Jan, why are you doing it the hard way?
The problem becomes much simpler if you divide and conquer; a multiloop amplifier topology separates ps load regulation and noise requirements.
A fast output amp can be supplied from a garden variety 3 term reg circuit – the input op amp open loop gain adds to the output amp PSRR eliminating the need for heroic measures to reduce ps impedance and noise simultaneously.
The input op amp is only driving Mohm||pF (at the output op amp’s input) so has negligible ps load regulation requirement allowing RC filtering of the 3-term reg ouput with as much as 100 ohm series R, a ordinary Zener shunt reg between the 3-term reg and the RC filter puts all ps effects way below op amp input noise for an input op amp with decent low frequency PSRR. Alternatively the input op amp could be supplied from a Cap multiplier (which is powered from the 3-term reg or intermediate Zener) for smaller component size and cost – many options open up for low noise input op amp power with virtually constant (and low) current load.
Separating the output stage from the input amp also increases the chances of approaching the high (-ly improbable ) calculated PSRR numbers in an actual layout that addresses supply/ground/load mutual impedances.
The output op amp also provides added loop gain which can be utilized for distortion/sensitivity reduction in a variety of ways ( plenty of material for another thread on that subject though )
The problem becomes much simpler if you divide and conquer; a multiloop amplifier topology separates ps load regulation and noise requirements.
A fast output amp can be supplied from a garden variety 3 term reg circuit – the input op amp open loop gain adds to the output amp PSRR eliminating the need for heroic measures to reduce ps impedance and noise simultaneously.
The input op amp is only driving Mohm||pF (at the output op amp’s input) so has negligible ps load regulation requirement allowing RC filtering of the 3-term reg ouput with as much as 100 ohm series R, a ordinary Zener shunt reg between the 3-term reg and the RC filter puts all ps effects way below op amp input noise for an input op amp with decent low frequency PSRR. Alternatively the input op amp could be supplied from a Cap multiplier (which is powered from the 3-term reg or intermediate Zener) for smaller component size and cost – many options open up for low noise input op amp power with virtually constant (and low) current load.
Separating the output stage from the input amp also increases the chances of approaching the high (-ly improbable ) calculated PSRR numbers in an actual layout that addresses supply/ground/load mutual impedances.
The output op amp also provides added loop gain which can be utilized for distortion/sensitivity reduction in a variety of ways ( plenty of material for another thread on that subject though )
Jan, why are you doing it the hard way?
You know I would agree with a lot of this if the op amp based regulators had not been tweaked by hundreds of people with sonic benefits from extremely low noise and high PSRR. You are talking about a circuit that has refined for over a couple of decades. These attention to minute details pay sonic dividends. Do some reading on the forum and you will see what I mean. I think you may be at about page 20 of a 400 page book on this one. Throw a schematic here and I bet we can find a dozen improvements in about 30 minutes. I will be nice because you seem to mean well but are maybe a bit unprepared to wade into this one without doing some studying I fear. References anyone?
You know I would agree with a lot of this if the op amp based regulators had not been tweaked by hundreds of people with sonic benefits from extremely low noise and high PSRR. You are talking about a circuit that has refined for over a couple of decades. These attention to minute details pay sonic dividends. Do some reading on the forum and you will see what I mean. I think you may be at about page 20 of a 400 page book on this one. Throw a schematic here and I bet we can find a dozen improvements in about 30 minutes. I will be nice because you seem to mean well but are maybe a bit unprepared to wade into this one without doing some studying I fear. References anyone?
the hard one
Hi jcx,
I see your point on the multi-feedback loop topology, but I have my reasons not to use it this time. Trust me, and all will be revealed in the fullness of time (I like that expression: the fullness of time).
And yes, the options you quote all have their pro's and con's, and if this pack gets their teeth into it will undoubly tear much of it apart.
What intrigues me is your statement that the open loop gain of the input opamp improves the PSRR of the output stage, if I read it correctly. If you mean that because of the light load on the input opamp it needs a much less sophisticated supply, I agree. But if you mean it literally, you lost me, please explain.
Jan Didden
jcx said:
Hi jcx,
I see your point on the multi-feedback loop topology, but I have my reasons not to use it this time. Trust me, and all will be revealed in the fullness of time (I like that expression: the fullness of time).
And yes, the options you quote all have their pro's and con's, and if this pack gets their teeth into it will undoubly tear much of it apart.
What intrigues me is your statement that the open loop gain of the input opamp improves the PSRR of the output stage, if I read it correctly. If you mean that because of the light load on the input opamp it needs a much less sophisticated supply, I agree. But if you mean it literally, you lost me, please explain.
Jan Didden
Jan (is it noisy in here or is it just me?)
