Sciencemadness Discussion Board

do-it-yourself nuclear magnetic resonance spectroscopy

 Pages:  1  2  

arsphenamine - 25-9-2010 at 21:36

Chemical shift describes how much a particular nucleus' magnetic
precessional frequency (or Larmor Frequency) is changed by its
electronic environment.

The Larmor equation is ω = γB


Knowing the 1H gyromagnetic constant, 2.675 222 099e8, you can figure
a proton Larmor Frquency for any magnetic field you can express in Tesla.

Brigham Young University has a handy Larmor frequency calculator.

For the 1H nucleus in water, it is 42.5MHz at 1T field, and 60MHz at 1.4T.
For benzene, the protons precessional frequencies are shifted 450 Hz
downfield on a 60MHz/1.4T NMR.

Since the precessional frequency changes with applied magnetic field, it
is more useful to express the chemical shift as ppm instead of Hz, so
that 450 Hz becomes 450/60Meg or 7.5 ppm.

When the French speak of their 1GHz NMR, they are essentially bragging
about its 23.5T superconducting magnet wherein a 7.5 ppm chemical shift
is 7500 Hz.

un0me2 - 26-9-2010 at 07:25

Now, getting back to where I originally suggested the same, the outgoing sine wave is generated as a sawtooth wave, then smoothed by a lowpass filter. Is there any reason we can't do with this PIC what has been done with others, ie. generate a hex-coded sine wave (43.xxxx MHz) for the signal we want and put it out through the DMA Controls? That would save fucking around with DDS, etc.

With regard to both the front & back-end could we fuck around with this, Silabs Hybrid TV Tuner, which is a system on a chip for the 42-1002MHz range?

OpenCore NMR comments

arsphenamine - 26-9-2010 at 08:46

Takeda has a site for his OPENCORE NMR. From the freely
distributable OpenCore NMR content, you infer that Takeda wants to bring
NMR to the masses.

A few years ago, there were a flurry of electronic designs that used an
FPGA for DDS and which relegated setup and off-board communication to
a small MCU. If/Once you get the FPGA program working, you have a
compact and relatively inexpensive hardware implementation.

I say 'relatively' because OpenCore NMR uses a $246 mixed-signal FPGA
that handles DDS, pulse generation, FFT, and outgoing data FIFO. Of
course, the price is low compared to even the smallest of commercial
NMR devices.

Designing an FPGA circuit is an onerous slog. Time-to-market on this sort
of project, when designed and built from scratch, takes about 1.5 years,
an interval that can kill a small company if they misjudge the market.

NB,
More than a few hobbyist USB oscilloscope+SigGen tools use an FPGA+CPU
topology since it so clearly brings down the build cost.


[Edited on 26-9-2010 by arsphenamine]

12AX7 - 26-9-2010 at 10:09

Quote: Originally posted by arsphenamine  
@12AX7
I agree that the johnson noise for a 500Hz window is quite low,
but neither the initial preamplification nor the saddle coil
will be that notchy -- the preamp will necessarily see a
weak signal at the Larmor frequency. The net amplification required
before the FID signal is downconverted may be enough to
make noise a factor.


Nope, bandwidth is bandwidth. If the coil's bandwidth is 400kHz (i.e., Q ~ 100), then the converter picks up only twice the noise, because of the in-band noise (i.e., 40.0000 to 40.0005 MHz) and the image (for a 1kHz IF and 39.9990 MHz LO, the 39.9980 to 39.9975 MHz band). If better filtering and several conversion stages are used, the images can be rejected much better and noise will be nominal.

Then you have to look at the contributions from oscillator, converter and amplifier, since extra stages makes for extra noise. A noisy converter may be worse than picking up the image.

Anyway, noise is evenly distributed, so bandwidth has absolutely, positively no effect on the NMR signal itself, because the FFT looks at slices of narrow bandwidth (perhaps 0.1Hz). If there's noise up at 2kHz or down at 0Hz, we just don't even care, because it's in some far off bin that we aren't even looking at.
Quote:

re: internal ADC suck-osity, oversampling mitigates that 2 LSB error
if you are willing to wait for it, although 16x oversampling is
admittedly (cough!) suboptimal.


2LSB error cannot be corrected by oversampling, but it can be corrected by calibration (using a piecewise-linear or polynomial fit, or conversion table), and quantisation improved somewhat by dithering. Statistics always wins, and you need exorbitant amounts of oversampling to turn a crappy 12 bit ADC into a good 16 bit. To the tune 64x, plus the extra DSP filtering.
Quote:
re: bit banging any protocol
Get serious.
It's a WOMBAT.
If your MCU can't handle USB, find one that DOES.
Drool-proofed $20 OTC implementations abound.


No, you get serious. Why pay twenty fucking bucks for something free?
http://www.obdev.at/products/vusb/index.html
Bit bang an ATtiny, that will give all the power you need for controlling the DDS and pulse gen. No ADC or DSP is necessary.
Quote:
There are abundant implementations of FFT code for x86 processors
and it would be good to take advantage of them, FFTW is one fine example.
Unless you like endlessly adjusting an embedded FFT implementation for the
application, it is simpler to use proven code as a pro-active hedge against
the "Shit Happens" Law.


So what the fuck? You have no confidence in analog circuitry at all (an unfortunate sign of the times), and you have no confidence in software either! What do you even trust? It's no wonder you're bouncing around in this thread, suggesting the waste of money on expensive unobtainium unsuited to the task!

This is a bizarre forum. To begin with, chemists have a peculiar arrogance, comparable to a physicist's, and not entirely undeserved, given the large amount of wrote knowledge required to work in the field. The peculiarity, however, is that they assume their arrogance extends to all fields, even fields they are sometimes trivially incompetent in (global warming, electronics, mechanics, etc.). This has been seen time and time again on this forum.

What defense do I have against such people? Perhaps I should just assert my own arrogance with an equally immalliable fist. In just a few months, I will officially be an Engineer, that's with a capital 'EE', and you will have no choice whatsoever but to take my word as the Holy, Certified Truth. You may have arrogance in your own field, but I demand nothing less on my home turf!

I say this not so much for this thread, which I admit has already had some delightful turns, but for other members who have crossed me before. There is something to be said, and I want to bring it into the light.

Tim

arsphenamine - 26-9-2010 at 12:25

@12AX7

re: bit banging
Belatedly, it occurs to me that you may have a different conception
of bit banging than was defined in the early days of expensive UARTs
when bit banging meant establishing bit periods by software timing loops.

It's a great way to turn a small preemptive multitasking OS into a
cooperative multitasker. Writing interrupt handlers for a decade or so will
sensitize you to unmaintainable WOMBATS, your own and that of others.

re: Ignorance of software
Heh! (snicker, chortle)

The more you work with software, the less you trust it.
I trust some software much more than others.

re: expensive unobtainium
You may have confused me with someone else, or misconstrued what I
have written. The fractured spelling 'cheep' has appeared more than
once in my posts. I concede your right to the misapprehension but would
appreciate your aiming that stream of urine somewhere else.:)

un0me2 - 26-9-2010 at 14:42

@12AX7,

I'm neither a chemist or an engineer, if what I write comes across as arrogant, then so be it, it most assuredly ain't intentional (I'm actually trying to learn this shit), that said I am "apparently" an arrogant SOB to start with.

For mine the sooner the signal is digitized the sooner it can be dealt with programmatically. The implementation of several algorithm's to reduce echo/noise/etc. as well as averaging the signal response sets. I'm not choosing the PIC32 because it will be optimal, but because it is cheap, available with built-in USB 2.0 for programming and is ready to go, plus it is certainly should be able to cope with what we require of it.

I'm actually wondering why we cannot do what used to be done with the Atmega/PIC controllers before and upload to them a digitized representation of our signal, frequency, signal-shape and duration in hex-codes, then push that through a DAC, then a low-pass filter. That would give us the signal we need with the duration, without the need for digital synthesis.

Here is a tutorial on using the PWM channels (1 or 2) to generate a sine wave, then there are several pages on doing so programmatically with the PIC18 (Example 1 & example 2). Smoothing out the resultant wave would be simple enough, it would also make the best possible use of what is on the board, instead of just arbitrarily adding more electronics when they aren't necessary.

Here is a single-chip MCU+Transmitter for the range we want (27-960MHz) which could presumably take in the timing signal from the PIC32 (it is an SOIC-14 Chip). IThey are ~$2.50US/ea in single quantities.

[Edited on 27-9-2010 by un0me2]

Twospoons - 26-9-2010 at 18:15

I'd just like to add the benefit of my experience with coding USB on micros : avoid if you can. It took two of us three months to create a working USB mouse device (using window's inbuilt HID driver). Along the way we created a device that put windows into an endless reboot cycle. If I were building this NMR device I would use high speed serial port with an OTC USB-Serial converter (given that so few PCs have RS232 these days). Its possible to go to 4Mbit/s with these (using a quality one). Bog standard serial is dead simple to send and recieve, And the "serial port" in the PC will buffer the data until windows gets around to storing it.
Unless, of course, you really really want to learn about USB.



[Edited on 27-9-2010 by Twospoons]

un0me2 - 26-9-2010 at 20:06

OH SHIT, you tell me that AFTER I ordered the USB 2.0 programmable PIC32? :o

Ok, well I guess I'm about to learn a whole lot about the problem:(

arsphenamine - 26-9-2010 at 20:40

Quote: Originally posted by un0me2  
OH SHIT, you tell me that AFTER I ordered the USB 2.0 programmable PIC32? :o

Ok, well I guess I'm about to learn a whole lot about the problem:(

If it's the Microchip PIC32 development kit , it looks as if they already did
the USB software API for you.

Unless you are crufting up a USB bootloader or mouse handler on a PIC12
(see "Trial Of Job") or some such, I wouldn't worry.

Microchip gives you source code for libraries and examples in the order
of 100's of Megs. Surely, there are pertinent USB examples in the
Wired Communications section.

Get USB protocol basics at Wikipedia. The primary specs are at usb.org.
Concentrate on USB 2.0 since so little USB 3.0 hardware exists.

Otherwise, have no fear and be patient.

un0me2 - 26-9-2010 at 22:23

There are some interesting files in the Microchip examples for certain, there is also a book (you'll find it on a torrent I'm sure;), Newnes, 'Programming 32-Bit Microcontrollers in C: Exploring the PIC32' 2008 or something like that).

I just found a pertinent link regarding the use of the PIC32's PWM ports for sine wave generation. If we could modify that for the relevant frequency(s) (supposing someone one day someone wants to try it with 2H/D, 13C, etc), that would save some board space, extra-chip programming, etc. It would also make fuller use of the capacity of what is already on the board.:)

PS Silabs also have an 8-bit Analogue-Heavy MCU that has a multiplexed 8-channel 24-bit ADC, with 2 8-bit DACS (C8051F35X NB uses the common 8051 core).

That with the MAX2306 should be a winner, 8 24-bit samples per xx ns, of a upconverted/downconverted, filtered radio frequency. It would actually probably be able to be shunted into the same space as the PIC32 (it is a 32KB chip itself, with 100 pins in the same package). Either replace the package or build a new board (sounds like fun, eh?) with the UART-USB2 210X Family.

[Edited on 27-9-2010 by un0me2]

Twospoons - 27-9-2010 at 14:09

I can't say for the Microchip solution, but we were using Silabs parts. Silabs did provide libraries and a driver for USB, but frankly they weren't very good. Example : my co-worker re-wrote the controller code to include 3 endpoints instead of just one, and at the same time cut the code size in half!
This was over 5 years ago, so hopefully things have improved.

On another note, using a 24 bit converter may let you get away with less signal amplification, and hence a lower overall noise figure. I know thats one of the things Silabs were pushing with their 24 bit converter. I think we still have the dev kit kicking around somewhere ...

un0me2 - 27-9-2010 at 18:54

Ok, you wouldn't happen to still have access to the adapted code? Because I intend to utilize Silabs component (USB 2.0 - UART) when I try and build a single board version of this - quite simply the 2 DAC's on the MCU pose an interesting question as to whether one could programmatically form a sine-wave by modifying the squared/sawtooth waveform & frequency by passing the square/sawtooth waveform through an amplifier IC with dedicated filters & a shitload more. That would take care of getting the amplitude of the signal (in dB) up while avoiding the need for DDS chips

Any further gripes with the SILABS components? I'd prefer to know what to be on guard for before I start:)

Twospoons - 27-9-2010 at 20:55

Somehow I knew you'd ask - sorry the code is not mine to give away. I've no other gripes - in fact I rather like the Silabs micros, given they have onboard regulators and trimmed oscillators, and a rather nifty crossbar switch that lets you move the peripherals around the pins. I guess my point is simply that company supplied code isn't always that great. You will probably find the UART code and driver work fine (our needs were a little more complex) - silabs have been doing USB-UART converters for years.

un0me2 - 27-9-2010 at 22:22

Thanks, you had to know I'd ask on this board:D With the dual 8-bit DAC's, couldn't we feed them directly into a Class D Amplifier as a full on, full off, waveform @ the appropriate frequency (or a modification thereof)? If so, 12V can give 40W (2 x 20W - MAXIM9744 of power, as an approximation of a sine wave. That would be MAXIM IC's for the front & back end, with the USB & MCU dealt with by Silabs IC's.

