AC constant current source

[anvoice]

I found this paper
https://www.sciencedirect.com/science/article/pii/S1631074819301055
which seems to give some insights regarding a potential architecture of a low-cost NRM machine, and requirements for the components. For instance

... the transmission power required to energize afrequency band of about 10 kHz (for the 1H nucleus) for a 1cm³ volume sample is about 11 W (@500 MHz or 11.7 T). For a low-field spectrometer, an excitation power of about 5 W for a pulse duration of about 20µs is sufficient togenerate an instantaneous tilting of 90° of M0, i.e.in the xy plane.

@anvoice, I have of course no idea how similar (or different) your machine will be in comparision to the machine described in this paper. But I guess it may be a reasonable starting point.

EDIT:
And browsing through Google search results, I get the feeling that coil design for NRM is a scienece on  its own.

That is reassuring. A 11.7T machine is out of the question (needs either very powerful permanent or high current cooled superconductor magnets), but the fact that a low-field one can get away with about 5W means what I have in mind may be feasible. I'll take a look at that paper right now.

Actually perhaps capacitor energy storage coupled with a resonant LC circuit could possibly do this.

Some of the strongest AC magnetic fields commonly in use are for induction heating. You can guess these are some very strong fields if anything highly conductive inside that field starts heating up so quickly. These things typically only use 1 to 10 turns but use massive currents. These massive currents come from a tank capacitor in parallel with the coil, this being a special capacitor designed for these high AC currents (In large high power machines even the capacitors are sometimes water cooled). The hundreds of amps of current are mostly supplied by the capacitor recycling energy back and forth while the power source is supplying a only few percent of it that ware lost to heat.

If you do wish to generate as powerful of a AC field as possible (Not sure if this is your requirement, or does only the permanent field need to as strong as possible) you could use the same coil + capacitor circuit as induction heaters (Likely reusing off the shelf induction heater parts), but instead of slowly pumping small amount of power into it to get it oscillating you instead slowly charge up a large capacitor. Then to start you dump the energy from the capacitor into the tank capacitor of the induction heater LC cirucit (This likely needs something like a thyristor to switch such a high peak current). This should get it oscillating within pretty much one cycle and then 5 cycles later to stop it you use a triac to short out the LC cirucit and stop it. This way you can charge the main storage capacitor slowly at the 10W you have avalable but then quickly dump all that energy into the AC field you need (Giving you many kW of peak power during that short time).

The higher the alternating field the shorter the pulse must be, which eventually becomes a problem of its own. About 0.5T will lead to a 10us pulse, which is good enough, and even smaller fields are possible with the only downside of a loss of some sensitivity in the final spectrum. So no, I don't need a very high AC field. Thanks for the detailed explanation though, perhaps I can adapt some of that into my design.

I don't think that a freely oscillating LC tank will suffice for the use case. If I understand correctly, the carrier frequency must match the Larmor frequency very closely (ppm or better?), and it needs to be finely adjustble (since the frequency is not fixed but depends on the strength of the static field, which must be - btw - as stable as well). So IMO the origin of the carrier needs to be a decent frequency synthesizer. Furthermore the carrier is not only needed for the excitation, but also for down-mixing the received signal (quadrature demodulation), since the spectral lines to be measured obviously reside in a very narrow frequency band (just a dozen of ppm or so) around the carrier frequency. I also did read in several papers that the envelope of the excitation pulses is not necessarily a rectangle, but "soft pulses" with different envelope shapes (e.g. sinc, gaussian, and even more complex ones) are used as well, in order to synthesize excitation pulses having dedicated frequency spectra.

Yes, the field should match the Larmor frequency quite well, which itself would change if the static field was not stable. I doubt I can afford a good frequency synthesizer for this project though, as it is a rough proof of concept. If a minimal (and cheap) configuration starts looking promising, more investment could be made. But the goal right now is to get a detectable signal in the detector coils (which may be the alternating field coils or another orthogonal set). If that is not possible with basic, affordable electronics, low-field NMR becomes much less lucrative.

You're right about the excitation pulse envelope. I may have to incorporate something like that eventually, but again, the first goal is to get any kind of signal out of a minimal setup.

[gf]

The higher the alternating field the shorter the pulse must be, which eventually becomes a problem of its own. About 0.5T will lead to a 10us pulse, which is good enough, and even smaller fields are possible with the only downside of a loss of some sensitivity in the final spectrum. So no, I don't need a very high AC field. Thanks for the detailed explanation though, perhaps I can adapt some of that into my design.

