how to perform sensor data filteration?

[hameeds01]

sensor data filteration

microcontroller is receiving some data via sensor & that sensor data has noise, now my question is how i can filter out this type of data ??

 

[s_cihan_tek]

Re: sensor data filteration

From the plot it seems like you can simply ignore the samples which are above or below a threshold value.

 

[doraemon]

Re: sensor data filteration

Hello!

It's difficult to choose a proper processing without knowing what you are
looking for. From this graph, since I don't know what signal you are measuring,
I cannot know what is signal and what is noise.

Noise can be reduced by using a filter. Period, that's all one can say about your
problem. Now if you want more accurate info, please tell us:

- What is the sampling frequency?
- What are your signal's minimum and maximum frequencies?
- What are the noise characteristics?

Dora.

 

[hameeds01]

Re: sensor data filteration

sampling time = 0.001;
i can't use a simple limiter bcz the data will be sitting in between 0 or 8

 

[bobcat1]

Re: sensor data filteration

Hi

The noise you are getting looks like a repeatable type & higher frequency then your data

Add low pass filter (hardware - r/c ) before the A/D converter

All the best

Bobi

The microcontroller specialist

 

[kender]

sensor data filteration

What's the source of noise? Why does your signal have these transients at regular intervals?

 

[doraemon]

Re: sensor data filteration

Hello!

Just a silly question. Your noise is a lot higher than your data.
And on top of that, your noise seems to be in steps of a fixed quantum.
Most of the noise is 1 steps, you have also peeks of 3, 4 steps, etc... I hope
you understand what I mean.
So here is my question: are you sure that the endianness of your signal is
right? It looks like for instance if your signal is 16 bit and if you invert
msb and lsb, you get that kind of signal. Could you try to invert your signal
samples?

If you know what is endianness, please skip the following lines.
For your information, on a PC, numbers are written in little endian. Which means
"little end come first". Example, if you want to write 0x0100 (or 256) in memory,
you need 2 bytes. The first will be 00 and the second 01.
Suppose you have a signal of 256 and a noise of 3 lsb. In this case, you
would get numbers between 0x00FD and 0x0103.
If you represent this on a graph the wrong way (little end first), then you will
get get large steps for the noise and a tiny signal..

Dora.