Module 03 / Transmission Lines & Matching

Real Transmission-Line Structures

A “50 Ω trace” is not a line until its fields, return conductor, materials, geometry, frequency, and transitions are named. Compare practical structures, expose modal and loss tradeoffs, and write a pre-layout decision that knows what still needs solving and measuring.

01 / 10

Why a nominal 50 Ω trace changed

The layout used the width from an earlier 50 Ω rule. Why did the assembled sensor node no longer look like 50 Ω?

Illustrative engineering case: a 2.450 GHz radio reaches a surface connector through a short PCB feed. The routing rule copied a nominal 0.30 mm width, but the released stackup changed finished dielectric height, laminate construction, and copper treatment. The connector pad also widened the path, while its ground pins and stitching vias defined a three-dimensional return-current transition. “0.30 mm” described only one dimension of several.

ArtworkWidth is intentional

But etch and plating determine the finished cross-section.

StackupFields set the ratio

Height, Dk, resin/glass geometry, and nearby metal share control.

LaunchUniformity ends

Pads, antipads, vias, connector pins, and enclosure currents reshape the mode.

Think about itIf finished width is held exactly, which single drawing note guarantees 50 Ω?
Answer

None. A controlled-impedance requirement must bind the complete structure and its evidence: named layer and reference, finished geometry, material data and method, tolerance, frequency band, launch boundaries, solver/coupon plan, and acceptance criterion.

Common misconceptionA controlled-impedance trace is 50 Ω because the CAD rule says so.

The rule is an input to fabrication, not a measurement. A nominal cross-section estimate, a fabricator field-solved target, a coupon result, and the impedance of the product route are distinct evidence with distinct reference planes.

02 / 10

A mode needs fields and a return conductor

Where does current return, and what exactly repeats as the wave advances?

A guided mode is a permitted electric- and magnetic-field pattern of the entire cross-section. Voltage is an integral of electric field between conductors; current is associated with the magnetic field around those conductors. The familiar characteristic impedance is therefore a modal voltage/current ratio—not a material property of the signal trace alone.

Zmode=Vmode+Imode+Z_{\mathrm{mode}} = \frac{V^{+}_{\mathrm{mode}}}{I^{+}_{\mathrm{mode}}}The voltage path, current definition, direction, and reference conductors belong to the result.
  • TEM: electric and magnetic fields are transverse to propagation. Ideal coax and ideal homogeneous stripline support this model.
  • Quasi-TEM: the dominant fields remain transmission-line-like, but inhomogeneous dielectric makes the exact mode hybrid and potentially dispersive. Microstrip and PCB GCPW use this approximation at sufficiently low normalized frequency.
  • Unwanted modes: slots, cavities, common-mode paths, higher-order modes, and radiation can appear when symmetry, boundaries, or electrical size no longer support the assumed single mode.
Four practical transmission-line cross-sections and their return conductorsCoax has a center signal surrounded by a return shield. Microstrip has a surface signal over a plane. Stripline has a signal between two planes. Grounded coplanar waveguide has a surface signal between grounded rails tied by vias to a backing plane. Cyan field strokes are conceptual, not calculated.COAXshield is the return conductorMICROSTRIPfields occupy air and dielectricSTRIPLINEhomogeneous ideal TEM regionGCPWrails need real via connections
Definition · idealized cross-sections. Navy denotes the intended conductors; cyan strokes show field topology only. The drawings are not to scale and contain no solver data.
Common misconceptionGround is an equipotential sink, so the return route does not matter.

At RF, return current follows the path that satisfies the electromagnetic boundary conditions with least impedance—not a schematic ground symbol. Plane splits, voids, sparse stitching, and connector geometry can force a longer or different mode.

03 / 10

Coax confines a TEM path

Why is coax often the cleanest reference structure, and where does that cleanliness stop?