Twospoons - 28-9-2010 at 14:01

I getting confused here - would you mind posting a complete summary of the system parameters? eg field strength, frequencies, power levels etc. Its all got a bit scattered through the thread, and I can't quite work out the current state-of-play. Don't want to make stupid suggestions based on wrong data ...

un0me2 - 28-9-2010 at 15:39

Ok, as it stands at present

1. The system will be utilizing a 1T (actually 0.989 ± 0.001T) field, with the proton frequency being ~43-44MHz, based upon a modified Halbach Array.

2. The system will be utilizing the simplest possible electronic spectrometer

(a) the original will be based upon a PIC32 Board from Sparkfun (availability)

(b) the next step will be a bread board design based upon PCB breakouts for the various chip packages, including, if needed, power supply, USB 2.0 (Silabs USB-UART), MCU (Silabs), Back End (MAX2306 + multiplexed 8 channel, 24-bit ADC's on the MCU), Front-End is based on a simple Class D Amplifier (MAX97xx), which can presumably take in a Digital signal (given the use of the same by the Class D) to output a sine-wave at nW to the speaker. The DAC can be programmed to give whatever frequency (based upon the onboard clocks) / speed of pulse, the length of the pulse can also be programmed based upon user input.

(c) the final design will be a fully custom designed PCB with design based upon the changes that end up being made in part (a) or (b). This may well include small Golay coils to improve the homogeneity of the field, we'll see if that is necessary, if it is a DAC would be useful.

If possible I'd like to avoid using excessive power, there are 2-3W available using a 3.3V-5V supply which is available from the USB 2.0 if I am reading the specs correctly.

I am trying to optimize the design so as to avoid under-utilization of the components, and to ensure the fullest possible utilization of what is on the board, while ensuring the highest possible sample bit-rate & speed (the Silabs MCU @ 24-bit is the best I've found for the price). The Digital Frequency Synthesis utilizing one of the DAC channels coupled with a D Class Amplifier (which already have to transform a digital to an analog signal) removes the need for external frequency synthesis.

arsphenamine - 28-9-2010 at 17:22

How to do you generate the 42.5MHz signal for excitation and downconversion?

un0me2 - 28-9-2010 at 18:06

With the Digital to Analog Converter /(ADC Tutorial, PWM Tutorial, Analog Devices, App. Note.928). Particularly with the Class D Amplifiers, which should take the rough edges off, you could pretty much just output a triangular waveform to the Amp, which would then convert it to an amplified analog signal. Here is a primer on Digital Synthesis (a Tutorial and a primer).

The point is that it is probably inefficient to filter it too much between the DAC & a Class D Amplifier, which essentially converts it back to digital, amplifies it and outputs it as an amplified analog signal. It has to be measured in Watts because it cannot be measured in dB, it is inaudible.

arsphenamine - 28-9-2010 at 18:44

If your 42.5MHz signal is generated by PWM, what are the pulse widths and clock rate of the PWM output?

Which chip's PWM had you planned on using?

un0me2 - 28-9-2010 at 20:09

With the PIC I'll have to use the PWM ports for Digital Synthesis, with the other chip (the Silabs MCU) one of the DAC's will be utilized. I haven't had a chance to look at the programming for the PIC, but there are several examples of using smaller PIC chips for sine wave generation.

arsphenamine - 28-9-2010 at 21:05

Look at the PWM clock frequencies required to generate a given sine wave frequency.

The PWM tutorial demonstrates how to generate a 4kHz sine by pumping a low pass filter at 30kHz or 500kHz.

That suggests that a 42.5MHz sine might require a ~320MHz PWM clock.

For one, if the internal CPU clock on the PIC32 is only 80MHz, there may be difficulties generating the 42.5MHz.
And two, the processor crystal's accuracy would determine that of the generated clock.

Is a DDS chip completely out of consideration, despite Analog Device's seeming monopoly on low RF DDS IC market?

Twospoons - 28-9-2010 at 21:18

Quote: Originally posted by arsphenamine  



For one, if the internal CPU clock on the PIC32 is only 80MHz, there may be difficulties generating the 42.5MHz.
And two, the processor crystal's accuracy would determine that of the generated clock.


Seconded: 80MHz clock will let you create 300kHz PWM with 8 bit resolution.
Crystal accuracy should not be an issue, ordinary crystals give 30ppm stability, a TCXO is even better (1-2ppm). Or, for lots of $, you can use an ovenised crystal (better than 1 ppb!). Phase noise might be more of an issue.

Also, that maxim amp is an audio amp and wont work at 40MHz.

Its looking more and more like DDS is your best bet. Or at least a fractional N synthesiser.

And I hope you realise that the 24 bit Silabs converter will only do 1ksps?



[Edited on 29-9-2010 by Twospoons]

un0me2 - 28-9-2010 at 21:43

Righto, what about DS1085L? That will do the job in terms of the frequency?

OK, what amplifier is needed?

As to the 24-bit resolution @1ksps, we are back to the trade-off that started the argument, a faster sampling rate or better resolution?

densest - 28-9-2010 at 22:10

The DS1085 spec doesn't include short-term frequency stability or phase noise. That suggests it isn't very good.

arsphenamine - 29-9-2010 at 06:45

Quote: Originally posted by densest  
The DS1085 spec doesn't include short-term frequency stability or phase noise. That suggests it isn't very good.
Agreed. The accuracy is not good enough for the application, either.
It synthesizes the frequency but has no external crystal.

If the clock source is selected for accuracy, the Maxim-IC tables
turn up chips like the MAX9450 family and MAX3674; purchased directly,
they cost $24 and $20 respectively, or nothing at all if you can
get a free sample.

un0me2 - 29-9-2010 at 07:11

That MAX2306 Complete IF Subsystem, has dual synthesizers and dual output. If one were to run one input into the coil post-pulse as a carrier wave, then it could be removed on demodulation, leaving noise + resonance. The sampling rates for the high-resolution sampling come down under the Nyquist boundary if the only thing left in the sample is the resonance and noise.

arsphenamine - 29-9-2010 at 08:34

That MAX2306 is tuned by a RLC tank resonator circuit which
is inferior in accuracy and stability to an OCXO/TCO system.

Can it use an external clock source instead a tank circuit?

not_important - 29-9-2010 at 10:28

The MAX23xx uses an external reference, that's where the TCXO/OCXO come in to provide the stability needed. The PLL(s) in the chip take care of short term noise; getting the desired offset for the IF desired may be a problem without using a switchable dividing PLL or DDS. Given that the FID is acquired over 1/2 to 2 seconds, you need around 1 part in 10^8 stability over that interval to get good resolution.

Generally PWMs aren't that good for signal generation, being intended for driving motors and such. In the case of excitation for a FT-NMR it might work, I've seen descriptions of such systems that used effectively narrow band noise for the excitation pulse.

Note that some of the devices you chose have LVPECL outputs.

Also "the simplest possible..." does not necessarily mean the fewest number of chips, particularly if those chips are being pressed into applications on the end of their design space. Attempting to get away with only one coil makes the receiver + signal processing more demanding (hint - it's more than just filtering, saturation and recovery time come into play).




un0me2 - 29-9-2010 at 14:06

Microchip have dedicated DAC's and ADC's that will solve both resolution problems, high-speed with high-ish bit rates. I'm seriously looking at them with the normal PIC32, the dsPIC's don't have enough benefits to justify getting rid of the memory & processing space (they can be run directly from the PIC32 through the SPI/I2C interfaces as slaved devices).

I have to look further at the outputs, I am seriously going through at the present time looking at how to avoid having parts from too many manufacturers, or the code to run this will be a Frankenstein-type monstrosity.

If I run dedicated high-speed DAC & ADC chips then the PWM channels on the PIC32 could be left alone waiting to be pressed into use as electromagnetic coil drivers.

As to the utility of multiple circuits, etc. I am looking into designing a Quadrature Modulation - Demodulation circuit.

What I am thinking is that with multiple DAC outputs it should be possible to build the circuit that is needed. I was thinking that if the receiver coil was running on the same rf signal as the larger rf pulse, it may cut down on echo & recovery time for the second coil. If the constant signal is run through the output coil to the input coil, then both are running on the same frequency.

It is designing a switchable, much higher magnitude pulsed signal that is going to be the hard part. That said, if the pulsed signal is the same frequency as the rf carrier signal, it should make it easier to filter out the echo from the pulse.

[Edited on 29-9-2010 by un0me2]

watson.fawkes - 29-9-2010 at 15:30

Quote: Originally posted by not_important  
Attempting to get away with only one coil makes the receiver + signal processing more demanding (hint - it's more than just filtering, saturation and recovery time come into play).
There's also testing involved. I can't imagine how to a single-coil system running without a second coil in the lab somewhere to test and exercise both the pulse generation and receiver circuits. And for a hobbyist who's making exactly one of these, there's no particular reason to make a test coil that won't end up in the final device. I just can't see why using two coils in the device is a downside in this scenario, where a certain amount of the test equipment cost should be considered part of the device cost.

un0me2 - 30-9-2010 at 16:34

Coil design is a LOOOONG way down the track, getting hold of the electronic components, etc. is taking time enough. Let's finish the electronics design argument, then we can get into the nitty gritty of the coil design.

not_important - 30-9-2010 at 17:59

Needed electronics depends a little on coil design. Coil impedance affects the transmitter and receiver designs, this isn't an electromagnet connected to the power mains. A single coil needs much better protection of the receiver section.

Quadrature demod is often part of receiver chain chips, the shifted signal may be derived by that or generated in the clock generation section depending on what you pick for each. Building it up from matched transistor such as Tim posted can give a bit better performance at the cost of more components and more tweaking; more fully integrated ones are usually easier to use if you're not experienced in the art.

I've worked on products that used processors from 4 different families, no code nightmare - each is a separate section, data and control signals going between them are interface protocols tht you have to understand anyway no matter how many types of processors.


un0me2 - 2-10-2010 at 07:20

Can someone please tell me I'm reading this wrong, I'm tired as hell and have been looking at this shit all day (so it is on the cards), but this Application Note (#2009) "Performance of the MAX2395 PLL with 80kHz Comparison Frequency" seems to me, to be suggesting that with some careful, well, there really isn't a better word for it, hacking of the external circuit - the [url=http://www.maxim-ic.com/datasheet/index.mvp/id/3974[/url] (rated @ 1920-1980MHz) can be made to output at significantly lower frequencies (500Hz-~20MHz). If that is the case, then that really does provide a good solution to the problem.

PS I have noted several online discussions about the utility of the Si5xx series of OX/VXCO's and am wondering if anyone has experience with the Si401x series of programmable (27-960MHz) transmitter on a chip? From what I can see of the diagram, a DAC could provide a high-powered pulsed-tone (for x-nsec) @ the excitation frequency & that would be that?

[Edited on 2-10-2010 by un0me2]

not_important - 2-10-2010 at 10:31

Why are you trying to run this at 500 Hz? I can't see a good reason to do so.

Likewise I'm puzzled by what you're trying to do with the oscillator. For the excitation coil you gate the power amp on for some number of cycles of the excitation frequency, the pulse is a number of cycles long. I.m not fitting that to what you propossed.

What are you hoping to be able to do with the NMR system you want to build? The intended application sets some requirements for things such as resolving power and stability of signals. Are you after measure water content in ice or hydrocarbons? Easy. Analytical and structural determination? Harder. Doing 1D or 2D? You need to set out your target clearly before you start doing designs.

un0me2 - 2-10-2010 at 23:13

No, not the point - they are achieving outputs several hundred MHz under what it is rated for.

As for the oscillator, I'm trying to time the DAC output, high powered tone @ the right frequency, added into the RF & then only the programmed period of output from the DAC, gives the power & the pulse, no?

That said, the dual-DAC's and the MCP3901 dual 16-24 ΔΣ ADC's (with dual output), should really improve the signal quality (so should proper separation of the Digital/Analog sections, grounding, etc.). The improvement in the Microchip Library means that the data can be dealt with (DFT/FFT/RADIXn) prior to sending it. Connecting the DAC to the PIC is dealt with in this Application Note, while some of the design considerations are dealt with in this Application Note. The code for several processors, the Texas Instruments one, DSP, etc. are on the web, while these IC's look promising (damn, a discrete chip returning the FFT of signals virtually in real-time? That should change shit:P)

With the transmission/reception modulation/demodulation, is it possible to transmit a baseband just under the LF, ie.40MHz? Then add the excitation pulse with 2.xxxMHz? That would allow for the same signal processing train...

[Edited on 4-10-2010 by un0me2]

not_important - 5-10-2010 at 18:09

Quote:
As for the oscillator, I'm trying to time the DAC output, high powered tone @ the right frequency, added into the RF & then only the programmed period of output from the DAC, gives the power & the pulse, no?


Still not making sense to me. If you have the RF, what is the DAC output tone for? If your thinking of using that as the pulse gate, it's something you'd use a DAC for given the roughly 4 or 5 orders of magnitude of off vs on times.

