In a post in another thread (http://www.diyaudio.com/forums/showthread.php?postid=334536#post334536), Fred D suggested starting a new thread on ways to improve the performance of three-terminal regulators (a potentially beneficial discussion in my opinion, provided we can keep it on topic):
quote: "Maybe a new thread on design, simulation, and test of high performance three terminal regulators would be worth starting, free and clear of the morass this thread became."
So here it is.
To start the ball rolling, taking the standard datasheet circuit, I would suggest that the resistor in the lower limb of the potential divider that determines the output voltage be replaced with a voltage reference (TL431/???, equal to Vout-1.25V). This should improve ripple/noise rejection and will enable a reduction in the value of the capacitor from ADJ to ground so that the phase margin of the LM3x7 can be maximised. An improved phase margin should reduce the amount of ringing on step loads.
quote: "Maybe a new thread on design, simulation, and test of high performance three terminal regulators would be worth starting, free and clear of the morass this thread became."
So here it is.
To start the ball rolling, taking the standard datasheet circuit, I would suggest that the resistor in the lower limb of the potential divider that determines the output voltage be replaced with a voltage reference (TL431/???, equal to Vout-1.25V). This should improve ripple/noise rejection and will enable a reduction in the value of the capacitor from ADJ to ground so that the phase margin of the LM3x7 can be maximised. An improved phase margin should reduce the amount of ringing on step loads.
This isn't exactly on topic, but... if noise is an issue, would the LM723 be a better choice?
http://www.national.com/ds/LM/LM723.pdf
http://www.national.com/ds/LM/LM723.pdf
The National App Note has a couple of circuits that use a zener as a reference. I remember that several on this forum have tried a zener for a reference, sometimes bypassing the zener with a capacitor for reduced zener noise and better regulation.
Are multiple regulators, a la the tracking preregulator (or even non-tracking) being considered in this thread? A standard tracking prereg per an App Note with a capacitor on each adjust pin ought to be fruitful. But I don't know how well that would help Zout and other factors.
Are multiple regulators, a la the tracking preregulator (or even non-tracking) being considered in this thread? A standard tracking prereg per an App Note with a capacitor on each adjust pin ought to be fruitful. But I don't know how well that would help Zout and other factors.
Mark:
>The National App Note has a couple of circuits that use a zener as a reference.<
Have a link, or would you know the application note number?
>Are multiple regulators, a la the tracking preregulator (or even non-tracking) being considered in this thread?<
I normally find both preregulation and multiple regulators to be useful from the performance aspect. But due to the penalty in component count and board surface area, their use should be considered on a case-by-case basis.
>But I don't know how well that would help Zout and other factors.<
You could also add an output follower, and drive the load from the follower.
regards, jonathan carr
>The National App Note has a couple of circuits that use a zener as a reference.<
Have a link, or would you know the application note number?
>Are multiple regulators, a la the tracking preregulator (or even non-tracking) being considered in this thread?<
I normally find both preregulation and multiple regulators to be useful from the performance aspect. But due to the penalty in component count and board surface area, their use should be considered on a case-by-case basis.
>But I don't know how well that would help Zout and other factors.<
You could also add an output follower, and drive the load from the follower.
regards, jonathan carr
I agree, The LM723 is still a very good choice for a low noise regulator, first because of the voltage reference that is used and secondly because the reference can be decoupled by an external cap.
I am always surprised when people is trying to use simple 3 pin regulators to feed sensitive audiocircuits, most 3 pins are unstable with certain loads, noisy and slow. For very critical applications where noise is an issue it is very difficult to remove noise from a 3 pin regulator, it is better to start with something better. LM723 with good external components is a good reference that can be used to compare other regulators with.
BTW, LM723 has a Zener reference which is although noisy, less noisy then the bandgap reference that is used in many other regulators.
