Skip to content
Tutorials Center
My learning

Topics

Lesson 20 of 36 · Fourier Transform

Properties of the Fourier Transform

Master Fourier transform properties: linearity, time shift, scaling, modulation, differentiation and the convolution theorem, each checked numerically in Python.

You rarely compute a Fourier transform from the integral. Instead you start from a handful of known pairs and apply properties that turn operations in time into simple operations in frequency. A delay becomes a phase factor, differentiation becomes multiplication by j2πf, and convolution becomes a product. These rules make the transform the main tool for analyzing LTI systems.

Below, x(t) ↔ X(f) means "x has transform X". The properties are proven for the continuous transform, but the discrete Fourier transform (DFT) obeys exact analogues, so you can verify each of them with NumPy's fft. That is what the examples do.

The core properties

Property Time domain Frequency domain
Linearity a x(t) + b y(t) a X(f) + b Y(f)
Time shift x(t - t0) X(f) e^(-j2πf t0)
Frequency shift (modulation) x(t) e^(j2πf0 t) X(f - f0)
Time scaling x(at) (1/|a|) X(f/a)
Time reversal x(-t) X(-f)
Differentiation dx/dt j2πf X(f)
Integration integral of x X(f) / (j2πf) + (X(0)/2) δ(f)
Convolution x(t) * y(t) X(f) Y(f)
Multiplication x(t) y(t) X(f) * Y(f)
Duality X(t) x(-f)

Three of these deserve special attention.

Time shift. Delaying a signal does not change which frequencies it contains, so the magnitude |X(f)| is unchanged. Only the phase changes, by a term that is linear in f. The slope of that line is the delay.

Time scaling. Compressing a signal in time (a > 1) stretches its spectrum, and the amplitude drops to keep the energy balance. This is the algebraic form of the inverse relationship between duration and bandwidth.

Convolution. The convolution integral is expensive, but its transform is a plain product. This is why filtering is usually thought of in the frequency domain: output spectrum = input spectrum times the system's frequency response.

Checking shift, modulation and reversal

For the DFT of length N the shift is circular, and a frequency shift of m bins corresponds to multiplying by e^(j2πmn/N).

Python Linearity, shift, modulation, reversal

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

Program input · 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 convolution theorem and differentiation

To convolve two finite sequences with FFTs, zero-pad both to length len(a) + len(b) - 1 first; otherwise you get circular convolution. For differentiation, multiply the spectrum by j2πf and transform back. For smooth signals sampled finely enough, this "spectral derivative" is extremely accurate.

Python Convolution theorem and spectral derivative

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

Program input · 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.

Scaling in action

For the one-sided exponential e^(-at) u(t) with transform 1/(a + j2πf), the scaling rule says that x(2t) = e^(-2at) u(t) has transform (1/2) · 1/(a + jπf). You can confirm it by writing the transform of e^(-2at) u(t) directly as 1/(2a + j2πf) and noting that it equals (1/2) · 1/(a + jπf). Both routes give the same answer, which is a useful sanity check when you doubt a factor of 2.

When a problem looks hard, try to express the signal as a combination of known pairs using linearity, shifts and scaling. Computing the integral directly should be your last resort.

Common mistakes

  • Treating a delayed signal as having a different magnitude spectrum. It does not.
  • Forgetting the 1/|a| factor in time scaling.
  • Using FFT products without zero-padding and then wondering why the result wraps around.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. What does delaying a signal by t0 do to its Fourier transform?

    A time delay multiplies the spectrum by a unit-magnitude phase factor, so |X(f)| is unchanged while the phase gets a linear term.
  2. 2. Differentiating x(t) corresponds to what in the frequency domain?

    dx/dt maps to j2πf X(f). This is why differentiation boosts high frequencies and why noise gets amplified.
  3. 3. Which operation in time becomes a simple product in frequency?

    The convolution theorem says x(t) * y(t) maps to X(f) Y(f). Multiplication in time maps to convolution in frequency.
  4. 4. If x(t) is compressed to x(2t), its spectrum is...

    Time scaling gives (1/|a|) X(f/a). With a = 2 the spectrum is twice as wide and half as tall.

Summary

  • Linearity, shifting, scaling, reversal, differentiation and convolution each have a clean frequency-domain counterpart.
  • A time shift only changes the phase; a frequency shift is modulation.
  • Convolution in time is multiplication in frequency, which makes FFT-based filtering fast.
  • The DFT obeys the same rules (with circular shifts), so NumPy can verify them.

Next: Transform Pairs and Parseval's Theorem