EC-4.3 Fourier Series, and Why a Periodic Signal Decomposes

What this is and why it exists

A periodic waveform can be written as a sum of sinusoids at multiples of one fundamental frequency. That is usually presented as a result to accept, which hides the one property that makes it work.

The property is orthogonality. Sinusoids at different multiples of the same fundamental integrate to zero over a period when multiplied together. Because of that, each coefficient can be extracted by a single integral without disturbing any of the others. Without it you would have to solve for every coefficient at once, and there would be no formula.

The series also matters because it is where a periodic signal stops being a shape and becomes a set of frequencies. That change of view is what makes filtering, modulation and spectral analysis thinkable at all.

The vocabulary

• Fundamental frequency — one divided by the period. The lowest frequency present.
• Harmonic — a sinusoid at a whole-number multiple of the fundamental.
• Orthogonality — the property that two different harmonics multiply and integrate to zero over one period.
• Fourier coefficient — the amount of one particular harmonic present in the waveform.
• Exponential form — the series written with complex exponentials rather than sines and cosines.
• Line spectrum — a plot of coefficient magnitude against frequency, showing isolated lines.
• Gibbs overshoot — the fixed percentage overshoot that persists beside a discontinuity however many terms are added.
• Half-wave symmetry — the property of a waveform that repeats inverted after half a period.

The mental model

Here is the mechanism, and it is the only thing in the topic worth committing to memory.

To find how much of a particular harmonic is present, multiply the waveform by that harmonic and integrate over one period. Every other harmonic contributes zero to that integral, because different harmonics are orthogonal. Only the one you multiplied by survives, and what comes out is its coefficient.

That is why there is a formula for each coefficient separately, and why finding the fiftieth does not require finding the first forty-nine.

The series can be written two ways. The sine-cosine form has two coefficients per frequency. The exponential form has one complex coefficient per frequency, where the magnitude carries the amount and the angle carries the phase. They hold identical information, but the exponential form makes shifting and scaling far shorter to write, and later subjects use it almost exclusively. Get comfortable with it now.

Two waveforms are worth knowing by heart because they appear constantly. A square wave has only odd harmonics, and their amplitudes fall as one over the harmonic number. A sawtooth has all harmonics, also falling as one over the harmonic number. Derive both once. After that, recognising them saves real time.

Symmetry is the practical shortcut. An even waveform has no sine terms. An odd waveform has no cosine terms. A waveform with half-wave symmetry has no even harmonics at all. Checking symmetry before integrating often removes half the work or more, and it is the first thing an experienced person does.

Convergence deserves a warning. Adding more terms improves the fit everywhere except immediately beside a jump. There, a fixed overshoot of about nine percent persists no matter how many terms you add. It merely gets narrower, never smaller. This is a genuine property of the series and not a numerical error. It reappears in filter design, where a sharp cutoff produces ripple for the same reason.

Finally, the picture. Plot coefficient magnitude against frequency and you get isolated lines at multiples of the fundamental, with nothing between them. A periodic signal occupies discrete frequencies only. That discreteness is exactly what the next topic gives up when the period is allowed to grow without bound.

What you should now be able to explain or do

• Explain why orthogonality lets each coefficient be found by one integral.
• Write the series in both forms and say why the exponential one is preferred.
• Compute the coefficients for a square wave and a sawtooth.
• Use symmetry to say which coefficients vanish before integrating.
• Say what the Gibbs overshoot is and why more terms do not remove it.

Check yourself