Journal notes, Improving Powered Coil Discharge
GELLER Labs "Backyard Science"
Thoughts on a proton precession magnetometer design - a Proton Magnetometer Project. Build an Earth's field magnetometer.
The FDM MAGNETOMETER1 project is a low cost high performance proton magnetometer (a digital magnetometer) kit under development for universities and amateur scientists to be able to accurately measure and monitor changes in the Earth's total magnetic F field and to observe geomagnetic storms. Magnetic storms can cause large excursions in the field and are of concern to interests ranging from electrical power grids, radio communications, and satellite operations, to aurora watchers and amateur radio operators.
1 Filter Diagonalization Method "FDM" (harmonic inversion), see Jan 21 and Jan 23 entries, based on: Vladimir A. Mandelshtam, Howard S. Taylor, Harmonic inversion of time signals and its applications, Journal of Chemical Physics (1997), Volume 107, Issue 17, 1997, Pages 6756-6769
(be sure to hit refresh to pick up our latest changes and entries)
Project Documentation (very early stages)
Journal Notes:
Monday, December 20, 2010
Overnight: PDF, TXT. Some minor slow excursions in the geomagnetic field overnight were preceded and followed by relatively quiet periods.
SWCTRL BOARD and The COIL ENERGY DUMP: This afternoon I looked more closely at the way we "fast" dump energy from the powered coil (of the counter-wound sensor coil pair). By floating a hp 54645A oscilloscope, I monitored the inductor voltage of the powered coil during coil discharge, by observing the voltage across the dump resistor R8. It turns out that the selection of the IRLIZ24N was fortuitous. The IRLIZ24N is actively clamping the powered coil discharge voltage at 55 V, well below the disharge voltage expectected for the passive clamp of the 402 ohm resistor alone (which completely explains why the dump resistor never gets warm). PDF Here is a more detailed description of these measurements.
Afternote: I suppose as long as one is careful to stay well within the safe operating area (SOA), one could choose any available avalanche voltage (VDSS) with a fully avalanche rated device to set a desired powered coil discharge time. There would still need to be a passive L/R resistive discharge, using the secondary dump resistor, from the VDSS active discharge endpoint to zero. Coil discharge graph I vs time for VDSS 55V and secondary dump R 402 ohms: PDF (17.6 mH, 1.55 A to zero current) Note, for that for this type of unclamped inductive operation, the SOA includes both the Maximum Effective Transient Thermal Impedance curves and the Junction-to-Case, and Maximum Avalanche Energy Vs. Drain Current curves.
Tuesday, December 21, 2010 - Noticed and Corrected High Frequency Current Noise at Powered Coil Discharge
Overnight: PDF, TXT. The geomagnetic field was very quiet over night.
Continuing to look at the polarization cycle in more detail. I added a temporary 0.1 ohm series shunt resistor to look directly at the powered coil current (the previous R8 secondary dump resistor waveforms showed the powered coil voltage). Once again the hp 54645A oscilloscope was floated to make this measurement. As expected, the current falls linearly since Vcoil is clamped at a constant voltage by the power FET avalanche mode at about 55 V. However, the high frequency noise/oscillation just as the FET turns off was a bit of a surprise. PDF (upper waveform discharge current, lowever waveform powered coil voltage)
Afternote: The very high frequency damped oscillation (>100 kHz) shown in the PDF and suppressed by the capacitor described below (< 2us period) is believed not to be directly caused by the powered coil self-resonance. Our powered coil self-resonance is believed to be around 16 kHz (needs to be measured, another job for the to do list).
Adding a 0.1 uF 100 V ceramic capacitor from the FET Drain to the digital ground plane JPG almost completely damped out the undesired high frequency current noise. If design verification shows this is a good change, we will revise the SWCTRL PCB. Note that this capacitor needs to be relatively small, as it causes a short circuit current when the FET turns on. The ESR of the capacitor should limit the i*t product for 0.1 uF to an insignificant pulse for the 14 A device, hopefully without needing to add a small series R. The first microseconds of the current waveform are much improved with the snubber in place PDF. Following the addition of the snubber, we had to reduce the amplifier (NBLNA) gain by about 10% to keep the precession signal below 2 V, so clearly there was a minor improvement in the proton spin alignment. As with other minor changes, it will be interesting to see if there are any noticable changes in the figure of merit (FOM) and/or the number of auto-retry cycles.