And what you want the NMR to be capable of doing is _very_ important, as that drives the design.

1) what do you want it to be able to do?
2) determine needed specifications - field strength and uniformity, frequencies, noise levels, phase stability, ...
3) figure out that hardware you want to use to satisfy 1) and 2)

aliced25 - 6-10-2010 at 01:33

Sorry cannot access other account (for anyone who is reading this 2+2 generally does =5, there is always a part of the calculation that comes down to common sense).

not_important,

What I am trying to realise is a high-speed, high accuracy digital-to-analog and analog-to-digital signal train. For the most part (except for one offering from Silabs) there really isn't a whole lot that suggests this is feasible from one manufacturer, which is a bugger. That said I have decided to "evaluate":D the AD9852ASTZ for the signal generation and the AD7606 8-channel 16-bit, high-speed ADS (ADC by any other name) for the return signal.

Without using a carrier wave that means there is going to be a god-awful discrepancy between the initial signal strength and the end-strength, so I'm thinking of looking at putting in a switch (to ground presumably), sending the overloaded signal generated immediately upon excitation (and a little while thereafter) into the dirt (so to speak), then passing the lot through a low-pass filter with a cut-off immediately above the highest resonance signal which is likely to be generated.

That would reduce the amplitude of the samples, thereby reducing the effect(s) of the echo/etc, while sacrificing some resonance data.

AD is coming out with a 400MHz Blackfin for ~$3-5 by the end of the year, which would see them squarely in the running for the MCU/MCP (at that speed the FFT/Radix-n of the samples should be almost real time).

On the other hand, Microchip are still in with a very good chance, the PIC32MX795F512H (with integral USB 2.0 transceiver), 512kB Flash & 128kB RAM can be used (I2C/SPI) with the MCP3901 Dual-Channel 16/24-Bit ΔΣ ADC's (up to 64ksps, up to 5 on a SPI/I2C IIRC), while the DAC - MCP4728 is not only a DUAL 12-bit DAC, but it is cheap & fast as hell (and according to this paper, should be quite capable of handling the DDS, while they provide a program to design low-pass & stop-pass filters that work up to 10MHz (presumably anything above that is discarded).

Working out how to design a decent board that will allow me to separate the analog & digital ground planes while breaking out the 64-100 XXQFP chips is going to be a problem, but I was thinking, once the digital end of this is set up to send analog signals to the analog end and to receive & process the responses to the same, that is a digital spectrometer, right there (actually, given the ability to do the FFT/RADIX-4 on chip, an FT-Spectrometer), which could then be utilized to control a laser (FT-Raman), a UV-Vis Light Source (FT-UV-Vis), an IR light source (FT-IR). The spectrometer is essentially the same (the digital end anyway), digital to analog control signals, analog to digital response signals and digital processing. With some thought, that could be taken way further, the pwm/dac outputs that aren't being used could be used to drive a moving stage, etc.

[Edited on 6-10-2010 by aliced25]

not_important - 7-10-2010 at 14:32

Quote: Originally posted by aliced25  
...
Without using a carrier wave that means there is going to be a god-awful discrepancy between the initial signal strength and the end-strength, so I'm thinking of looking at putting in a switch (to ground presumably), sending the overloaded signal generated immediately upon excitation (and a little while thereafter) into the dirt (so to speak), then passing the lot through a low-pass filter with a cut-off immediately above the highest resonance signal which is likely to be generated.

That would reduce the amplitude of the samples, thereby reducing the effect(s) of the echo/etc, while sacrificing some resonance data.
...


If by "initial signal" you mean the excitation bang, that's true. That's also why you isolate the receiver from the pick-up coil and have the pick-up coil at right angles to the driver coil. Isolation is done by several methods, a simple one is to put a diode across the input and run DC through it during the excitation pulse; the DC putting the diode into low-Z conduction mode, no DC and the low amplitude of the desired signal means the diode is a hi-Z shunt that doesn't affect the signal.

But it's still not making much sense to me as to what is you are attempting to do. Possibly it's my never made it past systems engineer in the world of electronics.



aliced25 - 8-10-2010 at 00:07

OK

I'm going to have to do up a diagram showing the concept I've come up with...

Can anyone help point me in the direction of where arsphenamine came up with the 160MHz figure? IIRC it was the maximum resonance return that could be expected from a 15ppm shift.

I would just like to confirm this (as I am using TI excellent FilterPro Program to design the filter for the analog return signal, I'm thinking put the passband @/~ 160MHz and the stop band @/- 240MHz (that, with a gain of 50v/V or 33.98dB (and a negative gain of -50dB for those wavelengths outside the band), returns as a 4-Stage (8th Order) Chebychev Filter (with a straight line @/~160MHz), so I'd like to make sure that I have it right before I source the components & build it).

I then have to determine how to synthesize an accurate frequency (the AD985x series aren't filtered).

I also then have to work out which amplifier fits with which ADC combination, the trick is going to be the bandwidth 160MHz is still pretty chunky, especially if the majority of the signals are 1/10th of that.

I'm looking toward the fast Delta-Sigma ADC's interfaced to the MCU (whichever variety), with several small (1/2 channel) ADC's running on an I2C/SPI/Parallel bus, sending the data into the processor. As the Delta-Sigma ADC's send, effectively, a noise-reduced, averaged signal, then collecting that will give a truer representation of what is in the mix.

But I'll quickly explain the system as I see it, purely the spectrometer end - the analogue end will have to be worked out later.

Interface + MCU

USB 2.0 Port <==> MCU (with USB Transceiver + SPI/I2C/Parallel buses).

Output

The output will consist of either dual interleaved DAC's with a sine table & oscillator or a Programmable Integer-N Synthesizer + at least 1 PLL with onboard clock, etc. That should really give a fairly accurate signal (based upon what I've read). As the FGPA's have really only turned the output on and off, I'm confident I can achieve that with the MCU. The output can then be amplified until it meets the needs of the coil. The output will then be passed to a 50 Ohm RF Connector Socket.

Input

The analogue input will be taken from a 50 Ohm RF Connector Socket & filtered down to 200MHz (max - see attached). The filtered analogue signal will then be further amplified. It will then be passed to the multiple Dual ADC's (either high-speed, high-resolution SAR/Delta-Sigma). The ADC's will then transmit the 16/24 bit signals to the MCU via the relevant bus (all the ADC's will be slaved).

Digital Signal Processing

The data stream from the ADC's will then be further processed, with 2 filter algorithms, one for echo cancellation, the other for noise cancellation. They will then be averaged across the time domain and the average will be kept. They will also be, best case scenario (according to the application notes it is well within the MCU's capacity), an FFT/DFT will be performed on the entirety of the sample/data-set, giving an accurate FT-NMR result which, with the sample average, can be sent to the PC/Laptop/SD Card/etc.

Other Issues

I will not know until I can test the board & then try it with a coil in the magnet assembly, whether the level of homogeneity is adequate. If it is not, Golay Coils will be needed, but as no use has been made of the MCU's onboard DAC/PWM, then it should be possible to utilize these controls for the Golay Coils.





Attachment: FilterPro.Design.Report.Schematic.Version.1.0.pdf (269kB)
This file has been downloaded 790 times

arsphenamine - 8-10-2010 at 06:36

Quote: Originally posted by aliced25  
Can anyone help point me in the direction of where arsphenamine came up with the 160MHz figure? IIRC it was the maximum resonance return that could be expected from a 15ppm shift.

No, because I never wrote that.

The signal return is estimated by the ratio of up-down proton spin states
and which, at its heart, contains a magnetic field term over a Boltzmann thermal distribution.

The ratio is very small, but the great number of protons in
organic matter means that it is sufficiently detectable.

The NMR wiki has links to good pedagogical material.

watson.fawkes - 8-10-2010 at 06:58

Quote: Originally posted by aliced25  
I then have to determine how to synthesize an accurate frequency (the AD985x series aren't filtered).
Yeah, that's kind of a problem, isn't it. You're trying to measure frequency shifts in the range of a few parts per million, and you need a frequency reference with stability of less than that. One bad stage anywhere in your transmit/receive chain and you'll lose it all. Since I haven't seen it mentioned before, only hinted at, drink deeply the concept of "error budget". I did a brief search and this introductory article from Electronic Design came up first. It's by Bob Pease, and it's reasonably motivational. Now with this project it's not just a single circuit, but an entire instrumentation chain. Nowhere in this whole thread have I seen any component choice constrained by error analysis. And I don't mean anything about bit lengths or bandwidth queenery. I'm talking about the analog stages. It doesn't matter if you're sampling 24 bits at 1 Ghz if your analog noise is up at 100 ppm.

I would heartily recommend doing this project in stages, and to make the first stage be a CW transmit/receive pair at the frequency of interest, paying special attention to frequency error issues. What's your calibration standard going to be for frequency stability? It's a pretty basic question, because it's relatively easy to make an oscillator that looks stable to the naked eye, which really means that it's at least as stable as your oscilloscope. Every oscilloscope has some internal timebase error, so you can't determine stability better than the capability of your scope. So, what's your calibration standard?

And I'm not really referred to absolute frequency measurements, because you can eventually determine that with known samples. I'm speaking of calibration for oscillator stability, such things as jitter, phase noise, frequency drift, and the like. You'll need a stable time base somewhere in your lab, or access to one in someone else's lab. You could, for example, build an OCXO unit or buy one. I did a little looking around, and the cheapest one Digi-Key sells near your band, AOCJY-100.000MHZ-F, is $144 retail. It's got +/- 30 ppb stability (that's "b" as in "billion"), so that'd be adequate for measuring. Or you could build one. An OCXO unit all by itself would be a solid little open hardware project. It'd be particularly useful because you could more easily get custom crystal frequencies. But back to the question: what's going to be your metrology standard for frequency stability?

After you've got CW working at some known stability, then you can worry about high power and pulsed switching. This isn't trivial either, since the transients induced by switching definitely have the capacity to wreak havoc with your frequency stability. But you can't possibly know if they are or not in the absence of a measurement and calibration regime.

not_important - 8-10-2010 at 17:40

Seconding watson.fawkes It's worth getting a highly stable frequency reference. Use a PLL+dividers to get other frequencies (f<sub>r</sub>*M/N ), use a DDS chip to get fine control of frequency and phase and to quickly shift between several pre-selected frequencies. Filter those outputs with bandpass or low pass filters to get good clean sine without higher frequency stuff.

Remember that for many of the applications you want to do, your talking about resolving sub-part-per-million values and collecting data over hundreds of milliseconds; if your base time/frequency reference isn't up to it, it matters not how good the analogue chain is or how fast your ADC and DSP are. A good OCXO is worth it, pick a useful frequency that conveniently supports the other frequencies and time intervals you need (20 MHz gives 50 nsec resolution and all that). This is especially true if you do spin-echo to compensate for inhomogeneity in the magnetic field.

I'd separate the ADC+FFT/DSP section, basically the digital stages, from the analogue front end - put them on separate PCBs with low-Z analogue signals plus SPI/I2C controls and maybe some other simple digital signals; USB goes on the digital board. This makes it easier to get low noise in the front end, keeps all the noise from DSPs and RAM away from it. Also there is fair to considerable differences in the analogue for the applications you listed, it would be simpler to do an application specific front end. FT-NMR needs strong signal generation plus low level amplification and translation to lower frequencies. FT-Raman sees a fairly narrow spectrum, input signals of under a MHz, and ramped or stepped power output; Ft-IR is similar except the spectrum is wider and the signal band of lower frequency due to slower detectors. Doing FT-UV-Vis is somewhat demanding on the optics including sensor, you're talking at least 300 to 800 nm, 200 to 1000 for a good system and sensors for that much range are likely rather pricey - the FT techniques often used for wideband UV-Vis typically use 2D detectors. For the optical stuff you often have a reference channel that is processed the same way as the sample one, twinning the front end. No offsetting from some reference frequency, no IF stages, oftn no I/Q signals.






arsphenamine - 9-10-2010 at 08:21

re: ppm vs. Hz calculations

Assumption: anything you want to see in 1H NMR is within 15 ppm of the Larmor Frequency,
although most of it is within 8 ppm.

ppm = parts per million from which you infer a 1e-6 multiplier.

15 ppm of a 42.5MHz Larmor Frequency = 15e-6 * 42.5e6 = 637.5 Hz.

The phrase downfield shift was current in pre-FFT (viz., continuous wave) NMR spectroscopy.
Back then, it was easier to decrementally sweep the magnetic field than with an RF oscillator.

A 15 ppm shift in 60MHz 1H NMR is equivalent to a 0.2 Gauss field change.

0.2 Gauss.
A miniscule 0.2 Gauss.
An amount slightly less than the minimum observed Earth magnetic field.
In a 14092 Gauss field.

You need a frequency standard that is at worst 100x as stable over
the measurement interval, i.e., no more than 150 parts per billion.