In an ideal uniform coaxial line, the outer conductor encloses the center conductor. The electric field runs radially between them and the magnetic field circles the center conductor; the return current occupies the shield. With a homogeneous dielectric, the dominant mode is TEM.

Z0=(60ϵr)ln(ba)Z_{0} = (\frac{60}{\sqrt{\epsilon _{r}}}) \ln (\frac{b}{a})Ideal lossless coax: a is inner-conductor radius [m], b is shield inner radius [m], and εr is homogeneous relative permittivity.

The symmetry offers good field confinement and a clear return boundary, but a real cable adds conductor and dielectric loss, roughness, dimensional tolerance, bends, connectors, and a higher-order-mode limit. The PCB launch breaks circular symmetry and must transform the coaxial mode into the board mode without creating a large excess inductance, capacitance, or alternate return path.

Go deeperWhy the shield can still carry external current

Ideal coax has equal and opposite currents on the center conductor and the shield’s inner surface. Common-mode current on the shield’s outer surface is a different mode. Cable routing, connector bonding, and enclosure current can excite it even while the internal coaxial mode remains well matched.

04 / 10

Microstrip and stripline make different trades

If both can be designed for 50 Ω, why choose one over the other?

Topology consequences · qualitative engineering comparison
QuestionMicrostripSymmetric stripline
Field regionAir plus dielectric; quasi-TEMHomogeneous dielectric; ideal TEM
AccessEasy probing, trimming, and surface launchingRequires transitions to an internal layer
ConfinementMore exposed to nearby copper, coatings, and radiationFields better confined between planes
Delay / dispersionEffective permittivity lies between air and substrate values and can vary with frequencyIdeal homogeneous model uses εeff = εr; real laminates still have material dispersion
Manufacturing leversFinished width, height, copper shape, mask, and material constructionThose plus centering and both plane spacings

A quasi-static formula is useful because it makes direction visible: wider trace generally lowers impedance; greater separation from the reference plane generally raises it; higher Dk generally lowers it and increases delay. But a formula’s variables and validity region are part of the result. A different thickness correction or Dk definition can move the answer without any arithmetic error.

Common misconceptionStripline is always lower loss because its fields are confined.

Confinement does not determine total attenuation by itself. Stripline places essentially all electric energy in dielectric, may require narrower conductors for a target impedance, and adds vias to reach the layer. Copper profile, Df, geometry, frequency, and route/transition length decide the budget.

05 / 10

Coplanar ground must actually be connected

Do two ground pours beside a trace automatically become the intended return conductor?

CPW places the signal and return conductors on one surface. Grounded CPW also includes a backing plane. That topology is attractive around surface connectors and components because the return can remain close to the signal—but only if the coplanar rails are electrically connected to each other and the backing plane. Otherwise, asymmetric slot voltages and slotline-like modes are possible.

  1. 01Define the gap

    Use finished signal-edge to ground-edge spacing, including etch shape and solder-mask policy.

  2. 02Stitch the rails

    Place return vias from field/current reasoning and solver evidence, not a universal folklore pitch.

  3. 03Guard the launch

    Connector ground pins, rail width, via antipads, and the backing plane form one three-dimensional transition.

  4. 04Check the enclosure

    Nearby walls and ground edges can become part of the cross-section or create another guided path.

The explorer’s GCPW equation assumes infinite, symmetric coplanar grounds perfectly tied to an infinite backing plane. It does not numerically model via pitch or rail width. Those omissions are dominant when the physical layout no longer resembles the assumed cross-section.

Common misconceptionA via fence spacing such as λ/10 is a universal GCPW design rule.

A wavelength fraction can be a screening heuristic only after the relevant medium, mode, frequency, boundary, leakage mechanism, and allowed error are named. Launch current density and local discontinuities can demand closer connections than a route-level wavelength rule suggests.

06 / 10

Differential pairs have odd and even modes

What does “100 Ω differential” conceal about the two-conductor system?