It is way too early to be drawing conclusions about the improvements described above, however, since 2 pm, there has been an uprecedented ratio of 86 successful measurements to 102 total measurements. (Afternote: down to 75% by 9 pm, might have just been another quiet EMI/RFI period. Visual inspection of the FID FDM amplitude seems to show some slight increased scatter this evening in this plot from about 2 pm to 9:30 pm PDF). Here is the plot at 1 nanoTesla per division (our daily view is usually 10 nT/div). PDF The spikes were caused by a school bus (observed) and probably some other momentarily stopped vehicle. The down turn after 4:30 was an arriving and parking vehicle, which then departed near 5 pm. Recall that "perfect" operation is not 100%, because the system is designed to reject some measurements at times of known inaccuracy, such as a nearby moving vehicle which can cause errors at least on the order 0.1 nT to 1 nT by causing a gradient field (in time and/or distance) at the counter-wound coil sensor pair. Here it the latest magnetogram (use the lower total "F" field plot) from the Canadian Ottawa observatory (about 200 miles to the North) for the same time period for rough comparison PDF . As an official Canadian government geomagnetic observatory, they have obviously been better able to better place their sensors to avoid nearby vehicles! The same plot a little later at out normal 10 nT resolution (but, still 7 hours as compared our typical 12 hour plot) PDF
Also, added some new pictures on the Docs page.
Wednesday, December 22, 2010 - The FDM Magnetometer Performance has been slightly improved.
Overnight: PDF, TXT, the geomagnetic field was very quiet overnight. Afternoon: sample log spectra PDF, spectra PDF
Yesterday, while continuing to take a closer look at the polarization coil current discharge curve, we noticed and corrected a short period of high frequency noise (microseconds) which ocurred just as the power FET was switched off. The entire powered coil discharge curve is about 500 us long. After a snubber capacitor was added, the noise was much attenuated and the amplitude of the proton precession signal increased by over 10%.
Having run with the slightly improved system overnight, it does appear that the number of auto-retry cycles is down and that the average figure of merit (FDM FOM) has improved now to about 6x10-7. 2x10-6 corresponds to 0.1 nT, so 6x10-7 is about 0.03 nT. FID FDM amplitude PDF and FOM PDF overnight. (The FDM FOM graph is clipped at 2x10-6, since this data is from the plotted data where one of the auto-retry criteria is an FOM less than or equal to 2x10-6.)
Thursday, December 23, 2010
Overnight: PDF, TXT, the geomagnetic field was very quiet again overnight. Here is a graph of FID Amplitude versus measurment number for about the last 1400 plotted measurments PDF. The ambient temperature has generally been in a range of about 20 F to 30 F. The black line shows a 50 point moving average. Last couple of days, several day view PDF. Still running at about a measurement 72% success rate (1616/2235) as compared with the earlier averages around 60%. The field has continued to be relatively quiet, giving us a chance to better evaluate the 12/21 high frequency snubber fix PDF (two hour snap shot, spike was caused by a vehicle).
While momentarity stopped or parked vehicles continue to be a minor annoyance (as disturbing the geomagnetic record), the FDM magnetometer has proved remarkably robust to all sorts of interference from thunderstorms to local EMI/RFI. I attribute this exceptional noise immunity (the precession signal has a peak amplitude of less than 10 uV) to a combination of a very low impedance source, good EMI/RFI filtering, relatively narrow noise bandwidth, and the extremely robust FDM algorithm. Note, however not to confuse noise immunity with undesired magnetic gradients across the sensor coils in time and/or distance. For example, most EFNMR systems designed for geophysics applications (as opposed to in laboratory NMR and EFNMR spectroscopy applications which use gradient coils to restore field uniformity), including our FDM magnetometer, no matter how robustly designed, will not operate with sensor coils indoors (because of the magnetic gradients in most rooms caused by ferrous building materials and iron pipes) or very near to many types of AC power lines. In other words, at some level of inhomogeneity (field gradient), when there is no longer a coherent precession signal, no amount of signal processing will find it.
