Spikes in DNL histogram from ADC of triangle wave

[SteveP]

I'm examining the uniformity of the distribution of an ADC's binary digital output code when it digitizes a triangle wave that spans its input range. This DNL histrogram graph is done by digitizing many cycles of a triangle wave. NI's references to DNL mention normalizing the histrogram. A graph of the raw histrogram shows periodic spikes in the distribution. A PCI-MIO-16E-4 has high spikes every 16 binary code channels. Most are followed by a negative spike, perhaps indicating that the particular adc bit that should be changing state is instead changing on a lower voltage than it should be. Data acquired at 10 kHz on a 27 Hz trianle wave. Similar results at other acquisition and waveform rates. Spikes also o
bserved on two NI6070E cards (one ISA and one PCI). The 6070 results spiked high every 128 channels and low every 64 channels. Spikes observed digitizing the triangle wave as wellas a random noise source.

Is this a typical characteristic representing DNL but now observed in DNL results due to their being presented in some normalized fashion ? Are there other adc's that do not exhibit this ?

[Matthew C]

Hello Steve,

Your code has basically the right idea. I started with the histogram data and wrote a new vi around it that processes it correctly. Also, I had to chop off a bunch of points at the ends because the source wasn't covering the whole ADC range nicely. If you think about what a triangle wave goes through when it reverses direction then you understand why there are nonlinearities there. You typically want the triangle wave to clip just above the input limits of the DAQ board.

Run the VI and you can see, the DNL actually looks okay with correct processing, typically less than .4 LSBs of error, which is decent. Those same spikes are still buried in there, but it's okay to get an extra 100 hits in a particular bin if the average number of hits
is 1000 (that corresponds to a DNL of only +0.1). I've also computed the INL, and here I can clearly see the effects of a very poor triangle wave source. The INL line should be flat, but instead there's the classic s-shape that all circuits eventually succumb to when they get near their limits.

I hope this helps!

Matthew C
Applications Engineer
National Instruments