Since your Larmor frequency is just ~500kHz, I'm wondering in fact whether a 10us pulse isn't much too short, as this leads to a pretty wide bandwidth of the pulse (-> about 20%, while 10us at say 100MHz rather leads to 0.1% bandwidth only). On the one hand this increases the chance that the spectrum of the pulse does include the Larmor frequency even in case of a bad match of the carrier frequency, but on the other hand it reduces the pulse's energy density per Hz of bandwidth.

I'm a complete NRM layman - just did read some basic stuff yesterday, and there is one thing I did not understand so far: In the literature, B1 is obviously trated as a constant value (although the excitation signal is actually AC and therefore a function of time). For a pure sine wave at the Larmor frequency, it makes sense that the value of B1 is simply the amplitude of the sinve wave. But how is the value of B1 defined for a spread-spectrum signal whose magnetic flow per Hz of bandwdith is a function of frequency? Which frequency band of the signal contributes to the value of B1? Do the protons have a particular (let me call it) "resolution bandwidth" inside which they react to the excitation, while any frequencies otside this spectral band are just wasted excitation, having has no effect?

To summarize: As far as I understand, the maximum pulse duration is eventually limited by the required bandwith (which depends in turn on the maximum chemical shift that needs to be captured). Reducing the pulse bandwith to 0.1% (which I guess is still sufficient?) whould lead to a 2ms pulse (instead of 10us) and would reduce 0.5T to only 250uT, right? And depending on the answer to my above question regarding the definition of B1, the smaller bandwidth may additionally help due to the lager magneric flow per Hz of bandwidth.

Yes, the field should match the Larmor frequency quite well, which itself would change if the static field was not stable. I doubt I can afford a good frequency synthesizer for this project though, as it is a rough proof of concept. If a minimal (and cheap) configuration starts looking promising, more investment could be made. But the goal right now is to get a detectable signal in the detector coils (which may be the alternating field coils or another orthogonal set). If that is not possible with basic, affordable electronics, low-field NMR becomes much less lucrative.

Every single-digit-$ radio tuner includes a reasonable frequency synthesizer today. I think that a $$ OXCO can be stabe enough for the use case (at least short term). For 500kHz, likely a DDS generator like AD9830 or similar (using an OXCO as reference for the clock) may well suffice, at least for generating the carrier for the excitation pulse.

The receiver side may be more challenging than the transmitter. Here I'm not sure yet reagarding the carrier phase noise requirements for your use case. And the major issue is likely that the signal is very weak. How big is actually the amplitude of the (~500kHz AC) magnitic flow which needs to be detected? Somewhere in the uT range, or even less? What's the area of the coil, btw?

EDIT: Btw, how many T is your B₀?

[Kleinstein]

The AC excitation field should be relatively small compared to the DC field. This is at least the general assumption for the theory. Anyway 0.5 T would be really high for an AC field in air -  remember a system made to get 0.1 T AC over a slightly larger volume: it was quite a beast with a few kW of RF power to a high Q resonator with water cooled coil. Kind of like induction heater, just higher frequency.

I have once used a system that seem to be made for low field NMR. It had an RF amplifier for some 2 kW for pulses up to 1 ms or so. This gives odd effects when one starts to see/hear arcing at BNC connectors. The NMR coil may not be well matched to the amplifier / cable impedance. One may have to consider this with the amplifier, not all power amps like a poorly matched load.

I don't think one would really need a pure sine wave, so square wave drive with a H bridge may be acceptable.
A point could be dampening the LC resonance at the end of the pulse.

[gf]

EDIT: Btw, how many T is your B₀?

I just recognized, 1H isotope has 42.55 MHz/T (γ=2.674e08), so 500kHz corresponds to only ~12mT. Is the static field really that low? Or do you have a different isotope in mind?

EDIT:
And solving "α = γ * B1 * tp" for B1 [where α=pi/2, tp=2ms (as I suggested earlier, for ~0.1% spread-spectrum bandwidth), and γ=2.674e08 for 1H isotope] results in a required RF field of only B1=2.94uT, in order to turn the spin vector to the x-y plane ("90° pulse"). (Please correct me if there is an error in the calculation - I could hardly believe the low number either.)

According to the calculator at https://www.accelinstruments.com/Magnetic/Magnetic-field-calculator.html, a magnetic field of 3uT should be achievable with a current of only 50mA, in the center of a short coil (air core) with a single turn and an assumed diameter of 20mm. The calculated inductance is 45nH then (-> impedance 0.14jOhm @500kHz, i.e. almost negligible). This would not reqire a powerful amp at all.

