I have to demonstrate that a function $x(t)$ and his Hilbert's transform are orthogonal, it is said:

$$\int^{\infty}_{-\infty} x(t) \cdot \hat{x}(t) dt = 0$$

I have tried the exercise using Parseval's theorem:

$$\int^{\infty}_{-\infty} x^2(t) \cdot \hat{x}^2(t) dt = \int^{\infty}_{-\infty} |X(f) \ast \hat{X}(f)|^2 df = \int^{\infty}_{-\infty} |X(f) \ast (-jsgn(f)X(f))|^2 df =$$

$$= \int^{0}_{-\infty} |X(f) \ast jX(f)|^2 df + \int^{\infty}_{0} |X(f) \ast (-jX(f))|^2 df$$

Using properties of convoution ($ax(t) \ast y(t)) = a(x(t) \ast y(t))$: $$\int^{\infty}_{-\infty} x^2(t) \cdot \hat{x}^2(t) dt = \int^{0}_{-\infty} | j(X(f) \ast X(f))|^2 df + \int^{\infty}_{0} |-j(X(f) \ast X(f))|^2 df =$$ $$=\int^{0}_{-\infty} |X(f) \ast X(f)|^2 df + \int^{\infty}_{0} |X(f) \ast X(f)|^2 df = $$ $$= -\int^{\infty}_{0} |X(f) \ast X(f)|^2 df + \int^{\infty}_{0} |X(f) \ast X(f)|^2 df = 0$$

But I don't know how to relationate both integrals. Maybe that isn't the way to demonstrate the exercise.

Thanks for your help.

  • 1
    $\begingroup$ See page 14 here. $\endgroup$ – MBaz Mar 28 '18 at 15:43

You are making this too complicated. Parseval's theorem in a more general form is $$\int_{t}^{ } u(t)^* \cdot v(t)\cdot dt= \int_{f}^{ } U(f)^* \cdot V(f)\cdot df$$

$$\int_{t}^{ } x(t)^* \cdot {\hat{x}}(t)\cdot dt= \int_{f}^{ } X(f)^* \cdot X(f)\cdot (-j \cdot sign(f))\cdot df $$

$$= -j \cdot \int_{f}^{ } |X(f)|^2 \cdot sign(f) \cdot df = 0$$

The final integral is zero only if $x(t)$ is real. This implies that $|X(f)| = |X(-f)|$ and hence the integral part can be split in two (below 0 and above 0) and because of the signum() function both parts cancel.


You have to use the following form of Parseval's identity:


where $^*$ denotes complex conjugation, and where $X(f)$ and $Y(f)$ are the Fourier transforms of $x(t)$ and $y(t)$, respectively. With $y(t)=\hat{x}(t)$ and $Y(f)=-j\,\textrm{sgn}(f)X(f)$, we get from $(1)$


For real-valued $x(t)$, $|X(f)|^2$ is an even function, and, consequently, the integrand on the right-hand side of $(2)$ is odd. This implies that the integral on the right-hand side of $(2)$ equals zero if it is interpreted as a Cauchy principal value.

Note that we've used the fact that $x(t)$ is real-valued. For complex-valued $x(t)$, the integrals in $(2)$ are generally not zero.


The Hilbert transform is a complicated operator. Its precise definition, and that of its Fourier transform, require a lot of care: this is a singular integral, and one should prove results in the proper function spaces (are functions integrable, in which $L_1$, $L_2$ spaces, etc.) There are two volumes by F. W. King on the Hilbert transform: Hilbert transforms.

The previous answers rely on Fourier/Parseval versions, and did not (so far) give the proper mathematical conditions. The good thing is that Fourier offers a much more intuitive vision. However, in the spirit of the law of parsimony, I would draw another version without Fourier (whose 250th anniversary is being celebrated in France), based on the symmetry/anti-symmetry properties in the time domain for real signals.

So, as a complement, I propose a pure time version, and as well without taking care about integration condition (bad bad me), but this can provide a different insight. Let us start from:

$$ \hat{x}(t)={\frac {1}{\pi }}\int _{-\infty }^{\infty }{\frac {x(\tau )}{t-\tau }}\,d\tau $$ then $$ P_x=\pi\int_{-\infty}^{\infty} x(t)\hat{x}(t)dt=\int_{-\infty}^{\infty} x(t)\int_{-\infty }^{\infty }{\frac {x(\tau )}{t-\tau }}\,d\tau dt =\int_{-\infty}^{\infty} \int_{-\infty }^{\infty } x(t){\frac {x(\tau )}{t-\tau }}\,d\tau dt$$

thus, exchanging integral signs (Fubini-without-a-care)

$$ P_x= \int_{-\infty}^{\infty} x(\tau )\int_{-\infty }^{\infty } \left({\frac {x(t)}{t-\tau }}dt \right) d\tau $$

then (switching the mute variables, not necessary, but to make notations clearer):

$$ P_x= \int_{-\infty}^{\infty} x(t ) \left(\int_{-\infty }^{\infty }{\frac {x(\tau)}{\tau-t }}d\tau\right) dt $$ thus $$ P_x= \int_{-\infty}^{\infty} x(t ) \left(-\hat{x}(t)\right) dt =-P_x\,.$$

Thus, $ P_x=0$. In other words, via basic manipulations, you turn a function into its opposite.

Let us finish with a more limited, but intuitive insight: if $x(t)$ is even (resp. odd), then $\hat{x}(t)$ is odd (resp. even ). So, the product is odd, with a zero integral. An example is the cosine, whose Hilbert transform (in some well-defined sense) is a sine. And cosine and sine are orthogonal. But combined, they yield the complex exponential $\cos t + i \sin t$.


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.