AD797 shows 110 MHz GBP and 20V/us slew rate, I think more than sufficient in regulated power supply.
The same about OPA350, 38 Mhz GBP and 22V/us slew rate.
BTW I think in a tipical feedback shunt section a very fast device is not an advantage, but exactly the contrary.
Not a bad idea to use an high hfe as pass device, but I choose D44H11 for its linearity and it has only 60 hfe. I should use a darlington configuration, but I'm not sure that's the right way in a series regulator. I read contrasting opinions about darlington in power supply regulation.
What do you think about?
Andrea
So, why on earth you need "GBW: 75 MHz" from an OpAmp in a power supply ?
Regulated linear power supply is a very badly abused amplifier - because its output is loaded by huge capacitance of electrolytic capacitors..
So, the whole thing has very limited GBW, and hence using OpAmps with good GBW is nonsense in this case.
So what ?
Bandwidth is determined by the electrolytic capacitors and power transistor through which they are charged.
It's like you have two cascaded LPFs.
For example, the first of them (the OpAmp) has, say, cutoff frequency of 1 MHz. The second of them (power transistor plus output electrolytic capacitor) has cutoff frequency, say, of 1 Hz.
You've decided to use a faster OpAmp which provides cutoff frequency of 10 MHz instead of 1 MHz - so what ?
I don't see any reason to not try to have the highest bandwidth and therefore the lowest impedance that you can get, at the power supply's output. After all, maybe the rails will be extremely short (or even planar) so it could do some good.
(Sorry if no one is interested, or if I am missing something. I will give it one last try.)
But, again, what actually matters is the power rails' impedance needed at, and seen from, the power supply connections RIGHT AT the load, i.e. directly across an active device's power pins.
For example, if you have an active load device and it can slew something in 1 μs, that implies an equivalent frequency of over 300 kHz. So six inches of typical wire or PCB trace, with 10 nH of inductance per inch, is going to add 0.060 μH with an impedance magnitude of up to roughly 0.12 Ohms, i.e. 120 mΩ, no matter how low the PSU's output impedance is. And that's for a relatively-slow type of device.
Looking at another part of the picture, if your load were able to need to quickly draw 2.5 A (a large digital chip, for example) and you wanted the 5V supply rail to stay within 1% of its intended voltage, then the impedance seen by the device's supply pins would need to stay at or below .05/2.5 = 20 mΩ, at least up to the frequency that is equivalent to the rise times, which, for a digital device with 2 ns rise times, could be around 160 MHz.
Not only would ANY power supply not be able to supply the needed current through the power-rail inductance of a rail length that was more than a small fraction of an inch without massive voltage spikes on the rail, it would even be difficult just to figure out how to connect the needed decoupling capacitance to have a low-enough overall inductance, even if using proper power and ground planes.
In any case, the voltage noise and fluctuations on the supply rail, due to changing load currents flowing through rail and/or decoupling network inductances, will probably be orders of magnitude greater than anything from the PSU.
Even if you don't think that I'm exaggerating, I suggest reading Chapter 11 of Henry W. Ott's latest edition of "Electromagnetic Compatibility Engineering" (the chapter on "Digital Circuit Power Distribution), and whatever one can find by Bruce Achambeault. Small bits are also on the web, from both of them. [It is also possible (and necessary) to extrapolate their analysis and design methods down into the audio frequency range and onto non-planar topologies.]
Last edited:
You're right, in fact i choose first LT1028 and now AD797 for its low noise, not for GBW.
That's why remote load sensing makes for a better sounding regulator. It helps if you decouple the sense circuitry with a pair of low value resistors.
Note that the Adcom GFP-565 uses a pair of brass buss lines for the poz and neg rails.
(Sorry about inflicting another extremely-long post on this thread.)
A remote sensing power supply is nice but it wouldn't help much in mitigating the rail inductance when powering a digital load (or a dynamic analog one, for that matter): By the time the sensing happens it's already too late.
That does nicely illustrate my (proposed) point, though:
NOTHING we could put at the PSU end of the power rails could ever help very much in supplying accurate transient currents when they are demanded by an active load, except for keeping the decoupling caps charged.
(Well, there is probably a way to do it. But it would involve adding a known delay time between the input and the reproduced output, so we could use knowledge of the input and the future output to be able to anticipate and pre-calculate the transient current demands, and then still have enough time to generate and provide them, accurately and precisely in both time and amplitude, directly at the point of load, without causing other problems. But then the relevant new parts of the PSU wouldn't be very much like a simple regulator, any more. So that's all irrelevant to this particular discussion. But it might be worth pursuing for an all-digital system.)
