Cathode resistors and bypass caps.

[Diabolical Artificer]

I'm after knocking up a front end for an amplifier and have been doing some reading. I've kind of used a cathode bypass cap by default, without much thought.

From what I've read and seen by experimenting we use a cap to bypass the cathode resistor to increase gain and keep the cathode voltage at a fixed voltage, however we loose quite a lot by doing so. First without the cap there some FB that reduces distortion and second frequency response is affected negatively, EG HF roll off.

Today I knocked up a cascode using a 6SN7, without the CBC (cathode bypass cap) THD was 0.8% for a 20v RMS OP, with (47u) , it was 7% - big difference. Admittedly the cascode isn't optimised, it could be tweaked.

There's a few things I'd like to try, firstly partial bypassing and secondly using a LED or diode instead of a resistor.

.Lastly when applying NFB you often see a 100 ohm R in series with the cathode resistor with the NFB being "injected" at the junction of the two. I've used a 100r as default, but is there any other consideration apart from minimising loss (as the two R's are a voltage divider) when choosing a value and what is the difference (apart from loss again) in applying NFB straight onto the cathode as opposed to in the junction of the two R's as mentioned previously

Havn't looked at frequency response yet. thoughts most welcome.

Andy.

[ionburn]

I think the cathode bypass cap is often not used on output stages for the reasons you indicate. In general terms the cap has to be a large value to give good low frequency response, although I think for good high frequency response the (larger) cap is also bypassed itself by a lower value for low inductance.

[Radio1950]

In general, bypass capacitors need to have a numeric reactance of one tenth, or less, of the resistance being bypassed, at the lowest frequency in use.
This is a design starting point, and then economic and practical realities have to be addressed.

[Biggles]

I was taught that the general rule of thumb is that the impedance of the cathode bypass capacitor should be a tenth of the cathode resistor value, but I don't recall at what frequency so the whole theory is a bit meaningless! Anything that cancels negative feedback would be expected to introduce more distortion though, I would have thought.
Alan.

[GrimJosef]

Originally Posted by Diabolical Artificer

... what is the difference (apart from loss again) in applying NFB straight onto the cathode as opposed to in the junction of the two R's as mentioned previously ...

If you apply the feedback across a 100R in series with (typically for a small-signal stage) a 1-2k cathode bias resistor then the 100R will sink almost all of the current that flows into that junction from the output stage and the bit that flows the other way, into the 1-2k, will be relatively insignificant. If, on the other hand, you apply the feedback across the whole cathode bias resistor then its impedance might be comparable with or even larger than the cathode impedance of the valve. So now you'd be driving current 'up' into the valve as well as 'down' into the resistor. This isn't quite such a clean situation as in the simple 100R case and you may find that working out the correct value for the feedback resistor is tricky.

And of course if you apply the feedback signal across the whole cathode bias resistor then you won't be able to short that resistor out with a bypass cap. So as well as the global NFB that you're applying you will also have to have the local NFB that the absence of a bypass cap gives you, and the correspondingly reduced gain which you might not want to sacrifice.

Cheers,

GJ

[trobbins]

Did you keep the output signal level constant between those two examples?

If you have a very low input signal level, then parasitic hum back through an unbypassed cathode resistor may become noticeable if the valve has poor Rhk.

As GJ indicates, calculating the gain of the lower triode in a cascode with feedback gets tricky when the cathode resistor is unbypassed, due to a mix of voltage and current feedback occurring at the cathode - the 3rd Ed of Morgan Jones book shows a method to calculate that gain if you ever get that interested!

You are likely to be operating a LED well above its V-I knee, and not have a signal level that pushes operation in to that knee at minimum current swing of the stage, so you should be able to get nice operation of the loadline in a clear region of curves when using a 2-3 LEDs in series (depending on LED colour). But of course you then have no cathode feedback node available unless you resort to including a 100 ohm series resistor type node.

[Oldcodger]

Originally Posted by Radio1950

In general, bypass capacitors need to have a numeric reactance of one tenth, or less, of the resistance being bypassed, at the lowest frequency in use.
This is a design starting point, and then economic and practical realities have to be addressed.

I also remember likewise on faulting courses, but the frequency used was the lowest design frequency for the Amp, working on the principle that the reactance of the cap should be 1/10 of the cathode resistor .

[Diabolical Artificer]

Not heard of the 1/10th rule of thumb, I've always used the usual formula to find the value, IE - 1/(2pi*f*r) with f being 5hz or lower.

"If you apply the feedback across a 100R in series with (typically for a small-signal stage) a 1-2k cathode bias resistor then the 100R will sink almost all of the current that flows into that junction from the output stage and the bit that flows the other way, into the 1-2k, will be relatively insignificant. " Are you saying 100r is too low a value?