BR Hans
Zener
A zener of voltage reference from the ADJ terminal will provide less variation of this voltage with changes the ADJ current. You will very likely face the penalty of a pretty large increase in broadband noise from the zener. The noise voltage for the LM317 is about 30uV per DC output volt. Capacitor bypass of the resistor to ground should drop the noise about 35 to 40uV over the frequency range where the impedance of the cap is lower than resistor to ground from the ADJ terminal. I have seen much bigger than 10uF caps used for the ADJ bypass. The limit seems largely determined by damage the discharge current can do the regulator during fault conditions like shorting the output or input of the regulator
I am looking at a bootstrapped current source instead of the resistor for Vout the ADJ which I have seen in another design. I have to do a startup circuit for it since it needs the output voltage there and will not start by itself.
Ben Duncan did a very good regulator article in HIFI News and Record Review but I don't have a copy anymore. I seem to remember it using small signal diodes with series resistors for the clamps from Vout to Vin and Vout to ADJ to minimized RFI coupling from the raw supply. Anybody have it or remember the circuit? Andy (AWL)?s
A zener of voltage reference from the ADJ terminal will provide less variation of this voltage with changes the ADJ current. You will very likely face the penalty of a pretty large increase in broadband noise from the zener. The noise voltage for the LM317 is about 30uV per DC output volt. Capacitor bypass of the resistor to ground should drop the noise about 35 to 40uV over the frequency range where the impedance of the cap is lower than resistor to ground from the ADJ terminal. I have seen much bigger than 10uF caps used for the ADJ bypass. The limit seems largely determined by damage the discharge current can do the regulator during fault conditions like shorting the output or input of the regulator
I am looking at a bootstrapped current source instead of the resistor for Vout the ADJ which I have seen in another design. I have to do a startup circuit for it since it needs the output voltage there and will not start by itself.
Ben Duncan did a very good regulator article in HIFI News and Record Review but I don't have a copy anymore. I seem to remember it using small signal diodes with series resistors for the clamps from Vout to Vin and Vout to ADJ to minimized RFI coupling from the raw supply. Anybody have it or remember the circuit? Andy (AWL)?s
PSRR and Zout Regulator test fixture
This simple test fixture can be used with a scope, meter, or PC based FFT for testing Three Terminal Regulators for PSRR, Zout, and Transient response (oscilloscope only). Op amp supplies and their decoupling not shown. I use 9 volt batteries and 0.01 decoupling caps. Split gain between two op amps to get good bandwidth. The current change is 10 mA per volt. Ground Vin for load current mosfet when testing for PSRR and input transient response. R0/R1 AND R9/R10 can be replaced with two 100 K pots but I would use 10 or 20 turn pots for this. The output mosfet needs to be based to a greater DC drain current the AC current swing used for testing don't forget to add the current to any DC load current you have on the regulator when logging your test results. The change in current on the output is about 10mA per volt for input voltage to the current mosfet gate via the 2.2uF cap. Monitor the voltage across the 100 ohm resistor for the exactly value.
I posted this to show how simple and cheap a fixture for testing regulators can be, and to use for the testing the circuit mods that I hope will follow in this thread. It can also be used to test other regulator circuits but you will to use low noise op amps for decent resolution.
This simple test fixture can be used with a scope, meter, or PC based FFT for testing Three Terminal Regulators for PSRR, Zout, and Transient response (oscilloscope only). Op amp supplies and their decoupling not shown. I use 9 volt batteries and 0.01 decoupling caps. Split gain between two op amps to get good bandwidth. The current change is 10 mA per volt. Ground Vin for load current mosfet when testing for PSRR and input transient response. R0/R1 AND R9/R10 can be replaced with two 100 K pots but I would use 10 or 20 turn pots for this. The output mosfet needs to be based to a greater DC drain current the AC current swing used for testing don't forget to add the current to any DC load current you have on the regulator when logging your test results. The change in current on the output is about 10mA per volt for input voltage to the current mosfet gate via the 2.2uF cap. Monitor the voltage across the 100 ohm resistor for the exactly value.
I posted this to show how simple and cheap a fixture for testing regulators can be, and to use for the testing the circuit mods that I hope will follow in this thread. It can also be used to test other regulator circuits but you will to use low noise op amps for decent resolution.
