Crosstalk in NI 9215 Signals

[riplakish]

I am currently using an NI 9215 module with BNC terminals to read the outputs of two different types of voltage sensors.  Sensor A is a differential o-scope probe (Tektronix P5200A) which has very good noise rejection, while Sensor B is an isolated hall-effect shunt-based measurement using a LEM LV20-P and a custom PCB, which has considerably lower noise rejection.  The noise in the circuit being measured is predominantly the result of an H-bridge inverter in the circuit which switches at 10kHz.  An image of both sensors measuring the same signal displayed an o-scope is shown below with the signal from Sensor A on top and Sensor B on the bottom.

As you can see there is significant amount of noise in Sensor B while Sensor A is mostly noiseless.  When I then connect both of these signals to my NI 9215 I get the signals shown below (sampling rate of 75 kHz), Sensor A is shown in white and Sensor B in green (ignore the differences in scaling, this is programmatic).

As you can see the level of noise in both is now comparably high.  However if we disconnect Sensor B from the 9215, the Sensor A signal then changes to the image shown below:

Although there is some noise present, the signal is significantly cleaner than before.  The natural conclusion that I draw from this is that there is significant crosstalk between the two signals.  The same cabling is being used for connecting to the 9215 as the o-scope, and both sensors use shielded twisted pair cables.  This amount of cross-talk seems very high compared to the -80dB listed in the specifications of the 9215.  Any ideas what could be causing this, or how to remedy it?  Unfortunately, I am currently unable to afford a second Sensor A.

[SteveD123]

A few thoughts:

1. Are you sure that there are no runt pulses on Sensor A/ Oscilloscope channel 1? Tektronix oscilloscopes have a triiger mode specifically for looking for these.

2. Is there a power supply involved? Could noise from it be getting into your signals?

3. By connecting to the oscilloscope, you are connecting the 0V lines of sensor A and Sensor B together, and possibly connecting them to mains earth, if it is an earthed oscilloscope. What happens if you try the same configuration with the 9215 module? According to it's data sheet "In addition, this module includes 250 Vrms channel-to-earth ground isolation for safety, noise immunity, and high common-mode voltage range." What happens if you ground the Sensor A 0V line? In fact the Tektronix P5200A spec states "Warning: For safe operation, do not use the P5200A High-voltage Differential Probe with oscilloscopes that have floating inputs (isolated inputs), such as the Tektronix TPS2000 and THS3000 Series oscilloscopes. The P5200A High-voltage Differential Probe requires an oscilloscope or other measurement instrument with grounded inputs. ", so by using it with an isolated DAQ module, you are putting yourself at risk.

4. My experience is that the very high dV/dt of inverters can get EVERYWHERE, and  is extremely difficult to get rid of. Check the feed to your PC, and the USB +5 V if you can.

[riplakish]

1.  I'm only vaguely familiar with the concept of runt pulses, but I'm not sure how they apply to an analog signal like this one.  If there are any pulses on the Sensor A signal, they are negligible, or at least much less than those on Sensor B signal.

2.  There are three power supplies involved. 1) a benchtop power supply powering the cRIO 2) a 9V wall wart powering the Sensor A probe, and 3) a +/-15V brick powering the sensor board.  I could measure the noise on these separately, however I'm sure that the vast majority is coming from the inverter, since I can see the noise appear/disappear when I turn the inverter off/on.

3.  I think you're getting to the root of the problem here.  I was unaware that the voltage probe required a grounded input.  I'm a little unsure how to go about grounding the negatives of my sensor inputs since they're BNC.  Below is a diagram from the 9215 specs.  Should I just solder a wire from the AI- to earth?

4.  You're right about the inverter.  I can see the noise messing with the pixels in my PC monitor when I turn it on.  We're currently working on adding a grounded metal enclosure around the inverter to help mitigate the noise before it gets to our sensors.

[Henrik_Volkers]

A central ground close to the 9215 can help. the +-15V should really be floating and the ground should have a close connection to central gnd.  (However see sig 😉 )

Adding chokes like this one at the 9215 input: A nice ferrite core and a BNC cable , 

can help too, but they work best for two wires, (shield??) .

You said shielded twisted pair, how is the shield connected?

[SteveD123 · Accepted solution]

1. By runt pulses, I was meaning extremely short pulses on Sensor A. If they are short enough you will not see them unless you look for them.

2. My concern is whether the switching noise is contaminating your PSUs through the incoming mains wiring. Of course good power supplies should filter this out, but it is just another thing to check.

3. The quick and dirty way would be to use a BNC T connector to connect to the oscilloscope, and ground that way.

Henrik's suggestion of ferrite chokes on the instrumentation is a good one.

My understanding is that this kind of inverter uses the load (normally a three phase AC motor) to filter the bridge frequency (10kHz in your case) down to the required frequency ( normally 50-60 Hz). This means that those high frequency currents go all the way to the motor, if they are not filtered out by the cables first. You cannot just screen the inverter enclosure, because high frequency currents down to the load are part of how it works. If you start screening things, you will have to screen all the way from the inverter to the load, and will not be able to break in to fit your instrumentation.

Standard warning: If you are tempted to connect directly to the inverter outputs, and attenuate down to the 9215 input voltage range:Firstly make sure someone qualified has checked your wiring. Second place fuses in all lines close to where the voltage is picked off. An inverter might have a fault current in the 100 A range and you do not want that down your instrumentation wiring. Not directly relevant from your posting, but I feel I have to state it.

[riplakish]

Sorry I misspoke earlier.  I'm using BNC cables for both connections like the ones in the picture you showed.  The +/-15V supply is floating.  I have connected the grounds of both sensors to earth gnd using T-connectors and BNC connection to my o-scope as SteveD123 suggested.  I assume that in your comments central=earth ground?  Obviously using the o-scope as I'm now doing is not very direct.  Do you have any other suggestions.  I have installed ferrite cores on both cables on the NI 9215 side as you suggested.  That along with grounding the AI- of each channel seems to have helped tremendously.  Thanks.

[riplakish]

1.  Gotcha, I have confirmed that such runt pulses do exist in the signal, but are small enough that they don't affect my application.

2. I assume I can test this by scoping the power supply outputs, any suggestions on fixing it if significant noise is present?  Like putting a choke on the supply wires?

3. This worked great!  The noise level on my Sensor A signal seems to be comparable to what I was seeing on the o-scope and the noise level is unaffected by the presence or absence of the Sensor B connection in the 9215.  The signal is now usable for my application.

In the case of my application the inverter has an output LC filter, and the capacitor voltage is the signal that I am trying to measure with Sensor A.  But you are correct in that the inductor current has high frequency components from the inverter switching.  From what you're saying it sounds like I would need to enclose my inductor along with the inverter (still not sure that that would solve it).  We're going to go ahead and try enclosing just the inverter and see if it helps, but as things stand we're in good shape even if it doesn't.

Thanks for the warning, but doesn't seem like it's an issue for our application, and thanks for all the help 🙂