"Did you keep the output signal level constant between those two examples?" Yes. I'm using DC for the heaters so that should reduce hum. I've been reading bits of that book Trobbins to design the cascode but tended not to analyse things too much, though I'm trying to look a bit deeper into the AC conditions side of things.

I started off with a cascode as I'm looking at Hedges cascode LTP to drive an OP stage, but think it might be better to use a common cathode gain stage to experiment with CBC's. Will do siome tests today and report back.

Thanks for your input all, Andy.

[GrimJosef]

Originally Posted by Diabolical Artificer

... Are you saying 100r is too low a value?

No. In this case low is A Good Thing. It means that the impedance that the incoming feedback signal sees is dominated by this 100R, which is purely resistive and whose value is well-known. It's low enough that even if you bypass the cathode bias resistor most of the NFB current will still flow into it rather than into the cathode of the valve above it. The cathode impedance is roughly 1/gm, whose value will depend on the operating point you choose. It's unlikely to be much bigger than the datasheet value though, and that's 2.5mA/V, so the cathode resistance will be at least 400R.

Cheers,

GJ

[kalee20]

Leaving the cathode resistor unbypassed can be an improvement, but you have to be careful!

It introduces NFB, so for the same output, the previous stage must work harder. Reduced distortion in the output stage is then offset by increased distortion in the driver stage. Just watch it can cope!

The NFB introduced is CURRENT feedback, so it linearises the output current which is another way of saying the output resistance is increased. Any non-linearity in the output transformer will make its presence felt all the more, whereas with the bypassed cathode resistor in a triode output stage, the transformer and load are driven by a reasonably constant voltage source, which reduces the effect of transformer core distortion.

The damping on the speaker is reduced. And the peaks, troughs, and resonances are accentuated. You may find the sound 'boomy'.

Replacing the resistor by a Zener diode or a series of LED's can be done, and it will give a nice steady bias. But, the compensation effect which takes place with a resistor, as different specimens of the same valve, will be absent - the valve is virtually operated with fixed bias.

[Diabolical Artificer]

Thanks for the clarification GJ.

"Replacing the resistor by a Zener diode or a series of LED's can be done, and it will give a nice steady bias." I suppose you could use a trimmer and diode in series?

I've a test circuit on the bench at present and will try out the various options.

A.

[Argus25]

There is one trap to watch out for, but it is seldom fallen into. While the absence of the cathode bypass cap lowers the gain and significantly lowers the distortion of a valve stage, the presence of the capacitor, if large enough in value, bypasses hum modulation off the cathode from microscopic heater cathode leakage. Different valve specimens have different amounts of this.

So while the capacitor can be omitted in many stages in an amplifier , or reduced in value to alter the frequency response, it should never be removed in a pre-amp input stage, especially one sensitive enough for a magnetic phono cartridge or hum balance issues will result. So this one is better always kept at 25 to 47uF unless the heaters are DC powered. This is the only occasion where I have seen the absence of a cathode bypass cap cause trouble, other than loss of gain.

[G8HQP Dave]

The rule of thumb comparing the reactance of the cathode bypass capacitor to the cathode resistor may be useful, but it is also deeply misleading. This is because the LF rolloff actually starts when the capacitor reactance is equal to the parallel combination of the cathode impedance (which in some cases is roughly 1/gm) and the cathode resistor.

(For a pentode cathode impedance is 1/gm. For a triode it is roughly 1/gm + Ra/(mu+1), where Ra is the anode load (typically the anode resistor in parallel with the grid resistor of the next stage).)

Comparing the capacitor with the resistor alone actually gives you the point where the LF rolloff stops, not starts. As it happens, the likely ratio of the cathode impedance and the cathode resistor can mean that choosing a cap in this way will give you a useful working circuit, but at the cost of creating misunderstanding.

The capacitor will have almost no effect on HF rolloff, apart from reducing the output impedance of the stage which may make it better at driving a capacitive load. On the other hand, by raising gain it will also raise Miller capacitance; whether this is a problem depends on the signal source impedance.

[kalee20]

Originally Posted by G8HQP Dave

The rule of thumb comparing the reactance of the cathode bypass capacitor to the cathode resistor may be useful, but it is also deeply misleading. This is because the LF rolloff actually starts when the capacitor reactance is equal to the parallel combination of the cathode impedance (which in some cases is roughly 1/gm) and the cathode resistor.

Quantitively, that's true! The 'rule of thumb' is just that. Though I reckon the OP was asking about the merits of no bypassing versus bypassing - with an implicit assumption that the bypassing was "sufficient". Actually calculating the roll-off point is, as you say, not a simple divide-by-ten.

For a pentode cathode impedance is 1/gm. For a triode it is roughly 1/gm + Ra/(mu+1), where Ra is the anode load (typically the anode resistor in parallel with the grid resistor of the next stage).