Attachments
Jonathan, the App note is for the LM117 and LM317, which I guess is technically a datasheet, not an app note....
http://www.national.com/ds.cgi/LM/LM117.pdf
Specifically the "High Stability 10V Regulator" on page 14. I don't see why they use a 2k resistor instead of the standard 240ohm from Vout to Vadj in this example, which raises the question: how much does changing the value of the Vout-Vadj resistor value affect the output characteristics? I think I'll go do some simulations before bed.
I notice that the Electrical Properties page lists PSRR specs for no Cadj and Cadj = 10uF.
Cheers,
Mark
http://www.national.com/ds.cgi/LM/LM117.pdf
Specifically the "High Stability 10V Regulator" on page 14. I don't see why they use a 2k resistor instead of the standard 240ohm from Vout to Vadj in this example, which raises the question: how much does changing the value of the Vout-Vadj resistor value affect the output characteristics? I think I'll go do some simulations before bed.
I notice that the Electrical Properties page lists PSRR specs for no Cadj and Cadj = 10uF.
Cheers,
Mark
As noone has posted the link for recent article series on just this subject on TNT-Audio here it is...
http://www.tnt-audio.com/clinica/regulators_noise1_e.html
It is of some interest, at least was for me
Ergo
http://www.tnt-audio.com/clinica/regulators_noise1_e.html
It is of some interest, at least was for me
Ergo
Re: Zener
The signal diode thing was documented in Duncan's HFN&RR series on the AMP2 preamp design in 1990 or so. But in this design he used the LT317/337 regs in standard configuration, i.e. with largish bypass cap, and about 330uF+33uF elcaps at the output, and presumably with a proper ground layout.
The protection signal diodes were used as they exhibit a lower junction capacitance than ordinary power diodes, and as such allow less ingress of HF/RF from the rectifiers.
Fred Dieckmann said:
The signal diode thing was documented in Duncan's HFN&RR series on the AMP2 preamp design in 1990 or so. But in this design he used the LT317/337 regs in standard configuration, i.e. with largish bypass cap, and about 330uF+33uF elcaps at the output, and presumably with a proper ground layout.
The protection signal diodes were used as they exhibit a lower junction capacitance than ordinary power diodes, and as such allow less ingress of HF/RF from the rectifiers.
And I was starting to think every one was asleep
"this example, which raises the question: how much does changing the value of the Vout-Vadj resistor value affect the output characteristics?"
You may have asked the magic question. The fairly large programing current makes the internal 50uA current that appears from the adjustment terminal look smaller in terms of an error current. For a adustment current of 1.25 V divided by 240 ohm we get 5.2mA. The current from the ADJ terminal (used to bias the regulator's1.25 V voltage reference) represents about 1% error in relation to the 5.2mA current. This current can vary in initial value up to 100uA. When picking a set value of resistors this 1% variation must be added to the resistor tolerance for the error budget of how much the output voltage value will vary in production. Using a larger resistor value means less current and that the 50uA variation is a larger error term. The use of a voltage reference from ADJ to ground shows much less variation since the reference has a much lower impedance. An LM 329 has under 1 ohm which gives a much smaller change than varying current through a resistor of several kilohms. Using a pot to set the desired voltage works around this issue but is not practicle in large scale production, except by other means than a standard trim pot which would have to be tweakied by hand for the desired voltage.
A smaller current gives greater PSRR at low freqencies for a given sized cap from ADJ to ground, since the cap impedance smaller compared to a larger resistor value. I have seen values of 1K used instead of 120 or 240 ohms for this purpose in some audio regulators. Another thing the relatively large program current does is provide the minimum load current that the regulator needs to operate correctly. This load can easily achieved by a resistor across the output of the regulator where it is not compromising the dynamic performance of the regulator if not derived form voltage adjust circuit . The ap note circuit is designed to deal with the issue of voltage tolerance and minimum load which are easy work arounds for small quantity production. The data sheet has a good description of this current error term on page 8 of the data sheet:
http://www.national.com/ds/LM/LM117.pdf
The 1.2 Volts is even enough current for certain (very specific devices) transistor current sources to be used between Vout and ADJ, an approach I am investigating, stay tuned.............
