with a PSU it seems very stable, and it is loaded with 100nF caps on the output of each chip because before I've had problems with HF oscillation
This circuit is prone to LF instability - you can see this in the original Mullard frequency response plots. There are three LF poles in the gain path: phase splitter (R6,C6), output grid drive (C8/9, R10/17), and the output transformer. Some people forget the PS one because they think that DC coupling avoids this - not true.
In addition, there are two LF lead-lag networks: C7, R13 and the cathode arrangements for the EL84s. I suspect your problem, as you say, is at the EL84 cathode. Try reducing the value of the 470uF decoupler. Alternatively, try changing (up or down) the values of C6 or C8/C9. The CCS turns the lead-lag into a pole so it changes the LF phase shift. Maybe you will have to deliberately degrade the CCS with a parallel resistor.
Another LF problem is that the forward path can have some gain down to very low frequencies which are not filtered by the HT decoupling. Mains voltage variations can enter the amp at the anode of V1.
In addition, there are two LF lead-lag networks: C7, R13 and the cathode arrangements for the EL84s. I suspect your problem, as you say, is at the EL84 cathode. Try reducing the value of the 470uF decoupler. Alternatively, try changing (up or down) the values of C6 or C8/C9. The CCS turns the lead-lag into a pole so it changes the LF phase shift. Maybe you will have to deliberately degrade the CCS with a parallel resistor.
Another LF problem is that the forward path can have some gain down to very low frequencies which are not filtered by the HT decoupling. Mains voltage variations can enter the amp at the anode of V1.
Hi DF96, thank you for your insightful reply!
After further scoping I think this may be the culprit. The HT decoupling, as you say, lets low frequencies through no problem. During oscillation, the supply to the EF86's anode resistor bobs up and down at the same frequency (but not in the same phase). I'm not sure if this the DIRECT cause or effect, but it certainly can't help, as I'm sure the pentode has barely any PSRR! The voltage going to the screen of the pentode in the phase splitter also bobs up and down. This is with the output valves being powered with an independent supply (I thought it might be the output stage causing feedback through the PSU), so it's sagging under its own signal conditions!
I have set to scope the grid of the triode side, that will be interesting. Like you said, C6 may be too big or too small. One of these many rolloffs is the problem!
After further scoping I think this may be the culprit. The HT decoupling, as you say, lets low frequencies through no problem. During oscillation, the supply to the EF86's anode resistor bobs up and down at the same frequency (but not in the same phase). I'm not sure if this the DIRECT cause or effect, but it certainly can't help, as I'm sure the pentode has barely any PSRR! The voltage going to the screen of the pentode in the phase splitter also bobs up and down. This is with the output valves being powered with an independent supply (I thought it might be the output stage causing feedback through the PSU), so it's sagging under its own signal conditions!
I have set to scope the grid of the triode side, that will be interesting. Like you said, C6 may be too big or too small. One of these many rolloffs is the problem!
I forgot to mention:
I reduced the 470uF bypass caps in the cathodes to 100uF, which unfortunately had little effect.
With a 1K resistor in parallel with the CCS, there is still a distinct ringing, but no sustained oscillation This obviously isn't ideal as the valves will draw appreciably more current.
When using the original bias circuit (470 ohm cathode resistors with 100uF bypass cap) there is a slight ringing, but well damped enough to never build into an oscillation.
I reduced the 470uF bypass caps in the cathodes to 100uF, which unfortunately had little effect.
With a 1K resistor in parallel with the CCS, there is still a distinct ringing, but no sustained oscillation This obviously isn't ideal as the valves will draw appreciably more current.
When using the original bias circuit (470 ohm cathode resistors with 100uF bypass cap) there is a slight ringing, but well damped enough to never build into an oscillation.
Last edited:
I went through some of these issues with my modified 5-20. I had increased coupling capacitors to get better LF, then found that a Spice model showed LF peaks at a few Hz. I also saw the output moving up and down as the mains voltage changed. I was using half an ECC81 as the first stage; the EF86 would be worse as it has a much higher anode impedance. I eventually decided that the original Mullard design was actually quite well balanced, so changing it has to be done with care.
I forgot that there is another LF lead-lag in the anode decoupling for the EF86, so changing the HT decoupling can affect LF stability too.
I forgot that there is another LF lead-lag in the anode decoupling for the EF86, so changing the HT decoupling can affect LF stability too.
I have also seen this on mine. In fact, a sudden increase or drop in the mains voltage sometimes seems to trigger the oscillation.
There's definitely something to be said for regulated or DC stabilized power supplies!
Cathode bypasses count as lead-lag networks, but the effect is very small. You'll need to look for something with bigger gain to be useful.
I don't like awful big RCs in supplies. Fewer the better. You could nix C1-R7 and increase C10 to 40uF or so (caps are cheaper than they were back in the day!). R4 and R5 will have to be larger to compensate for the bias change.
It's interesting that R9 is supplied from B+ directly. You could do the same thing for R4, moving it to C10 or B+, and making C2 larger to filter noise better (the screen is the second most important source of noise!). To keep the voltage the same, make R4 larger, or add a resistor to ground (like R32).
Tim
I don't like awful big RCs in supplies. Fewer the better. You could nix C1-R7 and increase C10 to 40uF or so (caps are cheaper than they were back in the day!). R4 and R5 will have to be larger to compensate for the bias change.
It's interesting that R9 is supplied from B+ directly. You could do the same thing for R4, moving it to C10 or B+, and making C2 larger to filter noise better (the screen is the second most important source of noise!). To keep the voltage the same, make R4 larger, or add a resistor to ground (like R32).
Tim
Generally true, but not when a bypassed CCS is used - it is a high-pass filter so adds phase shift exactly where you don't want it, just below the audio band where the output transformer is doing the same thing. It may be that an output stage CCS just won't work when global feedback is being used. If you put four high-pass filters in series with some gain and then add global feedback you need to ensure that they have very different cutoff frequencies, yet the instinct is to make them all go just below the audio band. Without a CCS you only have three HPFs, plus a lead-lag.
Bigger HT decouplers can help, provided they don't add more phase shift in the critical region from about 1-10Hz.
Something seems wrong. Have you really got 4 LM417s in parallel in one cathode, as you show? And why the series 220 ohm resistor? If you have insfficient voltage across the LM417s, they may not work as intended.
Incidentally, that isn't a Mullarg at all, but a Radford, and it has a step network coupling the first and second stages. That's not a bad thing in itself, but it does complicate the LF stability picture somewhat.
Incidentally, that isn't a Mullarg at all, but a Radford, and it has a step network coupling the first and second stages. That's not a bad thing in itself, but it does complicate the LF stability picture somewhat.
Agreed. The CCSs are fighting each other for control. You at least need resistors in series with each LED.
No. You can parallel CCSs - they only fight for control when in series. The only effect of adding resistors would be to reduce the voltage across each CCS and move some heat dissipation from the LM317 to the resistor.
- Status
- Not open for further replies.