High voltage digital volume control

[composition4]

Hi all,

Trying to come up with novel solutions for digital volume control of high voltage signal (tube amp voltages, up to 100V P-P)

Avoiding LDRs/Vactrols/etc (too much variance between parts for my application).  Avoiding attenuating signals, using a digital pot, then re-amplifying.

I've had the vague idea of using PWM to control the volume.  I would have a FET connected to shunt the signal to ground and use this FET in an mute/unmute manner. If my PWM signal was well out of the audio range (lets say 8MHz) and I applied a 50% duty cycle to it, would switching the FET on and off at this rate effectively halve the voltage of my signal?

Terrible schematic shown below, sorry but I'm at work and only have MS Paint.

Is there any merit to this idea or am I not thinking of some problems? I'm thinking the capacitance of the FET won't have any negative effect as I'm just using it to shunt signal to ground (albeit at quite a fast rate).



Thanks
Jonathan

[TejfolvonDanone]

With this topology you essentialy modulation the input signal with a square wave. There's no attenuation in that. Somehow you have to reconstruct the input signal from that.

[PRR]

Ultimately the chopped signal will reconstruct. The proposed 8MHz chop rate is far above what any guitar amp will pass.

But the stage just after the chopper will probably freek-out with 1% 100V pulses, and not act like it would with smoothly attenuated signal.

The 100V probably exceeds what any available FET will block. JFETs over 40V are almost obsolete. MOSFETs generally have body diodes which will conduct at 0.6V on the back swing.

No reasonable tube stage has 1K output impedance +and+ 100V swing. That would imply 100mA current! Yes, the Bassman 5F6a's tone driver can swing 100V, and has a SMALL-signal impedance near 1K. But only as long as current is <1mA. Much over that, and you face the poor conduction of an empty bottle, tens of Kohms.

And shunt choppers "should" have hi-Z sources, for reasons like this. This also suggests post-buffers to drive the next step in the chain. Also fairly heavy low-pass filters so those next stages don't freak-out with supersonic slams. (Series-shunt choppers dodge some of these faults by adding complicated drive.)

8MHz seems rather extreme. You need 2X the signal frequency; with precise filters you could chop at 20KHz and get guitar-range output. 48KHz samplers are all around us. At 8MHz, to get 40dB attenuation you need -narrow- spikes equivalent to 800MHz. You may find it difficult to control those, precisely, with any common parts.

An observation. "All" the slick active multipliers/dividers I can think of favor "small" input levels. A Volt, even 20mV, not 100V. Level control is basic audio design. Bring your level down, manipulate it, then (if essential) bring it back up large.

[ElectricDruid]

I agree with what Paul/PRR said. Or at least I agree with all the bits I understood. Some of the tube stuff is over my head.

One further point though - PWM doesn't generally offer great dynamic range. There was a discussion about this on the SynthDIY list the other day. For example, if you can vary your pulse width down as far as 1%, that's only -40dB. You need to get to 0.01% to get -80dB (which would be a reasonable expectation of a decent VCA). Even at much less than 8MHz, there's no way on earth your rise time will be short enough for you to get a pulse that short.

HTH,
Tom

[Transmogrifox]

Practically speaking you can't drive high voltage FETs at 8MHz even at 50% duty cycle.  Just to give you an idea how much power you burn charging and discharging the gate this frequently, we assume gate charge of, say 100 nC at 8V drive,
P = Q*V*Fsw
Q = 100 nC (gate charge)
V = 8 V Drive voltage
Fsw = 8MHz (switching frequency)

P = 6.4 watts in the FET just to turn the gate on and off.

A more reasonable frequency would be 100 kHz, which only requires 80 mW gate drive.  I would shy away from 20 kHz because you need some really extreme reconstruction filtering, and you probably won't get usable bandwidth above 7 kHz (which may be ok for guitar).  100 kHz gives enough margin that you can filter away the switching noise without such an extreme filter, and do it with simple RC ladder.

Most class D audio amps operate between 500 kHz to 1 MHz.  Not much above 1 MHz because of the practical limitations as pointed out above.  I have a class D amp board that operates at 800 kHz and the heat sink gets pretty warm even when there is no audio.  In fact, driving the speakers with full-on 50W power doesn't really seem to raise the temperature an appreciable amount, so at 800 kHz the FET heating is likely dominated by power to charge and discharge the gate.

Here is the basic idea using high voltage power MOSFETs.  As Paul pointed out, the MOSFET body diode causes problems, but this isn't insurmountable. 