But I could imagine noise problems in the detector if B0 is so low.

[anvoice]

The higher the alternating field the shorter the pulse must be, which eventually becomes a problem of its own. About 0.5T will lead to a 10us pulse, which is good enough, and even smaller fields are possible with the only downside of a loss of some sensitivity in the final spectrum. So no, I don't need a very high AC field. Thanks for the detailed explanation though, perhaps I can adapt some of that into my design.

Since your Larmor frequency is just ~500kHz, I'm wondering in fact whether a 10us pulse isn't much too short, as this leads to a pretty wide bandwidth of the pulse (-> about 20%, while 10us at say 100MHz rather leads to 0.1% bandwidth only). On the one hand this increases the chance that the spectrum of the pulse does include the Larmor frequency even in case of a bad match of the carrier frequency, but on the other hand it reduces the pulse's energy density per Hz of bandwidth.

I'm a complete NRM layman - just did read some basic stuff yesterday, and there is one thing I did not understand so far: In the literature, B1 is obviously trated as a constant value (although the excitation signal is actually AC and therefore a function of time). For a pure sine wave at the Larmor frequency, it makes sense that the value of B1 is simply the amplitude of the sinve wave. But how is the value of B1 defined for a spread-spectrum signal whose magnetic flow per Hz of bandwdith is a function of frequency? Which frequency band of the signal contributes to the value of B1? Do the protons have a particular (let me call it) "resolution bandwidth" inside which they react to the excitation, while any frequencies otside this spectral band are just wasted excitation, having has no effect?

To summarize: As far as I understand, the maximum pulse duration is eventually limited by the required bandwith (which depends in turn on the maximum chemical shift that needs to be captured). Reducing the pulse bandwith to 0.1% (which I guess is still sufficient?) whould lead to a 2ms pulse (instead of 10us) and would reduce 0.5T to only 250uT, right? And depending on the answer to my above question regarding the definition of B1, the smaller bandwidth may additionally help due to the lager magneric flow per Hz of bandwidth.

Yes, the field should match the Larmor frequency quite well, which itself would change if the static field was not stable. I doubt I can afford a good frequency synthesizer for this project though, as it is a rough proof of concept. If a minimal (and cheap) configuration starts looking promising, more investment could be made. But the goal right now is to get a detectable signal in the detector coils (which may be the alternating field coils or another orthogonal set). If that is not possible with basic, affordable electronics, low-field NMR becomes much less lucrative.

Every single-digit-$ radio tuner includes a reasonable frequency synthesizer today. I think that a $$ OXCO can be stabe enough for the use case (at least short term). For 500kHz, likely a DDS generator like AD9830 or similar (using an OXCO as reference for the clock) may well suffice, at least for generating the carrier for the excitation pulse.

The receiver side may be more challenging than the transmitter. Here I'm not sure yet reagarding the carrier phase noise requirements for your use case. And the major issue is likely that the signal is very weak. How big is actually the amplitude of the (~500kHz AC) magnitic flow which needs to be detected? Somewhere in the uT range, or even less? What's the area of the coil, btw?

EDIT: Btw, how many T is your B₀?

From my understanding, a pulse with a nonzero bandwidth (e.g. a sinc pulse ideally) is used for MRI for the excitation of particular slices of space. However, for simply detecting an NMR signal a pure Larmor frequency sine wave is used. In that case, a rotating reference frame is created within which the magnetic moment of the sample is pushed down from the z-axis into the xy-plane at a constant speed, i.e. as neatly a possible. At that point, the width of the pulse (which is independent of bandwidth) will determine whether the sample gets to start relaxing before it's fully in the xy-plane. If it does start relaxing (i.e. long pulse) the overall signal will be lower.

My AC field only needs to be about 0.5mT (millitesla), not 0.5T. You're right that the pulse width will be inversely proportional to field strength, so the longer the pulse the lower the field I get to use. At the expense of sensitivity though, which may already be limited with a weak field.

Will definitely look into these signal generators. You're right that the receiver will be a challenge: I was planning to start another topic a little later to figure out how to go about detecting the microamperes or even nanoamperes of signal I will get. I have an aged digital oscilloscope, but will need some sort of amplifier to be able to detect the low currents in the receiver coils. I have heard of Dave's ucurrent gold, which helps do this, but it's apparently out of stock.