Friday, December 24, 2010
Overnight: PDF, TXT, the geomagnetic field was quiet again overnight. There was one interesting very slow depression of the field around 11 pm followed by a positive going impulse ~5 nT PDF just before 12 am (0500 UTC). NRCan OTT PDF. Later PDF showing the beginning of the diurnal cycle. The measurment success rate continues at 72% (ratio of accepted 2 minute measurements/(accepted 2 minute measurements plus 8 second retry measurements). This is now improved over our past ~60% rate. Several day view PDF.
The auto-retry process could be sped up to about 5 seconds, however, slowly moving vehicles can take 10 to 15 seconds second to clear our area, so it is not clear that a faster auto-retry rate would be productive. While fast cycling is a possibility for future applications (as discussed in earlier posts), in our present setup our minimum cycle is on the order of 2 seconds polarize, plus 1 second digitize, plus 1 second for signal processing, plus a fraction of a second for various predetermined cycle delays so about 5 seconds minimum. Since the "official" cycle for government geomagnetic observatories is 20 seconds with moving averages, our single point accurate 2 minute measurment cycle / 8 second auto-retry cycle should be fine for now.
Recall that the auto-retry process does not do averaging, it merely accepts or rejects a given measurement, each of our measurements stands on its own. We can easily resolve to seven digits in frequency and magnetic field. Using a GPS stablized signal generator, we can demonstrate absolute accuracy of our FDM frequency estimator to 7 digits. However, beyond consistent observations made during the most extreme quiet geomagnetic periods (generally overnight), at present, we have no way to experimentally quantify the absolute accuracy or noise floor of the FDM magnetometer system as a whole.
As discussed earlier, we like the 2 minute measurment cycle as making for an easy to read 12 hour plot. It remains a topic of interest to add an "automatic transmission" so that on detection of the onset of a geomagnetic disturbance or geomagnetic storm, the FDM magnetometer would automatically move to a faster measurment rate, another project for the to do list!
I was looking again at the precession waveform this afternoon with the hp 3581A wave analyzer and the hp 54645A oscilloscope. Recall that our precession waveform envelope display somewhat mimics the function of the 3581A. The apparent delay in peak amplitude is an artifact of the moving average process. Going back to basics and reviewing the free induction signal (FID) in the time domain (example PDF the vertical lines are an artifact, zooming in revealed a well formed sine wave), I noticed that with improvements of recent months (including the change to (magnet enhanced) fast relay closure following the powered sensor coil discharge) and taking into account the moving average nature of the envelope display, perhaps the 350 millisecond delay before to the start of our 1.1 second (fs=10 kHz) digitization window is now much too conservative. As a test, I moved the start of digitization back to 50 ms from relay contact closure. To account for the now much larger FID amplitude, the NBLNA gain was again reduced, this time by over 20%. It will be interesting to see if there is any change instrument performance, perhaps as indicated by a combination of auto-retry rate and average FDM figure of merit (FOM). Perhaps we were wasting 300 milliseconds of best part of the FID waveform.
Evening: After about 7 1/2 hours of operation, we are running at about a 79% measurement success rate (212/270) with an average FOM of 5x10-7 (~0.03 nT). It will be interesting to see how these numbers hold up overnight. Here is the early part of the overnight run from 8 pm on, PDF (including one distant departed vehicle). Down to about 75% (251/332) at 1:12 am, with the FOM average still holding at 5x10-6 PDF.
Project Documentation (very early stages)
QUESTIONS/COMMENTS/notice of typos, etc. send email to joegeller at gellerlabs dot com
COPYRIGHT © 2009, 2010 JOSEPH M. GELLER, All rights reserved.