Differential Amp CCS Tail SVCS Plate - Noise Issues

[Wavebourn]

May be some transistor in SVCS is noisy? Or forgot to shunt some Zener by capacitor?
Zeners are noisy. Transistor in SVCS amplifies this noise supplying triode by noisy current.
If you have 2 SVCSs try to feed them from a single Zener string, in such case their noises will cancel each other partially.

 

[Rod Coleman]

Hi Ian,

If the circuit has no noise with cathode resistors - you can suppress/overcompensate the CCS circuit even further. Since the CCSs are bypassed with big capacitors, the CCS bandwidth is fairly unimportant. Just for fun, you could put 47uF all the way up to 220uF in C2 & C3, and see if it can be returned to resistor-tail levels of stability. If this approach works, we can develop a better method using the results you obtain.

 

[Hearinspace]

May be some transistor in SVCS is noisy? Or forgot to shunt some Zener by capacitor?
Zeners are noisy. Transistor in SVCS amplifies this noise supplying triode by noisy current.
If you have 2 SVCSs try to feed them from a single Zener string, in such case their noises will cancel each other partially.

Last week, when I first powered up the circuit I heard a little tik like the sound of something giving out from high V and I thought the noise might be coming from a not quite completely fried part so I replaced all the transistors and two of the electrolytics . It didn't have any audible benefit.

I wonder about my boards' stability because I made them using ExpressPCB mini boards and laid the circuit out for parts I had in the drawer - so the 470uF and 1uF are way over-rated for V and thus bigger than they need to be. The trace leading to the Q3 base is 1.5" long.

I can try sharing the zener string though I don't think your design is the problem.

If the circuit has no noise with cathode resistors

Well, it didn't but then it did so I don't think the resistors were the source.

Hi Ian,

. . . . . .you could put 47uF all the way up to 220uF in C2 & C3, and see if it can be returned to resistor-tail levels of stability. If this approach works, we can develop a better method using the results you obtain.

Hi Rod - Interesting and certainly easy enough to do. Will try in a day or two.


And about that 1M cap bias resistor. "Bias" is presently wandering up and down in a half volt bracket that is slowly dropping in voltage, right now centred at about 7.7 VDC.

 

[Hearinspace]

Hi Ian,

If the circuit has no noise with cathode resistors - you can suppress/overcompensate the CCS circuit even further. Since the CCSs are bypassed with big capacitors, the CCS bandwidth is fairly unimportant. Just for fun, you could put 47uF all the way up to 220uF in C2 & C3, and see if it can be returned to resistor-tail levels of stability. If this approach works, we can develop a better method using the results you obtain.

I clipped in 10uF as it was what I happened to have sitting on the bench next to the circuit. It's clear - clip on, noise gone. clip off, noise back. What interests me is that doing either side alone will also stop it.

1 uF also works
.22uF works and though I can't hear the noise, it somehow doesn't feel black quiet.
.15 works but careful listening with ear right to the speaker you can just detect the white noise at one small location
.022 the noise is easily heaard if right up to the speaker but at low level and has some lower frequency at a stronger level in the mix.
10uF is what I'm inclined to us as it's dead quiet.
None of these seems to affect the music level as far as I can tell by ear.

May be some transistor in SVCS is noisy? Or forgot to shunt some Zener by capacitor?
Zeners are noisy. Transistor in SVCS amplifies this noise supplying triode by noisy current.
If you have 2 SVCSs try to feed them from a single Zener string, in such case their noises will cancel each other partially.

Two Questions.

1) Do I try sharing the zener string by pulling the ground connection on one of the boards and running a jumper from the bottom of R7 (100K) over to the top of the zener string on the other (still grounded) board?

2) I have run these boards with lower than 500V B+ and didn't notice any problems. But now B+ is down to 425VDC. Could this be the source of the erratic adjustment ?


Man this thing sounds good!

 

[Michael Koster]

Hi Ian, do you want some rolloff on the output tail CCS? This one is not bypassed with big caps, like the input, so it may have an influence on sound. But you could try 22n up to around 68nF from drain to gate of the upstairs FET. I suspect it will not be likely to be useful - but no harm in trying.

...

I would only add external feedback capacitance if there is also a gate protection diode. otherwise it's a good way to blow up the 01N100 MOSFET.

Also adding feedback capacitance in this way eliminates the common tail and makes it act like a simple bypassed CCS. maybe that's what you're going for though.

