Gearslutz

Gearslutz (https://www.gearslutz.com/board/)
-   Geekslutz Forum (https://www.gearslutz.com/board/geekslutz-forum/)
-   -   When is a Feedback Loop Capacitor NOT Good Practice? (https://www.gearslutz.com/board/geekslutz-forum/1294281-when-feedback-loop-capacitor-not-good-practice.html)

Brian M. Boykin 12th January 2020 02:27 AM

When is a Feedback Loop Capacitor NOT Good Practice?
 
So in the world of swapping modern op amps into older designs I always add a .1uF cap from the +vcc and -vcc to ground and in recent years started adding a 33pf cap across the feedback resistor to further help with stabilization.

Questions:
1) Is there an instance when I should not add a feedback cap?
2) Knowing this cap can limit bandwidth, what kind of trouble can I get myself into from lowering a 100pf cap to 47pf or 33pf in an existing design to open up the bandwidth a little more?

As always,

Thanks in advance

Brian

johnnyc 12th January 2020 08:48 AM

1) you don't want a cap if the opamp is current feedback topology
2) it depends. You want extra bandwidth but not too much 5-10x is a good rule of thumb, meaning 100-200kHz for audio.

Brian M. Boykin 12th January 2020 01:32 PM

Quote:

Originally Posted by johnnyc (Post 14445548)
1) you don't want a cap if the opamp is current feedback topology
2) it depends. You want extra bandwidth but not too much 5-10x is a good rule of thumb, meaning 100-200kHz for audio.

1. Easy enough
2. You answered how to pick the cap value which was ultimately what I was hunting for. I usually go with 33pf as I’d rather have a little too much bandwidth than not enough.

Thanks

JohnRoberts 12th January 2020 03:10 PM

Quote:

Originally Posted by Brian M. Boykin (Post 14445118)
So in the world of swapping modern op amps into older designs I always add a .1uF cap from the +vcc and -vcc to ground and in recent years started adding a 33pf cap across the feedback resistor to further help with stabilization.

Questions:
1) Is there an instance when I should not add a feedback cap?

When using a decompensated op amp (like NE5534) that is not unity gain stable. The popular 5534 IIRC is only stable down to a gain of 10dB (3x).
Quote:


2) Knowing this cap can limit bandwidth, what kind of trouble can I get myself into from lowering a 100pf cap to 47pf or 33pf in an existing design to open up the bandwidth a little more?
generally no harm, but little benefit...

[edit] I am not a fan of second guessing design engineers but recall that capacitors cost money to purchase and then pay labor to insert, if design was better without them why waste money? [/edit]

JR
Quote:

As always,

Thanks in advance

Brian

Brian M. Boykin 12th January 2020 04:05 PM

Quote:

Originally Posted by JohnRoberts (Post 14445989)
When using a decompensated op amp (like NE5534) that is not unity gain stable. The popular 5534 IIRC is only stable down to a gain of 10dB (3x).

generally no harm, but little benefit...

[edit] I am not a fan of second guessing design engineers but recall that capacitors cost money to purchase and then pay labor to insert, if design was better without them why waste money? [/edit]

JR

Thank you John, I was hoping you’d post. My caution with second guessing the original designer stems from your advise on this topic. My goals have changed over the years and I find myself going for lower noise and lower distortion parts over trying to achieve some kind of “sound.” I’m mainly recapping because of age and swapping in low noise low distortion op amps and stabilizing them on a scope. I send gear out if I want it hot rodded.

I got a Peavey VMP-2 in today. I think I’ve read every thread you’ve ever commented on about this preamp. It went in the rack as is for now. We’ll put it through it’s paces to see if we even feel there’s any low hanging fruit. If I like it I’m gonna try for a VC/L-2. Those are harder to come by.

Cpl. Punishment 12th January 2020 04:38 PM

Good info here.

Quote:

Originally Posted by Brian M. Boykin (Post 14446096)
swapping in low noise low distortion op amps and stabilizing them on a scope.

What are you stabilizing them with? Are you breadboarding these and using various values for capacitance and resistance, and how/where are you generating and measuring signal?

Apologies in advance for the stupid questions...

Cabirio 12th January 2020 05:27 PM

Note that the value of the cap is meaningless without knowing the value of the feedback resistor. As a rule of thumb I use 22p for 10k and scale it accordingly, so 100p for 2k2, 4p7 for 47k, etc. This sets a -3dB point around 700kHz and a drop at 20kHz of around -0.003dB, hardly noticeable. Designs where this cap is so large that it limits the bandwidth significantly (i.e. audibly) are rare, and if you come across one, they've probably done it for a reason (see John's edit above), e.g. oscillation due to layout that couldn't be cured another way, so reducing it may get you in trouble.

The one change that may be beneficial is to replace class 2 ceramic caps (most old disc caps with no black colour band on top are) with NP0 / C0G of the same value. The former suffer from aging (they lose capacitance over time) and have a large voltage coefficient (dependance of capacitance on voltage) which may introduce distortion, while NP0 / C0G don't have these problems and modern ones are as small as old class 2 ones.

