Frequency counter unreliable using rotary vane anemometer

[mints]

I have set up an anemometer to measure air flow speed, with the signal being acquired by a 9402 module in a cDAQ-9174 chassis (4 slot). I am using Signal Express 2011 to program the instrument.

I have set up a frequency counter task, using a maximum frequency of 1.8 kHz a minimum frequency of 250 mHz, rising edge and 1-counter (low frequency). The output is scaled using y = 0.0111x + 0. This gives 20 m/s when the frequency is 1.8 kHz.

The readings given by Signal Express give a very noisy signal, at a frequency much higher than the bandwidth of the anemometer. The range is in the order of 20% of the mean. I have attached a PDF of the signal, for two different air flow sources, the seproj file and the tdms file for one of the runs. I have also observed that the counter output is rounded to the nearest multiple of 10, e.g. 490, 470, 480, 480, 470, etc.

Is this consistent with a digital bounce issue? Is this consistent with an earthing issue? Is this consistent with a sample rate issue? Any other ideas for investigating this issue?

thanks

[Nick-C]

This could be a grouding or earthing issue. Is there anyway you could put together a wiring diagram showing how you have the device connected to the 9402? Also, if you could post the model of the anemometer you are using that would be helpful.

-Nick-

[mints]

The anemometer is a Schiltknecht MiniAir60 Mini, giving 0-4 V pulses at a frequency range of 0-1.8 kHz over an air speed range of 0 - 20 m/s.

Currently the sensor is earthed to its power supply, I will investigate electrical connections and alternative earthing arrangements. The signal is carried from a junction box to the cDAQ on a shielded cable and connects with the appropriate BNC connector.

I have also observed the frequency count using a Fluke 175 digital multimeter. By this method, the mean was 472.5 Hz, the maximum was 496.8 Hz and the minimum was 483.9 Hz. Using the NI cDAQ, the mean was 482.6 Hz, the max was 542.8 Hz and the min was 425.2 Hz. These are similar means and very different max and min readings. To my mind this is consistent with the Fluke filtering out digital bounce whereas the NI cDAQ does not. The similar means (which corresponds to a wind speed reading verified by a separate handheld anemometer) indicates to me that any noise is less significant than the signal. I am hoping that the NI cDAQ can be configured to give some sort of digital debounce or can be configured to be less sensitive to the noise.

I plan to observe the pulse train directly using the NI 9239 module that I also have in the system, when I have that data I will make it available to you.

thanks

[mints]

I have logged the voltage signal coming from the anemometer, using our 9239, a report of a few cycles is attached as a PDF. The graph clearly shows a small digital bounce. The noise is small compared to the signal, is there any way it can be filtered when acquired by the 9402?

Other tests that I ran showed a background noise in the order of 50 microVolts and a frequency of 50 Hz, this is very likely a background noise from a power supply (we run at 50 Hz in Aus) but is many orders of magnitude smaller than the signal, I assume it would be ignoredby the 9402.

[Brian-G]

Hi mints,

The cDAQ-9174 chassis does support digital filtering (debounce filtering) on its PFI lines.  Please search "PFI filtering" in the specifications below for more information:

NI cDAQ-917x User Manual and Specifications:  http://www.ni.com/pdf/manuals/372838c.pdf

I have not found a way to access these settings from signal express though.  The only way I believe it's possible is through LabVIEW or C.

Brian

[Kevin_Price]

I wouldn't suggest you *shouldn't* try to address the noise/ringing visible in your analog capture, but it doesn't look to me to be the root of your problem.

The scaled windspeed data from your original post appears to have a big component of quantization noise.  This is pretty visible in the "Run 2" data where there are a lot of instantaneous step changes in the measured value.  Your observation that count values are always multiples of 10 also suggest quantization (though 10 counts out of about 500 only accounts for a little of the overall variation.)

The fact that you can confirm the mean measurement with your Fluke also points toward a non-cumulative source of quantization error.  I've never used Signal Express, but I would look for a way to improve the measurement resolution.  Right now it kinda looks like it's counting # of anemometer edges per 1/10 sec interval.  That turns integer # edges into multiples of 10 for freq.  What *I* would do in LabVIEW is count timebase cycles between anemometer edges.   (With a 20 MHz timebase and a <2 kHz signal, quantization error would drop to less than 1 part in 10000.)    Sorry, but I don't know how to set that up in Signal Express (or even if you *can*).

-Kevin P

[mints]

Kevin

thanks for the idea for investigation. My understanding is that the counter I've set up in Signal Express will give a quantization error of less than 0.5%, but we're seeing an error of +- 5%. The signal is at about 480 Hz, the timebase is 100 kHz.

I will try to other counter configurations and observe the effects; as you say, the sudden and rapid changes in count are consistent with a quantization error.

[mints]

Brian

thanks for confirming this state of affairs, I also cannot find access to the digital debounce through Signal Express.

I may have to try exporting the project to LabView and completing the configuration there.

[mints]

Using a two-counter high frequency method, with a measurement time of 1s reduces the noise on the signal. This is consistent with reducing quantization error. However, this should not be needed with a signal of 480 Hz, as the 1-counter method using a timebase of 100 kHz should give acceptable results.

I also generated a pulse train using a second 9402, at 480 Hz; the one-counter method work perfectly, with no quantization error (or any other noise, for that matter).

I have no conclusive evidence to show whether this is a signal noise problem, or a quantization error problem, unless someone can interpret my data in a different way.

Is there any way to make a 9402 less sensitive so that the digital bounce is ignored?

[Brian-G]

Hi Kevin,

To try and confirm our suspicion that we are having debouncing problems, you can try to create a version of your project in LabVIEW.  A digital filter with a minimum pulse width can be applied using the example below.

     Digital Filtering (Digital Debounce Filtering) with M Series and CompactDAQ:  http://digital.ni.com/public.nsf/allkb/9ECE7DBE7DF338BD86256F8600734A41

Brian