Low noise cable - SMA coax vs twisted pairs

[ricko_uk]

Hi,
I have a very tiny and noise sensitive magnetic sensor that needs to be connected to an INA amplifier PCB and I need to keep the noise as low as possible.

To keep noise picked up by the cable as low as possible, I am not sure if the best option is an SMA coax cable or a twisted pair with shielding braid (grounded on the INA side).

Any suggestions on pros and cons of each?

Thank you

[Freesurfer]

Remember that a coax and a twisted pair cable use two very different approaches to achieve the same thing, and which one you end up with, depends on your circuit that is going to receive and amplify your signals.

Twisted pair is designed in such a way that any noise it picks up will be identical on each wire. The twisting ensures that each wire picks up identical noise, so that when you measure the difference between the wires, the noise on each cancels/zeroes each other out, thereby giving you a clean signal..

But one thing is important to make this work correctly: You need to make sure that each wire is treated identical, floating or differential, as in you cannot connect one wire to a ground plane and leave the other floating to amplify the signal. Doing so will give each wire a different impedance in such a way that the grounded wire may pick up less noise than the floating wire, thereby defeating the purpose of a twisted pair. Some designs use a little transformer just before the amplification stage to ensure that the 2 wires are indeed treated equal.

A Coax is different, it is designed to connect the shield to a ground plane with the idea that it will shield out any noise from the center conductor. It may be an easier design, but isn't perfect.

[ricko_uk]

Thank you Freesurfer

[KT88]

Twisted pair offers more opportunities to mitigate disturbances. If you have a differential signal source like your coil, every external influence appears as a common-mode signal that can be cancelled out (to certain extent).
As always there are some imperfections in the approach... Most of it comes from subtle differences in impedance against shield and ambient. Fotunately the effect of these imperfections can be reduced by a more or less sophisticated compensation networks and common-mode fliters.
The other challenge is the limited CMRR over frequency of the amplifiers. To mitigate these limitations you could isolate the cable from the amplifier with a transformer and/or use an amplifier that is fast enough (CMRR). In a similar application the ADA4897 worked very well as it is also very low (voltage-) noise. Current noise is less of a concern for that kind of use case.
I used it in a symmetrical arrangent (like the first stage in an in-amp) all the way into the ADC (18-Bit, 2MSPS). Of course, the ADC required a dedicated differential AMP to drive it...

Cheers
Andreas

[ricko_uk]

Thank you Andreas.

Did you use the ADA4897 in front of an instrumentation amplifier or instead of the instrumentation amplifier?

Also using a transformer, it would work only above certain frequencies, correct? How do I see the operating frequency band and how do I choose a transformer to put in front of an instrumentation amplifier?

Thank you

[KT88]

instead of an integrated one. I'm not aware of any integrated in-amp with the desired specs. You can also stay differential for the following gain stage(s)... The first stage shouldn't have too high gain to awoid too much bandwidth reduction. Note that you get about 10x lower input referred noise contribution from the next stage at a gain 3 already (root sum square).

[Vovk_Z]

So I'll try to conclude: there isn't any 'the best' option between a coax and a twisted pair. There is only a dependence if it is a balanced input or not. Otherwise, it is better to test both in your design.
As for me, I prefer to use the option that is most convenient for me: the most convenient wire for 'inside the case use' is a self-made twisted pair. And the most convenient for interconnecting (outside) one-signal cable is a coax.

[ricko_uk]

Thank you KT88 and Vovk_Z

[ElectronRob]

Just to add to the mix,

I have also used two coax cables from sensor to instrument amp before, each core is the input to each buffer, the outer braid is connected at one end (usually sensor end) but could likely also be grounded. The two coax cables are run together (side by side) with occasional heatshrink to keep them in place. You can decide whether to use the braid as shielding or for lower frequencies you can use the shield to inject the same noise into each core.

It goes without saying that your goal isn't just to attenuate or shield from noise but to make every effort to achieve the same noise (amp and phase) on each input to the INA. Shielded twisted pair is great, but I've struggled a little with size and connectors, with two coax you can terminate with MCX while fully shielded connectors to terminate STP aren't as wide spread (or small), of course if your using a large connector or just soldering I'd likely start with STP.

