Question about proper decoupling

Mateusz Kaluzny17

Other Parts Discussed in Thread: OPA1604, OPA1622, LM317, LM2674, LM2675

Hello dear Ti community.

I know decoupling thread was explained thousand times and there is some general rules that should be followed.

But I've got some other questions that might be important as well.

So, I will try to explain my design and why I really care about those.
My desing is builded around few opamps and analog muxes.

I will look like this :

1x OPA1604 (audio input stage) 3x paralleled OPA1622 which will drive load directly (headphones), and analog mux around OPA1604 for gain manipulation (powered from bipolar supply).
This is one channel only.

I've got rather good PSU builded around ultra low noise LDO's.

Most common technique is to place 100nF capacitor as close as possible to IC pins (with opamps that are hi current output it's often recommended to place a bulk capacitor to provide enough charge).
And I already made that, each opamp got it's own bulk capacitor (10uF) followed by 100nF capacitor + another smaller like 1000pF.
My PCB is 4 layer board, so I can use internal layer for power to avoid thin/long power trace with significant inductance, or I can use outer layer to route fat enough power traces.
So PCb could be handled well by me and it shouldn't be an issue here.

Question is:

- should I isolate each opamp power supply from each other via ferrite bread (even if they are tends to work at higher freq) or resistor (I will affect dynamic range).
- or I should forget about it?

Should I include that kind of isolation? Imporant question is where should I care about it and when it's not required.
From theoretical point of view it's look promising, because it will improve opamp PSRR and it will make also make power supply cleaner, because there will be no "coupling contamination" that could harm other opamps.
I thinkg this is important because my output stage is AB class amplifier, that draw only half rectified current.
I also heard that placing multiple small value capacitors (like 100nF) for each power supply IC pins could resonate if connected to same power plane.
To avoid this I will need to add some damping.

So I'm looking for some more extended explaination of my post, because I'm aware about some things but I assume that it's still not enough.

over 7 years ago

0 Ron Michallick over 7 years ago

TI__Guru**** 157456 points

Mateusz,

Decoupling is used to make voltage more consistent. The amount needed is based current requirement that exceeds the ability of the traces (inductance and resistance) to get the current form the next capacitor. Adding separation between between caps (like beads) reduces spreading of noise for one place to another. Every bit helps but there will be a diminishing return as you add more.

I will transfer post to another team so you can get an additional answer. None of the integrated circuits mentioned were in my group.

0 Mateusz Kaluzny17 over 7 years ago in reply to Ron Michallick

Expert 1065 points

Hi Ron.

But adding some isolation via resistor or bread (which is often recommended for an MCU or hi speed OPA etc.) will make a tight local loop, so noise will stay there and will not flow over whole power supply line.

Am I right?

I'm still interested about theory when to use that isolation and maybe someone will be able to provide some information about it.

0 kai klaas69 over 7 years ago in reply to Mateusz Kaluzny17

Guru 140200 points

Hi Mateusz,

the following explanations are very simplified but are good enough to show the sticking points.

Assume you have a perfect power supply with presents 0R source impedance even at the highest frequencies. Further assume that you have an OPAmp circuit which is located about about 20cm away from the power supply. You take a 5mm wide copper trace for the supply line and have a continuous ground plane. You plan to decouple the OPAmp with a 100nF X7R. Everything looks fine but what you have forgotten is the inductance of copper trace of supply line. 20cm long copper trace gives about 200nH. In combination with the 100nF a resonance can be formed at arround 1MHz:

Now assume that you use a decoupling cap which has some ESR, like a 10µ tantal. Then the resonance almost disappears:

And if you finally add some series resistance the resonance can be furtherly suppressed:

One of the advantages of such a 1th/2th order low pass in the supply line is, that the OPAmp no longer draws all the supply current along the 5mm wide copper trace. At high frequencies most of the supply current is drawn from the decoupling capacitor and only a fraction of the supply current is drawn from the power supply being 20cm away. In the following a supply current of 1mA/10kHz sine is simulated:

By chosing the right decoupling capacitances this performance can be furtherly improved.

But you have to be careful, because the supply current of an OPAmp caused by the load is not necessarily of sine shape. This is only true for small load currents when the output stage is still running in class A mode. For higher load currents the output goes into class AB mode and the supply currents become distorted. But in the most cases this is not so much of a problem, because modern OPAmps can show a very good PSRR which suppresses this source of distortion. The pros of this methode must be carefully weighed against the cons and is part of the development.

