Electronics Guide

Spectral Content of Clock Signals

An ideal square wave contains energy only at odd harmonics of the fundamental frequency, with amplitudes that decrease as 1/n, where n is the harmonic number. The third harmonic therefore sits about 9.5 dB below the fundamental and the fifth about 14 dB below. Real clock signals deviate from the ideal square wave, with finite rise and fall times, duty-cycle errors, and asymmetries that modify the harmonic content.

The rise and fall times of a clock signal determine the bandwidth of its spectral content. Faster edges produce harmonics that extend to higher frequencies before rolling off. For a trapezoidal waveform with equal rise and fall times, the envelope of the spectrum has two breakpoints. It remains flat up to a first breakpoint at 1/(πτ), where τ is the pulse width, then falls at 20 dB per decade. At a second breakpoint at 1/(πt_r), where t_r is the rise time, the slope steepens to 40 dB per decade. A 50 MHz clock with a 50 percent duty cycle has a pulse width of 10 ns, which places the first breakpoint near 32 MHz; with 1 ns edges the second falls near 320 MHz, and with 0.5 ns edges near 640 MHz. Because faster edges push this second breakpoint higher in frequency, logic families with sub-nanosecond transitions sustain significant harmonic energy into the GHz range.

Duty-cycle error introduces even harmonics into the spectrum. A waveform contains only odd harmonics when it has half-wave symmetry, which for a clock requires both a 50% duty cycle and equal rise and fall times; departing from either condition adds even harmonics. The effect of duty-cycle error is readily quantified. With a 51/49 duty cycle, the second harmonic lies about 30 dB below the fundamental, while a 45/55 clock puts it only about 16 dB below, stronger than the seventh harmonic of an ideal square wave, which sits about 17 dB below. Even harmonics become particularly problematic when they fall where emission limits are most stringent or where the system has resonant structures that amplify specific frequencies.

Clock Distribution Networks

Clock distribution networks act as efficient antennas, coupling clock energy to the surrounding environment. Each clock trace and its return path form a current loop, and for a loop small compared with the wavelength the radiated field grows in proportion to the loop area, the current, and the square of the frequency. Long clock traces, and any return path forced away from the trace, enlarge that loop. Each branch of a clock tree adds another radiating loop, and the branches together often create complex radiation patterns.

Impedance discontinuities in clock distribution networks cause reflections that increase overshoot and ringing. The ringing concentrates energy near the resonant frequencies of the mismatched line, which are set by its electrical length, and so raises harmonics above the envelope that the edge rates alone would predict. Because the reflected waves carry current back and forth along the line, they elevate both conducted and radiated emissions.

Clock skew compensation circuits and delay-locked loops (DLLs) can introduce additional emission sources. A digitally controlled loop that corrects its delay in periodic steps modulates the phase of the clock at its update rate, creating sidebands around each clock harmonic, offset from it by the update rate and its multiples. These sidebands spread spectral energy to frequencies that a list of clock harmonics alone would not predict.

Spread-Spectrum Clocking

Spread-spectrum clocking intentionally modulates the clock frequency to reduce peak spectral emissions. By spreading the clock energy over a wider bandwidth, the peak amplitude at any single frequency decreases. Typical implementations use a modulation depth of roughly 0.5 to 2 percent at a modulation rate in the tens of kilohertz; PCI Express and SATA, for example, specify 30 to 33 kHz.

The reduction depends on how wide the spread is compared with the resolution bandwidth of the measuring receiver. When the spread bandwidth B_spread greatly exceeds the resolution bandwidth B_RBW, the reduction R in decibels approaches:

R = 10 log₁₀(B_spread / B_RBW)

Because the nth harmonic is spread over n times the absolute bandwidth of the fundamental, the benefit grows with harmonic number. With the 120 kHz resolution bandwidth used for radiated measurements from 30 MHz to 1 GHz, a 1 percent spread of a 100 MHz clock gives at most about 9 dB at the fundamental and about 16 dB at the fifth harmonic. A 0.5 percent spread of a 25 MHz clock, only 125 kHz wide, gives almost nothing at the fundamental. Above 1 GHz the resolution bandwidth widens to 1 MHz, so the same spread yields less. The formula is an upper bound: the modulation profile and rate determine how closely a real clock approaches it.

