1. Limited time only! Sign up for a free 30min personal tutor trial with Chegg Tutors
    Dismiss Notice
Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

I Truncated Fourier transform and power spectral density

  1. Mar 31, 2017 #1
    Hello,

    I am trying to find an expression for the signal-to-noise ratio of an oscillating signal on top some white noise. In particular I would like to know how the SNR scales with the integration time. It is well known that during some integration time ##T##, the SNR increases as ##T^{1/2}## (because the noise increases as ##T^{1/2}## and the signal increases at ##T##). I am trying to prove this with, my limited skills in maths, using the power spectral density formalism.

    The integration time is ##T##. I assume some signal: ##x(t)=A \sin(2 \pi \omega_0 t)## on top of some white noise of power spectral density ##S##. I am guessing that I should compare the power spectral density of the noise vs signal:
    $$P(f)=X(f)^2$$
    where ##X(f)## is the truncated Fourier transform:
    $$X(f)=\frac{1}{\sqrt{T}}\int_0 ^{T} x(t) e^{-i \omega t} dt$$
    In the limit where ##T## is large, the integral is a Fourier tranform, yielding a delta function:
    $$X(f)=\frac{A}{\sqrt{T}} \delta (\omega - \omega_0)$$
    Ence:
    $$P(f)=\frac{A^2}{ T } \delta (\omega - \omega_0)$$
    $$P(f_0)=\frac{A^2}{ T }$$

    Here I have that the power spectral density at the oscillation frequency decreases linearly in time. Can someone explain me where my misunderstanding is please ?
     
  2. jcsd
  3. Apr 1, 2017 #2

    blue_leaf77

    User Avatar
    Science Advisor
    Homework Helper

    It can be that my memory is playing trick in my head but as I recall the white noise variance in frequency domain increases as ##T^3##. So, do you have any reference that states what you wrote there?
     
  4. Apr 3, 2017 #3

    jasonRF

    User Avatar
    Science Advisor
    Gold Member

    I think the issue is that you are playing fast and loose with delta functions. The integral is simple enough that you may as well just do it. If I do a simplified case of ##x(t) = A e^{i\omega_0 t}## and let t run from -T/2 to T/2, then
    $$
    X(\omega) = \frac{A}{\sqrt{T}} \int_{-T/2}^{T/2} dt \, e^{i(\omega_0-\omega)t} = A \sqrt{T} \left[\frac{\sin\left(\frac{1}{2} T (\omega-\omega_0) \right)}{\frac{1}{2} T (\omega-\omega_0)} \right]
    $$
    The term in square brackets is a very nice number when ##\omega\rightarrow\omega_0##.

    jason
     
  5. Apr 3, 2017 #4

    jasonRF

    User Avatar
    Science Advisor
    Gold Member

    By the way, for the noise I get ##\langle |X(\omega)|^2 \rangle## to be independent of time, so the SNR ##\propto A^2 T##, as I would expect. You should do the calculation.

    Jason
     
Know someone interested in this topic? Share this thread via Reddit, Google+, Twitter, or Facebook

Have something to add?
Draft saved Draft deleted



Similar Discussions: Truncated Fourier transform and power spectral density
  1. Fourier Transformations (Replies: 10)

  2. Fourier Transform (Replies: 5)

Loading...