A filter is a divider that listens to frequency
In the last guide we found that a capacitor's opposition to AC — its reactance — is not a fixed number. A capacitor's reactance Xc = 1 / (2 times pi times f times C) shrinks as the frequency f climbs, so to a fast signal a capacitor looks nearly like a short, while to a slow one it looks nearly like an open. Now put a resistor and a capacitor in series across a signal and tap the joint between them. You have built a voltage divider — exactly the share-the-voltage trick from the DC rung — except one leg's impedance depends on frequency. The divider ratio is no longer fixed; it changes with pitch. That is the whole idea of a filter.
Take the resistor on top and the capacitor on the bottom to ground, with the output read across the capacitor. At low frequencies the capacitor's reactance is huge, so it hogs almost the entire signal and the output nearly equals the input — low notes pass straight through. At high frequencies the capacitor's reactance collapses toward zero, shorting the fast wiggle to ground before it can reach the output. Low frequencies pass and high ones are shunted away: this is a low-pass filter. Swap which element is the small reactance and you simply change which frequencies win.
Swap the legs: high-pass, the corner, and the phase
Now flip the two parts: capacitor on top in series, resistor on the bottom to ground, output across the resistor. At low frequencies the capacitor's big reactance blocks the signal, so little reaches the output; at high frequencies the capacitor becomes a near-short and lets the signal pass to the resistor. High frequencies pass and low ones are blocked — a high-pass filter. Remarkably the corner sits at the very same f_c = 1 / (2 times pi times R times C); swapping the legs keeps the corner and only flips which side of it survives.
Let us put real numbers on it. Choose R = 1.6 k and C = 100 nF (which is 100 times 10^-9 farad). Then f_c = 1 / (2 times pi times 1600 times 100 times 10^-9) ≈ 1 / (1.0 times 10^-3) ≈ 1 kHz. As a low-pass this passes the bass and gently mutes everything above a kilohertz; as a high-pass (same parts, swapped) it does the reverse. Designing a first-order filter is really this small: pick the corner you want, choose a sensible R (a few k keeps currents tiny without being so large that stray pickup matters), then solve C = 1 / (2 times pi times R times f_c).
A filter does not only change amplitude — it also shifts timing. Because a capacitor's current leads its voltage, the output sine wave comes out nudged in phase relative to the input. Far inside the passband the shift is near zero; far into the stopband it approaches 90 degrees; and exactly at the corner it is 45 degrees, the halfway mark. This is why no filter can have a truly flat passband and an instant cutoff at the same time: amplitude and phase are linked, and bending one bends the other.
The decibel: measuring gain in ratios and logs
Filters and amplifiers span enormous ranges — an output might be a thousandth of the input here and a hundred times bigger there. To tame that range we measure the ratio on a logarithmic scale, the decibel (dB). For a voltage ratio, dB = 20 times log10(Vout / Vin); for a power ratio it is 10 times log10(Pout / Pin). The two differ by the factor of two because power goes as voltage squared, and log turns that square into a doubling. The lovely payoff: gains that you would multiply together become decibels you simply add.
A few magic numbers are worth memorizing. A ratio of 0.707 is -3 dB — and since 0.707 squared is 0.5, that is exactly half the power, which is why the corner is the "-3 dB point." Halving the voltage is -6 dB; one tenth is -20 dB; one hundredth is -40 dB. (When you see dBm, it is the same idea but power relative to a fixed 1 milliwatt, so it names an absolute level, not a ratio.) Be honest about what -3 dB means: the voltage is down to 0.707, not to half — RMS, average, and peak are different beasts, and so are voltage decibels and power decibels.
|Vout/Vin|
(dB)
0 +--------*. passband: flat, ~0 dB
| '.
-3 +.......... o <- corner at f_c (0.707, half power)
| '.
| '. slope = -20 dB / decade
-20 +................ '. <- one decade above f_c
| '. ( = -6 dB / octave )
-40 +....................'.
+----+-------+-------+-------+--> f (log scale)
0.1 f_c f_c 10 f_c 100 f_cPlot decibels against a logarithmic frequency axis and you get a Bode plot, the single most useful picture in analog design. A first-order RC filter shows a flat passband, a knee at f_c, and beyond it a dead-straight fall: its roll-off is 20 dB per decade (a tenfold change in frequency), equivalently 6 dB per octave (a doubling). The beauty is that you can sketch the whole response by hand from two straight asymptotes that meet at the corner — no calculation, just intuition.
Add an inductor: resonance and Q
An inductor is the capacitor's opposite twin. Its reactance XL = 2 times pi times f times L rises with frequency, exactly the mirror of the capacitor's falling Xc. Put the two in a loop and something special happens: there is one frequency where XL and Xc are equal in size but opposite in nature, so they cancel. The energy no longer just dissipates — it sloshes back and forth between the inductor's magnetic field and the capacitor's electric field, the electrical echo of a child pumping a swing at just the right rhythm. This is resonance.
Setting XL = Xc and solving gives the resonant frequency f_0 = 1 / (2 times pi times sqrt(L times C)). With L = 10 mH and C = 100 nF, f_0 = 1 / (2 times pi times sqrt(10 times 10^-3 times 100 times 10^-9)) ≈ 5 kHz. In a series R-L-C loop the two reactances cancel at f_0, leaving only the resistance, so the impedance dips to a minimum and current peaks — that selectivity makes a band-pass filter that favors one frequency. (Wire L and C in parallel instead and you get the dual: a high impedance at f_0, a "tank" that rejects that frequency.) This is exactly how a radio plucks one station out of the air.
How sharp is the peak? That is the quality factor, Q — the ratio of energy stored to energy lost each cycle, and equivalently Q = f_0 / bandwidth, where the bandwidth is the width between the two -3 dB points around the peak. A high Q means a tall, narrow, very selective resonance; a low Q means a broad, gentle bump. If our 5 kHz circuit has Q = 10, its bandwidth is 5000 / 10 = 500 Hz. Be honest about the limit: real Q is capped by losses — the resistance of the coil's wire, the capacitor's leakage — so a perfectly sharp, lossless resonance exists only on paper.
Honest limits, and filters in the wild
There is no such thing as a brick-wall filter. A first-order RC rolls off a gentle 20 dB per decade; to fall faster you cascade more stages or use resonance to build higher-order filters — but every extra bit of steepness is paid for. Push the passband flat and steep at once (a Butterworth-to-Chebyshev move) and ripple appears in the response; chase the sharpest amplitude cutoff and the phase distorts, so sharp-edged pulses ring and smear in time. Designs like Bessel deliberately give up amplitude sharpness to keep the timing clean. The engineering is always choosing which imperfection you can live with.
These passive R, L, and C filters are cheap and rugged, but they sag when a load draws current, they cannot provide gain, and good inductors are bulky and lossy — which is why later rungs rebuild filters around the op-amp as active filters. Yet the simplest filter of all is everywhere already: the decoupling capacitor sitting beside every chip is a high-pass-to-ground that shorts fast supply noise away while acting as a local water tank, delivering a gulp of current the instant the chip gets thirsty so the long supply pipe never sags. Just remember real parts are not ideal — a capacitor has a little series resistance and inductance of its own, so above some frequency it stops behaving like a capacitor at all.