I am looking for the Hilbert transform of the following function:

\begin{equation} \mathcal{H}\bigg\{ \sin\Big(At^2 + Bt + \frac{\pi}{4}\Big) \bigg\} \end{equation}

where $A$ and $B$ are constants with $A<0$ and $B>0$.


It is well known that $\mathcal{H}\{ \sin(Bt) \} = -\cos(Bt)$, which can easily be shown by rewriting the Hilbert transform as a convolution with $1/{\pi t}$ and using the spectral representation as shown hereinafter:

\begin{equation} \mathcal{H}\{ \sin(Bt) \} = \sin(Bt) *\frac{1}{\pi t} \end{equation}

where $*$ denotes the convolution operator. Consider the following Fourier pair:

\begin{equation} \mathcal{F}\left\{\frac{1}{\pi t}\right\} = -\mathrm{j}\,\mathrm{sgn}(\omega) \end{equation}

With this the problem can be solved in the spectrum as follows:

\begin{equation} \mathcal{F}\big\{\sin(Bt) *\frac{1}{\pi t}\big\} = \frac{\pi}{\mathrm{j}} \big(\delta(\omega-B) - \delta(\omega+B)\big) \; \big(-\mathrm{j}\,\mathrm{sgn}(\omega)\big)\\ % = -\pi\, \big(\delta(\omega-B) + \delta(\omega+B)\big) \end{equation}

Here, the two Dirac delta pulses are located at $+B$ and $-B$ angular frequency and hence the sign function is directly applied. Therefore, we get

\begin{equation} \mathcal{H}\{ \sin(Bt) \} = \mathcal{F}^{-1}\Big\{ -\pi\, \big(\delta(\omega-B) + \delta(\omega+B)\big) \Big\} = -\cos(Bt) \end{equation}


However, the same principle cannot be applied to $x(t) = \sin(At^2 + Bt + \pi/4)$, since its spectral function are two superimposed complex Gaussian functions also shifted to $+B$ and $-B$ angular frequency:

\begin{equation} \mathcal{F}\{x(t)\} = \frac{\sqrt{-\frac{\pi}{A}}}{2 \mathrm{j}} \bigg( \exp\Big(-\mathrm{j}\frac{1}{4A}(\omega-B)^2\Big) - \exp\Big(+\mathrm{j}\frac{1}{4A}(\omega+B)^2\Big) \bigg) \quad \text{if} \quad A<0 \end{equation}

Each complex Gaussian function is defined for all frequencies and hence the application of the sign function does not simplify or solve the problem. I have also tried to directly solve the Hilbert transform integral with no success. I appreciate any help.

  • 1
    In this case, w(t) = 2*A*t + B. Could you use that info to help you? – Ben Dec 5 at 16:34
up vote 3 down vote accepted

Directly calculating this seems difficult.

My argumentation is the following. For a signal $s(t)$, the so-called analytic signal $s_{\mathrm{an}}(t)$ can be obtained by \begin{equation} s_{\mathrm{an}}(t) = s(t)+\mathrm{j} \mathcal{H}\{s(t)\}\,\,, \text{where}\,S_{\mathrm{an}}(\omega) = 0\,\forall\, \omega < 0 \end{equation} The analytic signal essentially corresponds to the spectral content of $s(t)$ in the positive frequencies only.

For your first example of a simple sinusoid, you can also come to the result for the Hilbert transform if you consider the analytic signal. The real sinusoid consists of the frequency components $\pm B$. The analytic signal should then be the component at $B$, which is then obviously a complex exponential, yielding the result for the Hilbert transform.

Now for your chirp signal $x(t)$, the situation is somewhat more complicated. If we think about a virtual "instantaneous frequency course" of the signal, it is \begin{equation} \omega_{x}(t) = \pm (2At+B)\,. \end{equation} Now this is somewhat strange, corresponding to two linearly changing components of opposite slope, crossing the zero-frequency point at $\omega_{x}(t=-\frac{B}{2A})=0$.

Now the analytic signal would have to represent the part of this frequency course above the zero-line of the $\omega-t$ plane (I might add some plots later). This means it would have to first have a negative slope, go down to frequency zero, and then abruptly change to a positive slope!

This means that the analytic signal would have to look something like \begin{equation} x_{\mathrm{an}}(t) = c_1 \exp{(-\mathrm{j} (At^2 + Bt + \frac{\pi}{4}))}\,\forall\,t < -\frac{B}{2A} \end{equation} and \begin{equation} x_{\mathrm{an}}(t) = c_2 \exp{(\mathrm{j} (At^2 + Bt + \frac{\pi}{4}))}\,\forall\,t \geq -\frac{B}{2A}\,\,, \end{equation} where the $c$ are some constant with $\vert c \vert = 1$.

Now we can determine the Hilbert transform of $x(t)$ by observing and checking in the equation for the analytic signal. This yields \begin{equation} \mathcal{H}\{x(t)\} = \cos{(At^2 + Bt + \frac{\pi}{4})}\,\forall\,t < -\frac{B}{2A}\,,\,\text{with}\,c_1 = -\mathrm{j}\,, \end{equation} and \begin{equation} \mathcal{H}\{x(t)\} = -\cos{(At^2 + Bt + \frac{\pi}{4})}\,\forall\,t \geq -\frac{B}{2A}\,,\,\text{with}\,c_2 = \mathrm{j}\,. \end{equation}

One can probably also write these as one equation with the absolute value function. In any case, the point is the Hilbert transform seems to contain a discontinuity, which is what makes this particularly confusing to calculate I suspect.

I know it is somewhat "handwaivy", but I think the general idea/result is correct, so hope this helps!

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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