JohnRoberts 12th January 2020 07:07 PM

Capacitance loss in cheap ceramic discs is measured in percent per decade, so generally not an issue for decoupling caps across PS rails. I would avoid ceramic caps for any audio band filter pole. Of course NPO/COG are the possible exception.

JR

Jim Williams 12th January 2020 07:26 PM

Feedback resistance and poles count. Above about 3k ohms stray capacitance across the resistor body can create a pole, then you need to cancel it with a small cap or ringing/oscillations can occur.

Other factors are involved such as bias currents, operational bandwidth and the output stage current. PCB layouts can create more poles to solve as stray capacitance can be created by proximity to other components or even ground planes placed directly underneath a film resistor.

Brian M. Boykin 12th January 2020 07:29 PM

Quote:

Originally Posted by Cabirio (Post 14446289)
Note that the value of the cap is meaningless without knowing the value of the feedback resistor. As a rule of thumb I use 22p for 10k and scale it accordingly, so 100p for 2k2, 4p7 for 47k, etc. This sets a -3dB point around 700kHz and a drop at 20kHz of around -0.003dB, hardly noticeable. Designs where this cap is so large that it limits the bandwidth significantly (i.e. audibly) are rare, and if you come across one, they've probably done it for a reason (see John's edit above), e.g. oscillation due to layout that couldn't be cured another way, so reducing it may get you in trouble.

The one change that may be beneficial is to replace class 2 ceramic caps (most old disc caps with no black colour band on top are) with NP0 / C0G of the same value. The former suffer from aging (they lose capacitance over time) and have a large voltage coefficient (dependance of capacitance on voltage) which may introduce distortion, while NP0 / C0G don't have these problems and modern ones are as small as old class 2 ones.

Great info. Up until now I’ve road the coattails of the original designer using values they used for the feedback resistor in parts of the design they added them. It’s served me well. They picked those values for a reason so I tend to stick with them. I’ve never changed an existing feedback cap, only added them. I never use ceramics, only NP0/C0G, however, I’ve never changed ceramic disks. I’ll start doing that. There so cheap.

Brian M. Boykin 12th January 2020 07:35 PM

Quote:

Originally Posted by Jim Williams (Post 14446571)
Feedback resistance and poles count. Above about 3k ohms stray capacitance across the resistor body can create a pole, then you need to cancel it with a small cap or ringing/oscillations can occur.

Other factors are involved such as bias currents, operational bandwidth and the output stage current. PCB layouts can create more poles to solve as stray capacitance can be created by proximity to other components or even ground planes placed directly underneath a film resistor.

This is where I’m wanting my understanding to get to. I have known I’m a bit reckless with my modding but it’s so much fun. I’ve never ruined a piece of gear and everything plays nice together in my studio, but I’ve also scaled back how radicle I go.

Brian M. Boykin 12th January 2020 07:42 PM

Quote:

Originally Posted by Cpl. Punishment (Post 14446169)
Good info here.



What are you stabilizing them with? Are you breadboarding these and using various values for capacitance and resistance, and how/where are you generating and measuring signal?

Apologies in advance for the stupid questions...

I’m only upgrading older gear. I recap electro’s, films, and swap in lower noise lower distortion op amps and have recently started switching metal film resistors into the audio path. Transistors has been tough because of all the fakes. If I could ever find a reliable source of sa1084’s or 85’s I’d buy a nice lot of them, as well as MAT 02’s and 03’s. I’ve tried NOS and they never drop in. It’s a bummer.

Anyway, no bread boarding. I add a local psu cap and the feedback cap. I’m not great with an o-scope but I’m checking for overshoot on a square wave. I’m not even sure I do it correctly. I really need to take a class or find someone to walk me through it.

lame pseudonym 16th January 2020 05:33 AM

.
Dig:

https://oriongateway.org/hubble-supe...rdick-rf-gain/

Cpl. Punishment 16th January 2020 06:14 AM

Quote:

Originally Posted by lame pseudonym (Post 14458901)

That inset section "Taming Op Amp Distortion" was both informative and timely.

Thank you sir.

Brian M. Boykin 16th January 2020 04:32 PM

For those still following, what spec in the op amps specs dictates how much gain an op amp can give before it clips? As in, the example in the article is for a 30dB gain increase. What limits a 40dB or 50dB gain in the op amps design?

JohnRoberts 16th January 2020 05:38 PM

Quote:

Originally Posted by Brian M. Boykin (Post 14460424)
For those still following, what spec in the op amps specs dictates how much gain an op amp can give before it clips? As in, the example in the article is for a 30dB gain increase. What limits a 40dB or 50dB gain in the op amps design?

Clipping or saturation is not a function of gain but output voltage (or current).

Running out of gain, like occurs if the closed loop gain (feedback network) is asking for more gain, than the available open loop gain the gain stage can provide. When that occurs the result is a more graceful tapering off of delivered gain. Since most op amps use dominant pole internal compensation their open loop gain rolls off at very high frequency to prevent instability.