Kai

0 Mateusz Kaluzny17 over 7 years ago in reply to kai klaas69

Expert 1065 points

Jesus Kai!
I think Ti should hire you or you should be marked at this forum as a living encyclopedia ;)

I'm very glad that you're helped me with those, so I wil ltry to ask another question, OFC I'm asking about your opinion.

So in your opinion, when or where should I go for that kind isolation?
We're not talking about MCU/Hi speed(RF) amplifiers or anything like that, because I'm aware of that those kind of curcuit require special care.

In my case I'm using 4 layer PCB, with continous ground plane (inner layer) and power plane that could be placed inside or as a fat traces also could be routed on outer layers (top and bottom).
I'm using few diffrent capacitors size and values as a decoupling so I can play with those easily, all of them are placed as close as possible (even reservoir).

I will try to reveal more info, so I'm using input section wrapped around quad opa and a buffer to drive load.
I was thinking about isolating buffer or input section via resistors because they're powered from same power supply.
So I found that I can use ferrite bread which will start to work at certain frequency but it will also not affect my headroom.
OFC there is an option with resistors (discussed above) or some people were talking about using additional power supply/regulator to achieve even better isolation.

I was thinking about some isolation, because I'm using input stage and output stage which are power from mutual power supply.
Both stages are placed on the same board with power supply included.
After reading some other posts all over internet I'm aware of that most people are not even thinking about it and it's often recognized as a not important.
But after few hours of reading I also found some posts that were written in diffrent flavour, and they recommend isolation.

As you might see I'm a rather curious person and I'm always looking for more knowledge and proper answers.

0 kai klaas69 over 7 years ago in reply to Mateusz Kaluzny17

Guru 140200 points

Hi Mateusz,

when you discuss the stability of an OPAmp, it's not the signal band of interest (which is the audio band in your case) which counts but the unity gain bandwidth of OPAmp. Even if you amplify only DC voltages, if you do it with a HF-OPamp, this automatically turns your circuit into a HF-application. So, whenever you use a HF-OPAmp you must treat it like a HF-design.

I have learned that even the least amount of isolation is so helpful in increasing stability, that I always use it. In mixed analog digital applications I use for all involved chips power supply filters, even for the digital chips. It's the only way to keep the noisy supply currents within the local decoupling loops and to prevent the ground return currents from flowing back to the voltage regulators and contaminating the whole signal ground wiring. Have a look at my last simulation: It shows that the amount of supply current which flows through R1 (AM1) is only a fraction of the supply current drawn by the OPAmp. But remember, the current AM1 is the same current which flows across the signal ground back to the regulator, called the "ground return current". By this, the power supply filtering minimizes the current "traffic" on signal ground and helps to keep signal ground quiet and free from noise.

Keeping the signal ground as noise free as possible is the most crucial point in an ultra low distortion audio application. This cannot be said too often. Because of this I avoid supply planes and use instead multiple ground planes. As I use RC-filters in the supply lines of OPAmps anyway, I don't need the ultra low impedance the supply plane is offering. But this is my personal taste. I know that other designers are thinking differently.

What supply filtering is good for your circuit cannot by said by a rule of thumb. It depends on your actual circuit. In any case you must follow the decoupling recommendations given by the datasheet. Usually a low ESR 100nF/X7R is recommended. Then you can put a cap with much higher capacitance and much higher ESR in parallel to this cap. Put them close together and place them as close as possible to the supply pins of OPAmp. The filter resistor should also be located near the decoupling caps but need not to sit directly at the supply pins of chip. The value of supply filter resistor mainly depends on the load current.

It's not wise to place an armada of decoupling caps at the supply pins of OPAmp. Too many decoupling caps can hinder each other. If the 100nF caps cannot be placed directly at the supply pins, the armada is worthless. Also, several low ESR caps can show nasty resonances when being paralleled, eroding the decoupling performance for certain frequency bands.

Kai

0 Mateusz Kaluzny17 over 7 years ago in reply to kai klaas69

Expert 1065 points

Your answers are what I was looking for :)
I'm very glad that you had time to answer me properly and deeply about very important things not only in audio but overall.

I know Kai, that armada of decoupling caps are not wise choice.
I'm using 3 caps per line, smallest (in terms of size and value) is placed as close as possible to IC power pin.
Followed then by bigger ones.

It's look like this 0603 (because it's rather small value, so 0603 package is a good choice), 0805 (because some fancy capacitors like film Panasonic or Cornell are in 0805 or even 1206 for 100nF, it's also true for C0G capacitor, they rather big compared to X7R).
Last one is SMD A or B case tantalum capacitor.
To be honest I'm always trying to place every capacitor as close as possible, so 0603 package is often placed within 2-3mm from IC, sometimes even below 2mm if it's possible and will not harm with routing other traces.
It's also very elastic option to manipulate with diffrent types of capacitors (tantalum, film, ceramic) and values.

