Unity gain buffer on AI lines

[Vikash_Kumar]

There is no good news but I realized something stupid that I was doing.

For the debugging purposes and the data/graphs I provided above, I was reading using a task with only 8 channels (AI0:7) instead of all 80 channels. I was NOT setting the convert clock explicitly, allowing NI to use its default. For 8 channels it was defaulting to 90901 Hz, effectively providing long settling time than what it will be when all 80 are being samplled.

I am now sampling all 80 channels and the convert clock now defaults to 740741Hz. It also appears that arbitrary convert clock values are now allowed. When I tried setting it to 72000, it defaults to 740741. Anything below 72000 kHz results in error (expected). Next value that it can be set is 769231 (for my rate and #channels).

Settling time of this order makes the problem even more dire. The correlation got worse. 

Note: To accommodate the correlations, Y axis range has been increased.

For sanity check, I reduced the sampling rate to 500 and let NI pick its own convert clock. Its picked up 90909 Hz. Effectively NI has bigger settling time between each measurements and correlations dies (there is still little noise). Lowering the convert clock (using  DAQmxSetAIConvRate) to 500*80 = 40000 Hz didnt have much of a difference. Here is the one with the 90909 Hz.

I also tried the star ground connection without the buffers. No good news there as well. Furthermore, (1) playing with the settling time changes the correlation effects (2) Multiplexed vs Non-Multiplexed experiment above also converys that its something in the NI that is messing things.

Things otherwise are fairly clean. I have a NI chassi with 7 cards. There are about 200 sensors being read and 50 actuators being commanded. My NI codes reliably works for everything else. Its just this one particular sensor that is being painful. I have tried all these experiments in isolation wrt of the rest of the system as well. I still get the correlations.

I have tried it with following cards PXI-6255, PXIe-6363 and USB-6259 , PCI-6259 and in isolation with everything else (other sensors and actuators) I have. No luck.

[Henrik_Volkers]

---

@Vikash_Kumar wrote:

Things otherwise are fairly clean. I have a NI chassi with 7 cards. There are about 200 sensors being read and 50 actuators being commanded. My NI codes reliably works for everything else. Its just this one particular sensor that is being painful. I have tried all these experiments in isolation wrt of the rest of the system as well. I still get the correlations.

I have tried it with following cards PXI-6255, PXIe-6363 and USB-6259 , PCI-6259 and in isolation with everything else (other sensors and actuators) I have. No luck.

---

This correlation seems to be mostly due to multiplexing and electical crosstalk. Can you isolate the sensor to make shure that there isn't a mechanical correlation? (Just thinking of vibration etc.)

Since the sensor can drive 2k Ohm, have you tried a 2k load (sig+ to sig-(??GND) ) ?  Keep in mind that your GND from the sensor now is carrying that current...

[Vikash_Kumar]

Henrik,

Thats exactly what I mean when I said that I have tried them in isolation wrt to the rest of the system - away from any mechanical or electrical systems.

I am not electrical/ electronics oriented by training. I'll be upfront and accept that I don't completely understand what you recommended in the later half of you response. Are you recommending to use a 2kohm load in series with the sensor? It would be best if you could provide a schematic drawing of you recommendation so that there is no point of confusion.

Vikash

[Catherine-B]

Hi Vikash,

Can you try feeding a signal other than the magnetic sensors into the AI of the card (such as a sine wave from a function generator with a similar frequency)? It is starting to seem like those particular sensors may not be able to handle a 10kHz sampling rate. Have you tried sampling rates between 500 and 10kHz to see if there is a frequency cutoff where the ghosting becomes more apparent?

[Vikash_Kumar]

I dont have a signal generator. I am trying to get hold of one. We will have to wait a little for that test. 

Meanwhile, I tried different sampling rates. Its seems like after 2Khz the correlation effects become really prominent. Still, things below this rate doesnt look that promising either. Here are the plots

[CoffeeBreak]

Vikash_Kumar,

First, I wanted to point out that the data sheets for the sensor you provided earlier have a sampling time of 1 ms. This corresponds to a sampling rate of 1kHz and thus I am not sure we can rely on measurements faster than this sampling rate.

MPS Data Sheet #1

MPS Data Sheet #2

According to this KnowledgeBase Article: Incorrect Readings when Scanning Unconnected or Open Channels, it is very possible we are seeing the elevated voltage due to several reasons. If we connect another MPS to AI2 what does the data look like?

[Vikash_Kumar]

1) All the channels from AI0-AI79 are conencted to the MPS sensors. There is no floting terminal being read.

2) My control loop runs in those ranges and needs precise measurements. I want to over sample a little to average out the white noise. I understand that the sensor data will still update at 1kHz but oversamplling trick shoud still work.

In any case even at 500 - 1000 Hz we are seeing correlations beyond acceptance.

[CoffeeBreak]

Vikash_Kumar,

Were you able to obtain a function generator that would be capable of a frequency similar to the sensor output frequency?

[Vikash_Kumar]

No I wasn't but here is what I did, which makes up for The absence of a function generator.

I took another NI card and used its AO channel to generate a 5 Hz sine wave signals of 4.5v amplitude. This signal was externally wired to The AI-port of The NI 6255 board under use for this study. This signal is marked as "NI-AO" in The plots bellow. The MPS magnetic sensors are marked as "MPS" in the plots below.  

For the experiments, all 80 channels was sampled at various frequencies(500/ 1000/ 9000 Hz). Different combinations between NI-AO and MPS were tried at channels NI-AI0 and NI-AI1, which will be clearly marked in the plots below. NI-AI2 : NI-AI79 were always connected to MPS sensors.

1) This is a baseline plot with both AI0 and AI1 externally driven by a different cards NI-AO channel. Both were kept fixed at 5v and sampled at 2Khz. 

2) This is a plot with both AI0 and AI1 externally driven by a different cards NI-AO channel. One kept fixed at 5v and other with "NI-AO" input (i.e.. 5 Hz sine wave from external NI card)

3) These are plots with one MPS(fluctuating) and one NI-A0 (fixed at 5 v)

4) This are plots with one NI-A0 (fluctuating) and one MPS (fixed) i.e.. The MPS sensor values are fixed but NI provides a fluctuating reading. Why? That what we are trying to find (probably ghosting effect or something else)

5) With experiments mentioned earlier in the thread, we have seen that buffering does help a little with the correlations. I repeated the tests where both the NI-AO and the MPS channels was buffered. For low samplling rates, things are not so bad but once you crank up the samplling rates things go crazy. Here are the plots

[CoffeeBreak]

Vikash_Kumar,

Would you mind uploading some screenshots of the data that includes a task that only samples AI1 connected to an MPS sensor? Make sure that you change the physical channel input in your VI to just AI1 instead of including AI0-AI79.

Please include the 500 Hz, 1 kHz, 2 kHz, and 9 kHz sample rates you have used in the tests above.