0
$\begingroup$

Why does a pure sine wave starting and ending at 0 in the sample window of a DFT but containing 1.5 cycles (or periods) produce the same spectral leakage (i.e. not more) as a pure sine wave with integer number of periods in the sample window of a DFT.

$\endgroup$
  • $\begingroup$ The integer number of periods (of a pure sinewave) has to be equal to the length of the DFT (e.g. N, not (N-1)), for the DFT result to end up in a single bin. $\endgroup$ – hotpaw2 Jun 12 '16 at 15:33
1
$\begingroup$

That's not true. Take a simple example:

N = 8; n = 0:N-1;
w1 = pi/4; w2 = 3*pi/8;
s1 = sin(n*w1); s2 = sin(n*w2);
S1 = fft(s1); S2 = fft(s2);
subplot(2,2,1),stem(n,s1)
subplot(2,2,2), stem(n,s2)
subplot(2,2,3), stem(n,abs(S1))
subplot(2,2,4), stem(n,abs(S2))

enter image description here

$\endgroup$
  • $\begingroup$ Thank you Matt for the great answer - so when I see in a Matlab DFT Demo a similar result for Integer and Integer and a half, but much worse for integer and 1/3, 1/4 etc does it mean that 1.5 integer cycles is the least worst case for discontinuities (i.e. sig is at 0 - no "corner " so less high freq content ), and a window is optimised to make it look as good as an integer cycle case? $\endgroup$ – BillO Jun 12 '16 at 19:32
  • $\begingroup$ @BillO: You're right; think about it in continuous time: 1.5 periods result in no discontinuity (only a discontinuity in the derivative), whereas in the other cases you have a discontinuity leading to large spectral leakage. $\endgroup$ – Matt L. Jun 12 '16 at 19:36
  • $\begingroup$ @BillO: Also take a look at this answer to a related question. $\endgroup$ – Matt L. Jun 12 '16 at 19:39
  • $\begingroup$ Thanks again Matt for the informative answer - aploogies for more questions but I wonder how does the cyclical nature of the DFT (in that sample 0 is considered adjacent to last sample N) cope with the discontinuity in derivative? Is it because it the derivative is no different in magnitude but only in sign (i.e. +ve to -ve or vice versa). Is the spectral leakage you see with 1.5 integer periods due to the extra quanity of +ve (or-ve) sig energy not being cancelled out by its inverse (i.e. the missing half cycle)? $\endgroup$ – BillO Jun 16 '16 at 17:50

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.