Is $y[k] = y[k-1] + x[k]$ an integrator?

Question

It looks exactly like an integrator to me. Since $$y[k] = y[k-1]+x[k] = y[k-2] + x[k-1] +x[k] = \sum{x}$$

Applying the Z-transform gives \begin{align} Y(z) &= Y(z)\cdot z^{-1} + X(z)\\ \Rightarrow\frac{Y(z)}{X(z)} &= \frac{1}{1-z^{-1}} \end{align}

When I convert $\frac{1}{1-z^{-1}}$ into Lapace transform in Matlab using d2c, it does not return $\frac{k}{s}$. Instead, it returns $\frac{1+s}{s}$. It seems there is a proportional term inside, when thinking in PID control point of view.

Could anyone explain on that?

The system as a pole at z = 1 so it's unstable and, strictly speaking, the Z transform isn't the right tool since the region of convergences doesn't include the entire unit circle — Hilmar, Apr 01 '16 at 13:59
@Gilles: Thanks for the edit; I just corrected the first occurrence of "Laplace transform" to "Z-transform". — Matt L., Apr 01 '16 at 18:08

Matt L. · Accepted Answer · 2016-04-03T10:57:00.563

9

The system

$$y[n]=y[n-1]+x[n]\tag{1}$$

is an ideal accumulator, i.e., it computes the cumulative sum of the input samples:

$$y[n]=\sum_{k=-\infty}^nx[k]\tag{2}$$

It is in a way analogous to a continuous-time integrator, but this doesn't mean that you will necessarily obtain an ideal integrator by transforming the discrete-time system to a continuous-time system. There are several methods to do the conversion, and I'm not sure which one you used when calling d2c.

In any case, the properties of the transformed continuous-time system not only reflect the properties of the original discrete-time system, but also the properties of the transformation. So the properties of a discrete-time system should be investigated directly in the discrete-time domain, not by transforming it to a continuous-time system.

edited Apr 03 '16 at 10:57

answered Apr 01 '16 at 13:08

Matt L.

78,282
5
66
149

Thanks for your reply. It seems that any method for d2c does not give a 1/s. – richieqianle Apr 01 '16 at 14:15
Actually this is an intermediate problem of mine and is not the final one. My real problem is actually to design a discrete controller which is to be transformed from s domain. I am really thinking if I shall use $y[k] = y[k-1] + x[k]$ or $y[k] = y[k-1] + x[k-1]$ or $y[k] = y[k-1] + 0.5*x[k-1]+0.5*x[k]$ for $1/s$ part of the controller. – richieqianle Apr 01 '16 at 14:16
Is "when calling `d2m`" a typo and should be "when calling `d2c`" instead? – Matthias W. Apr 03 '16 at 10:40
1

@MatthiasW.: That was indeed a typo, thanks. – Matt L. Apr 03 '16 at 10:57

score 6 · Answer 2 · answered Apr 03 '16 at 06:14

6

Your simple integrator is called a "Rectangular Rule" integrator. There are more complicated (and more accurate) integrators called "Trapazoidal Rule", "Simpson's Rule", and "Tick's Rule" integrators.

answered Apr 03 '16 at 06:14

Richard Lyons

4,305
11
24

It should be mentioned that the choice of integrator's depends upon presumptions of the behavior between samples; that is input characteristics. – rrogers Apr 06 '16 at 16:31

score 1 · Answer 3 · answered Apr 01 '16 at 14:09

1

Yes, the system should be an integrator.

What method did you use in the call to d2c.

answered Apr 01 '16 at 14:09

Peter K.

21,266
9
40
78

Thanks for your reply. It seems that any method for d2c does not give a 1/s. – richieqianle Apr 01 '16 at 14:12

Is $y[k] = y[k-1] + x[k]$ an integrator?

3 Answers3