Build an Earth s Field Magnetic Observatory!

Transcription

1 P a g e 1 Build an Earth s Field Magnetic Observatory! PART VII, Software Most of the FDM magnetometer signal processing, as well as the display functions are performed by a LabView Windows executable program. Our present plan is to distribute the compiled version of the LabView program as a Windows executable, so that builders do not need to own a LabView development system to build and operate our FDM magnetometer kit. 1 Our LabView program controls a NI USB 6008 (12 bit ADC and digital I/O module) 2. The USB 6008 digital outputs control one of our two hardware modules, the switch control (SWCTRL) module. Each measurement cycle, the LabView program calls a frequency estimator, the FDM Windows executable. 3 The FDM executable is a very sophisticated, yet easy to use ultra high resolution frequency estimator. Our customized FDM executable reads as input, digitized data in the time domain (the precession signal), and outputs a single frequency with milli Hz resolution (the Larmor frequency). Professor Mandelshtam of UC Irvine most generously provided FDM source code from his landmark work 4 in mathematical methods related to NMR spectroscopy for the FDM magnetometer project. Our FDM windows executable was compiled from a modified version of the professor s FDM FORTRAN source code. 5 MEASUREMENT FLOW CHART The measurement flow chart below shows the operation of the FDM magnetometer software. With the exception of the FDM executable, the blocks represent functions performed by the LabView program. The time line, as described in more detail below, provides the timing for both the USB-6008 and SWCTRL modules.

2 P a g e 2 A sample display screen from the current version of our working FDM proton precession magnetometer prototype.

3 P a g e 3

4 P a g e 4 The digitizer digitizes the proton precession waveform, also known as the free induction decay (FID) signal. The digitizer is software triggered at the end of the running of the time line. The LabView program reads the signal sampled points from the digitizer and saves the integer values to a signal data text file. The LabView program also uses this one dimensional (vector) 1D data array for further processing and to generate the waveform displays. The precession waveform envelope is computed and displayed by the LabView program and the raw signal modified by an exponential window (apodization) and both waveforms are displayed on the front panel. Each measurement cycle, the LabView program calls the FDM executable, which reads the signal data file and returns a single fundamental frequency, a figure of merit (FOM), and a narrow band signal to noise ratio (NB S/N). These values are saved to the field_data text file. The auto-retry process (within the LabView program) evaluates the returned FDM data to test if it is within acceptable limits. If within limits, on the following measurement cycle after a measurement delay of about 2 minutes, the values are added to a plot data array and also plotted to the display graph and written to a field_data_plot text file. If the most recent acquired data is out of limits, the auto-retry process repeats the measurement at a relatively fast rated (e.g. every 8 seconds), until an acceptable measurement is achieved. revised to here 3/20/11 INITIALIZATION The following section of the LabView program is for program initialization. Initialization (not shown in the measurement flow diagram above) precedes the running of the time line. The two flat sequence structures at the left open the field_data.txt and field_data_plot.txt files. The green wire from both of these boxes carries the file name reference number (refnum) into the main program while loop. The flat sequence structure at the bottom of the initialization drawing reads a user edited file FDM_FRNG which allows the user to set the FDM frequency range. This frequency range should be a 300 Hz range roughly centered at the local expected fundamental Larmor frequency. It can be slightly skewed should there be an especially problematic interfering spur at the low or high edge of the frequency range of interest. For example, our local Larmor frequency is presently around 2287 Hz and we are using a range entered in khz in the FDM_FRNG.txt file as: 2.1d0 2.4d0 (2100 Hz to 2400 Hz). In our experience, for reliable operation, once robust operating parameters are found, it is best to leave them set as is, rather than to continue to attempt to tune them.

5 P a g e 5 Thus, some parameters, while user settable, once set, are best left alone as a fixed file entry (that is why they are not an active front panel control).

