A divider that listens to frequency
You already own the one idea a filter is built on. In the AC rung you saw that a capacitor does not fight a signal with a fixed resistance; it fights with reactance, an opposition that shrinks as the frequency rises. A capacitor is nearly an open door to fast wiggles and nearly a brick wall to slow ones. Now put that frequency-sensitive part into the simplest circuit you know — a voltage divider — and something wonderful happens: the divider no longer splits every signal the same way. It splits by frequency.
Picture a resistor on top and a capacitor on the bottom, with the output taken across the capacitor. At low frequency the capacitor's reactance is huge, so it grabs almost all the voltage and the output is full-sized. At high frequency its reactance has collapsed to nearly zero, so it keeps almost none and the output withers away. Swap the two parts — capacitor on top, resistor on the bottom — and the behaviour flips: lows are killed, highs sail through. That is the whole secret. A filter is just a divider whose two arms disagree about frequency.
One thing this little divider does not do is amplify. At its very best a simple passive filter only passes signals it likes and attenuates the rest — think of it less as a megaphone and more as a doorman who never makes the music louder, only decides which frequencies get in. (An active filter can add gain with an op-amp, but the selecting underneath is still pure attenuation, and we save that distinction for guide 2.)
The four shapes a filter can take
Every filter, no matter how fancy, sorts the frequency axis into a region it keeps and a region it throws away. There are exactly four useful ways to draw that line. A low-pass keeps everything below a chosen frequency and rejects what is above — the RC divider we just built. A high-pass does the mirror opposite: highs through, lows out. A band-pass keeps only a band in the middle and rejects both far ends, like a radio tuned to one station. And a band-stop is its inverse, punching a narrow hole in an otherwise open response.
RESPONSE passes ... blocks ... a job it does
----------- ----------------------- --------------------- -----------------------
low-pass below the cutoff above the cutoff smooth a DAC, anti-alias
high-pass above the cutoff below the cutoff block DC offset / rumble
band-pass a band in the middle both far ends tune one radio channel
band-stop both far ends a narrow band kill 50/60 Hz mains hum
amplitude
^ low-pass high-pass band-pass band-stop (notch)
| ----\ /---- /--\ ----\ /----
| \ / / \ \/
+------ f --> ---- f --> ---- f --> ---- f -->The narrow band-stop has its own nickname — the notch — and it earns its keep removing one specific pest, most famously the 50 Hz or 60 Hz hum that the power grid sloshes into sensitive measurements. A band-pass and a notch are really just teamwork: stack a low-pass and a high-pass and where their passbands overlap you get a band-pass; where their *stop*bands overlap you get a notch. Two shapes you already understand combine into the other two.
Passband, stopband, and the -3 dB cutoff
Now the vocabulary, because filter people speak a precise dialect. The passband is the stretch of frequencies the filter lets through at roughly full strength. The stopband is where it shoves signals down to nearly nothing. Between them is no sudden wall but a sloping transition band — and this is the first honest truth of the subject: real filters do not switch from on to off at a single frequency. They fade.
So where exactly does the passband end? By universal convention, at the cutoff frequency (also called the corner frequency): the point where the output has fallen to 0.707 of its passband value, which is exactly -3 dB on the decibel scale. Why that oddly specific number? Because 0.707 is 1 over the square root of 2, and power goes as voltage squared, so 0.707 in voltage is half the power. The cutoff is the half-power point: the frequency at which the filter is letting through exactly 50 percent of the signal's power. In the RC low-pass it is precisely where the capacitor's reactance equals the resistance, so the two divider arms are evenly matched.
For that RC low-pass the cutoff sits at f = 1 / (2 times pi times R times C). Pick R = 1.6 k and C = 100 nF and you get f = 1 / (2 times pi times 1600 times 100 times 10^-9) which works out to about 995 Hz — call it 1 kHz. Below 1 kHz the signal passes nearly untouched; right at 1 kHz it is down to 0.707 (-3 dB); above it, it keeps sliding away. Designing the filter is really just choosing R and C so that this corner lands where you want it.
- Decide the cutoff you need. Say you want to pass speech but trim hiss, so put the corner at f = 1 kHz.
- Choose a capacitor in a sane range (about 1 nF to 1 uF — too small and stray capacitance dominates, too large and the resistor must be tiny). Pick C = 100 nF.
- Solve for R: R = 1 / (2 times pi times f times C) = 1 / (2 times pi times 1000 times 100 times 10^-9) which is about 1592 ohm, so reach for a standard 1.6 k.
- Sanity-check loading. The thing driving the filter should be much stiffer than 1.6 k, and the thing it feeds much higher, or the corner shifts. When that is impossible, a buffer (the op-amp follower from the last rung) fixes it.
Roll-off, order, and the no-brick-wall truth
Past the cutoff, how fast does the response fall? That slope is the roll-off, and it is the headline number of any filter. The plain RC low-pass rolls off at 20 dB per decade — that is, for every tenfold rise in frequency the output drops by 20 dB, a factor of 10 in amplitude (the same slope can be quoted as 6 dB per octave, per doubling). So one decade above our 1 kHz corner, at 10 kHz, the signal is down about 20 dB to a tenth of its size; at 100 kHz, down 40 dB to a hundredth. Steady, but gentle.
Gentle is often not enough. To get a steeper cliff you raise the filter's order: each order adds another 20 dB per decade to the slope. A first-order filter (one R and one C) gives 20 dB/decade; a second-order gives 40; a fourth-order gives a brisk 80 dB/decade. You build high orders by cascading simple sections, the way you stack lenses to sharpen an image — and the practical workhorse section, the second-order Sallen-Key built around an op-amp, is the whole subject of guide 3.
Going second-order unlocks a new dial: Q, the quality factor, which sets how the response behaves right at the corner. Its flip side is damping, and the two trade off — high Q means light damping. A low-Q section eases smoothly into the stopband; a high-Q one peaks up sharply near the cutoff before diving, the same resonance ring you met with an RLC loop. Choose Q too high and the filter overshoots and rings on a fast signal; too low and the corner sags early. That single dial is what separates the famous filter 'flavours' — Butterworth, Chebyshev, Bessel — that guide 4 lays out.
Where filters earn their keep
Filters are everywhere a circuit needs to care about which frequencies survive. In audio, a crossover is a low-pass and a high-pass splitting the music so the woofer gets the bass and the tweeter the treble; the tone and bass knobs on an amplifier are just adjustable filters. In a power supply, the reservoir-plus-resistor smoothing you met earlier is a low-pass killing ripple. And in almost any instrument, a notch sits ready to strangle mains hum before it swamps a faint signal.
The two jobs that brush right up against the digital world come at the end of this ladder. Before any signal is digitised, an anti-alias filter — a low-pass — must strip away everything above half the sampling rate, because any higher frequency would fold back and masquerade as a false low one (this is aliasing, and the half-rate limit is the Nyquist line you will meet in the data-converter rungs). After a digital signal is turned back into voltage, a DAC needs a 'reconstruction' low-pass to smooth its jagged staircase output into a clean waveform. Same humble low-pass, two indispensable jobs.