General power supply and op-amp noise reduction techniques

[decc]

I've been searching the forums this morning for these subjects and was hoping for corrections/additions. And if this thread can aggregate things for others even better.

You will see some points are incomplete, especially the discrete transistor section since I haven't looked in to that very much.

All of this of course applies to pedal design. There's a wealth of information in various app-notes that don't necessarily apply to our frequency/voltage levels/etc.

As I understand it:

Power

• Small (< 100-ohm, 20-30 maybe?) resistor between input and circuit. Note that there will be a voltage drop the larger the resistor, while more power dissipated the smaller the resistor.
• Better than the resistor, use a small inductor (1000uH? What kind/part number?). This filters AC noise without a DC voltage drop.
• Bypass after the resistor/inductor with a 22-100uF electrolytic and .01-.1uF ceramic.

Grounding

• If using a ground plane: make sure it isn't split. Otherwise you will have ground loops.
• If not using a ground plane: keep power, analog, and digital grounds separate and route in a star configuration.

Op-amp buffer/gain stages

• Choose a low-noise chip such as TL072, (others?).
• Every IC should also have a .01-.1uF ceramic as close as possible to its power pins.
• High frequencies should be filtered to help prevent oscillation.
     
  
     
  • Non-inverting op-amps should have a small (~47pF) ceramic cap across the output and the non-inverting input.
  • Inverting op-amps should have the cap from the inverting input to ground.
  • The small cap can be omitted if the op-amp is part of a tone-shaping network that already has a high-frequency roll-off.
  
  


• The resistance (impedance?) to the + and - inputs should be matched.
     
  
     
  • In an inverting configuration the - input is a "virtual ground", therefore the + input should be tied directly to the reference voltage.
  • In a non-inverting configuration the pull-up from the + input to V_ref should be equal to the parallel combination of the resistors setting the gain on the - input.
  
  

[R.G.]

I normally think that trying to avoid/defeat noise, so I was surprised at myself not being happy about this approach to noise reduction.

After some speculating, I think it just struck me that a checklist approach to noise is not going to be very productive. Some of these things need actual understanding to do right. F'rinstance:

Small (< 100-ohm, 20-30 maybe?) resistor between input and circuit. Note that there will be a voltage drop the larger the resistor, while more power dissipated the smaller the resistor.

This particular technique is for two purposes, reducing local power supply impedance, and for isolating between sections. The series resistor/inductor and shunt cap form a voltage divider that divides more as frequency goes up. It's particularly good for isolating circuit sections at mid audio and above. It's mandatory for any high impedance circuit with more than about three stages. However, feedback around three or more stages isolated this way will cause motorboating oscillation.

Better than the resistor, use a small inductor (1000uH? What kind/part number?). This filters AC noise without a DC voltage drop.

And better because the impedance of the inductor goes up as frequency increases, while the cap's is going down. However, at some frequency they resonate, and you can cause oscillation right there. It's practically mandatory to be able to damp the resonance with resistance unless the resonance is way out of the audio spectrum - which makes choosing that inductor trickier.

Bypass after the resistor/inductor with a 22-100uF electrolytic and .01-.1uF ceramic.

... because electro caps quit looking like capacitors at some frequency as the ESR and ESL swamp the capacitive reactance. Ceramic disks have low ESR and help on out through RF.

If using a ground plane: make sure it isn't split. Otherwise you will have ground loops.

Ground planes are OK, but are not all that great at audio frequencies. There is a continuum of grounding from star to plane, with star better at low frequencies and plane at high. Ground loops are not necessarily a problem with even split planes at audio.

If not using a ground plane: keep power, analog, and digital grounds separate and route in a star configuration.

If not using a ground plane, know the difference between power, sewer, and voltage reference "grounds", know what currents are going into which kind of ground.

Choose a low-noise chip such as TL072, (others?).

Practically every opamp since the 741 says it's "low noise". Makes separating out the good ones harder.

Every IC should also have a .01-.1uF ceramic as close as possible to its power pins.

... for stability reasons, not hiss.

High frequencies should be filtered to help prevent oscillation.

It would be much better to think of this as "Never use more bandwidth than you need. Otherwise you let in more noise than you have to." It's probably more oriented to removing hiss, hum, and 1/f noise than oscillation.

Non-inverting op-amps should have a small (~47pF) ceramic cap across the output and the non-inverting input.

This is primarily to swamp any inter-trace and output pin capacitance.

Inverting op-amps should have the cap from the inverting input to ground.

I'm not sure what this achieves.

In an inverting configuration the - input is a "virtual ground", therefore the + input should be tied directly to the reference voltage.

Well, it ought to be tied to someplace quiet and that generally means low impedance. If your reference voltage is noisy, it injects noise when you do this.

In a non-inverting configuration the pull-up from the + input to V_ref should be equal to the parallel combination of the resistors setting the gain on the - input.

This is primarily for DC accuracy. Using the resistor for DC offset balance because of input currents is how this started. It adds in the thermal noise of the resistor directly into the high impedance + input, though.

Some of these things are not two-edged swords, they're five or six edged swords. 

[Cliff Schecht]

I was about to type up a long winded answer like R.G. did, but there is so many different topics that low-noise op amp design encompasses. I'm instead just going to recommend a good book on the topic - "The Art of Electronics" by Horowitz. I know I recommend this book a lot here but IMO there isn't a better source for general, easy to understand electronics information. In the book, there is a whole chapter on noise, where it comes from and how to prevent it.

