I always go for 'A' and 'B' together for an NE5532.
As Doug Self has explained in the past, both are necessary if you want to prevent (A) the possibility of the NE5532 internally "hooting", and (B) HF disturbances from the supply rails entering the supply pins. The requirement for 'A' will depend largely upon how the NE5532 is externally configured but, for the sake of a few pennies, better safe than sorry in my opinion. Keep 'A' as close to pins 4 & 8 as possible.
Note: The supply pins are 8 (+) & 4 (-), not 4 & 5 as shown in your schematics.
Under some circumstances, either may be considered overkill and may make no audible let alone measurable difference, but my gut tends to go for a 'best practice' approach and do both anyway as it can save having to troubleshoot later. They're still an astonishingly good opamp when configured appropriately, so you might as well give them the best chance you can to show you all they've got.
Box polyesters are probably a sensible choice if you want to avoid capacitors with unusually high Qs.
As Doug Self has explained in the past, both are necessary if you want to prevent (A) the possibility of the NE5532 internally "hooting", and (B) HF disturbances from the supply rails entering the supply pins. The requirement for 'A' will depend largely upon how the NE5532 is externally configured but, for the sake of a few pennies, better safe than sorry in my opinion. Keep 'A' as close to pins 4 & 8 as possible.
Note: The supply pins are 8 (+) & 4 (-), not 4 & 5 as shown in your schematics.
Under some circumstances, either may be considered overkill and may make no audible let alone measurable difference, but my gut tends to go for a 'best practice' approach and do both anyway as it can save having to troubleshoot later. They're still an astonishingly good opamp when configured appropriately, so you might as well give them the best chance you can to show you all they've got.
Box polyesters are probably a sensible choice if you want to avoid capacitors with unusually high Qs.
This looks like a misunderstanding of what "as close as possible" means. It's the whole loop distance, not just from each supply to the nearest capacitor. It's not just the wording of the datasheet that the capacitor should be "as close as possible" to the device, but ALSO that the TRACES for both connections between the capacitor and the device should be as short as possible. Schematics don't really show this properly as it's assumed every connection has zero resistance and zero inductance. A PCB layout demonstrating "as short as possible" would be much more instructive. This may or may not be critical, depending on the bandwidth of the op-amp.
Loop area is the critical factor, and why ground planes are so important.
The image currents in the ground plane reduce the remaining trace inductance.
https://www.amazon.com/High-Speed-Digital-Design-Handbook/dp/8131714128/ref=asc_df_8131714128/
The image currents in the ground plane reduce the remaining trace inductance.
https://www.amazon.com/High-Speed-Digital-Design-Handbook/dp/8131714128/ref=asc_df_8131714128/
Last edited:
If he was wrong, then AudioPrecision's best testgear is wrong also... He really knows what he's doing and has the testgear and experience and publishes the results of his research. Do you have any results to share about this?
Just wondering if anybody has any information about the 'Hooting'?
What is it, what causes it, any reference material to look at, for example.
I've not been able to find anything online so far and wonder if it is still valid with today's production items.
What is it, what causes it, any reference material to look at, for example.
I've not been able to find anything online so far and wonder if it is still valid with today's production items.
avtech - The cause is apparent from the NE5532's schematic.
The 37pF capacitor couples a signal referenced to the negative rail to a base of the PNP stage. The 36pF capacitor couples the positive rail to the other PNP base. The result is that at high frequencies, the PNP stage sees the difference between the supply rails as a signal.
This is a bad design from the point of view of PSRR.
Ed
The 37pF capacitor couples a signal referenced to the negative rail to a base of the PNP stage. The 36pF capacitor couples the positive rail to the other PNP base. The result is that at high frequencies, the PNP stage sees the difference between the supply rails as a signal.
This is a bad design from the point of view of PSRR.
Ed
If you look at what happens when uou get current draw, and there is high frequency, then you’ll want that decoupling cap between the two power supply pins (which can be +V and gnd if that is the power) so low impedance wave then propgates to the other side of the cap..
If your opamp only has + and - then that’s the only return paths for the power noise (ie the drawn power) to loop back.
What I have always seen is in amps, the power caps are rail-to-ground, even in bipolar otherwise it creates more noise perhaps?
