Hello Jan, maybe I am not seeing this properly, but I do not quite understand how the current values for R7 and R8 would really work out in the first place
For R7 there is just the VBE of Q2 and R8 should - unfortunately - be even larger than it is now since the emitter of Q1 cannot move closer to the rail as it is limited by the voltage drop across CCS of R1, Q4, and Q7 (-VBE of Q1)Without changing the current layout one could off course change Q2, either to a Darlington, or - as Rüdiger’s suggested – to a FET. With an IRF FET we would at least have some 4 volts to drop across R7. That still leaves the question of what to do with this CCS?
MRupp said:
Hello Jan, maybe I am not seeing this properly, but I do not quite understand how the current values for R7 and R8 would really work out in the first place
Without changing the current layout one could off course change Q2, either to a Darlington, or - as Rüdiger’s suggested – to a FET. With an IRF FET we would at least have some 4 volts to drop across R7. That still leaves the question of what to do with this CCS?
You're absolutely right - that's why I put a zener in series with R8 to give the cascode some breathing room, and decrease R8 to 120 ohms, but it can probably be made even smaller to get more loop gain.
A FET would help with the level, but not with the gain. I'll look at it again tomorrow when I get time.
Jan Didden
Did some more measurements. I discovered that R5, 1k in the schematic, was actually 9.1k on the board 
. You guys really have to get to grips with version control. I have 2 schematics, one layout and one pcb and all four are different....
I changed R5 to 3k, so the base current of Q6 and Q5 is less influencing the CCS current. Also measured the output ripple directly on the top of R4 and bottom of RV15, as these are the sense points for the regulator.
Changed R8 (in series with the zener) down to 50 ohms. That helped, (more loop gain), but now there is a tendency to oscillate depending on the input headroom, so that has to go back up some.
It measures a bit better: at 1k, -51dB, at 10k, -39dB. Still not too hot.
OTOH, lets for the sake of argument take -60db because it is a nice round number: 1/1000. Now remember we measured for Zout something like 20milliohms. That x1000 puts the CCS impedance at 20k. So, although it is VERY rough, it doesn't look too improbable, that maybe this is as far we can get with this discrete solution.
Thoughts?
Jan Didden
. You guys really have to get to grips with version control. I have 2 schematics, one layout and one pcb and all four are different....I changed R5 to 3k, so the base current of Q6 and Q5 is less influencing the CCS current. Also measured the output ripple directly on the top of R4 and bottom of RV15, as these are the sense points for the regulator.
Changed R8 (in series with the zener) down to 50 ohms. That helped, (more loop gain), but now there is a tendency to oscillate depending on the input headroom, so that has to go back up some.
It measures a bit better: at 1k, -51dB, at 10k, -39dB. Still not too hot.
OTOH, lets for the sake of argument take -60db because it is a nice round number: 1/1000. Now remember we measured for Zout something like 20milliohms. That x1000 puts the CCS impedance at 20k. So, although it is VERY rough, it doesn't look too improbable, that maybe this is as far we can get with this discrete solution.
Thoughts?
Jan Didden
Perhaps we could disconnect bottom of R5 ( at 3K ) from ground and connect to it a 1.5K resistor to ground. Then connect a 12V zener with a small cap across it from the meeting point of these resistors to i/p voltage 20V
I think this will improve PSRR
A big cap across the zener will help even more.
what you think ?
mike
I think this will improve PSRR
A big cap across the zener will help even more.
what you think ?
mike
Considering the number of members on the group buy this thread is surprisingly quiet? The regulator in it’s current form is “not quite” ready for production as both the voltage regulator and the current source would need some further amendments to be fully functional and achieve their optimal performance, irrespective of how that would compare to other possible designs.
I for my part do not feel quit comfortable with the boards I have ordered (ok I could have looked more closely beforehand) and I find we need to decide where we go from here ???
I for my part do not feel quit comfortable with the boards I have ordered (ok I could have looked more closely beforehand) and I find we need to decide where we go from here ???
Likely most of us don't really know about voltage regulation and noise reduction enough to participate in the conversation. At this point, I am happy to let this process play itself out until there is consensus that the best boards are going to be built. With that, the law of diminishing return should be kept in mind. In other words let's not kill ourselves over minute details that may or may not be noticed in the real world.
In other words let's not kill ourselves over minute details that may or may not be noticed in the real world.
Well my feeling is that the process is coming to a halt. I think that Jan has spent more than enough time on the current prototype that he was testing and that it would be tedious for him to get much further.
Regarding the diminishing returns, if it were just that I would totally agree. Unfortunately, some elements cannot stay as they are right now. Example: the Q4 / Q7 current source eats up way too much voltage, with the ensuing problems, and it's high performance is not even needed in this place. On the other hand the (Q5/Q6) CCS feeding the regulator does not seem to scale well with current and has it's own problems. Lastly, we seem to run into stability issues when raising the gain of the error amp. So amending those areas would be a prerequisite before taking a final decision.