There are literally dozens upon dozens of good books and app notes on the topic however, so if you're looking for anything more specific, I might be able to dig up something in my ebook collection.

[Cliff Schecht]

Also, R.G., still interested in those books I mentioned a few years ago? I've got a little over a 1000 ebooks on electrical engineering now if you're interested .

[km-r]

we should have a sticky about this or something...

[R.G.]

Also, R.G., still interested in those books I mentioned a few years ago? I've got a little over a 1000 ebooks on electrical engineering now if you're interested .

Yea. Send me a link.

I'm old school you know. I probably have close to a tone of real, paper techie books. They work well, but the search function seems to be erratic. 

[Cliff Schecht]

I've got a decent amount of paper on the bookshelves myself, although probably not as much as yourself. My dad may have you beat OTOH, he's got a massive collection of books. Fortunately, he doesn't care if I pick and choose what I want when I'm in Austin so it works out quite nicely.

[decc]

Small (< 100-ohm, 20-30 maybe?) resistor between input and circuit. Note that there will be a voltage drop the larger the resistor, while more power dissipated the smaller the resistor.

This particular technique is for two purposes, reducing local power supply impedance, and for isolating between sections. The series resistor/inductor and shunt cap form a voltage divider that divides more as frequency goes up. It's particularly good for isolating circuit sections at mid audio and above. It's mandatory for any high impedance circuit with more than about three stages. However, feedback around three or more stages isolated this way will cause motorboating oscillation.

When you say "stages" do you mean high-gain stages or even simple buffers/followers? If the former I don't think that comes up very often,  if the latter I'm surprised I haven't seen this used all the time between buffers. Then again multiple op-amp ICs have common power pins so I guess one doesn't have the option most of the time.

Better than the resistor, use a small inductor (1000uH? What kind/part number?). This filters AC noise without a DC voltage drop.

And better because the impedance of the inductor goes up as frequency increases, while the cap's is going down. However, at some frequency they resonate, and you can cause oscillation right there. It's practically mandatory to be able to damp the resonance with resistance unless the resonance is way out of the audio spectrum - which makes choosing that inductor trickier.

I don't even know if this would be the right ballpark, but this Mouser Page has chokes with resonances in the MHz range. Would picking one involve choosing 1) highest inductance with 2) current capability to match the circuit, and 3) self resonance frequency outside the audible range?

Bypass after the resistor/inductor with a 22-100uF electrolytic and .01-.1uF ceramic.

... because electro caps quit looking like capacitors at some frequency as the ESR and ESL swamp the capacitive reactance. Ceramic disks have low ESR and help on out through RF.

Oh good, looks like I got one right.

High frequencies should be filtered to help prevent oscillation.

It would be much better to think of this as "Never use more bandwidth than you need. Otherwise you let in more noise than you have to." It's probably more oriented to removing hiss, hum, and 1/f noise than oscillation.

How does one achieve this? An RC low-pass filter on every stage?

Non-inverting op-amps should have a small (~47pF) ceramic cap across the output and the non-inverting input.

This is primarily to swamp any inter-trace and output pin capacitance.

Inverting op-amps should have the cap from the inverting input to ground.

I'm not sure what this achieves.

This was following my reasoning that the small cap was a LP filter. Is it correct that in both inverting and non-inverting arrangements the cap should be from output to -? What problems do inter-trace and output-pin capacitance cause that we want to counter-act here?

In an inverting configuration the - input is a "virtual ground", therefore the + input should be tied directly to the reference voltage.

Well, it ought to be tied to someplace quiet and that generally means low impedance. If your reference voltage is noisy, it injects noise when you do this.

A properly filtered reference (electrolytic and ceramic caps) at the voltage divider would help to remove noise, correct?

In a non-inverting configuration the pull-up from the + input to V_ref should be equal to the parallel combination of the resistors setting the gain on the - input.

This is primarily for DC accuracy. Using the resistor for DC offset balance because of input currents is how this started. It adds in the thermal noise of the resistor directly into the high impedance + input, though.

Since we don't care about DC accuracy would any old resistor be "good enough"? Would a low value or high value be better?

Some of these things are not two-edged swords, they're five or six edged swords. 

Yeah, I guess putting together a to-do list isn't going to happen. But a list of things to try, that may or may not work, would be good too. Thanks for all the help though!

I was about to type up a long winded answer like R.G. did, but there is so many different topics that low-noise op amp design encompasses. I'm instead just going to recommend a good book on the topic - "The Art of Electronics" by Horowitz.

Thanks, I'll definitely check this out.

[biggy boy]

Power

• Small (< 100-ohm, 20-30 maybe?) resistor between input and circuit. Note that there will be a voltage drop the larger the resistor, while more power dissipated the smaller the resistor.

Hi decc
When you say put a "resistor between input and the circuit" Are you referring to the power supply output that feeds the Dc voltage to the circuits (tube or transistor)?
Or are you referring to the input of a circuit from the audio signal going into a guitar amp jack for example?

Glen[/list]

[decc]

Hi decc
When you say put a "resistor between input and the circuit" Are you referring to the power supply output that feeds the Dc voltage to the circuits (tube or transistor)?
Or are you referring to the input of a circuit from the audio signal going into a guitar amp jack for example?

Glen

The former. The resistor should be the first thing right at the point where power from the DC jack connects.