However, I like the idea of using a shunt regulator. If fed from a constant current source, it will draw a constant current, keeping the main supply clean. This is a very interesting property. Although it needs more parts, a current source... but that's not difficult, just a bunch of cheap parts...
So, I stuck a huge booster on the poor thing. See schematic.
The triple-CFP controls the output at high frequencies. It is extremely fast. It looks a bit like a Salas Simple Shunt.
The TL431 controls the triple-CFP, and brings voltage precision, temperature stability, etc.
It is a nested feedback system. The dominant pole of the inner feedback loop (CFP) is created by the output cap, which needs to be low-ESR, low-ESL, and more than about 100µF. No problem with that, I'm designing this regulator to suit the caps I'll use...
The outer feedback loop goes through the TL431. C20, which needs to be around 200-400 µF, filters out the TL431 high frequency noise, and provides the dominant pole.
There are no capacitors which would need to be X7R (ie, annoying values like 100nF or 1µF) anywhere near the feedback loop. Accordingly, when tapping the board with a pencil, it does not produce a HUGE voltage spike like the LT1761 does... (I used a 10nF X7R for the noise bypass cap)...
Of course, I built it following the Tradition of Ugly.
So, I stuck a huge booster on the poor thing. See schematic.
The triple-CFP controls the output at high frequencies. It is extremely fast. It looks a bit like a Salas Simple Shunt.
The TL431 controls the triple-CFP, and brings voltage precision, temperature stability, etc.
It is a nested feedback system. The dominant pole of the inner feedback loop (CFP) is created by the output cap, which needs to be low-ESR, low-ESL, and more than about 100µF. No problem with that, I'm designing this regulator to suit the caps I'll use...
The outer feedback loop goes through the TL431. C20, which needs to be around 200-400 µF, filters out the TL431 high frequency noise, and provides the dominant pole.
There are no capacitors which would need to be X7R (ie, annoying values like 100nF or 1µF) anywhere near the feedback loop. Accordingly, when tapping the board with a pencil, it does not produce a HUGE voltage spike like the LT1761 does... (I used a 10nF X7R for the noise bypass cap)...
Of course, I built it following the Tradition of Ugly.
Attachments
So what is the performance of this ugly bastard ?
First, Noise : about 20µV RMS in the same bandwidth as the previous measurements.
So, thanks to the filtering and the lower current going through the TL431, it actually has less noise with the booster than without... Much of the high-frequency noise has been removed.
First, Noise : about 20µV RMS in the same bandwidth as the previous measurements.
So, thanks to the filtering and the lower current going through the TL431, it actually has less noise with the booster than without... Much of the high-frequency noise has been removed.
Attachments
Now, output impedance...
Black is "raw" TL431, blue is with booster.
Pretty low impedance, identical to simulation results. In the noise floor at DC, and something like 5 mOhm above. Smooth transition to polymer capacitor...
Transient response, same as before. The shunt current is stepped between 23mA and 40mA, a 17mA step. The blue trace shows the load steps.
First scope shot : There is a problem here... The 500µV/div setting on the scope barely shows anything useful... I had to use averaging, which screws up the RMS values on display. The peak to peak measurement seems to agree with the measured impedance, though.
Next 2 scope shots : using x100 preamp confirms the amplitude of voltage variations.
Black is "raw" TL431, blue is with booster.
Pretty low impedance, identical to simulation results. In the noise floor at DC, and something like 5 mOhm above. Smooth transition to polymer capacitor...
Transient response, same as before. The shunt current is stepped between 23mA and 40mA, a 17mA step. The blue trace shows the load steps.
First scope shot : There is a problem here... The 500µV/div setting on the scope barely shows anything useful... I had to use averaging, which screws up the RMS values on display. The peak to peak measurement seems to agree with the measured impedance, though.
Next 2 scope shots : using x100 preamp confirms the amplitude of voltage variations.
Attachments
Yep, the TL431 output voltage. You forgot the feedback loop...
Attached is a scope trace at 500ns/div to show the transient. Nothing to see really. The little undulations after the load step are present also when the regulator is off, so they probably come from the step generator...
Not bad for about $2 in parts...
Attachments
I've played with regulators in a TL431 loop in simulation, and had a similar circuit, but the TL431 would clip itself trying to drive the 470u cap on certain signals. I would make sure it can't clip on any of the signals the regulator might normally receive, and you should specify what kind of load is enough to clip it.
The transresistance of the TL431 alone is about 1/Zout, which is around 7.3R. Since the TL431 is biased at about 1.9mA, 1.9mA*7.3R=14mV - if Vref goes lower than this the TL431 will clip off. You use a 3/4 divider, increasing that to 18mV. So, any time your voltage at R18 decreases by 18mV for longer than 10uA or so (response time of the TL431), your TL431 will clip off. So make sure that doesn't happen in normal operation. One easy way to do this would be to increase the TL431 bias.