Consider the PSRR of the output amp, it defines the equivalent disturbance voltage at the input to the output amp due to power supply voltage fluctuations at the output op amp ps pins – call the output op amp _0, input op amp _1
PSD_0 = power supply disturbance at the output op amp ps pins (dBV)
input voltage disturbance(d in MathCad below): PSDin_0 = PSD_0 – PSRR_0
To refer this voltage to the input op amp’s input terminals we divide by the gain of the input op amp = A_1 (input op amp open loop gain)
PSDin_1 = [ PSDin_0 - A_1 ] + [ PSD_1 – PSRR_1 ]
PSDin_1 = [ PSD_0 – ( PSRR_0 + A_1 ) ] + [ PSD_1 – PSRR_1]
So I think it is fair to say that when considering the overall PSRR of the composite amplifier that the gain of the input op amp (in dB) adds to the PSRR of the output op amp, of course the last bracketed term above is the PSRR of the input op amp rejecting the PSD at its ps pins from the (hopefully quieter) input op amp supply which isn’t improved by the composite topology
At AC ps ripple frequencies the input op amp (OPA227, AD8610) gain can easily be ~80-100 db, which added to a fast output op amp’s relatively poor 70 dB PSRR (LM6171 or AD811 for instance) gives insane numbers for line harmonic rejection at the output op amp ps pins, even at the high end of the audio spectrum the input op amp adds 50-60 dB to the output op amp PSRR (still at ~ 70 dB) allowing output ps Zout of a few tens Ohms to put load regulation artifacts down >140 dB with your example 600 ohm load (not that I recommend the output ps Zout be as large as a few Ohms)
Attached, a little MathCad on multiloop
Consider the PSRR of the output amp, it defines the equivalent disturbance voltage at the input to the output amp due to power supply voltage fluctuations at the output op amp ps pins – call the output op amp _0, input op amp _1
PSD_0 = power supply disturbance at the output op amp ps pins (dBV)
input voltage disturbance(d in MathCad below): PSDin_0 = PSD_0 – PSRR_0
To refer this voltage to the input op amp’s input terminals we divide by the gain of the input op amp = A_1 (input op amp open loop gain)
PSDin_1 = [ PSDin_0 - A_1 ] + [ PSD_1 – PSRR_1 ]
PSDin_1 = [ PSD_0 – ( PSRR_0 + A_1 ) ] + [ PSD_1 – PSRR_1]
So I think it is fair to say that when considering the overall PSRR of the composite amplifier that the gain of the input op amp (in dB) adds to the PSRR of the output op amp, of course the last bracketed term above is the PSRR of the input op amp rejecting the PSD at its ps pins from the (hopefully quieter) input op amp supply which isn’t improved by the composite topology
At AC ps ripple frequencies the input op amp (OPA227, AD8610) gain can easily be ~80-100 db, which added to a fast output op amp’s relatively poor 70 dB PSRR (LM6171 or AD811 for instance) gives insane numbers for line harmonic rejection at the output op amp ps pins, even at the high end of the audio spectrum the input op amp adds 50-60 dB to the output op amp PSRR (still at ~ 70 dB) allowing output ps Zout of a few tens Ohms to put load regulation artifacts down >140 dB with your example 600 ohm load (not that I recommend the output ps Zout be as large as a few Ohms)
Attached, a little MathCad on multiloop
Attachments
The off topic part of this thread has been moved to the Off Topic section of the forum: http://www.diyaudio.com/forums/showthread.php?s=&threadid=15565&perpage=15&pagenumber=3
(is it noisy in here or is it just me?)
"And yes, the options you quote all have their pro's and con's, and if this pack gets their teeth into it will undoubly tear much of it apart.'
Jan behave yourself! I offer only a few gentle nibbles, like a puppy playing, to someone with this level of design and analytical skill who is quite qualified to hand me my hat in this exchange. He has thought this out and I love the intellectual stimulation. I say this in deference without the least intension of sarcasm. He has made a pretty convincing case in theory.
Jbx, yes it is noisy and is mostly me. Your math models and rational appear to be faultless. If I might be permitted to summerize in a little simpler layman's terms (something I very seldom do, although often asked to). The introduction of the composite amp allows the U1 op amp (with greater PSRR) to increase the PSRR of U0, by reducing the power supply error voltage at the output of U0 by the amount of loop feedback (open loop gain divided by closed loop gain) of U1. The other result, the greatly reduced loading for U1 by driving UO's high input impedance instead of the load is maybe even more beneficial. The less current U1 has to supply the larger the power supply impedance increase from RC filter can be for a given signal induced noise voltage on the supply terminals for U1. In all, a very symbiotic arrangement for both op amps, in theory.
A little history on the development of the Jung Sulzer regulator. Original had an RC filter for the increase of the PSRR for op amp. The next article discussed to use of a preregulator and a two transistor darlington to increase output current capacity.
Some suggestions for better preregulators and voltage references were examined, well as the use of faster and lower noise op amps were examined. The addition of remote sensing to put the power supply wire impedance (which was greater than the regulator impedance and therefore the limiting factor) inside the feedback loop was added. The use of a buffer transistor to increase load impedance for the op amp and decrease the modulation of the op amp supply terminals was added along the lines of what was outlined above. The addition of a level shifting circuit to put the op amps DC output voltage below the regulators output voltage allowed the regulator to supply its op amp with a very clean and low impedance supply, this also a very symbiotic relation of circuits. Keeping the circuit small and relatively simple allowed stability to be optimized when using fast op amps and remote sensing. It also minimized the pick up of stray noise which is essential for a circuit whose feedback signal are in the microvolt range. some of these changes were made concurenty and possibly out of the sequence descibed, as some one will probably point out.
The original circuit by Mr. Sulzer was published in the March 1980 issue of the Audio Amateur. this gives pause to thing that people have been using this design for nearly a quarter of a century. Even more effort has been devoted to increasing the PSRR, than to the extremly low output impedance and noise. The sonic importance of these parameters was discussed in the original article. I can't think of too many circuits whose measured parameters have been correlated with sonics to the degree that this one has. When people come along and write it off I just have to sit and shake my head. They say familiarity breeds contempt, but I really don't understand why unfamiliarity should breed contempt. Not for such a classic circuit as this one. I am not implying that to be the case here, but there have been some others.......
The composite op amp scheme is an impressive circuit. It has been used with quite some success in preamplifiers from the reports I have read. It is also one of the most challenging circuits around in terms of design, decoulping, and PCB layout.
