Three doorways for noise
By now you know where noise comes from — the intrinsic hiss of thermal noise from guide 1, plus all the external mess of motors, clocks and radios — and you know how to keep your ground from turning that mess into hum with a ground loop from guide 3. This guide is about the journey in between: once noise exists on one wire, how does it climb aboard a neighbouring wire that it is not even connected to? There are only three doorways, and once you can name the door, you know which cure to reach for.
The first doorway is conducted coupling: the noise simply shares a wire with you — a common supply rail, a shared ground return — and rides in on the conductor itself. That is largely what guides 2 and 3 fought, with the decoupling capacitor (the local water tank that keeps the long supply pipe from sagging) and with careful grounding. The other two doorways are sneakier, because no copper joins the two circuits at all. The noise leaps the gap through a field — either an electric field (capacitive coupling) or a magnetic field (inductive coupling). Those two field handshakes are the heart of this guide.
Capacitive coupling: the electric handshake
Any two pieces of metal with a gap between them form a tiny capacitor whether you drew one or not — this is parasitic capacitance, and on a crowded board it is everywhere. When a nearby 'aggressor' wire swings its voltage, that stray capacitance does exactly what a capacitor always does: it passes a current proportional to how fast the voltage changes. The current that leaks across is i = Cm times dV/dt, where Cm is the few-picofarad parasitic between the wires. Notice it is the speed of the swing that matters, not the voltage's size — a slow 100 V change can be harmless while a fast 3 V logic edge is vicious.
Put real numbers on it. Say two PCB traces run side by side with Cm = 1 pF between them, and the aggressor carries a logic edge that swings 1 V in 1 ns — that is a slew rate of 1 V / 10^-9 s = 10^9 V/s. The injected current is i = 1 pF times 10^9 V/s = 10^-12 times 10^9 = 10^-3 A, a full milliamp spike of charge shoved into the victim. Now the victim's own impedance decides the damage. If the victim node is a low 50 ohm, that milliamp makes only 50 mV of blip; but if it is a high-impedance 100 kohm input — a sensor, an op-amp's + input — the same milliamp tries to make a huge spike and the line jumps wildly. That is why capacitive coupling hunts high-impedance nodes.
The cures follow straight from the equation i = Cm times dV/dt. Shrink Cm by moving the wires apart or crossing them at right angles instead of running them parallel (parallel wires share the most area, hence the most stray capacitance). Slow the aggressor's edge if you can afford to. Lower the victim's impedance so the injected current makes less voltage. And the decisive cure — slip a grounded conductor between aggressor and victim, so the electric field terminates on the shield and is swept to ground before it ever reaches the victim. That grounded barrier is the seed of the whole idea of shielding, which we develop fully below.
Inductive coupling: the magnetic handshake
The second field doorway is the mirror image of the first. A changing current in the aggressor wraps a changing magnetic field around itself, and any nearby loop of wire that the field threads through has a voltage induced in it — exactly the flywheel-like reluctance of magnetism to let flux change suddenly. The coupling is set by the mutual inductance M between the two loops, and the induced victim voltage is v = M times dI/dt. Where capacitive coupling cared about voltage slew, this one cares about current slew — the fast-switching supply currents of a motor driver or a switching regulator are the classic aggressors.
Numbers again. Let the mutual inductance between an aggressor loop and a victim loop be M = 2 nH, and let the aggressor's current snap by 0.2 A in 2 ns, a slew of 0.2 A / 2x10^-9 s = 10^8 A/s. Then v = 2 nH times 10^8 A/s = 2x10^-9 times 10^8 = 0.2 V — a 200 mV glitch conjured in a wire carrying no signal of its own. The key knob here is M, and M grows with the area of the victim loop: a big rectangle of signal-wire-out and ground-return-wire-back catches a lot of flux, a thin one catches little. Minimize loop area and you starve the magnetic handshake at its source.
