lets assume following signal $$x=a_1 \sin(2\pi f_1 t) + a_2 \sin(2\pi f_2 t).$$ This is passed through a simple single-pole filter with cut-off frequency $f_c$. How is $x$ delayed by the filter (peak of the cross-correlation)?

In the case $a_2=0$, the time delay between the original and the filtered frequency is $$T_1=\arctan(f_1/f_c)/(2\pi f_1).$$

In the case $a_1=0$, the time delay between the original and the filtered frequency is $$T_2=\arctan(f_2/f_c)/(2\pi f_2).$$

But how can we calculate the time delay $T_{12}$ in the case $a_1\neq0$ and $a_2\neq0$?

It is not the group delay: It must depend in some way on the values of $a_1$ and $a_2$, because if $a_1>>a_2$, then $T_{12}\approx T_1$.

Any light on this would be greatly appreciated!


1 Answer 1


The answer is that the output signal for the case $a_1\neq 0$ and $a_2\neq 0$ is not simply a delayed version of the input signal anymore, as it would be if the input were just a single sinusoid. With $T_1$ and $T_2$ as given in your question, the response to the combined input signal is

$$y(t)=\frac{a_1}{\sqrt{1+(f_1/f_c)^2}}\sin(2\pi f_1(t-T_1))+\frac{a_2}{\sqrt{1+(f_2/f_c)^2}}\sin(2\pi f_2(t-T_2))$$

which cannot be written as $x(t-T_x)$ with some unknown delay $T_x$.

But as you suggested, you could define some 'global delay' by the first maximum of the cross-correlation between the input signal and the output signal. This cross-correlation is

$$R_{xy}(\tau)=\frac{a_1^2}{2\sqrt{1+(f_1/f_c)^2}}\cos(2\pi f_1(\tau-T_1))+\\+ \frac{a_2^2}{2\sqrt{1+(f_2/f_c)^2}}\cos(2\pi f_2(\tau-T_2))\tag{1}$$

For $a_2=0$, there is obviously a maximum of $R_{xy}(\tau)$ at $\tau=T_1$. The same is true for $a_1=0$, which gives $\tau=T_2$ as the location of the maximum. However, for $a_1\neq 0$ and $a_2\neq 0$ I don't see how to compute the first maximum of (1) analytically. A numerical solution is of course possible.

  • $\begingroup$ Thanks Matt. But when we calculate the cross correlation, the peak will be somewhere between $T_1$ and $T_2$. Is there no analytic solution to find that peak? Btw, shouldn't the new amplitudes be $1/\sqrt(1+f_1^2/f_c^2)$ ? $\endgroup$ Aug 16, 2014 at 22:27
  • $\begingroup$ @AnkaiosArgo: Thanks for pointing out the magnitudes, I was just focusing on the delays. Fixed it. I'll need to think about what you say about the cross-correlation. $\endgroup$
    – Matt L.
    Aug 16, 2014 at 22:33
  • $\begingroup$ Sorry, I didn't want to be picky :-) Would be great, if you find a solution... In order to find the peak in the cross-correlation, maybe we can calculate the cross-spectra? $\endgroup$ Aug 16, 2014 at 22:43
  • $\begingroup$ @AnkaiosArgo: I added some info about cross-correlation to my answer. Unfortunately, I can't see how to find an analytical solution. $\endgroup$
    – Matt L.
    Aug 17, 2014 at 0:10
  • $\begingroup$ Thanks, this already helps... would you mind to elaborate, how you determined the $R_{xy}(\tau)$? The auto-correlation of a sum of sinus is a sum of cosines, see e.g. chem.purdue.edu/courses/chm621/lecture/electronics/… , but in the case of the cross-correlation and the time-shift $T_i$ it does not seems very trivial to me. $\endgroup$ Aug 17, 2014 at 15:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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