If you're asking for the area of the transmitter coil, it will probably be a few square centimeters. As far as the receiver goes, likely even smaller: maybe a square centimeter, maybe less.

Yes, the static field (B0) will indeed be about that low. Creating a much higher field in air requires prohibitively high currents for an air-cooled setup.

Thank you for taking the time to do this much research into the topic!

The AC excitation field should be relatively small compared to the DC field. This is at least the general assumption for the theory. Anyway 0.5 T would be really high for an AC field in air -  remember a system made to get 0.1 T AC over a slightly larger volume: it was quite a beast with a few kW of RF power to a high Q resonator with water cooled coil. Kind of like induction heater, just higher frequency.

I have once used a system that seem to be made for low field NMR. It had an RF amplifier for some 2 kW for pulses up to 1 ms or so. This gives odd effects when one starts to see/hear arcing at BNC connectors. The NMR coil may not be well matched to the amplifier / cable impedance. One may have to consider this with the amplifier, not all power amps like a poorly matched load.

I don't think one would really need a pure sine wave, so square wave drive with a H bridge may be acceptable.
A point could be dampening the LC resonance at the end of the pulse.

Unit error. It's 0.5mT. You're totally right, the field would be insane at 0.5T.

That's a very interesting point. From what I understand, a square way may theoretically work (not 100% sure). It's usually done via sinusoid as that creates a rotating reference frame within which the magnetic moment goes from the z-axis to the xy-plane at constant speed. I believe at the very least, it would exhibit jerky movement in a square wave field. In a perfect world, I would use a sine wave so everything is as neat as possible, but if the square wave is many times easier to implement I could definitely consider it. I'm therefore officially welcoming suggestions for how to create an efficient square wave alternating field in the transmitter coils, it may be worth a try.

[gf]

However, for simply detecting an NMR signal a pure Larmor frequency sine wave is used.

Do you only want to detect 1H resonance at all?
Is your aim not to get a chemical shift spectrum?
You get it for almost free anyway if you sample the down-mixed / quadrature demodulated signal and do a FFT.
Note that a time-limited pulse with rectangular envelope has a (sinc-shaped) spread spectrum anyway in the frequency domain (-> 1000ppm bandwidth requires a pulse width of about 2ms, @500kHz, and shorter pulses would even have wider bandwidths - IMO wider than necessary).
A sinc pulse (in the time domain) has the property that its spectrum is rectangular in the frequency domain, w/o any side lobes, so all energy is concentrated (and evenly distributed) inside this rectangle.

Btw, I did update my previous message while your message arrived. Please check.

EDIT:

From my understanding, a pulse with a nonzero bandwidth (e.g. a sinc pulse ideally) is used for MRI for the excitation of particular slices of space

The spread spectrum is used in FT-NMR in order to excite and capture the whole chemical shift spectrum at once, avoiding the need to sweep through frequencies in small steps.

The location of MRI slices in space is IMO done by adding a gradient field to B0, making the Larmor frequency location-depedent (thus only the slice with the right Larmor frequency will respond to the excitation). Gradients can also be extended to 3D in order to address indivudial voxels in space. I guess that a spread spectrum can be used here as well in order to capture multiple neighbor voxels at once, but I don't known any details regarding MRI.

From what I understand, a square way may theoretically work (not 100% sure).

My thougths: A square wave is eventually a sum of sine waves (fundamental + harmonics).
The harmonics would (undesirably) excite isotopes whose gamma is by a factor of n higher (where n in 3,5,7,...), if such isopes exist.
Otherwise the harmonics will likely have no effect.

[ssmirnov56]

EDIT:
And solving "α = γ * B1 * tp" for B1 [where α=pi/2, tp=2ms (as I suggested earlier, for ~0.1% spread-spectrum bandwidth), and γ=2.674e08 for 1H isotope] results in a required RF field of only B1=2.94uT, in order to turn the spin vector to the x-y plane ("90° pulse").

Guten tag! Very interesting thread!

I just have one quick point: the pi/2 pulse duration of milliseconds might be so long that the B0 and B1 relaxation eliminates much of the sample induced magnetization. If relaxation has enough time to progress, the intensity of the induced and detected signal will go down substantially (=low sensitivity). This is one reason why for the pi/2 B1 pulse duration better be as short as technically possible (especially if the goal is to ping only one type of the nuclei and not get the spectrum).

Vielen dank!

[anvoice]

I just recognized, 1H isotope has 42.55 MHz/T (γ=2.674e08), so 500kHz corresponds to only ~12mT. Is the static field really that low? Or do you have a different isotope in mind?

