A Heretical Unity Gain Line Stage part III

[SY]

Brian, I think that combining the cathode stopper and the bias is a very clever idea! Mind if I steal it?

[Brian Beck]

Mind? Heck no, that's what this forum is all about.

[Brian Beck]

SY,

While we’re on this roll, I hope you won’t mind a couple of comments on your B+ modulation circuit. I know that your posted schematic is only a caricature of the final design, but it seems to me that there are a couple of “opportunities for improvement”.

First, I think ripple rejection and raw supply modulation rejection is compromised by the error amp, through the collector resistor. Ripple will be coupled from the collector of the series pass transistor to its base, with not a lot of shunting action from the high dynamic resistance of the lower transistor’s collector resistance. This might have been an issue with your original non-modulated PS design too, but there you had immediate feedback to clean things up. With the servo design, feedback is too slow to remove ripple. There are many ways to overcome this, including providing some RC filtering prior to the collector resistor of the error amp, or by using another PS design altogether.

The second comment is that I think it’s wise to make the transfer function of the servo loop as linear as possible to avoid distortion in the bass. The servo will increasingly try to track (null) the bass signals as you go lower and lower in frequency. You are driving the base of the error amp transistor directly by the opamp, with the LED providing essentially a low-Z voltage source at the emitter. Gain will be high, but the transconductance transfer function will be highly non-linear – exponential in fact. As the servo follows low frequency signals, the exponential transfer function will generate a harmonic series with plenty of higher harmonics. It is arguable whether this is a practical/audible problem, but there is no reason not to fix it now. BTW, with the current source modulation approach, the emitter resistor of a few hundred ohms provides linearizing local feedback which helps tremendously to linearize the servo transfer function. Something like that could be applied in the B+ modulation approach.

[SY]

Brian, sorry to be so slow to respond, but I wanted to have a chance to chew over your excellent suggestions.

Your point about ripple is well-taken; experimentally, the ripple was under 0.1mV at the regulator output. As a belt-and-suspenders approach, I suppose that the collector resistor could be swapped out for a current source. When I get home, I may also try moving the resistor to the output side of the regulator and see if it remains stable.

The LED in the emitter circuit accomplishes two things. First, as usual, it gives a visual indication that the circuit is operating. Secondly, it lifts the base a bit farther away from ground. Some discussions with John Curl were helpful in this regard- he feels that getting away from the zero-crossing part of the op-amp's output stage is beneficial (analogous to the pull-up resistor mods one sees on occasion). One could add some resistance in series with the LED to linearize the op-amp's load, and that might indeed be beneficial. The same thing could be accomplished by adding a series resistor between the op-amp and the transistor base. The LED has a dynamic impedance of something like 50 ohms at 1mA; multiplied by a typical hfe and we're in the 5-10K range. Even with that load nonlinear, it's high enough that the op-amp shouldn't be breathing too hard.

While crammed into a 737 today, I sketched out a few ways to modulate the current source instead of the B+. Unfortunately, the best ways I could think of make using a cascoded CCS something of a problem unless I want to add a battery. And I hate batteries.

[Brian Beck]

SY,

I too have spent many an hour in a cramped airplane seat scratching out circuits on a piece of paper, or doing simulations on my laptop, so I can empathize. More than once, out of the corner of my eye, I’ve seen the person sitting next to me gazing at my screen, probably wondering just what kind of video game I was playing.

When I get home, I may also try moving the resistor to the output side of the regulator and see if it remains stable.

Here’s a trick: if there are start-up problems with putting the resistor after the regulator, I’ve used a dual diode switch that initially draws power from the unregulated side. Then as the regulated voltage comes up, the diodes switch and the power is taken from the regulated side. I'll post a picture later.

You may have other start-up glitches as the tube heats up. So it may be useful to bracket the range of servo control of the B+ so that it starts at some functional value even with the servo pegged before settling.
.

Some discussions with John Curl were helpful in this regard- he feels that getting away from the zero-crossing part of the op-amp's output stage is beneficial (analogous to the pull-up resistor mods one sees on occasion).

Can’t argue with that concern. But no matter what you put in the emitter circuit of the error amp transistor, the base will be at least 0.6 volts above ground, and since it’s always pulling some base current, the opamp is always sourcing current from its own output NPN transistor in class A, albeit weakly. You can add a resistor from the base to -12V to pull a bit more current out of the NPN if you wanted to, but I don’t think it’s necessary.

