Taming Shutdown Transient Spike?

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

barefoot

Well-known member
Joined
Jun 4, 2004
Messages
93
Location
Portland, OR
I have an amp configured like the schematic below. It start's up softly, but shuts down with a big transient spike. The transient is definitely coming from the input stage. If I disconnect the 100n cap between the input and power stages I get no transient in the speaker. Also, if I tie the two inputs together, I get no transient.

The power supplies come from external sources, so I have no direct control over their startup and shutdown speeds. And it seems like the +/- 12V supplies collapse much faster than the +48V supply, since the spike is transmitted through the output stage before it has a chance to shutdown.

softshutdown.jpg


Any ideas on how to soften the shutdown of the input stage?

I was thinking about putting resistors in series with the +/- 12V supplies then using larger bypass caps to increase the time constant. But maybe this would compromise linearity or add a lot of noise?

Thanks!
Thomas
 
> maybe this would compromise linearity or add a lot of noise?

How?

> 50 ohms, 47uFd

That won't do squat.

Use diodes, so the collapsing 12v rails can't drag your rails down. Say your opamp is TL072, 2.4mA drain. It acts sorta like a 24V/2.4mA= 10K resistor. You have not said how fast the rails are coming down... I'll say you need to stay up for 10 seconds to allow for the slow fall of an AB amp working at negligible signal. 10K and 1,000uFd gives a 10 second time constant, meaning the "12V" will be down to 4.4V after 10 seconds, which happens to probably be a workable rail for a TL072 driving a power amp but not being tested for point-oh-oh THD. I said 24V but it is really two 12V rails so it is two 2,200uFd caps.

Otherwise: clamp your output when your rails are low. It is not trivial to stay clamped as your rails fall to zero. A relay is the obvious approach since it can stay ON without power.

Why not run your part on the +48V rail? Dropped to 30V if you like. The "ground" on the "+" leg can be a +15V reference, then the OUT pin will also sit at +15V for most likely input sources.
 
Put the FET part of an H11F1 coupler in series with the 100n capacitor. Drive the LED from the +/-12V rails via an 18V zener and a 680 Ohm resistor. As the rail-rail volts collapse from 24V to about 20V, the coupling to the output stage slowly opens.
 
nice idea boswell!

Edit:

however, I wouldn't add a fet in series if I didn't *have* to, no sense in adding an additional distortion source if possible.

Maybe a similar circuit with the fet pulling the signal down through a resistor without cutting the actual signal? We don't know anything about the amp stage so we can't say if that will cause a DC problem on the amp output.

I think we need a little more info.
 
How odd---my post on this got eaten! Oh well... What I had said was:

Boswell's suggestion is a good one. There will be the opportunity for a bit of distortion but if the input Z of the PA is dominated by the 50k shown it will be small. OTOH the high frequency energy will be coupled through more effectively the higher the input Z is, due to capacitances, so there is a tradeoff.

The other remark is that slowing down rail voltages, though superficially appealing and intuitive, almost never works and in fact often makes things worse. Opamps in particular and circuits in general don't degrade gracefully as their rails get arbitrarily reduced, in most cases. The stability margin may degrade and oscillations ensue, or even worse transients propagate. Slower rail motion can just prolong the agony.

The business of making circuits power up and down without pops and thumps is a subtle art, and the more so for usually being treated as an afterthought towards the end of the design process, when it is much harder to fix.
 
[quote author="bcarso"]
The other remark is that slowing down rail voltages, though superficially appealing and intuitive, almost never works and in fact often makes things worse. Opamps in particular and circuits in general don't degrade gracefully as their rails get arbitrarily reduced, in most cases. The stability margin may degrade and oscillations ensue, or even worse transients propagate. Slower rail motion can just prolong the agony.[/quote]
I tried PRR's idea of using diodes and 1000uF caps. It definitely slowed the shutdown, but the transients are still there. I tried slowing the power stage as well - transients still. I got them so slow that I could roughly watch the rise and fall on a DVM. But there's always a point where things shut off hard ( about +/-6V on the input stage). So I think you're right about this.

[quote author="bcarso"]The business of making circuits power up and down without pops and thumps is a subtle art, and the more so for usually being treated as an afterthought towards the end of the design process, when it is much harder to fix. [/quote]
Any general design approaches that you know of? Or a good resource? Application notes just don't seem to deal with this issue.

Thanks!
 
I didn't really introduce my suggestion properly. From design experience going back several decades, I've found that you can't de-plop an amplifier by monkeying around with the rates of rise or fall of the power rails. You have to be pro-active and put in place a circuit that gives slow release on power-up and fast cut on power down.

Two techniques for this are the series FET such as the H11F1 that I mentioned, or a shunt depletion-mode opto-FET such as a Panasonic AQV414. The shunt device has to short in the absence of power, so the depletion-mode is the obvious choice, and it's off when powered, so no added distortion. However, in your circuit, the series FET gives 0dB loss. A shunt device needs reistance to work with, and this gives you a loss under normal operation, hence my recommending the series FET for your circuit.
 
[quote author="barefoot"]
Any general design approaches that you know of? Or a good resource? Application notes just don't seem to deal with this issue.
[/quote]

Short answer: No. Longer: probably some material here and there about noiseless muting within a signal chain. But the devilish problem of how to design a given amplifier to turn on and off quietly I've never seen treated. Philips Semiconductor, among others, have spent a lot of effort on this and probably have achieved the best results so far for some of their monolithic power amps. Needless to say they consider the details highly proprietary.

The automotive market in particular has pushed increasingly demanding specs on this parameter, with the Japanese automakers in particular wanting products where you can have your ear immediately adjacent to a speaker in a stationary vehicle with the engine off and hear nothing when the power is cycled. Now that is tough.

Consider just one aspect of what you have to do to achieve this: even if your rail voltages are already up and stable, if there is any d.c. offset in the amp output, the transition from off to on has to ramp to this final state slowly enough to reduce the high frequency energy that would be produced by an abrupt change. And this is assuming that you don't have any other issues associated with power cycling, like RF energy propagating and being detected at various nonlinear circuit nodes (one designer I know forswore all bipolar input opamps in favor of FET inputs owing to their markedly lower RF detection sensitivity).

A lot of people just default to a muting relay in the output line to the speaker, but this will have potential low voltage nonlinearity problems, long term degradation from comtamination and outgassing adsorption, maybe problems with inadequate contact current ratings, and also can't actuate fast enough on mute to get rid of RF detection transients. And there's the problem of d.c. offset producing a click/pop.

Another psychological phenomenon gets at work here too: the amazing sensitivity range of human hearing tends to drive ever-increasing levels of intolerance to any perceptible levels of this sort of noise. Since it is fairly difficult (though less so today with transient-recording scopes and fourier analyzer software) to be quantitative, the engineer may reduce an initial problem by say 20dB, an enormous improvement, only to have the bar moved again after the next listening test.
 

Latest posts

Back
Top