To model inject a small voltage into the control loop -- in real life this can be done with a wide-band (and expensive) transformer.
Here's one application note with some illustrations: http://www.omicron-lab.com/fileadmi...notes/App_Note_Noninvasive_Stability_V1_0.pdf
See also Rogowski current transformers and Frank Weidmann's discussion of loop gain measurement in LTSpice.
I spent some time on Friday digesting the first 2 of the Tim Green series on op amp stability. Thanks for the references. There's some of it that I need to go back to but I got a good deal of it.
I then spent some time with LTspice last night. Unfortunately Tim's guidance re inserting an inductor and cap in Tina Spice didn't translate well into LTspice, however after looking up a couple of LTspice tutorials and, more importantly, some guidance from Helmut Sennewald on the LTspice Yahoo group I made a lot of progress, the results of which are rather interesting.
Helmut was kind enough to edit my botched attempt at .ac analysis (although in the interim I had made better progress). He also edited the circuit to fix a few things as mine, which did not include the 22uF cap between J310 source and ref of the SPX431, was unstable. (Recall Lasse's tests, namely the one he labelled 302EX, cap C3.)
One of the things he changed was to replace the J310 with a simple 470 ohm resistor. He said: "I don't see any advantage from J310 with Vds of 1V and Vgs=0V. It just behave like a resistor with +100%/-50% tolerance. Nobody likes that." I'm still trying to get my head around the comment but the guy knows his stuff.
Second, rather than using a 22uF cap between the adj pin of the LM317 and the ref pin of the TL431/SPX431, he "added a RC compensation to get less than -180° at 0dB loop gain." The resulting circuit (with the AC injection) is below:
It is interesting to compare the transient response of this circuit with one where R6, C2 and R5 are replaced with a 22uF cap.
First, the circuit above:
Then with the 22uF cap instead:
Looks better to me.
I'd like to ensure that I understand the RC network that has been inserted. R5 and C2 create a low pass filter into the ref of the TL431 so high frequency noise gets shunted into the 47K resistor???? Doesn't seem right as better to shunt the noise to ground. So I am confused. And I certainly don't understand how it helps stability. (Sometimes I feel like I will never learn enough to stop taking in water…)
S
PS Here is the AOL chart for the above circuit:
I then spent some time with LTspice last night. Unfortunately Tim's guidance re inserting an inductor and cap in Tina Spice didn't translate well into LTspice, however after looking up a couple of LTspice tutorials and, more importantly, some guidance from Helmut Sennewald on the LTspice Yahoo group I made a lot of progress, the results of which are rather interesting.
Helmut was kind enough to edit my botched attempt at .ac analysis (although in the interim I had made better progress). He also edited the circuit to fix a few things as mine, which did not include the 22uF cap between J310 source and ref of the SPX431, was unstable. (Recall Lasse's tests, namely the one he labelled 302EX, cap C3.)
One of the things he changed was to replace the J310 with a simple 470 ohm resistor. He said: "I don't see any advantage from J310 with Vds of 1V and Vgs=0V. It just behave like a resistor with +100%/-50% tolerance. Nobody likes that." I'm still trying to get my head around the comment but the guy knows his stuff.
Second, rather than using a 22uF cap between the adj pin of the LM317 and the ref pin of the TL431/SPX431, he "added a RC compensation to get less than -180° at 0dB loop gain." The resulting circuit (with the AC injection) is below:
An externally hosted image should be here but it was not working when we last tested it.
It is interesting to compare the transient response of this circuit with one where R6, C2 and R5 are replaced with a 22uF cap.
First, the circuit above:
An externally hosted image should be here but it was not working when we last tested it.
Then with the 22uF cap instead:
An externally hosted image should be here but it was not working when we last tested it.
Looks better to me.
I'd like to ensure that I understand the RC network that has been inserted. R5 and C2 create a low pass filter into the ref of the TL431 so high frequency noise gets shunted into the 47K resistor???? Doesn't seem right as better to shunt the noise to ground. So I am confused. And I certainly don't understand how it helps stability. (Sometimes I feel like I will never learn enough to stop taking in water…)
S
PS Here is the AOL chart for the above circuit:
An externally hosted image should be here but it was not working when we last tested it.
Last edited:
Not sure if you meant R5 or R6. R6+C2 form a local loop around the 431 and likely blow off some of its loop gain, which is a common (but always not ideal) solution to stability in composite amplifiers. To see what specifically they're doing with respect to gain and phase margin just change the circuit and re-sim.
Does this make sense for the former?
An externally hosted image should be here but it was not working when we last tested it.
LTSPICE has a feature that lets you plot the power dissipation of primitive elements (such as resistors) directly. You hold down some combination of SHIFT, CTRL, ALT keys while either clicking the left or right mouse button on a resistor, and a green-and-red thermometer icon appears. This indicates you have selected the "plot power dissipation" option. And what appears in the plot window is power, in watts.
The user manual and/or help file can assist your search for the magical combination of keypresses, or you could just experiment.
I prefer to select resistors whose wattage rating is 3X higher than the expected operating conditions; that way they'll be cool to the touch, and will operate reliably for a very long time. But I'm not a pinchpenny mass market manufacturer competing in a cutthroat market where halfpennies matter.
The user manual and/or help file can assist your search for the magical combination of keypresses, or you could just experiment.
I prefer to select resistors whose wattage rating is 3X higher than the expected operating conditions; that way they'll be cool to the touch, and will operate reliably for a very long time. But I'm not a pinchpenny mass market manufacturer competing in a cutthroat market where halfpennies matter.
. Average dissipation is also displayed in the status bar while you hover the cursor over a specific part.
I'm using the Mac version, the one where you have to guess secret key strokes to display stuff. In this case shift left click works. I can plot power dissipation. I believe I was just confusing myself by thinking that power dissipation of resistor X would depend on the load the regulator is driving. But the currents through the resistors would remain the same regardless of load. It's only the LM317 which has variable power dissipation. Duh.
Am I modelling output impedance correctly?
Is it worth trying to improve this? Small bypass cap on output?
Am I modelling output impedance correctly?
An externally hosted image should be here but it was not working when we last tested it.
Is it worth trying to improve this? Small bypass cap on output?
Mostly. The denominator should be the total load current, not just I2. It's also worth checking the 317 on its own is in reasonable agreement with the datasheet as the part model may not be accurate in this regard.
Oh, it makes quite a difference from a few MHz up. There's an abundant literature on bypass cap locality and ESL which you may want to consult to understand what's missing in the sim.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.