The fineness of field and frequency control required is extreme
even by today's technological standards.


not_important - 9-10-2010 at 08:54

Note that you don't need the high quality reference to start with, you can use a simple crystal oscillator to get things debugged, and even see some NMR results. But for any sort of analytical work you'll need to sub in a hig-stab source.
Quote:
analog return signal, I'm thinking put the passband @/~ 160MHz and the stop band @/- 240MHz

'The results in your PDF are for KHZ, not MHZ ...

arsphenamine - 9-10-2010 at 16:19

Quote: Originally posted by watson.fawkes  
An OCXO unit all by itself would be a solid little open hardware project.
+1

A search on "crystal oven" turns up some delightful projects ranging from
a pure analog design that heat the crystal with a power transistor and sense by thermistor,
to microcontroller PID systems using the venerable LM35 linear temp sensor.

aliced25 - 9-10-2010 at 16:37

Quote: Originally posted by not_important  
Note that you don't need the high quality reference to start with, you can use a simple crystal oscillator to get things debugged, and even see some NMR results. But for any sort of analytical work you'll need to sub in a hig-stab source.


Quote:
analog return signal, I'm thinking put the passband @/~ 160MHz and the stop band @/- 240MHz

'The results in your PDF are for KHZ, not MHZ ...


Um, yeah sorry about that - I was thinking something else entirely and wrote the wrong thing (My bad), the actual filter design was intended to cut-off from 240kHz with the 0-160kHz pass. I do seem to "recall" reading that some of the expected return was in the 150kHz region, if not that makes life easier, the smaller the bandwidth of the signal fed to the ADC, the smaller the reference bandwidth and the better the result.

I'm actually looking for a way to narrow the generated signal - ie a lowpass filter, at the requisite band, which would also help to ensure stability in the bandwidth (but not necessarily the strength transmitted) of interest.

As to physically separating the digital & analog, I was thinking that if separate ground planes were used it should be possible to keep the noise down (especially if the power source is separated), these have everything on one board and I'm presuming the signal integrity would be at least as high. It's also good to see things like that, it helps me to realise that I'm not "inventing the wheel" (for all that I AM reinventing it somewhat:D)

not_important - 9-10-2010 at 17:44

Separate grounds is a requirement given the low level signals. But NMR will need a bunch of more-or-less analogue functionality that the other instruments won't, makes more sense to build that into a head for NMR and send the AF-ish signals to the digital section over low-Z twisted pair.

Also note the phase shift for the amp design, that might be important in some applications.

Note that crystal oscillators may have short term jitter and hase instability, syncing a PLL with a rather low pass feedback can improve those factors.

As those pages note, good insulation surrounding the OCXO is helpful. Enclosing that in a container with high volumetric heat capacity (Cu or Fe) can also be helpful as it reduces the impact of sudden changes in external temperature.
Quote:
I'm actually looking for a way to narrow the generated signal - ie a lowpass filter, at the requisite band, which would also help to ensure stability in the bandwidth (but not necessarily the strength transmitted) of interest.


Which generated signal? The received one? That's what one of more IFs help with, followed by low pass filtering of the 'audio', high rate oversampling of the signal (means that a wider range of frequencies above the band of interest get properly digitised without folding over) followed by digital low-pass filter.




[Edited on 10-10-2010 by not_important]

aliced25 - 9-10-2010 at 19:00

I was actually talking about the 43-44MHz signal, the use of a filter on that to keep the output within a specific narrow bandwidth is what I'm looking for.

As for filtering and amplifying the return signal, if the resonance signal is in the "audio" range, then the number of options in the way of low-cost, high-precision filters, amplifiers, etc. grows exponentially (some go to 100-120kHz). High speed ADC's, including MSPS ΔΣ ADC's with oversampling, are much easier to match to the audio bandwidth than any other (obviously, that is the purpose behind the manufacture thereof). High-end audio components cost next to nothing when dealing with IC's, they are high-volume products.

I'm looking at Digital Interpolating Filters for the DAC's and/or integer-n PLL's (all digital PLL's are available) but fully integrated circuits aren't available at the 40-50MHz range unfortunately (well some are, but they still need cleaning up). High speed interleaved 12-bit Dual DAC's with PLL & a filter would allow for much easier pulse-generation & timing than would be possible with other components. Just pre-program the table look up and the pulse length & subsequent wait period, with the number of pulses also set by programming.

12AX7 - 9-10-2010 at 21:39

http://wingzero.ath.cx/index.php?page=xtal-osc-varactors

not_important - 10-10-2010 at 11:52

Quote: Originally posted by aliced25  
I was actually talking about the 43-44MHz signal, the use of a filter on that to keep the output within a specific narrow bandwidth is what I'm looking for.


It's called a tuned circuit, or a bandpass filter - both done with inductors and capacitors. With most M/N PLL and DDS chips, the leftovers from generating the target frequency are much higher than the desired output, simply filters take car of them.
Quote:
As for filtering and amplifying the return signal, if the resonance signal is in the "audio" range, then the number of options in the way of low-cost, high-precision filters, amplifiers, etc. grows exponentially (some go to 100-120kHz). High speed ADC's, including MSPS ΔΣ ADC's with oversampling, are much easier to match to the audio bandwidth than any other (obviously, that is the purpose behind the manufacture thereof). High-end audio components cost next to nothing when dealing with IC's, they are high-volume products.


The FID signals are standard NMR, plug in your magnetic field and see what the value will be - 10s of MHz for the fields you've been talking about. You need to mix the received signal with the high stability reference to generate the lower frequency stuff. You still need to worry about the image on the other side of the reference, which is why there's generally at least one IF stage - take a, say, 80 MHz FID signal, mix it down to 10.7 MHz, the image frequency is 21.4 MHz away, 10.7 MHz on the other side of the reference, so your RF front end just needs to block that range well (in this case 1/8 the reference or desired signal, no prob). Then demod the IF to AF, or use a second lower frequency IF followed by down-conversion to AF. As noted earlier, generation a reference + 200 Hz (for example) frequency for the down conversion means that the signals you want will all be in the audio range with no DC or really low frequencies to worry about.
Quote:

I'm looking at Digital Interpolating Filters for the DAC's and/or integer-n PLL's (all digital PLL's are available) but fully integrated circuits aren't available at the 40-50MHz range unfortunately (well some are, but they still need cleaning up). High speed interleaved 12-bit Dual DAC's with PLL & a filter would allow for much easier pulse-generation & timing than would be possible with other components. Just pre-program the table look up and the pulse length & subsequent wait period, with the number of pulses also set by programming.


I still don't see this, it sounds quite complex for just generating pulses of RF; DACs are just not making sense. Just use a diode bridge or balanced mixer to gate the RF. The RF should be derived from your master reference, and you don't turn oscillators on and off if you're expecting any sort of solid, stable signal out of them. The control pulses come from timers counting at a high enough frequency to give the needed resolution, also derived from the master reference. Some signal generator (PLL/DDS) chip include smart gating that allows near zero-crossing gating; or you can Schmitt trigger the RF and count cycles to generate gating signals locked to the RF.

BTH - the DACs you referenced in a previous message won't hack it - check the settling times, and how long it takes to get a setting into the DAC even at max control line rate.





[Edited on 10-10-2010 by not_important]

Twospoons - 10-10-2010 at 14:33

Regarding OCXOs: Not a trivial project by any means - I spent 6 months working for Rakon on their double-oven oscillator. That's an ovenised oscillator inside another oven!. One of the tricks is using an SC cut crystal, which has a cubic Tc, and operating it at a temperature which matches one of the turning points of the Tc curve. Managed to get down to 0.2 ppb with the best one. To measure it we used a Trimble 10MHz GPS reference oscillator - which makes use of the 30 odd atomic clocks circling overhead.
The crucial parameter for this project is going to be phase noise, as this will contribute directly to spreading of the demodulated signal, and therefore its visibility amongst the noise. I suggest finding the lowest phase noise TCXO you can and make a small oven for it. You should be able to get 0.1C stability on your oven without too much trouble, and let the TCXO take care of the rest. I'm thinking that as long as the short term stability is very good, any absolute error can be compensated for later.

aliced25 - 10-10-2010 at 18:33

Ok, found a processor & a half, have a look at this "FFT Implementation on the TMS320VC5505, TMS320C5505, and TMS320C5515 DSPs" (SPRABB6A) from Texas Instruments, the FFT is hardware optimized & it has some 500kB of memory on disk - with USB 2.0. Difficulty, the TMS320C5515 is a 196 NFBGA (14x14) chip, but that can be overcome (not quite trivial though) It is the one that the ECG Solution and the discussion vis-a-vis that is rather interesting to read. That uses the ADS1258 16-Channel, 24-bit ΔΣ ADC running at 23.7KSPS/Channel (so 11.8kHz would be the Nyquist boundary?), which is amplified by the INA128 Precision, Low Power Instrumentation Amp with the REF5025 Precision, Low-power, Low-drift, Voltage Ref (which looking at what else it is used for would seem to the the pick).

Twospoons - 10-10-2010 at 20:31

I'd be inclined to do all the processing in a PC, and focus the design effort on the analogue and data acquisition side. You have a lot more options for processing software in a PC, as well as stuff all resource limitations. Looking at ECG solutions is a good idea - its the same sort of problem finding a small signal amongst a lot of noise.

not_important - 10-10-2010 at 21:07

I think you'd best read that document yourself. ECG apps typically deal with nothing higher than around a KHz, and that on auxiliary signals; the main signals top out around a tenth of that.

Yo're still going at this wrong way around. Effort in designing a system to satisfy unspecified requirements is mostly wasted. There's a load of rather importan electronics ahead of the DSP section of FT-NMR.

aliced25 - 11-10-2010 at 00:09

Yes, but the concept is similar enough to grab my attention - the need to attenuate a high-volume pulse, the need to filter out the majority of the noise, all pretty much what we are dealing with in this application, the size of the bandwidth of interest is the only thing that really changes...

Actually, on that subject, I'm looking everywhere I've been told to look, I'm not seeing a whole lot on the actual breadth of the bandwidth come to that, what is the area of interest? Failing that, how do I work it out?

The difference in specifications & price when dealing with 10's of KSPS to several hundred KSPS, to the MSPS range makes this kind of imperative. I know (now) that I have to sample at at least twice the bandwidth, but I'm fucked if I can work out where the bandwidth ends.

watson.fawkes - 11-10-2010 at 05:19

Quote: Originally posted by aliced25  
all pretty much what we are dealing with in this application, the size of the bandwidth of interest is the only thing that really changes...
Except for the analog section, which is completely different, harder to get right than the digital one, and where almost all of the stability, metrology, and calibration issues are.

arsphenamine - 11-10-2010 at 09:54

Quote: Originally posted by aliced25  
... I'm not seeing a whole lot on the actual breadth of the bandwidth come to that, what is the area of interest? Failing that, how do I work it out?
*facepalm!*

Understand the Larmor Frequency calculation.
Learn to perform ppm to Hz interconversions on NMR spectra.
Observe the practical ppm range of published 1H spectra.
Bandwidth figures will be incidental and obvious then.

I strongly urge that you gain practical experience with an NMR device.
That is to say that you must physically lay hands on the machine.

You need an experiential skeleton to flesh out with theoretical concepts,
otherwise this DIY NMR is only intellectual masturbation -- feels good, but
nobody else cares.

Many universities and community colleges have ageing functional NMR's
with which they train students. Arrange for a session or two.

Come back when you can explain the splittings and integration line on the 1H-NMR spectrogram of ethanol below.


Notes:
Sample was 5% in CCl4 from which you infer minimal intErmolecular hydrogen bonding from the -OH moeity.
0.0 ppm is the tetramethylsilane calibration standard;
1.2 ppm is a triplet,
3.7 ppm is a quartet,
4.8 ppm is a singlet.

Don't tell us about it.
Just get the skill and figure it out.

arsphenamine - 11-10-2010 at 16:56

@aliced25

You need this.
It interconverts field strength and Larmor Frequency.

[code]<script language="JavaScript"> function btof() { var bfield = document.ppfcalc.bfield.value; var freq = 267522205 * bfield / (2 * Math.PI * 1e6); document.ppfcalc.freq.value = freq.toFixed(9); } function ftob() { var freq = document.ppfcalc.freq.value; var bfield = 2 * Math.PI * freq * 1e6 / 267522205; document.ppfcalc.bfield.value = bfield.toFixed(9); } </script> <form name="ppfcalc"> <table style="text-align: left; margin-left: auto; margin-right: auto;" border="2" cellspacing="1"> <br> <tbody> <tr> <td style="text-align: center; vertical-align: middle;"> <input name="bfield" size="11" type="text"> Tesla</td> <td style="text-align: center;"> <input size="2" value="&gt;&gt;" onclick="btof();" type="button"> </td> <td style="text-align: center;"> <input size="2" value="&lt;&lt;" onclick="ftob();" type="button"> </td> <td style="text-align: center;"> <input name="freq" size="11" onchange="ftob();" type="text"> MHz<br> </td> </tr> </tbody> </table> </form>[/code]

Twospoons - 11-10-2010 at 17:10

That's interesting. It shows a shift of several hundred kHz with just 0.01T field change at 0.8T. So that magnetic field has to be exceptionally flat, otherwise the signals are going to be spread all over the band, and will be invisible amongst the noise. In fact, I would hazard a guess that flatness is two or three orders of magnitude more important than peak field strength.

arsphenamine - 11-10-2010 at 17:33

Quote: Originally posted by Twospoons  
... So that magnetic field has to be exceptionally flat, otherwise the signals are going to be spread all over the band, and will be invisible amongst the noise..
Ayup.