The LED has a dynamic impedance of something like 50 ohms at 1mA; multiplied by a typical hfe and we're in the 5-10K range. Even with that load nonlinear, it's high enough that the op-amp shouldn't be breathing too hard.

I wasn’t worried at all about the opamp driving the base. I was worried that even if the opamp puts out a perfectly zero distortion voltage waveform, the base-emitter diode will convert that linear voltage into an exponentially-defined base current that gets multiplied by hfe and ends up at the pass transistor’s base as a distorted voltage. A resistor in the emitter circuit that is much larger than the dynamic emitter resistance (and in series with an LED if you wish), will swamp out much of the exponential base-emitter diode effect and greatly linearize the error amp. Gain will drop, but with an “infinite gain” opamp, all you have to do is to readjust the cap and feedback resistors to restore the loop parameters.

Unfortunately, the best ways I could think of make using a cascoded CCS something of a problem unless I want to add a battery. And I hate batteries.

Agreed on the batteries. I guess I don’t see why or where you’d need a battery, but maybe I’m missing something. If I get time I will sketch out a notional CCS modulated servo based on your original CCS, but it’s late now and I’m going to bed…

[SY]

On my last trip, I did get into an interesting conversation with an engineer in the next seat who was quite incredulous upon seeing me reading Reich.

The idea of the LED was to get that base more than a 0.6V drop away from ground. The LED gets the voltage solidly positive, around 2.5V. That way, the upper half of the emitter follower in the op-amp output stage is carrying the load, so to speak.

Let's see who has the time to sketch out the modulated CS quicker... I'm handicapped, being 3000 miles from my lab at the moment, so I think this is a contest you'll win!

[Brian Beck]

Let's see who has the time to sketch out the modulated CS quicker... I'm handicapped, being 3000 miles from my lab at the moment, so I think this is a contest you'll win!

SY, no contest, just teamwork! This is your excellent Heretical Unity Gain Line Stage Design thread after all. I don't want to steal your thunder, so I'm using your complete original cascode CCS configuration and just adding the servo feedback to the lower base.

CAUTION: I quickly whipped up this circuit. I did do a quick simulation, but I haven't built it or listened to it. The sim looks good. But there may be errors!

The resistors and C1 that I selected place the servo pole (-3dB roll-off) at about 1.5Hz. Adding optional C2 introduces a filter pole at 160Hz. This is two orders of magnitude above the servo pole and the resultant bump in the response is negligible - about 0.06dB at 45Hz. C1 and C2, if used, should be decent film audio grade caps.

I didn't search for the best opamp part. The LF411 was in my sim library already. I'd suggest looking for a single-package dual low-offset FET-input opamp. Having the additional inversion obviates the need for the single-stage non-inverting integrator that would require two large 1uF caps. And it gives you the option to add the second pole.

The +/-12 volt supplies should be made pretty quiet (as SY designed them I think they are) so as not to introduce noise, especially via resistors R3, R4, and R11.

OK, now let the tweaking begin…

[gingertube]

About to start on a new preamp/line stage so will give this one a go.

A small hint from the "old farts" school of op-amps.

In any circuit where you have a cap connected O/P to -ve input (i.e. as an integrator) a small value resistor, say 10R to 47R, in series with the cap on the OP Amp O/P side will help both for stability and linearity.

I agree with Brian about the LED current, 9.5mA is way more than required. Cheap RED LEDs are quite happy at 1mA.
UNLESS:
SY
Do you have any opinion/evidence to suggest better voltage stabilty or noise performance from the LED at higher current?

Cheers,
Ian

[wrenchone]

I would expect lower incremental impedance using 10ma vs. 1ma bias current. I don't know about noise, as the conduction mechanism is different for a forward biased LED than for a zener. I do know from sad experience that starved zeners can be really noisy. I always use 10-20ma for LED references, and get about 1.6V drop for the old GaAsP deep red LEDs, and almost exactly 2V for the yellow-green GaP devices.

[SY]

Originally posted by gingertube

I agree with Brian about the LED current, 9.5mA is way more than required. Cheap RED LEDs are quite happy at 1mA.
UNLESS:
SY
Do you have any opinion/evidence to suggest better voltage stabilty or noise performance from the LED at higher current?

Cheers,
Ian

Morgan Jones did some measurements of impedance which showed a strong dependence on current. It flattened out at 5-10mA. I'll see if I can find his graphs somewhere. This was quite consistent with curves shown by Bob Pease in his book on the Vf versus I of a variety of diode types.