Active removal of noise in DC power

[mike7877]

This could be for anything, but here I'm thinking: most useful for things which deal with analogue audio signals.
Power output transistors, the gain stages before them, tone controls, wherever there are op-amps. Oh, the analogue supply to DAC chips...

The idea: isolate the AC noise on the rail in question (with a capacitor..), use the noise as the control signal of a high-frequency amplifier as, essentially, negative feedback. The amplifier is then connected in reverse phase to the rail so that it counters any AC noise on it with the opposite energy- negating it.

Before this, try to do the best job possible with passive components first (within reason) focusing especially on attenuating the highest frequencies and reducing overall amplitude. Why? So the 180 degree amplifier (TM) {lol...} doesn't have to do this actively (which would increase its cost)

My oscilloscope isn't of the highest frequency variety... When I bought it I wasn't thinking of using it to do work on power supplies. Fortunately, I like to buy nicer things than I need, so I did still get a 200MHz scope. But unfortunately... What I found when I eventually used it to observe some of my many SMPSes, is the bulk of a lot of the noise from the transistor switching is up at 50+ MHz. My excessive scope, was barely excessive enough to register the dang things! Due to this (and my less than ideal leads compounding the problem, probably), I don't have an accurate estimation of the amplitude of these things... Do we only need 50mA to void this crap? If so, this thing could only need to add $20 to the manufacturing of something that includes it.
But if you need a few watts? Now, THAT's going to be a few hundred!
(for 10Ghz BW, which, off the top of my head, should be fast enough for the bulk of these 50-200MHz peaks, but I could be wrong, I'm not an injuneere!)

I assume that the noise is somehow proportional to the current being drawn as the switching transistors switch ( Low power circuit? Low power noise. )
And whatever isn't thrown into the environment by wire induction transmitting it - won't need to be attenuated... But we're not talking long runs of anything inside of amplifiers, probably only traces mere inches in length


How do you feel about this idea? If my first draft implementation is crappy, imagine it's not - what do you think of just the idea itself? Useful?

Question:
Say the +50VDC rail in a class ab amplifier is from a regulated SMPS using standard transistors, and it's capable of up to 5.1A.
The amplifier is outputting 18kHz at 15W. The load on the SMPS looks almost constant due to the 20,000uf audio grade electrolytics buffering the output transistor.
The current at the SMPS output measures 2.0A, voltage ripple is 7mV so we'll pretend it doesn't exist in this situation.
We connect a 4GHz scope on the SMPS output and the AC we see this:

Very simple, the width of the spike is 1ns, the height is 2.3V

Of course we know that we don't need neg 2.3V @ 5A to null this momentary spike, even though it appears to be 52.3V with 5A being drawn. The impedance of the circuit isn't the same at 250MHz, especially just a quarter cycle!

How would we estimate the current required? Is this something that is just... known? Meaning is it known by those who make power supplies how much energy is actually contained within this noise? Like is there a rule, a percentage, something like "if switching at 20% of its maximum rating in the upper kHz-lower MHz range, you take the transistors's maximum rating, divide it by 500 and multiply by 0.2, then you have your answer in mA." something like that? lol


Like I said above, a solution like this doesn't have to cost much if the power required is only in the milliwatt range. Once in the watt range, it looks to me like it'd be getting expensive - that's the impression I got from looking at ready-made solutions on Mouser anyway - for example, a 20MHz to 10GHz amplifier capable of 3 watts costs about $200. If there are cheaper options (discrete gain section after low power, maybe?) that'd be great, but what'd be best is if only milliwatts are needed. And that this is a good idea in the first place lmao!

What do you think?

 

[AnalogSteph]

The idea: isolate the AC noise on the rail in question (with a capacitor..), use the noise as the control signal of a high-frequency amplifier as, essentially, negative feedback. The amplifier is then connected in reverse phase to the rail so that it counters any AC noise on it with the opposite energy- negating it.

So basically, like this?

(The article is OLD, it mentions a CA3094 which I think has been discontinued for two decades. So my guess is mid-'80s to late '90s.)

 

[SSS]

So basically, like this?