Altus Audio's lead engineer Hilton and ubertech Ritchie order parts from DigiKey for power supply test jig.
"this example, which raises the question: how much does changing the value of the Vout-Vadj resistor value affect the output characteristics?"
You may have asked the magic question. The fairly large programing current makes the internal 50uA current that appears from the adjustment terminal look smaller in terms of an error current. For a adustment current of 1.25 V divided by 240 ohm we get 5.2mA. The current from the ADJ terminal (used to bias the regulator's1.25 V voltage reference) represents about 1% error in relation to the 5.2mA current. This current can vary in initial value up to 100uA. When picking a set value of resistors this 1% variation must be added to the resistor tolerance for the error budget of how much the output voltage value will vary in production. Using a larger resistor value means less current and that the 50uA variation is a larger error term. The use of a voltage reference from ADJ to ground shows much less variation since the reference has a much lower impedance. An LM 329 has under 1 ohm which gives a much smaller change than varying current through a resistor of several kilohms. Using a pot to set the desired voltage works around this issue but is not practicle in large scale production, except by other means than a standard trim pot which would have to be tweakied by hand for the desired voltage.
A smaller current gives greater PSRR at low freqencies for a given sized cap from ADJ to ground, since the cap impedance smaller compared to a larger resistor value. I have seen values of 1K used instead of 120 or 240 ohms for this purpose in some audio regulators. Another thing the relatively large program current does is provide the minimum load current that the regulator needs to operate correctly. This load can easily achieved by a resistor across the output of the regulator where it is not compromising the dynamic performance of the regulator if not derived form voltage adjust circuit . The ap note circuit is designed to deal with the issue of voltage tolerance and minimum load which are easy work arounds for small quantity production. The data sheet has a good description of this current error term on page 8 of the data sheet:
http://www.national.com/ds/LM/LM117.pdf
The 1.2 Volts is even enough current for certain (very specific devices) transistor current sources to be used between Vout and ADJ, an approach I am investigating, stay tuned.............
Altus Audio's lead engineer Hilton and ubertech Ritchie order parts from DigiKey for power supply test jig.
Attachments
Current source
I have had limited success with the LM334 from stability and noise issues but would not rule it out. I am a little reluctant to introduce another device with feedback into the feedback loop of the regulator and even would be careful about adding capacitance to this part of the circuit. I had something else in mind. A clue below............. (the graph is 1mA and 1 volt per division)
I have had limited success with the LM334 from stability and noise issues but would not rule it out. I am a little reluctant to introduce another device with feedback into the feedback loop of the regulator and even would be careful about adding capacitance to this part of the circuit. I had something else in mind. A clue below............. (the graph is 1mA and 1 volt per division)
Attachments
Uh Oh.............
You are going in the right direction.........
More questions to think about:
1. What happens when the output is shorted, will the jfet handle the peak current of the ADJ capacitor discharging through the now forward biased diode structure of the reversed polarity jfet? It very likely won't.
2. What the effect of a few picofards capacitance from the output to the noninverting input of the error amps in the regulator? As I remember Ben Duncan thought this was a significant issue when adding protection diodes across a three terminal regulator.
3. How to achieve the goal in number 1 while minimizing the effect of number 2?
4. Is the jfet still a constant current source at 1.2 Volts and less or has output impedance started to become a fairly low resistance?
5. how much does the internal current source move with temperature? Since it is a much larger portion of the current and the resistor to ground is larger, how much does the regulator output voltage change with temperature?
Not as easy as everyone thought, I'll bet. Really tweaking the last bit of performance from the LM 317 with a fairly simple circuit and with capacitor values on the ADJ terminal that won't damage the regulator is more involved than anyone thought I'll bet. The use of cap values below 5 uF would make the use of film caps practical, to avoid leakage currents in electrolytics that cause DC drift and noise. You are going to have to go further than the things the simple Spice model tells you about this part.......... That's what I get for working on protection circuit for a few years.
You are going in the right direction.........