I need to ask (but I assume that I'm right here anyway).
Even if I use fre buffers paralleled to improve load abilities you will still recommend to isolate each one of those?
I assume that connecting few buffers via one pair of resistors will affect dynamic range even more due to current that will be flowing trough the resistor.

If this technique (isolation) is so good and cheap (due to resistor cost) why it;s not widely used in audio?

0 kai klaas69 over 7 years ago in reply to Mateusz Kaluzny17

Guru 140200 points

Hi Mateusz,

in the following I try to explain a bit how the supply filter should be designed.

First, the OPA1622 shall be discussed. Assume a unity-gain buffer with an output signal of 1Vp sine into a load of 32R. This gives a load current of 31mAp. As this load current is much higher than the idle supply current of OPA1622 we want to assume the worst case, which means a supply current of 31mAp on each supply line looking like a single wave rectified sine. The main distortion of this single wave rectified sine is k2 which is about 1/5 of the supply current. By other words, these 31mAp contains k2 with an amplitude of about 6mAp. A supply filter resistor of 10R would give a voltage drop of 60mVp. The OPA1622 has a fantastic PSRR of about 100dB at 20kHz. So, even without considering the filtering effect of the supply filter cap the 60mVp voltage drop will only appear as an unwanted signal of 0.6µVp. This unwanted signal has to be treated like input reffered noise or offset voltage. So, by the introduce of 10R supply filter resistor you have added an "unwanted" signal of 0.6µVp (k2) to your "wanted" signal of 1Vp. This looks like an additional k2 distortion of 0.00006%.

But now the supply filter cap comes into play. A 10µF tantal shows an impedance of about 1R at 20kHz. In combination with the 10R supply filter resistor this gives a damping of voltage drop of a factor of 10. So, the added k2 decreases to 0.000006%.

For the OPA1604 a similar calculation is valid. It should be kept in mind that the PSRR is not as high as for the OPA1622, though. It's "only" 60dB at 20kHz. But if you take into consideration that the load impedance of a typical OPA1604 circuit is much higher than 32R, then a decreased PSRR can still give brilliant results. If you have a load of 2k driven by the OPA1604 then the PSRR is allowed to be 2000/32=63 times lower (36dB) to yield the same results. So, 60dB + 36dB = 96dB, which is almost 100dB again! Also, the load current of a typical OPA1604 circuit isn't usually much higher than the idle supply current. It can be even smaller, so that the output is not in full class B mode. So, the supply current can be less distorted and can contain much less k2.

In audio circuits I usually take 47µ...100µ aluminium electrolytic caps in parallel with 470n/X7R/0805. I wouldn't add a third 1000p cap. This isn't really needed. For the supply filter resistance I take 10...100R, depending to the load currents. I use a supply filter for each OPAmp chip. Only by this a full isolation can be achieved. For standard tasks I take dual OPAmps. Only for very high gain applications I use single OPAmps. But I seldom use quad OPAmps. Layouting them is often impractical for me.

Kai

0 Mateusz Kaluzny17 over 7 years ago in reply to kai klaas69

Expert 1065 points

Hello Kai.
Sorry for my late reply, but your help is very apprieciated :)
That kind of replies are what I was looking for quite some time.

Taking advantage of the opportunity I've got another question.

Let's imagine a typical power supply with center-tapped winding which is followed then by diodes and then a bulk capacitor.
Question is how much ripple is allowable?

I know that capacitor calculation will be load dependent.
I found a equation :

C = 0.7 * I /(ΔV * F)

I"m also aware of that ripple voltage should be "added" to drop-out voltage.
So if my LDO will work perfectly fine when VIN is 1V higher than VOUT then I should add riple to that 1V to achieve good performance.

And here we go again, I also know that ripple will be capacitance dependent, so if I pick bigger capacitor then I will get less ripple.
But this is somethimes a waste of money and space.
My question is rather simple, how much ripple is allowable (in your option) especially when we're going to feed this "raw" supply in a front of hi performance LDO.
Or maybe that thread is more expanded and I also should consider ripple and a regulator performance to find a best price/performance point?
If there will be some diffrences between examples then let's assume typical circuit with LM317/LM78xx and something better like TPS7A4701 (both hi current).