Crosstalk: when the two doorways meet on a board
On a real PCB the two field handshakes happen together between adjacent traces, and the combined leak of an aggressor signal into a victim trace is called crosstalk. It is the everyday face of coupling: a fast clock line whispering into the analog line beside it, a digital bus making your ADC reading jitter. The table below is the field guide for telling the two contributions apart, because the cure depends entirely on which one dominates — and the honest answer is usually 'both, in proportions set by the impedances'.
CAPACITIVE coupling INDUCTIVE coupling
------------ ---------------------- ----------------------
field electric (E) magnetic (B)
driven by dV/dt (voltage swing) dI/dt (current swing)
couples via stray capacitance Cm mutual inductance M
injected i = Cm x dV/dt v = M x dI/dt
worst victim HIGH-impedance node LOW-impedance node,
large loop area
main cures guard/shield, distance, shrink loop area,
cross at 90 deg, twisted pair,
lower node impedance ground plane
aggressor =========================== (fast edge, dV/dt + dI/dt)
| Cm )( M
victim --------------------------- (picks up i and v)
GND plane ########################### (return path right below)The layout cures for crosstalk are cheap if you do them up front and painful if you retrofit. Pull aggressor and victim traces apart — a common rule of thumb is the '3W rule', keeping edge-to-edge spacing at least three trace widths, which knocks coupling down to roughly one percent; be honest that this is a guideline, not a law, and tight, fast, or long parallel runs need more. Drop a grounded guard trace between them to catch the electric field. Above all, keep that solid ground plane unbroken beneath the signals: a slot or gap in the plane forces the return current to detour around it, ballooning the loop area and turning a quiet board into a crosstalk and emissions problem in one stroke.
Shielding, twisted pair, and beating noise by symmetry
When the noise is out in the air rather than on the next trace over, you build a wall: a conducting enclosure that surrounds the sensitive circuit is a Faraday cage. An external electric field cannot penetrate a closed conductor — the field pushes charges around on the shell until they cancel it inside — so anything within the metal box is screened. The catch is the word closed: every seam, slot and cable hole is a leak, and a gap behaves like a slot antenna for any wavelength comparable to its size. A shield with a long crack in it can be worse than no shield, because it can resonate. Good shielding is as much about continuous seams and bonded covers as about the metal itself.
Cable shields raise a subtle, much-argued question: ground the shield at one end or both? Tie it at one end only and it drains capacitively-coupled electric fields beautifully while breaking the ground loop that would otherwise sing 50/60 Hz hum into your signal — the right default for low-frequency, high-impedance audio and sensor lines. But a one-end shield does little against magnetic fields. At high frequencies you often must bond both ends so shield current can flow and cancel the magnetic pickup, accepting the ground loop because at radio frequencies the loop's inductance chokes the hum current anyway. There is no universally correct answer; the choice is an honest tradeoff between electric and magnetic threats.
The most elegant defence does not block the noise at all — it makes the noise cancel itself. Send your signal differentially, as the difference between two wires twisted together, and let a difference amplifier (or its precise cousin the instrumentation amplifier) at the far end subtract them. Because the twisted pair holds both wires equally close to the interference, capacitive and magnetic coupling land on both wires almost identically — as a common-mode voltage. The receiver amplifies the difference and ignores what is common, and its common-mode rejection ratio (CMRR) measures how well. A CMRR of 100 dB attenuates that shared noise by a factor of 10^5: a 1 V interfering common-mode wobble shrinks to a 10 uV blip on the real signal. This is why microphone cables, USB, Ethernet and industrial sensor loops are all balanced twisted pairs.
One more workhorse belongs here: the ferrite bead, a small lossy core slipped over a wire. It looks like a near-zero resistor at DC and audio but becomes a high resistance at radio frequencies, where it turns unwanted high-frequency current into a little heat — perfect for killing noise riding on a cable or supply line without disturbing the real signal. Note the honest limits, though: a bead does nothing at low frequencies, it can resonate with nearby capacitance, and it is a band-aid, not a substitute for a good loop area and a clean ground. Coupling problems are won at layout time; the bead, the shield and the twisted pair are how you handle what layout alone cannot — which is exactly the spirit guide 5 carries into formal EMC and ESD testing.