So here is a (proposed) major point:
Decoupling, and specifically the inductance between the decoupling capacitors and an active load's power/gnd pins, is at least as important as the power supply.
Maybe this should be in a separate thread. But, on the other hand, if the power supplies being discussed are going to be implemented and used, then addressing these issues could significantly improve the resulting systems.
Please allow me to rant, for a short time. It will help to illustrate my concerns, which should help the high-fidelity power-distribution efforts of anyone who learns to share them.
WHY does everyone seem to take such a "voltage-centric" view of power supplies, anyway? The SIGNAL PATH in a power amplifier is the CURRENT through the PSU, which then runs through the active power devices, and through the speakers. THAT is the signal that we actually HEAR.
The PSU voltage is relatively boring, by comparison. It is held as constant as possible so that there will be a mostly-linear CURRENT response through the controllable-resistance CURRENT VALVES that we call transistors (although feedback will compensate for most of the unexpected variations in the current flow that are due to rail-voltage fluctuations, anyway, making rail voltage changes and noise even less important).
The real "action", the main high-precision dynamic action that's occurring, in the PSU of an audio amplifier, is the supplying of high-fidelity CURRENT, in exact proportion to the signal being reproduced. Make no mistake about that. Both BJT and FET types of transistors are CURRENT VALVES. One is current-controlled and one is voltage-controlled. But they are both just controllable-resistance current valves.
If you think about it, in an audio power amplifier, where else could such a high-power, high-fidelity "force" be coming from? The inputs and their small-signal chain have no power to speak of. And the power supply's voltages are supposed to be nearly unchanging. The only thing left that could be doing the job is a massive, variable, high-fidelity flow of current, straight from the power supply.
But what does it take, to get "high fidelity", anyway? What could make one amplifier so much better-sounding than many others? It has become quite clear that many competent amplifiers exist that can reproduce any audio-frequency steady-state "pure sine wave" tone with extremely-low distortion, much lower than our ears can detect. So, their reproduction of an unchanging pure tone is, in a practical sense, "perfect", and we should detect virtually no difference between such amplifiers when they are reproducing a steady-state tone. And yet, when playing music, one amplifier might sound much better than many of the others.
Therefore, the difference between "merely competent" amplifiers and exquisite-sounding high-fidelity amplifiers MUST lie in the accuracy of their transient reproduction. (Those who have studied Signal and System Theory already know that everything an amplifier can possibly do is included in either its "steady-state response" to a sine or its "transient response".)
And that could make sense, intuitively, too: Without accurate transient reproduction, the conceptual Fourier components of a rising edge, for example, would have incorrect phase angles and amplitudes, not only distorting the amplitude profile of the edge and what's nearby but also distorting the timing cues that create a good soundstage image.
That is a major part of why the impedance seen by the power devices' power and ground pins when looking upstream, right at the pins, is VERY IMPORTANT. When the power transistors are suddenly commanded to drastically lower their resistances, thereby "opening the floodgates", a massive current needs to suddenly flow, but with absoultely-accurate timing and amplitude.
With a large-enough ideal decoupling capacitor and zero inductance at the power rail end of a power output transistor, for example, the rail voltage would only need to go down by a very tiny amount to make the capacitor provide a massive current. But a few too many tens of nanoHenries of inductance can ruin everything. Not having enough capacitance is almost as bad. But even with more-than-enough capacitance, a little too much inductance still ruins the potential for accuracy of the resulting current flow. Not only that, it also can create relatively-large voltage spikes because of the current that does eventually manage to flow through it. (Remember, also, that the amplitude of the voltage across an inductance depends only on how fast the current is changing, and NOT on how MUCH current flows.)
I guess my point is also that a high-performance power supply might often be necessary but is basically always not sufficient, by itself, if the system handles dynamic-enough signals.
Additionally, making sure that there is low-enough inductance but also sufficent decoupling capacitance presented to a dynamic active load's power/ground pins is very-often not trivial! (Basically, EVERY power amplifier PCB layout that I have seen on diyaudio has failed to implement proper decoupling of the active power devices, when it probably would have been possible by changing the layout and the capacitors used.)
It's a little bit like the idea of putting a high-performance race car engine into a stock automobile. To really make use of it and gain all of the potential benefits you then must also upgrade everything else in the drive train, all the way from the engine to the wheels, as well as the suspension, and possibly other subsystems such as brakes, steering, seating/harnesses, et al.