Last edited:
The belief is the NE553x issue is with very fast oscillations (perhaps microwave region) internal to the chip when the supply rails have too much inductive reactance, the effect is not easily observable other than by a small but distinct increase in distortion. Its a reasonably common situation for unwanted Colpitts and Hartley oscillators to exist at VHF/UHF in low frequency circuits, sometimes going unnoticed. I suspect it varies between devices, batches, manufacturers and with supply voltage
Having the cap ceramic and directly between the rails minimizes the series inductance so is the best approach for this condition. Decoupling at audio frequencies is another matter, related to the overall circuit.
Having the cap ceramic and directly between the rails minimizes the series inductance so is the best approach for this condition. Decoupling at audio frequencies is another matter, related to the overall circuit.
It's a term that was widely used in the UK for at least half a century to describe self-oscillation. I hadn't realised it had fallen out of favour. Apologies if I was unclear.
Mark Tillotson and EdGr explain why and how best to avoid it with an NE5532.
I had a lot of results on this with AP gear. But they where done in the previous Millenium. These days, I'm a beach bum.
There was a huge thread on decoupling OPAs in a mixer where someone followed my suggestions and found them to work but you'll have to dig it up yourself. Many gurus (including Scott Wurcer) and pseudo gurus (who shall remain nameless) joined in.
Self damns with faint praise some older OPAs that in practice outperform the newer uber OPAs especially in circuits using more than one package. That's cos these are less fussy in their decoupling requirements. 5532 counts as 'moderately fussy'.
As I said, if you can measure very low THD, you can confirm the worth of my suggestions. If not, well it probably doesn't really matter to you.
There's also the proper use of ground planes for low audio THD ... but as I can no longer show results, I'll just shut up 🙂
Wayne Colburn's already been there: the WHAMMY features 47 ohms in series with 220 uf to ground in both opamp power lines, although in this case I suspect that's to help with high frequency oscillation since the outputs are on the same lines as the opamp and opamp PSRR above 1 MHz is typically not great.
Along similar lines, Analog Devices recommend thinking carefully when using their LT3081 and LT3091 low-dropout regulators. Look for the data sheet section called "Stability and input capacitance". The short skinny is that power lines have inductance, which creates a tuned circuit with bypass caps and possibly leading to instability. A few measures can be taken: using tantalums instead of ceramics since they have higher ESR, increasing the value of the bypass cap as power lines get longer, or louse up the tuned circuit with a bit of series resistance before the bypass.
I slap 3-10 uF ceramics on opamp bypasses and feed them with a half-ohm resistor in series with the power feed. Works a treat and the bigger caps are marginally more expensive than the 0.1 uF items. Only thing you gotta watch is backwash through the regulators when the power supply is switched off, easily cured by a reverse-biased diode across the reg which conducts only when the voltage after the regulator is greater than before it.
Decoupling becomes ever more critical when using the latest generation of opamps. These have much greater gain bandwidth than the exceptionally long lived and excellent 5532. They have much lower distortion, and often much lower voltage and current noise, For example the OPA1611/1612 or OPA828. Surface mount of course, whereas the older opamps are still available in through hole versions. Get decoupling wrong with the new generation opamps and they will oscillate at RF.
This is at https://groupdiy.com/threads/opamps-and-local-decoupling-of-rails-some-questions.37307/page-3 Read it all from at least #41 to pick up all the pearls of wisdom. 🙂
The one I'm thinking off is OPA2132. It is a secret ingredient used by some designers to get better performance than with da uber OPAs. It is also more resistant to EMI/RFI
I had some correspondence with Guru Scott Wurcer, who designed the AD797 and some of the other lowest noise OPAs in the known universe, about its use in the Audio Precisions. He said with hindsight, he should have been a lot more precise and insistent about the recommended decoupling schemes in the datasheets.
This is another useful resource, despite the D word in the title and that it's 30 years old:
High Speed Digital Design: A Handbook of Black Magic
https://www.amazon.com/High-Speed-Digital-Design-Handbook/dp/0133957241
It's basically about PCB design. The TL;DR is that all 'fast' (switching or square wave) signals are R-C time constants, the R being the output impedance of the driving device, and C being the capacitance between the trace and ground.
More pertinent to this discussion, the second half discusses ringing of power traces (which being wires are naturally inductive), and that adding bypass capacitors may not reduce the amplitude, but only change the frequency. The given solution is adding a series R-C circuit between power and ground (at one or more points), with R roughly in the 0.1 to 2 ohm range, and C the usual 0.01uF to 1uF to damp such ringing.
Checking bookfinder, I see used copies starting at $25.
High Speed Digital Design: A Handbook of Black Magic
https://www.amazon.com/High-Speed-Digital-Design-Handbook/dp/0133957241
It's basically about PCB design. The TL;DR is that all 'fast' (switching or square wave) signals are R-C time constants, the R being the output impedance of the driving device, and C being the capacitance between the trace and ground.
