Raw data is recorded in dBm, and then we have a script which converts to current noise with the option of integration.
August 20 Update:
We found this note that the Durfee group published in 2011, which already addresses what we've been learning about choosing vales for the so-called "snubber network" for a high-speed op amp driving a capacitive load (see here or here for instance).
Using their values for the snubber (R2 and C3 on the MIT Rev03 board) and output series resistor (R7) when an AD8671 and IRF9Z14 are used, we see a quite well-behaved version of the circuit (no instabilities at low currents, which we had been seeing; and small amplitude noise at high currents).
It does appear that the capacitance added to the LM317 regulator subcircuit
further mitigates the effect of a noise peak appearing as the current
approaches the maximum current setting. Some plots demonstrating this
can be found
here.
Before measuring the linewidth of a laser driven by the MIT/Northwestern Libbrecht-Hall style current controller, we thought it'd be good to investigate some of the noise spectrum characteristics and compare it to the Steck lab current driver (which isn't as quiet since the voltage drop across a four-point sense resistor is amplified during feedback, adding extra noise absent in the LH style driver) and a $3800 ultralow noise current supply from Vescent Photonics.
This has led to some additinal poking to see if other simple changes could reduce the noise of the MITRv03 circuit, which are discussed on this page.
Some documents first, just so they're all in one place.
To measure a current controller's spectral noise density, we use an LED
to mimic the laser diode and convert fluctuations in current to fluctuations
in voltage using a 1/2 Watt, 10 ohm resistor. A capacitor is used as a
dc-block before monitoring the fluctuating voltage on an Tektronix RSA3408A
spectrum analyzer (SA). Not included in the figure below is a 7.5V
Zener diode that protects the spectrum analyzer from an overvoltage
(30 dBm max).
Raw data is recorded in dBm, and then we have a script which converts
to current noise with the option of integration.
To get a quantitative sense of the current noise, we integrate to find the RMS current noise over a particular bandwidth, subtract the associated RMS background current noise and report the result.
We first test the various current supply circuits, after attempting to track down some ground loops in our setup. (We purchased an isolation transformer which served to nicely remove some of the biggest backgrounds. There are still a few resonances in the dc-block circuit described above, but they are fairly constant and subtract away nicely. One interesting tidbit here is that the fluorescent lights in the office where much of the data are taken broadcast a signal of about 48 kHz and result in a large, but easily-removed background.)
The following plot shows traces of current noise density for the following
The improvements between the Rv00 and Rv03 designs are significant, and the Rv03 outperforms the Steck Lab controller at frequencies up to about 1.5 MHz (crossing not included in the above plot). The Vescent supply, meanwhile, is much quieter than any of the homebuilt supplies.
Integrated noise values will be more meaningful later as we see how changes in Rv03 alter current noise, but the above plot
This section is not as significant as the next and can probably be skipped, but it is included anyway.
Mostly for fun, we thought it might be interesting to change the operational amplifier primarily responsible for stability of the LH style circuit. This is IC2 in the MIT Rv03. We tried the two included in the package from Chris Seck, and tried a few other "low noise" amplifiers we had lying around the lab.
There isn't really much to be learned here except to confirm that generic purpose operational amplifiers don't perform as well as their low-noise counterparts (revolutionary!).
Here we begin making small tweaks to the MIT Rv03 circuit to see how the current noise changed, with the goal of potentially reaching Vescent-level stability over many frequencies.
When Chris sent us the circuit, it had an AD711 as IC2 and INA217s as IC3 and IC6. Additional pieces included were an LT1028 amplifier (to try as IC2 and use with a LRF9Z14 MOSFET, which is faster than the original IRF5305), and INA114s to swap out for IC3 and IC6.
After changing out IC6, it seemed reasonable that the modulation section of the circuit could be injecting noise into the current output, and we decided to remove that chip from the circuit entirely (clearly this disables the modulation feature, but it isolates the noise of the driving MOSFET and feedback amplifier nicely).
We also noticed that C4 was not in the original LH design, and had maybe been added at some point to address an instability in the feedback loop.
The following traces are in the plots below, with integrated current noise included (over the range of the plot, 10 kHz to 1 MHz integration bandwidth)
Clearly, removing the modulation section greatly reduces the integrated RMS current noise of the Rv03 circuit. It appears that the instrumenatation amplifiers inject a substantial amount of noise into the circuit. Dan compared the MIT Rv03 design to the original LH current modulation method (see his notes here), and confirmed that IC3 and IC6 are responsible for much of the observed noise.
We acquired an assortment of op-amps to try as IC2 to see if there's a quieter option for this circuit. Here we are using the "simplified RV03", with IC6 and C4 removed, but with the IRF9Z14 MOSFET.
The following traces are in the plots below, with integrated current noise included (over the range of the plot, 10 kHz to 1 MHz integration bandwidth)
So, it appears that the AD8671 and LT1128 are the quietest of the batch, with the Libbrecht-Hall LT1028 being quite a bit worse. The noise bump apparent in the plots rises out of the background as the current approaches the limit setpoint; this will be discussed a bit more in the next section.
Log frequency scale (click for larger image)
A correction the Northwestern group made to the MIT design was adding a bypass capacitor on the LM317 adjustment pin (C11). This improved the issue (that can be seen in Rev00 above) of large current noise on the supply as the setpoint reaches the current limit. This noise peak rises out of the background around 100 kHz as the current is about 80% of the maximum value. As current is increased further, the peak translates to lower frequencies, eventually burying itself in the 1/f noise as the max current is achieved.
To reconfirm the benefits of this bypass resisotr (and to illustrate the trend of a downward-moving noise peak, I'll take data for the quietest op-amp from above, the AD8671 (from Durfee paper), and record traces for current values of 75, 85, and 90 mA (this final trace will be when the current has reached its ceiling set by the I_MAX trimpot). The first plot has no C22 capacitor and has a lot more noise, but note that adding the capacitor doesn't quite kill the noise bump in the second plot.
Data are averaged 500 times and again integrated from 10 kHz to 1 MHz.
The drop in integrated current noise in the last entry is not surprising. This is because the lower bound for our integration band is 10 kHz, and as the current setpoint saturates the maximum value, part of the noise peak is below this lower bound and doesn't contribute.
Log frequency scale with C11 capacitor excluded
