Because I work in a lab where the instrumentation of choice is Nuclear Magnetic Resonance (NMR) spectroscopy, you can say I’ve grown fond of NMR, as well as all the engineering challenges NMR instrumentation faces and overcomes. Truth be told, sometimes I find the circuitry and signals in NMR spectrometers more interesting than the sample… Don’t tell my boss. ;)
Of course, my level of understanding when it comes to circuits is not yet advanced enough to tackle a full pulsed NMR design. I figured I’d give myself a simpler challenge: a proton precession magnetometer.
I absolutely could not have achieved success in this project without the insight, guidance, support and encouragement offered by Joe Geller at Geller Labs. Thanks Joe!
Without going into a good amount of quantum mechanics, the idea behind proton NMR is this: a proton (hydrogen nucleus), when placed in a static magnetic field ($\mathbf{B}_0$), will be polarized by that field. If a stronger static magnetic field ($\mathbf{B}_1$) is applied orthogonally to the first field, the protons will essentially be re-polarized by the stronger field, much the way a compass needle points to magnetic north. If the stronger field is then turned off sufficiently quickly, the proton will begin to precess about the original field at its characteristic frequency:
\[\omega = -\gamma \| \mathbf{B}_0 \|\]In the above equation, gamma is called the gyromagnetic ratio, which is a scale factor that simply relates field strength to precession frequency. The entire process is described a bit more completely by the phenomenological Bloch equations:
\[\begin{pmatrix} \frac{d M_x(t)}{dt} \\ \frac{d M_y(t)}{dt} \\ \frac{d M_z(t)}{dt} \end{pmatrix} = \gamma \left( \mathbf{M}(t) \times \mathbf{B}(t) \right) - \begin{pmatrix} \frac{M_x(t)}{T_2} \\ \frac{M_y(t)}{T_2} \\ \frac{M_z(t) - M_{eq}}{T_1} \end{pmatrix}\]Solution of the Bloch equations for the Earth’s field along $z$ and the polarizing field along $x$ yields a situation in which the same solenoid used to generated the polarizing field may be utilized to detect the precession signal as it swings through the $xy$ plane. Thus, a properly constructed solenoid that holds at least 500 mL of water and sits at a right angle to the local field vector should be capable of both polarizing and detecting the sample protons.
Of course, it’s much easier said than done. Other factors dictate the use of specialized circuitry and design. For example, while polarization turn-on may be slow, turn-off must be as fast as possible, or the magnetization will decay away without precessing (more on that later). Furthermore, the signal is tiny: on the order of nanovolts to microvolts. Thus, placing both digital and analog components on the circuit board requires careful attention to grounding and signal paths. Finally, the device should be powered solely from a PC USB port, with the exception of the polarization battery. This dictates the use of ultralow-noise switching regulators to turn the nasty USB +5Vdc into a clean +/-5Vdc.
If all that isn’t enough, construction of the coil itself has broad impacts on the entire system. It’s a big task, no doubt…
This whole project was kind of dead in the water until I made this board. I had been working for months on self-contained low-noise preamplifiers designed around multiple LT1028 opamps, and had gotten nowhere. This board renewed my faith in the feasibility of the project just enough to get started again.
The FILv1r0 board is derived from an 8th order Chebyshev bandpass filter in a multiple-feedback topology proposed by the TI FilterPro software tool. The design targets were: 1.4 kHz passband center, 3.7 kHz passband width, and +/- 0.5 dB passband ripple. In fact, the board – realized around an Analog Devices OP462 – achieved the targets just fine. The -3 dB rolloff frequencies of the filter of 500 Hz and 4.0 kHz translate to a magnetic field (as measured through the proton gyromagnetic ratio) that can range from 11.75 to 93.95 microtesla and remain detectable, far outside the typical values for earth’s field.
Excited from the success of the filter design, I went to work on a new instrumentation preamplifier circuit, this time simplifying around the fixed-gain (66 dB) AD8428 low-noise instrumentation amplifier IC. Thanks to the exemplary application notes on noise, grounding and instrumentation amplifiers put out by Analog Devices and Linear Technology, I was able to incorporate a functional differential filter into the design to limit any RF interference that could try and creep in. And it worked!
This oscillator was a quick circuit to test the true noise floor of the INA amplifier and PPM analog signal chain. It was designed to put out a 1.0 kHz sine wave at 10 microvolts, but the output voltage ended up around six times that. Good enough. I have plenty of inline attenuators that can kick that down plenty.
Once the input precession signal is suitably amplified, it has to be captured somehow! I decided to use an external 16-bit SAR ADC attached to an ATmega32U2 to get the analog data into digital bits on a host computer.
First off, I was blown away at how much nicer it is to build around the new USB-enabled ATmega chips. Programming and communication can now both be done over the USB cable – there’s no longer any need for that stupid ISP, unless you’re doing something wacky with fuses.
I rewrote the usb-serial software stack from the Teensy 2.0 to use in the DIG board. (I say rewrote because it was literally all typed in again, and I did change a significant amount of the code to make it more readable and understandable…) The end result was a USB device that you could ‘cat’ and get a timed binary dump of ADC register values from. Neat!
Once I had built up and tested the designs above, I was left with a choice: keep building modular boards for the analog power supply and current sink, or just wing it and build up a complete magnetometer prototype. I was a bit impatient, so I chose the latter. I got a bit lucky. After swapping out the current sink driver opamp to a slower chip, the whole board was completely functional!