The modulation profile significantly affects spread-spectrum effectiveness. Triangular and Hershey-Kiss profiles distribute energy more evenly than simple sinusoidal modulation, which lingers at the frequency extremes and leaves peaks there. Center-spread modulation maintains the average frequency at the nominal value, while down-spread modulation keeps the maximum frequency at the nominal value, which may be important for timing-critical applications.

Spread-spectrum clocking does not eliminate emissions; it redistributes them. The total energy is unchanged, and the level rises at the neighboring frequencies inside each spread band. A nearby receiver tuned within a spread band still picks up the interference, so the technique helps a compliance measurement more than it helps coexistence with sensitive equipment. Spread-spectrum clocking also consumes timing margin, can complicate analog measurements, and cannot be applied to clocks that must remain spectrally pure, such as the references for radio transceivers and precision data converters.

Transient Current Demands

When a CMOS gate switches states, it must charge or discharge the load capacitance connected to its output. This charging current flows through the power distribution network, creating voltage drops across parasitic inductances and resistances. The rate of change of current (di/dt) determines the magnitude of inductive voltage drops, which in turn affect both signal integrity and emissions.

The current waveform during a switching event resembles a narrow pulse, with duration roughly equal to the signal transition time. The spectrum of such a pulse stays nearly flat up to a frequency on the order of the reciprocal of its duration and rolls off above it, so the shorter the pulse, the higher its spectrum extends. Faster logic families therefore generate higher-frequency spectral content for equivalent switching activity.

The peak current during switching depends on the output drive strength and the load. A capacitive load draws current only while the output is changing, with a value of I = C dV/dt: charging 10 pF through a 3.3 V swing in 1 ns takes an average of 33 mA during the edge. A resistive load, including the characteristic impedance that a driver sees when it launches an edge into a transmission line, draws current for as long as the voltage is applied. Mixed loads combine capacitive spikes with sustained resistive components.

Crosstalk-Induced Switching

Crosstalk between adjacent signal traces induces transient currents on victim lines, even when those lines are not intentionally switching. These induced currents contribute to the overall switching noise budget and can cause false switching in extreme cases. The coupled energy adds to conducted emissions on the victim line and radiates from both aggressor and victim structures.

Forward (far-end) crosstalk travels with the aggressor's edge and arrives at the far end of the victim as a single pulse roughly as wide as the signal rise time. Its amplitude grows in proportion to the coupled length and falls as the edge slows. In a homogeneous dielectric such as stripline, the capacitive and inductive contributions to forward crosstalk cancel, while microstrip, with air above and laminate below, shows both. Backward (near-end) crosstalk appears at the near end as soon as the edge enters the coupled section and lasts for the round-trip delay of that section, twice its one-way delay. Its amplitude saturates once the round-trip delay exceeds the rise time, beyond which added length widens the pulse without raising it. Both mechanisms inject energy into the victim circuit, contributing to the overall noise environment.

Differential signaling reduces crosstalk-induced emissions in both directions. The equal and opposite currents of a differential aggressor produce fields that largely cancel a short distance away, and a tightly coupled victim pair picks up external noise mostly as a common-mode voltage, which the receiver rejects. However, common-mode noise from imperfect balance can still couple to adjacent circuits and contribute to radiated emissions through cable connections and other common-mode paths.

Data-Dependent Emissions

The spectral content of data signals depends on the bit patterns being transmitted. Random data produces a continuous spectrum with energy distributed according to the data rate and transition density. Repetitive patterns concentrate energy at discrete frequencies related to the pattern repetition rate, potentially creating strong emissions at unexpected frequencies.

Idle patterns and training sequences used in high-speed serial interfaces often have repetitive characteristics that generate discrete spectral lines. These emissions may differ significantly from those produced during normal data transmission, requiring testing across multiple operating modes to fully characterize the emission profile.

Data encoding schemes affect emission characteristics by controlling transition density and pattern distribution. 8b/10b encoding, for example, limits runs of identical bits to five and keeps the line DC balanced, which removes low-frequency content and sets a minimum transition density. Scrambling randomizes the data sequence to suppress discrete spectral lines, spreading the energy more uniformly across the signal bandwidth.