Two coupled traces support two natural symmetric modes. In the odd mode, trace voltages are equal and opposite; strong electric field occupies the gap. In the even mode, the traces move together relative to the reference conductor. Their field patterns differ, so their impedances and effective permittivities generally differ.

Odd excitationVA = −VB

Zdiff = 2Zoo

Even excitationVA = VB

Zcommon = Zoe/2

Symmetric referenceZSE,coupled = √(ZoeZoo)

Not the isolated-line Z0

The displayed relationships use the lesson’s stated modal voltage/current definitions. A field solver or datasheet can use another normalization, so import the definitions with the values. When odd and even phase velocities differ, a waveform that contains both can accumulate modal skew.

Think about itWhat happens as the gap between two identical traces becomes very large?
Answer

Coupling weakens. Odd- and even-mode impedances converge toward the isolated-trace impedance, so Zdiff approaches twice that impedance and Zcommon approaches half. The pair does not lose its modal description; the two modal solutions become nearly degenerate.

Common misconceptionDifferential current means the reference plane carries no current.

An ideal perfectly symmetric odd mode may concentrate return between the conductors, but real PCB pairs still interact with planes. Asymmetry, bends, launches, skew, packages, and common-mode content drive reference-plane current and conversion.

07 / 10

Loss mechanisms change with frequency

Why is one room-temperature loss number not a broadband interconnect model?

ConductorSurface resistance rises

For smooth thick copper, Rs ∝ √f. Finite thickness, edge crowding, proximity, plating, and roughness change the current distribution.

DielectricPolarization dissipates

For a small, constant loss tangent, attenuation trends roughly with f·tanδ and electric-energy participation. Real Dk and Df depend on method and frequency.

Radiation / leakageBoundaries stop confining

Open structures, bends, slots, plane edges, cavities, and unintended modes can move energy outside the intended line.

δ=1πfμ0σRs=πfμ0σ\delta = \frac{1}{\sqrt{\pi f\mu _{0}\sigma}} \qquad R_{s} = \sqrt{\frac{\pi f\mu _{0}}{\sigma }}Smooth bulk-copper trend with σ = 5.8 × 10⁷ S/m and μr = 1; δ is skin depth [m], Rs is Ω/square.

The lesson adds a finite-sheet factor coth(t/δ), then uses a deliberately simple series-resistance proxy. This exposes why thickness stops buying much when copper is many skin depths thick, but it cannot predict rough-copper or edge-current loss. Material suppliers likewise caution that laminate Dk/Df values depend on test method, frequency, and construction; use the value that matches the modeling purpose.

Go deeperSeparate attenuation from distortion

If all spectral components see the same attenuation and a linear phase slope, the waveform can scale and delay without changing shape. Frequency-dependent attenuation, nonlinear phase, modal delay differences, and reflections reshape it. “Low loss” and “low distortion” are related but not interchangeable requirements.

08 / 10

A uniform line ends at every discontinuity

Where should the 2D cross-section model stop and the 3D model begin?

A uniform line model assumes the same transverse field solution at every position. A pad changes width; a via changes direction and exposes a barrel; an antipad changes plane capacitance; a bend redistributes current; a connector transforms geometry and reference conductors. Each is a local network between explicit planes, even if its physical length looks small.

  1. R1Uniform board line2D cross-section evidence
  2. R2Pad + via field3D transition evidence
  3. R3Connector launchVendor/model correlation
  4. R4Cable interfaceCalibration / de-embedding plane

Reference planes make separate evidence composable. They also prevent a successful cable calibration from being misreported as proof of the un-de-embedded PCB launch. IEEE 370 provides standardized context for interconnect characterization and fixture de-embedding; it does not make an unevaluated launch correct by declaration.

Common misconceptionA connector footprint is validated because it matches the vendor drawing.

A footprint drawing controls manufacturability and compatibility. RF performance also depends on the chosen layer transition, stackup, antipads, ground-via arrangement, launch orientation, enclosure, fabrication, and measurement planes. Use the vendor geometry as an input to the actual 3D problem.