More questions to think about:
1. What happens when the output is shorted, will the jfet handle the peak current of the ADJ capacitor discharging through the now forward biased diode structure of the reversed polarity jfet? It very likely won't.
2. What the effect of a few picofards capacitance from the output to the noninverting input of the error amps in the regulator? As I remember Ben Duncan thought this was a significant issue when adding protection diodes across a three terminal regulator.
3. How to achieve the goal in number 1 while minimizing the effect of number 2?
4. Is the jfet still a constant current source at 1.2 Volts and less or has output impedance started to become a fairly low resistance?
5. how much does the internal current source move with temperature? Since it is a much larger portion of the current and the resistor to ground is larger, how much does the regulator output voltage change with temperature?
Not as easy as everyone thought, I'll bet. Really tweaking the last bit of performance from the LM 317 with a fairly simple circuit and with capacitor values on the ADJ terminal that won't damage the regulator is more involved than anyone thought I'll bet. The use of cap values below 5 uF would make the use of film caps practical, to avoid leakage currents in electrolytics that cause DC drift and noise. You are going to have to go further than the things the simple Spice model tells you about this part.......... That's what I get for working on protection circuit for a few years.
Sorry that this is just a loose conceptual idea with no simulations
or measurements backing it up, and that I am taking the risk of
being wrong by posting late at night.
The idea of using a CCS from Vout to Adj and a voltage reference,
a zener a string o LEDs or whatever, from Adj to ground has been
suggested both here and in an earlier thread that I linked to in
the "parent thread" of this one. One potential problem here, that
Fred also brought up, although as a question to us rather than
a claim, is about CCSs for such small voltages as 1.2V. There might
perhaps be JFETs working in the saturation region already for such
small voltages, at least if used at a sufficiently small Id, I don't know
since I haven't checked it. An altenative would be a BJT-based CCS,
but we will probably be out of the linear region also here for most
or all BJTs. Maybe this is still sufficiently good. Howver, maybe there
is a different approach to this.
Since the regulator attempts to maintain a fixed voltage difference
between Vout and Adj, it seems to me that it is not really necessary
that the CCS is connected from Vout to Adj. Couldn't it just as
well be connected from Vin to Adj, which would give it a larger
voltage drop to work with? This may on the other hand introduce
noise since Vin is not regulated. However, if one anyway uses
a preregulator this might perhaps be an interesting approach.
Alternatively one could use an extra dedicated regulator
(an LM317L would suffice) just to feed the CCS.
or measurements backing it up, and that I am taking the risk of
being wrong by posting late at night.
The idea of using a CCS from Vout to Adj and a voltage reference,
a zener a string o LEDs or whatever, from Adj to ground has been
suggested both here and in an earlier thread that I linked to in
the "parent thread" of this one. One potential problem here, that
Fred also brought up, although as a question to us rather than
a claim, is about CCSs for such small voltages as 1.2V. There might
perhaps be JFETs working in the saturation region already for such
small voltages, at least if used at a sufficiently small Id, I don't know
since I haven't checked it. An altenative would be a BJT-based CCS,
but we will probably be out of the linear region also here for most
or all BJTs. Maybe this is still sufficiently good. Howver, maybe there
is a different approach to this.
Since the regulator attempts to maintain a fixed voltage difference
between Vout and Adj, it seems to me that it is not really necessary
that the CCS is connected from Vout to Adj. Couldn't it just as
well be connected from Vin to Adj, which would give it a larger
voltage drop to work with? This may on the other hand introduce
noise since Vin is not regulated. However, if one anyway uses
a preregulator this might perhaps be an interesting approach.
Alternatively one could use an extra dedicated regulator
(an LM317L would suffice) just to feed the CCS.
question to us rather than a claim..... the data sheets claim.
"One potential problem here, that Fred also brought up, although as a question to us rather than a claim, is about CCSs for such small voltages as 1.2V. There might perhaps be JFETs working in the saturation region already for such small voltages, at least if used at a sufficiently small Id, I don't know since I haven't checked it."