 

[Michael Koster]

I clipped in 10uF as it was what I happened to have sitting on the bench next to the circuit. It's clear - clip on, noise gone. clip off, noise back. What interests me is that doing either side alone will also stop it.

1 uF also works
.22uF works and though I can't hear the noise, it somehow doesn't feel black quiet.
.15 works but careful listening with ear right to the speaker you can just detect the white noise at one small location
.022 the noise is easily heaard if right up to the speaker but at low level and has some lower frequency at a stronger level in the mix.
10uF is what I'm inclined to us as it's dead quiet.
None of these seems to affect the music level as far as I can tell by ear.
...
Man this thing sounds good!

Congratulations on getting the noise to stop.

I'm not clear on exactly what you have in the input cathode circuit.

As I understand you now have the BJT based CCS on the input cathodes as per post 22. Do you also have the big caps and 1M resistor to AC couple the cathodes together?

Since you don't need the common AC tail to do splitting (your input transformer does this) the stiff cathode bias with no common high impedance tail should be just fine.

 

[Hearinspace]

I guess I should put up a new sketch. The circuit is changing so much it's a choice between drawing or soldering.

Yes, the input stage cathodes have Rod's CCS as in post 22 with 2.8mF between the two cathodes.

I tried it without the caps before and got much louder white noise and no music at all. Not sure why. I thought that with the input transformer I'd just be getting a little negative feedback from the unbypassed cathodes.
I should try it again now that a way to stop the white noise has been found.

 

[Rod Coleman]

Ian, excellent! Glad to hear it is under control now.

I know for sure that the CCS you are using is completely stable on its own account, and the explanation I will offer is that the 8233 pair is the real origin of the problem. This is no surprise with 50mA/V valves, and having 2 differentially-connected increases the opportunity for undesired feedback paths. The 10uF works by throttling the bandwidth available through the CCS. The large 2.8mF electrolytics do not form much of an alternative path for high frequencies, since their inductance is quite large. The fact that 1 cap suffices to stop the oscillation suggests differential-mode action, but I would fit the same C to each CCS.

How did the circuit perform for current balance in the two valves? It should track temperature fairly well, if the two NPNs are epoxied together. To get even better match with a PCB layout, you can use two-in-a-pack NPNs from Diodes inc (DMMT51?) that mount consecutively manufactured dice in an SMD outline.

 

[Michael Koster]

I guess I should put up a new sketch. The circuit is changing so much it's a choice between drawing or soldering.

Yes, the input stage cathodes have Rod's CCS as in post 22 with 2.8mF between the two cathodes.

I tried it without the caps before and got much louder white noise and no music at all. Not sure why. I thought that with the input transformer I'd just be getting a little negative feedback from the unbypassed cathodes.
I should try it again now that a way to stop the white noise has been found.

There was no audio with no caps installed because there are 2 separate CCS with no common tail and thus no current variation possible in the tubes to pass signal.

Putting in the cap between the 2 cathodes creates a common tail for differential AC signal, but if it's an electrolytic it should be done with the 2 caps and resistor to DC bias them.

Putting in 10 uF C2 and C3 from collector to base basically "rolls off", effectively bypasses, each CCS through the whole audio band and beyond, creating a CCS for DC but a low impedance path ("stiff" as they say) for AC. This is where the unwanted feedback loop is being successfully interrupted.

With large values for C2 and C3 you shouldn't notice any difference AC coupling the cathodes or not.

You don't really need a differential amplifier phase splitter anyway, as your input transformer does the phase inversion.

PS just occurred to me that the input transformer could also be playing a role in the noise/oscillation problem. Perhaps it's leakage inductance or maybe just the fact that it couples the 2 sides together grid to grid. I made an oscillator one time when I tried to put a CCS under a self-splitting anti-triode circuit with an opamp providing the dominant pole...

 

[Michael Koster]

...I thought that with the input transformer I'd just be getting a little negative feedback from the unbypassed cathodes.
...

Or positive feedback at some high frequency after accounting for some reactive components of the transformer

If I recall correctly, my circuit was common mode bouncing on the leakage inductance of my OPT. The opamp provided the extra gain needed and the CCS removed the damping from the system.

 

[Hearinspace]

The large 2.8mF electrolytics do not form much of an alternative path for high frequencies, since their inductance is quite large. The fact that 1 cap suffices to stop the oscillation suggests differential-mode action, but I would fit the same C to each CCS.

