Elegant NMR
Register a new account · Login to your account Welcome, Guest

Electronic Datasheet and Parameters, as for June 2017 RSS Feed available

Top   #1

ilgis

**
Posts: 19
Joined: August 30, 2016
Group: Administrators

June 5, 2017 @ 8:22 PM
Hi,

let me share current technical parameters of electronics inside the spectrometer.

We have:

1. two computational units,
1.1. one based on Cyclone 5 has 132 SIMD cores running at 160MHz with one MADD per core, i.e. ca. 40GFlop/s real performance,
1.2. dual Intel Atom 500MHz with 6GFlop/s peak performance that achieve ca. 1GFlop/s (we do not need more), with 4GB/90MB/s NAND and 1GB RAM,

2. clock with 90fs jitter and 2ppb stability on well conditioned temperature range (don\'t forget about liquid cooling!!!),
3. dual 16bit ADC running at 160MHz with 40dB preamplifier, so, we can really see below 1mkV fluctuations on receiving coils.
4. two receiving and four transmitting coils as it is shown at the figure 5 at https://arxiv.org/abs/1706.00237 paper.

Finally the spectrometer cannot operate from standard USB, because it consumes ca 27Watt, so, we are modifying interconnect to 15V/3A + USB.

By plugging to USB, the spectrometer is recognized as USB disk with GUI software and free space for the spectra. You can connect the device to USB without external power supply to work with the setup and read collected spectra, but for the proper operation you need to power it with 15V/3A PSU.

Sincerely,

Ilgis
Reply

Top   #2

James_Keeler

*
Posts: 1
Joined: August 15, 2017
Group: Members

August 15, 2017 @ 4:03 AM
Hi Ilgis,
I am a rf enthusiast and would like some more information on your electronics system so I can understand it better. How do you generate your rf signals and amplify them, and what amplifier design do you use? Do you have a transmit receive switch? I see you use a fast ADC to digitize signal, do you amplify the nmr signal before and also filter? What rf parts do you use there? I don\\\\\\\'t see any coax cables in the pictures you took, even to the coil. How could this possibly work?

Also, can you elaborate on magnet design? Seems like you mentioned 1.2 T on the website but says this corresponds to 42 MHz - it\\\\\\\'s actually 51 MHz for 1H. What kind of magnets and arrangement do you use. I understand you do something with parallel transmitting to compensate for the inhomogeneous field, but I\\\\\\\'m confused as to how this works. I didn\\\\\\\'t find an explanation in the papers you cited. How can you do NMR without having a tuned coil? I see the figure in your paper of the key idea, but for me, the nmr coil was always tuned and matched to a frequency so how do you make-do without it?

Finally, it seems like your cost of electronic components is very high to sell your product at $1000. I doubt very much you can make any money on what you said you are including, and would anticipate you losing money on each sold stick. Comments?

Thanks,

James


QUOTE: ilgis, June 5, 2017 @ 8:22 PM (view in context)
Hi,

let me share current technical parameters of electronics inside the spectrometer.

We have:

1. two computational units,
1.1. one based on Cyclone 5 has 132 SIMD cores running at 160MHz with one MADD per core, i.e. ca. 40GFlop/s real performance,
1.2. dual Intel Atom 500MHz with 6GFlop/s peak performance that achieve ca. 1GFlop/s (we do not need more), with 4GB/90MB/s NAND and 1GB RAM,

2. clock with 90fs jitter and 2ppb stability on well conditioned temperature range (don\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\'t forget about liquid cooling!!!),
3. dual 16bit ADC running at 160MHz with 40dB preamplifier, so, we can really see below 1mkV fluctuations on receiving coils.
4. two receiving and four transmitting coils as it is shown at the figure 5 at https://arxiv.org/abs/1706.00237 paper.

Finally the spectrometer cannot operate from standard USB, because it consumes ca 27Watt, so, we are modifying interconnect to 15V/3A + USB.

By plugging to USB, the spectrometer is recognized as USB disk with GUI software and free space for the spectra. You can connect the device to USB without external power supply to work with the setup and read collected spectra, but for the proper operation you need to power it with 15V/3A PSU.

Sincerely,

Ilgis
Reply

Top   #3

ilgis

**
Posts: 19
Joined: August 30, 2016
Group: Administrators

June 6, 2018 @ 12:34 AM
Hi James,

thank you for your kind reply and please, very-very excuse me for such a long delay with the answer, I mainly was active in ResearchGate and did not visited this forum for a long time.

