Correct. There is a small difference at high frequency that is unimportant. Funny thing is, that was exactly my starting point in the simulation, but afterwards R11 was moved to the Q5 collector. Not enough time to play with.
Ok.
About a no compromise negative power supply using TL431:
I think I found a way. Using a current mirror.
The anode of TL431 is tied to Vout-, the REF input tied to a two resistors divider, just the same as done in the positive psu.
The current mirror is tied to ground.
The cathode of TL431 sends its current to the mirror, the mirror sends the same current value to the pass transistor. For the pass transistor this looks as if it were connected to the Cathode of a complementary TL431.
In other words, the current mirror makes a complementary TL431.
Current mirrors made of discrete components are not accurate, but I hope accuracy is not needed here. I hope, a two transistors mirror tightly thermal binded should be enough.
There are certainly variants, and I think solutions that do fully solve the load regulation issue will use some form of current mirroring.
About a no compromise negative power supply using TL431:
I think I found a way. Using a current mirror.
The anode of TL431 is tied to Vout-, the REF input tied to a two resistors divider, just the same as done in the positive psu.
The current mirror is tied to ground.
The cathode of TL431 sends its current to the mirror, the mirror sends the same current value to the pass transistor. For the pass transistor this looks as if it were connected to the Cathode of a complementary TL431.
In other words, the current mirror makes a complementary TL431.
Current mirrors made of discrete components are not accurate, but I hope accuracy is not needed here. I hope, a two transistors mirror tightly thermal binded should be enough.
There are certainly variants, and I think solutions that do fully solve the load regulation issue will use some form of current mirroring.
calculate a current mirror's "sensitivity" to VBE mismatch; this is (partial derivative of (I1-I2) with respect to (VBE1-VBE2)); for several different values of degeneration resistor. It may elicit a Eureka! response
All roads lead to Rome.
I started looking in the voltage inverter and current mirror solutions as well.
I started looking in the voltage inverter and current mirror solutions as well.
I am not familiar with current mirrors.
According to Rod Eliott, ESP, my bible, degeneration resistors don't help much.
He says the Beta and Vbe should be matched firstly.
I see that one transistor will have little self heating ( the one with the base tied to the collector ) while the other will have a lot, Vce close to the PSU voltage, so there is no hope for equal junction temperatures. Vbe cannot be the same, how can this work ? And it is certainly like this in other current mirror applications. I am puzzled.
According to Rod Eliott, ESP, my bible, degeneration resistors don't help much.
He says the Beta and Vbe should be matched firstly.
I see that one transistor will have little self heating ( the one with the base tied to the collector ) while the other will have a lot, Vce close to the PSU voltage, so there is no hope for equal junction temperatures. Vbe cannot be the same, how can this work ? And it is certainly like this in other current mirror applications. I am puzzled.
It is a question to ask your guru, Rod Elliott.
Circuit design textbooks cover current mirrors relatively thoroughly, but if you want Rod Elliott's explanations and tutoring, there's only one place to get it. From Rod Elliott himself.
_
Circuit design textbooks cover current mirrors relatively thoroughly, but if you want Rod Elliott's explanations and tutoring, there's only one place to get it. From Rod Elliott himself.
_
Last edited:
Thanks. Good idea about using RC to drop the voltage. Assuming 39R to get to 24V what size C would you suggest? Looks like I'm pretty limited in non-crapicon >25V caps too.
Some other points on the general implementation that seem a bit unclear:
How precisely do R4 and R5 need to be 1K8 and 180K? I got 5%s which measure at 1750ohms and 177Kohms. Alternatively I could use trimpots.
Does the type of cap used at C2 (22uF) matter? I've got Nippon Chemicon SHL and SMS ones I can't find datasheets for.
Does 270uF Nichicon HV and a 1uF film sound good for input? Have a wide variety of films, electrolytics not so much.
Assuming I can't find proper fixed resistors to set VOUT, is it preferable to use a pot for R1 or R2?
Is the "protection diode" that's ubiquitous in the the LM317 datasheet not necessary in this design?
Hopefully that's all.
I simulated the negative regulator as said in post #1102.
I get exactly the same results as the positive regulator.
I am amazed to see the very same PSRR figures.
The mirror does mirror the TL431. As expected it makes a complementary TL431.
