opamps and local decoupling of rails, some questions

[Twenty Log]

And looking at more (deeper detailed) capacitor sheets shows interesting things happening beyond 25 degree C and beyond 20kHz, especially if yer using switching power supplies (despite a linear post reg that has no quieting above 1MHz and the noise feeds into and modulates the A/D stage thought the power supply rails)...

That's why the more anal designers start by trading a bit of SMPS efficiency for slower switch slewing, add passive filters + shielding before the linear postreg, consider shunt regs to keep the SMPS noise spectrum load-invariant and synchronize the SMPSes to the converter clocks.

JD 'loop area!' B.

Indeed!  I haven't looked at the design for this A/D box in a while since the custom silicon I am waiting for has been delayed another year...  Sheesh...  I seem to recall borrowing some filtering from Howard Johnson's book, or maybe some "bassterdization" there of, can't remember, but I remember obsessing over the regulator response (and impedance) above 50~70kHz...  Less loop area is conducive for the "path of least inductance" for sure ... Easy to say now, until one gets to PCB layout hehehe....

 

[Kingston]

I just bumped into this thread and thought I'd update some of my experiments a bit, even though it's been years.

No matter how technically sound it might be to sometimes decouple rail to rail with 1uF (or so) film caps in power lines with plenty of rail-to-ground-to-rail decoupling, I've found it a generally bad idea. Some opamps will happily commit suicide in this environment. And some opamps that are resilient enough to stay on might scream oscillation, burning hot when touched.

Can't say I've found any info on opamp datasheets as to why this sometimes happens. It also seems to depend in the PSU. But to be on the safe side, plain rail-to-ground-to-rail decoupling has never caused any problems whatsoever in my projects.

 

[JohnRoberts]

it seems rail to rail capacitance has been well discussed in another thread. With one well know author reportedly endorsing it, and others (like me) suggesting caution, other OK with it in moderation.

Sorry i don't recall the thread topic but probably another PS decoupling discussion

JR

 

[Andy Peters]

I just bumped into this thread and thought I'd update some of my experiments a bit, even though it's been years.

I'll just point out that the 1uF 0603 and 0402 cap is quite the standard thing these days. We tend to use a single 1uF cap per FPGA I/O bank supply and it works well.

-a

 

[Kingston]

I just bumped into this thread and thought I'd update some of my experiments a bit, even though it's been years.

I'll just point out that the 1uF 0603 and 0402 cap is quite the standard thing these days. We tend to use a single 1uF cap per FPGA I/O bank supply and it works well.

-a

FPGA I/O bank, a single ended 3.3V supply, right?

 

[mulletchuck]

But to be on the safe side, plain rail-to-ground-to-rail decoupling has never caused any problems whatsoever in my projects.

I have a question about this.    when i plugged my rev-eng of the Gar2520 into a circuit sim, it said that there was a Capacitor Loop if I included the 0.1uF capacitors that sit between the voltage rails and the ground pin.  Once i remove those from the schematic in the simulator, the simulator works perfectly fine.  sort of...

 

[Samuel Groner]

Some opamps will happily commit suicide in this environment. And some opamps that are resilient enough to stay on might scream oscillation, burning hot when touched.

Can you give detailed info with which opamps and in which circuits this happend?

It also seems to depend in the PSU.

Which is a strong indicator that not the "differential" decoupling is the culptry, but rather than it exposed some PSU flaws which were not apparent without.

Samuel

 

[Kingston]

Some opamps will happily commit suicide in this environment. And some opamps that are resilient enough to stay on might scream oscillation, burning hot when touched.

Can you give detailed info with which opamps and in which circuits this happend?

In a Studiomaster trilogy: http://www.groupdiy.com/index.php?topic=44841.0

It has a mixed environment of rail-to-rail decoupling and sometimes rail-to-ground-to-rail. The PSU itself is pedestrian and very common. Some of those rail-to-rail 100nF caps I replaced with 1uF as an experiment. I had two ADA4898-2 commit suicide in the group bus section (input buffer and fader buffer), burned to the core. Convert everything to rail-to-ground-to-rail decoupling and all opamps were happy.


Another case that I can't unfortunately verify in much detail was when adding 1uF rail-to-rail decoupling to RME HDSP sound card. The PSU was some standard ATX type, coming through the motherboard. It was running happily. All opamps were LME49860.