09 / 10

Coupling and asymmetry create another path

What changed when the intended line still looks uniform, but nearby routing or an asymmetric launch appears?

Every nearby conductor can participate capacitively and inductively. Coupling does not merely add an error voltage; it creates a larger multi-conductor modal system. Symmetry can keep desired and common modes separate. Width/gap imbalance, length skew, unequal via fields, plane openings, and connector imbalance mix those modes.

The explorer intentionally does not invent a coupling or conversion percentage: that needs a multi-conductor or 3D model with the aggressor, victim, boundaries, ports, and frequency band. Its coupled-pair calculation does reveal the modal basis and the sensitivity of Zdiff to width, height, gap, and Dk under exact symmetry.

Think about itCan a route meet its differential-impedance target and still radiate or fail a common-mode requirement?
Answer

Yes. Zdiff describes one modal ratio in a cross-section. Asymmetric discontinuities can convert some incident differential energy to common mode even when the nominal odd-mode impedance is correct. Conversion, return continuity, and radiation require their own evidence.

10 / 10

Select and document the node feed

Which structure earns a place in the layout, and what would make the team reopen that choice?

Use the same starting dimensions to expose different field assumptions, then vary one input at a time. The comparison is intentionally not an optimization: equal numeric dimensions do not mean equal manufacturability, equal target impedance, or equal launch quality. The output is a traceable pre-layout trade record.

Class 2 interaction · derived teaching evidence

Structure Trade Explorer

Hold a compact cross-section contract, compare four line families, and rank exact one-at-a-time tolerance endpoints. Recalculate explicitly so every displayed result corresponds to one submitted input set.

structure-trade/1.0.0
Cross-section
Same numeric inputs are reinterpreted under each named geometry contract.
0.05 to 3 mm
0.05 to 2 mm · signal layer to backing plane
5 to 105 µm
Material, frequency, and sweep
2 to 12
0 to 0.03
0.1 to 20 GHz
0 to 20 %

Entry limits prevent nonphysical controls. Formula-region warnings are evaluated after calculation and do not silently clip the result.

Default microstrip result ready.

Derived · within declared teaching region

Microstrip

Hammerstad–Jensen zero-thickness quasi-static microstrip

Z0
56.58 Ω
Quasi-static wave impedance
Effective ε
3.0249
Quasi-static propagation value
Delay
5.801 ps/mm
From c / √εeff
Loss trend
11.80 dB/m
Conductor + dielectric proxy
Selected transmission-line cross-sectionOne surface trace over one reference plane, with fields shared by air and dielectric. Field strokes are explanatory and are not solver data.εr = 4.00E / H field sketchh
One surface trace over one reference plane, with fields shared by air and dielectric. Cyan field strokes are conceptual, not computed field magnitude.
Conductor trend6.61 dB/mSmooth copper · finite-sheet coth(t/δ)
Dielectric trend5.19 dB/mSmall-loss participation proxy
Skin depth1.335 µmσ = 5.8 × 10⁷ S/m

Validity audit

  • w/h = 1.6667Inside declared 0.1 to 10 region

Warnings

No formula-region warning for this submitted point.

