How to interpret the spectra of signal ?

[Kroun1]

How can you correctly interpret the spectra of any signal?
This may seem a simple question, but when I worked on my project about OFDM I had to think about it. In engineering practice the Fourier transform is used alot. It represents the sum of many harmonic signals, but isn’t it only mathematical substitution ? Is there only one way to interpret any signal? I don’t think so.
Remember power spectral density. It’s continuous function which represents power across frequency.
The other problem is what about aperiodic signals? Any data signals are aperiodic because they are random so you can’t find a period within them. The period is necessary to compute a spectral component, but how should it be computed if you don’t have any period? So my question is, what should I use for representation of data signals?

[charan langton]

I use spectrum pictures only in qualitative sense. They tell me if carriers are spaced properly, their relative power levels, the amount of noise present, out of band rejection, filter response, etc.

It is not a useful tool for baseband signals where eye diagram (or constellation) is more useful in telling you about quality of signals.

Spectrums are tricky to interpret also because of the video signal bandwidths of the analyzers being used. Another issue as you have mentioned is that a FFT is not really the power spectrum of the signal because it assumes that the given length of the signal is periodic, which of course it is not. For this reason, a periodogram is often used as an estimate of the spectrum of random signals. Windowing can be used to get closer to the actual estimate, but for this you need to know what the real signal spectrum looks like. Various algorithms such as Welch are used to do this estimate and they all have problems and can not be interpreted consistently in all bands of a signal. Some algorithms are good in band but flatten out of band and give a wrong picture.

In doing a periodogram, you can change the tap length which will change the way a spectrum looks by reducing the resolution, smoothing the figures so it appears to have less noise but this does not in any way change the underlying signal nor increases its frequency resolution.

Let me just say that use FFT for a spectra based on that or qualitative feel only. Thats its best use.

Charan Langton

[sathiskumar]

is powerspectrum and powerspectral density refer to the same at all times?.
plz explain.

[charan langton]

Yes, as computed by FFT they mean the same thing. The only difference can be that the PSD is often normalized. Both are total power as distributed into the number of frequency bins.

[neetikabhalla]

Respected Sir, I am working on Spectral encoding in optical CDMA. Sir, I want your help regarding the Spectrum and phase of optical signals.