I know that group delay is defined as :

$$\begin{align} \tau(\omega) = grd [H(e^{j\omega})] &=-\frac{d[\arg[H(e^{j\omega})]]}{d\omega}\end{align}$$

But if what is available is group delay itself, is it possible to recover the phase?

According to Discrete-Time Signal Processing 3rd edition by Alan Oppenheim equation 5.86 the expression would look like this:

$$\begin{align} \arg[H(e^{j\omega})] = -\int\limits_{0}^{\omega} \end{align} grd [ H(e^{j\phi})]\, d \phi + \arg[H(e^{j0})] $$

and my doubt is:

Where did the second term $$\arg[H(e^{j0}) ]$$ came from?

  • 1
    $\begingroup$ That's basic math – the antiderivative of a function $f(t)$ is always only certain up to a constant $C$ as in $\int f(t)\,dt +C$. That $C$ must fulfill the boundary conditions, and in this case, that condition is given by the the phase shift at DC. $\endgroup$ – Marcus Müller Aug 29 '17 at 20:03
  • $\begingroup$ Isn't the +C constant only for indefinite integrals? This one has integration limits. Why would there be a +C? $\endgroup$ – VMMF Aug 30 '17 at 13:23
  • 1
    $\begingroup$ no, it's not just for indefinite integrals. It's just that limited integrals usually inherently cancel that $C$. $\endgroup$ – Marcus Müller Aug 31 '17 at 1:04
  • $\begingroup$ If after integrating, when it comes to evaluate in 0 and W (integration limits) I get C - C then in the end there's no C constant remaining right? $\endgroup$ – VMMF Jan 29 '18 at 4:33

Note that if $F(x)$ is an antiderivative of $f(x)$, i.e., if $F'(x)=f(x)$, then from the fundamental theorem of calculus we have


Since the phase $\phi(\omega)$ is an antiderivative of the negative group delay $-\tau(\omega)$ we get from $(1)$


With $b=\omega$ and $a=0$ we get from $(2)$


By adding $\phi(0)$ on both sides of $(3)$ we obtain the given formula:


| improve this answer | |

When integrating any expression, there is a constant term that is added. This represents an offset that may exist. For example, if I take the derivative of the function $y=x^2+3$. The derivative $\frac{dy}{dx} = 2x$. If I integrate $\frac{dy}{dx}$, then I end up with the expression $y=x^2+c$, where $c$ is an unknown constant. In your question, $arg[H(e^{j0})]$ is the phase offset at DC. This is the missing constant offset when you integrate the group delay. Without it you, you can only know the phase at some frequency $\omega1$, relative to some other frequency $\omega2$.

I'll see if I can clarify (hopefully not confuse) and answer the comment below. Let's look at a different problem. Let's say I want to integrate velocity $v$ over time to find out my position $p$ on the highway. I started at time 0 and drove for $\omega$ seconds. My position relative to my starting position is $p=\int\limits_{0}^\omega vdt$. The key here is relative to my starting position. If I don't know my position at time 0, I do not know my absolute position. I can tell you I travelled 10 miles, but I don't know where from and therefore where to.

The same is true for integrating the group delay to produce a phase. If we wanted phase offset from DC, the $arg[H(e^{j0})]$ wouldn't be necessary. If we want absolute phase, we must add it.

| improve this answer | |
  • $\begingroup$ Thank you very much for your answer. I know that for indefinite integrals (no integration intervals) after integrating a +C constant is added. But this integral has integration intervals between 0 a w. Why would there be a +C? $\endgroup$ – VMMF Aug 30 '17 at 13:20
  • $\begingroup$ @VMMF: because the real phase response is not necessarily zero at zero. $\endgroup$ – sellibitze Sep 30 '17 at 17:16

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.