Let x=[1, ⋯ ,1, 0, ⋯ ,0] be a window vector of length N, which consists of B consecutive 1s and the remaining N-B consecutive 0s.

I took the N-point DFT on x and got X=[X_0, X_1, ⋯, X_(N-1)] which is the result of the DFT.

I tried to show that the maximum absolute value of X_k, where k=1, ⋯, N-1, is always when k=1, for any B=1, ⋯,N.

(when k=0, it is trivial that X_0=B)

My approach was as follows:

Click and see the "The maximum absolute value of DFT of window vector" post: http://taekyo11.tistory.com/

Could anybody please help me figure out |X_k| is at peak when k=1?

I am strongly convinced that it is true through computer simulations and my intuition with the shape of cosecant function also suggests the same result, but I could not be successful mathematically.

May some kind person help me!

Thank you very much.


Think of the coefficients $\mathrm e^{2\pi\mathrm ik/N}$ as vectors in the complex plane, like the spokes of a wheel. It's geometrically clear that for any number $m$, the greatest magnitude of any sum of $m$ of these vectors is achieved if you add $m$ consecutive vectors. This is exactly what the transform with $k=1$ does. The only thing to check is whether it's possible to do better by having several copies of the same vector in the sum. But if the sum for some $k$ contains the same vector twice, the vectors in between (including one of the copies) add up to zero, so the sum is equal to a sum of a smaller number of unique vectors. For $B\le N/2$, the sum of a smaller number of consecutive vectors has a smaller magnitude, and we can reduce the case $B\gt N/2$ to the other one by substracting $1$, multiplying by $-1$ and shifting by $N-B$, none of which changes the magnitudes of the DFT coefficients.

  • $\begingroup$ I am very happy about getting help from you, Mr. joriki. I was thinking of the same idea as you but I couldn't organize it. Thanks to you, I can be sure about it and I think your explanation is mathematically neatest even without many expressions. Thank you very much Mr. joriki, have a nice day. $\endgroup$ – Taekyo Lee Oct 13 '12 at 5:31
  • $\begingroup$ @TaekyoLee: You're very welcome, 이태교씨! $\endgroup$ – joriki Oct 13 '12 at 6:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.