6 P a g e 6

7 P a g e 7 The multiplication factor above the flat sequence structure at the bottom of the initialization drawing is the correction for the actual measured sample rate of any given USB 6008 digitizer (due to the fixed clock manufacturing scale error). We have found the USB 6008 clock to be stable with time and temperature to better than a ppm, however, for best reporting of the Larmor frequency, where an experimenter can measure the scale error of the clock frequency (as with a calibrated signal generator, or synthesizer using a GPS disciplined oscillator frequency reference) the scale error should be entered. This sample rate scale factor is a work in progress and needs to be moved either to a readable file or a front panel user entry. Our USB 6008 sample frequency was found to be low by a scale factor of about 3.54x10-5 (i.e. when we software set the USB 6008 sample rate to 10 khz, it is actually 10, Hz). The corrected sample rate is sent on to later routines that take actual sample rate as an input. The time functions above the sample rate correction provides actual time in an initialized time array to the program while loop. Also, the number of samples to be digitized by the USB 6008 and then used by the following routines is defined here as 11,000 points (another user settable factor can be added later). MEASUREMENT TIME LINE The running of the measurement time line refers to the sequenced operation of the zero-current hybrid FET circuit of the switch control module (SWCTRL). At present, the time delay values for each sequence block are fixed programmed, however some or all of these values could also be converted over to user settable values for the final distribution software. In our present design, the small signal relay is de-energized at the beginning of the measurement time line. A delay of 200 milliseconds ensures that the relay contacts are in a stable normally opened position (relay coil deenergized). Since the polarization circuit for the powered coil (of the counterwound coil pair) uses the relay s normally open contacts, the powered coil can now be energized by turning the power FET on. Following the polarization cycle (typically 2 seconds), the power FET is turned off. Once the power FET is turned off, the powered coil discharges via the dump resistor on the SWCTRL board, and the other resistances in the system, including the coil and cable resistance. There can be some oscillatory nature in the form of an RLC discharge since although the resonating capacitance is not in the circuit during the polarization cycle, there can be significant capacitance

8 P a g e 8 attributed both to the powered coil inter winding capacitance as well as the capacitance of the sensor cable. By the time the coil discharge delay time of 20 milliseconds has passed, the powered coil has been completely discharged. With zero coil current, the relay coil can now be re-energized. Notice that the relay contacts never change state while there is coil current flowing for zero-current relay switching. Zero-current relay switching is very important to provide a very long relay life, not limited by electrical contact wear (pitting due to arcing, etc.) and limited only by the mechanical life of the moving contacts as related to metal fatigue. In the normally closed position, the relay configures the counter-wound coiled pair as a center-tapped coil (analogous to a center-tapped transformer secondary winding). The center-tapped coil pair is configured in parallel with the resonating capacitor bank and coupled to the input of the NBLNA differential input.

9 P a g e 9

10 P a g e 10 Now turning to the measurement time line section of the LabView program below, it can be seen that for convenience, we used higher level LabView Express vi s for both USB 6008 writes and reads as well as for the timers. In the future we might break one or more of these Express vi s into a more basic LabView G code for efficiency. However, early compiled versions have performed well with no noticeable delays, therefore conversion of the Express vi s to raw G code might not be needed. The Boolean F to the Relay OPEN Express vi sends a low via a USB 6008 digital output (Port 0, line 6 (P0.6)) to the input relay 2N3904 transistor causing the relay on the SWCTRL board to de-energize. The Wait for Relay to Close Express time delay vi provides the first 200 ms delay. The Polarize ON Express vi sets Port 0, line 7 (P0.7) to low (Boolean F ) which turns ON the power FET via an inverting 2N3905 (or 2N3906) PNP inverter stage on the SWCTRL board. The ALIGN PROTONS time delay Express vi sets the polarization time (typically 2 seconds, probably later user settable). Following the ALIGN PROTONS time delay, the Polarize OFF Express vi sets Port 0, line 7 (P0.7) to high (Boolean T ) which turns the power FET OFF. The Wait for Discharge of Sensor Coils time delay vi pauses while the powered coil (of the counter-wound coil pair) completely discharges. For those less familiar with inductive circuits, note that a coil is charged with a current, just as the more familiar situation where a capacitor is charged with a voltage. Following complete discharge of the powered coil, the relay coil is re-energized to close the relay (contacts in the normally closed position) by the Relay Close Express vi setting the USB 6008 digital output (Port 0, line 6 (P0.6)) sending a high to the input relay 2N3904 transistor causing the relay on the SWCTRL board to energize. Note that when Port 0, line 6 (P0.6) is set to high, current flows to the 2N3904 relay control transistor base via a 4.7 kilo ohm pull up resistor within the USB Finally, following a Pause Before Observing Exponential Decay of Proton Spin Alignment time delay vi, the DAQ assistant Express vi commands the USB to digitize the precession (free induction decay (FID)) signal, presently for 11,000 samples at 10 khz sample rate for a measurement window of 1.1 seconds.