In fact for a pentode, it is more complicated because the screen-grid current needs to be taken into account, and also whether the screen-grid is bypassed to the cathode, or to 0V. A moment's thought has convinced me that if well-bypassed to the cathode, cathode impedance is 1/gm x Ia/(Ia + Ig2), on the assumption that the ratio of Ia to Ig2 stays constant - and also in such case, the cathode resistor should be considered to have the screen-grid feed resistor in parallel with it. If bypassed to 0V, then complexity grows for an exact solution. And if the reactance of the bypass capacitor is not negligible, then the equations get really tedious!

[G8HQP Dave]

Yes, for pentodes it gets complicated. Scaling 1/gm by the current ratios seems about right, provide that the screen is decoupled to ground. If decoupled to the cathode then there is some positive feedback so I'm not sure what happens then.

Anyway, my main point is that the oft-quoted rule is useful but misleading. It took me years to realise that!

[Diabolical Artificer]

Here testing, trying to find the best circuit and trying to answer the cap or no cap question. However just a simple sweep test sees a dire frequency response.

For simplicity I'm using an ECC83, 100k anode R, 1k5 cathode unbypassed, 1m grid R. Anode de-coupled by a 0.1u cap. 300v HT

My test gear is a bit crap, my sig gen's OP drops with higher frequency, I'm having to compensate, increasing the IP a tad to make sure IP at the valve grid is constant.

Anyhoo, with a 1k reference @ 0dB, OP is pretty flat from 1k to 15hz ( rises a tad by 1/2 a dB). Going the other way the 3dB point is at 3800hz with a steady roll off to 12dB down at 20khz. This is terrible, is this normal?

I'm using a Advance H-1 sig gen, sweeping frequency by hand and monitoring the sig gen OP on a Marconi TF2331 distortion meter on the voltmeter. I'm using a B&K Type 2416 AC voltmeter to measure the OP, set to RMS, then using the meter scale marked in dB. I did check it a while back by calculating a drop in voltage of 3dB on a calculator, the meter read true.

So if my test setup is ok, what's the craic? Surely a simple valve voltage amp should have a better frequency response than this?

Andy.

[ms660]

Originally Posted by Diabolical Artificer

Here testing, trying to find the best circuit and trying to answer the cap or no cap question. However just a simple sweep test sees a dire frequency response.

For simplicity I'm using an ECC83, 100k anode R, 1k5 cathode unbypassed, 1m grid R. Anode de-coupled by a 0.1u cap.

So if my test setup is ok, what's the craic? Surely a simple valve voltage amp should have a better frequency response than this?

Andy.

I'm surprised you got much at all with a 0.1uf decoupling the anode?

Lawrence.

[GrimJosef]

Any source of audio can be modelled (simplistically) as a perfect voltage generator in series with an output impedance. In the case of a triode valve where the output signal is developed across an anode load resistor, the output impedance (again simplistically, and in the absence of global feedback) will be the load resistance in parallel with the valve's anode resistance. The anode resistance for an ECC83 is unusually high - around 62.5k, depending on the exact operating point. That would give an output impedance for the circuit of 62.5k in parallel with 100k which is 40k or so.

A problem now arises if you have a load which has any significant input capacitance. Charging that capacitance through the 40k source resistance works OK at low frequencies but becomes increasingly difficult as the frequency rises. At higher and higher frequencies the capacitor gets better and better at shorting the audio signal to ground. If you are down 3dB at 3.8kHz then that could be explained by an RC time constant of about 40 microseconds. With a source resistance of 40k that indicates a capacitance of about 1nF or so. Coaxial cable can have a capacitance of the order of 0.1nF per metre length so for testing the output of an ECC83 it pays to use short cables. The test meter will have some input capacitance (although, to be honest, probably not much). Are there any other stray capacitances on the output circuit (e.g. other test gear) ?

EDIT: I've just checked and although the feedback associated with the unbypassed cathode resistor is applied at the cathode, it does have an effect on the anode resistance. Morgan Jones in Valve Amplifiers calculates the effect for pretty much exactly your ECC83 circuit and his result for ra when the cathode resistor is unbypassed is around 220k. Now you'd need less than 200pF of load capacitance to explain your treble roll-off and I can easily imagine that you will have that.

Cheers,

GJ

[GrimJosef]

Further EDIT: The circuit's output resistance will actually be 220k in parallel with the 100k load which comes to about 70k. So you'd in fact need less than 600pF of load capacitance to explain your treble roll-off. That's not as easy to account for as 200pF would have been, but I can still imagine that you could have it.

Cheers,

GJ

[Diabolical Artificer]

L, said decoupling cap is de-coupling/blocking DC from test lead, not in series with the anode, as you probably knew/guessed.

Thanks GJ, that gives me a handle on things, I'd given little thought to the blocking cap or test lead length, which is long. I'm also running the OP off the anode without a resistive load. I'll go back and sort all these problem's out as well as doing the sums for the various impedances.

Andy.