How did the circuit perform for current balance in the two valves? It should track temperature fairly well, if the two NPNs are epoxied together. To get even better match with a PCB layout, you can use two-in-a-pack NPNs from Diodes inc (DMMT51?) that mount consecutively manufactured dice in an SMD outline.

Thanks Rod!
The 2.8mF I listed is actually 4 x 700uf. They are non-polar and put together as L-canceling pairs. (at least according to Rubycon!) I did put in a small bypass cap at one point but it had no audible effect nor any influence on the noise problem so as things got tried it came out and never went back in.

Yes, it didn't seem right to put the cap on only one side.

The circuit seems fine as is. It runs a little more current than the 317 set-up did but that is probably better anyway. They seem stable without any need to glue them together as far as I can tell.

 

[Hearinspace]

There was no audio with no caps installed because there are 2 separate CCS with no common tail and thus no current variation possible in the tubes to pass signal.

Ahaah! I see!!

Putting in the cap between the 2 cathodes creates a common tail for differential AC signal, but if it's an electrolytic it should be done with the 2 caps and resistor to DC bias them.

This is what I don't understand and why I was posting the voltages above. People call it a bias resistor but from the little I know I can't see how the term fits. Measuring this circuit on the output tube cathodes , the neg. terminal of the caps goes up to near cathode voltage (170VDC) on power up but then immediately starts to discharge through the resistor. After a few hours it's down close to ground. Yesterday, after letting it run until the voltage was down to about 5VDC I put a jumper across the bias resistor (to ground) and there was no audible difference.

With large values for C2 and C3 you shouldn't notice any difference AC coupling the cathodes or not.

OK I can try that in a few days

PS just occurred to me that the input transformer could also be playing a role in the noise/oscillation problem. Perhaps it's leakage inductance or maybe just the fact that it couples the 2 sides together grid to grid. I made an oscillator one time when I tried to put a CCS under a self-splitting anti-triode circuit with an opamp providing the dominant pole...

You know, before I tried these little input transformers in this circuit I put them into a single ended amplifier I use in a biamped system to see what it would do. The amp started oscillating.

I'm learning a lot from this! I really appreciate you guys taking the time! Thanks

 

[Michael Koster]

This is what I don't understand and why I was posting the voltages above. People call it a bias resistor but from the little I know I can't see how the term fits. Measuring this circuit on the output tube cathodes , the neg. terminal of the caps goes up to near cathode voltage (170VDC) on power up but then immediately starts to discharge through the resistor. After a few hours it's down close to ground. Yesterday, after letting it run until the voltage was down to about 5VDC I put a jumper across the bias resistor (to ground) and there was no audible difference.

I see your confusion. It's the capacitors that are being biased.

Electrolytic capacitors perform much better with some DC voltage ("bias") across them. The bias is to provide some DC across the electrolytic capacitors but still lets them act like a single capacitor (series connection) across the cathodes for AC. This locks the 2 current sources together for AC but allows them to produce slightly separate DC voltages, thus perfect DC current balance with an AC coupled tail. The node where the resistor connects still passes signal.

That you were able to short this node to ground and prevent it from passing any signal and not notice anything is because your 2 CCS already each have their own low impedance AC path to ground created by the hfe of the transistor amplifying the C-B capacitor to a huge effective value.

PS in your circuit there is no particular advantage in the idle current being perfectly matched across the input tubes. The plate voltage is stabilized by the gyrators.

 

[Rod Coleman]

<snip>
PS just occurred to me that the input transformer could also be playing a role in the noise/oscillation problem. Perhaps it's leakage inductance or maybe just the fact that it couples the 2 sides together grid to grid. I made an oscillator one time when I tried to put a CCS under a self-splitting anti-triode circuit with an opamp providing the dominant pole...

Michael, I agree, the input-trafo circuit is a very promising vector to provoke oscillation. The coupling of phase-shifted components of each side will easily take place, and the leakage inductance of the trafo + wires plus the various stray capacitances will give the required shift. Normally there's no problem, but with 50mA/V and CCS anode loads, its a different matter.

It's worth mentioning, as the sound may improve if the provocatory coupling is reduced.

In the first place, I would locate the trafo 150mm or more from the 8233s, to prevent direct coupling from stray fields of the trafo. Next, wire each grid in high quality cable, with dense screen - from the trafo shell all the way to the grid stopper. Separate screened cable for each grid. Connect screens to the trafo housing at one end, the CCS return at the other, along with the centre-tap connexion.

 

[Hearinspace]

I see your confusion. It's the capacitors that are being biased.

