NE5532 decoupling capacitors

[kgrlee]

Ok, so is this mainly a problem then designing HF circuits in the Mhz range. Or could this also be a problem for audio circuits, for example making them unstable and oscillate?

Yes

 

[sawyers]

Some opamps also need correct sequencing of the power rails. The 5532 is a case in point, needing a few volts on the negative voltage before the positive is connected to prevent a big output transient. This is from Burwen's design notes for the Cello Audio Palette

For decoupling, Self says (of the 5532) "The essential requirement is that the +ve and -ve rails are decoupled with a 100nF capacitor between them at a distance of no more than a few mm from the opamp. Normally one such capacitor is fitted per package. It is not necessary, and often not desirable, to have two capacitors going to ground. Every capacitor between a supply rail and ground carries the risk of injecting rail noise into the ground." (Small Signal Audio Design. second edition)

Of course the 5532 has pretty benign characteristics (10MHz unity gain BW and 9V/us slew).

The much more recent (bipolar) OPA1612 is 40MHz unity gain BW and 27V/us, and the (jfet) OPA828 45MHz and 150V/us. These need more extreme decoupling considerations to tame them, but benefit by being SM only, so decoupling SM capacitors can be got much closer to the active device. The data sheets have design examples of how to do this correctly.

Craig

 

[Bonsai]

Recommendation is because of parasitic capacitances forming between tracks and ground plane. They are easily 3 – 10 pF. Main issue is parasitic capacitance effect on negative feedback network, where it causes gain peak at high frequency (MHz range) and can provide conditions for circuit oscillation.

That’s why often the recommendation is to clear the GP around the feedback node - ie the inverting input. You can in any event compensate for this type of problem in a VFA with a small capacitor across the feedback resistor. Unfortunately, CFA’s don’t lend themselves to this type of approach.

 

[Bonsai]

Funny, I’ve never had distortion or decoupling problems with the AD797. On all my opamp circuits, I always use 0805 decoupling caps placed directly under the opamp so the traces to the supply pins and ground are really short. When laying a board out, these are the first components placed after the opamps. I guess when you have 50 opamps on a mixer board, the problem is really a lot more difficult.

 

[bengtssk]

On all my opamp circuits, I always use 0805 decoupling caps placed directly under the opamp so the traces to the supply pins and ground are really short.

What type and size do you usually use?

 

[HalFoster]

I would suggest taking the datasheets recommendation:

https://www.analog.com/media/en/technical-documentation/data-sheets/AD797.pdf

page 13.

Hal

 

[Mark Tillotson]

I've just been measuring some LM4562's for distortion and found that 100nF across the supply pins is not enough to prevent oscillation with this opamp - adding 1µF to ground from each supply rail fixed the issue (which showed up as clearly raised distortion levels). So beware replacing an NE5532 with an LM4562 - the decoupling requirements are definitely more stringent.

[ The datasheet for the LM4562 shows using parallel 100nF and 1µF on each rail to ground in one of the several application circuits. ]

And of course X7R ceramics are just fine for this, no special requirements on them other than low inductance.

The plots show the correctly functioning LM4562 distortion is not even measurable with the QA403 analyzer, but without the extra decoupling its woefully falling short of its specs (without any obvious other indication).

So I can see most opamp-rollers being completely caught out by this.

[ where "not-decoupled" means only 100nF between the power pins, "decoupled" is this plus 1µF on each rail to ground, "loopback" is the distortion floor of the QA403 ]

 

[cvanc]

Fascinating and thank you but please one more data point: No caps to ground, increase to 1uF between the power pins. That would really bound the issue.

 

[abstract]

There was a whole thread on decoupling capacitor techniques. Apparently the performance may vary, at least in part because of asymmetric internal op-amp circuitry.

Edit, wait! Refer to page 1 of this thread XD.

 

[Mark Tillotson]

Fascinating and thank you but please one more data point: No caps to ground, increase to 1uF between the power pins. That would really bound the issue.

Just tried that - no dice, it seems the positive rail definitely needs decoupling to ground, negative less clear.

 

[Markw4]

IIRC, negative power terminal of most opamps is the reference terminal. Maybe a good idea to have a low-Z path to ground?
An exception is OPA1622 which has a ground pin for the internal compensation cap. Lower distortion that way, again IIRC.

 

[DSP_Geek]

I've just been measuring some LM4562's for distortion and found that 100nF across the supply pins is not enough to prevent oscillation with this opamp - adding 1µF to ground from each supply rail fixed the issue (which showed up as clearly raised distortion levels). So beware replacing an NE5532 with an LM4562 - the decoupling requirements are definitely more stringent.

[ The datasheet for the LM4562 shows using parallel 100nF and 1µF on each rail to ground in one of the several application circuits. ]

The TI datasheet shows 0.1 uF ceramic in parallel with a 10 uF electrolytic on page 25 here. Are your 1 uFs ceramic or electrolytic?