Insert the card in a whole new system, different ATX PSU, different motherboard, and all of the sudden HDSP is not even booting, unrecognizable to the system. All opamps were running red hot and I measured oscillation there + 10mhz if I recall correctly at the opamps max output level. They were way too hot to even touch. Remove those 1uF I added and the card was working fine again.

I admit these experiments are not as scientific as I'd like, but it's proof enough for me that rail-to-rail decoupling is a touchy thing to implement, and not well fitting to very generic designs. The advantages are negligible compared to a system that already has rail-to-ground-to-rail throughout.

(note, in all above contexts "ground" is referring to PSU ground return path)

 

[JohnRoberts]

Some opamps will happily commit suicide in this environment. And some opamps that are resilient enough to stay on might scream oscillation, burning hot when touched.

Can you give detailed info with which opamps and in which circuits this happend?

It also seems to depend in the PSU.

Which is a strong indicator that not the "differential" decoupling is the culptry, but rather than it exposed some PSU flaws which were not apparent without.

Samuel

+1

Not to rehash that other thread, but opamps are 5 terminal devices, with most (all?) circuits including a 6th ground node. While opamps can operate without a ground node being defined, their loads are generally returning current into some ground, so the relationship (compliance) between the load's ground and power pins, and complete current path back to the power supply can affect circuit operation.

Modern opamp design generally treats the power supplies as black boxes that don't affect opamp performance. Prudent design considers the entire current path (and even voltage interactions with the PS terminals PSRR). I have had this bite my ass more often in very large (console) designs than small relatively simple product designs.  The higher the current the more it matters, and with larger physical systems, ground path or power rail inductance can be more of an issue.

Solid low impedance PS rails should not be affected by adding a modest rail to rail cap. A soft rail can act differently depending on how it is soft. A rail to rail cap could alter the PS terminal voltage if say a HF cap rail to rail, is shunting across two LF caps to ground and wiring impedance.  Note: a rail can be solid at LF and soft at HF based on type of caps used. 

Not trying to overstate the complexity, but it can matter in marginal designs.

JR

PS: Putting on my aluminum foil hat to better understand the tweaker advice to add rail to rail capacitance. If we take as given that opamp PS rails have a real impedance, there will be a terminal voltage component representative of load current delivered.  With standard independent supplies, this terminal voltage will look like a half wave rectified replica of the output (more or less). The opamp's PSRR tells us that this PS terminal voltage will be reflected into the input as an error signal ( a very small error signal). This means we will have two half wave rectified errors, one from each rail showing up in our input as an error term and therefore in the output. But... opamps do not generally have identical PSRR from + and - PS rails so these will not add nicely to make a coherent error signal, but the distortion will be distorted. By coupling the + and - rail strongly to each other, we can make this PS rail voltage more coherent with the output current, so input error is more coherent.  This is akin to managing the angels footsteps on the pin head, but as long as the PS compliance to ground is solid and this terminal voltage small, this system should be stable. I personally prefer to just keep PS rail terminal voltage as low as possible, and use opamps that have good PSRR. Note: the significance of input referred errors like this is amplified by the closed loop gain so more of an issue (perhaps) with very high closed loop gain circuits (like sum bus or high gain stages).

 

[mulletchuck]

But to be on the safe side, plain rail-to-ground-to-rail decoupling has never caused any problems whatsoever in my projects.

I have a question about this.    when i plugged my rev-eng of the Gar2520 into a circuit sim, it said that there was a Capacitor Loop if I included the 0.1uF capacitors that sit between the voltage rails and the ground pin.  Once i remove those from the schematic in the simulator, the simulator works perfectly fine.  sort of...

bump

 

[Andy Peters]

I just bumped into this thread and thought I'd update some of my experiments a bit, even though it's been years.

I'll just point out that the 1uF 0603 and 0402 cap is quite the standard thing these days. We tend to use a single 1uF cap per FPGA I/O bank supply and it works well.

-a

FPGA I/O bank, a single ended 3.3V supply, right?

The supplies are all single-ended, by which I guess you mean not bipolar. We use bank supplies of 3.3V, 2.5V, 1.8V, whatever is required by whatever we're interfacing with. We do lots of LVDS which can be either 3.3V or 2.5V. We directly interface to a fiber transceiver module which uses CML (basically PECL) at 3.3V with AC coupling.