EDIT:
And solving "α = γ * B1 * tp" for B1 [where α=pi/2, tp=2ms (as I suggested earlier, for ~0.1% spread-spectrum bandwidth), and γ=2.674e08 for 1H isotope] results in a required RF field of only B1=2.94uT, in order to turn the spin vector to the x-y plane ("90° pulse"). (Please correct me if there is an error in the calculation - I could hardly believe the low number either.)

But I could imagine noise problems in the detector if B0 is so low.

No, your calculations are correct. However, because the magnetic moment can relax during the pulse and not after it, sensitivity suffers greatly. I believe 2ms might be at a point where the trade off is too much. Could try to go for half a millisecond though: the field would then only need to be 12uT.

Do you only want to detect 1H resonance at all?
Is your aim not to get a chemical shift spectrum?
You get it for almost free anyway if you sample the down-mixed / quadrature demodulated signal and do a FFT.
Note that a time-limited pulse with rectangular envelope has a (sinc-shaped) spread spectrum anyway in the frequency domain (-> 1000ppm bandwidth requires a pulse width of about 2ms, @500kHz, and shorter pulses would even have wider bandwidths - IMO wider than necessary).
A sinc pulse (in the time domain) has the property that its spectrum is rectangular in the frequency domain, w/o any side lobes, so all energy is concentrated (and evenly distributed) inside this rectangle.

Yes, the idea for the proof of concept is to detect the hydrogen resonance. If that succeeds, I would go for a spectrum. That is why I was not worrying about the receiver bandwidth spectrum at the moment. Basically, I don't know if I will get a signal at all: the idea of a basic prototype is to test that.

The spread spectrum is used in FT-NMR in order to excite and capture the whole chemical shift spectrum at once, avoiding the need to sweep through frequencies in small steps.

The location of MRI slices in space is IMO done by adding a gradient field to B0, making the Larmor frequency location-depedent (thus only the slice with the right Larmor frequency will respond to the excitation). Gradients can also be extended to 3D in order to address indivudial voxels in space. I guess that a spread spectrum can be used here as well in order to capture multiple neighbor voxels at once, but I don't known any details regarding MRI.

Right, and B0 is varied by using variable step coils. Not something I can manage right now, and again, this is far beyond the prototype's point. If there is no signal using a ~10mT B0 field, there is no point trying to get a spectrogram.

My thougths: A square wave is eventually a sum of sine waves (fundamental + harmonics).
The harmonics would (undesirably) excite isotopes whose gamma is by a factor of n higher (where n in 3,5,7,...), if such isopes exist.
Otherwise the harmonics will likely have no effect.

As for the harmonics, you're right that they would be a parasitic in the spectrum, but actually a boon in the prototype, because they would add to the initial signal intensity.

To summarize: you're making very valid points for a fairly advanced NMR instrument capable of resolving various molecules, but that is step 2. If step 1, which is proof of concept, fails, investing effort into a more sophisticated design will be for naught.

[Berni]

If weaker fields allow you to go for a longer pulse then this might be helpful in the reception later on.

Demodulating a weak signal after it has been received back is a lot easier if you have a longer slice of time to work with. Signal processing in receivers (be it radio or radar or ultrasound) tend to have some sort of averaging like behavior present somewhere along the chain. So the longer you can look at a narrow slice of the spectrum the cleaner the resulting signal gets.

This allows receivers to pull out signals that are well below the noise floor, so this is a great help when looking for really weak signals as the amplifier sensing the signal does have limits on how low noise it can get (Without cryocooling and all that jazz)

[anvoice]

If weaker fields allow you to go for a longer pulse then this might be helpful in the reception later on.

Unfortunately, the sample relaxation time (at which point the signal will be generated) after the excitation pulse does not depend on the transmitter pulse width. It will certainly change the spectrum if one is present because the excitation bandwidth will be different, but I don't think it will help with sensitivity to have a longer pulse. The opposite is true because the sample's magnetic moment will start the relaxation before it's fully adjusted to the xy-plane.

I will start another topic about the receiver side of things so as not to clutter this one.

[gf]

No, your calculations are correct. However, because the magnetic moment can relax during the pulse and not after it, sensitivity suffers greatly. I believe 2ms might be at a point where the trade off is too much. Could try to go for half a millisecond though: the field would then only need to be 12uT.

I'm just wondering, aren't relaxation time constants rather in the 100ms to seconds range? Then I would not expect more than a few percent of exponential decay happening already during a 2ms pulse (i.e. rather negligible). Or do I miss anything?