Thanks for the tip: that was an easy one to miss. I sure did.

And of course X7R ceramics are just fine for this, no special requirements on them other than low inductance.

Careful with that axe Eugene!

X7R parts are nice & compact but their capacitance decreases somewhat with voltage, as explained in an Analog Devices blog, at Murata, and at Vishay (showing off their own parts, of course). The effect is more pronounced with smaller MLCCs (multi layer ceramic capacitors) of the same rating, so for example an 0603 part could lose 70% of its value against 20% for a 1206 part at the same DC bias.

 

[Mark Tillotson]

I used 1uF ceramic as they were lying around. The stray inductance usually dominates decoupling cap impedance at high frequencies so the droop in capacitance with voltage may not be a big factor - in fact the NE5532 survives happily in the same setup with a single 100nF cap across the pins (so 36V across it, not 18). I'm sure you can used stacked film caps too as they are low inductance, but many film caps are unclear about their construction.

My PSU is remote so there's no bulk decoupling nearby, which is probably unrepresentative of most setups, but there's a clear difference between the two chips' sensitivity to decoupling.

Incidentally the oscillations could be seen on the 'scope at low amplitude at about 20MHz.

 

[DSP_Geek]

The OPA1656 datasheet sneaks in something similar for the EMI Rejection Ratio spec:

Now I'm hunting for similar caveats in other spec sheets. It's annoying: "Oh yes, the car will go perfectly well on these tires but you'll need the extra special rubber if you want to make turns." Given your distortion measurements, I wonder how much performance is lost in other designs because of reductionist bypassing recommendations.

Edit: the OPA1602 and OPA1612 spec sheets don't seem to have this gotcha. Nevertheless, it's at least 1 uF ceramics on all PSU pins from now on. I had a few 3 uFs lying around from other projects so I slapped them into my headphone preamp bypasses, which seems to have been the accidental right thing to do.

 

[Michael Bean]

When all other options have been exhausted...try following the manufacturer's recommendations. 😊

Mike

 

[DSP_Geek]

Following the manufacturer's recommendations would be ever so much simpler if those weren't carefully hidden.

OPA1656:

Connect low-ESR, 0.1-μF ceramic bypass capacitors between each supply pin and ground, placed as
close to the device as possible. A single bypass capacitor from V+ to ground is applicable for single-
supply applications.

However the EMIRR schematic also shows a 10 uF electrolytic capacitor beside the 0.1 uF ceramic.

LM4562:
Only a demo schematic, which shows 0.1 uF in parallel with 10 uF. No reason given.

LM49720, supposedly the LM4562 in drag:

The LME49720 requires adequate power supply decoupling to ensure a low total harmonic distortion (THD).
Place a low equivalent-series-resistance (ESR) ceramic capacitor, typically 0.1 μF, within 2 mm of the V+ and V-
pins. This choice of capacitor and placement helps with higher frequency transients, spikes, or digital hash on the
line. In addition to the 0.1 μF ceramic capacitor, it is recommended to place a 2.2 μF to 10 μF capacitor on the
V+ and V- pins. This larger capacitor acts as a charge reservoir, providing energy faster than the board supply,
thus helping to prevent any droop in the supply voltage.

it's not only opamps. Relays also have carefully concealed gotchas, such as this beauty on page 13:

If a current is applied to the coil over a long period of time, the coil temperature rises, promoting generation of organic
gas inside the relay, which may result in faulty contacts. In this case, use of a latching relay is recommended.

 

[Mark Tillotson]

Sounds like a reason to choose a different brand of relay(!).

I suspect going with the LM4562 demo schematic is wise, even if its overkill for most opamps. The larger caps aren't really about voltage droop, they are to suppress unintended feedback at ultrasonic frequencies via the rails which could lead to oscillation. Typically one of the rails has a much lower PSRR than the other, so you might get away with one cap for this, but caps are cheap and symmetry feels good(!)

 

[DSP_Geek]

They're KEMET relays so not junk. A graze through other DIP/SMT relay specs shows much the same language.

I'm closing in on this bypass scheme: 200ish uF electrolytic || 10 uF ceramic from V+ to V-, 2.2 uF ceramic from V+ to ground, 2.2 uF ceramic from V- to ground, with the whole thing fed by 47 ohm resistors from regulators to local power. My headphone preamp is close to that layout and it's quiet.

 

[Markw4]

Sounds like a reason to choose a different brand of relay(!).

Omron says the same thing, and they make very good small signal relays. They are just giving engineering info. They recommend their latching products if that is a better fit to customer needs. Personally, I always avoided magnetically latching relays. Tried some recently though. Decided I like 'em.

 

[Mark Tillotson]

Latching relays aren't appropriate for speaker protection as you need to know the relay is open-circuit at power-up. Relays able to handle 100% duty cycle without cooking themselves definitely exist!