Don't forget you need decoupling on your FPGA core supply, which is 1.2V in many cases (Spartan 3A, 6, Virtex 4). An Actel part I used needs a 1.5V core supply.

There are way too many signaling standards. It gets messy.

-a

 

[ricardo]

I was a big fan on rail to rail decoupling in da old days ... I didn't want to sully my nice clean earths with crap from the supply rails.  Then I did a very comprehensive circuit analysis simulation of Power Amplifier performance which showed that rail to ground decoupling was essential for unconditional stability under all operating conditions.

Going back to OPAs, I found going back to rail to ground from rail to rail made everything much more predictable and explained some instances of having to take extra measures to ensure stability and good thd.  OPAs are small power amplifiers.

An important finding was that you needed a bit of resistance in the decoupling capacitance too so small aluminium electrolytics are just the job.  I like to use 1u or 10u  near each OPA, bigger for better HF THD but small size is important too.  100n films or ceramics aren't usually required but don't make things worse.  So add what you like as long as you don't leave out the electrolytics.

But when you decouple to to earth/ground, you have to be careful that the crap on the supplies isn't coupled to your nice clean earths.  I always define 2 separate nets; clean and dirty.  Clean is used only for feedback & signal.  These are only connected at the output of the 'box' and this is also the connection to the main PSU capacitor.  ie star earthing.

This is complicated by some new chips that MUST have a 'clean' earth pin connected to something directly 'coupled' to the rails.  eg THAT 1646
http://www.proaudiodesignforum.com/forum/php/viewtopic.php?f=6&t=464&p=5177&hilit=decoupling#p5177 but these are rare.

There are many disbelievers of star earths from the digital gurus who like earth planes.  But probably the most useful advice is JR's ..
Earth/ground is not a fixed voltage but has resistance & inductance (even earth planes) so any currents flowing in it will induce voltages.

That's not to say, earth planes can't give good results but they have to be given the same careful consideration as a star earth system ie you have to know the exact path your currents are taking.

So my $0.02 ...

• decouple both rails to earth near each OPA
• use small Aluminium electrolytics
• think where your decoupled currents are going

 

[mulletchuck]

So, no one is gonna answer my question about the "Open Capacitor Loop" in the circuit simulator when I include the 0.1uF caps between the +/- rails and the ground?

But to be on the safe side, plain rail-to-ground-to-rail decoupling has never caused any problems whatsoever in my projects.

I have a question about this.    when i plugged my rev-eng of the Gar2520 into a circuit sim, it said that there was a Capacitor Loop if I included the 0.1uF capacitors that sit between the voltage rails and the ground pin.  Once i remove those from the schematic in the simulator, the simulator works perfectly fine.  sort of...

 

[Kingston]

So, no one is gonna answer my question about the "Open Capacitor Loop" in the circuit simulator when I include the 0.1uF caps between the +/- rails and the ground?

Perhaps you should make a new thread for your off topic thing. You will get more eyes that actually know that particular simulator you use and its bugs and workarounds.

You didn't even say what software you are using.

 

[tv]

rail-to-rail is good if your supply is good to start with and groundloops are benign. It lets you to decouple "rails" to dirty-gnd and the opamp itself, "r2r".

if you look at the "decoupling" situation from a different prespective, inductance is split between the two rails, and the rest is vaguely similar.

http://en.wikipedia.org/wiki/File:Cc_colp2.svg

@ricardo: any amp will need a C as transient energy storage between the "rail" and the common, "gnd" - that's likely the only way to deliver the energy to the load.

IOW, in a single-ended "opamp" amp, you need only one ceramic to do the wole thing - and to add the insult to the injury, the damn thing has a potential to sound sweeter. Unfair, I know..