With an output impedance of 6mR max within the frequency response of the TL431, any reasonable load variation would be far below this. So you are actually pretty safe, mainly because the booster does a lot of work so the TL431 isn't stressed. In my application I had to lower the TL431 gain with an RC from Vref to ground.
With an output impedance of 6mR max within the frequency response of the TL431, any reasonable load variation would be far below this. So you are actually pretty safe, mainly because the booster does a lot of work so the TL431 isn't stressed. In my application I had to lower the TL431 gain with an RC from Vref to ground.
OK ! I had wondered about that but I didn't understand you were talking about the same thing.
If the load draws huge di/dt this may be a problem. I will measure the di/dt of the load. I could also add a small capacitor between VRef and GND, but the pole would need to be high enough not to make an oscillator out of the 431...
If the load draws huge di/dt this may be a problem. I will measure the di/dt of the load. I could also add a small capacitor between VRef and GND, but the pole would need to be high enough not to make an oscillator out of the 431...
That's why you use an RC instead of just a C. A 150R RC would increase tolerance to over 180mV by increasing the divider by about 11 times.
You could also put a cap across the Vref and anode of the TL431, so it would only regulate the 470uF cap above a certain frequency.
Before you go nuts, consider whether your shunt regulator is better than a cm of wire or not, and if so, use force/sense lines. The circuit is already amazingly fast. Do you know what exactly is being improved, and how much it can be improved before reaching diminishing returns? How will you know when it will be more practical to work on another part of the circuit?
Sure, it has great performance for such a simple circuit !
Using a lower noise reference and an opamp could be nice for AVCC supply. 431 has rather high LF noise.
The reason i'd like to use faster transistors is to use less capacitors. There is not much space around ES9018 for the digital supply decoupling...
Using a lower noise reference and an opamp could be nice for AVCC supply. 431 has rather high LF noise.
The reason i'd like to use faster transistors is to use less capacitors. There is not much space around ES9018 for the digital supply decoupling...
In my simulation your regulator already has less than 1nH of inductance. That's much less than a cm of wire IIRC. That's why I asked whether it was better than a cm of wire or not. The wiring to the DAC will swamp out the reg's virtual inductance unless you use a ground plane. The reg's inductance is probably no worse than that of the cap, so adding the cap may not improve much.
I'm back from the beach ! Nice holidays.
So, I ran a test without the 470u polymer cap at the output, just the 10u X7R 1206 cap. You were right, the 470u is unnecessary. With just the ceramic cap, it works almost as well. No significant change in transient response... Output impedance measurement attached.
I will breadboard a prototype with SMD transistors on a proper low-inductance fixture (this test fixture has rather high inductance, perhaps a few nH). I want to be sure it doesn't oscillate !...
Actually the regulator (as measured) seems to have around 5nH output inductance, test fixture included, which is really very low for an active regulator.
1cm of wire is 5-10 nH depending on the wire diameter, but that's not very interesting since this is a local regulator so it will be on the PCB and close to the chip.
1cm of trace is about 2nH (4 layer, 1mm wide, 0.2mm over ground plane).
I will use power planes (or in the case of VDD, a copper pour under the DAC, opposite side of PCB).
The chip's own pin, mounting and bondwire inductance, is also a few nH.
The inductance of the via connecting the chip's VDD pin to this copper pour is ~ 1nH...
So it isn't useful to go on a capacitor binge, or use any fancy low-ESL ceramics here... it will be swamped by the connection inductance anyway.
For the AVCCs it is different, to prevent crosstalk between the DAC channels, the AVCC should be as stiff as possible...
So, I ran a test without the 470u polymer cap at the output, just the 10u X7R 1206 cap. You were right, the 470u is unnecessary. With just the ceramic cap, it works almost as well. No significant change in transient response... Output impedance measurement attached.
I will breadboard a prototype with SMD transistors on a proper low-inductance fixture (this test fixture has rather high inductance, perhaps a few nH). I want to be sure it doesn't oscillate !...
Actually the regulator (as measured) seems to have around 5nH output inductance, test fixture included, which is really very low for an active regulator.
1cm of wire is 5-10 nH depending on the wire diameter, but that's not very interesting since this is a local regulator so it will be on the PCB and close to the chip.
1cm of trace is about 2nH (4 layer, 1mm wide, 0.2mm over ground plane).
I will use power planes (or in the case of VDD, a copper pour under the DAC, opposite side of PCB).
The chip's own pin, mounting and bondwire inductance, is also a few nH.
The inductance of the via connecting the chip's VDD pin to this copper pour is ~ 1nH...
So it isn't useful to go on a capacitor binge, or use any fancy low-ESL ceramics here... it will be swamped by the connection inductance anyway.
For the AVCCs it is different, to prevent crosstalk between the DAC channels, the AVCC should be as stiff as possible...
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.