Fourier Series Coefficient Confusion

37 Views Asked by At

In the context of the Discrete Fourier Transform, a function in time may be expanded as a Fourier series,

$$ f_m = \frac{1}{T} \sum_{n=0}^{N-1} \tilde{f}_n e^{2\pi inm/N}, $$

where $ f_m $ is the function in the time domain in time step $ m\Delta t $, $ f_m = f(m\Delta t) $, $ T = N \Delta t $ is the period, $ f(m\Delta t) = f(m\Delta t + T) $ and $ \tilde{f}_n $ are the frequency-resolved weights.

The last, are computed in terms of the time-resolved function as

$$ \sum_{m=M}^{M+N-1} f_m e^{-2\pi in'm/N} \Delta t = \frac{1}{N} \sum_{n=0}^{N-1} \tilde{f}_n e^{2\pi i(n-n')M/N} \sum_{m=0}^{N-1} e^{2\pi i(n-n')m/N} = \tilde{f}_{n'}, $$

for any given integer $ M $.

Assuming the above is correct, $ \underline{\text{my question is the following}} $: If we can get the same coefficient $ \tilde{f}_n $ for ANY given integer $ M $, we get that

$$ \sum_{m=M}^{M+N-1} f_m e^{-2\pi inm/N} \Delta t = \sum_{m=0}^{N-1} f_m e^{-2\pi inm/N} \Delta t $$

$$ \Rightarrow \left( e^{-2\pi inM/N} - 1 \right) \sum_{m=0}^{N-1} f_m e^{-2\pi inm/N} = 0 $$

$$ \Rightarrow Mn = \mathbb{Z} N \ \lor \ \tilde{f}_n = 0, $$

which it seems to me obviously wrong since $ M $ can be any integer.

What am I missing?

1

There are 1 best solutions below

0
On BEST ANSWER

Your last step is the error. You can't subtract the two sums and assume the values are the same.

If we change the variable in the first sum to $p=m-M,$ you get:

$$\sum_{m=M}^{N+M-1} f_m e^{-2\pi mni/N}=\sum_{p=0}^{N-1}f_{p+M}e^{-2\pi (p+M)ni/N}$$ Now we can subtract this sum from the right sum term by term.

Your result uses $f_p$ where you should have $f_{p+M}.$ You substituted correctly in the exponent, but failed to substitute in the subscript.