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!

Optics convolution

  1. Feb 4, 2015 #1
    1. The problem statement, all variables and given/known data
    According to my notes, if we have a sinusoidal aperture/transmission function of the form a(x)=1+sin(wx) and a 'top-hat' aperture function given by b(x)=1, -0.5d≤x≤0.5d, b(x)=0 otherwise, then their convolution should give a finite sinusoidal aperture function, i.e sinusoidal but only over a certain range - outside of this it is zero.

    2. Relevant equations
    Convolution of a(x) and b(x) is ∫-∞a(y)b(x-y)dy

    3. The attempt at a solution
    We need ∫-∞[1+sin(wy)]b(x-y)dy, and my thinking is that the integrand is zero everywhere apart from between x-0.5d and x+0.5d, where it is simply [1+sin(wy)]. Then the convolution integral reduces to ∫x-0.5dx+0.5d[1+sin(wy)]dy giving a-{cos[w(x+d/2)]+cos[w(x-d/2)]}/w which is not the finite sinusoidal aperture function I've been told I should get. Is this right?
    Last edited: Feb 4, 2015
  2. jcsd
  3. Feb 4, 2015 #2


    User Avatar
    2017 Award

    Staff: Mentor

    The convolution of a signal with a "box" is not the signal restricted to a specific range. If you want that, you have to multiply the two functions.
    "a" seems to have two different meanings which is confusing.
  4. Feb 4, 2015 #3
    Changed a to d, didn't realise that.

    I'm not clear - the function b(x-y) is a box centred on y=x with width d, right? Then the integrand [1+sin(wy)]b(x-y) surely vanishes everywhere but for where the box is non-zero, and so basically we can write b(x-y)=1 as long as we integrate over the box, from x-0.5d to x+0.5d. Then I just do the integral. I'm not sure what you mean...
  5. Feb 4, 2015 #4


    User Avatar
    Science Advisor
    Homework Helper
    Gold Member

  6. Feb 4, 2015 #5

    Ray Vickson

    User Avatar
    Science Advisor
    Homework Helper

    Yes, except for a sign. If ##C(x)## is the convolution we have
    [tex] C(x) = a+\frac{\cos(wx - wa/2) -\cos(wx+wa/2)}{w}[/tex]
    However, using standard trigonometric additional formulas, this can be manipulated to give ##C(x) = a + K \sin(w x)##, where ##K## is some constant that I will leave you to compute.

    Note, however: the formula for ##C(x)## applies on the whole real line ## -\infty < x < +\infty##; it does not vanish outside of some finite interval. This follows from the fact that
    [tex] C(x) = \int_{-\infty}^{\infty} a(y) b(x-y) \, dy = \int_{-\infty}^{\infty} a(x-y) b(y) \, dy [/tex]
    (by a simple change of variables). The second form gives
    [tex] C(x) = \int_{-a/2}^{a/2} [1 + \sin(w(x-y)) \, dy [/tex]
  7. Feb 4, 2015 #6
    Thanks for your reply! So my notes say that it should give me a sinusoidal function, BUT, only over a finite interval - so we agree that they must be wrong then?
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: Optics convolution
  1. Convolution Question (Replies: 0)

  2. Convolution problem (Replies: 4)

  3. Convolution Integral (Replies: 8)

  4. Convolution Problem (Replies: 3)

  5. Help with convolution (Replies: 1)