More stable biasing for common emitter stages?

[Fancy Lime]

Hi there,

I've recently been messing with discreet transistor designs again, on account of liking the sound the make when clipping. Don't judge me. Anyway, one problem is that the biasing is, at least to some degree, dependent on the supply voltage. A stage that is biased to, say, a collector voltage of 4.5V with 9V supply voltage will generally not have 9V at the collector when running at 18V supply voltage. This can be mitigated to some extent by choosing the bias resistors wisely (see Joe Davisson's great tool here: https://www.diystompboxes.com/analogalchemy/emh/emh.html) but I was wondering if there is a way of dealing with this problem properly. Opamp designers seem to manage it, don't they? So my question is: How do I go about designing a stage that works from 9V to 18V without having to readjust the bias. If I remember past experimentation correctly, Joe Davisson's CCS Drive (https://www.diystompboxes.com/analogalchemy/old/ccsdrive.gif) works fairly well with a range of supply voltages but I cannot find my note on that anymore. I makes sense to me that a constant current source might help stabilizing the bias at different supply voltages, no? This would be a sufficiently simple solution to the problem. Of course there will be some trade-off between complexity and stability.

Also: how would I best go about biasing a stage like the CCS Drive with a voltage divider bias instead of feedback bias?

Also also: We could put a CCS in place of the emitter resistor instead, couldn't we? I've never seen a design doing that and wonder why. Does anyone know an example? I'm asking because that could be useful for a long tailed pair, which I might want to use as a phase splitter.

Cheers,
Andy

[amptramp]

One way to stabilize the operating point of the input transistor is to use a cascode circuit with a Zener-regulated bias voltage for the upper cascode transistor.  The lower transistor would never know what the voltage was powering it.  You could use a current source load as with the CCR circuit so that you still have high gain using three transistors.  The collector of the bottom transistor would run at a fixed voltage regardless of the power supply.  The cascode transistor in the middle would run at fixed base and emitter voltages.  The upper transistor would be just like that in the CCR circuit allowing for a high gain.  Everybody wins.

[R.G.]

Opamp designers seem to manage it, don't they?

Not really, no. An opamp can be driven from one end of its power supply to the other with a few millivolts of offset between the two input terminals. It has a huge amount of feedback and this is sufficient so that a few of the millivolts of the output signal can be compared to a bias voltage set to a fraction of the power supply, and the large gain and feedback cooperate to make the set bias come true.

A CE stage doesn't have nearly the gain or the differential input flexibility to do this all by itself. There are ways to tinker it in, but it is tinkering and won't in general be possible with only the single transistor. One simple way to get a half-power-supply bias from a CE is to use a sledgehammer - use an opamp to sense the average DC level at the collector and have its output drive the base through a high resistance to force the voltage on the collector to come true. Some people would complain about using an opamp to bias a transistor as too complicated, but an opamp really has only two more pins than a bipolar, those being the power + and power - pins.

Constant current biasing helps, but the problem you face is that the collector load has to be either a resistor or a CCS itself, and that means the DC bias point will vary a lot as you change power supply voltage.

I used CCS biasing in my onboard preamp circuit, but I was after a different result and had a relatively fixed 6V supply.

[PRR]

> I makes sense to me that a constant current source might help stabilizing the bias at different supply voltages, no?

No!

Think. If supply voltage changes, transistor current *must* change, to keep the same % bias.

Most opamps do NOT use resistor loading, and this is one reason why.

You could go crazy making an active current source proportional to supply voltage etc yadayada. But a current proportional to voltage is a stupid resistor!

If hFE is known even approximately, you can get "half supply" bias over a very wide range of B+, and a fair range of hFE (note '2222 biases OK with resistor picked for '3904).

[Rob Strand]

I'm with PRR on this one.

The opamp with resistor divider is the best example of what you would like.  At least as a first order approximation.

BTW, biasing at exactly to Vcc/2 isn't always the best place if there are bypassed emitter/source resistors.  Although you could do a lot worse.

A side effect of having the bias point proportional to the supply rail is the supply ripple finds its way to the output and produces hum in the audio.  In the opamp case the bypass cap across the bias resistors lets the DC track but stops the AC modulating the output.  With transistor stages it's not so simple keeping the supply ripple out of the audio.  That's why you see RC networks on the PSU rails in the solid state designs and tube stages.  Of course a modern way to attack that would be to use a regulator ... but that goes against have the output track the DC supply rail      So the next step is an active RC type filter  ... which you see in some Boss pedals.

[PRR]

> biasing at exactly to Vcc/2 isn't always the best place if there...

Aside from any voltage slack in the stack: depends on Loading.

I had a 10K resistor for the collector. If the next-stage load were 10k, then the optimum for maximum swing is 33% not 50%.

[amptramp]

You can get reasonable power supply rejection in a cascode by following the scheme of:

http://tubecad.com/march99/page2.html

This is a tube amplifier which has the low output impedance of a triode but you can use the same trick with the higher impedance of a transistor of feeding a bit of the power supply voltage to the base of the transistor so it will cancel the effect of the power supply ripple.  Transistors should be inherently better at power supply rejection but if you want to get the last bit of ripple gone, this offers a method.

