Logo   Information, Signal, Images et ViSion C.N.R.S.   GdR   M.E.S.R.


Properties

Here is a list of the main properties of the WVD [Fla93].
  1. Energy conservation: by integrating the WVD of x all over the time-frequency plane, we obtain the energy of $x$:

    \begin{displaymath}E_x = \int_{-\infty}^{+\infty} \int_{-\infty}^{+\infty} W_x(t,\nu)\ dt\
d\nu\end{displaymath}

  2. Marginal properties: the energy spectral density and the instantaneous power can be obtained as marginal distributions of $W_x$:

    \begin{eqnarray*}
\int_{-\infty}^{+\infty} W_x(t,\nu)\ dt &=& \vert X(\nu)\vert^...
...\int_{-\infty}^{+\infty} W_x(t,\nu)\ d\nu &=& \vert x(t)\vert^2
\end{eqnarray*}


  3. Real-valued:

    \begin{displaymath}W_x(t,\nu)\ \in {\cal R},\ \forall\ t, \nu\end{displaymath}

  4. Translation covariance: the WVD is time and frequency covariant:

    \begin{eqnarray*}
y(t)=x(t-t_0) &\Rightarrow & W_y(t,\nu)=W_x(t-t_0,\nu)\\
y(t)=x(t) e^{j2\pi \nu_0 t} &\Rightarrow & W_y(t,\nu)=W_x(t,\nu-\nu_0)
\end{eqnarray*}


  5. Dilation covariance: the WVD also preserves dilations:

    \begin{eqnarray*}
y(t)=\sqrt{k}\ x(kt)\ ;\ k>0\ \Rightarrow\ W_y(t,\nu)=W_x(kt,\frac{\nu}{k})
\end{eqnarray*}


  6. Compatibility with filterings: it expresses the fact that if a signal $y$ is the convolution of $x$ and $h$ (i.e. the output of filter $h$ whose input is $x$), the WVD of $y$ is the time-convolution between the WVD of $h$ and the WVD of $x$:

    \begin{displaymath}y(t)=\int_{-\infty}^{+\infty} h(t-s)\ x(s)\ ds\ \Rightarrow\
W_y(t,\nu)=\int_{-\infty}^{+\infty} W_h(t-s,\nu)\ W_x(s,\nu)\ ds\end{displaymath}

  7. Compatibility with modulations: this is the dual property of the previous one: if $y$ is the modulation of $x$ by a function $m$, the WVD of $y$ is the frequency-convolution between the WVD of $x$ and the WVD of $m$:

    \begin{displaymath}y(t)=m(t)\ x(t)\ \Rightarrow\ W_y(t,\nu)=\int_{-\infty}^{+\infty}
W_m(t,\nu-\xi)\ W_x(t,\xi)\ d\xi\end{displaymath}

  8. Wide-sense support conservation: if a signal has a compact support in time (respectively in frequency), then its WVD also has the same compact support in time (respectively in frequency):

    \begin{eqnarray*}
x(t)=0,\ \vert t\vert>T &\Rightarrow & W_x(t,\nu)=0,\ \vert t...
...,\ \vert\nu\vert>B &\Rightarrow & W_x(t,\nu)=0,\ \vert\nu\vert>B
\end{eqnarray*}


  9. Unitarity: the unitarity property expresses the conservation of the scalar product from the time-domain to the time-frequency domain (apart from the squared modulus):

    \begin{displaymath}\left\vert\int_{-\infty}^{+\infty} x(t)\ y^*(t)\ dt\right\ver...
...y}
\int_{-\infty}^{+\infty} W_x(t,\nu)\ W_y^*(t,\nu)\ dt\ d\nu.\end{displaymath}

    This formula is also known as the Moyal's formula.

  10. Instantaneous frequency: the instantaneous frequency of a signal $x$ can be recovered from the WVD as its first order moment (or center of gravity) in frequency:

    \begin{displaymath}f_x(t)=
{\int_{-\infty}^{+\infty} \nu W_{x_a}(t,\nu)\ d\nu
\over
\int_{-\infty}^{+\infty} W_{x_a}(t,\nu)\ d\nu}\end{displaymath}

    where $x_a$ is the analytic signal associated to $x$.

  11. Group delay: in a dual way, the group delay of $x$ can be obtained as the first order moment in time of its WVD:

    \begin{displaymath}t_x(\nu)={\int_{-\infty}^{+\infty} t\ W_{x_a}(t,\nu)\
dt\over \int_{-\infty}^{+\infty} W_{x_a}(t,\nu)\ dt}\end{displaymath}

  12. Perfect localization on linear chirp signals:

    \begin{displaymath}x(t)=e^{j2\pi \nu_x(t) t} \mbox{ with } \nu_x(t)=\nu_0+2\beta t\
\Rightarrow\ W_x(t,\nu)=\delta(\nu-(\nu_0+\beta t)).\end{displaymath}

Eric Chassande-Mottin 2005-10-26

© GdR ISIS - Contact