Similarly, using a high-performance power supply might involve a significant amount of wasted effort, or at least the loss of some potential performance improvements, if the decoupling networks are not also redesigned and upgraded.
Anyway, I hope that I have convinced you that the inductance between the decoupling capacitors and an active load's power and ground pins is at least as important as the power supply.
IF SO, there are two basic paths that can be taken. One, maybe mostly for electrical-engineer types (or maybe anyone who can handle a calculator and read data sheets), is to remember to determine the worst-case transient conditions and calculate the maximum impedance that should be seen by an active load's power/gnd pins and the frequency up to which that impedance must not be exceeded, and calculate the minimum decoupling capacitance needed and the maximum tolerable inductance in the decoupling network (which determines the allowable maximum round-trip connection length for the decoupling capacitance). Then determine whether or not those requirements dictate the use of multiple parallel decoupling caps instead of a single cap, and calculate the required maximum round-trip conductor lengths that would enable meeting the inductance spec with the correct number of appropriately-sized capacitors. And THEN try to find a way to physically fit the required number of capacitors and their connections into the real circuit.
To see how to do the simple calculations needed, look at all of posts number 5 and 8, at least, and also follow all of the links given in post 5, in the thread at the following link:
http://www.diyaudio.com/forums/power-supplies/208579-30vdc-10a-psu.html#post2942537
"A second way" is available for those who don't want to do or understand any math or physics and just want to go straight to the physical design and layout of the circuit. This only really involves finding ways to get decoupling capacitors as close as possible to device power and ground pins, while also knowing that using multiple lower-value capacitors with separate parallel connections all the way to the device pins can give much lower total inductance than using one larger capacitor of equivalent value. See the link above for a little more qualitative detail about what to do, and when to do it.
If you get really ambitious, or if you just like "overkill" for its own sake, you might want to also try extending the concept of using multiple parallel decoupling caps with parallel connections that stay separate all the way to the device pins (for lowest inductance), by extending it all the way back to the power supply! i.e. Use multiple separate parallel conductors for EACH voltage and ground rail, with a smoothing cap across each voltage/ground rail conductor pair on the PSU end and a decoupling cap across each pair at the load end.
Theoretically, that topology would enable us to make the impedance seen by the load arbitrarily low. i.e. We could make it as low as we wanted, just by adding more separate parallel power/ground rail conductors and capacitors. (In reality, of course, we might run out of space.)
First, one also needs to realize that anywhere there might be a sudden demand for current, there should probably be decoupling capacitance.
(I am truly sorry for having blathered-on for so long, again.)
Cheers,
Tom
Last edited:
The reason that Gootee and others have to keep repeating this decoupling/inductance message is that most Members here refuse to do the research.
The information is available. One must take responsibility for informing oneself to allow decisions to be taken.
Instead many just copy what is seen stated on the internet, without trying to sort the wheat from the chaff.
The information is available. One must take responsibility for informing oneself to allow decisions to be taken.
Instead many just copy what is seen stated on the internet, without trying to sort the wheat from the chaff.
Andrew, Tom,
your suggestions are welcome and useful, I'm taking and I'll take them in the right consideration.
BTW I'm trying to design and build a good enough power supply regulator, useful for digital and analog devices, not the best PSU.
SO, first, I have to make some choises about circuit topology and devices to use.
In a previous post you say that it is not worth wasting time to discuss about op map to choise. I think instead that it is important too, because their the electrical characteristics greatly affect the result.
I'm not used to get ideas from thousands of circuits on the internet, usually I take informations from manufacturers documentation, datasheet and application notes.
Once defined a basic circuit, I'll build a wired prototype, I'll test the results and I'll experiments on it. Then I'll build the PCB, and at that moment I'll take the maximum care to every path of the PCB.
Maybe I'm wrong.
Let me share one of my experience, also if that's not inherent this thread.
20 years ago I designed a power amplifier for a brand that marketed the product (original pen made schematic attached). The very first listen left everyone perplexed, so we started to experiment on it. Finally, simply adding 2 capacitors (I remember 1 x 2.2 uF Wima MKP2 and 1 x 100 nF Wima MKS4) directly on the drain of parallel mosfets and returning them directly to PSU filter capacitors, the sound changed drastically. Better transient response, bass more controlled, high frequencies more defined and smoothed, music scene more realistic.
Simply adding 2 capacitors!
Now I'm still reviewing the regulator circuit. Some people, in another forum, have rightly pointed out that I'm using a too noisy voltage reference compared to the quiet AD797 used as feedback control device. So, I'm trying to design a quieter voltage reference. Also I'm seriously considering to eliminate the shunt stage.
