spectral analysis of a signal

Hello

I have a situation. I have a signal stored in a .csv file. The sample interval is 2us which means a sample taken every 2 micro seconds. I have a million samples in the file. Essentially the signal is in the time domain (x(t) vs t). I have to acquire the spectrum of the signal to analyze it. Can anyone please explain how to do this?

Answer 1

[Honglei Chen]

You can use cvsread to read the signal into MATLAB first. If you have Signal Processing Toolbox, I'll suggest you to use spectrum object to compute the spectrum. See

>> help spectrum

for detailed help. If you do not have Signal Processing Toolbox, then you may need to use fft to compute the spectrum. Given you have a large data set, you may want to divide your data into segments, compute the spectrum and then average them.

[vsee]

Thank you. I did use the cvsread function to import the signal in Matlab. I do have the sig processing toolbox. I'll try the help spectrum and let you know.

[vsee]

I tried spectrum and I see there are various functions used to find PSD etc. That's not what I am trying to do. I got a signal from the front end of a receiver using an oscilloscope. I have the data stored as a csv file. When I plot time vs amplitude in Matlab there are my points of interest where there is a signal pulse and there is also a lot of noise. How can I possibly filter this and get a clean plot of my signal alone with respect to time? Thanks.

[Honglei Chen]

If you do not know which frequency your signal is at, then plotting PSD is a good way to find out roughly where it is. You can then try to design a filter to filter out the noise that is out of interested band. You can use either fdesign or fdatool to design the filter. The details of these tools can be found using

>> doc fdesign

>> doc fdatool

[vsee]

I plotted the fft and the psd of the signal I have. Here is the link to the plot.

http://www.yorkphoto.com/york/slideshow/AlbumID=3615814002/PictureID=206725940002/a=100160453_100160453/

Although I don't see a peak in the fft plot, the psd plot does show a peak and then rolls off. Should I design a low pass filter to this spectrum to filter out the noise? If yes, how do I find out what my cutoff frequencies should be? Thanks again.