The amp inside the loop must be much faster (or slower) than the op amp controlling the loop; since you now have another pole in the open loop response. Current feedback op amps by themselves are very demanding to design with and the number app notes of what not to do with them is a pretty good indication of that. The AD811 pulls about 15mA of bias current and operation at greater than +/- 5 volts warrants a heat sink. Pulling load current from it will probably lead to hotter operation. I have some 811s and if am ever feeling dangerously cheerful, I will drag them out for a design to restore my grumpiness. The use of current feedback op amps in a power supplies invokes that little voice in my head that says "don't go there!" (a literary device, I do not actually hear voices, as some of you imagine). The even greater challenge, a composite op amp base supply with a current feedback op amp, brings to mind that ancient Chinese curse "May you live in interesting times." As Chuck Yeager used to say after some adventure in an airplane from which he narrowly escaped with his life, " It can be done but I don't recommend it."
"And yes, the options you quote all have their pro's and con's, and if this pack gets their teeth into it will undoubly tear much of it apart.'
Jan behave yourself! I offer only a few gentle nibbles, like a puppy playing, to someone with this level of design and analytical skill who is quite qualified to hand me my hat in this exchange. He has thought this out and I love the intellectual stimulation. I say this in deference without the least intension of sarcasm. He has made a pretty convincing case in theory.
Jbx, yes it is noisy and is mostly me. Your math models and rational appear to be faultless. If I might be permitted to summerize in a little simpler layman's terms (something I very seldom do, although often asked to). The introduction of the composite amp allows the U1 op amp (with greater PSRR) to increase the PSRR of U0, by reducing the power supply error voltage at the output of U0 by the amount of loop feedback (open loop gain divided by closed loop gain) of U1. The other result, the greatly reduced loading for U1 by driving UO's high input impedance instead of the load is maybe even more beneficial. The less current U1 has to supply the larger the power supply impedance increase from RC filter can be for a given signal induced noise voltage on the supply terminals for U1. In all, a very symbiotic arrangement for both op amps, in theory.
A little history on the development of the Jung Sulzer regulator. Original had an RC filter for the increase of the PSRR for op amp. The next article discussed to use of a preregulator and a two transistor darlington to increase output current capacity.
Some suggestions for better preregulators and voltage references were examined, well as the use of faster and lower noise op amps were examined. The addition of remote sensing to put the power supply wire impedance (which was greater than the regulator impedance and therefore the limiting factor) inside the feedback loop was added. The use of a buffer transistor to increase load impedance for the op amp and decrease the modulation of the op amp supply terminals was added along the lines of what was outlined above. The addition of a level shifting circuit to put the op amps DC output voltage below the regulators output voltage allowed the regulator to supply its op amp with a very clean and low impedance supply, this also a very symbiotic relation of circuits. Keeping the circuit small and relatively simple allowed stability to be optimized when using fast op amps and remote sensing. It also minimized the pick up of stray noise which is essential for a circuit whose feedback signal are in the microvolt range. some of these changes were made concurenty and possibly out of the sequence descibed, as some one will probably point out.
The original circuit by Mr. Sulzer was published in the March 1980 issue of the Audio Amateur. this gives pause to thing that people have been using this design for nearly a quarter of a century. Even more effort has been devoted to increasing the PSRR, than to the extremly low output impedance and noise. The sonic importance of these parameters was discussed in the original article. I can't think of too many circuits whose measured parameters have been correlated with sonics to the degree that this one has. When people come along and write it off I just have to sit and shake my head. They say familiarity breeds contempt, but I really don't understand why unfamiliarity should breed contempt. Not for such a classic circuit as this one. I am not implying that to be the case here, but there have been some others.......
The composite op amp scheme is an impressive circuit. It has been used with quite some success in preamplifiers from the reports I have read. It is also one of the most challenging circuits around in terms of design, decoulping, and PCB layout.
The amp inside the loop must be much faster (or slower) than the op amp controlling the loop; since you now have another pole in the open loop response. Current feedback op amps by themselves are very demanding to design with and the number app notes of what not to do with them is a pretty good indication of that. The AD811 pulls about 15mA of bias current and operation at greater than +/- 5 volts warrants a heat sink. Pulling load current from it will probably lead to hotter operation. I have some 811s and if am ever feeling dangerously cheerful, I will drag them out for a design to restore my grumpiness. The use of current feedback op amps in a power supplies invokes that little voice in my head that says "don't go there!" (a literary device, I do not actually hear voices, as some of you imagine). The even greater challenge, a composite op amp base supply with a current feedback op amp, brings to mind that ancient Chinese curse "May you live in interesting times." As Chuck Yeager used to say after some adventure in an airplane from which he narrowly escaped with his life, " It can be done but I don't recommend it."
Fred,
I thought your last a most interesting and challenging post, and thoroughly complement you for it! BRAVO!! I particularly enjoyed your description of being 'dangerously cheerful', and 'designing to restore grumpiness'! Wonderful..... Another genuine human being, not quite content unless he is a little miserable. Seminal stuff!
I've done a little work on power supplies for preamps myself, and finished with a simple, thoroughly decoupled zener voltage reference driving an emitter follower. Depressingly simple, but by God it sounded better than shunt regs, Sulzer regulators, and the whole shooting match. I concluded from this that a very high PSRR and linearly falling Zout with current draw (Zout of an emitter follower is essentially 26/Ie, with Ie in mA) was vital to good sonics. The very good frequency response of the emitter follower seems to be instrumental in this too. Don't ask me why, but it seems to work.