11 P a g e 11

12 P a g e 12 FID SIGNAL PROCESSING The digitized free induction decay (FID) signal (the precession signal) is processed separately in two paths. One path generates the diagnostic displays of the precession waveform filtered envelope and the precession waveform with apodization (the FID multiplied by an exponential window). Another path builds a data input file for the filter diagonalization method (FDM) executable and conveys the output of the FDM procedure to a standard output which is then read by the LabView vi as it resumes execution and further processes the FDM results. Note that using FDM, generally more points are better. There is no need to use 2 N points as with some Fourier based frequency estimators. Also windowing is not used to control leakage as in Fourier methods. We chose to multiply our FID window by an exponential waveform for several reasons. First, the exponential window causes additional emphasis on the earlier part of the FID waveform having a better signal to noise ratio than the latter portion of the partially digitized (first 1.1 seconds) FID waveform. Also, since FDM solves only for a decaying waveform, the exponential window allows for use of a continuous wave (CW, constant amplitude) calibration signal. In some very unusual circumstances, such as in very high noise areas, there might be some advantage to reducing or eliminating apodization. We begin a more detailed description of FID signal processing with the routine that provides the precession waveform filtered envelope display on the application display screen. The purpose of this signal processing is to roughly emulate the combination of a frequency selective voltmeter and an oscilloscope display of AC amplitude over a narrow frequency band about the fundamental frequency. To create this display, first take the absolute value of all of the digitized data. Then in a moving window process, the RMS voltages of each subset of points (the moving window) are calculated, and finally we apply a low pass filter to produce the envelope display. The precession waveform filtered envelope display can be helpful for tuning the resonating capacitor bank, as well as for monitoring the health of the instrument and evaluating interfering signals that can cause the envelope to become distorted.

13 P a g e 13

14 P a g e 14 The section at the top of the LabView vi below shows the precession waveform filtered envelope display routine. The first FOR loop creates an array of the absolute value of the signal waveform digitized data. The second FOR loop calculates the RMS value for a moving window (here 600 points wide). Finally, a low pass FIR filter provides the final envelope display. FID SIGNAL FILE (READ BY THE FDM EXECUTABLE) Another branch of the overall vi, the lower section of the vi page, shows the creation of the signal text file which is read each measurement cycle by the FDM executable. The present offset integer format of the signal file is as follows with zero set at 8192: d0 2.4d (11,000 points total) The first line has the number of samples followed by the exact (actual) sample period (1/f sample ). The third line (second line blank) is the FDM frequency range as provided in the user file (a user edited file FDM_FRNG, as described above). This signal text file is then written as the file signal (no file extension) and the file is closed. The signal text file is written or overwritten every measurement cycle.

15 P a g e 15

16 P a g e 16 Referring to the LabView drawing below, the LabView call to the FDM executable is now described in more detail. In the first frame of the flat sequence structure (lower left), the LabView System Executable (System Exec) vi calls the FDM executable. The FDM executable reads the signal file written and described above. The FDM executable outputs as standard output an ASCII string including the FDM precession frequency, FDM figure of merit (FOM), FDM narrow band signal to noise ratio (S/N), and FDM amplitude. The LabView System Exec vi accepts the ASCII string into the overall FDM magnetometer vi and passes it to the second frame in the flat sequence structure, which parses the FDM output variables and converts them to real numbers.

17 P a g e 17

18 P a g e 18 AUTO-RETRY The Boolean logic to the right of the second frame in the flat sequence structure performs the auto-retry operation as shown in the Auto-Retry block diagram below. The concept behind auto-retry is that in addition to computing the fundamental frequency of the FID signal, FDM also provides an estimate of the quality of each measurement. The figure or merit (FOM) provides a rough estimate of the confidence of the measurement to a desired resolution. Our FDM FOM cutoff of 2x10-6 corresponds roughly to 0.1 nt. Since a passing vehicle causes a spread in the Larmor frequency over a digitization interval, the FOM filter largely rejects measurements made in the presence of nearby moving vehicles. (Unfortunately once the vehicle is stopped, the FOM filter can no longer filter the vehicle presence and then there is a corresponding shift in the magnetic field reflecting the new actual total field F scalar at the counter-wound coil sensor pair, as caused by the vehicle presence.) The S/N cutoff value of a S/N ratio of 5 db or better, helps to filter out occasional in band noise sources, since the presence of a large in band interfering signal is generally accompanied by a family of in band signal spurs. Since a properly installed and operating system generates a relatively high amplitude FID signal, the 100 mv amplitude threshold further filters out failed FID signal acquisitions, whatever the cause. If a given measurement survives the auto-retry evaluation process, the new data is added to the current measurement array and the normal measurement cycle timer is started (typically 2 minutes). On the successive measurement, the new data is added to the field_data_plot file as well as plotted on the FDM magnetometer front panel magnetogram. This one measurement cycle delay allows for some additional post processing (not yet implemented). If a given measurement does not survive the auto-retry evaluation, the measurement cycle is immediately restarted. Note that with the ultra high resolution capability of FDM, there is no averaging needed nor used. Once a given single measurement passes the autoretry cutoff values, it is the final measurement reported by the instrument. Depending on EMI/RFI local conditions and artificial changes in the local total magnetic F scalar (e.g. nearby vehicle traffic), there can be on average fewer or greater auto-retries. We find in our local home office installation that over 24 hour period, very roughly, 40% or our measurements result in auto-retry operations.