Mechanisms of Ground Bounce

When an output driver pulls its output low, current flows from the load through the driver to the ground pin. The inductance of the bond wire, package lead, and PCB trace connecting the driver ground to the power supply return impedes instantaneous current changes. The voltage developed across this inductance causes the local ground to rise above the system ground reference:

V = L di/dt

where L is the total inductance of the ground path and di/dt is the rate of change of the current through it. A single output whose current ramps by 20 mA in 1 ns through 5 nH of ground-path inductance lifts its local ground by 0.1 V.

The magnitude of ground bounce depends on the current slew rate and the total ground path inductance. High-speed drivers with large output currents and fast edge rates produce the most significant ground bounce. The package usually contributes the largest share of that inductance. In through-hole packages such as the dual in-line package, the long lead frame dominates. Surface-mount leaded packages shorten the leads, leaving the bond wire, at roughly 1 nH per millimeter of length, as a comparable or larger share. Flip-chip packages replace bond wires with short solder bumps and spread power and ground connections across the die, cutting the inductance substantially.

Ground bounce can propagate to other circuits sharing the same ground path, causing noise injection into sensitive analog circuits and potentially triggering false switching in digital inputs referenced to the bouncing ground. The distributed nature of ground bounce makes it particularly challenging to measure and characterize at the board level.

Impact on Emissions

Ground bounce creates voltage differentials between different ground regions of a system. These differential voltages drive common-mode currents through any paths that connect the regions, including cables, chassis connections, and parasitic capacitances to the enclosure. Common-mode currents are particularly efficient radiators because they do not cancel as differential currents do.

The periodic nature of ground bounce associated with clock signals concentrates emission energy at clock harmonics. Data-dependent ground bounce adds broadband noise content to the emission spectrum. The combination of discrete and continuous spectral components creates a complex emission signature that varies with operating conditions.

Ground plane discontinuities near high-current switching circuits can channel ground bounce currents in unexpected directions, creating slot antennas and other radiating structures. Proper ground plane design with solid returns under high-speed signals minimizes these effects and contains ground bounce within localized regions.

Mitigation Approaches

Reducing ground path inductance provides the most direct mitigation of ground bounce. Multiple parallel ground pins, distributed throughout the package, reduce total inductance by providing parallel current paths, although mutual inductance between neighboring pins keeps the reduction short of the ideal factor equal to the number of pins. Modern IC packages include numerous ground connections specifically to minimize ground bounce effects.

Slowing edge rates reduces di/dt and proportionally reduces ground bounce magnitude. Many drivers offer selectable edge rate control or multiple drive strength options to allow optimization for each application. The trade-off between signal integrity margins and EMC performance must be carefully balanced.

Decoupling capacitors provide local charge storage that shortens the path the transient current must take. Board-level capacitors keep the current out of the board traces and the supply path, but the package inductance still lies between them and the die, so they cannot reduce the bounce that develops across it; only on-die and in-package capacitance, more ground connections, and slower edges do that. Effective board decoupling requires capacitors with low equivalent series inductance (ESL), mounted with short connections to the planes as close as possible to the power pins. Multiple capacitors of different values address different frequency ranges in the emission spectrum.

Aggregation Effects

When N outputs switch simultaneously, the total switching current can approach N times the single-output current. In a simple lumped model the voltage disturbance scales in proportion, so eight outputs sharing the 5 nH ground path of the earlier example, each ramping 20 mA in 1 ns, lift the local ground by 0.8 V. In practice the growth is somewhat less than linear, because the bouncing ground reduces each driver's gate drive and slows its edge, but the disturbance can still be many times larger than single-bit switching would produce. Wide data buses and memory interfaces are particularly susceptible to SSN effects.

The worst-case switching scenario depends on the circuit topology. For many designs, all outputs switching in the same direction (all rising or all falling) produces the maximum aggregate current. However, designs with significant output-to-output coupling may exhibit different worst-case patterns depending on the relative timing and direction of transitions.

Statistical correlation between data bits affects the probability of worst-case switching patterns. Random data produces statistically independent switching events, while structured data with high correlation increases the probability of many simultaneous transitions. An incrementing address bus flips every bit at once when a carry ripples through all of them, as from 0x7FFF to 0x8000, and a bus that alternates between complementary values such as 0x0000 and 0xFFFF produces the worst case on every cycle.

