LM4562, Decoupling, Starpoint, Noise and Oscillation

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
According to Marks proposal, I think the schematic attached is a good solution.
If you are using more than 1 LM4562, this is the way to do it for low THD and less stability/oscillation problems. You need those 2x100u electrolytics AT each package; within 1cm.

You very rarely need the supa low noise of LM4562 so why not use 5532 for less problems? But doing this with 5532 will also give better THD.
 
You need 2x100nF, not 100uF
Actually you need the 2x100uF. Include the 100n s or other Golden Pinnae stuff or not as you wish but DON'T LEAVE OUT THE 2xELECTROLYTICS AT EACH OPA.

This isn't a digital circuit which has different rules.

Kingston had an excellent thread on what is REALLY important for noise & THD; opamps and local decoupling of rails, some questions

opamps and local decoupling of rails, some questions

Many true gurus chime in. It proves how OPA rolling takes a VERY poor second place to correct earthing, layout & decoupling. It’s a long thread but read the whole thing from #41 to find pearls of wisdom.

If you are not interested in reliable good THD performance, then ignore everything I say.
 
Actually you need the 2x100uF. Include the 100n s or other Golden Pinnae stuff or not as you wish but DON'T LEAVE OUT THE 2xELECTROLYTICS AT EACH OPA.

This isn't a digital circuit which has different rules.

Kingston had an excellent thread on what is REALLY important for noise & THD; opamps and local decoupling of rails, some questions

opamps and local decoupling of rails, some questions

Many true gurus chime in. It proves how OPA rolling takes a VERY poor second place to correct earthing, layout & decoupling. It’s a long thread but read the whole thing from #41 to find pearls of wisdom.

If you are not interested in reliable good THD performance, then ignore everything I say.

You need bulk decoupling but certainly not 100uF. 1-10uF is probably adequate and there are billions of designs out there with just the local 0.1uF from each rail to ground that work perfectly fine. The vast majority of extremely high speed op-amp eval boards use something like 0.1 || 10uF both to ground and at the pins or pours. More capacitance and the rail-to-rail cap may be helpful depending on load and circumstances. I am looking right now at the eval board of a high end 24-bit industrial ADC with THD < -120 dB and there is, in fact, not a single electrolytic capacitor on the board.

Thank you for the link though, Samuel's advice is always welcome.

I’m not saying the 100uF are harmful or wrong, but I would probably hesitate in leaving out the 0.1uF caps if there is no other low inductance decoupling.
 
Last edited:
What you want to prevent is mixing the current loops. Here you can clearly see power, output and input grounds.
View attachment 744820 View attachment 744821


Are the GNDs really separate? I see just one and only GND.

You need 2x100nF, not 100uF


I need the 100uF because the load can go down to 600 ohm (23mAp)

If you are using more than 1 LM4562, this is the way to do it for low THD and less stability/oscillation problems. You need those 2x100u electrolytics AT each package; within 1cm.

You very rarely need the supa low noise of LM4562 so why not use 5532 for less problems? But doing this with 5532 will also give better THD.


NE5534: lowest noise
LM4562: lowest THD
 
Are the GNDs really separate? I see just one and only GND.

Just trace the current loops to see what is going on. Small loop area with minimal overlap and inductance.

Things could be improved by moving the output connectors to the right of the three resistors.
Riaa preamp1.png Riaa preamp2.png
 
Last edited:
I managed the oscillation now, but I need some measurements. I put 2x100uF from pin 4 and 8 to load GND and in front of the op a low pass filter with corner freq. ~3.3 MHz.

100uF caps are going to be sorta big, and will have a self resonant frequency near the 'danger zone' of several MHz. Smaller caps will probably work better - you're not trying to hump huge currents here, you're trying to reduce extremely HF impedance. Physically large capacitors are counter to that goal.

However, I guess it worked, so congratulations! Small components with short lead lengths (ideally surface mount) are going to be your friends here, as will PCBs with ground and power planes.

National should have done the op amp with less bandwidth.