Thanks for spend time to share your opinions
Andrea
<grin> I would have. But I had already implied that I wouldn't do it again, in an earlier post.
People talk about inductance/chokes apparently forgetting that real life chokes:
1) have parasitic capacitance;
2) tend to have mutual inductance with other chokes;
2) suck in EMI.
So, installing chokes may sometimes worsen the situation.
...
I think good HW utilizes distributed regulation model, i.e. regulators are put near consumers.
...
Regarding chokes again - datasheets often list their self-resonant frequency - which is present because of parasitic capacitance. Above the self-resonant frequency (or whatever it's called) the choke looks like capacitive load.
...
I strongly recommend to design from discrete components and assemble a switching power supply with switching frequency of, say 5KHz. Not because it is going to be good, but because it will allow to step on all the rakes and watch the results with oscilloscope. I.e. to design and assemble it for learning - not for practical use.
1) have parasitic capacitance;
2) tend to have mutual inductance with other chokes;
2) suck in EMI.
So, installing chokes may sometimes worsen the situation.
...
I think good HW utilizes distributed regulation model, i.e. regulators are put near consumers.
...
Regarding chokes again - datasheets often list their self-resonant frequency - which is present because of parasitic capacitance. Above the self-resonant frequency (or whatever it's called) the choke looks like capacitive load.
...
I strongly recommend to design from discrete components and assemble a switching power supply with switching frequency of, say 5KHz. Not because it is going to be good, but because it will allow to step on all the rakes and watch the results with oscilloscope. I.e. to design and assemble it for learning - not for practical use.
I have searched through all five of my posts.
I cannot see where I said that.
read again.
I have said that Gootee's information is relevant to the good operation of the circuit.
I have said that Members ignore Gootee's information.
I have said that Members seem to unaware of the effect of appropriate decoupling.
I have NOT said that
Is this an english language problem, or are you just creating trouble?
....."They are still worried about the opamps".....
For me, it means that opamps are not a problem to discuss, they are not relevant for this project.
BTW if I misunderstand, please excuse me for the mistake.
This is a forum to discuss about audio, not for create trouble.
And again all your comments and suggestions are always useful for this project.
Andrea
That is certainly not what is meant.
I could see that you and others were ignoring good advice and directing all your efforts in one direction. Looking at the opamp exclusively is not the way to "solve" the PSU supply problem. You have to look at the system. That means examining both ends and the interconnection, the PSU, the cables/traces and the client circuit. You must make both ends "work" however close together, or far apart, they are.
I could see that you and others were ignoring good advice and directing all your efforts in one direction. Looking at the opamp exclusively is not the way to "solve" the PSU supply problem. You have to look at the system. That means examining both ends and the interconnection, the PSU, the cables/traces and the client circuit. You must make both ends "work" however close together, or far apart, they are.
Andrew,
I appreciate that you have tried to help steer their interest toward the issues I have raised.
I do believe that Andrea understands, based on the product capacitor-mod example given. If not, it will probably become obvious, later, anyway.
And while I would have liked some feedback or debate or collaboration, because I am still forming the concepts and knowledge-base in my own mind and there is still some uncertainty on my part, that is not a necessity, and cannot be required in any case. I am satisfied-enough, just because the material was presented and is now on record in this context, for future readers (and present ones).
And in one sense I was intruding, with possibly-"inconvenient" information, and, while people like you and I might nearly always jump at the chance of finding new pieces of the puzzle, it's difficult for me to want to blame anyone else for not suddenly showing an avid interest in a freshly-opened can of worms when they are still enjoying being fixated on the one they've already got. <smile>
No worries,
Tom
I appreciate that you have tried to help steer their interest toward the issues I have raised.
I do believe that Andrea understands, based on the product capacitor-mod example given. If not, it will probably become obvious, later, anyway.
And while I would have liked some feedback or debate or collaboration, because I am still forming the concepts and knowledge-base in my own mind and there is still some uncertainty on my part, that is not a necessity, and cannot be required in any case. I am satisfied-enough, just because the material was presented and is now on record in this context, for future readers (and present ones).
And in one sense I was intruding, with possibly-"inconvenient" information, and, while people like you and I might nearly always jump at the chance of finding new pieces of the puzzle, it's difficult for me to want to blame anyone else for not suddenly showing an avid interest in a freshly-opened can of worms when they are still enjoying being fixated on the one they've already got. <smile>
No worries,
Tom