Magritek has a pleasant video on inhomogeneity.

It is part of a series of NMR/MRI video tutorials.

not_important - 11-10-2010 at 19:33

Quote:
So that magnetic field has to be exceptionally flat


Thus shimming, twisting screws and inserting brass shims to get the magnet pole pieces as close to parallel as possible. Also thus the spinning of the sample to expose it to in effect an averaged magnetic field in the X and Y axis; really useful only after shimming out larger field imperfections.

The crystal structure of permanent magnets create some inhomogeneities in the field, one problem with some designs using NbBFe magnets.

For some applications there's also a mid- to long- term stability issue, thus systems that maintain a lock on the <sup>2</sup>H in deuterated solvents, using an axillary coil to tweak the field strength in real time - sensing lock via CW NMR methods.


-----------------------------------------------------------------

You need to sample (at least) a bit more than twice the highest frequency of interest; both to reduce distortion and because real filters are not brick walls.

If you sample a sine wave at a rate exactly equal to the wave's frequency, you get a DC level. For signal frequencies slightly lower or higher than that you will see a low frequency - the difference between the sample rate and the signal being sampled; as you move away from the sample rate the apparent frequency increases. This means that you want the anal signal to have as little energy as possible above (or even at) the Nyquist frequency (1/2 the Nyquist rate). Rather than designing complex analogue filters, it's easier to oversample quite a bit with reasonable analogue filtering to take out frequencies a ways about the highest frequency of interest but still well below the Nyquist, then use digital filtering to improving the filtering of the frequencies about those of interest. End result is elimination of wrap around, generation of spurious signals in the result. Typically the filtered digital data is decimated before performing the FFT, as this reduces memory and processor speed requirements.



aliced25 - 11-10-2010 at 21:25

That is what I was looking for not_important, the lower end, medium speed ADC cited above is capable of operation well and truly above the Nyquist envelope, but I'm looking at the trade-off in speed:accuracy in ENOBs, how far above the Nyquist envelope is too much of a good thing?

I'm looking at the filter on that ECG - the one that is there to grab the over-voltage from a single-shot pulse and take it to ground, plus the low-pass filter implementation (such as I can generate with FilterPro). As I make clear (to arsphenamine below), all I need to know is the maximum ppm that anyone would want out of this unit - I'm presuming that noone is likely to want to try and source 13C at home? (then again, presumption would be the bastard-brother of all fuckups). If I only have to design for 1H-NMR, then 10 or 15ppm as a baseline reference bandwidth?

I was looking very closely at the digital filter specifications, that with the decimation and the ability to average the result set in almost real time (thereby saving each cycle as one averaged, decimated & noise reduced 32-bit word for later processing).

Any experience with digital isolators? This would appear to be precisely the sort of application that they would be targeted at? Here is an App. Note from TI (SLLA198). There are two reasons I ask, firstly, the suggestion that 100+ gauss within any manufactured device (fail-point) is unlikely (not here it ain't - there is a magnet not too far away that will be producing about 10 times that) is unlikely, but that this IC will damp EMF & EMI - ie. noise. I'm presuming it probably should straddle bare (copper free) board in the middle in order to maximize the isolation, but I noted one figure in the cited paper where they use precisely this IC to separate the Analog input from the Digital processor.

As for magnetic homogeneity, you'll note in the original paper from Danieli, et al, on the small magnet they designed, they were discussing how the design allowed them to get rid of inhomogeneity in the order of 20K ppm, well according to the model, this design can get down to 1K or even 500ppm. From that point, the only way forward (without designing a motor to spin the fuck out of the sample) is shimming. The major difficulty I'm facing is the 1% (minimum, generally well above that) rating on field sensors, Hall effect sensors, etc. It would appear that the best - indeed possibly the only - way to test the magnet assembly and get it down to the minimal ppb range would be to actually test a water sample (or a solution) as used by Dogan, et al. Jachmann, et al, describe an extremely interesting solution, a small Halbach array within a Halbach array which would be interesting to model, but I've promised myself that I'll work on the basic electronic component design area before I fuck around with anymore modeling (with too much open this PC does not like FEMM, tends to freeze up).

I actually suggested that the pwm/dac drivers on the DSP/MCU could be used to drive small shim coils, in fact I cannot see why it would be impossible (difficult yes, impossible we'll see) to design the system so that it senses inhomogeneity itself and auto-shims to correct it.

@ arsphenamine,

Thank you indeed for illustrating the various ppm shifts of two methylene groups and the OH, not precisely what I had in mind, but a nice start. What I was after is more of an indication of precisely what is the maximum ppm shifts one would expect OF a portable NMR device? Quite frankly given the time & effort it is going to take to design this sucker (and the board it goes on), I might as well look at shifting some populated/unpopulated boards for anyone who is interested in order to keep my costs down (and theirs). Fuck all point of that if the unit doesn't perform at the minima one would expect is there?:P

[Edited on 12-10-2010 by aliced25]

Attachment: Danieli.etal.Mobile.Sensor.for.High.Resolution.NMR.Spectroscopy.and.Imaging.pdf (746kB)
This file has been downloaded 1240 times

Attachment: Danieli.etal.Small.Magnets.for.Portable.NMR.Spectrometers.pdf (107kB)
This file has been downloaded 735 times

Attachment: Dogan.etal.Development.of.Halbach.Magnet.for.Portable.NMR.Device.pdf (783kB)
This file has been downloaded 920 times

Attachment: Blumich.Casanova.Appelt.NMR.at.Low.Magnetic.Fields.pdf (800kB)
This file has been downloaded 1166 times

Attachment: Jachmann.etal.Quadupolar.Order.Shimming.of.Permanent.Magnets.Using.Harmonic.Corrector.Rings.pdf (457kB)
This file has been downloaded 845 times

arsphenamine - 12-10-2010 at 07:14

Quote:
...all I need to know is the maximum ppm that anyone would want out of this unit - I'm presuming that no one is likely to want to try and source 13C at home? (then again, presumption would be the bastard-brother of all fuckups). If I only have to design for 1H-NMR, then 10 or 15ppm as a baseline reference bandwidth?
Pay. Attention. This. Time.

U Wisconsin has a good page on 1H-NMR chemical shifts.
Look at the shift tables at the top.

For ordinary organics (C,H,O), everything of interest is between 0 and 12
ppm. Don't take my word for it. Gloss over the charts.
15 ppm is easier for estimation purposes. So is 20 ppm.

Anything outside 0-12 ppm is usually a hetero (B,P,N) or metallic/semi-metallic that
you are unlikely to encounter or must not encounter outside of a dry box.
Quote:
I was looking very closely at the digital filter specifications, that with the decimation and the ability to average the result set in almost real time (thereby saving each cycle as one averaged, decimated & noise reduced 32-bit word for later processing).
In practice, it often reduces to something like this:

unsigned int sum = 0;
for (int i=0 ; i<16 ; i++)
{ sum += readADC() ;}

sum >>= 4 ;
return (sum) ;
Quote:
As for magnetic homogeneity, you'll note in the original paper from Danieli, et al, on the small magnet they designed, they were discussing how the design allowed them to get rid of inhomogeneity in the order of 20K ppm, well according to the model, this design can get down to 1K or even 500ppm.
Homogeneity coils for a small Halbach or Stelter array
will be small but easier to execute as printed circuits,
perhaps copper on Kapton.

First you make a Z axis coil...

Halbach Arrays are sexy, but I would consider designing a Stelter Array
since it uses alloy pole faces which, incidentally, improve field homogeneity.
The Stelter is also easier to design, uses block parts, and it, too,
is proven in NMR/MRI devices.
Quote:
From that point, the only way forward (without designing a motor to spin the fuck out of the sample) is shimming.
To this day, liquid samples spin under air pressure. See the original
patent for drawings, US 2,960,649 "Line narrowing gyromagnetic apparatus".
The practice is to spin the sample around 10 rps, fast enough to time average
out some field inhomogeneity, but not so fast that sideband peaks appear from
liquid turbulence. The sample tube mounts through the center of a plastic sawtooth
gear that spins from tangential air flow -- essentially a small turbine.
Quote:
It would appear that the best - indeed possibly the only - way to test the magnet assembly and get it down to the minimal ppb range would be to actually test a water sample (or a solution) as used by Dogan, et al.
Hold that thought.
Testing is your friend.
Quote:
Jachmann, et al, describe an extremely interesting solution, a small Halbach array within a Halbach array which would be interesting to model...
Figure that the Jachmann,et.al. design of static correction rings placed
them in the hands their machinist when it came time for executing the design.



not_important - 12-10-2010 at 09:44

Quote: Originally posted by aliced25  
...- I'm presuming that noone is likely to want to try and source 13C at home? (then again, presumption would be the bastard-brother of all fuckups).
...
It would appear that the best - indeed possibly the only - way to test the magnet assembly and get it down to the minimal ppb range would be to actually test a water sample (or a solution) ...


<sup>13</sup>C shifts are very roughly an order of magnitude larger then <sup>1</sup>H shifts. Both <sup>13</sup>C and fluorine could be done on the same unit as ordinary proton NMR, provided the electronic - coils and RF stage amps - can be tuned for them. The low concentration of <sup>13</sup>C does make it more difficult, but FT-NMR with good frequency/time stability can do it.

See the attached for a discussion of shimming, as well as other things that impact homogeneity, including the makeup of the coils themselves, dioxygen dissolved in the sample, and so on.



Attachment: shimming.pdf (639kB)
This file has been downloaded 875 times

aliced25 - 12-10-2010 at 21:22

<sup>13</sup>C shifts are an order of magnitude larger than the <sup>1</sup>H shifts, then aren't they going to be rather abruptly cut off by the multi-part active filter on the analog input? I'm not sure how I'd go about having a programmable active filter, at least on the analog side.

By the way, when I had another look at the ECG Solution (just down the page, under signal acquisition challenges (dot point #4), one of the design issues was that the design had to allow for the potential use of 0.5-1,000Hz in diagnostic mode. As the signals we want are well under the 1kHz range (even @20 ppm), then the question is, wouldn't the solution for that apply here? If that is the case, presumably we could design an active filter with the bandpass so it wouldn't cut off the <sup>13</sup>C resonance, incorporate the gross-overvoltage filter, and allow the 24-bit ΔΣ ADC to do the same job it does there, getting rid of any remaining noise with the digital filter & the noise & echo reduction algorithms?

I'm just trying to work out why so many separate shims are needed X,Y,Z I get, why multiples?

arsphenamine - 13-10-2010 at 08:22

13C resonance signals?

Don't worry about it.

Since there is usually less carbon than 1H in a compound,
and 13C's natural abundance is only 1.1% that of 12C,
and 13C's relative sensitivity is 1.6% that of 1H,
and its signal bandwidth is >10X that of 1H ...

13C signals appear as noise in the 1H spectrum of interest,
in the absence of clever excitation tricks and expensive low-noise amps
on the front-end of the receiver coil.

arsphenamine - 13-10-2010 at 09:11


Simplified shim coils on printed circuit boards:
Attachment: PCBshims_pat3735306.pdf (412kB)
This file has been downloaded 705 times

Note that this patent references several others:

The first two show the progress of simple magnet shim coils and suggest where the
current 18-30+ coil shim systems came from.

The last is by the venerable Golay and has diagrams of second and third order effects
from the initial XYZ homogeneity corrections. It shows you _why_.

You don't need to address them all (or any) so much as be aware of the limitations.

Lacking a sophisticated shim coil implementation, solid sample NMR is a waste of time.

Absent any shim coils, only low resolution spectra of physically spun liquid samples
are practical. Not too shabby, actually.

In accord with the Principle of Maximum Laziness, I'd go with a PCB version of one
of the simpler versions having only 2-3 coil types.

Finally, all these patents antedate superconducting solenoid magnets which use
more coils and more complicated shapes whose patents don't seem directly relevant to this project.

not_important - 13-10-2010 at 13:58

Quote: Originally posted by aliced25  
<sup>13</sup>C shifts are an order of magnitude larger than the <sup>1</sup>H shifts, then aren't they going to be rather abruptly cut off by the multi-part active filter on the analog input? I'm not sure how I'd go about having a programmable active filter, at least on the analog side. ...


Ignoring the <sup>13</sup>C shifts temporarily, consider a <sup>1</sup>H system that yields a Larmor frequency of around 30 MHz. If you were to RF amplify the signal, then mix it against a 30 MHz reference, the desired data would be in that audio range on up around 500 Hz (17 ppm * 30 Hz-ppm = 510 Hz).

But remember that when you mix (same as multiply) that reference against the input RF you get say 75 Hz from both a signal 57 Hz above the reference and 75 Hz below the reference. You just don't get filters narrow enough up in that part of the RF to dump the undesired frequencies while passing the desired ones.