Good luck
 

[tchiwam]

I am currently trying to design a power source for sensitive AD and amplifiers... Just trowing some random stuff together, maybe you would like the 4xLT3045 config:

My draft is here: https://github.com/tchiwam/LNA/tree/main/Kicad

[jonpaul]

Well, First, decide if CM or DM noise is the primary  issue. Then the freq band of the noise you need to reduce. EG mains 50/60 Hz, audio, RF, UHF, microwave?

CM noise: Induced by flux  common to both conductors, noise  current flows equally on BOTH conductors.

Use shielded TP, and balanced connectors, diff amp type RX or transformer isolation with shield.

DM (normal mode) noise: Flows as the signal current in a loop, and hard to separate from the signal current.

Use Coax or triax, cable and connectors.

One can design baluns, beads and chokes to reduce either CM, DM or both at higher freq.

Suggest to see the fine books by Ott, etc, on noise reduction.



Jon

[David Hess]

The lowest noise common option is usually shielded twisted pair with a floating sensor and singled ended operation, which means no instrumentation amplifier or differential input.  The reason for this is that differential inputs are always noisier than single ended inputs.  This assumes that the sensor is floating although there are ways to use a single ended signal chain with bridge sensors by using common mode suppression.

Usually differential is preferred to suppress common mode noise but if you can arrange for there to be no common mode noise, then singled ended is better.

[sam[PS]]

If really ultra low noise is required i would suggest to go active = to place the first amplification stage on the sensor side so the signal that goes along the wire is higher level and/or lower source impedance.

So for very low noise with cheap and easy sourced cabling i would use a piece of cat6/cat7 cable that cost close to nothing (actually nothing for short piece as there is so many opportunity to grab some leftover for free/a bear and a smile). Use one pair for power supply, one for the signal, you got two spare.
Plus for super cheap you can have a guaranteed matched impedance connector that both lock, is easy to unplug when needed, can't be plugged inverted, ... Yes it's a rj45, your system integrator will thank you.

Now for the amp on the sensor side i would use a single ended amplifier as they are easier to build and most time lower noise, that's one place where discreet transistor/jfet still rules (bjt have lower noise but jfet have higher input impedance so the impedance of your source will dictate the best choice). And  to keep things balanced in the pair just measure (or estimate) the output impedance of your amp and place the closest E24 resistor in series with the return wire and it's pretty much as good as it can as long as you are <1MHz.

For extra low noise i would pay some good attention to the power supply for the amp, at the very least good decoupling on the sensor side, "capacitor multiplier" are very good in this kind of application and only cost you 1 high hfe bjt, and if any RF or switching stuff could ever get around (i guess you have some microcontroler somewhere in the box ?) ferrite bead are unbeatable.

You should end up with 3-4 discreet and 10 passive elements, all easily fit on couple cm² smt board and the total cost is probably less than a meter of decent quality coax cable + sma connector. Plus this solution would still be valid if on a future project you need to get 20m cable from the sensor to the main board.

Just my 2cts...

[ricko_uk]

Thank you all very much. All very useful info!!

[djerickson]

These are all great ideas. I'd like to add my experience designing fA sensitive Photodiode amps. Gain was 1e11 V/A (including a lock-in amp), then into a 24 bit ADC.  'Normal' RG58 coax has 90-95% coverage, and leaked lots of room 60Hz into the input at the pA level. I had to use either double shielding (copper tape over normal coax) or solid metal BNC adapters. Good Triax would have worked. Semi-rigid or rigid coax may have worked, I didn't try them. 

This thing was soooo sensitive. One day I tried using a thermistor just inside the metal box to measure temperature. The thermistor wiring made a nice 60Hz antenna and coupled into the amp. Grounding one thermistor connection to the box fixed it.  Don't let anything float: metal hardware, PCB traces, wires... Even plastic stuff can accumulate charge.