Frequency Characteristics

SSN produces spectral content concentrated at frequencies related to the switching rate and the resonant frequencies of the power delivery network. The current transient excites resonances between plane capacitance and via inductance, between decoupling capacitors and their mounting inductance, and between package and board structures.

Anti-resonances in the power delivery impedance can amplify specific frequency components of SSN, creating emission peaks at unexpected frequencies. An anti-resonance is a parallel resonance, where the inductance of one stage, such as a group of capacitors and their mounting, resonates with the capacitance of another, such as the plane pair or the die. The network's impedance peaks there, so any switching current at that frequency produces a correspondingly large voltage.

The broadband nature of the switching transient means that SSN contributes to emissions across a wide frequency range. Unlike clock harmonics, which are concentrated at discrete frequencies, SSN creates a noise floor that can raise emission levels throughout the spectrum.

Design Constraints

Device vendors, particularly of FPGAs and ASICs, publish SSO guidelines indicating the maximum number of outputs that may switch simultaneously, typically per I/O bank or per power and ground pin pair, for each I/O standard and drive strength. These limits depend on the package inductance, driver characteristics, and the target noise budget. Exceeding SSO limits can cause signal integrity failures and excessive emissions.

Staggering output switching times reduces peak aggregate current by spreading transitions over a wider time window. Some output drivers include intentional skew to prevent simultaneous switching, while system timing may naturally distribute transitions across multiple clock phases. Coding can also bound the activity: bus-invert coding transmits the complement of a word, with a flag bit, whenever more than half of its bits would otherwise toggle, which caps the number of simultaneous transitions at about half the bus width.

Power delivery network design must account for worst-case SSN scenarios. Target impedance methodology sets impedance limits across the frequency range of interest to ensure that current transients produce acceptable voltage disturbances. The basic target is Z_target = ΔV / ΔI, the allowed voltage variation divided by the transient current: a 1.0 V rail held within 5 percent while handling 10 A transients needs 5 mΩ. Meeting such targets typically requires careful coordination of decoupling capacitors, plane design, and package selection.

Switching Regulator Noise

Switching regulators generate emissions at their switching frequency and harmonics. The discontinuous current drawn from the input and delivered to the output creates high di/dt events that produce both conducted and radiated emissions. The switching frequency, typically in the range of 100 kHz to several MHz, places the fundamental and its low harmonics in the 150 kHz to 30 MHz band covered by the conducted-emission limits of CISPR 32 and FCC Part 15, while the fast switching edges carry harmonics up into the radiated-emission range.

Input and output ripple currents flow through circuit traces that can act as loop antennas. The magnetic field created by these current loops couples to nearby circuits and radiates into the surrounding environment. Minimizing loop areas through careful layout significantly reduces switching regulator emissions.

Switch node ringing, caused by parasitic inductances and capacitances interacting with the rapid voltage transitions, can produce emissions at frequencies well above the fundamental switching frequency. Snubber circuits and gate drive optimization can reduce ringing, while spread-spectrum modulation of the switching frequency reduces peak emissions.

Load Transients

Sudden changes in load current produce transient voltage deviations that propagate through the power distribution network. When a processor enters or exits a low-power state, or when a high-speed interface begins or ends activity, the step change in current demand creates voltage transients that affect all circuits sharing the power domain.

The power delivery network response to load transients depends on its impedance versus frequency characteristics. At low frequencies, the voltage regulator feedback loop maintains regulation. At higher frequencies, the regulator cannot respond quickly enough, and bulk capacitance provides current. At still higher frequencies, local decoupling capacitors and plane capacitance must supply the transient current demand.

Load transient events produce broadband spectral content that adds to the emission noise floor. Repetitive load transients, such as those associated with periodic processing tasks or communication frame structures, create discrete spectral components that may coincide with EMC measurement frequencies.

Power Plane Resonances

Power and ground plane pairs form resonant cavities whose resonant frequencies depend on the plane dimensions and the dielectric properties of the separating material. At these resonant frequencies, relatively small excitations produce large voltage variations that can dominate the emission spectrum.