There are jfets that are still acting in the current source mode at voltages of a volt and below. One could also use voltage reference like a LM329 biased from the regulators output, filtered, and multiplied by an op amp running from the regulated output. There are many op amp who's output will swing to less than 1.2 volts from the supply, like the AD820 or (AD823 dual) for example. I was thinking along the lines of some thing simple that used 2 or three extra transistors and a larger R (and smaller C) from ADJ to ground. I actually have a design but would like to play with the proto before posting.
"One potential problem here, that Fred also brought up, although as a question to us rather than a claim, is about CCSs for such small voltages as 1.2V. There might perhaps be JFETs working in the saturation region already for such small voltages, at least if used at a sufficiently small Id, I don't know since I haven't checked it."
There are jfets that are still acting in the current source mode at voltages of a volt and below. One could also use voltage reference like a LM329 biased from the regulators output, filtered, and multiplied by an op amp running from the regulated output. There are many op amp who's output will swing to less than 1.2 volts from the supply, like the AD820 or (AD823 dual) for example. I was thinking along the lines of some thing simple that used 2 or three extra transistors and a larger R (and smaller C) from ADJ to ground. I actually have a design but would like to play with the proto before posting.
What are the requirements that we need? Some kind of spec to meet? I like 3 terminal regulators because they are simple and don't take up much board space. Sometimes simple means reliable. Nice benefits of the IC regulators are packaging, current limit and thermal shutdown. If circuit size, current limit, or thermal shutdown is not a requirement, use a precision reference, OPA134, and a series pass transistor. Add one more transistor and have a low voltage dropout regulator. You can have current limit/shutdown, and thermal shutdown with more added discretes.
However, the top of discussion is how to improve the 3 terminal regulator. Maybe by the time you get it improved, you won't need it.
If you want lower noise at low current levels, use an RC low pass filter near the sensitive circuit. I usually use a 10, 50, or 100 ohm resistor in series and a .1u and .01u (or some other combination)in parallel to ground. Using surface mount parts, you can really reduce the noise. I use power planes above ground planes on most of my PCBs as good quality capacitance.
Using 2 or 3 different valued capacitors in parallel at the output of a 3 terminal regulator and good PCB layout techniques is good for enough for some applications.
However, the top of discussion is how to improve the 3 terminal regulator. Maybe by the time you get it improved, you won't need it.
If you want lower noise at low current levels, use an RC low pass filter near the sensitive circuit. I usually use a 10, 50, or 100 ohm resistor in series and a .1u and .01u (or some other combination)in parallel to ground. Using surface mount parts, you can really reduce the noise. I use power planes above ground planes on most of my PCBs as good quality capacitance.
Using 2 or 3 different valued capacitors in parallel at the output of a 3 terminal regulator and good PCB layout techniques is good for enough for some applications.
Fred,
OK sounds good if there are JFETs that make it possible to build
low-voltage-drop CCSs. To bad JFETs are getting hard to buy
easily for many of us.
My main point however was to point to a possible alternative
route to go. All designs I can remember seeing are essentially
based on the standard connection by replacing one or both
resistors in the voltage divider with other types of components.
My point was: Do we really need any component at all between
Vout and Adj if we have some kind of voltage reference connected
to Adj. It seems we don't. Yes, this is just a loose conceptual idea
as I said and, no, it is not a full design. If that bothers some people,
sorry. I seem however to remember one of our famous talented
commecial designers asking for "ideas, simulations or
measurements" of things
recently, but that must have been in the thread this one originated
from.
OK sounds good if there are JFETs that make it possible to build
low-voltage-drop CCSs. To bad JFETs are getting hard to buy
easily for many of us.
My main point however was to point to a possible alternative
route to go. All designs I can remember seeing are essentially
based on the standard connection by replacing one or both
resistors in the voltage divider with other types of components.
My point was: Do we really need any component at all between
Vout and Adj if we have some kind of voltage reference connected
to Adj. It seems we don't. Yes, this is just a loose conceptual idea
as I said and, no, it is not a full design. If that bothers some people,
sorry. I seem however to remember one of our famous talented
commecial designers asking for "ideas, simulations or
measurements" of things
recently, but that must have been in the thread this one originated
from.