Multifunction DAQ

cancel
Showing results for 
Search instead for 
Did you mean: 

adjacent channel interference in NI9205

I am using cDAQ-9172 and NI9205, there are 12 peizo sensors which are used to count impacts on the surface of the sensor. I hooked up these sensor to the DAQ by 12 amplifiers. The sensors worked very well individually, however, when they worked together, the first 8 channels cross talked to each other and the last 4 also cross talked. I use RSE wiring, btw. I just don't understand why this happened, can anyone give me a hand? Thanks
0 Kudos
Message 1 of 9
(4,512 Views)

Hi Zeph,

 

What voltage levels are you seeing?  What voltage levels are you expecting?  Is there anything different from the first 8 in comparison to the last 4?

 

Here is a detailed KnowledgeBase article on crosstalk and the general causes:

http://digital.ni.com/public.nsf/allkb/B9BCDFD960C06B9186256A37007490CD?OpenDocument

 

If you have further questions on this article please let me know.

 

Sincerely,

Jason Daming

Sincerely,
Jason Daming
Applications Engineer
National Instruments
http://www.ni.com/support
0 Kudos
Message 2 of 9
(4,490 Views)

I'm seeing a similar problem, maybe the same thing.  I have used DAQ Assistant to create a task. I start the task, then DAQmx Read (Analog 1D DBL NChan 1 Samp) runs in a loop.  What I notice is that the value for channel 0 shifts to channel 1, then channel 2, untill the first 8 channels are reading the same. The task acquisition mode is "1 sample (On Demand)".   When I use the test panel in MAX, of course I can only read one channel at a time, but those look okay.

0 Kudos
Message 3 of 9
(4,144 Views)

This is normal behavior for multiplexed ADCs.  It is called "ghosting".  There are certain physical limitations to a MUX system.  Trying to measure 9.8 Volts on channel 1 and then quickly jumping to channel 2 to measure 0.0098 Volts for instance is a BAD idea.  You have to let the MUX settle first to get good accuracy. 

 

One easy way to avoid ghosting is to use grounded "guard channels" between your actual measurement channels.  See your DAQ manual for more info and/or search NI for ghosting and crosstalk.

 

EDIT:  BTW, there's no point posting an image of a DAQ Assistant.  Smiley Wink

LabVIEW Pro Dev & Measurement Studio Pro (VS Pro) 2019 - Unfortunately now moving back to C#, .NET, Python due to forced change to subscription model by NI. 8^{
0 Kudos
Message 4 of 9
(4,139 Views)

I am writing in this string because I think I have encountered similar issue with the DAQ I have been using. I have been using a multiplexed (non-simultaneously sampled device, PXI-6251). I didn't pay attention to the error until late when I started to analyse the data offline in MATLAB. I found the channel measuring a signal two orders of magnitude larger than the other channel was breaking into the second channel.

 

I am more cautious now to first establish if the two channels I am using are showing me the correct signals. Should this be happening for simultaneously sampled device? I am trying to do my earlier measurement using PXIe-6124 which  has 4 simultaneously sampled AIs. I have a feeling that it still happens if not careful. I am using BNC-2120 connector block with the 6124 card. Can you tell me if this is not ideal?

 

If I write a simple VI with only one AI channel configured in an express VI, and if I place two channles, I see the larger signal breaking into the smaller signal which is going through a different channel altogether! I don't get this. Surely there must be a way around, especially with an expensive alternative!?

 

0 Kudos
Message 5 of 9
(4,039 Views)

I am writing in this string because I think I have encountered similar issue with the DAQ I have been using. I have been using a multiplexed (non-simultaneously sampled device, PXI-6251). I didn't pay attention to the error until late when I started to analyse the data offline on MATLAB. I found the channel measuring a signal two orders of magnitude larger than the other channel was breaking into the second channel.

 

I am more cautious now to first establish if the two channels I am using are showing me the correct signals. Should this be happening for simultaneously sampled device? I am trying to do my earlier measurement using PXIe-6124 which is has 4 simultaneously sampled AI. I have a feeling that it still happens if not careful. I am using BNC-2120 connector block with the 6124 card. Can you tell me if this is not ideal?