The lowest resonance of a rectangular plane pair occurs when its longer dimension equals half a wavelength in the dielectric. For a pair of planes measuring a by b, separated by a dielectric of relative permittivity ε_r, the resonant modes fall at:

f_mn = (c / 2√ε_r) × √[(m/a)² + (n/b)²]

where c is the speed of light and m and n are integers, not both zero. A 20 cm by 15 cm plane pair in FR-4, with ε_r of about 4.3, resonates first near 360 MHz, then near 480 MHz and 600 MHz, squarely within the radiated-emission band. The spacing between the planes does not enter the formula, so thinning the dielectric lowers the impedance peaks without moving them. Irregularly shaped planes have more complex resonant mode structures that may be difficult to predict analytically.

Plane resonances are excited by any switching activity that injects energy at the resonant frequencies. Once excited, the resonant modes persist until losses dissipate the energy, creating sustained oscillations that can last for many clock periods. Damping techniques, including resistive termination at plane edges and distributed resistive loading, reduce resonance amplitudes.

Common-Mode Currents on Cables

Differential signals on cables produce magnetic fields that largely cancel in the far field because the forward and return currents are closely spaced and flow in opposite directions. Common-mode currents, however, flow in the same direction on all conductors, and their fields add constructively, creating significant radiated emissions even with modest current levels.

Common-mode currents develop from multiple sources: ground potential differences between connected equipment, unbalanced coupling to signal traces, and intentional signals that have common-mode components. The radiation efficiency is striking. For a cable short compared with a quarter wavelength, the maximum field from a common-mode current I at frequency f on a cable of length L, measured at distance r over a reflecting ground plane, is approximately:

E = 1.257 × 10⁻⁶ × f × I × L / r

with E in volts per meter, f in hertz, I in amperes, and L and r in meters. On a 1 m cable at 48 MHz, where a quarter wavelength is about 1.6 m, a common-mode current of only 5 µA produces about 100 µV/m at 3 m, enough to reach the FCC Class B limit of 100 µV/m that applies from 30 to 88 MHz. Henry Ott's widely used guideline accordingly keeps the common-mode current on any cable of 1 m or longer below about 5 µA for a Class B product. This explains why a single weakly coupled noise source can dominate a product's emission profile.

The cable acts as a monopole antenna driven against the system ground plane. The radiation efficiency depends on the cable length relative to wavelength: it first peaks when the cable length equals a quarter wavelength, near 75 MHz for a 1 m cable, and above that frequency it passes through a succession of resonances and antiresonances while remaining an effective antenna. Multiple connected cables create a complex antenna structure with radiation characteristics that depend on the relative lengths and orientations.

Shield Effectiveness

Cable shields suppress common-mode radiation when they are terminated to the chassis at both ends. At high frequencies, skin effect separates the inner and outer surfaces of the shield: current on the inner conductors returns along the inside surface, and only the portion that leaks through the shield, or that a poor termination drives onto the outer surface, reaches the outside, where it radiates. Shield effectiveness therefore depends on shield coverage, transfer impedance, and termination quality.

Shield terminations must provide low-impedance connections to the chassis or ground reference at all frequencies of concern. Pigtail connections that extend the shield conductor beyond the connector introduce inductance that degrades high-frequency performance. Full 360-degree terminations, which contact the shield around its whole circumference through an appropriate connector backshell, maintain shield effectiveness to much higher frequencies.

Transfer impedance quantifies how effectively a shield prevents external fields from inducing voltages on internal conductors and vice versa. Braided shields have higher transfer impedance at high frequencies than solid shields because of coupling through the braid apertures. Combination shields using foil with an overlying braid offer improved high-frequency performance.

Cable Filtering

Ferrite cores placed around cables increase the common-mode impedance, reducing the common-mode current for a given driving voltage. The ferrite material provides frequency-dependent impedance that typically peaks in the range of 10 to several hundred MHz, depending on the material formulation.

Multiple turns through a ferrite core increase the impedance in proportion to the square of the number of turns, at least at frequencies well below self-resonance. However, parasitic capacitance between turns creates a self-resonance that limits effectiveness at high frequencies, and because each added turn adds capacitance as well as inductance, the self-resonant frequency falls as turns are added.