Thank you for a seminal posting.
Cheers,
Hugh
I thought your last a most interesting and challenging post, and thoroughly complement you for it! BRAVO!! I particularly enjoyed your description of being 'dangerously cheerful', and 'designing to restore grumpiness'! Wonderful..... Another genuine human being, not quite content unless he is a little miserable. Seminal stuff!
I've done a little work on power supplies for preamps myself, and finished with a simple, thoroughly decoupled zener voltage reference driving an emitter follower. Depressingly simple, but by God it sounded better than shunt regs, Sulzer regulators, and the whole shooting match. I concluded from this that a very high PSRR and linearly falling Zout with current draw (Zout of an emitter follower is essentially 26/Ie, with Ie in mA) was vital to good sonics. The very good frequency response of the emitter follower seems to be instrumental in this too. Don't ask me why, but it seems to work.
Thank you for a seminal posting.
Cheers,
Hugh
I believe that one reason simple circuits like the ones Hugh describes work so well is because their impedance tends to remain constant within the audio band. When combined with a circuit that is fairly immune to modulation on the rails, it can work quite well.
But back to CFB again. I hate to sound like a broken record, but I strongly urge against using them for audio. Their main selling point is bandwidth that remains constant as gain is changed. This bandwidth is almost always far more than is necessary for audio. These devices are best saved for RF work, where the banwidth is needed. Also, if you must use one, the non-inverting mode is harder to stabilize, as it is very sensitive to stray capacitance at the non-inverting node. I would not use it in darn near any application.
And did I mention that they sound odd.......
Yeah, I think that I did.
But back to CFB again. I hate to sound like a broken record, but I strongly urge against using them for audio. Their main selling point is bandwidth that remains constant as gain is changed. This bandwidth is almost always far more than is necessary for audio. These devices are best saved for RF work, where the banwidth is needed. Also, if you must use one, the non-inverting mode is harder to stabilize, as it is very sensitive to stray capacitance at the non-inverting node. I would not use it in darn near any application.
And did I mention that they sound odd.......
Yeah, I think that I did.
Emittor Follower Regulator
Hi Hugh,
I wish I could confirm your findings but that is only partial. I think it depends on what the regulator is powering.
The TDA1543 reacted allergic to my Jung like regulator and does sound much better powered from the triple Darlington scheme I posted on this forum.
However I just modified my J. regulators; powering IV-converters and outputbuffers of my DAC; to triple Darlington followers and now the sound is definitely worse. Weird and not what I did expect.
I also used Jung like regulators for my Sansui TUX1 tuner and that one almost had a resurrection in sound!
http://db.audioasylum.com/cgi/m.pl?forum=tweaks&n=40852&highlight=elso+sansui&r=&session=
In my preamp including balanced MC phonostage the Jung regulators sounded better than a LM317/337 combo with bypassed adjustment pin.😉
Cheers
AKSA said:
Hi Hugh,
I wish I could confirm your findings but that is only partial. I think it depends on what the regulator is powering.
The TDA1543 reacted allergic to my Jung like regulator and does sound much better powered from the triple Darlington scheme I posted on this forum.
However I just modified my J. regulators; powering IV-converters and outputbuffers of my DAC; to triple Darlington followers and now the sound is definitely worse. Weird and not what I did expect.
I also used Jung like regulators for my Sansui TUX1 tuner and that one almost had a resurrection in sound!
http://db.audioasylum.com/cgi/m.pl?forum=tweaks&n=40852&highlight=elso+sansui&r=&session=
In my preamp including balanced MC phonostage the Jung regulators sounded better than a LM317/337 combo with bypassed adjustment pin.😉
Cheers
Sideline: On tuners & power
In my experience, no other unit sends as much garbage out to the AC power-line as tuners (and TV).
So almost any PSU/filtering-upgrade usually will help...
Arne K
In my experience, no other unit sends as much garbage out to the AC power-line as tuners (and TV).
So almost any PSU/filtering-upgrade usually will help...
Arne K
I find the “economy” of using one added op amp in the multiloop to get the added PSRR at both amplifier pins simultaneously (inside the loop, no concerns about track resistance or extra Kelvin sensing traces) nearly enough to tip scale in favor of trying the multiloop before going on to exotic ps such as the high feedback, multiloop super-regulator on each ps; add in any one or two of the half-dozen other advantages of the multiloop as a amplifier and the scales come crashing over in favor of starting any demanding op amp based amplifier design with a multiloop topology. Of course the benefits of feedback only go so far (in frequency anyway) and attention to ps impedance and RF/EMI rejection are still important – and potentially simplified by the LF PSRR boost of the multiloop amplifier toptology.
As for the added complexity of multiloop amplifier design, I’m afraid it’s just the cost of doing business (its not like the principles haven’t been known for 60+ years) and the availability of very fast voltage (or current) feedback op amps simplifies the stability design of multiloop amplifiers for audio applications where most of the input op amps you might consider have 8-25 MHz unity gain stable GBW and can serve as the dominant pole compensating element with 100+ MHz output op amps in the loop.
I’m sorry I mentioned the AD811 (Jung will have to apologize on his own behalf)
As for the added complexity of multiloop amplifier design, I’m afraid it’s just the cost of doing business (its not like the principles haven’t been known for 60+ years) and the availability of very fast voltage (or current) feedback op amps simplifies the stability design of multiloop amplifiers for audio applications where most of the input op amps you might consider have 8-25 MHz unity gain stable GBW and can serve as the dominant pole compensating element with 100+ MHz output op amps in the loop.
