1

EDIT: I've made adjustments based on your answers, while I couldn't get it to be evenly spaced, I managed to get a better curve with more samples per cycle. Turns out I need to remove the print function from my program so that the Pi could focus more on reading and plotting.

See ((Final adjustment)) at the bottom

Question:

Thanks for reading this. I'm doing a project using Raspberry Pi and MCP3208 (ADC) and currently, I'm testing my program with NI myDAQ's function generator to supply a 10Vpp to the ADC at 50Hz. I've tried using NumPy and managed to get some readings. However, when I got to my supervisor I noticed that my frequency reading is not exactly at 50Hz. Then I tried to change the input to 25Hz to see if it works at a lower frequency. Proceeding to 10Hz, I noticed that it is true! It is more accurate at a lower frequency.

My friend tried using another ADC chip and it worked accurately (or at least more accurate than mine) so I am wondering where did I go wrong with this? Below are my results and I've placed my code underneath.

READING AT 50HZ

Reading at 50Hz

READING AT 10HZ

Reading at 10Hz

FINAL ADJUSTMENT

Reading at 50Hz

Hazman Hussin
  • 11
  • 4
  • 1
    The more samples in a waveform you have OR the more exact that those samples fit the waveform cycle precisely, the better the outcome. You also appear to be missing samples: your 25 Hz screenshot is missing a sample at close to the far right position. There appear to be samples missing in your 50 Hz picture too. – Andy aka Mar 02 '20 at 16:15
  • 1
    You understand that the RPi is not an RTOS with hard-timing limits right? Your readings in Linux, especially if running X or GUI apps, can vary up to about 30ms (or more). – Ron Beyer Mar 02 '20 at 16:33
  • Measuring frequency very accurately is not at all as simple as it seems. Google “Bogdan Kasztenny Frequency”. I was looking at a couple of modern microprocessor relays this morning and they are using Clarke components to create a voltage waveform which they then use to time zero crossings. – relayman357 Mar 02 '20 at 16:46
  • @Andyaka from what I understand, the FFT is obtained from a reading that is beyond what I have shown in the diagram. I only made the window small so that I could count how many samples per cycle. I've updated the samples obtained in a larger window but I think you're right, I am missing samples. – Hazman Hussin Mar 02 '20 at 17:09
  • @RonBeyer my colleague managed to get a more accurate reading and she was reading at 100Hz. I've read multiple online posts and articles on raspberry pi limitation and came to the same conclusion but just wondering if I made a mistake here as my friend managed to get better results than mine. – Hazman Hussin Mar 02 '20 at 17:12
  • Hmm, your samples seem to be unevenly spaced. The FFT does not account for that if I'm not mistaken - it assumes it is time series data with even spacings.You do record the times you've taken the samples though, so you can try to resample the `ch0` buffer to make it even-stepped - using some form of interpolation. Then the FFT should be sharper and more precise. – anrieff Mar 02 '20 at 17:13
  • @relayman357 thank you, I'll take a look at it tho the article is long – Hazman Hussin Mar 02 '20 at 17:14

0 Answers0