Also as Tom mentioned dynamic range with PWM is not so great.  Even at something close to 100 kHz it's hard to get nice square pulses that narrow.  However if you are fine with larger-grained steps, like 100%, 75%, 50%, 25%, Mute, then you can do it fine with PWM.  You will be able to move pretty smoothly between 10% to 90%, but it's hard to have much control over that last little bit.  Your PWM ramp generator can be designed to put spikes on the start and end that emulate something like a tan() function.  This would prevent the "click" you get when you run off the end, but it also is likely to result in audible distortion since the MOSFETs won't be switching fully on or off during those short pulses.

Here's a way to work around it, but notice the whole circuit requires 4 power MOSFETs and (a) gate drive isolation transformer(s) (either one with 2 output windings, or 2 transformers).  The simulation runs at 160 kHz.  You could lower the frequency and increase the aggressiveness of the filter, for example employ some cascaded Sallen-key active filters for  more of a maximally-flat (butterworth) response.  Probably 6th or 8th order filter would be best to get noise down.

Also notice this could be used with an inductor and capacitor to drive speakers directly with a power amp output.  In that regard it's very similar to a Class D audio amp. The speaker will likely reject most of the high frequency stuff so you don't need to filter quite as aggressively (but if you didn't filter at all you would end up generating a lot of heat in your speaker coils).

This is actually a nearly identical topology to a synchronous Buck Converter, except that it is bipolar.  If your signal is 0V to 100V and never goes lower than ground then the whole thing gets much simpler because you don't need to block the negative direction.


Here are the simulation output traces demonstrating proof-of-concept with 50% duty cycle, showing 100V input and 50V output on the filter:

[merlinb]

Are you only interested in smooth volume control, or would a stepped attenuator do?

[TejfolvonDanone]

I have to point out that it's not really PWM. If you subsitute the MOSFET with an ideal voltage controlled switch the output would be equal the input signal if the switch is off and 0 is the switch is on. So it's really just an AM modulation with a square wave carrier.
In 'real' PWM: it's amplitude is CONSTANT only it's DUTY CYCLE varies corresponding to the input signal amplitude.
So this is more like a chopper amplifier than a class D.

[ElectricDruid]

Trying to come up with novel solutions for digital volume control of high voltage signal (tube amp voltages, up to 100V P-P)

Avoiding LDRs/Vactrols/etc (too much variance between parts for my application).  Avoiding attenuating signals, using a digital pot, then re-amplifying.

How about trying a resistor divider where you can switch resistors in or out? Something like this:



It should be possible to build such a thing that can stand 100V. Alternatively, you could parallel the resistors up into a high-voltage tube mixer.

[merlinb]

Since the grid of a valve can only accept a few volts peak before clipping, why do you even need to control 100Vpp?

[Transmogrifox]

How about trying a resistor divider where you can switch resistors in or out? Something like this:

Each one of those switches represents a pair of head-toe power MOSFETS (at least as far as high voltage switches I am aware of) at maybe $2 to $3 each.  This gets expensive.  The cost is compounded by the need for an isolated gate drive circuit.  Most FETs can't handle more than +/- 30V gate-source, so you need the gate drive to float relative to the source.

For perceptually smooth control you need at least 12 bits so the cost (and complexity) climbs rapidly.

The other challenge I have encountered with bit-stepping attenuators is the question of how fast you want it to be able to slew.  For example for 8 bits, you need to be able to switch the LSB at 256 Hz just to ramp from 0V to 100V in 1 second...and now you have to worry more about clock feed-through and you can't filter a 256 Hz pulse train out of an audio signal.  For those don't follow how this gets into the audio path, the switch edges feed through parasitic capacitance on the FET gate to channel...so then you need to slow down the switching speeds...but then that causes some distortion if you make it too slow because it passes through a region where it will not be linear.  This will be most audible when switching the MSB's on and off.

Can you implement an R2R ladder effectively?  Probably, but this is some food-for-thought to shoot down any delusions that it will be easy.

I liked this suggestion by Tom, seems less troublesome:

Alternatively, you could parallel the resistors up into a high-voltage tube mixer.

And this goes along well with the comment from merlinb

Since the grid of a valve can only accept a few volts peak before clipping, why do you even need to control 100Vpp?

The only reasons I can think is you just want to. 

Avoiding attenuating signals, using a digital pot, then re-amplifying.

Many tube amps implement LED/LDR tremolos and the signals on the LDR usually come after a tonestack, so it's attenuated down to several volts here and followed by the tonestack's gain-recovery stage.  Usually the volume controls don't even operate up to 100V.  Again, it's usually a lower level signal followed by a gain recovery stage.  So you almost always already have the attenuated signal and post-amplification built into the tube amp.  Just need to put your digital control in the right place and then it's not quite as difficult to make something that operates at say, 15V max instead of 100V.