I’m sorry I mentioned the AD811 (Jung will have to apologize on his own behalf)
This is in regard to the calculation of the input distuirbance voltage d. According to the definition of PSRR in the Walt Jung article here:
http://www.analog.com/Analog_Root/static/pdf/amplifiersLinear/training/Sensor_sect3.pdf
"If a change of X volts in the supply produces the same output change as a differential input change of Y volts, then the PSRR on that supply is X/Y". So if you take PSD_0 and divide it by the PSRR, you'll get the difference mode input disturbance, that is, the voltage between pins 2 and 3 of U0 (lets call it Vdm). In order to get the input voltage d from Vdm, use
Vdm = d/(1+AB)
where A is the open-loop gan of U0 and B is the feedback divider ratio of U0. Let's call the power supply disturbance Vsd. Then
d = (1+AB)Vsd/PSRR
So it looks like there's a (1+AB) factor missing from d.
http://www.analog.com/Analog_Root/static/pdf/amplifiersLinear/training/Sensor_sect3.pdf
"If a change of X volts in the supply produces the same output change as a differential input change of Y volts, then the PSRR on that supply is X/Y". So if you take PSD_0 and divide it by the PSRR, you'll get the difference mode input disturbance, that is, the voltage between pins 2 and 3 of U0 (lets call it Vdm). In order to get the input voltage d from Vdm, use
Vdm = d/(1+AB)
where A is the open-loop gan of U0 and B is the feedback divider ratio of U0. Let's call the power supply disturbance Vsd. Then
d = (1+AB)Vsd/PSRR
So it looks like there's a (1+AB) factor missing from d.
andy_c
Thanks for the excellent link. It is required reading for all fellow analog propeller heads.
jbx's analysis for PSSR was for a composite op amp. The purpose was to examine the effect of the additional oop gain firm the input amp on the power supply rejection for the output amp. The power supply terminals or the input op amp were assumed to be from a very clean supply and the PSRR of the input op amp is ignored for the sake of analysis. The fact that the input amp is driving only the very high input of the output amp also makes this a reasonable condition for ignoring the PSRR of the input amps and examining only the effect of its loop gain on the PSRR for the whole compost amp.
From andy-c's post:
"So I think it is fair to say that when considering the overall PSRR of the composite amplifier that the gain of the input op amp (in dB) adds to the PSRR of the output op amp, of course the last bracketed term above is the PSRR of the input op amp rejecting the PSD at its ps pins from the (hopefully quieter) input op amp supply which isn't improved by the composite topology"
The analysis on page 17 of your linked reference AMPLIFIERS FOR SIGNAL CONDITIONING has been modified by jbx to describe a composite amp with conditions above. An analysis could be done, ( by jbx.... my brain would melt) but the model he included is not really over simplified or academic in light of the actual circuit topology described. I believe it is correct in theory and my comments were mainly on the difficulty all but the very skilled with excellent test equipment wood have in actually building it. thanks again for to reference on the actual definition of PSRR in terms of the closed loop gain of the op amp. Thanks also for the opportunity for civil discourse on analog design, which some think me so incapable of. 🙂
Thanks for the excellent link. It is required reading for all fellow analog propeller heads.
jbx's analysis for PSSR was for a composite op amp. The purpose was to examine the effect of the additional oop gain firm the input amp on the power supply rejection for the output amp. The power supply terminals or the input op amp were assumed to be from a very clean supply and the PSRR of the input op amp is ignored for the sake of analysis. The fact that the input amp is driving only the very high input of the output amp also makes this a reasonable condition for ignoring the PSRR of the input amps and examining only the effect of its loop gain on the PSRR for the whole compost amp.
From andy-c's post:
"So I think it is fair to say that when considering the overall PSRR of the composite amplifier that the gain of the input op amp (in dB) adds to the PSRR of the output op amp, of course the last bracketed term above is the PSRR of the input op amp rejecting the PSD at its ps pins from the (hopefully quieter) input op amp supply which isn't improved by the composite topology"
The analysis on page 17 of your linked reference AMPLIFIERS FOR SIGNAL CONDITIONING has been modified by jbx to describe a composite amp with conditions above. An analysis could be done, ( by jbx.... my brain would melt) but the model he included is not really over simplified or academic in light of the actual circuit topology described. I believe it is correct in theory and my comments were mainly on the difficulty all but the very skilled with excellent test equipment wood have in actually building it. thanks again for to reference on the actual definition of PSRR in terms of the closed loop gain of the op amp. Thanks also for the opportunity for civil discourse on analog design, which some think me so incapable of. 🙂
multiloop to loop for thrill seekers
I will not dispute your argument on the use of multiloop op amps as do able and useful circuits. I was assuming your use for the composite amp PS regulator circuit included its use with an output pass transistor, an assumption I now wonder if was correct on my part. It is very interesting that the goals of minimal loading of the error op amp and increasing the effective PSRR are both goals pursued vigorously by the Jung Sulzer regulator through bootstrapping; and were even considered very early on in the evolution of the Sulzer regulator begining over two decades ago. The advantages of composite op amps and multiple feedback loops are not limited to just those composites made of two op amps. In fact, the use of a single op amp in combination with one or more ......... uh oh, Andy W. might read this and have a contract hit put out on me. Never mind ..................
jcx said:
I will not dispute your argument on the use of multiloop op amps as do able and useful circuits. I was assuming your use for the composite amp PS regulator circuit included its use with an output pass transistor, an assumption I now wonder if was correct on my part. It is very interesting that the goals of minimal loading of the error op amp and increasing the effective PSRR are both goals pursued vigorously by the Jung Sulzer regulator through bootstrapping; and were even considered very early on in the evolution of the Sulzer regulator begining over two decades ago. The advantages of composite op amps and multiple feedback loops are not limited to just those composites made of two op amps. In fact, the use of a single op amp in combination with one or more ......... uh oh, Andy W. might read this and have a contract hit put out on me. Never mind ..................
