2

Suppose $\mathcal{L}$ be invertible system with memory. Does $\mathcal{L}^{-1}$ have memory necessarily?

Intuitively I think the answer is "yes". There are many examples showing that. For instance $\mathcal{L}(x(t)) = x(t-2)$ and $\mathcal{L}(x(t)) = x(\frac t 3)$. Another example which seems problematic to me is $$\mathcal{L}(x(t)) = \int_{-\infty}^{t}x(\lambda)d\lambda$$The inverse is $$\mathcal{L}^{-1}(x(t)) = \frac{dx(t)}{dt}$$Does differentiator have memory? Of course the main question here is about memory of an invertible system which has memory. Note that here $\mathcal{L}$ can be nonlinear as well.

For clarity, I add some related definitions from Oppenheim's book:

Invertible system: A system is said to be invertible if distinct inputs lead to distinct outputs.

Causal system: A system is causal if the output at any time depends only on values of the input at the present time and in the past.

Memoryless system: A system is said to be memoryless if its output for each value of the independent variable at a given time is dependent only on the input at that same time.

S.H.W
  • 726
  • 2
  • 7
  • 20
  • Regarding the memory of the differentiator, see https://dsp.stackexchange.com/questions/58533/is-the-first-derivative-operation-on-a-signal-a-causal-system – MBaz Oct 09 '20 at 18:21
  • The inverse of a time delay is a time advance, right? Does the time advance have memory? – MBaz Oct 09 '20 at 18:27
  • @MBaz Thanks. I've seen that. My question is more general. Also there are many answers in that link which really confuses me. – S.H.W Oct 09 '20 at 18:33
  • @MBaz "A system is said to be memoryless if its output for each value of the independent variable at a given time is dependent only on the input at that same time." This is the definition which I'm using. So time advance have memory. – S.H.W Oct 09 '20 at 18:35
  • Well, I don't agree with that definition. Seeing into the future is not "memory", it is "non-causality". – MBaz Oct 09 '20 at 18:37
  • @MBaz Actually it's the definition which Oppenheim's book gives. I agree it seems unnatural. – S.H.W Oct 09 '20 at 18:40
  • Yeah, it's not an uncommon definition. I still disagree with it. – MBaz Oct 09 '20 at 22:47
  • I assume that invertible also means that both are causal? So I think you would be limited to transfer functions with the same order in the numerator as denominator. – fibonatic Oct 11 '20 at 12:15
  • @fibonatic I've added some details. Please take a look at it. – S.H.W Oct 11 '20 at 12:27

1 Answers1

1

For time-based systems, I understand that it is difficult to imagine a memory of the future. But for general systems, $-t$ and $t$ are just left and right (think of a spatial system). Other discussions are in LTI system $y(t)=x(t−T)$ with or without memory, What is a memory less system?, or A question about the concept of the time.

By definition of invertibility, $\mathcal{L}^{-1}$ is such that $\mathcal{L}^{-1}( \mathcal{L}(x))=x$. But also that $\mathcal{L}( \mathcal{L}^{-1}(x))=x$ (by the way, derivatives and integrals are not inverses). Let us suppose the converse: $ \mathcal{L}^{-1}$ has no memory. Hence $\mathcal{L}^{-1}(x[n]))$ can only use the present state, and $\mathcal{L}$ as well to yield $(x[n])$.

So, if $\mathcal{L}^{-1}$ is memoryless, $\mathcal{L}$ is memoryless as well. By contraposition, the converve is true

Laurent Duval
  • 31,850
  • 3
  • 33
  • 101
  • Thanks. I don't understand why $\mathcal{L}$ has to have memory in order to $\mathcal{L}( \mathcal{L}^{-1}(x[n]))=x[n]$ holds. Would you elaborate, please? – S.H.W Oct 11 '20 at 20:59
  • I used an argument based on logic. It is not constructive in the common sense. I suppose the converse. It entails that the initial hypothesis on $\mathcal{L}$ cannot be verified. Hence, my initial supposition is false – Laurent Duval Oct 12 '20 at 21:30
  • 1
    I see. I don't understand the part "and $\mathcal{L}$ as well". Why $ \mathcal{L} $ can only use the present state to yield $x[n]$? Maybe it uses future or past values of the input as well and still yields $x[n]$. – S.H.W Oct 12 '20 at 21:39
  • Because (from my hypothesis) $\mathcal{L}^{-1}$ has no memory. So $\mathcal{L}^{-1}(x)$ can only use the present state. Hence, $\mathcal{L}$ is only given sometimes related to the present state – Laurent Duval Oct 12 '20 at 21:46
  • It's a real shame for me that I don't get this part of your answer still. Anyway, thank you so much. – S.H.W Oct 12 '20 at 23:47