QUOTE: James_Keeler, August 15, 2017 @ 4:03 AM (view in context)

How do you generate your rf signals and amplify them, and what amplifier design do you use?

we have 4 transmitting coils, directly connected to 4 full-bridges. It leads to larger excitation sequences, but it saves component costs.

QUOTE: James_Keeler, August 15, 2017 @ 4:03 AM (view in context)

Do you have a transmit receive switch?

Yes, incorporated in amplifier and controlled numerically. One bug in a numerical algorithm lead to complete burnout of complete analog part of the device.

QUOTE: James_Keeler, August 15, 2017 @ 4:03 AM (view in context)

I see you use a fast ADC to digitize signal, do you amplify the nmr signal before and also filter?

Amplified with 40dB, with no filter - just as it is. Filter is prohibited, otherwise it is impossible to use our numerical method.

QUOTE: James_Keeler, August 15, 2017 @ 4:03 AM (view in context)

What rf parts do you use there?

16Bits 160MHz 2 channel ADC.

QUOTE: James_Keeler, August 15, 2017 @ 4:03 AM (view in context)

I don\'t see any coax cables in the pictures you took, even to the coil. How could this possibly work?

Sure, it is one of the key advantage. We have coil situated only in 3.5cm far from an amplifier, so, roughly we can work without coax cables up to 120MHz. It saves a lot of space and component costs.

Some photos of our PCBs are published in ResearchGate.

QUOTE: James_Keeler, August 15, 2017 @ 4:03 AM (view in context)

Also, can you elaborate on magnet design? What kind of magnets and arrangement do you use.

It is patent pending magnetic material (Mn-Bi + Co-Fe + In-Bi) and patent pending design (casted magnet on irregular magnetic field that forms spherical Halbach structure).

QUOTE: James_Keeler, August 15, 2017 @ 4:03 AM (view in context)

Seems like you mentioned 1.2 T on the website but says this corresponds to 42 MHz - it\\\\\\\'s actually 51 MHz for 1H.

sorry, we are playing now with Halbach spheres, and sometimes achieving 3.5T (with said magnetic material and design), but I am tending to more pessimistic scenario to have 1.2-1.8T on a manufacturing.

QUOTE: James_Keeler, August 15, 2017 @ 4:03 AM (view in context)

I understand you do something with parallel transmitting to compensate for the inhomogeneous field, but I\'m confused as to how this works.

Key idea is to measure two and more NMR responses in the same time, and make numerics on them. It is easy to prove that we can compute magnetic field and oscillator deviation and apply compensation. We have oscillator that cost only $30, you may imagine how worse signal can be without such compensation.

We need several coils to find best point for measurement in case if magnetic field is not too smooth spatially.

Second reason (hope to succeed on mass production manufacturing on it also) is to have two different poles of magnets so that the same isotope (mainly 1H) deliver two responses. If the responses are scaled, they are very close to each other. Since spectra for different excitations differ only in the J-coupling parts, it is clear that these spectra build a low rank object that may be used on another compensation method.

The third and simplest compensation method is to incorporate marker isotopes into the walls of measuring camera. We are using Eu and Ag isotopes right now and it works fine.

The main disadvantage of these compensation methods are high numerical costs, we need ca. 3*10^9 computations of sin/cos per second + ca 2*10^10 multiplications and additions per second that leads to multicore parallel computing (i.e. need to have portable supercomputer in the stick).

QUOTE: James_Keeler, August 15, 2017 @ 4:03 AM (view in context)

How can you do NMR without having a tuned coil? I see the figure in your paper of the key idea, but for me, the nmr coil was always tuned and matched to a frequency so how do you make-do without it?

yes, clear that it is difficult to work, but the signal can be detected at 0.3uV LSB taking into account that we are using 40dB pre-amplified 16Bits ADC. We need the signal as it is, without tuning to predetermined frequency.

QUOTE: James_Keeler, August 15, 2017 @ 4:03 AM (view in context)

Finally, it seems like your cost of electronic components is very high to sell your product at $1000. I doubt very much you can make any money on what you said you are including, and would anticipate you losing money on each sold stick. Comments?

You are almost right, but let me try, may be with mass production we can make money on it.
Reply

Tags:

Quick Reply Add Reply New Topic

Users viewing this topic

0 members and 1 guests; 1 in total