Comparison of LM317 capacitors

We wanted to confirm that capacitance on the LM317 voltage regulator affects the current noise behavior of the Libbrecht-Hall-style current controller. We choose the values discussed in this note as a baseline for comparison.

The following alterations have been made to the MIT Rv03 circuit we received from the NW group (some of these are remenants of attempts to reduce the noise in the circuit):

We use an AD8671 as IC2 and a IRF9Z14 as Q_CTL
We remove IC6
We remove wire jumpers and solder parts directly onto the PCB
We add a discharge diode from pin 3 to pin 1 of the LM317
C4 is removed
C3 is set to 33 nF

There are two capacitors in question: one on the adjust pin of the LM317 and one on the output pin (C11 and C7 on the MIT Rv03 board, respectively). The first is the component responsible for drastic stabilization of the MIT circuit by the guys at NW. The second we noticed could stand to be larger based on a discussion of LM317 voltage regulation by Horowitz and Hill (page 345). Previously, we've tried several values, and noticed that increasing beyond several tens of uFs does not affect the behavior. (Note that the "Steck" version below uses an absurdly large capacitance for C7; this is a remnant of confirming the previous statement.) The Durfee group has each of the two capacitors in question valued at 10 uF. These are the four different setups:

"Durfee": C11=10uF, C7=10uF
"Odom00": (or MIT-V00 as in plot below) C11=0 F, C7=0.1uF
"Odom03": (or MIT-V03 as in plot below) C11=22uF, C7=0.1uF
"Steck": C11=22uF, C7=940uF

Linear frequency scale (click for larger image)

Obviously, the Rv00 capacitor configuration is terrible, but note how the larger capacitance values for C7 and C11 reduce/alter the noise peak. Due to real-time observation of the noise behavior as the current is tuned, we are suspicious that the capactor values change how close the current must be to I_max before the noise bump appears. We investigate this next.

For simplicity, we now only look at the Durfee and Steck capacitor values (ie current published status quo vs. our best efforts) to see a direct comparison of changing these capacitors.

The plots below show the nosie peak when the current is within about 5% of the maximum set point. We take a trace at several current values for each of the Steck and Durfee capacitor configurations. They are combined in a single plot at the end for convenience.

Linear frequency scale (click for larger image)

It appears that additional capacitance on the adjust and output pins of the LM317 not only reduce the amplitude of the high-frequency tail of the noise peak, but also push the onset of such a noise peak to current values closer to I_max.

We thought we'd try to see if changing the snubbing capacitance from the Durfee group suggested value altered the noise peak behavior at all. We investigate this by making a rather drastic change from 33 nF to 100 nF. The amplitude of the 50 kHz peak is suppressed nicely in the plot below. We plot the following four traces for visual comparison of the effects of certain capacitors in the circuit:

"Steck": C11=22uF, C7=940uF, C3=33nF
"Durfee": C11=10uF, C7=10uF, C3=33nF
"New snub": C11=22uF, C7=940uF, C3=100nF
"Durfee with new snub": C11=10uF, C7= 10uF, C3=100nF

Linear frequency scale (click for larger image)

Note that the last two traces were taken on a different day than the rest, and that the background in the lab has drifted significantly around 20kHz. Adding the larger snubber capacitor certainly mitigates the noise (compare blue with red, and green with magenta), but the LM317 adjust and output capacitors still have an affect (compare red with magenta).