The principle works.
But, here the simulation gives a perfect current mirror, the two transistors are perfectly matched and at the same temperature.
The work to be done is to see the sensitivity of the circuit to not so perfect mirrors
I get exactly the same results as the positive regulator.
I am amazed to see the very same PSRR figures.
The mirror does mirror the TL431. As expected it makes a complementary TL431.
The principle works.
But, here the simulation gives a perfect current mirror, the two transistors are perfectly matched and at the same temperature.
The work to be done is to see the sensitivity of the circuit to not so perfect mirrors
Here low ESR should be beneficial. Your suggestion of 270uF+1uF should work.
These are just fine.
This should work also.
R2
You can add it. Please remember that Elvee's original circuit is just an add-on to existing LM317 power supplies. It was not meant to be an example of a complete PS.
How about uploading some of the asc files showing those results with the TL431s?
Then we can also try them and simulate.
Then we can also try them and simulate.
How to make a negative regulator from an existing positive regulator based on TL431.
( A positive regulator like at post #1040 )
A difficulty is: Complementary TL431 does not exist.
The trick is to use a current mirror.
The procedure is 3 steps.
Step 1:
Implement a complementary of the pass circuitry at the negative rail. Using a complementary pass single transistor or Sziklai or Darlington with a bias resistor or a CCS.
Step 2:
Duplicate the TL431 and associated circuitry from the positive regulator. The original sits between ground and the positive rail, set the copy between the negative rail and ground.
Step 3:
Connect the cathode of the TL431 to the pass transistor via a current mirror. The current mirror sits at ground, the TL431 cathode sends it's current to the mirror, the mirror sends a current of same value to the pass circuitry.
Here is the magic: The pass transistor sees through the mirror a complementary circuitry of TL431 with the resistor divider and the denoiser.
This works like wonder in simulation. The only difference is stability not as good, which was expected because of some delay from the addition in the regulation loop of the current mirror.
So here we need a very good current mirror.
A real one has not equal current, some current offset, some gain current not exactly 1, some light distortion. All these are swallowed by the loop gain.
The phase shift is an issue because it directly decreases the phase margin of the loop.
In this design with typical component values the 0dB open loop gain is around 250KHz. So we need a current mirror with very low phase shift around this frequency.
( A positive regulator like at post #1040 )
A difficulty is: Complementary TL431 does not exist.
The trick is to use a current mirror.
The procedure is 3 steps.
Step 1:
Implement a complementary of the pass circuitry at the negative rail. Using a complementary pass single transistor or Sziklai or Darlington with a bias resistor or a CCS.
Step 2:
Duplicate the TL431 and associated circuitry from the positive regulator. The original sits between ground and the positive rail, set the copy between the negative rail and ground.
Step 3:
Connect the cathode of the TL431 to the pass transistor via a current mirror. The current mirror sits at ground, the TL431 cathode sends it's current to the mirror, the mirror sends a current of same value to the pass circuitry.
Here is the magic: The pass transistor sees through the mirror a complementary circuitry of TL431 with the resistor divider and the denoiser.
This works like wonder in simulation. The only difference is stability not as good, which was expected because of some delay from the addition in the regulation loop of the current mirror.
So here we need a very good current mirror.
A real one has not equal current, some current offset, some gain current not exactly 1, some light distortion. All these are swallowed by the loop gain.
The phase shift is an issue because it directly decreases the phase margin of the loop.
In this design with typical component values the 0dB open loop gain is around 250KHz. So we need a current mirror with very low phase shift around this frequency.
Last edited:
"Current feedback opamps" with bandwidths of >100 MHz certainly do exist, and have been commercially available starting in 1999. Since these amplifiers depend upon current mirrors for their successful implementation, it seems to me that's conclusive evidence: High bandwidth current mirrors, with low phase shift at 0.25 MHz, do exist.
_
_
Attachments
Indeed ICs highly use current mirrors, an ideal situation because on the same die.
Mark, I followed your advice about resistor degeneration.
Simulating with the default transistors with no degeneration the negative regulator using LT431 with a current mirror works fine.
With 2N3906 models, fine too, adding 100ohm degeneration, fine too. But with 1000 ohm degeneration, it rings. I think this comes from the transistor output capacitance. So, what is needed is a low capacitance transistor, well, a transistor good at a VAS.