Less bandwidth equals more distortion. There are plenty of slower amps to choose from, but (surprise), they produce more distortion. The facts of life :)
 
100uF caps are going to be sorta big, and will have a self resonant frequency near the 'danger zone' of several MHz. Smaller caps will probably work better - you're not trying to hump huge currents here, you're trying to reduce extremely HF impedance. Physically large capacitors are counter to that goal.

No. What you are trying to do is to damp the power rails. Electrolytics have just the right amount of ESR to do this. If you use films without electrolytics, you will have resonance, possible instability & poorer THD.

Yes. Size is important. The correct capacitors are the biggest physical size electrolytics that you can get adjacent to the OPA.

If you can measure very low levels of THD like Kingston, try the circuit with just 2x100n and then with just 2x100u. His thread has more tests & details.
...... as will PCBs with ground and power planes.
If you do use EVIL ground/power planes, make sure you have big cutouts under the input pins of the OPA and any lines connected to them.
 
Actually, modern aluminum electrolytics don't have much ESR. Polymer and the new generation wet electrolytics frequently have ESR < 20mOhm for larger values.

I guarantee if you break open an Audio Precision they don't stick 100uF crap electrolytics across the rails of every single op-amp because they are too lazy to figure out how to design a PDN correctly. You can just actually use an R or use a network analyzer and figure out if you actually have a problem. Agree that film alone is the worst case. Higher L than Class II MLCC in small packages, even less loss.

Your advice is fine for people like the OP who can't even get a tame op-amp like LM4562 to work. In the real world, we don't need to waste vast amounts of PCB space for huge electrolytics on the power pins of circuits that may source almost no current.
 
Last edited:
Actually, modern aluminum electrolytics don't have much ESR. Polymer and the new generation wet electrolytics frequently have ESR < 20mOhm for larger values.
Today, it may be more space efficient to use SMD MLCC ceramics with series SMD resistors to provide the damping ... and indeed I have done just this for a recent design.

I guarantee if you break open an Audio Precision they don't stick 100uF crap electrolytics across the rails of every single op-amp because they are too lazy to figure out how to design a PDN correctly.
Presumably, you work for AP. And presumably, you've measured the THD difference in a complex system like kingston's between only 100n ceramics and only 100u electrolytics.

What is a PDN?
 
I don't work for AP, and that's not a complex system. Some kind of mixer with 10 op-amps and LM3xx regulated rails is just not demanding from a power perspective.

No one is arguing for only 100n ceramics anyway, but 100uF is possibly overkill. I guess TI and AD just make up their datasheet figures when their eval boards have 10uF at most?

I don't know this Kingston, nor do I care what he measured in one specific instance. This is all well understood.

I mean, we agree on the core principle here now that it's been explained, so I guess there isn't much to say. I thought your initial advice sounded like you must have huge capacitance at every single op-amp which isn't really the case.

PDN = power distribution network

You can also find that low value tantalum caps and sometimes polymer tantalums are in a better ESR range for this if you don't want to use the additional series resistor.

only 100n ceramics

I agree with you on the ubiquitous 0.1uF cap. It's like folklore at this point.

I have worked with engineers who will freak out if you take away the 0.1uF caps. As a rule of thumb I try to get the highest value cap I can find in the lowest inductance package. If we have 0402 1uF caps, do we really need a 0.1uF? The impedance at high frequencies is dominated by the mounted inductance anyway.
 
No. What you are trying to do is to damp the power rails. Electrolytics have just the right amount of ESR to do this. If you use films without electrolytics, you will have resonance, possible instability & poorer THD.

Yes. Size is important. The correct capacitors are the biggest physical size electrolytics that you can get adjacent to the OPA.

If you can measure very low levels of THD like Kingston, try the circuit with just 2x100n and then with just 2x100u. His thread has more tests & details.

If you do use EVIL ground/power planes, make sure you have big cutouts under the input pins of the OPA and any lines connected to them.