The sensor coil current sink did prove to be a bit touchier than I expected. I had originally designed in an AD8591 3 MHz 250 mA rail-to-rail opamp into the circuit, without any compensation or high frequency rolloff. The result was pretty severe oscillations in the output current around 3 MHz.
Swapping the AD8591 with the AD8541 1 MHz opamp and upping the compensation capacitor to 10 nF quenched the oscillations with a resistive load and the sensor coil, so I went with that.
The second revision magnetometer was relatively similar to the original version, with the only major changes being:
Of course, I wish I’d tested the current sink with the actual sensor coil before finalizing the second revision, because the behaviour changes wildly between a resistive load and an inductive one. I learned that the hard way, and ended up having to go back to compare polarization waveforms for both versions, shown below:
The final verdict from these comparisons was that placing a huge inductor inside the feedback path of a current-boosted opamp was probably a poor decision. Experimentation with several possible alternative current sink circuits led me to decide on the simplest possible method: directly switching the maximum current available by Ohm’s law. I’m expecting inductive fall times around ten microseconds using this method in the next revision.
The major change in the third revision board focus on the redesigned coil polarization scheme, which does away with the 12 bits of current resolution in exchange for certainty that the coil will experience a flat polarization envelope with rise and fall times below a couple hundred microseconds.
A less serious revision involved a lowering of the input instrumentation amplifier’s 3 dB rolloff from 300 kHz to 15 kHz in an attempt to further reduce high-frequency noise and aliasing in the input signal chain.
Lo and behold, the third time was a charm! After several trials with the new board (including realizing I had forgotten to fill the coil with that magic water!) I collected a hopelessly smeared NMR line in the basement of the chemistry building at UNL. Subsequent tests in a more remote region of Nebraska resulted in a sharper, more intense line: confirmation that the magnetometer works.
Because DIG was a true proof of concept, I only wanted to test the very basic and critical features of the design first: transferring a live stream of ADC values back to the host in real time. So the first walk of testing DIG was to just cat it’s device file. How elegant! :P
Of course, my first real attempt at talking with the DIGv1r0 board was through a tiny set of command line programs. There is really no need to build up a complete graphical interface for debugging the prototype. The functions of the PPM are boiled down into three commands: rpar, wpar, and zg. (Yes, it’s a tiny hat-tip to the folks at Bruker.) The parms are transferred in 15-byte packets (six parameters) using rpar and wpar, and zg initiates execution of an acquisition. Signal averaging of multiple transients is also supported.
As an example, the following code shows the process of reading current parms, writing new parms, and running a four-scan acquisition with the new parms. The new parms are for a 5-second polarization, 32k samples at 20 kS/s, a 10-millisecond dead time before and after polarization, and a 1-amp coil polarization current:
rpar
RPAR: 62 800 16384 25 25 16 OK
wpar 306 800 32768 400 400 2048
WPAR: OK
zg 4
RPAR: 306 800 32768 400 400 128 OK
ZG: POL ACQ POL ACQ POL ACQ POL ACQ OK
awk 'END {print NR, NF}' fid
32768 2
I got pretty tired of translating between DAC and timer/counter register values and wrote a few helper functions to convert to and from human- readable versions of the PPM parameters, like so:
wparh 10 22.05 32768 1 10 0.5
WPAR: OK
rparh
RPAR: 610 726 32768 40 400 1024 OK
Polarization time 9.994 s
Acquisition rate 22.039 kS/s
Sample points 32768
Polarization deadtime 1.000 ms
Acquisition deadtime 10.000 ms
Polarization current 0.500 A
Conversion into the frequency domain was the next software design target, so I wrote ft, hrft and wfp to address that need. In short, ft just calculates a Fast Fourier Transform of the time-domain data, hrft calculates a ‘High Resolution Fourier Transform’ of the data, and wfp calculates a Short-Time Fourier Transform ‘waterfall’ from the data. Once an acquisition is complete, running them is as easy as:
ft
wfp
hrft 950 1050
Finally, if the wparh statement above didn’t give you the hint, there is a tool to save the acquired time-domain data as a ‘wav’ file: snd.
Once I had a first complete prototype PPM board, I quickly became frustrated with having to run a shell command to acquire and then jump into gnuplot for visualization. It was time for a graphical interface to the device.
May I present backspin. It’s really just a quick hack in GTK3 that ties into the API I had already written for the CLI utilities and provides a snazzy way of visualizing acquired and transformed data without tabbing between a ton of windows. The following images detail a typical progression through the backspin program:
Because why stop at a small set of inflexible command-line and graphical tools? I couldn’t help myself! I wrote a C extension in Python (2.7 and 3.3) that links into the PPM goodness, allowing complete control over a connected magnetometer from inside the Python scripting language.
For example, here’s the Python code for acquiring and plotting an acquired spectrum:
import pyppm
import numpy as np
import matplotlib.pyplot as plt
device = pyppm.PPM()
device.setparm(pyppm.POLARIZE_TIME, 10)
device.setparm(pyppm.POINT_COUNT, 32768)
(t, a) = device.acquire()
(f, A) = pyppm.fft(t, a)
l, = plt.plot(np.array(f), np.array(A))
plt.show()
From time to time I’d post a video on part of the project, so here are all the videos I’ve put up relating to the magnetometer…
This is an open-source project, from the software to the hardware. The whole shebang is wrapped up into a single tarballed source tree, with a few scripts inside to simplify getting started. The circuit schematics can be read and edited with gEDA gschem and the board designs can be read and edited with gEDA PCB.
Enjoy! :D