[tubegeek]

A side effect of having the bias point proportional to the supply rail is the supply ripple finds its way to the output and produces hum in the audio.  In the opamp case the bypass cap across the bias resistors lets the DC track but stops the AC modulating the output.

If you filter the bias resistors, which of these is best practice?

On the bias network:
• one cap from + to - supply
• two caps, one from + to bias junction, one from bias junction to -
• one cap across one of the resistors (which one?)
• two caps, one from + to - supply, one across one of the resistors (which one?)

If you filter the bias resistors, do you have to filter the virtual ground? Are these needed?

On the virtual ground:
• two caps, one from + to virtual ground (after the op amp), one from virtual ground to -
• one cap across one of the above (which one?)

Or is it best to do both? (which way?)

I should look and see what The Art says about this, too, huh? Must be in there....

[jonny.reckless]

I've used a simple current mirror to allow variation in supply voltage in the past. This technique is more useful with JFETs due to the varying nature of Vgs threshold and transconductance between instances of the same part, making JFETs notoriously hard to bias, but could be applied to BJT common emitter stages. You have to bypass the emitter / source else you will get zero AC gain due to the constant current sink from the current mirror.

You don't need precisely matched transistors since a few millivolts mismatch is fine in most applications like this.

[Rob Strand]

If you filter the bias resistors, which of these is best practice?

On the bias network:

• one cap from + to - supply

Putting caps across the rails is generally a good practice but that on its own doesn't guarantee filtering of AC ripple.
To get good AC ripple filtering you need a resistor in series with the supply, like the old tube amps.
Putting resistors in the supply causes a voltage drop.   Suppose you used 100 ohm to keep
the voltage drop low, to get the same filtering as a 10k+10k+10uF (+Vref to gnd) you would
need a 500uF cap,  quite large!  If you can get away with only filtering the biasing circuit
you can use smaller caps than filtering the supply rail.   For a transistor etc stage you may
be forced to filter the supply, so that means some juggling of supply series series
and caps to find the most economical solution.  For transistor/JFET/tube stage you really investigate how
much ripple is getting in through the biasing and how much is getting into the supply then make the best choice.
That could be only filtering the supply for higher level circuits and perhaps both for low level circuits.
For buffers supply filtering doesn't help unless it's a PNP transistor with the emitter resistor on the + rail.

• two caps, one from + to bias junction, one from bias junction to -

You see that in circuits but to me there is no technical rational behind it.  (In rare cases you will see deliberate injection of supply ripple to cancel out the hum.) 
It actually degrades the ripple rejection a lot.
If you have a single supply rail the single cap on the bias ckt is the best solution.

• one cap across one of the resistors (which one?)

Never bad.
Always the one going to ground (except in odd cases where the ground reference is really +V)

• two caps, one from + to - supply, one across one of the resistors (which one?)

Same as previous comment.  The cap across the + and - provides additional benefits, as in mentioned on first case.
The need for a bias cap and a supply cap are treated separately.

If you filter the bias resistors, do you have to filter the virtual ground? Are these needed?

On the virtual ground:
• two caps, one from + to virtual ground (after the op amp), one from virtual ground to -
• one cap across one of the above (which one?)

Or is it best to do both? (which way?)

Pretty much same as previous comments.  One cap from Vref to ground.  Two caps has no technical rationale behind it (unless you can see a specific reason in a specific case).

[Eb7+9]

cool idea Jonny!
still, I'd stick 100r's on the emitters to nail in that R/4R ratio ...

[tubegeek]

Thanks for the detailed answer, Rob. I see that I didn't specifically ask about the split-supply case - did I waste a cap on this then? Seems like one would want both in this circuit but you've made me wonder.

https://www.diystompboxes.com/smfforum/index.php?topic=123650.0

[jonny.reckless]

I would imagine crossover distortion in that output stage would make it sound quite harsh. Even with a fast op-amp in the feedback loop like an NE5534, you will have some pretty nasty high order distortion remaining since the op-amp is neither infinitely fast nor infinitely high gain. It's especially noticeable at low volumes when your ear is most sensitive.

[Rob Strand]

Thanks for the detailed answer, Rob. I see that I didn't specifically ask about the split-supply case - did I waste a cap on this then? Seems like one would want both in this circuit but you've made me wonder.

https://www.diystompboxes.com/smfforum/index.php?topic=123650.0

For a *supply splitter* ideally you would not use bypass caps on the divider since even if ripple is present on the rails you want to track the middle of the ripple.     However a similar circuit is where you have buffered Vref circuit using an opamp.  In this case you would follow the rule of bypassing between the Vref divider and ground.

For both cases, any additional caps on the opamp output are more to do with reducing noise and ensuring the buffered Vref is low impedance at high frequencies (opamps aren't low impedance at high freqs).    Thus brings up can of worms since the buffer adds noise to Vref and the added caps can affect the stability of the opamps.   In general I don't like buffered Vrefs *unless* you need the current on Vref and even then you could probably just use the opamp to feed the high current part of the circuit.  The reason I don't like them is they only add noise and don't add any extra "goodness" .

[tubegeek]

OK that all makes good sense. Thanks for the detailed answer, much appreciated.