If a designer ignores the inadequate gain and asks for more gain than the op amp can deliver, the result will look like a low pass filter as the circuit delivers less output at high frequency.

This is a simplified explanation but close enough for the WWW.

JR

Jim Williams 16th January 2020 05:42 PM

Former BurrBrown EE Jerald Graeme's book "Optimizing Opamp Performance" on McGraw-Hill is a complete treatment of opamp stability issues. I discusses phase lead and phase lag compensation plus noise gain. It has all the math you need.

Brian M. Boykin 16th January 2020 06:42 PM

Quote:

Originally Posted by JohnRoberts (Post 14460805)
Clipping or saturation is not a function of gain but output voltage (or current).

Running out of gain, like occurs if the closed loop gain (feedback network) is asking for more gain, than the available open loop gain the gain stage can provide. When that occurs the result is a more graceful tapering off of delivered gain. Since most op amps use dominant pole internal compensation their open loop gain rolls off at very high frequency to prevent instability.

If a designer ignores the inadequate gain and asks for more gain than the op amp can deliver, the result will look like a low pass filter as the circuit delivers less output at high frequency.

This is a simplified explanation but close enough for the WWW.

JR

Simplified or not it helps. This thread is connecting some dots.

Thank you

Brian M. Boykin 16th January 2020 06:43 PM

Quote:

Originally Posted by Jim Williams (Post 14460818)
Former BurrBrown EE Jerald Graeme's book "Optimizing Opamp Performance" on McGraw-Hill is a complete treatment of opamp stability issues. I discusses phase lead and phase lag compensation plus noise gain. It has all the math you need.

I bought this book many years ago per your suggestion. I’ve tried reading it on several occasions since. Time for another run.

tINY 16th January 2020 07:12 PM

Quote:

Originally Posted by JohnRoberts (Post 14460805)
If a designer ignores the inadequate gain and asks for more gain than the op amp can deliver, the result will look like a low pass filter as the circuit delivers less output at high frequency.

This is a simplified explanation but close enough for the WWW.


No treatise on slew-rate, gain, and frequency?

It can make for some interesting distortions....



-tINY


JohnRoberts 16th January 2020 07:46 PM

Quote:

Originally Posted by tINY (Post 14461272)

No treatise on slew-rate, gain, and frequency?



No.... There is only a weak association between slew rate and GBW as decompensated op amps with smaller dominant pole capacitors exhibit both higher slew rate and higher HF open loop gain, of course with less stability margin.
Quote:

It can make for some interesting distortions....
Slew rate was more of a concern back in the old days, but since the 70s jelly bean op amps have been faster than audio so not much of a concern.

In fact experiencing slew rate limiting distortion is arguably a design flaw. Dr. Marshall Leach (RIP) wrote an AES paper on the subject decades ago. If an input signal/gain stage is properly bandwidth limited, it can not be made to slew limit with valid input signals...

Since the unwashed audiophools still insisted on slew rate specs, some amplifier designers had to defeat the input bandwidth limiting just to make max slew rate measurements, then restored it.
Quote:



-tINY

[edit] if you want to know what gross slew rate limiting sounds like, dust off an old delta modulation delay line and run the clock frequency too low... ;) [/edit]

JR

Brian M. Boykin 29th January 2020 03:25 AM

2 Attachment(s)
So I just went through an Ashly SC63 parametric EQ. Turned out great but some curve balls were thrown at me. The first op amp IC1a has a 27pf cap across the 10k resistor but there’s a resistor following it. Does this count? Will it still stop stray capacitance and tame the bandwidth? For the following op amps.... if it wasn’t an EQ circuit I’d just drop the cap on the foil side across pins 1&2 and 6&7 accordingly for any feedback loop with a resistor of 3k or higher. But there’s a variable resistor and an entire network. Do I still just drop it across 1&2 and 6&7, check with a scope and call it a day? Or is it now more complicated? I used lm4572’s throughout this section of the EQ.

For the filter section, ICA has no cap, so with all the extra circuitry, do I drop the cap across pins 1&2? ICB has no resistor but has a cap, so I leave it as is? I used lm4562’s here also which may not have been the greatest choice because of the 100k VR and the fact I replaced an JFET with a bipolar/BJT and biasing could raise noise. Correct???

Any tips welcome.

Jim Williams 29th January 2020 07:07 PM

ADA4625-1/2 will do very well. 50V/us slew rate, 3.3 nv noise, jfet inputs, duals or singles. Yes, they are very expensive.
Check with a scope, don't guess.

Brian M. Boykin 29th January 2020 11:46 PM

Quote:

Originally Posted by Jim Williams (Post 14497528)
ADA4625-1/2 will do very well. 50V/us slew rate, 3.3 nv noise, jfet inputs, duals or singles. Yes, they are very expensive.
Check with a scope, don't guess.

Do these need a gap on the adapter and a heat sink like the ADA4898-1/2’s? I’m gonna grab some on the next Mouser order and try my hand at dropping them on a Browndog adapter. Mastering that would open up a world of new opportunities.