Cable filtering at the connector interface, using filtered connectors or common-mode chokes, addresses emissions before they can propagate along the cable. A filter capacitor bonded to the chassis at the enclosure wall shunts the noise before it reaches the cable, whereas a ferrite only adds series impedance, whose effect depends on the impedance of the rest of the common-mode circuit. For that reason, connector-level filtering is often more effective than a ferrite added to the cable afterward.

Slot Antenna Principles

A slot in a conducting surface radiates according to Babinet's principle, which relates slot antenna behavior to complementary dipole antenna behavior. A half-wavelength slot has characteristics similar to a half-wave dipole, including relatively high radiation efficiency and defined polarization.

The slot radiation pattern depends on the slot orientation and the field distribution along its length. Horizontal slots radiate primarily with vertical polarization, while vertical slots produce horizontally polarized emissions. The surrounding structure affects the pattern, causing directivity variations that depend on the observation angle.

Slot resonance occurs when the slot length approaches half a wavelength: a 10 cm slot resonates near 1.5 GHz. At resonance the slot's reactance cancels, leaving a purely resistive input impedance of several hundred ohms, the complement, by Booker's relation, of the roughly 73 Ω of a half-wave dipole. The voltage that internal currents and fields develop across the slot then drives energy efficiently into radiated emissions. Non-resonant slots also radiate but with lower efficiency.

Common Slot Sources

Enclosure seams, particularly those between removable panels, form slots that can extend the full length of the seam. Even with apparent metal-to-metal contact, surface oxide layers and imperfect mating create high-impedance gaps that behave as slots at high frequencies.

Ventilation openings represent intentional slots that must balance thermal requirements with EMC performance. Large openings provide better airflow but also more efficient radiation. Honeycomb structures and waveguide-below-cutoff designs can provide ventilation while maintaining shielding effectiveness.

Cable entry points create slots when cables penetrate the enclosure through oversized holes or when multiple cables pass through the same opening with air gaps between them. Proper cable entry treatment using appropriate glands, grommets, or filtered connectors maintains enclosure integrity.

Mitigation Techniques

Keeping every slot shorter than half a wavelength at the highest frequency of concern avoids resonance, but a sub-resonant slot still leaks, so practical designs keep openings much shorter still. For a single aperture whose longest dimension is L, the shielding effectiveness is approximately:

SE = 20 log₁₀(λ / 2L)

which falls to zero when L reaches λ/2. A common guideline limits openings to λ/20, which gives about 20 dB; at 1 GHz, where the wavelength is 30 cm, that means 15 mm. Multiple short slots radiate far less efficiently than a single long slot with equivalent total area, though their leakage adds: for n identical apertures closely spaced in a row no longer than half a wavelength, the shielding effectiveness falls by a further 20 log₁₀(√n), about 6 dB for four apertures. This principle guides the design of ventilation patterns and seam treatment.

Conductive gaskets at seams maintain electrical continuity even when mechanical contact is imperfect. The gasket must provide low-impedance contact at the frequencies of concern, which typically requires proper selection of gasket material and compression to achieve adequate contact pressure.

Waveguide-below-cutoff principles apply to openings that must remain open, such as ventilation holes. An opening with depth behaves as a short waveguide, and below its cutoff frequency, roughly 17.6 GHz divided by the diameter in centimeters for a circular tube, waves cannot propagate through it and are instead attenuated exponentially with depth. Well below cutoff, a circular tube attenuates by about 32 dB for each diameter of length, so the diameter of an opening sets its cutoff and its depth sets the attenuation. Honeycomb structures extend this principle by creating many small waveguide sections in parallel.

Origins of Common-Mode Currents

Common-mode currents arise from asymmetries in ostensibly balanced systems. Differences in trace routing, component placement, or termination impedances create imbalances that convert a portion of differential-mode energy to common-mode. Even small percentage imbalances can produce significant common-mode currents because the common-mode radiation efficiency far exceeds that of differential-mode signals.

Ground potential differences between different parts of a system drive common-mode currents through any available path. When two circuits reference different ground points that have voltage differences due to current flow, cables connecting the circuits carry common-mode current proportional to the voltage difference divided by the common-mode impedance.