19 P a g e 19

20 P a g e 20

21 P a g e 21 Peak Envelope Voltage (PEV) Servo The peak amplitude of the exponentially decaying precession signal is a strong function of outdoor air temperature. This is because tau 1 of the working NMR fluid changes with temperature. For example, when tau1 is relatively small at lower temperatures, a higher percentage of the proton spins are aligned in a shorter time. The result is a widely varying peak envelope voltage over temperature for a fixed polarization time. One way to manage the peak envelope voltage within a desired range is to manually adjust the NBLNA gain to give a desired peak voltage at some temperature. When the envelope becomes too small or too large at other temperatures, the gain can be further adjusted. However, to avoid the need for periodic gain adjustments, we implemented a negative feedback loop (a servo ) to continuously hold a relatively constant peak envelope voltage. Instead of automatically adjusting the NBLNA gain, we programmatically adjust the polarize time. For a lower amplitude free induction decay (FID) signal (the precession waveform), we set a shorter polarization time. For a higher amplitude FID, we use a longer polarization time. The sensitivity of the amount of change in FID peak amplitude to change in polarization time is non-linear. The curve is non-linear in a way analogous to charging a capacitor in an R-C circuit. Thus, there is more sensitivity for changes in shorter polarization times on the order of one time constant or the tau 1 of the working NMR fluid (at that temperature). Beyond several tau 1, there is relatively little increase in FID peak amplitude for longer polarization times. For this reason, we set an upper limit at about 3.5 seconds, which is generally below 4 tau 1. Normal operation should be kept below about 3 seconds. The peak envelope voltage (PEV) servo loop uses a moving average of N past measurements of the filtered envelope maximum voltage as a feedback sensed or measured parameter. The control variable is the polarization time. In the Labview block diagram, on the left (early in time), we enter the PEV servo with the array of past filtered envelope maximum voltage values and the present polarization time (already used earlier in the present measurement cycle). N points (the number for the moving average) are extracted and the mean value of the N points is calculated. Until the first valid moving average, the loop uses a fixed polarization time. After N points have been measured and the first moving average value is valid, the loop goes into operation. The difference of the set PEV value and the present moving average value is multiplied by -0.1 and a loop gain value

22 P a g e 22 (typically 5). That value is subtracted from the present polarization time to complete the loop calculation. The resulting computed polarization time will be used at the beginning of the following measurement cycle as the polarization time. Hard limiting clamps the computed value between 0.5 and 3.5 seconds. If the PEV servo operates at either of these clamped values, the loop has failed in an open-loop mode. The peak envelope voltage and the computed polarization time are recorded each measurement cycle to the data text files. The blue curve is a plot of 640 points of peak envelope voltages. The red curve is a plot of the computed polarization times. The red curve generally follows the temperature of the working NMR fluid. With the PVC coil form, plastic sensor coil container, and plastic rain cover, solar radiation as well as outside air temperature all drive the temperature of the working NMR fluid. With only modest optimization of loop parameters (e.g. number of moving averaged points, and loop gain), the FID peak envelope voltage is generally held to +/- 0.1 Volt.

23 P a g e 23 ALTERNATIVE SOFTWARE AND ADC/DIGITAL I/O MODULES We acknowledge that some experimenters might prefer to write the software application in frameworks other than National Instrument s LabView and/or use other digitizer and I/O modules. We believe that our hardware kit (which does not include the USB 6008 module) will be amenable to alternate software development approaches. Once a suitable sensor is in operation (generally a counter-wound sensor coil pair as described in our Part IV article), our hardware modules provide a standard analog precession signal (the free induction (FID)) exponentially decaying waveform. Therefore, any suitable signal procession system can be used with our hardware kit. Of course, experimenters using alternative software approaches will need to provide appropriate system timing as described above. 1 Using National Instruments Application Builder The FDM frequency estimator is a windows executable which performs the filter diagonalization method (FDM). 4 Filter Diagonalization Method "FDM" (harmonic inversion), 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 We plan to share our vi source code for those who have access to a LabView development system, the compiled LabView executable, and the FDM executable. However, since we do not own rights to the FDM source code, at this time we are not able to share the FDM FORTRAN source code. Our software runs on the Microsoft Windows operating system. All of our recent testing was done using Windows 7. Our understanding is that an alternative FDM implementation in LINUX is available for experimenters wanting to use FDM signal processing on other platforms (Google MIT harminv).