All About Circuits site states that waveforms that are symmetrical above and below their horizontal centerlines contain no even-numbered harmonics. Can somebody explain this mathematically, or point to a resource? I do not know much about dsp, but understand what a Fourier transform is.

  • $\begingroup$ 0 Hz is an even (0th) harmonic and if you have a constant bias you will have it. To close this loophole, "horizontal centerline" can be replaced by "zero level". $\endgroup$ Feb 11 '19 at 6:37

The complex Fourier coefficients of a $T$-periodic function $f(t)$ are given by

$$c_n=\frac{1}{T}\int_{0}^Tf(t)e^{-j2\pi nt/T}dt\tag{1}$$


$$f(t)=\sum_{n=-\infty}^{\infty}c_ne^{j2\pi nt/T}\tag{2}$$

The coefficients with even indices are

$$\begin{align}c_{2n}&=\frac{1}{T}\int_{0}^Tf(t)e^{-j4\pi nt/T}dt\\&=\frac{1}{T}\left[\int_{0}^{T/2}f(t)e^{-j4\pi nt/T}dt+\int_{T/2}^{T}f(t)e^{-j4\pi nt/T}dt\right]\\&=\frac{1}{T}\left[\int_{0}^{T/2}f(t)e^{-j4\pi nt/T}dt+\int_{0}^{T/2}f(t+T/2)e^{-j4\pi nt/T}e^{-j2\pi n}dt\right]\\&=\frac{1}{T}\int_{0}^{T/2}\left[f(t)+f(t+T/2)\right]e^{-j4\pi nt/T}dt\\&=\frac12\frac{2}{T}\int_{0}^{T/2}\left[f(t)+f(t+T/2)\right]e^{-j2\pi nt/(T/2)}dt\\&=\frac12 d_n\tag{3}\end{align}$$

where $d_n$ are the complex Fourier coefficients of the $T/2$-periodic function $g(t)=f(t)+f(t+T/2)$. These coefficients can only be zero if $g(t)=0$, i.e., if $f(t)=-f(t+T/2)$. This latter condition is exactly the symmetry condition you mention in your question. Consequently, the even Fourier coefficients are zero if (and only if) $f(t)$ satisfies


Or, in other words, a $T$-periodic function $f(t)$ has only odd harmonics if it is $T/2$-antiperiodic. I.e., if you shift the function by half its period and flip it across the horizontal axis, it must look the same as before.


An $M\text{-}$periodic waveform that has half-periods of length $M/2$ that are of opposite signs but of equal absolute value are by definition $M/2\text{-}$Bloch-periodic with a coefficient of $-1:$ Each period of length $M/2$ is identical to the previous period multiplied by the coefficient.

A generalized frequency-shifted discrete Fourier transform (DFT) of length $M/2$ with a frequency shift of $0.5$ bin widths, or $\frac{0.5\times2\pi}{M/2}$ in angular frequency, has extended basis functions that are $M/2\text{-}$Bloch-periodic with a coefficient of $-1.$ The $M/2\text{-}$Bloch-periodic waveform can be represented by a weighted sum of the extended basis functions. The basis functions have angular frequencies $\frac{0.5\times2\pi}{M/2}, \frac{1.5\times2\pi}{M/2}, \frac{2.5\times2\pi}{M/2}, \ldots.$ This can be rewritten as $\frac{1\times2\pi}{M}, \frac{3\times2\pi}{M}, \frac{5\times2\pi}{M}, \ldots.$ The extended basis functions do not contain even harmonics of the $M\text{-}$periodic angular frequency $\frac{2\pi}{M}.$ This property is retained by the waveform which is their weighted sum.

A function that is $N\text{-}$Bloch-periodic with coefficient $-1$ is also called an $N\text{-}$antiperiodic function.


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.