OE-8.4 FIR Filter Design

What this is and why it exists

This filter family has one selling point that decides most of the cases where it is chosen: exact linear phase.

Linear phase means every frequency is delayed by the same amount. So the waveform's shape survives the filter — it comes out delayed, not distorted. In data transmission and image processing that is not a refinement, it is the requirement.

The window method is then a controlled compromise, and knowing which part of the window controls which part of the result is the whole skill.

The vocabulary

• Finite impulse response — the response to an impulse ends after a fixed number of samples.
• Linear phase — delay is the same at every frequency.
• Group delay — the actual delay the filter imposes.
• Symmetry — the coefficient pattern that guarantees linear phase.
• Window — a shaping function applied to an ideal response.
• Main lobe — the central part of the window's own spectrum.
• Side lobe — the smaller surrounding parts.
• Transition width — how quickly the filter moves from passing to blocking.
• Stop-band attenuation — how much the blocked band is suppressed.

The mental model

Start with why linear phase matters, because it is often stated and rarely explained. Delay different frequencies by different amounts and the components of a waveform arrive out of their original alignment. The shape changes, even though every frequency is present in the right amount. A square pulse comes out rounded and ringing. For a filter whose job is to clean a signal without altering what it looks like, that is a failure.

A filter of this family achieves exact linear phase from one condition. Its coefficients are symmetric. That is a structural guarantee rather than an approximation, and it is the family's real advantage.

The design method is then honest about being a compromise. The ideal frequency response you want corresponds to an impulse response that is infinitely long, so it cannot be built. Truncating it abruptly causes ripples. A window tapers the truncation instead.

Here is the part worth memorising, because it turns design from guessing into choosing. The window's main lobe width sets the transition width. Its side lobe height sets the stop-band attenuation. Those are two separate knobs on two separate outcomes. A wider main lobe gives a lazier transition; lower side lobes give deeper suppression; and for a given window, improving one worsens the other. Choosing a window is choosing where on that trade you sit, and lengthening the filter is what buys you a better position on both.

The four shapes are low-pass, high-pass, band-pass and band-stop. All come from the same procedure with a different ideal response, so learning the low-pass case properly gives you the rest.

And the comparison with the other family: this one gives guaranteed stability and exact linear phase, and needs more coefficients for the same sharpness. The other is more efficient and offers neither guarantee.

What you should now be able to explain or do

Explain why linear phase preserves waveform shape, and give a case where that is required. State the coefficient condition that guarantees it. Say why the ideal response cannot be built and what a window does about it. Name which window property sets transition width and which sets stop-band attenuation. Design all four filter shapes from the same procedure. Justify choosing this family over the alternative.

Check yourself