Replacing one of the two transistors by other transistor models, makes no difference, this proves to me there is no need for transistor matching. I can even replace the transistor that has it's base wired to it's collector by a diode.
All this gives solutions for a real current mirror that is good in this design.
Mark, I followed your advice about resistor degeneration.
Simulating with the default transistors with no degeneration the negative regulator using LT431 with a current mirror works fine.
With 2N3906 models, fine too, adding 100ohm degeneration, fine too. But with 1000 ohm degeneration, it rings. I think this comes from the transistor output capacitance. So, what is needed is a low capacitance transistor, well, a transistor good at a VAS.
Replacing one of the two transistors by other transistor models, makes no difference, this proves to me there is no need for transistor matching. I can even replace the transistor that has it's base wired to it's collector by a diode.
All this gives solutions for a real current mirror that is good in this design.
The trouble is my LTSpice is on a notebook under Windoze XP. DIYaudio is not accessible using Chrome which is the browser with least drawbacks under WinXp.
Sharing my files is a tedious extra work.
There is not much to do for simulating.
Use the schematic in post #1040.
From this there is not much to change following my detailed explanation to get at my other simulations.
I looked at stability with a Tian probe, this is a bit involved and is worth sharing. Meanwhile googling with "Middlebrook's method VS Tian's...." one can find, well explained, how to implement a Tian probe in LTSpice.
Here is a well explained, how to install the Tian probe in LTSpice.
loop gain - Middlebrook's Method VS Tian's
Method VS Middlebrook's General Feedback Theorem - Electrical Engineering Stack Exchange
The tedious work is typing the long equation. Once debugged and saved in the plot settings you are all done.
To turn off the probe actually needs .param prb = 0
I inserted the probe at the REF pin of the TL431 with it's current source away from that pin.
loop gain - Middlebrook's Method VS Tian's
Method VS Middlebrook's General Feedback Theorem - Electrical Engineering Stack Exchange
The tedious work is typing the long equation. Once debugged and saved in the plot settings you are all done.
To turn off the probe actually needs .param prb = 0
I inserted the probe at the REF pin of the TL431 with it's current source away from that pin.
Last edited:
LTSpice simulation is attached. This is still not the actual circuit to build, but is close enough.
As for the current mirror, unanswered question is why to use this implementation? What are the benefits, how is load and voltage regulation? In my intended usage scenario with 1,5 V headroom this works weird.
In the attached simulation, positive and negative regulator results are well matched.
As for the current mirror, unanswered question is why to use this implementation? What are the benefits, how is load and voltage regulation? In my intended usage scenario with 1,5 V headroom this works weird.
In the attached simulation, positive and negative regulator results are well matched.
My goal is explained in the post #1092. Based on the actual chosen semiconductors, some part values may differ and schematics may be altered. As said, it is close enough.
OK, first simulation with Denoised TL431.
What's the big deal about it?
Just on my first try, PSRR is worst than any Denoiser or Dienoiser with 317 or any other regulator.
Didn't get into noise sims yet, but the question is because it can be used with higher voltages?
Please, I'm not criticizing. I just want to know what's being looked for with this arrangement.
What's the big deal about it?
Just on my first try, PSRR is worst than any Denoiser or Dienoiser with 317 or any other regulator.
Didn't get into noise sims yet, but the question is because it can be used with higher voltages?
Please, I'm not criticizing. I just want to know what's being looked for with this arrangement.
OK. Please, present solution with LM regulators that delivers:
Several A output
Over 100 dB PSSR at that currents
Stability with capacitive load ESR from 2 mΩ to ohm range
0,1 uV total noise
Much less than 1 mΩ impedance up to 1 MHz
Fast transient response
All that with only 1,5 V drop
And finally, some people don’t like completely IC voltage regulators, like some are bent on Jung types.
You have simulation so play with it.
Several A output
Over 100 dB PSSR at that currents
Stability with capacitive load ESR from 2 mΩ to ohm range
0,1 uV total noise
Much less than 1 mΩ impedance up to 1 MHz
Fast transient response
All that with only 1,5 V drop
And finally, some people don’t like completely IC voltage regulators, like some are bent on Jung types.
You have simulation so play with it.
Last edited: