Proper supply impedance for opamp

[chip_mk]

Ideally + and - supply pins of an audio opamp should see zero impedance, or the lower the better. But at some point we get in the territory of diminishing return.

Supply impedance is result of combined Zout of the voltage regulator, PCB traces and bypassing capacitors near opamp. Pursuing low supply impedance (assuming relatively noise free supply) is not just costly but might result in resonant peaks as side effect. It seems wise to look for a compromise - not to low impedance, ohmic at low frequencies followed by decreasing capacitive impedance of bypass caps, making sure to become ohmic again (ESR kicks in) before L of the supply rail becomes dominant (this is achievable by inserting small value resistor in supply rails.)

Is there some guideline to determine max supply impedance that is still good enough for opamp in serious audio application? Is it OK to arrange for 0,1R, 1R or even 10R at low frequencies and much lower due to bypass caps at HF?

 

[AndrewT]

just ignore all that and use a reasonable regulator to ensure the opamp is NEVER exposed to more voltage than it can survive with.

A Zener diode and resistor makes a pretty good non feedback (no resonances to worry about) regulated supply that works with virtually any opamp.

 

[peufeu]

Honestly you don't need milliohms, superregs or even low noise regs for any opamp having decent LF PSRR.

What opamps really dislike is resonance peaks at HF, because their HF PSRR is crap, so it all goes straight in the output. Those are very easy to avoid : just don't do anything stupid like putting too many (like, 2) fancy super low ESR caps or MLCC or film in parallel, or forgetting to read the "stability conditions" paragraph in your regulator datasheet, or using an audiophile regulator which has no specified stability conditions, or putting inductors in the power line without damping them, etc, etc.

For example if you consider this guy (lol), doing listening tests on caps, well he makes a LC resonant circuit, and selects by ear the cap which has the best amount of ESR to damp it...

The main difference between "audio grade" caps and standard ones, besides fairy dust, and possibly a tiny amount of THD, is that audio grade caps usually have huge ESR, which avoids problems.

If you don't believe it, try ABXing a black gate versus a 10 cents low-esr cap plus a resistor to match the ESRs.

 

[Mark Johnson]

A quite good book by Graeme (Link) advocates a power supply bypassing strategy whose goal is to maintain a flat and constant 1.0 ohms of supply impedance (at the opamp pins!) across the entire bandwidth of the amplifier. He accomplishes this by including more than one bypass capacitor, where each capacitor has a series resistor chosen such that the impedance-vs-frequency of that RC network, has a nice wide, flat zone at 1 ohms. Then appropriately stagger the capacitor values such that the +20dB/decade rise of the Nth-largest capacitor's RC, is cancelled by the -20dB/decade fall of the (N+1)th-largest capacitor's RC. Voila, smooth and flat and purely resistive (i.e. no reactive peaks or troughs) across frequency.

He even explains why he feels 1.0 ohms is a good and righteous choice of target. Does his reasoning include PSRR? Yes. Yes it does. Get the book, it's a worthwhile investment.

 

[chip_mk]

A Zener diode and resistor makes a pretty good non feedback (no resonances to worry about) regulated supply that works with virtually any opamp.

Sounds reasonable, just need to keep ZD close to opamp, close to bypass caps. If properly bypassed ZD noise could probably be ignored. Just two things to consider:
a) PSR is probably not good enough for precision circuits so preceding liner regulator chip is commendable I guess
b) not sure if the opamp performance will be affected by several ohms dynamic impedance of the ZD (the question from the OP)

 

[AndrewT]

There are two other posters telling you you don't NEED ultralow impedance. They are both telling you that non resonance is more important.

That makes three non resonance advices to zero advices for adopting the risk of resonant behaviour.

 

[chip_mk]

I was considering a case when a low Z regulator is somewhat distant from the opamp and usual bypass caps. In that case, rail inductance isn't quite negligible and can produce impedance peak at HF. Adding small serial resistance can prevent this and might be reasonable trade off.

 

[chip_mk]

There are two other posters telling you you don't NEED ultralow impedance. They are both telling you that non resonance is more important.

That makes three non resonance advices to zero advices for adopting the risk of resonant behaviour.