So generally you don't mix against the zero-shift Larmor frequency of 30 MHz, but rather an offset frequency of perhaps 40.7 MHz, giving an intermediary frequency of 10.7 (used by broadcast FM receivers). For this the unwanted image frequencies would be 10.7 MHz on the other side of te reference, 40+10.7 MHz, or 21.4 MHz different than the NMR ones. That's plenty far away that simple RF filtering will reject the undesired frequencies in the RF section _before_ they get mixed with the reference - they just don't show up in the IF.

Then after a bit of amplification and filtering, you might mix against another reference to give an output in the range of 455 KHz - commonly used by AM broadcast and various 2-way communication systems. Why that range? First the image will be 910 KHz away from the desired signals in the 10.7 MHz IF, again fairl easy to reject by filtering; filters that are fairly low in cost can be used. And second, that's low enough that you'll find it easy to have a fairly narrow bandpass at 455 KHz, with low cost commercial filters common.

Now finally you convert the 2nd IF to 'audio. But once again you don't want to end up with the no-shift Larmor at 0 HZ, signals near DC are more difficult to deal with and best avoided if practical. And then there's the image problem, 2 Hz below and 2 Hz about the Larmor frequency (or its 455 KHz substitute) will both give 'audio' of 2 Hz.

Instead once again pick a reference that's offset from the Larmor, Let's pick a frequency 5 KHz away from the Larmor stand-in frequency, meaning NMR data in the 5 KHz plus/minus a few tens of Hz.

So now your analog audio filter is centered around 5 KHz. what does its bandpass need to look like? Let us assume a 16 X oversampling, meaning we sample at about 80 K samples per second. We want 'zero' energy up at that point, and as little as possible at the Nyquist - say around 7 KHz. It's not too difficult to design an analogue filter gives good performance to fit. But we really don't need to cut off that quickly, the analogue filter could have a high rolloff starting at say 10 KHz and still give good suppression at 40 KHz. Low side blocking is less critical, as the DSP side can deal with it.

Ru in through the ADC and you've a 80 K samples/sec data stream. Pass it through some digital filtering to isolate the NMR range of interest - around 5 KHz. Decimate, FFT, only grab the bins corresponding to the frequencies of interest, then renumber them to actual NMR frequencies or shifts in ppm.


In this fashion you can pick the final frequencies you want, easily changing them by tweaking the digital filtering while leaving the analogue filters fixed.

Having the analogue bandwidth greater than needed for <sup>1</sup>H NMR has advantages in that early set-up of the instrument is easier, handy for mapping the magnet field where field strength varies enough to give shifts much greater than those for actual NMR work, and so on. For that reason you might wish to pick as high an ADC rate as economic for the desired number of bits, allowing a wide band look at the <sup>1</sup>H signals for development and shimming, then 'zoom in' for actual NMR work.

Example filters

10.7 MHz http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&...

455 KHz the CFL455G3 on this page http://www.surplussales.com/filters/filters-1.html

12AX7 - 13-10-2010 at 17:57

Downside: two conversions adds two mixers' worth of noise, plus filters. I wonder how noisy ceramic resonators are.

The only signal below 0ppm is noise, so you're weighing converter noise versus double thermal noise over the same bandwidth. There is no external noise (interference), since the whole system shall be very nicely shielded.

Tim

not_important - 14-10-2010 at 14:52

True, and I'd use just one IF myself, with a really good RF stage. It can be difficult to avoid image problems without an IF, unless you offset the signals and do a fairly high speed ADC

During shimmy you can get shifts below apparent zero, it's handy to see them. Plus offsetting zero to see a bit of the negative shift range gives you some feedback that the system is functioning propperly - lack of anything but noise in the negative zone is reassuring.

12AX7 - 14-10-2010 at 16:11

Which is another good reason to set 0ppm to a few kHz, so you have the bandwidth below to see any shifts.

Tim

densest - 15-10-2010 at 21:17

If you have a "good enough" bandpass filter on your input, there is no energy in the images produced by subsampling. I've been sampling a bit of wine and beer tonight so please forgive any stupid mistakes:

A technique to avoid multiple conversions is subsampling: deliberately sample at a submultiple of the Nyquist limit of your carrier frequency.

Current audio converters give 100 dB s/n up to 96KHz and a few op amps can work there without degrading things below 90 dB or so.

30MHz +- 12 ppm in (signal) or 30MHz +- 360 Hz

Nothing outside of 30 MHz +- 1 KHz is of interest, so throw it away!

Filter to pass 30MHz +- 250 KHz - bandwidth is wide to accommodate magnet variation. Each magnet set would have to be calibrated to set the stimulus pulse and the filters trimmed to match.
Convert to 455 KHz
Filter +- 8 KHz (using AM radio technology)
Sample at (waving my hands) 48 KHz.
Digital filtering removes anything over 360 Hz.

This approach is commonly used in "software defined radios", removing one or all of the frequency conversions. If your input bandpass filter is narrow enough and linear phase enough, sampling in the audio range yields good data.

Waving hands again: since we control the carrier phase and frequency, it is possible to null it out during one of the early conversions, drastically improving the signal to noise (calling the carrier noise) ratio.

It isn't necessarily easy to do all of this well, but I've played with the math some and have played with circuitry and filters in the required ranges enough to think it's possible. Phase shifts are deadly if not compensated for; luckily, our required bandwidth is so narrow that simple filters will work.

Actually, 360 Hz around 30 MHz is narrow enough that crystal filters (see amateur radio circa 1960) would knock out everything but the useful signal and provide excellent S/N - a single conversion afterwards would suffice. The tricky part would be adjusting the magnets to hit the bandpass of the crystals. The advantage would be removing a great deal of the noise early in the chain. Such a filter is very very stable. Crystals ground to exact frequency used to be amazingly inexpensive - a few dollars each.

Ceramic filters are good (60+ dB s/n) or better. I believe that there are significant differences between manufacturers and styles of filters: SAW filters are very good at higher frequencies and have good S/N but are expensive unless you're using one designed for a big demand. L/C filters are way less fun (many stages, critical tuning, and awful tempco even with T/C capacitors (N750, anyone?) but can be quieter if one invests in really stable coil forms and lots of mechanical, electrical, and thermal shielding.

I am drifting in and out of this discussion and haven't read enough lately. Has anyone quantified the likely signal amplitude????? femtovolts, nanovolts, or microvolts? Anything above 10 microvolts can be considered to be trivial to work with in this context, since we are looking at multi-millisecond to fractional second data acquisition intervals.

arsphenamine - 16-10-2010 at 13:27

N52 Halbach segments for sale, 1.125T field claimed.

Plan B -- Stelter Array using 8 N52 1" cubes

Homogeneity is comparable at the 5mm gap core, the diameter of a standard NMR liquid sample tube.

FEMM4.2 file for stelter array:
Attachment: STELTER.FEM (4kB)
This file has been downloaded 937 times

aliced25 - 17-10-2010 at 06:14

Ok, for the frequency synthesis side of things, here is something very interesting - the use of 2 PLL's in the one IC (here) and another paper on the use of 2 PLL's and DDS (here). Now both claim serious reductions in phase/spurious noise, while having read both there is an obvious road to reducing complexity by adding in just a high speed DAC (instead of the DDS chip) between the 2 PLL sections of the one IC. The filters that are recommended aren't negotiable, although I suspect we could work around the need for the mixer (from Paper 1 - NB It isn't used in Paper 2). If the DAC were simply spitting out a monotonic output (of pretty much any frequency, wouldn't that translate to a monotone at the mixed frequency (ie. mixed via the PLL)?, which after filtration should give a nice sine wave?

I'm also wondering, instead of fucking around with multiple filters, why aren't SAW filters used in this regard? They appear to have a smaller bandwidth, so I'm obviously in the dark on something...

Frequency locked to B field

arsphenamine - 17-10-2010 at 10:35

&Delta;B/&Delta;T is important.

NdBFe temperature coefficient is -0.0011% T/°C,
or 11 gauss per degree for a 1T magnet which
corresponds to a -47 kHz shift in the Larmor Frequency.

Back when NMR electromagnets weighed a ton, it was easier to adjust the field directly. Varian described this in their "field-lock" patent 3,109,138 in which they determined the frequency drift from a built-in water sample and adjusted the field accordingly.

Given today's DDS oscillators, it is easier to frob the frequency knowing
that pure water's chemical shift is 4.8 ppm.

Modern NMR devices use a deuterium reference and call it 0 ppm.

Varian "Gyromagnetic Methods and Apparatus" patent
Attachment: 3109138__VARIAN_.pdf (433kB)
This file has been downloaded 735 times

[Edited on 17-10-2010 by arsphenamine]

Yet Another Halbach Array Simulation

arsphenamine - 17-10-2010 at 13:34

This is a field plot from the .25" center of aliced25's Halbach array, whose FEM file is attached at the end.
There were changes:
As you can see, the field is 10501.6 gauss +/- 0.3 gauss,
a range similar to the earth magfield strength magnitude.
In a 1T field, a +/- 0.3 gauss range is equivalent to a 2.5kHz shift.

This image was generated by gnuplot since Femm42's graph
generator won't resolve y-axis legends finer than 1 Gauss.



Femm42 base file Attachment: halbach02.fem (4kB)
This file has been downloaded 871 times

12AX7 - 17-10-2010 at 22:05

That's pretty good for a simulation. Now you just need something to cancel the hump, maybe put a current through it (once you've got 1T or so to begin with, a few mT here and there is only a few mA away) so it can be varied / tuned.

Tim

aliced25 - 17-10-2010 at 23:08



In terms of frequency synthesis, the programmable CDCE-706 3-PLL Multiplier/Divider used in the Diophantine paper looked good, then I found the use of all 3 PLL's in a Cascaded Diophantine setup, where the mixer generates a complete frequency with the addition of the 3rd PLL input. Now that is effectively modulation is it not? We could run the DAC in through the 3rd PLL and that could take it right up to the necessary excitation/pll frequency? Does the DAC have to be outputting the 43MHz all by itself or can it be set on a carrier band of, say, 43MHz? That would mean that fairly low speed (ie. Audio) DAC's could be used, if the carrier wave was left on during the collection of the resonance signal, wouldn't that help to cancel out a lot of the echo/noise? This is what it says on the 3-PLL sheet:
Quote:

To achieve an independent output frequency the reference divider M and the feedback divider N for each PLL can be set to values from 1 up to 511 for the M-Divider and from 1 up to 4095 for the N-Divider. The PLL-VCO (voltage controlled oscillator) frequency than is routed to the free programmable output switching matrix to any of the six outputs.


Now I cannot understand why there are products from 3 different manufacturers on the board with the Diophantine synthesizer, I'd be of a mind to utilize as few suppliers as possible. That said, the MAX-2306 IQ Demodulator would enable us to remove the carrier wave and access the signal.

There is also a 2-PLL version of the CDCE-706, being the CDCE-925, which we could presumably use for the direct frequency of the carrier wave (per the Diophantine model) and add the signal to be carried at the mixer (10kHz sine wave @ several watts - ain't hard with audio speakers, you can even get Digital Input Class D Analog Output). (Dumbarse - I put 100kHz & that ain't audio).

PS The use of a PLL to clean up after DDS ain't exactly novel (It is in products like here, while Analog Devices discusses the use of upconversion of a DDS Signal by multiplication thereof to the UHF band.

In terms of this project, that would work too - low speed audio DACs at full-throttle, upconverted to 0.43 or 4.3GHz, then downconverted through the dual PLL (/10 or /100), multiplication adds noise, division reduces it, so a SAW filter (several MHz bandwidth - Oscilent has claims on +/- 2MHz bandwidth on the 430MHz range) between the two would be good. Ideally if the bandwidth was narrowed to several MHz prior to downconversion (and everything is in phase due to the PLL), it should be possible to narrow the bandwidth of the downconverted signal significantly (+/- 200kHz?, so what 0.4MHz wide?).

@ 12AX7 & arsphenamine

You can cancel the hump out by turning the 4 circular magnets, however it will only go so far. That said the circular magnets take up the manufacturing tolerances (in practice) as well (I've been quoted on a shitload of magnets and most want 5% tolerances). Be aware that as you turn each magnet it alters the field around the others (as do any shims).

PS If anyone has knowledge of a working torrent for FEMLAB/COMSOL I'd be willing to view it (not that I'd misuse it, promise:P). I'd be interested to model the whole thing in 3D

[Edited on 18-10-2010 by aliced25]

[Edited on 18-10-2010 by aliced25]

arsphenamine - 18-10-2010 at 08:13

Quote: Originally posted by 12AX7  
That's pretty good for a simulation. Now you just need something to cancel the hump, maybe put a current through it (once you've got 1T or so to begin with, a few mT here and there is only a few mA away) so it can be varied / tuned.Tim
Thanks.

Note that the simulation includes some realistic air space since
any one smooth machined surface is usually good within 1 mil flatness
irrespective of its major dimensional (in)accuracies.

Estimations for a simplified shim coil for +/ 2.5 Gauss correction
revealed that they generate a little heat that must be accounted for.