Hi Fred,
I wasn't questioning jcx's assumptions or your comments about them. My concern was only with the details of one specific aspect of his computation - that of the power supply disturbance of U0 referred to its input - the parameter which he calls "d" in his schematic. It appears that he's computing d by taking the power supply disturbance and dividing it by the PSRR (subtracting dB) in his equation:
But if I assume Jung's defiintion for the PSRR in the referenced link is correct, then dividing the power supply disturbance by the PSRR does not give the power supply disturbance referred to the input of the closed-loop amplifier, but rather it gives that disturbance referred to the floating differential input of the op-amp itself (Vpin3-Vpin2). Referring to U0, let's call this difference-mode voltage Vdm, the output voltage Vo, and the input voltage of the closed-loop amplifier Vin. Let's also call A the open-loop gain of U0, and B the feedback factor of U0. Since
Vo = Vin*A/(1+A*B) and
Vdm = Vo/A
then Vdm/Vin must be:
Vdm = Vin/(1+A*B) or
Vin = Vdm*(1+A*B)
So to sum up, Jung's definition of PSRR in the referenced link says that dividing the power supply disturbance by the PSRR gives the differential voltage Vdm, not Vin. So unless I'm making some boneheaded mistake here (perfectly possible I'm sure 😉 ), one must take Vdm and multiply it by (1+A*B) to get the disturbance referred to the input of the closed-loop amplifier U0 (what I call "Vin" above, and what jcx calls "d"). So according to this analysis, his computation of "d" is optimistic by a factor of (1+A*B).
I didn't try to go further into his derivation though.
I wasn't questioning jcx's assumptions or your comments about them. My concern was only with the details of one specific aspect of his computation - that of the power supply disturbance of U0 referred to its input - the parameter which he calls "d" in his schematic. It appears that he's computing d by taking the power supply disturbance and dividing it by the PSRR (subtracting dB) in his equation:
But if I assume Jung's defiintion for the PSRR in the referenced link is correct, then dividing the power supply disturbance by the PSRR does not give the power supply disturbance referred to the input of the closed-loop amplifier, but rather it gives that disturbance referred to the floating differential input of the op-amp itself (Vpin3-Vpin2). Referring to U0, let's call this difference-mode voltage Vdm, the output voltage Vo, and the input voltage of the closed-loop amplifier Vin. Let's also call A the open-loop gain of U0, and B the feedback factor of U0. Since
Vo = Vin*A/(1+A*B) and
Vdm = Vo/A
then Vdm/Vin must be:
Vdm = Vin/(1+A*B) or
Vin = Vdm*(1+A*B)
So to sum up, Jung's definition of PSRR in the referenced link says that dividing the power supply disturbance by the PSRR gives the differential voltage Vdm, not Vin. So unless I'm making some boneheaded mistake here (perfectly possible I'm sure 😉 ), one must take Vdm and multiply it by (1+A*B) to get the disturbance referred to the input of the closed-loop amplifier U0 (what I call "Vin" above, and what jcx calls "d"). So according to this analysis, his computation of "d" is optimistic by a factor of (1+A*B).
I didn't try to go further into his derivation though.
PSRR for poets
I believe you are correct. I trust the guys at Analog Devices to know the definition for PSRR. You equations for the gain look fine.
jbw's error looks to been along the lines of a math error more than any misunderstanding of the mechanism of PSRR. the discussion of the effect of the composite op amp topology (with the correction for the definition of d) still appears to be valid. The first equation in his Mathcad model described relationship between Vo and d appears fine but I did not work through the algebra further than that as what he called "a little hand massaging" appeared to be a hell of a lot of work. I am moderately lazy as I run Spice models for head scratchers like this.
His math error is much more forgivable than my haste in reading your first post which was clear to someone who read it closely. I ask your forgiveness for jumping to assumptions about the assertion you made in your post. It was case of not seeing the tree for the forest in this case. I would much rather be wrong in discussion with people with the level of knowledge of both you and jwb, than to be right in an argument with some of the idiots I have tangled with on this forum. Anyone who ask if I am talking about them in particular, should pause and think of the status that will attach to you for asking, even I don't answer. I once had a well known troublemaker ask me this when I thought it pretty clear that complaining about another well known denizen. Guilty conscience I guess.
Thanks for the link and for your detailed look at jwb's analysis. I'll bet even he will thank you.
I believe you are correct. I trust the guys at Analog Devices to know the definition for PSRR. You equations for the gain look fine.
jbw's error looks to been along the lines of a math error more than any misunderstanding of the mechanism of PSRR. the discussion of the effect of the composite op amp topology (with the correction for the definition of d) still appears to be valid. The first equation in his Mathcad model described relationship between Vo and d appears fine but I did not work through the algebra further than that as what he called "a little hand massaging" appeared to be a hell of a lot of work. I am moderately lazy as I run Spice models for head scratchers like this.
His math error is much more forgivable than my haste in reading your first post which was clear to someone who read it closely. I ask your forgiveness for jumping to assumptions about the assertion you made in your post. It was case of not seeing the tree for the forest in this case. I would much rather be wrong in discussion with people with the level of knowledge of both you and jwb, than to be right in an argument with some of the idiots I have tangled with on this forum. Anyone who ask if I am talking about them in particular, should pause and think of the status that will attach to you for asking, even I don't answer. I once had a well known troublemaker ask me this when I thought it pretty clear that complaining about another well known denizen. Guilty conscience I guess.
