Signal Reflection In Transmission Lines: Causes And Termination

Every discontinuity on a transmission line sends part of the incident wave back toward the source. The mechanism is not mysterious: the wave carries a ratio of voltage to current set by the line, and when it reaches a region where that ratio cannot be maintained, the excess energy has nowhere to go but back. What matters in practice is how much comes back, at what frequency, and what it does to the receiver’s decision.

This article covers the quantities used to describe signal reflection, the discontinuities that cause it on a real board, and the termination strategies that suppress it without consuming more power than the design can afford.

Describing Reflection Numerically

The reflection coefficient is the ratio of the reflected voltage to the incident voltage, and it is computed from the load impedance and the characteristic impedance of the line. When the two match, the coefficient is zero and no energy returns. When the line is open at the far end, the coefficient is one and the reflected wave doubles the voltage at that point. When the line is shorted, the coefficient is minus one, and the reflected wave inverts.

Power reflection follows the square of that coefficient, and the related figure used in RF work is return loss, expressed in decibels as the ratio of incident to reflected power. A return loss of twenty decibels sounds good and corresponds to a reflection coefficient of about a tenth, which still leaves a visible perturbation on a fast edge.

What Actually Creates The Discontinuity

On a printed circuit board the largest discontinuities are usually deliberate features rather than defects. A via that changes layers presents a short section of different impedance and an associated capacitance. A connector presents a transition between two geometries. A stub left on a trace for a test point presents an open circuit branch that radiates energy back at a frequency determined by its length.

The subtler causes are the ones that do not appear in a schematic. A trace that passes over a split in its reference plane loses its return path and behaves as though it had a much higher impedance. A section of line that widens briefly, whether from a thermal relief or from the pad of a small component, adds capacitance. A connector footprint whose ground pins are sparse adds inductance. Each of these is small on its own, and together they consume the margin that the receiver needed.

Why Reflections Matter More At Higher Speed

A reflection is a delayed copy of the signal. If the round trip time to the discontinuity is short compared with the edge transition time, the reflected wave arrives while the edge is still transitioning and merely changes its shape slightly. If the round trip is comparable to the transition time, the reflected energy arrives as a separate event and produces a step or a plateau in the waveform.

That is why the same board layout works at one hundred megabits per second and fails at one gigabit per second without any change to the routing. The physical discontinuity is unchanged; the edge is faster, so the round trip time is now a significant fraction of the transition. The waveform disturbance moves from an invisible perturbation to a decision point error.

Termination Strategies

Series termination places a resistor at the driver, close to the source, so that the incident wave is halved and the reflection at the far end restores the full amplitude. It costs no static power, uses one resistor per net, and is the natural choice for a point to point link with a single receiver. It requires that the load be essentially open at the receiver, which is true for a CMOS input.

Parallel termination places a resistor to a supply rail at the receiver, absorbing the wave and dissipating power continuously. Thevenin termination splits the resistor between the rail and ground, centring the bias at half the supply and halving the static current. Both are appropriate where several receivers share a bus, and both fail where the driver cannot supply the additional DC current.

Termination On Buses And Multi Drop Nets

A multi drop bus is the hardest case, because the loading of every receiver and the length of every stub are part of the geometry. Keeping stub lengths short relative to the rise time is the first rule, and it usually matters more than the choice of termination. Where the bus must be long, terminating both ends matched to the line impedance gives the cleanest result at the cost of two resistors and their standing current.

Daisy chaining the receivers rather than stubbing them off a trunk is the usual fix, and it converts a bus into a series of short segments whose reflections are absorbed by the terminator at the far end. When the total load capacitance per unit length becomes comparable to the line capacitance, the effective impedance of the loaded line falls, and the termination value must be recalculated rather than taken from the unloaded figure.

Matching Lengths And Matching Impedance

Length matching and impedance matching solve different problems and are often confused. Matching lengths keeps the arrival times of parallel bits aligned so that the receiver samples a coherent word. It does nothing for signal reflection, and the serpentine used to add the extra length can itself create reflections if the spacing between the meanders is too tight or the geometry too abrupt.

Matching impedance keeps energy from returning to the source. A controlled impedance board with a defined characteristic impedance on every high speed net, continuous reference planes and via transitions designed for the layer change will show clean edges without any discrete terminator at all, provided the driver’s own output impedance is part of the calculation.

Diagnosing Reflection On A Real Board

Time domain reflectometry is the direct measurement: a fast step is launched into the trace and the reflected waveform is examined. A rise in impedance appears as a positive reflection, a capacitive dip as a negative one, and the distance to the feature can be read from the time delay. On a finished board the measurement is usually made through a test coupon or a connector.

Where a coupon is not available, the practical approach is to compare a known good net with the suspect one, and to look for differences in the eye diagram as the data rate is swept. A defect that only appears above a certain rate is a reflection that the edge transition time has finally exposed, and the high frequency trace routing rules that govern via design and reference continuity are where the fix is found.

FAQ

Do I need termination on every high speed net? No. Short point to point nets whose round trip time is a small fraction of the edge transition time work without termination, provided the driver impedance and the line impedance are reasonably matched. Termination earns its cost on long nets, buses and fast edges.

How much return loss is acceptable? It depends on the standard and on the link budget. Many high speed serial specifications require better than ten decibels across the band of interest, and demanding links require fifteen or twenty. The number should come from the interface specification, not from a rule of thumb.

Can a serpentine cause reflections? Yes, if the meander spacing is tight enough for adjacent segments to couple, or if the geometry changes abruptly. Keeping the spacing at least three trace widths and the amplitude consistent avoids most of the problem.