Just my 2 cents.
Jan,
This schematic is what I can reconstruct of changes you have made to the 30V regulator. And if I understand correctly these changes will make this regulator perform better overall.
My question now would be if what you have measured (seen) up to this point will make this regulator work well on audio circuits or would you suggest other topology changes?
This schematic is what I can reconstruct of changes you have made to the 30V regulator. And if I understand correctly these changes will make this regulator perform better overall.
My question now would be if what you have measured (seen) up to this point will make this regulator work well on audio circuits or would you suggest other topology changes?
First of all; thanks a lot to Jan Didden for testing the Toolereg.
The results raises a few questions. First of all, is it worthwhile to pursue the exsisting topology further? A decision must be taken, based on Diddens measurements.
Secondly, how does other shuntreg's measure under the same conditions? ( Ofcourse excluding any preregs that some designs are based on) The spec's of e.g. a Borbely shunt, would be nice to have for comparison.
Since we at this point has no spec's to compare with, I find it a bit of a tuff decision to make.
The results raises a few questions. First of all, is it worthwhile to pursue the exsisting topology further? A decision must be taken, based on Diddens measurements.
Secondly, how does other shuntreg's measure under the same conditions? ( Ofcourse excluding any preregs that some designs are based on) The spec's of e.g. a Borbely shunt, would be nice to have for comparison.
Since we at this point has no spec's to compare with, I find it a bit of a tuff decision to make.
I think we definitely made progress. The changes Tony mentioned above improve the regulator, and I'm sure that it will be a good regulator for audio. Is it the best that is possible, technically? Probably not. Is it the best that is possible, technically, with a relative simple discrete design? I think it is very close.
You guys put a lot of effort in this, and the result is pretty good. It would be a shame to aboandon it and start with something else. What I would like to try one more time is to replace R3 and R10 with more than one diode, to see if that improves the PSRR. But even at -40dB that's an attenuation of 100 times. In addition to that there is also the PSRR of the powered circuit of course.
So, give me a few more days to fit this in.
On the last schematic, shown by Tony above:
If you want to use this for 5V supplies, the zener in series with R8 can be better replaced by an LED. I'll try that as well.
I think you should also make provisions on the board for a resistor in place of VR15. Absolute value is unimportant and a simple resistor is cheaper and more reliable than a pot. The resistor can be easily calculated knowing that it has Vz+0.6V across it.
Are you also planning to revise the pcb? There may be some gain in optimizing the ground loops.
Jan Didden
You guys put a lot of effort in this, and the result is pretty good. It would be a shame to aboandon it and start with something else. What I would like to try one more time is to replace R3 and R10 with more than one diode, to see if that improves the PSRR. But even at -40dB that's an attenuation of 100 times. In addition to that there is also the PSRR of the powered circuit of course.
So, give me a few more days to fit this in.
On the last schematic, shown by Tony above:
If you want to use this for 5V supplies, the zener in series with R8 can be better replaced by an LED. I'll try that as well.
I think you should also make provisions on the board for a resistor in place of VR15. Absolute value is unimportant and a simple resistor is cheaper and more reliable than a pot. The resistor can be easily calculated knowing that it has Vz+0.6V across it.
Are you also planning to revise the pcb? There may be some gain in optimizing the ground loops.
Jan Didden
Hi,
I didn't intend to diss this regulator. The proposed regs (and mine as well) are all relativly close (also add the one of lumba here). They have a main CCS, a voltage reference, a gain boosting stage and a main shunt device. That's it.
We probably only have to chose the stiffest CCs, the lowest impedance and least noisy reference, reasonable gain factor and a large bandwith shunt device, all in the boundaries of KISS.
Just a proposal. I'm not in the GB (yet) but I'm trying the same since some time (making a few parts, best possible shunt reg).
Rüdiger
I didn't intend to diss this regulator. The proposed regs (and mine as well) are all relativly close (also add the one of lumba here). They have a main CCS, a voltage reference, a gain boosting stage and a main shunt device. That's it.
We probably only have to chose the stiffest CCs, the lowest impedance and least noisy reference, reasonable gain factor and a large bandwith shunt device, all in the boundaries of KISS.
Just a proposal. I'm not in the GB (yet) but I'm trying the same since some time (making a few parts, best possible shunt reg).
