Skip to content
Tutorials Center
My learning

Topics

Lesson 9 of 38 ยท Discrete Signals

The z-Transform Basics

Learn the z-transform, transfer functions H(z), poles and zeros, stability, and how the frequency response comes from the unit circle, with Python checks against SciPy.

Convolution and difference equations describe a system in the time domain. The z-transform turns both into algebra. Convolution becomes multiplication, a delay becomes multiplication by z^-1, and a filter becomes a ratio of two polynomials. From that ratio you can read off the stability and the frequency response by looking at where its roots lie.

Definition

The z-transform of a sequence x[n] is

X(z) = sum over n of x[n] * z^(-n)

where z is a complex variable. Each sample becomes a coefficient of a power of z^-1. For the finite sequence [2, -1, 4] at n = 0, 1, 2, we get X(z) = 2 - z^-1 + 4 z^-2.

Three rules do most of the work:

Operation in time Effect on X(z)
Delay x[n-k] multiply by z^-k
Sum a x1 + b x2 a X1(z) + b X2(z)
Convolution x * h X(z) * H(z)

A few useful pairs: the impulse d[n] has transform 1; the one-sided exponential a^n u[n] has transform 1 / (1 - a z^-1) (valid for |z| > |a|); the delay d[n-k] has z^-k.

The transfer function

Apply the rules to the general difference equation from the previous lesson. With Y(z) = H(z) X(z):

H(z) = (b0 + b1 z^-1 + ... + bM z^-M) / (1 + a1 z^-1 + ... + aN z^-N)

So H(z) = B(z) / A(z), built directly from the coefficient arrays b and a that you pass to lfilter. The inverse transform of H(z) is the impulse response h[n].

flowchart LR
    A["difference equation<br/>coefficients b, a"] --> B["H(z) = B(z) / A(z)"]
    B --> C["poles and zeros"]
    B --> D["frequency response<br/>z = e^(jw)"]
    B --> E["impulse response h[n]"]

One compact object gives the stability, the frequency response and the impulse response.

Python Polynomial multiplication is convolution

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Poles and zeros

Factor the numerator and denominator:

H(z) = b0 * (1 - q1 z^-1)(1 - q2 z^-1)... / ((1 - p1 z^-1)(1 - p2 z^-1)...)

  • The zeros q are values of z where H(z) = 0: the filter blocks these.
  • The poles p are where H(z) becomes infinite: they create resonance and memory.

NumPy finds them as polynomial roots. Note that np.roots expects coefficients in powers of z (not z^-1), so for H(z) with equal-length numerator and denominator the arrays b and a can be passed directly, and for different lengths pad with zeros to the same length first.

Python Poles and zeros of a filter

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Stability

A causal LTI system is stable if and only if all poles lie strictly inside the unit circle (|p| < 1). Zeros can be anywhere. A pole on the circle gives a sustained oscillation (marginal), and outside the circle the response grows without bound. FIR filters have poles only at z = 0, so they are always stable.

Python Check stability from the poles

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Frequency response from the z-plane

The z-transform evaluated on the unit circle z = e^(j*w) is the frequency response H(e^(jw)). As w runs from 0 to pi we walk along the upper half of the circle from the frequency 0 point z = 1 to the Nyquist point z = -1. A pole near the circle makes the response large nearby; a zero on the circle makes it exactly zero there.

The magnitude |H| is the gain at each frequency, and the angle of H is the phase shift. Computing it by hand is just plugging z^-1 = e^(-jw) into the polynomials.

Python Frequency response by hand versus freqz

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

scipy.signal.freqz(b, a, worN) does the same and is what you will use. With fs given (freqz(b, a, fs=1000)), the frequency axis comes back in hertz.

Poles, zeros and the shape of the response

A geometric rule helps design: the gain at a frequency w is proportional to the product of distances from e^(jw) to the zeros divided by the product of distances to the poles. So:

  • Put a zero at a frequency you want to kill (a hum at 50 Hz).
  • Put a pole just inside the circle at a frequency you want to boost (a resonance).
  • Keep every pole inside the circle.
Python A notch filter from one pair of zeros and poles

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The gain at 50 Hz falls to a very large negative number (a deep notch), while the rest of the spectrum is close to 0 dB because the pole sits next to the zero and cancels its effect at other frequencies.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. A delay of k samples corresponds to which operation on X(z)?

    x[n-k] has transform z^-k X(z); that is why z^-1 is drawn as a delay block.
  2. 2. When is a causal LTI system with transfer function B(z)/A(z) stable?

    Poles with magnitude below 1 give decaying terms p^n; zeros do not matter for stability.
  3. 3. The frequency response is the z-transform evaluated at:

    Setting z = e^(jw) gives the gain and phase for a sinusoid of digital frequency w.
  4. 4. Where must you put a zero to completely block a frequency w0?

    H(e^(jw0)) = 0 only if the numerator vanishes at that point on the circle.

Summary

  • X(z) = sum x[n] z^-n; a delay multiplies by z^-1, and convolution becomes multiplication.
  • H(z) = B(z)/A(z) comes straight from the difference-equation coefficients.
  • Zeros block frequencies; poles create resonance; stability needs all poles inside the unit circle.
  • The frequency response is H(z) at z = e^(jw), computed by scipy.signal.freqz.
  • Pole-zero placement is a practical design tool (a notch filter needs one zero pair and one pole pair).

Next: Fourier Series and Complex Exponentials