For a 1" diameter coil of 100 turns of #42 in 10 layers at 10 turns per layer,
generating 2.5 Gauss incurs a self-heating of 5 &deg; C per 2.75 hours.
(specs: 43 ohms, at 2ma, 87mv, a copper volume of 0.5cc)

The magnet array conducts most of this and reaches an undetermined
steady state temperature from air convection cooling. Its field degrades
0.11%/&deg;C, some 11 Gauss/&deg;C for the 1T Halbach, or -47 kHz/&deg;C.

The shim coils may be better off in aluminum formers instead of on Kapton.

A prudent course may be to treat the inhomogeneity sepArately from bulk
field drift due to temperature changes.

Of course, these crappy little heating effects went away when NMR
technology shifted to superconductors.

arsphenamine - 18-10-2010 at 10:14

Quote: Originally posted by aliced25  
You can cancel the hump out by turning the 4 circular magnets, however it will only go so far. That said the circular magnets take up the manufacturing tolerances (in practice) as well (I've been quoted on a shitload of magnets and most want 5% tolerances). Be aware that as you turn each magnet it alters the field around the others (as do any shims).
1.Your Halbach design is amenable to a few 1/4'dia by 1/32" thick button magnets.
FEMM results suggest +/- 0.1 Gauss at the sweet spot.

2. Where the fsck are you spec'ing magnets?!
I hope that 5% refers to remanence/coercivity and not physical dimensions.

Note that the last two sig figs on the B axis are fractional Gauss.



Perhaps adding a 1/8" wide 5 mil steel strip at the magnet center might
flatten it usefully.

Stelter arrays are capable of stronger fields but with order of magnitude
worse homogeneity.

Tweaked FEM file: Attachment: halbach03.fem (5kB)
This file has been downloaded 911 times

[Edited on 18-10-2010 by arsphenamine]

not_important - 18-10-2010 at 12:33

I find myself amused by your concept of simplifying - replacing a DDS chip with one or more PLLs, plus a divider, buffer, and low pass filter per PLL, and a mixer. Even with the dividers in the package, it's still more parts. Your trying to use a DAC leads to even more parts, as you need either a really high speed clock and a lot of memory, and/or a finely tunable clock, to get the control you want.

On top of that most DDS chips have 20 to 50 bit long counters, while those TI chips have 12 and 9 bit dividers. This means more messing about when you want fairly wide range and high resolution when you use the DFS method

BTW - I find it interesting that this is being treated as new, when I was taught about PLLs some 45 years ago this combination approach was one way given to generate a range of frequencies.



For a homebrew rig it's likely you'll want much wider tuning range and something of a 'zoom out' function for use during development, than is needed for actual NMR spectroscopy. I suspect that actual field strength and homogeneity will be unknown until measured, thermal and mechanical drift will have to be learned on the fly. Once you've determined those, then you can go to really narrow band filters. I'd refer you to an image I posted re theory & practice, but I've no idea what thread it was in.

Twospoons - 18-10-2010 at 14:07

Quote: Originally posted by aliced25  


Now I cannot understand why there are products from 3 different manufacturers on the board with the Diophantine synthesizer, I'd be of a mind to utilize as few suppliers as possible.



I cannot understand why you seem to be fixated on using parts from one manufacturer only! There is no benefit in doing this. In my 20 odd years of engineering I've always used whatever was right for the job, regardless of who the supplier was. If there were multiple sources for the same part (equivalents) then so much the better.
Single source parts can easily become a manufacturer's nightmare.

12AX7 - 18-10-2010 at 16:51

Quote: Originally posted by Twospoons  

I cannot understand why you seem to be fixated on using parts from one manufacturer only! There is no benefit in doing this. In my 20 odd years of engineering I've always used whatever was right for the job, regardless of who the supplier was. If there were multiple sources for the same part (equivalents) then so much the better.
Single source parts can easily become a manufacturer's nightmare.


And so many of them from Maxim, no less. Half the engineers I know have had to design out Maxim parts that are no longer in production, or which have dropped out of stock. Maxim is great on samples, yes... place an order for a thousand, and you'll be waiting a very long time.

Fortunately, the most common chips, like MAX232, are multiple sourced and available by the shovelful.

Tim

arsphenamine - 18-10-2010 at 17:08

Quote: Originally posted by 12AX7  
Quote: Originally posted by Twospoons  

I cannot understand why you seem to be fixated on using parts from one manufacturer only! There is no benefit in doing this.


And so many of them from Maxim, no less. Half the engineers I know have had to design out Maxim parts that are no longer in production


Y'know, you probably won't listen to this either, but sepArating the circuitry into functional modules would allow for facile revisions as obligated by hardware (un)availability.

Oscillator, transmitter, demux, converter, CPU -- each gets their own PCB.

Like, duh, how much exposition does the value of modular design require?

Twospoons - 18-10-2010 at 18:37

Quote: Originally posted by aliced25  


PS If anyone has knowledge of a working torrent for FEMLAB/COMSOL I'd be willing to view it (not that I'd misuse it, promise:P). I'd be interested to model the whole thing in 3D

[Edited on 18-10-2010 by aliced25]


3d modelling will not just be interesting, it will be essential! End effects will be very significant, given the sensitivity of the system to any inhomogenity in the field. Unless you simply make the magnet array 1 metre deep!

aliced25 - 18-10-2010 at 19:51

@ twospoons,

I'm trying my best to get hold of a 3D modeling solution, when I get it I'll use it. As to the wanting to minimize the number of components from various manufacturers, it is a personal preference, where there are multiple products with similar specifications, I'll always choose one that I know will work with the board (ie. from the same manufacturer). It ain't that hard to work out why, surely?

@ not_important

With the DDS chips being unfiltered and having multiple harmonic images, there is little reason to choose them over a multiple PLL design using a close-tolerance crystal running a DAC, multiply the original sine-wave to about 10-20 times the necessary wavelength (PLL 1), run it through a close-ish tolerance SAW filter (+/- 2MHz) and back through the IC (PLL 2) to divide by 10/20. That will give a much narrower bandwidth than the DDS chips are capable of (0.4-0.2MHz bandwidth). The memory for the DAC can be onboard with some DAC's or can be sent via SPI (and kept on the processor chip). Having just come out of a filtered PLL (ie. PLL 2), the synthesized frequency should be in phase and pretty well noise free. A line-driver into that to pump up the juice and bam.

Twospoons - 18-10-2010 at 20:24

Years ago I had access to 3d FEA (Ansys ? I think it was). Had the nice feature of starting from a coarse grid, then progressively refining the grid only in those areas that needed it. Took some serious cpu horsepower though - would take all day to run one analysis. Interesting it was a Halbach array I was analysing (to find leakage induced eddycurrent losses in the motor housing). Don't have the software anymore, sorry.

As for using parts from one mfgr - thats no guarantee that things will work, or that you'll end up with the best solution. Every manufacturer has their strengths and weaknesses in their product portfolio.

12AX7 - 19-10-2010 at 08:12

Quote: Originally posted by arsphenamine  

Y'know, you probably won't listen to this either, but seperating the circuitry into functional modules would allow for facile revisions as obligated by hardware (un)availability.

Oscillator, transmitter, demux, converter, CPU -- each gets their own PCB.

Like, duh, how much exposition does the value of modular design require?


Yuck, that's a retarded idea. Just shove signal quality out the door, why don't you?

Tim

not_important - 19-10-2010 at 08:56

Quote: Originally posted by aliced25  
...

With the DDS chips being unfiltered and having multiple harmonic images, there is little reason to choose them over a multiple PLL design using a close-tolerance crystal running a DAC, multiply the original sine-wave to about 10-20 times the necessary wavelength (PLL 1), run it through a close-ish tolerance SAW filter (+/- 2MHz) and back through the IC (PLL 2) to divide by 10/20. That will give a much narrower bandwidth than the DDS chips are capable of (0.4-0.2MHz bandwidth). The memory for the DAC can be onboard with some DAC's or can be sent via SPI (and kept on the processor chip). Having just come out of a filtered PLL (ie. PLL 2), the synthesized frequency should be in phase and pretty well noise free. A line-driver into that to pump up the juice and bam.



Uh, you need to study how DDS circuits work a bit more, you seem to have some misconceptions on their capabilities; plus really try to design that concept of yours.


Anything in this image look familiar?




DDS.png - 6kB

aliced25 - 19-10-2010 at 23:48

@ not_important

Have a look at that paper where the OUTPUT of the DDS chip was run back in through a second PLL to get everything back in phase. Now with those PLL's being dual fractional/integer-N type (each), then the DAC could be used with a quality crystal to generate a sine wave at a decent rate, say 10MHz, well within the range of mid-rate DAC's (which use RAM/ROM as the sine table - which is merely a recorded set of integers from full high to full low and back). As Analog Devices are using their own DDS chips for UHF frequencies by running them through a multiplier/integer-N PLL (and if that works, a second fractional-N PLL would also work), multiply the frequency by say 43, ie. 10MHz * 43 = 430MHz, through a 4MHz wide SAW Filter, then back through a fractional-n PLL to divide the signal by 10 (while the phase filter takes out/adjusts the crud), that should "hypothetically" give a 0.4MHz wide sine wave @ 43MHz (with everything in phase).

The SAW filter on the output of the first stage of the PLL should get rid of the kinks from the DAC every bit as well as a low-pass filter, while the Ref. Clock is the decent quartz crystal (actually 12MHz is needed for USB, so it is already on the board - something to consider). The Phase Accumulator/Sine Table are one and the same and are taken care of via scripting the DAC (actually some use just 1/4 of the table and reverse/invert the numbers to suit).

@ Twospoons

I'm trying my best to get hold of a 3D modeling program, every torrent I've used so far is fucked, while Radia works after a fashion, but it needs some funky ass scripting & Mathematica (the torrent for it works, or so I've been told) & I'm clueless on how to get it working at present.

@12AX7,

Using separate boards for ALL of the components seems excessive, especially when the data has to be sent from one to another inside a box that is going to be awfully close to a very strong magnet and a shitload of noise. I'm reading up on various grounding schemes at present, the noisiest part of the whole affair (on the analog side especially, exactly where it is NOT needed) is going to be powering the components, as each board would still need an LDO Regulator, I cannot see the sense in multiplying the noise. If anyone has a scheme whereby THAT can be filtered out on a separate board and supplied as and where needed with minimal fuss, that I'd like to see (notice they don't separate boards on phones? Analog, Digital, Frequency Synthesis, power supply, battery charging, etc. all on one board)...

PS Can anyone tell me why in god's name people are using 0.25mm traces (0.034mm) instead of 0.25mm Dia Wire? In terms of resistance the wire is about 1/10 of the trace, then again, the pipe is wider isn't it? Anyway, this is good for ideas.

watson.fawkes - 20-10-2010 at 06:24

Quote: Originally posted by aliced25  
as each board would still need an LDO Regulator
Low dropout is more-or-less irrelevant to this application, because low-dropout doesn't mean the same as low-noise. Some LDO devices have low-noise characteristics, but not all do. LDO is principally an energy-efficiency technique, and it's principally useful for battery-operated devices. The other thing to remember is that low-noise for a consumer device, such as a cell phone, may not be considered low-noise in an instrumentation context.

The relevant concepts here are twin. For the power supply, you want high load rejection, which is the sensitivity of the output to changing load. On the powered-device side, you want a high power supply rejection (PSR), which is the sensitivity of the device behavior to changes in power supply. Note: the power supply itself has a PSR, often called line rejection, for example, the ability to reject spikes in the mains. A key point about consumer-grade chips is that they rarely need to worry about load rejection, particularly for digital loads.

This article has some example circuits and commentary about how to cut down regulator noise: http://www.wenzel.com/documents/finesse.html. Note the use of a pass resistor to cut down noise; this is absolutely not a lower-power technique.

Pretty much every accurate power supply circuit I've examined in any depth has two common features: a voltage reference and an error amplifier. To a first approximation, you can get a sense of the quality of a power supply by looking at the number and quality of the error amplifiers. For most applications a single error amplifier suffices. For some, though, you want cascaded ones.

I feel the need to stress that instrumentation electronics is significantly more than basic electronics, more of everything, particularly analysis. There's a whole chapter about it in The Art of Electronics, a book I recommend wholeheartedly. The authors, both at Harvard, have long experience designing instrumentation circuitry for physics and astronomy research.

arsphenamine - 20-10-2010 at 09:51

A discussion of several low noise DIY regulators is here

Collected circuit schematics are here

A recurrent theme is pre-regulation (LM317 or similar) followed by a low
noise wideband error amp. The op amp choice is important; the AD797
was ultimately rejected as too sensitive for high EMI environments and
was replaced by the AD817 or AD825.

Walt Jung, a primary reference for linear power supply regulation,
has collected some of his DIY regulator articles here

arsphenamine - 20-10-2010 at 10:23

Quote: Originally posted by 12AX7  

Yuck, that's a retarded idea. Just shove signal quality out the door, why don't you?
It's easier to make a few small widgets and then a large widget, than to make a large widget.

Modular design means each module can be tested and troubleshot separately. (Testing? WTF is that?)
If you have finite resources and want to complete a prototype more quickly,
it is good sense to make it work first and optimize it second.

The history of premature optimization is heavily littered with mistakes.