 

If I write a simple VI with only one AI channel configured in an express VI, and if I place two channles, I see the larger signal breaking into the smaller signal whihc is going through a different channel altogether! I don't get this. Surely there must be a way around, especially with an expensive alternative!?

 

0 Kudos
Message 6 of 9
(4,037 Views)

That IS unusual.  I have used S-series cards from NI and have never experienced any ghosting across individual channels.  TBH, I've never had an app that really pushed the issue though.  Most of the signals I've collected are all in the 1-3 volt range (AND I was collecting differentially) so I may just have not noticed a small ghost effect. 

 

I suggest you check your own wiring for any potential crosstalk first.  Then, maybe try to set up two dedicated signals (that are carefully wired with good shielding), one large and one small on adjacent channels.  If you still see ghosting between them I would suspect the card and contact NI...

Good Luck!

LabVIEW Pro Dev & Measurement Studio Pro (VS Pro) 2019 - Unfortunately now moving back to C#, .NET, Python due to forced change to subscription model by NI. 8^{
0 Kudos
Message 7 of 9
(4,016 Views)

I want to get one thing cleared out on this. The S-series card I am using has 4 differential Analog channels. I have two analog signals to acquire from two different sensors which are essentially single ended. However, they come out as BNC out since that's how I designed the electronics. I am using a a BNC connector block (BNC2120) which connects with the card through 68-pin high density conectors.

Am I supposed to do anything different other than simply plug the two channels through the BNC  connectors on the block?


I know my card only does differential but I don't think with the hardware I have, I can access individual inverting and non-inverting pins within the card (AI0+, AI0- etc).

 

 

0 Kudos
Message 8 of 9
(4,013 Views)

Came across this old thread and want to supplement with an important detail about "ghosting" which is not mentioned.

 

The effect is not only dependent on the potential difference between adjacent channels but also on the impedance of the driving circuit and the sampling rate. The mux and adc node's potential has to settle before sampling. This is in the simplest approximation an RC time decay where R is the driving resistance and C the internal capacity of the mux and adc. So R should be small for fast settling.

Having a source with capacitive impedance (an RC low-pass filter for instance)  complicates the model as there will be charge redistribution between the mux/adc node capacity and the external source capacity at the mux switch-on time followed by the RC time decay where C is the sum of both capacities and R the source resistance. Typically this time constant will be much longer than the sampling time (that's the reason for the filter in the first place), so the voltage is essentially constant during sampling time. Only the initial charge distribution is of significance.

Say external capacity is Cex, and mux/adc capacity is Ci, the external voltage which we want to measure is Vext and the previous channels voltage (ghost) Vgh

Then the step in Vext called delta_Vext due to charge distribution between the two capacitors will be 

Initial step error from ghost channel chargeInitial step error from ghost channel charge

 

Example with made-up but typical values:

 Ci = 30pF, Cext = 3nF, Vgh=10V, Vext = 0V, then the voltage step will be

delta_Vext = 30pF/3030pF * 10V = 100mV! which is quite a lot when you want to measure in the mV range like ni9205.

As mentioned, the initial step will be leveled out over time due to the driving resistance of the RC filter with a time constant equal to R*(Cext+Ci) = R Cext.

 

Suppose the lowpass filter is at 5kHz then R=10kOhm and the time constant is therefore 10k*3n = 30us. So for instance in 30us the Vext error will have decayed by one time constant towards the correct value and will  become and error of 100mV/e = 100mV*0.37 = 37mV. 30us correspond to a sample rate of 33kHz.

Clearly still a large measurement error whose amplitude is proportional to the ghost voltage whatever that happens to be.

 

In this particular example increasing the Cext by 100 and decreasing R by 100 lowers the error by a factor 100 to 0.37mV.

So keep source resistance as low as possible is also important to have accurate measurements.

0 Kudos
Message 9 of 9
(2,916 Views)