(The article is OLD, it mentions a CA3094 which I think has been discontinued for two decades. So my guess is mid-'80s to late '90s.)

Yes, this can work for audio purpose. Question is whether there is really much noise past a series regulator IC in order to countermeasure? On switched regulators the spikes may be higher but did not do any measurements. If the audio output is clean then I guess the device is properly designed and there is no reason to modify the power supply.

 

[antcollinet]

Acutally a feed forward rather than a feedback. Also no benefit in principle to any active linear or switch mode regulator which generates an error signal by comparing the output voltage with a stable reference, and feeds back the error to the amplifier.

EG (for a very simple example)

Also - why do you think you need to be trying to actively control radio frequency noise (20MHz to 10GHz????) in an audio amplifier? Passive filtering is more than adequate to negate these frequencies to levels which are no issue for audio circuits. Remember - you cannot hear them.

Further - your 1ns spike - where is that coming from. Think about how much current is needed in 1ns to make a significant voltage increase in your 20,000uF capacitance. What is the (noise) energy source to create that current? That 20,000uF combined with the impedance to whatever your noise source is, is already a very effective filter.

 

[Cbdb2]

Your device shunts noise to ground, if it thinks your charging current voltage spikes are noise, likely, then it will shunt that current to ground, reducing the charging, which reduces the voltage, which will cause the regulater to increase the charging pulses etc. and something burns up.

 

[solderdude]

What do you think?

I think you don't need to worry about the MHz crap you see on scope screens.
Just connect the ground and probe together on a component of the DUT and you'll probably will still see those spikes.
That is just common mode crap from elsewhere and does not exist IRL.
Any 10nF decoupling cap on a power rail will effectively remove those frequencies.

Depending on the PSRR of the design small variations on the power supply line are lowered by the design itself.
When your design has a crappy PSRR you should redesign it when it becomes an audible problem.
One should also realize that when outputting a lot of power the added distortion is already in the measurement so if you change something in the power supply and distortion numbers do not change there is no reason to resort to heroic measures.
Thinking that one cannot measure it but can clearly hear it is well...

There are plenty of well designed regulators and parallel regulators to be found.

That's what I think and have experienced over the tens of years designing, measuring and testing electronics.

 

[DVDdoug]

This is confusing... All DC power supplies have filtering. Low-pass filtering (DC is zero Hz) is more effective on higher frequencies.

Voltage regulators (which most audio electronics have as part of their power supply) tend to be very effective filters since their whole purpose is to hold the DC voltage constant. If there is noise coming-out of the power supply and ineffectively filtered-out by the capacitors, adding a regulator will usually kill it.

A long time ago when phono preamps were common, I saw two approaches for additional filtering for the preamp section. One used a resistor and capacitor as a regular-old low-pass RC filter. Since the preamp doesn't consume much current, a resistor can be placed in series with only a small voltage loss. The other approach was double regulation with +/-15V feeding most of the circuit and another pair of +/-12V regulators for the phono preamp.

 

[solderdude]

Regulators don't like varying current draw and have a reaction time.
Fortunately for digital and pre-amp circuits there is mostly a constant current draw so there is no issue.
A capacitor multiplier also can greatly reduce ripples. (the circuit mentioned above by DVDdoug)
It could be for power amps.
HF is taken care of by decoupling caps which can be done with compound constructions.

As said... the PSRR of the circuit is often forgotten to be there which can be ranging from -80dB to -120dB for lower frequencies.
10mV (at 100Hz) ripple thus is merely 10nV at the output of such an opamp.

 

[AnalogSteph]

As said... the PSRR of the circuit is often forgotten to be there which can be ranging from -80dB to -120dB for lower frequencies.
10mV (at 100Hz) ripple thus is merely 10nV at the output of such an opamp.

Conversely, if you ever see the noise floor going up in the treble (like in my old Behringer mixer), odds are the supplies are a tad noisy...
(This can also be caused by direct rail noise injection into the ground via decoupling caps though. A real peril of shared grounds when rails are low impedance. Seasoned designers would have a habit of RC filtering their rails for that exact reason.)

