My understanding of BIBO stability is that if a bounded input is applied to a discrete-time, linear, time-invariant BIBO system, the output will also be bounded. A sufficient condition for this is that the poles of the transfer function are within the unit circle.

Suppose I have a system with the transfer function $$ H(z)={(1−2z^{−1})(1−{\frac{1}{6}} z^{−1}) \over (1−\frac{1}{2}z^{−1})(1−\frac{1}{3}z^{−1})} $$ The poles of this system are within the unit circle and so is a BIBO stable system.

If I apply an input of $x[n]=(\frac{1}{3})^n$, which is bounded, the output would involve $H(\frac{1}{3})=\infty$, which is unbounded - i.e. a bounded input produced an unbounded output.

Obviously, there is something amiss in my understanding here, but what?


The signal $x[n] = (\frac{1}{3})^n$ is not bounded. As $n \to -\infty$, $x[n] \to \infty$; therefore, the system's response to your candidate signal $x[n]$ has no bearing on its BIBO stability.

A bounded, causal version of the signal (and likely what you were thinking) is $x_b[n] = (\frac{1}{3})^n u[n]$, where $u[n]$ is the unit step function. The response of the system to this signal is bounded as you would expect.

  • $\begingroup$ D'oh! D'oh! D'oh!!!!! $\endgroup$
    – Westerley
    Aug 11 '15 at 19:23
  • $\begingroup$ Also, you don't get the response of the system to $x[n]$ by evaluating $H(z)$ at some value of $z$... instead, you take the z-transform of $x[n]$, multiply it with $H(z)$, and call the result $Y(z)$. Now, you can do an inverse transform to get $y[n]$, which you will see is bounded. $\endgroup$ Aug 11 '15 at 21:53
  • $\begingroup$ I understood from this question that the output of a system when the input is a 'pure' exponential such as $\frac{1}{3}^n$ would involve evaluating $H(\frac{1}{3})$. If the input was $\frac{1}{3}^nu[n]$, then it would be as you say $\endgroup$
    – Westerley
    Aug 12 '15 at 1:28
  • $\begingroup$ @CMDoolittle: The OP is correct; the $z$ transform can be used in this way since exponential functions are the eigenfunctions of discrete-time LTI systems. The value $H(z)$ corresponds to the eigenvalue associated with the function $x[n] = z^{-n}$. $\endgroup$
    – Jason R
    Aug 12 '15 at 1:32
  • $\begingroup$ Wow. I did not know that. Thanks y'all! $\endgroup$ Aug 12 '15 at 3:28

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.