 

[keefaz]

I saw in an analog app note, that there would be an advantage in decoupling rail to rail a double opamp, if the second opamp is in inverting mode, like in the picture

http://www.google.com/url?sa=t&rct=j&q=decouplinf%20of%20negative%20supply%20virtual%20ground&source=web&cd=2&ved=0CFUQFjAB&url=http%3A%2F%2Fwww.analog.com%2Fstatic%2Fimported-files%2Fapplication_notes%2FAN-202.pdf&ei=KWAiULPpEYnH0QW38oCADQ&usg=AFQjCNEuAZ6KQKcJQa01OSCAbt1ibkmG3A&cad=rja

 

[JohnRoberts]

I killed a few too many brain cells following my own advice to "follow the current" inside consoles. I looked seriously at local current flow within a console module, and currents sent feet away to various bus masters and whatever.  I ended up abandoning trying to come up with a neat complete cancellation while large net imbalances can matter. Note: a balanced dual polarity bus structure could net out bus currents sent off the module but at a prohibitive cost for that relatively modest benefit. Perhaps if the inputs were 20' feet away.  8)

This actually is more important inside power amplifiers where circulating currents are amperes not milliamperes. In fact power amps actually deliver slightly more power in bridge mode when the outputs are opposite polarity so drawing from both supplies at the same time, not one or the other alternately,

...as this veers further off topic.

JR.

 

[tv]

...as this veers further off topic.

tsk, tsk - not so fast!
I think that any "simple" balanced line driver (i.e. made with a dual-opamp) steps into this territory as well.

 

[ricardo]

I don't think there is ANY practical case where 2x10u electrolytics decoupling both rails to ground near the OPA gives worse results than ANY rail to rail system.  (provided you don't put sewage on your nice clean earths  However, there are MANY cases where rail to rail may have oscillation, poor HF & other THD etc as Kingston and other people including myself have found.

If anyone has different experiences please do tell.  Real life stuff please rather than pontificating.
PS  There is a case for BOTH rail to ground AND rail to rail capacitors where board space is limited in a power amplifier but even this is dubious.

I'll second JR's exhortation to watch current loops in your supplies.  This is a source of THD in power amplifiers, first pointed out by Baxandall (of course) and later formalised by Cherry, Self & others.

This is on topic cos the local decoupling closes your current loop

In fact power amps actually deliver slightly more power in bridge mode when the outputs are opposite polarity so drawing from both supplies at the same time, not one or the other alternately,

Bob Carver made use of this in some of his stereo amplifiers which were inverting on only one channel and corrected at the speaker terminals ...  or maybe he just wanted to confuse the unwashed masses  ;D

 

[Samuel Groner]

However, there are MANY cases where rail to rail may have oscillation, poor HF & other THD etc. as Kingston and other people including myself have found.

If anyone has different experiences please do tell.  Real life stuff please rather than pontificating.

I use rail-to-rail decoupling all the time, with excellent results. Typically I have 2x 100 nF caps to ground, and 10-100 uF rail-to-rail for each chip. Very important is the provision for enough damping of the decoupling system--either by some small resistance in series with the rail, or the use of low-Q (electrolytic) capacitors. As far as I can tell from the limited info given about the systems where rail-to-rail decoupling has failed, insufficient damping must have been the cause. In a power supply system where no provisions for damping have been taken, the addition of a high-Q (film, ceramic) capacitor can easily provoke instability by shifting the resonance frequency into a region where the opamps have less PSRR. However, this is not the fault of rail-to-rail decoupling per se, but the omission of damping. I have implemented several GHz opamps and composite opamp topologies (two opamps within one global loop, not for the faint of heart) where I too used rail-to-rail coupling, and this stuff was stable even on the first prototype board.

Key to understanding the advantages of rail-to-rail decoupling (or the general problem of decoupling for low distortion) is the separation of the signal current loop and the current loop of the harmonics which are generated because of the class B nature of typical output stages (and occur, to a surprising amount, even with class A designs). The signal current loop flows from the supply through the output stage, through the load, and through ground (let's ignore the case where the load is connected to a virtual ground for simplicity) back to the supply. The loop of the class B harmonics however never passes through the load (as the load hopefully get a clean signal delivered), but travels from one supply through the entire output stage right to the other supply rail.

An ideal decoupling system could completely isolate these loops, and keep both as short as possible for minimal crosstalk/distortion. The signal current loop would be bypassed to the load, and the class B harmonics loop to the other supply. Unfortunately it is not easy to completely separate these paths, and usually some compromise must be made. IME, the approach described above does pretty well usually.

Bottom line is that decoupling must be engineered, not left to a few "rule-of-thumbs" and guesses. No approach will be optimum in any case. As time permits I hope to write more on this in an essay, but I suspect this will have to wait another year or two...

Samuel