Another question might be kinda newbie.
But most audio eq. got very big transformers inside (oversided), question is how to pick a proper power ratio?
Let's assume that I will spend all of LM317 current and it will be set to 15-18V.
Transformator windings will be picked to mantain a proper voltage regulator and to not overheat regulator (high VIN compared to VOUT).

0 kai klaas69 over 7 years ago in reply to Mateusz Kaluzny17

Guru 140200 points

Hi Mateusz,

this depends on so many factors that there is no simple answer. What you also must take into consideration is mains voltage fluctuations of at least +/-10%. So, your headroom of the regulator's input voltage must be much wider: 1V for the drop out voltage, 1...3V for the ripple voltage and 2...3V for mains voltage fluctuations.

How you proceed mainly depends on the maximum supply current of your application and what you can spend for the cooling. In many of our products we generate a somewhat higher supply voltage at the storage cap behind the bridge rectifier and use a DC/DC switcher as pre-regulator to create a rather stable voltage for the LDO. This two stage design has many advantages: The switcher has a much higher efficieny and needs much less cooling and the LDO provides a noise free supply voltage without having to deal with high differential input to output voltages. A design could have the following parameters: Generating 25V at the storage cap, using a LM2674 to generate 18.5V and finally taking a LM317/LM7815 for the fine regulation. Furthermore, placing a suited pi-filter between the LM2674 and the LDO.

But again, it depends on the maximum supply current of your application and on what you can spend for the cooling.

Kai

0 Mateusz Kaluzny17 over 7 years ago in reply to kai klaas69

Expert 1065 points

This is indeed very useful information.
I was never thinking about step down converter, but I'm aware that they're exist.

Your option is somehow similiar to placing SMPS like MeanWell IRM series.

Should I consider higher switching freq for a DC/DC converter?

0 kai klaas69 over 7 years ago in reply to Mateusz Kaluzny17

Guru 140200 points

Hi Mateusz,

primary switched-mode power supplies can produce nasty interference and noise. As consequence common mode noise can contaminate your signal ground. The LM2674, on the other hand, produces significantly less noise, provided a proper layout and an inductance with shielded core is used.

To demonstrate what is possible with the LM2674 and the todays filtering components: We have a detector application where the OPAmps are powered by the LM2674 as pre-regulator. The detector signal is in the 100nV range...

Kai

0 Mateusz Kaluzny17 over 7 years ago in reply to kai klaas69

Expert 1065 points

I believe you :)
Do you know something similiar to LM2674 but with higher current output?
I will take something like this in to consideration and I will probably use it later in my desing.
But I need higher current than 0.5A.
It will be nice if there will be an option to build similiar circuit on VEE line aswell.

I've got another question, this one is about decoupling, because I can't find anything useful about it.
How about placing capacitor between VCC and VEE line?
I found some books with this method described and proposed as an additional decoupling to establish low AC impedance between those lines.
But does it make any sense? I don't saw any new design with 3rd capacitor.
Few years ago I was reading some datasheet (TI or Linear) when they described that 3rd capacitor will improve THD (2nd harmonic).
On some forum I also found info that decoupling between power rails should be used when balanced signal is handled.

But I'm pretty sure that those PDF was refering to xDSL/ADSL driver which is kinda diffrent IC (CFB) and it's designed to handle diffrent signal than typical audio IC.

I'm just looking for clarification, maybe you will know something that will be worth trying or considering.

0 kai klaas69 over 7 years ago in reply to Mateusz Kaluzny17

Guru 140200 points

Hi Mateusz,

there's also the LM2675 which can 1A. Today, TI has lots of newer switchers, though.

Yes, the LM2674 can also be used as negative pre-regulator. In this case you must use two separated secondary windings of mains transformer. You fabricate two identical supplies and hook them together with their outputs in such a way, that a positive and a negative supply voltage is created.

Connecting a decoupling cap between VCC and VEE was recommended in some older datasheets. This technique was widely used in the 80s. But at that time PCBs had no solid ground plane and strict star point grounding was used with two seperate grounds, one signal ground and one power ground for the supply voltage. As there was no connection of signal ground and power ground directly at the OPAmp at all, the decoupling performance was heavily impaired and an added decoupling cap between VCC and VEE could surely improve the stability in certain applications.

But with the todays EMC-friendly PCB design with all the decoupling caps referenced to a solid ground plane I don't think that the decoupling cap between VCC and VEE really plays an important role today.

Kai

0 Dennis Goeke over 7 years ago in reply to kai klaas69

TI__Genius 11135 points

Mateusz

We haven't heard back from you so we assume this answered your questions. If not, post another reply below.

Thanks
Dennis

Amplifiers

Amplifiers forum

Question about proper decoupling