EC-4.4 The Fourier Transform

What this is and why it exists

The Fourier transform is the series with the period allowed to grow without bound. Watching that limit happen is the difference between understanding the transform and merely using it.

It matters because most real signals are not periodic. A pulse, a burst of speech, a step: none of them repeat, so the series has nothing to work with. The transform extends the same idea to signals that happen once.

And it matters for a second reason that turns out to be the important one. Convolution in one domain becomes multiplication in the other. That single property is why the transform is central to engineering rather than merely elegant.

The vocabulary

• Transform pair — the forward transform and its inverse, taken together.
• Spectrum — the transform of a signal, viewed as a function of frequency.
• Continuous spectrum — a spectrum with content at every frequency, rather than at isolated lines.
• Sinc function — the sine of x divided by x. The transform of a rectangle.
• Impulse — an idealised spike of zero width and unit area.
• Convolution — an operation that slides one function across another, multiplying and integrating at each shift.
• Parseval's relation — the statement that total energy is the same computed in either domain.
• Duality — the symmetry between the forward and inverse transforms, which lets one result give two.

The mental model

Start with the limit. Take a periodic signal and lengthen its period. The fundamental frequency falls, so the harmonics crowd closer together. Each individual coefficient shrinks, because the same energy is spread over more of them. Let the period grow without bound and the lines merge into a continuous curve, while the sum becomes an integral.

The line spectrum has become a continuous spectrum. That is the whole conceptual step, and once you have seen it the transform pair looks inevitable rather than invented.

A warning before the properties. Where the factor of two pi sits, and whether frequency is measured in radians per second or in cycles per second, differs between textbooks. A formula copied from one source into a calculation set up from another can be wrong by a constant. Fix one convention, state it, and stay with it.

Now the properties, which get used far more often than the defining integral.

Linearity means the transform of a sum is the sum of the transforms. Shifting in time multiplies the transform by a complex exponential, so the magnitude is untouched and only the phase changes. That is worth pausing on: moving a signal in time does not change which frequencies it contains, only their relative alignment. Scaling compresses one domain and stretches the other by the same factor. A shorter pulse has a wider spectrum, always, and no amount of design escapes it.

Three standard pairs cover most of practice. A rectangle transforms to a sinc. An impulse transforms to a constant, meaning it contains every frequency equally. A Gaussian transforms to another Gaussian, which is unusual enough to be memorable.

Convolution in one domain is multiplication in the other. An awkward sliding integral becomes an ordinary product. This is why the transform earns its place in engineering. The reading of it in terms of systems is deliberately not taken here. That reading, which explains what a filter does to a signal, belongs to the signals and systems subject.

Parseval's relation closes the topic. Total energy computed by integrating the squared signal equals the energy computed from the squared spectrum, up to the convention's constant. It is both a physical statement, energy does not depend on how you count it, and a useful arithmetic check on work done by hand.

What you should now be able to explain or do

• Explain how the series becomes the transform as the period grows without bound.
• State the shifting, scaling and linearity properties and say what each does to a spectrum.
• Recall the transforms of a rectangle, an impulse and a Gaussian.
• State the convolution property and say why it makes the transform useful.
• Use Parseval's relation as a check on a computed spectrum.

Check yourself