I have a LTV (linear and time-varying) system. So, $h(\tau, t)$ is the "instantaneous impulse response" at time $t$ such that if the input signal is $x(t) = \delta(t - t_0)$ (an impulse at time $t_0$), the output signal should be $y(t) = h(t - t_0, t_0) = h(\tau, t_0)$ where $\tau = t - t_0$ is the delay/lag from the impulse.

This makes sense to me, intuitively. Obviously, if you input an "impulse at $t_0$", the output signal will be the "impulse response at time $t_0$".

This also has the nice property that if the input is some sort of comb (sum of many impulses with arbitrary complex-valued amplitudes), $$ x(t) = \sum_i A_i \delta(t - t_i) $$ then the output signal is $$ y(t) = \sum_i A_i h(\tau, t_i) $$ which also seems very reasonable to me.

However, this does not agree with the treatment given in most literature that deals with LTV systems where: $$ y(t) = \int_{-\infty}^{+\infty} h(\tau, t) x(t - \tau) d\tau $$

As above, if we have an impulse at $t_0$ such that $x(t) = \delta(t - t_0)$ then $$ \begin{align} y(t) &= \int_{-\infty}^{+\infty} h(\tau, t) \delta(t - \tau - t_0) d\tau \\ &= h(t - t_0, t) \\ &= h(\tau, t) \end{align} $$

This does not match what I expect above!

Am I misunderstanding something? I have a nagging feeling that this is somehow just a matter of definition/convention.

However, I want to make sure I am not accidentally making an implicit assumption of some sort that I don't have to make.


Okay, if I define the integral differently, I get what I expected: $$ y(t) = \int_{-\infty}^{+\infty} h(\tau, t - \tau) x(t - \tau) d\tau $$

Is this wrong? How does this differ from the usual treatment of time-varying impulse response?

One immediate problem is that this no longer looks like a "convolution".

  • $\begingroup$ If $x(t)=\delta(t-t_0)$, then $x(t-\tau) = \delta(t-\tau-t_0)$. $\endgroup$
    – Ash
    Oct 6, 2022 at 21:55
  • $\begingroup$ Yes, that's what I used to compute the integral. I edited my question to show this is what I am doing. $\endgroup$
    – XYZT
    Oct 6, 2022 at 21:56
  • $\begingroup$ Your last equation is correct. The response for an impulse at a delay of $\tau'$ is only valid for the impulse response defined at that time delay (i.e. $h(\tau, t-\tau')$). A response $h(\tau, t)$ would only be realized by an impulse occuring at $\delta(t)$, not $\delta(t-\tau')$. Variable overload, adding prime for input delay $\endgroup$
    – Ash
    Oct 6, 2022 at 22:11
  • $\begingroup$ Hmm, but it does not look like a convolution - which I thought would be the case if it was correct. $\endgroup$
    – XYZT
    Oct 6, 2022 at 22:24
  • $\begingroup$ Your analysis agrees with this blog post. $\endgroup$
    – Ash
    Oct 6, 2022 at 22:43

1 Answer 1


The problem is that there are two common definitions of the impulse response of an LTV system, resulting in the following input-output relations:

$$ y(t)=\int_\tau h_1(t,\tau)x(\tau)d\tau\tag{1}$$


$$ y(t)=\int_\tau h_2(t,\tau)x(t-\tau)d\tau\tag{2}$$

In the first one, the impulse response (integration kernel) $h_1(t,\tau)$ is the response at time $t$ to an impulse at time $\tau$. In the second, $h_2(t,\tau)$ is the response at time $t$ to an impulse at time $t-\tau$. The impulse response $h_2(t)$ is also called the input delay-spread function.

The relationship between $h_1(t,\tau)$ and $h_2(t,\tau)$ is

$$h_1(t,\tau)=h_2(t,t-\tau),\quad h_2(t,\tau)=h_1(t,t-\tau)\tag{3}$$

The conditions for time-invariance are as follows. For the first definition we require that $h_1(t,\tau)$ only depends on the difference $t-\tau$:


The second definition results in an impulse response that is independent of $t$:


The causality conditions are $h_1(t,\tau)=0$ for $t<\tau$ and $h_2(t,\tau)=0$ for $\tau<0$.


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.