My favorite one is where testing couldn't
save an application from optimization.
Someone profiled live code to find hot spots,
found something executing 10k/sec.
The code block got hand-optimized.
Now executing at 50k/sec, it still had no effect on the system performance.
Wiser heads noted that the code under examination was the system idle loop.

ps, my chromosome 21 is just fine.

not_important - 20-10-2010 at 10:49

Quote: Originally posted by aliced25  

Have a look at that paper where the OUTPUT of the DDS chip was run back in through a second PLL to get everything back in phase.


No. A PLL contains a controlled oscillator, limited by a LP filter in the rate of change in frequency and phase. The PLL locks to the signal applied, in the case you refer to that's the output of the DSS. The PLL is reducing jitter/short term instabilities, the same is done for xtal oscillators - a following PLL gives short term stability controlled by its LP filter while the crystal supplies long term stability.
Quote:
Now with those PLL's being dual fractional/integer-N type (each), then the DAC could be used with a quality crystal to generate a sine wave at a decent rate, say 10MHz, well within the range of mid-rate DAC's (which use RAM/ROM as the sine table - which is merely a recorded set of integers from full high to full low and back).


Note that a DDS generally contains a sine wave lookup table, or similar source of sine data, and a DAC converter - the same as you want to cobble together out of separate parts. I suspect the DDS will do better as there's no inter-package delay to deal with.

Much of the rest of the DDS is logic to generate the desired frequency - the clocking rate of the DAC in effect. If I understand what you're trying to do, you want to clock the DAC at a constant rate and change the lookup table contents. With a 10 MHZ clock you can only change your waveform cycle in steps of 100 nsec; 2 samples per cycle for 5 MHZ, 3 for 3.33333... MHz, 100 for 100 KHZ, 99 for 101,0101... KHz, 101 for 99,0099099... KHz. Without tweaking the clock you can't hit exactly 99 KHz.
Quote:
As Analog Devices are using their own DDS chips for UHF frequencies by running them through a multiplier/integer-N PLL (and if that works, a second fractional-N PLL would also work), multiply the frequency by say 43, ie. 10MHz * 43 = 430MHz, through a 4MHz wide SAW Filter, then back through a fractional-n PLL to divide the signal by 10 (while the phase filter takes out/adjusts the crud), that should "hypothetically" give a 0.4MHz wide sine wave @ 43MHz (with everything in phase).


A pure sine wave has a width of zero, a "0.4MHz wide sine wave @ 43MHz" is jumping around over 1% of its frequency - rather noisy. Are you trying to say 'tunable over 0,4 MHz' ? If so, then consider using better terminology.
Quote:
The SAW filter on the output of the first stage of the PLL should get rid of the kinks from the DAC every bit as well as a low-pass filter, while the Ref. Clock is the decent quartz crystal (actually 12MHz is needed for USB, so it is already on the board - something to consider). The Phase Accumulator/Sine Table are one and the same and are taken care of via scripting the DAC (actually some use just 1/4 of the table and reverse/invert the numbers to suit).


1) A SAW at that point is overkill, more expensive than simple LP filters and possibly requiring buffer amps for matching.

2) The crystal needs for a USB interface is much less demanding than the reference for a NMR. Remember everyone else has been talking OCXT to meet the demands.

3) How are you proposing to "reverse/invert" the table data? If not done in hardware then the delays at transition points will cause jitter and uncertainty in the frequency.


Quote:
PS Can anyone tell me why in god's name people are using 0.25mm traces (0.034mm) instead of 0.25mm Dia Wire? In terms of resistance the wire is about 1/10 of the trace, then again, the pipe is wider isn't it? Anyway, this is good for ideas.


Because traces are repeatable and automatble, wires aren't. For commercial products those are very important, for research work at that size of connections PCBs can be more convenient as design software can generate the trace layout.


As for voltage regulators - that's been pretty well covered.
Quote:
(notice they don't separate boards on phones? Analog, Digital, Frequency Synthesis, power supply, battery charging, etc. all on one board)...

However mobile phones are intended to slip into a pocket; I doubt you're planning to do so with a NMR, the magnet size along is well beyond that.* Also the designers expend effort and time to make those designs work well, I doubt you have the tools - software and hardware - used to do that job.


* this excludes some of the micro-NMR designs being kicked about. Fabbing those might be a bit of a challenge.



[Edited on 20-10-2010 by not_important]

arsphenamine - 20-10-2010 at 14:33

Quote: Originally posted by aliced25  
You can cancel the hump out by turning the 4 circular magnets, however it will only go so far.

How does one know when the hump is canceled,i.e.,homogeneity improves?

Perhaps a test method needs discussion.

How does one safely rotate and stabilize magnets that exert over 100 lbs attractive force?

These questions give me cause for concern.

aliced25 - 20-10-2010 at 15:48

@ arsphenamine

The Magnets are aligned with great care, the main difficulty lies in determining the resultant field, Teslameters/Magnometers/Reed Switches, etc. are all built with a 1% accuracy, that is way too low for this application - however several groups have tested and adjusted their field by performing <sup>1</sup>H-NMR experiments on water/soluble copper and compared their results to what is expected. As to rotation, there is going to need to be some way of holding the magnet sufficiently tightly so that it remains in the orientation in which it is aligned, and does not self-align - another part of the problem.

@ not_important

This tutorial, MT-031 Grounding Data Converters and Solving the Mystery of "AGND" and "DGND" pretty much explains what we are trying to do, there is a DSP in the center of the two planes, with both DAC's & ADC's plus power supply to deal with. There are numerous Tech. Bulletins on the subject, including the use of an isolator to physically separate the Analog & Digital parts of the board. There are even online books & chapters on the subject.

As to the use of the DAC, running a 100MSPS DAC to output a 10MHz signal (AD App. Note) is not exactly unusual (AD Tutorial, neither is the use of what is essentially playback - a set of values is "played" via the DAC to give the signal output (Basics of DDS). However the images & harmonics that come with that make the digitally approximated sine wave less than narrow-bandwidth (thus my problem with the DDS chips - there are circuit notes where they discuss using these for synthesizing much higher bands than are possible with the first-order). They utilize multiple-component 7-8th Order Filters as Low-Pass filters, whereas the SAW filters have smaller bandwidth passbands, are cheaper and use less components. Running the generated signal through a PLL to increase it to the UHF Band is also well and truly discussed by the AD Team, running that wide bandwidth signal through a SAW Filter to remove the images, spurs, etc. seems logical, running the narrow-bandwidth UHF Signal through a fractional-N PLL to get the smaller band signal also seems intuitive.

As the various DDS Chips (with the associated phase & spur/harmonic noise) have been used in micro-NMR before, a lower cost, narrower bandwidth solution using minimal parts appeals. The use of the DAC generated signal as the input for the PLL is not new, nor is the fractional/integer-n modification of it. The SAW filter is less demanding than that multipart analog filter design & a decent crystal should suffice, given the width of the bandwidth used by others using a DDS chip without serious filtration.

So the only reason why traces are used is the difficulty in reproducing the same wire width? Seems awfully odd, especially given that wire is used in chip manufacture (with ultrasonic welding). I'm just trying to comprehend why it is better to purchase copper foil, a photopolymer, make a negative, expose it, remove the unexposed layer, etch the copper back off the board, then remove the rest of the polymer, tin the copper and then lacquer the board... It hardly seems efficient does it?

[Edited on 20-10-2010 by aliced25]

12AX7 - 20-10-2010 at 18:30

So what is all this DAC BS about?

Has someone forgotten that a squarewave is 90% fundamental?? You put a filter on it. 4th order bandpass will do at least -30dB attenuation of harmonics, more if you add a trap for the 3rd harmonic, and still more if you add a resonator to accentuate the fundamental. What do the harmonics matter, anyway? There aren't harmonic protons. So what if it's wasted power, the harmonics ares down >= 10dB so they hardly affect efficiency.

RF excitation does not need to be very accurate. The spectrum of a 10us RF burst is necessarily spread out by the very act of windowing it. The spectrum looks like a sinc function, with -6dB points at f_o +/- 60kHz, regardless of whether the signal came from NIST or a redneck's car radio!

The ONLY thing that a reference is required for is conversion. The long thing. The FID capture that takes ten seconds. The signal that's actually long enough to observe narrow peaks.

Power supplies are even less of a concern than in cell phones (which is a big concern, BTW, because they pick signals out of the noise floor). With only two or three discrete frequencies of interest, it is ridiculously easy to filter.

This isn't audio, where you're filtering a three decade band, or an oscilloscope, where you're watching DC to light. The preamp, converter and FFT are all completely insensitive to interference outside of the passband. There could be gobs of trash at say, 41MHz, which ends up as a 1MHz image after the converter, which is about 23 times above the sample rate and trivial to filter.

Aliced25: Beats me, but considering that 99.97% of all electronics manufactured today are made of printed circuit boards, I think you'll discover the answer easily enough.

Tim

arsphenamine - 21-10-2010 at 08:35

Quote: Originally posted by aliced25  
The Magnets are aligned with great care...
How? By big chunky guys wielding velvet-covered pipe wrenches?
Quote:
the main difficulty lies in determining the resultant field, Teslameters/Magnometers/Reed Switches,etc. are all built with a 1% accuracy, that is way too low for this application
Accuracy != Precision != Resolution.

My bathroom scale is 5 lbs off, but I'm impressed that it detects
a difference when I add or remove a beer by the usual methods.

For field homogeneity measurements, repeatable resolution is more important than absolute accuracy.

IF you have a Hall sensor (not an IC), you can detect sub-Gauss field differences.
Quote:
- however several groups have tested and adjusted their field by performing 1H-NMR experiments on water/soluble copper and compared their results to what is expected.
What are the expected results? How did they adjust it?

Those indeterminate groups are most likely filling the field bore with a water bottle, in which case, the field is averaged over most of the bore volume.

You infer field inhomogeneity from peak height, broadening and slope asymmetry (and there are patented statistical analytic methods to do this), but the method is spatially too lo-rez to determine the inhomogeneity location.

Spinning an ISO 5mm sample tube in the field sweet spot is more germane but the same arguments apply.

If you can't find it, you can't correct it.

Since this design uses magnets of unknown homogeneity, a direct test method is essential.
You gotta poke a tiny sensor in the magnet bore.

So you may wonder, "Um. Now how do I position a Hall sensor, and with what resolution?"

An XY table or jig will do.

Your required resolution is the sensor die size, about 0.3mm square, about ~ .012 inch in each axis.

It's hard to find a commercial XY table that crappy but it's easy to build one.

Instructables has many variations on the XY position methods. By the Principle of Maximum Laziness, I favor the ones that use page scanners -- gut the electronics and put in two cheap stepper controllers.
Quote:
As to rotation, there is going to need to be some way of holding the magnet sufficiently tightly...
Talk to a machinist. Get a shop tour if you can, because machinists have some of the coolest toys in the world.

not_important - 21-10-2010 at 15:28

Well, according to Analog Device's "A Technical Tutorial on Digital Signal Synthesis"
Quote:
In its simplest form, a direct digital synthesizer can be implemented from a precision reference clock, an address counter, a programmable read only memory (PROM), and a D/A converter.

In this case, the digital amplitude information that corresponds to a complete cycle of a sinewave is stored in the PROM. The PROM is therefore functioning as a sine lookup table. The address counter steps through and accesses each of the PROM’s memory locations and the contents (the equivalent sine amplitude words) are presented to a high-speed D/A converter. The D/A converter generates an analog sinewave in response to the digital input words from the PROM. The output frequency of this DDS implementation is dependent on 1.) the frequency of the reference clock, and 2.) the sinewave step size that is programmed into the PROM. While the analog output fidelity, jitter, and AC performance of this simplistic architecture can be quite good, it lacks tuning flexibility.
...
With the introduction of a phase accumulator function into the digital signal chain, this architecture becomes a numerically-controlled oscillator which is the core of a highly-flexible DDS device.


So what you've been proposing to build from parts is a simple DDS, save that you are not using a PROM to hold the waveform. It will have similar problems to a conventional DDS chip unless you use a wide and deep memory. The outboarded PPLs used to generate the clock(s) and clean up the output would do the same for an integrated DDS chip.


You might find it useful to do a perusal of this http://www.xs4all.nl/~martein/pa3ake/hmode/index.html


Twospoons - 21-10-2010 at 21:50

Quote: Originally posted by aliced25  

So the only reason why traces are used is the difficulty in reproducing the same wire width? Seems awfully odd, especially given that wire is used in chip manufacture (with ultrasonic welding). I'm just trying to comprehend why it is better to purchase copper foil, a photopolymer, make a negative, expose it, remove the unexposed layer, etch the copper back off the board, then remove the rest of the polymer, tin the copper and then lacquer the board... It hardly seems efficient does it?

[Edited on 20-10-2010 by aliced25]


Two words: mass production.

There was a time when wirewrap was the way to build electronics, and to be fair it has a couple of advantages: lower crosstalk, reliability. But it is slow to build, and bulky, and silver plated wire and connectors aren't exactly cheap.

PCBs give you repeatability, a mounting surface, fast assembly, variable trace widths, controlled impedance, delay matching, heatsinking ....
And its cheap because everyone does it.
 Pages:  1  2