Best of luck,
Dave
www.djerickson.com

[David Hess]

'Normal' RG58 coax has 90-95% coverage, and leaked lots of room 60Hz into the input at the pA level.

...

Good Triax would have worked. Semi-rigid or rigid coax may have worked, I didn't try them.

RG-223 which is essentially doubled shielded RG-58 might have worked.  Double shielded cable is practically a requirement when dealing with a dynamic range greater than 90 dB.

[cncjerry]

 +1 on the triax, assuming cable pickup is the source. The connectors aren't fun to put on.

[ricko_uk]

Thank you both

Do you connect the shield/s on both sides or only one side? Also do you connect to chassis or PCB ground or one to chassis and one to PCB ground?

David,
I looked up RG-223. In the pictures I came across it looks like the two shields are in contact with each other (i.e. no isolation between them - which seems pointless). Is it just he pictures I came across that don't show properly the isolation between the two shields or is there perhaps some particular shielding effect in having two layers such as perhaps the overall thickness?

Thank you

[David Hess]

David,
I looked up RG-223. In the pictures I came across it looks like the two shields are in contact with each other (i.e. no isolation between them - which seems pointless). Is it just he pictures I came across that don't show properly the isolation between the two shields or is there perhaps some particular shielding effect in having two layers such as perhaps the overall thickness?

The shields are in contact.  The difference is that the double shields provide full or at least better coverage so there is less leakage.  They also make coaxial cable which includes a foil shield under the woven shield but it has limited flexibility.  Essentially double shields in contact provide close to the shielding of rigid cable while being still being flexible.

In RF receiver design, it is common to work with signal levels which are 90 dB or more lower than the signal source and double shielding or better is absolutely required because otherwise signals leak through the cables and around the assemblies.

Triaxial cable is better for low frequency applications where what is really required is guarding but double shielding is an improvement also and much easier to take advantage of since it only requires a different cable.

[ricko_uk]

Thank you David!

As I researched it I found all the way up to quad-shield. Multiple shields must make quite a difference. Very interesting.

Would quad shield be even more effective at low frequencies? And would increase shielding of even lower frequencies than dual or triple?

Also with those cables do you connect the shield on one side (i.e. sensor side) or both (sensor and amplifier)? Or perhaps just to shielding box? I am asking because I think ground loops might affect low signal / low noise circuits?

Thank you

[jmelson]

Hi,
I have a very tiny and noise sensitive magnetic sensor that needs to be connected to an INA amplifier PCB and I need to keep the noise as low as possible.

To keep noise picked up by the cable as low as possible, I am not sure if the best option is an SMA coax cable or a twisted pair with shielding braid (grounded on the INA side).

Actually, the best is a combination.  So, get shielded twisted pair cable (digital audio cable is good for this), but you won't be able to use a standard SMA.  If possible, make a shield that wraps over the sensor, but you may have to make a split in the shield so it doesn't cause a shorted turn.  That depends on the type of sensor.

You want everything balanced as much as you possibly can, so any interference is identical on both wires of the pair.  Then, you may need to put low-pass filters at the input of the INA.  The components on the filters may need to be matched by value to not ruin the balance of the common mode signal, so the INA can remove it.

Jon

[David Hess]

As I researched it I found all the way up to quad-shield. Multiple shields must make quite a difference. Very interesting.

Another reason I like double shielded coaxial cable is that the thicker shield makes for a stronger connection at the connector.

Would quad shield be even more effective at low frequencies? And would increase shielding of even lower frequencies than dual or triple?

I think so but there are diminishing returns.  Better screening is more effective than poorer screening.

Also with those cables do you connect the shield on one side (i.e. sensor side) or both (sensor and amplifier)? Or perhaps just to shielding box? I am asking because I think ground loops might affect low signal / low noise circuits?

If there is no ground loop then the shield is connected at both ends.  If there is a ground loop, then other considerations apply and connecting at both sides will cause more problems.  Analog Devices has some application notes on this subject.

[ricko_uk]

Thank you David and Jon!