Electrolytic capacitors perform much better with some DC voltage ("bias") across them. The bias is to provide some DC across the electrolytic capacitors but still lets them act like a single capacitor (series connection) across the cathodes for AC. This locks the 2 current sources together for AC but allows them to produce slightly separate DC voltages, thus perfect DC current balance with an AC coupled tail. The node where the resistor connects still passes signal.

That you were able to short this node to ground and prevent it from passing any signal and not notice anything is because your 2 CCS already each have their own low impedance AC path to ground created by the hfe of the transistor amplifying the C-B capacitor to a huge effective value.

PS in your circuit there is no particular advantage in the idle current being perfectly matched across the input tubes. The plate voltage is stabilized by the gyrators.

No,sorry Michael, though both stages have the same basic topology, it was the output stage I was measuring and talking about. It has the cascode mosfet CCS shown in post 1 with no capacitor bypass on the MosFets themselves. (I know I ought to put up a new drawing but my computer set-up is antique and it's tough to get the time to draw then scan, import etc.)

.... Anyhow, it's as in the drawing in post 1 but now with 2 x 440uF caps in parallel from each cathode, connected to each other as those in the input stage are and with their combined neg rail to ground through a 1M resistor.

That's why I'm interested in people's calling the 1M resistor from the inter-cathode capacitors' neg terminal to ground a bias resistor. What is the bias Voltage? It starts at near cathode voltage which in the direct coupled 2A3 is about 170VDC and then slowly falls to ground over several hours. So which voltage is the intended bias, the high voltage or the low one?

 

[Hearinspace]

Michael, I agree, the input-trafo circuit is a very promising vector to provoke oscillation. The coupling of phase-shifted components of each side will easily take place, and the leakage inductance of the trafo + wires plus the various stray capacitances will give the required shift. Normally there's no problem, but with 50mA/V and CCS anode loads, its a different matter.

It's worth mentioning, as the sound may improve if the provocatory coupling is reduced.

In the first place, I would locate the trafo 150mm or more from the 8233s, to prevent direct coupling from stray fields of the trafo. Next, wire each grid in high quality cable, with dense screen - from the trafo shell all the way to the grid stopper. Separate screened cable for each grid. Connect screens to the trafo housing at one end, the CCS return at the other, along with the centre-tap connexion.

The transformer is very small and mu metal shielded, with the shield taken to the local audio circuit ground with the centre tap. It's about 10cm away from the 8233's themselves and about 3cm from the ground trace on the adjacent circuit board.

I can try the shielded cable but I'd have to ask why. If the circuit no longer oscillates why continue trying to fix it. My real concern is the apparent unwillingness of the SVCS to adjust in a lineal fashion. I think I first need to remove it from the circuit and try it with a load resistor to see if it behaves that way on its own. If so the rest is moot. I'll try it tonight.

Thnx

 

[Rod Coleman]

Ian, there is no reason to fix it harder - except that it may sound better. Not certain to, of course. But having what is effectively an unwanted feedback mechanism has the potential to sound bad, at least in a minor way, I suspect.

Glad to hear that the trafo is well-shielded, though.

 

[Rod Coleman]

...but anyway, does the amp sound good??

 

[Michael Koster]

.... Anyhow, it's as in the drawing in post 1 but now with 2 x 440uF caps in parallel from each cathode, connected to each other as those in the input stage are and with their combined neg rail to ground through a 1M resistor.

That's why I'm interested in people's calling the 1M resistor from the inter-cathode capacitors' neg terminal to ground a bias resistor. What is the bias Voltage? It starts at near cathode voltage which in the direct coupled 2A3 is about 170VDC and then slowly falls to ground over several hours. So which voltage is the intended bias, the high voltage or the low one?

The 1M resistor is supposed to pull the junction of the 2 caps all the way to DC ground, while allowing the 2 caps to pass AC signal in series. The capacitors are biased with the same voltage as is seen on the respective cathodes. You shouldn't notice any change in sound once there are a couple of volts across them, and it's mainly a long term thing to keep the caps oxide layer formed.

 

[Shoog]

Don't be tempted to replace the 1M resistor with a smaller value in a direct coupled amp. I tried this once as an experiment. If for some reason the voltage on the cathode drifts high (which can easily happen in a DC coupled amp) the caps will breakdown and start to fizz.
I typically use 63V caps and haven't had any failures when the amps are running normally.

The 1Meg resistor is ESSENTIAL.

Shoog