Rüdiger
Onvinyl said:
apassgear said:
I have search through these and some other discrete designs,
It seems real life performances are not readily available
:1. Zenmod:
Shunty
See post#96 in this thread for performance
http://www.diyaudio.com/forums/showthread.php?postid=1330390#post1330390
His older Shinny
http://www.diyaudio.com/forums/showthread.php?postid=1507958#post1507958
2. G.Kleinschmidt:
"Stability is rendered unconditional by the 470uF output shunt capacitor and PSRR (or ripple rejection) simulates better than 120dB out to 20kHz", also see stimulated graphs in the thread.
http://www.diyaudio.com/forums/showthread.php?postid=1356700#post1356700
3. Wrenchone:
no performance parameters quoted
http://www.diyaudio.com/forums/showthread.php?postid=1731758#post1731758
(see first post for his older design, second post for Salas' shunt
)another version:
http://www.diyaudio.com/forums/showthread.php?postid=735035#post735035
4. Carlos
see his simulated performance
http://www.diyhifi.org/forums/viewtopic.php?f=5&t=1093&st=0&sk=t&sd=a&start=60
(there is also a discrete design from Jam/Fred ??? on the same page)
5. On-semi's TL431
There is a representative schematic in the datasheet PDF. Of course it includes performance figures.
http://www.onsemi.com/pub_link/Collateral/TL431-D.PDF
6. Werner's (TNT) simple two transistor shunt
Stimulated performance graphs.
Ripple rejection:
http://www.tnt-audio.com/clinica/regulators3_ripple_e.html
output impedance:
http://www.tnt-audio.com/clinica/regulators2_impedance4_e.html
Noise:
http://www.tnt-audio.com/clinica/regulators_noise4_e.html
I may have missed some other important designs.
Best Regards,
KF Tam
More measurements.
Changed R8 to 82 ohms + yellow LED (~2V). Works fine, and will allow use for 5V reg.
Changing R3 to LED only increases dropout, not usefull.
Then I replaced R5 with an LM334 CS (Rset=12 ohms, ~6mA). Some improvement, not much.
Then I measured the PSRR directly across the regulator sense points, between top of R4 and bottom of VR15. Bingo: a more than 20dB improvement! PSRR at 1kHz now -65dB. Residual close to noise (~ 60uV).
So the problem is the layout. If you want good performance at the output terminals, you must sense the errors at the output terminals. Top of R8 with its own trace to the output terminal, bottom of VR15 and bottom of ref zener Z with their own traces to the output gnd terminal. E & C Q2 with their own traces directly to the output terminals. Also C2.
I bet that Rout will also be much lower with those changes.
Jan Didden
Changed R8 to 82 ohms + yellow LED (~2V). Works fine, and will allow use for 5V reg.
Changing R3 to LED only increases dropout, not usefull.
Then I replaced R5 with an LM334 CS (Rset=12 ohms, ~6mA). Some improvement, not much.
Then I measured the PSRR directly across the regulator sense points, between top of R4 and bottom of VR15. Bingo: a more than 20dB improvement! PSRR at 1kHz now -65dB. Residual close to noise (~ 60uV).
So the problem is the layout. If you want good performance at the output terminals, you must sense the errors at the output terminals. Top of R8 with its own trace to the output terminal, bottom of VR15 and bottom of ref zener Z with their own traces to the output gnd terminal. E & C Q2 with their own traces directly to the output terminals. Also C2.
I bet that Rout will also be much lower with those changes.
Jan Didden
I have some more suggestions that might help:
Replacing the zener with a diode is a very good idea as it has much lower dynamic impedance but the issue is still the high voltage drop across the CCS. One suggestion would be to reshuffle Q4 and Q7 to form a “one VBE” type current source with a bootstrapped bias resistor. This is shown in the follow-up letter to Walt Jungs current source articles, page 4, fig. 1A: http://www.waltjung.org/PDFs/AX_Letters_0907.pdf - the performance is on page 5 and looks quite good and is as simple as the current solution. And we could use a cheap red LED with lower impedance beneath R8.
A further idea is to put a small source resistor under Q2, one reason is to measure the current, as Allen Wright suggested, but the impedance of the base of Q2 would also be higher and more linear. Maybe this helps with the stability?
P.S. I still think we should look in alternatives to the Q5 Q6 current source as it is not very good at higher currents.
Replacing the zener with a diode is a very good idea as it has much lower dynamic impedance but the issue is still the high voltage drop across the CCS. One suggestion would be to reshuffle Q4 and Q7 to form a “one VBE” type current source with a bootstrapped bias resistor. This is shown in the follow-up letter to Walt Jungs current source articles, page 4, fig. 1A: http://www.waltjung.org/PDFs/AX_Letters_0907.pdf - the performance is on page 5 and looks quite good and is as simple as the current solution. And we could use a cheap red LED with lower impedance beneath R8.
A further idea is to put a small source resistor under Q2, one reason is to measure the current, as Allen Wright suggested, but the impedance of the base of Q2 would also be higher and more linear. Maybe this helps with the stability?
P.S. I still think we should look in alternatives to the Q5 Q6 current source as it is not very good at higher currents.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.