I share the same view. What I'm trying to learn is how high rail impedance is still tolerable (assuming no peaks an being ohmic at LF)? Is there any limit or some guideline?

 

[peufeu]

Where you want low Z is if you got several loads (like, several opamps) feeding from the same supply, and you feel that one of the opamps will draw some ugly current (say, it's doing DAC IV duty). But then, you could put a zener on each, if you like zeners.

Remember the bondwires in any chip will add something like 0.1 ohm. So if you add fancy 5 mOhm ESR capacitor, improvement relative to 100 mOhm capacitor is not so spectacular seen from inside the opamp (but the huge spike from resonance with nearby 100nF caps is)

A quite good book by Graeme (Link) advocates a power supply bypassing strategy whose goal is to maintain a flat and constant 1.0 ohms of supply impedance (at the opamp pins!) across the entire bandwidth of the amplifier. He accomplishes this by including more than one bypass capacitor,

You can do that with small SMD tantalums. Very low ESL, and a bit of ESR, makes a wide band flat bottom cap. Or you can do it with any cap, just add a resistor. I'm not so sure "flat and constant" is that important though. I'd rather bet on absence of resonant peaks. But who knows.

There is a nice paper from audio precision which explains that, with examples and measurements... you may tr to google it...

 

[Elvee]

I share the same view. What I'm trying to learn is how high rail impedance is still tolerable (assuming no peaks an being ohmic at LF)? Is there any limit or some guideline?

You would be surprised at how high it could go. That said, it isn't certainly a good idea to aim at such a target. For example, I am quite certain that an ordinary M3 bolt made of ordinary steel could easily support my weight, but I would certainly not rely on it as an attachment to a zip-line (if I am the user of course ).
You have to try to make the supply impedance as low as reasonable for a good range of frequencies, in particular the higher ones.
If you have to parallel capacitors, for example because one of them is located too far from the POL, they should be in a 1:100 ratio to damp possible resonances.
Search the forum, Eva, myself and others have published valuable advices.
I have posted hundreds of VNA measurements (on another website, tens here).
A small (1 ~10µf) SMD solid tantalum is good tradeoff in practically all cases, even when paralleled with other types. Peufeu said nothing else...

 

[chip_mk]

Where you want low Z is if you got several loads (like, several opamps) feeding from the same supply, and you feel that one of the opamps will draw some ugly current (say, it's doing DAC IV duty). But then, you could put a zener on each, if you like zeners.

Yes in that case I reckon it would be good strategy to have central low Z regulator, then serial resistor toward each opamp and shunt reg (ZD is also shunt reg) near by plus 100-1000uF bypass cap. This looks like good inter opamp decoupling.

 

[abraxalito]

Remember the bondwires in any chip will add something like 0.1 ohm.

I'm surprised to see such a high figure - are they really that bad? What's the gauge of them?

If we take the thinnest wire mentioned on this chart - American Wire Gauge table and AWG Electrical Current Load Limits with skin depth frequencies and wire breaking strength (0.08mm dia) that's showing at 3.4ohms per metre. I can't imagine that any bond wire inside a chip is going to be 1/34m long (29mm). Or do they use much thinner wire or much lower conductivity material than copper?

To answer the OP's original question - the impedance you need on your supplies depends on the loading your opamp is driving and also the sound quality you're aiming for. Don't spend time and money optimizing your opamp supply down to the last mohm if you're going to feed that signal to a normal chipamp with unoptimized supply. The critical supply in the system is the most heavily loaded supply - which is normally the poweramp's because there the load is measured in single digit ohms. Whereas your typical opamp load is in the kohms range.

 

[DF96]

See Wire bonding - Wikipedia, the free encyclopedia.

 

[peufeu]

I'm surprised to see such a high figure - are they really that bad? What's the gauge of them?

The thing is, gold is expensive.

0.1 ohm is just an estimation, you can get something like 10-100 mOhms. If it's a power device, then much less. They got rid of the bondwires entierly in the latest packages like directfet.

Just the inductance of the pin and leadframe and bondwires at a few MHz is going to be more than that anyway. That's why you're not gonna see a GHz opamp in DIP, and big fat FPGAs have like 200 GND/power pins (or more)...