Thanks for the link and for your detailed look at jwb's analysis. I'll bet even he will thank you.
Ah Fred,
Noble sentiments, quite well expressed! You have humility, and that's always a good sign.
However, I take you to task on the 'idiots'.
I thank the Lord for these guys. They make me look and feel good, and as king of the heap, Fred, you surely should be grateful for them too. Intelligence and ability is all relative, and relatively speaking you are 1 smart guy, but amongst Alpha Centaurians you might well be an idiot yourself. (God knows where that puts most of us, but frankly it's not something we need worry about!!)
Elso, do you mean for every new circuit I devise I must revisit the power supply? Dammit!! I thought I'd saved myself some R&D!
Let me ask a question. Is it reasonable to suppose that a power supply with reducing Zout with increasing current draw, such as an emitter follower, would have different sonics to an active, opamp-based Sulzer with its vanishingly low but essentially constant Zout?
Cheers,
Hugh
Noble sentiments, quite well expressed! You have humility, and that's always a good sign.
However, I take you to task on the 'idiots'.
I thank the Lord for these guys. They make me look and feel good, and as king of the heap, Fred, you surely should be grateful for them too. Intelligence and ability is all relative, and relatively speaking you are 1 smart guy, but amongst Alpha Centaurians you might well be an idiot yourself. (God knows where that puts most of us, but frankly it's not something we need worry about!!)
Elso, do you mean for every new circuit I devise I must revisit the power supply? Dammit!! I thought I'd saved myself some R&D!
Let me ask a question. Is it reasonable to suppose that a power supply with reducing Zout with increasing current draw, such as an emitter follower, would have different sonics to an active, opamp-based Sulzer with its vanishingly low but essentially constant Zout?
Cheers,
Hugh
Hi Fred,
No apologies necessary whatsoever 😉 ! My original post on this was pretty darned terse, so that made it virtually impossible to determine the motivation for the statements I was making. With the way these forums go sometimes, statistics would point to such statements being some kind of jab at someone. I try my best not to do that, but I may not always succeed.
Before reading this thread, I didn't know what the formal definition of PSRR was. I had to look it up on the web, and that was when I found the link I posted. I'll have to get around to reading the rest of that app note.
But back to the PSRR subject! It looks like all that remains to calculate the equivalent PSRR of the composite amp is to take the power supply disturbance voltage "d" and refer it to the differential input of the first op-amp U1. Assuming I didn't make any error in my previous post, d should be:
d = (1+A0*B)*Vsd/PSRR0
Where A0 is the open loop gain of U0, PSRR0 is the PSRR of U0, B is the feedback factor of U0, and Vsd is the power supply disturbance voltage of U0. Now it looks like all we have to do is divide "d" by the open-loop gain of U1 to get the disturbance referred to the difference-mode input of the composite op-amp, since we're assuming that no power supply disturbance occurs at U1. We can call this input-referred voltage Vid for the input disturbance voltage. Then we can take Vsd and divide it by Vid to get the PSRR for the composite op-amp. So it looks like
Vid = Vsd*(1+A0*B)/(PSRRO*A1)
where A1 is the open-loop gain of U1. Now the PSRR of the composite should just be:
PSRR_total=Vsd/Vid
PSRR_total= PSRR0*A1/(1+A0*B)
In the special case of identical op-amps (A1=A0=A), and with A very large, it looks like
PSRR_total ~= PSRR0/B
In other words, the PSRR of the second op-amp has been improved by an amount equal to its closed-loop gain. For non-identical op amps, it's kind of hard to see what's going on.
One thing that bothers me about the formal definition of PSRR from the Jung article is that at first glance, it doesn't appear to be compatible with current feedback op-amps. The relationship between output voltage and differential input voltage isn't explicitly specified for the current feedback case. Since the open loop transfer function of CFB op-amps is not unitless, I'm a bit confused as to how to deal with it mathematically for computing PSRR.
I welcome all corrections and comments to this, as I'm trying to learn more about this stuff. I've been thinking about the design of a high-voltage version of the Sulzer/Jung design (+/-95 V for the low-current portion of a power amp), so I hope to get as many ideas as possible.
No apologies necessary whatsoever 😉 ! My original post on this was pretty darned terse, so that made it virtually impossible to determine the motivation for the statements I was making. With the way these forums go sometimes, statistics would point to such statements being some kind of jab at someone. I try my best not to do that, but I may not always succeed.
Before reading this thread, I didn't know what the formal definition of PSRR was. I had to look it up on the web, and that was when I found the link I posted. I'll have to get around to reading the rest of that app note.
But back to the PSRR subject! It looks like all that remains to calculate the equivalent PSRR of the composite amp is to take the power supply disturbance voltage "d" and refer it to the differential input of the first op-amp U1. Assuming I didn't make any error in my previous post, d should be:
d = (1+A0*B)*Vsd/PSRR0
Where A0 is the open loop gain of U0, PSRR0 is the PSRR of U0, B is the feedback factor of U0, and Vsd is the power supply disturbance voltage of U0. Now it looks like all we have to do is divide "d" by the open-loop gain of U1 to get the disturbance referred to the difference-mode input of the composite op-amp, since we're assuming that no power supply disturbance occurs at U1. We can call this input-referred voltage Vid for the input disturbance voltage. Then we can take Vsd and divide it by Vid to get the PSRR for the composite op-amp. So it looks like
Vid = Vsd*(1+A0*B)/(PSRRO*A1)
where A1 is the open-loop gain of U1. Now the PSRR of the composite should just be:
PSRR_total=Vsd/Vid
PSRR_total= PSRR0*A1/(1+A0*B)
In the special case of identical op-amps (A1=A0=A), and with A very large, it looks like
PSRR_total ~= PSRR0/B
In other words, the PSRR of the second op-amp has been improved by an amount equal to its closed-loop gain. For non-identical op amps, it's kind of hard to see what's going on.
