OK, I'm looking at the signal you are showing, which shows about 10k samples at 5.5, about 15k samples that have slow wave and 50 Hz negative pulses, and about 8k samples mostly near -5. I'm "guessing" that you are using 50 Hz power (as opposed to my devices plugged into 60 Hz power) and that your acquisition system is unshielded or improperly grounded (you aren't recording "single-ended", are you?). I'm also guessing that the relatively slow fluctuations (ignoring the negative pulses) are the signal you are trying to study.
Although I'm not an engineer, I learned at an early age that you want to start with as clean a signal (no "mains" noise, for example) as you can before you try to filter out "noise" (eliminating noise before you sample is always best). For your signal, I would start by carefully examining the signal to see if my supposition that it is contaminated with a largely-negative narrow pulse at (almost exactly) 50 Hz. Let me assume that this is true. I would make the following recommendations:
- First, and most important -- try to eliminate the source of the noise. Find an Electrical Engineer who understands "mains noise", shielding, and the difference between single-ended and differential recording modes.
- If you'd determined that the noise really does come from the mains, you want to eliminate it before you "acquire" it. Learn about "anti-aliasing filters" that come before the acquisition system. Choose an appropriate corner frequency, which will probably be <100 Hz.
- If you do choose to implement a post-acquisition digital filter, "think smart". If your main "noise" is "mains noise", it will occur quite reliably at 20 msec (if 50 Hz mains) intervals. You could take a 20 msec interval of data, compute the mean and S.D. of the data, and look for a region where the signal was consistently >3 SD from the mean -- this is your Noise Pulse. Replace these data with the mean of, say, the 10 points before and 10 points after the pulse, or, perhaps even better, replace the entire 20 msec sample (of 320 points) with a single point consisting of the mean of all the points except those identified as "part of the noise peak". This will appropriately low-pass (and "mains-noise-remove") your data.
I would hope you'd be able to write the code in Step 3 by yourself. Do try -- if you get stuck, attach your entire VI and come back and show your efforts. But it is most important that you do the first step of identifying (and, I hope, eliminating) the noise before trying a post-acquisition low-pass "decimating" filter (I call it that because it will replace you 35000-point record with 100-200 points).
Bob Schor