 

[audio_tony]

A long time ago when phono preamps were common, I saw two approaches for additional filtering for the preamp section. One used a resistor and capacitor as a regular-old low-pass RC filter. Since the preamp doesn't consume much current, a resistor can be placed in series with only a small voltage loss. The other approach was double regulation with +/-15V feeding most of the circuit and another pair of +/-12V regulators for the phono preamp.

The capacitance multiplier was also a useful circuit for filtering in low current circuits.

This was a very popular method to provide smoother supplies before we had really cheap three terminal regulators.

 

[AnalogSteph]

The cap multiplier has some unique advantages though, like being able to clean up variable input voltages with fairly low dropout and being very low-noise. Arguably you'd want something more complex than the single transistor variety these days though (maybe a Sziklai, a Darlington or a MOSFET), and if you mess up the choice of resistor you can end up with too little dropout for it to be effective. BTW, the RC can also be a higher-order filter.

 

[HaveMeterWillTravel]

Acutally a feed forward rather than a feedback. Also no benefit in principle to any active linear or switch mode regulator which generates an error signal by comparing the output voltage with a stable reference, and feeds back the error to the amplifier.

EG (for a very simple example)



View attachment 399926

Also - why do you think you need to be trying to actively control radio frequency noise (20MHz to 10GHz????) in an audio amplifier? Passive filtering is more than adequate to negate these frequencies to levels which are no issue for audio circuits. Remember - you cannot hear them.

Further - your 1ns spike - where is that coming from. Think about how much current is needed in 1ns to make a significant voltage increase in your 20,000uF capacitance. What is the (noise) energy source to create that current? That 20,000uF combined with the impedance to whatever your noise source is, is already a very effective filter.

The frequency content of EMI often has very little to do with the frequency of the manifestation.
I wouldn’t doubt that some of the more “esoteric” designs have zero knowledge of RFI rectification in OpAmps and other common semiconductor devices. This will even affect an active filter stages similar as the transfer function is not much of a factor.

OTOH: 20mF of capacitance effective at 10 GHz would be eye watering complex and expensive! Capacitors suitable for 10 GHz signal paths are typically only available up to 100s of pF. Perhaps 1 or 2 uF.

HMWT

This is confusing... All DC power supplies have filtering. Low-pass filtering (DC is zero Hz) is more effective on higher frequencies.

Voltage regulators (which most audio electronics have as part of their power supply) tend to be very effective filters since their whole purpose is to hold the DC voltage constant. If there is noise coming-out of the power supply and ineffectively filtered-out by the capacitors, adding a regulator will usually kill it.

A long time ago when phono preamps were common, I saw two approaches for additional filtering for the preamp section. One used a resistor and capacitor as a regular-old low-pass RC filter. Since the preamp doesn't consume much current, a resistor can be placed in series with only a small voltage loss. The other approach was double regulation with +/-15V feeding most of the circuit and another pair of +/-12V regulators for the phono preamp.

It's effective at higher frequencies and then it isn't. The impedance of a capacitor typically used between (+) and (-) will decrease to some value at some frequency. Then factors such as it's parasitic inline inductance starts to dominate (in a negative way). One reason why electrolytic capacitors are not as effective for say, bypass caps for an op-amp, their minimum impedance is at a significantly lower frequency. Just the case size and lead-wire lengths contribute.

 

[Gary_G]

Is this what Benchmark does?

 

[antcollinet]

The frequency content of EMI often has very little to do with the frequency of the manifestation.
I wouldn’t doubt that some of the more “esoteric” designs have zero knowledge of RFI rectification in OpAmps and other common semiconductor devices. This will even affect an active filter stages similar as the transfer function is not much of a factor.

OTOH: 20mF of capacitance effective at 10 GHz would be eye watering complex and expensive! Capacitors suitable for 10 GHz signal paths are typically only available up to 100s of pF. Perhaps 1 or 2 uF.

HMWT


It's effective at higher frequencies and then it isn't. The impedance of a capacitor typically used between (+) and (-) will decrease to some value at some frequency. Then factors such as it's parasitic inline inductance starts to dominate (in a negative way). One reason why electrolytic capacitors are not as effective for say, bypass caps for an op-amp, their minimum impedance is at a significantly lower frequency. Just the case size and lead-wire lengths contribute.