More pertinent to this discussion, the second half discusses ringing of power traces (which being wires are naturally inductive), and that adding bypass capacitors may not reduce the amplitude, but only change the frequency. The given solution is adding a series R-C circuit between power and ground (at one or more points), with R roughly in the 0.1 to 2 ohm range, and C the usual 0.01uF to 1uF to damp such ringing.
Checking bookfinder, I see used copies starting at $25.
Although a little off topic (and out of bandwidth) I've designed a clock and isolator PCB for 24MHz. I found that a good tuned inductor in the power line with decoupling caps on the end made a large difference to the power line noise whilst maintaining a clean signal. I tried two SMT inductors - one in the ohm range and the other 600R impedance at 24MHz but lowish impedance at lower frequencies. The 600R made the most difference.
The PCB was designed on a 4 layer board with a focus on component placing to minimise loops (as it gets below 100KHz) and reducing distance for return currents under the traces above 100KHz.
I like discussions like this. Everyone has an opinion, and many of the more experienced here are right.
Doug was right, but remember that he was looking for the absolute minimum required, because the multi-channel analogue mixing desks he was working on could not afford the luxury of NE5532s over TL072s unless really needed for output drive and noise reasons. The fact that he found a pair of electrolytic caps to ground could keep stable multiple NE5532s up to certain distance, but far more TL072s, so the extra caps needed for NE5532s added to the cost. He outlined minimum best practice, mostly on through hole 1 or 2 layer PCBs. As said modern opamps, on SMT PCBs in todays RF rich environment, minimum best practice can give way to better practice when we have cheaper passives and multi-layers available.
Also the topology makes a difference. An inverting opamp stage has a direct ground reference, whereas a non-inverting or differential does not on the pins of the opamp. So the PSRR is different depending on topology, and therefore the minimum requirement varies.
Please remember, at low frequencies PSRR is very high, and different for each rail, but lowers at about 6dB/oct 20dB/decade. At some point it can go below 0dB, meaning the PSU pins actually amplify! So HF decoupling is vital.
Paralleling caps can lower ESR, but remember, the same size MLCC cap has roughly the same minimum frequency ESR. So parallel different footprints if you really want to get a spread of low ESR at HF.
X7R etc. are OK for decoupling, as while they add THD and are piezoelectric in the signal path, decoupling this is non invasive. However, they have terrible dC/dV. So choosing a cap to be used near its rated voltage means you may get less than half of the expected capacitance, AND considerably lower its lifespan.
So use plenty of decoupling: A or B depending on topology, and put regular electrolytics to ground, but make it a star PSU ground if you have a signal ground following the signal.
As Groucho said: These are my opinions, and if you don't like them, I have others... or summat...
Doug was right, but remember that he was looking for the absolute minimum required, because the multi-channel analogue mixing desks he was working on could not afford the luxury of NE5532s over TL072s unless really needed for output drive and noise reasons. The fact that he found a pair of electrolytic caps to ground could keep stable multiple NE5532s up to certain distance, but far more TL072s, so the extra caps needed for NE5532s added to the cost. He outlined minimum best practice, mostly on through hole 1 or 2 layer PCBs. As said modern opamps, on SMT PCBs in todays RF rich environment, minimum best practice can give way to better practice when we have cheaper passives and multi-layers available.
Also the topology makes a difference. An inverting opamp stage has a direct ground reference, whereas a non-inverting or differential does not on the pins of the opamp. So the PSRR is different depending on topology, and therefore the minimum requirement varies.
Please remember, at low frequencies PSRR is very high, and different for each rail, but lowers at about 6dB/oct 20dB/decade. At some point it can go below 0dB, meaning the PSU pins actually amplify! So HF decoupling is vital.
Paralleling caps can lower ESR, but remember, the same size MLCC cap has roughly the same minimum frequency ESR. So parallel different footprints if you really want to get a spread of low ESR at HF.
X7R etc. are OK for decoupling, as while they add THD and are piezoelectric in the signal path, decoupling this is non invasive. However, they have terrible dC/dV. So choosing a cap to be used near its rated voltage means you may get less than half of the expected capacitance, AND considerably lower its lifespan.
So use plenty of decoupling: A or B depending on topology, and put regular electrolytics to ground, but make it a star PSU ground if you have a signal ground following the signal.
As Groucho said: These are my opinions, and if you don't like them, I have others... or summat...