Capacitive coupling from noise sources to cables and chassis creates common-mode currents that may be unrelated to the intended signal paths. Stray capacitance from switching regulators, high-speed logic, or other noise sources to nearby cables injects common-mode current directly onto the cable structure.

Common-Mode Rejection

Differential signaling provides inherent common-mode rejection when the receiver responds only to the voltage difference between the two conductors. This rejection protects signal integrity from common-mode noise but does not prevent the common-mode currents from radiating. Reducing common-mode current generation remains essential for EMC compliance.

Common-mode chokes present high impedance to common-mode currents while allowing differential-mode signals to pass unimpeded. The choke consists of two or more windings on a common magnetic core, wound so that differential-mode flux cancels and only common-mode flux magnetizes the core. This selectivity makes common-mode chokes effective filters without affecting signal integrity.

Balanced driver and receiver design minimizes common-mode generation by ensuring symmetric operation. Matched termination impedances, symmetric trace routing, and balanced component values all contribute to maintaining the differential-mode nature of the signal and minimizing common-mode conversion.

Measurement Considerations

Common-mode current measurement using current probes requires careful interpretation. The probe measures the net current passing through it, the sum of the currents on all enclosed conductors with their phases, which is the common-mode component when the probe is clamped around a whole cable. Separating individual conductors allows measurement of each current, from which common-mode and differential-mode components can be calculated.

The relationship between measured common-mode current and radiated emissions depends on the cable configuration and frequency. The short-monopole formula given under Cable Radiation converts a current-probe reading into an order-of-magnitude field estimate, which makes the current probe a practical pre-compliance tool. It holds only while the cable is short compared with a quarter wavelength, and accurate prediction requires consideration of the cable's position relative to ground planes and the proximity of other conductive structures.

Diagnostic techniques using near-field probes can locate common-mode current sources before they propagate to cables. Comparing emissions with and without cables connected identifies whether cables or the product enclosure dominates the emission path, guiding mitigation efforts toward the most effective interventions.

Frequency Planning

Strategic selection of clock frequencies and their harmonics can avoid placing strong emissions at frequencies where limits are most stringent or where the system has resonant structures. Frequency planning requires knowledge of the emission spectrum, the applicable limits, and the frequency-dependent characteristics of the system's radiating structures.

EMC standards specify measurement bandwidths and detector types that affect how emissions at different frequencies contribute to measured results. The CISPR 16 receiver uses a 9 kHz resolution bandwidth with quasi-peak and average detectors for conducted measurements from 150 kHz to 30 MHz, 120 kHz with a quasi-peak detector for radiated measurements from 30 MHz to 1 GHz, and 1 MHz with peak and average detectors above 1 GHz. A narrowband clock harmonic registers its full amplitude in any of these bandwidths, whereas broadband noise and spread-spectrum energy register only the share that falls within the bandwidth. Understanding these measurement parameters helps predict which emission sources will dominate test results and guides mitigation priorities.

Layer Stackup Design

PCB layer stackup significantly affects emission sources by determining loop areas, return path inductances, and parasitic coupling. Closely spaced power and ground planes reduce power delivery inductance and contain fields between the planes. Signal layers adjacent to reference planes minimize loop areas and provide well-defined return paths.

Interplane capacitance between power and ground planes provides distributed decoupling that supplements discrete capacitor placement. This capacitance is most effective at high frequencies where discrete capacitor inductance limits performance. Because capacitance varies inversely with plane spacing, thin dielectric layers maximize it: a 0.1 mm FR-4 layer provides about 38 pF per square centimeter, and halving the spacing doubles it.

Component Placement Strategy

Placing high-speed circuits and potential emission sources away from board edges keeps their fields over the reference planes. Near an edge, part of a trace's field extends beyond the plane, where it radiates more readily and can couple into the edge of the plane pair, and edge locations usually sit closer to the connectors where cables attach.

Separating analog and digital circuits reduces coupling of switching noise into sensitive measurement circuits. This separation also contains emission sources within defined regions where targeted mitigation measures can be applied without affecting the entire design.

Locating connectors and cable attachment points away from high-speed circuits reduces direct coupling of emissions to cables. When proximity is unavoidable, proper filtering at the connector prevents noise from reaching the cable structure.