I'm not following the point you are trying to make.

Are you arguing in favour of the OP's plan to somehow cancel out RF PSU noise using active negative feedback of the noise in the PSU of an audio device?

 

[mike7877]

I'm not following the point you are trying to make.

Are you arguing in favour of the OP's plan to somehow cancel out RF PSU noise using active negative feedback of the noise in the PSU of an audio device?

OP here - the first words in OP were "this could be for anything" (though I do recognize this is an audio forum, posted in the general audio subforum)

So while it might not be a practical solution for any common problem with audio devices, "anything" is very wide open.


The reason I thought this might be viable is with the new(ish) high speed GaN transistors being able to generate the inverse of the noise efficiently.
It may not necessarily be more effective, but it might be a cheaper alternative (if not now, eventually - the circuit shouldn't be too complicated)

 

[egellings]

Acutally a feed forward rather than a feedback. Also no benefit in principle to any active linear or switch mode regulator which generates an error signal by comparing the output voltage with a stable reference, and feeds back the error to the amplifier.

EG (for a very simple example)



View attachment 399926

Also - why do you think you need to be trying to actively control radio frequency noise (20MHz to 10GHz????) in an audio amplifier? Passive filtering is more than adequate to negate these frequencies to levels which are no issue for audio circuits. Remember - you cannot hear them.

Further - your 1ns spike - where is that coming from. Think about how much current is needed in 1ns to make a significant voltage increase in your 20,000uF capacitance. What is the (noise) energy source to create that current? That 20,000uF combined with the impedance to whatever your noise source is, is already a very effective filter.

It's just a unity gain buffered Zener diode.

 

[mike7877]

As said... the PSRR of the circuit is often forgotten to be there which can be ranging from -80dB to -120dB for lower frequencies.
10mV (at 100Hz) ripple thus is merely 10nV at the output of such an opamp.

Except for top-end, specifically low-noise op-amps, that's effectively in the grass, so great.

But imagine a simple active circuit before the op-amp which reduced that 10mV by 40dB!
(10mV is a really small amount of ripple, say it was 30-40mv, that'd be detectable on the output, and then: no more!)
Though originally I was thinking more for extremely high frequencies.

 

[solderdude]

The seen 'noise' is clearly measurement error (common mode switching noise being picked up and appearing as if it is differential across the probe) so there is nothing that needs to be corrected in the first place.
I have seen this way too much on scope screens on outputs of circuits only to find out that it really does not matter where you connect the probe to see it is everywhere not just on the output but in equal 'amount' on other parts of the circuit too. It would be a hard sell to state a circuit has 0dB PSSR.

Of course it is easy to build a circuit with low PSSR and get those pulses into the audio path. The point is the measured peaks would have to be there to begin with.

 

[mike7877]

The seen 'noise' is clearly measurement error (common mode switching noise being picked up and appearing as if it is differential across the probe) so there is nothing that needs to be corrected in the first place.
I have seen this way too much on scope screens on outputs of circuits only to find out that it really does not matter where you connect the probe to see it is everywhere not just on the output but in equal 'amount' on other parts of the circuit too. It would be a hard sell to state a circuit has 0dB PSSR.

Of course it is easy to build a circuit with low PSSR and get those pulses into the audio path. The point is the measured peaks would have to be there to begin with.

I've heard there are tricks to see how much power is in the pulse on the circuit - I forget the details exactly - add a resistor somewhere and measure it/across it.
I know that the current behind these measured peaks is often, essentially, nothing, but I think the closer to the output transistor making the noise, the higher it is. What would be best would be to (well, first, find out the current behind it) absorb it as close to there as possible so that inductance doesn't transmit it into the air to be absorbed somewhere else past anywhere that can be filtered, or into a chip causing problems.

This is more just hypothetical, see how viable it is to create the opposite of noise actively, up in the MHz to GHz frequencies

 

[solderdude]

First thing to do is rule out measurement error.
When it is there is no reason to look for a solution.