To each his own. My "evil" power and ground planes on 4 layer PCBs, driven by modern, low impedance regulators like the Analog ADP7142 and ADP7182, with only the bare minimum of bypass capacitance to assure regulator stability (3x Kemet 1µF X7R C1206C105K1RACTU per rail) is giving me 2nd and 3rd harmonic distortion numbers close to -150dBc with the National LM4562, Analog AD797, and TI OPA1611 as measured on an APx-555 at +6dBV drive level. That's basically the same as the APx residual, so the numbers are suspect but I can assure you that "evil" power and ground planes driven by modern regulators can get you below the inherent distortion levels of any monolithic op amp available.

Right now, the AD797 seems to outperform the APX-555 analyzer by a good margin, and reliable numbers for that amp using the "evil" power and ground planes and minimal X7R bypass caps are only obtainable by chaining several amplifiers in series.The OPA1611 and LM4562 are close, but not quite as good, but still well below -140dBc distortion levels.

Attached are some graphs, obtained from the average of 32x of the APx HD sweep tests. The green points are the APx residual and the red points are an "evil" Superbal amplifier built with an AD797, 2kΩ resistors, and power and ground planes with the regulators and bypass caps listed above. The error bars are the '1 sigma' standard deviation of the measurements - at these levels, distortion is about 40dB below the broadband noise, so averaging has to be done to extract useful data from this sweep test.

You can see that the AD797 Superbal amp is basically equal to the APx-555 residual, with only a little added "distortion" due to the noise of the circuit. The 797 amplifier distortion starts to exceed the APx residual above 10kHz, but the APx residual is already becoming high in that range anyway. Also included are similar curves for the OPA1611, which performs close, but not as good as the AD797. For contrast, also included are tests done with the ADA4898-1 amplifier, which is really nice, except that it has a lot of 2nd harmonic distortion.
 

Attachments

  • 797 2nd harmonic sweep.jpg
    797 2nd harmonic sweep.jpg
    73.2 KB · Views: 194
  • 797 3rd harmonic sweep.jpg
    797 3rd harmonic sweep.jpg
    73.8 KB · Views: 165
  • 1611 2nd harmonic sweep.jpg
    1611 2nd harmonic sweep.jpg
    68.4 KB · Views: 160
  • 1611 3rd harmonic sweep.jpg
    1611 3rd harmonic sweep.jpg
    65.5 KB · Views: 149
  • 4898 2nd harmonic sweep.jpg
    4898 2nd harmonic sweep.jpg
    69 KB · Views: 161
  • 4898 3rd harmonic sweep.jpg
    4898 3rd harmonic sweep.jpg
    65.7 KB · Views: 47
If we have 0402 1uF caps, do we really need a 0.1uF? The impedance at high frequencies is dominated by the mounted inductance anyway.

Check the curves of capacitance vs. DC bias - most X7R caps "squish up" under bias, and if you use them with high voltage (e.g. ±15V) rails, you'll find that many high K caps provide only a small fraction of their faceplate capacitance.

I've been using 3 Kemet C1206C105K1RACTU 1µF 50V X7R MLCCs in parallel to achieve a guaranteed 1.5µF capacitance to stabilize the ADP7142 and ADP7182 regulators. I'd like to use only one part per rail to achieve this guaranteed 1.5µF capacitance at 15V bias, but that is not possible given the actual parts we have available. If I'm wrong, please let me know and I'll send you a nice gift! :)

A number of vendors like Kemet and Murata have simulators online that will show you the expected behavior of their parts under DC bias. I think it's worthwhile to use these tools to see what the part will actually do when employed as a power supply bypass cap under DC bias, since some caps squish down to only 20-30% of their faceplate capacitance under bias.

I use these caps to stabilize low impedance regulators, so the bulk of the low rail impedance is from the regulators. However, above 10kHz or so, the bypass caps do 'take over' and dominate the rail impedance as the regulators lose gain bandwidth and feedback margin. Still, an 0402 1uF cap is gonna get destroyed by 15V rails, and will not provide much capacitance, regardless of its geometry.

Not to beat a dead horse, but I cannot find a single cap that will provide 1.5µF of guaranteed load, life, and tolerance capacitance at 15V+ bias. Please correct me if I'm wrong - it's tedious to have to place that many 1µF parts on a PCB!
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.