One thing that bothers me about the formal definition of PSRR from the Jung article is that at first glance, it doesn't appear to be compatible with current feedback op-amps. The relationship between output voltage and differential input voltage isn't explicitly specified for the current feedback case. Since the open loop transfer function of CFB op-amps is not unitless, I'm a bit confused as to how to deal with it mathematically for computing PSRR.
I welcome all corrections and comments to this, as I'm trying to learn more about this stuff. I've been thinking about the design of a high-voltage version of the Sulzer/Jung design (+/-95 V for the low-current portion of a power amp), so I hope to get as many ideas as possible.
andy_c;
My use of d = PSD – PSRR (in dB) is consistent with the industry standard definition in the AD app note, this “input referred “ definition of PSRR is not as obvious as offset or noise voltage which naturally appear at the op amp input. It is usually too awkward to carry the real PSD voltage input and separate Vout/Vpsd gain term around in addition to the ideal op amp diff input and open loop gain, for most usages the input referred PSRR shares the calculation convenience of the other input referred terms that are simply viewed as independent voltage sources in series with the op amp differential input. (Predicting the stability of ps bootstrapped amplifiers is one application where it is better to use Vout/Vpsd as a separate input gain) I think you are confusing the usage of the “real” Vout/Vpsd gain term and the “input referred” PSRR which has been divided by open loop gain already so that is can be used to simply calculate the voltage source in series with the op amp input.
TI/BB app note “Op Amp Performance Analysis” http://focus.ti.com/lit/an/sboa054/sboa054.pdf shows input referred error voltage analysis, http://focus.ti.com/lit/an/sboa015/sboa015.pdf shows some multiloop compensation approaches using Bode plots rather than heavy math
For current feedback op amps I hope the PSRR is referred to the high impedance noninverting voltage input as a voltage, you presumably could “transfer” the input referred voltage to the minus input as a current by dividing by the input differnential resistance Rin or R0 ( ~ 50-100 ohms)
My use of d = PSD – PSRR (in dB) is consistent with the industry standard definition in the AD app note, this “input referred “ definition of PSRR is not as obvious as offset or noise voltage which naturally appear at the op amp input. It is usually too awkward to carry the real PSD voltage input and separate Vout/Vpsd gain term around in addition to the ideal op amp diff input and open loop gain, for most usages the input referred PSRR shares the calculation convenience of the other input referred terms that are simply viewed as independent voltage sources in series with the op amp differential input. (Predicting the stability of ps bootstrapped amplifiers is one application where it is better to use Vout/Vpsd as a separate input gain) I think you are confusing the usage of the “real” Vout/Vpsd gain term and the “input referred” PSRR which has been divided by open loop gain already so that is can be used to simply calculate the voltage source in series with the op amp input.
TI/BB app note “Op Amp Performance Analysis” http://focus.ti.com/lit/an/sboa054/sboa054.pdf shows input referred error voltage analysis, http://focus.ti.com/lit/an/sboa015/sboa015.pdf shows some multiloop compensation approaches using Bode plots rather than heavy math
For current feedback op amps I hope the PSRR is referred to the high impedance noninverting voltage input as a voltage, you presumably could “transfer” the input referred voltage to the minus input as a current by dividing by the input differnential resistance Rin or R0 ( ~ 50-100 ohms)
Okay, I've looked at this some more. The definition of PSRR in the app note as stated by Jung et al is this:
"If a change of X volts in the supply produces the same output change as a differential input change of Y volts, then the PSRR on that supply is X/Y".
This definition strongly suggests that this so-called "differential input change" is just that - the change in voltage at the differential input. But checking into this further shows that this can't possibly be the case. Suppose you have an op-amp with an open-loop gain of 90 dB and a PSRR of 90 dB - reasonable numbers. If I were to take the definition above literally, the differential input voltage due to the power supply disturbance would be 90 dB below the disturbance. But the output voltage would be 90 dB above the differential input voltage, so the output voltage change would be the same as the power supply voltage change - a nonsensical result.
I wish the app note had either provided a formula, or an explicit statement that the so-called "differential input change" is the output voltage divided by the closed-loop gain - and as such is not a differential input voltage at all.
"If a change of X volts in the supply produces the same output change as a differential input change of Y volts, then the PSRR on that supply is X/Y".
This definition strongly suggests that this so-called "differential input change" is just that - the change in voltage at the differential input. But checking into this further shows that this can't possibly be the case. Suppose you have an op-amp with an open-loop gain of 90 dB and a PSRR of 90 dB - reasonable numbers. If I were to take the definition above literally, the differential input voltage due to the power supply disturbance would be 90 dB below the disturbance. But the output voltage would be 90 dB above the differential input voltage, so the output voltage change would be the same as the power supply voltage change - a nonsensical result.
I wish the app note had either provided a formula, or an explicit statement that the so-called "differential input change" is the output voltage divided by the closed-loop gain - and as such is not a differential input voltage at all.
- Status
- Not open for further replies.