The calculated coil was only a single turn, so it should be easy to increase B1 by a factor of N by using N turns instead (w/o need to increase the current).

Basically, I don't know if I will get a signal at all.

Can you actually calculate an estimate for the expected signal level (magnetic flux through the pick-up coil) for the planned setup?

As for the harmonics, you're right that they would be a parasitic in the spectrum, but actually a boon in the prototype, because they would add to the initial signal intensity.

While they add to the total pulse intensity, I don't think they are helpful for the excitation, since their frequencies are likely too far off to excite 1H.

[gf]

Yet another low-cost design:
https://iopscience.iop.org/article/10.1088/1742-6596/1380/1/012012/pdf

[anvoice]

I'm just wondering, aren't relaxation time constants rather in the 100ms to seconds range? Then I would not expect more than a few percent of exponential decay happening already during a 2ms pulse (i.e. rather negligible). Or do I miss anything?

Looking up the T1 relaxation time of water, turns out it's 4 seconds. So you are correct, it is probably negligible. I will be using a longer, weaker pulse then.

Can you actually calculate an estimate for the expected signal level (magnetic flux through the pick-up coil) for the planned setup?

Yes, it should be possible but slightly nontrivial. Give me a little time to answer this question: I will need to go from the Boltzmann distribution of hydrogen spins within the B0 magnetic field, and figure out what sort of signal (a damped sine curve) will be traced out as the magnetic moment relaxes back to the z-axis after the B1 pulse. Water should give a pretty good signal compared to other substances, but if I had to answer right away I'd say tens to hundreds of nanoamperes is the best I can count on.

While they add to the total pulse intensity, I don't think they are helpful for the excitation, since their frequencies are likely too far off to excite 1H.

Right, assuming they excite anything at all though, they will add to the output signal intensity. Not helpful in the long run for sure.

Yet another low-cost design:
https://iopscience.iop.org/article/10.1088/1742-6596/1380/1/012012/pdf

That is extremely helpful, thank you. It may be a good idea to use a setup similar to theirs since it's known to work, although their permanent field is many times stronger than what I can get (more sensitivity).

[gf]

@anvoice, I guess you might be interested in this one as well. In particular, it aims at a rather low frequencies, in your range.
It is very broadband, though (I'm not sure whether you want that).

https://42f0c01f-a-62cb3a1a-s-sites.googlegroups.com/site/casecirc/publications/mandal_2014_jmr_nr_mr_final.pdf?attachauth=ANoY7cqXB7TX6lCRnwNJm0yjB64NFpEA5feF0JxI0GcUX6_GfuW2Hms64hOb4FwEX5iyp43mufIIXgMIDTKpyxevbhqjNJYz_8LE-BL-8_8F2glzf80F5s80uWb_gKfxm--B_FY4E__UAsolxwosjdR6zkQuar9aPJTd3nfEqRTIrbtFN29OuiemN6YLW-uPvlF2R6FXI7MGq-vI7X_FRYgeLPtot_HDKs5wREEX60uVhB2pxlUp71gz8c1AczRRBBopfncA2g-x&attredirects=0

[anvoice]

@anvoice, I guess you might be interested in this one as well. In particular, it aims at a rather low frequencies, in your range.
It is very broadband, though (I'm not sure whether you want that).

https://42f0c01f-a-62cb3a1a-s-sites.googlegroups.com/site/casecirc/publications/mandal_2014_jmr_nr_mr_final.pdf?attachauth=ANoY7cqXB7TX6lCRnwNJm0yjB64NFpEA5feF0JxI0GcUX6_GfuW2Hms64hOb4FwEX5iyp43mufIIXgMIDTKpyxevbhqjNJYz_8LE-BL-8_8F2glzf80F5s80uWb_gKfxm--B_FY4E__UAsolxwosjdR6zkQuar9aPJTd3nfEqRTIrbtFN29OuiemN6YLW-uPvlF2R6FXI7MGq-vI7X_FRYgeLPtot_HDKs5wREEX60uVhB2pxlUp71gz8c1AczRRBBopfncA2g-x&attredirects=0

Yes, does seem interesting. Makes me wonder how many more such papers are floating around that are not in the open domain. Then again, those who publish in large journals usually have the funding to not build their own NMR.

I was thinking of trying the setup in the first paper you found, with components that will work around my frequency. Still trying to understand everything completely though, and I have doubts about the AD9959 module as another forum member made a thread concluding they're not too well-made.