LM4562 decoupling...

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Another interesting post. Thanks :)

As I understand it decoupling is basically the process of isolating one stage of an amplifier or circuit from another.

Yes, but decoupling also has the role of creating a nice low impedance power source at higher frequencies for the opamp. Its the optimisation of both of these which we try to achieve.

This is traditionally done by a locally placed cap from rail to ground. Its supposed to stop noise entering the next stage right?

I take the view that the purpose of such caps is the lowering of the power supply impedance at HF. If there is enough inductance in the power wires then it will help to keep HF current loops small, which is what we want. So no, its secondary function is preventing noise transfer to the next (or more importantly, the previous) stage.

How does a cap from +VE to -VE have the same effect in bi-polar situations?

It keeps the HF current loop area small - if the opamp isn't putting out appreciable current to ground, then its all we need. In the case where the opamp does put appreciable current to ground, then it turns the noise on the power supply into common mode noise, affecting both rails equally. Here it might be helpful to refer to an opamp datasheet graphs of PSRR and CMRR. They're closely related but normally the CMRR is better than the PSRR. The PSRR is measured for each supply separately, but the CMRR shows what happens when each supply wiggles up and down the same.

Also, if your LTspice sims prove that the traditional method dumps noise on the GND (which its supposed to do no?), then where does the noise go if the cap is between +/- rails?

Actually, it just shows that the ground gets as noisy as the power rails at high frequencies because they're both in effect wires with about the same series inductance. The noise goes to the other rail, rather than to ground in my preferred decoupling arrangement. I don't think the intention of people putting two caps around their opamp is to end up with a noisy ground - its an unintended consequence because they didn't quite understand enough circuit theory.

We do need to distinguish between sources of noise though - mainly it comes in via the (toroidal) mains transformer and 'leaks' through the regs as they have insufficient HF rejection. Some is generated in the rectifier. And some noise comes onto the supply from the opamps themselves, depending on the application.
 
abraxalito, what is your opinion about the finesse "regulator" which supposed to deal with HF garbage?

I first saw this a couple of years ago and conceptually I liked it a lot. In the intervening time I've done more work on power supplies myself and come to appreciate its limitations. Its a typical feedforward scheme, all fine provided we can predict the precise noise level at the output and cancel it. Like all feedforward schemes, it falls over if the matching at the output fails - which in real life is most of the time as few loads present an invariant impedance. If they did then let's face it, they wouldn't need decoupling at all. So I think this cute circuit is useful when dealing with a very quiet circuit - one that generates none of its own noise. Which perhaps is what a crystal oscillator circuit is like.

I don't think it will be particularly good at HF because good cancellation at HF does require amazingly close phase matching at the output, and going through a transistor stage can't avoid adding some phase shift. Within the audio band though I'm sure it does fine.

As regards the Bob Pease stuff, I'll read with interest and get back to you:D
 

iko

Ex-Moderator
Joined 2008
The finesse circuit is not something I would use, but I asked, what the hey :) Regulator is a poor name for it, I think. Active low-pass filter is more like it. It's beyond me why the author would opt for this active filter instead of just a regular LC, LR, or whatever passive low-pass filter.

Regarding the Bob Pease article; haven't tested the circuit, but thought it was interesting when I came across it.
 
The finesse circuit is not something I would use, but I asked, what the hey :) Regulator is a poor name for it, I think. Active low-pass filter is more like it. It's beyond me why the author would opt for this active filter instead of just a regular LC, LR, or whatever passive low-pass filter.

Its not really a low pass filter - its actually a high pass filter for noise as it does a good job of cancelling the audio bandwidth noise, but gets more and more hopeless as the frequency goes up. I believe its used to power Wenzel's crystal oscillator circuits - which doubtless operate in pure class A so present a largely invariant load. Any variation in load impedance and the carefully trimmed gain noise from the transistor stage becomes an enemy - it can put out a LOT more noise than goes in once the balance is lost. So it does need to be employed with extreme caution.:zombie:

Regarding the Bob Pease article; haven't tested the circuit, but thought it was interesting when I came across it.

Did not see the pic you included on that article page. Is that from another article?
 

iko

Ex-Moderator
Joined 2008
That's Figure 2, in the article.

As for it being which kind of filter, have a look at this.

166302d1270772758-my-take-discrete-shunt-voltage-regulator-finesse.png


166303d1270772758-my-take-discrete-shunt-voltage-regulator-finesse-psrr.png


The similarity in frequency response between a regular LC low-pass filter and the finesse "regulator" is quite amazing. Can make them exactly the same this way:

166312d1270779560-my-take-discrete-shunt-voltage-regulator-lowpass-lr.png


166313d1270779560-my-take-discrete-shunt-voltage-regulator-lowpass-lr-vs-finesse.png
 
As for it being which kind of filter, have a look at this.

But as that shows no rejection at all in the audio band, you must have got the feedforward gain wrong somewhere. It should show no rejection at DC [0dB](as the feedforward is AC coupled) but then reject 40dB or so across the audio band (according to how well the gain has been adjusted for the particular output impedance).
 
Yes, but decoupling also has the role of creating a nice low impedance power source at higher frequencies for the opamp. Its the optimisation of both of these which we try to achieve.



I take the view that the purpose of such caps is the lowering of the power supply impedance at HF. If there is enough inductance in the power wires then it will help to keep HF current loops small, which is what we want. So no, its secondary function is preventing noise transfer to the next (or more importantly, the previous) stage.



It keeps the HF current loop area small - if the opamp isn't putting out appreciable current to ground, then its all we need. In the case where the opamp does put appreciable current to ground, then it turns the noise on the power supply into common mode noise, affecting both rails equally. Here it might be helpful to refer to an opamp datasheet graphs of PSRR and CMRR. They're closely related but normally the CMRR is better than the PSRR. The PSRR is measured for each supply separately, but the CMRR shows what happens when each supply wiggles up and down the same.



Actually, it just shows that the ground gets as noisy as the power rails at high frequencies because they're both in effect wires with about the same series inductance. The noise goes to the other rail, rather than to ground in my preferred decoupling arrangement. I don't think the intention of people putting two caps around their opamp is to end up with a noisy ground - its an unintended consequence because they didn't quite understand enough circuit theory.

We do need to distinguish between sources of noise though - mainly it comes in via the (toroidal) mains transformer and 'leaks' through the regs as they have insufficient HF rejection. Some is generated in the rectifier. And some noise comes onto the supply from the opamps themselves, depending on the application.

Thanks :)
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.