Same entered dimensions · different geometry contracts
StructurePrimary impedanceεeffDelayLoss trendDecision cue
MicrostripWithin regionZ0 56.58 Ω3.02495.801 ps/mm11.80 dB/mEasy surface access; fields sample air and nearby materials.
Symmetric striplineWithin regionZ0 36.98 Ω4.00006.671 ps/mm16.80 dB/mStrong confinement; requires an internal layer and controlled plane spacing.
Grounded coplanar waveguideWithin regionZ0 55.04 Ω2.92285.703 ps/mm11.81 dB/mSurface launch-friendly only when coplanar grounds and the backing plane are connected well.
Coupled microstrip pairWithin regionZdiff 102.06 Ω2.99745.771 ps/mm11.86 dB/mCarries odd and even modes; symmetry and gap control the differential/common result.
Exact one-at-a-time ±10.0% endpoints · ranked by primary-impedance span
Rank / inputLow input → impedanceNominalHigh input → impedanceZ spanLoss low → high
1 · Trace width0.2700 mm60.05 Ω0.3000 mm56.58 Ω0.3300 mm53.53 Ω11.517%12.0711.58 dB/m
2 · Reference-plane height0.1620 mm53.21 Ω0.1800 mm56.58 Ω0.1980 mm59.71 Ω11.488%12.2611.42 dB/m
3 · Relative permittivity3.600 59.27 Ω4.000 56.58 Ω4.400 54.23 Ω8.900%11.2112.36 dB/m
4 · Copper thickness31.50 µm56.58 Ω35.00 µm56.58 Ω38.50 µm56.58 Ω0.000%11.8011.80 dB/m
5 · Loss tangent0.01350 56.58 Ω0.01500 56.58 Ω0.01650 56.58 Ω0.000%11.2812.32 dB/m
6 · Frequency2.205 GHz56.58 Ω2.450 GHz56.58 Ω2.695 GHz56.58 Ω0.000%10.9412.64 dB/m

What the calculator includes

  • Named zero-thickness quasi-static impedance models
  • Odd/even modes for the symmetric pair
  • Smooth-copper and dielectric loss trends
  • Exact deterministic one-at-a-time endpoints

What it deliberately omits

  • finite copper thickness in impedance
  • dispersion, solder mask, weave, anisotropy, and roughness
  • ground-current spreading, radiation, bends, pads, vias, and launches
  • No yield, compliance, or fabrication-release claim.

Decision rule: use this result to expose trade direction and sensitivity. Replace it with fabricator-confirmed geometry/material data, field solving, launch models, coupons, and measurement correlation before release.

Server-rendered default · printable / no-script record

Default microstrip estimate: Z0 = 56.58 Ω

Model structure-trade/1.0.0: w = 0.30 mm, h = 0.18 mm, t = 35 µm, εr = 4.00, tanδ = 0.015, f = 2.45 GHz, and exact one-at-a-time ±10% endpoints. Hammerstad–Jensen zero-thickness quasi-static impedance; thickness enters only the smooth-copper loss trend.

Effective ε
3.0249
Delay
5.801 ps/mm
Conductor trend
6.61 dB/m
Dielectric trend
5.19 dB/m
Same default dimensions · geometry meanings remain structure-specific
StructurePrimary impedanceεeffLoss trend
MicrostripZ0 56.58 Ω3.024911.80 dB/m
Symmetric striplineZ0 36.98 Ω4.000016.80 dB/m
Grounded coplanar waveguideZ0 55.04 Ω2.922811.81 dB/m
Coupled microstrip pairZdiff 102.06 Ω2.997411.86 dB/m
Default microstrip · exact one-at-a-time ±10% endpoints
Rank / inputLow → Z0Nominal → Z0High → Z0Z spanLoss low → high
1 · Trace width0.2700 mm60.05 Ω0.3000 mm56.58 Ω0.3300 mm53.53 Ω11.517%12.0711.58 dB/m
2 · Reference-plane height0.1620 mm53.21 Ω0.1800 mm56.58 Ω0.1980 mm59.71 Ω11.488%12.2611.42 dB/m
3 · Relative permittivity3.600 59.27 Ω4.000 56.58 Ω4.400 54.23 Ω8.900%11.2112.36 dB/m
4 · Copper thickness31.50 µm56.58 Ω35.00 µm56.58 Ω38.50 µm56.58 Ω0.000%11.8011.80 dB/m
5 · Loss tangent0.01350 56.58 Ω0.01500 56.58 Ω0.01650 56.58 Ω0.000%11.2812.32 dB/m
6 · Frequency2.205 GHz56.58 Ω2.450 GHz56.58 Ω2.695 GHz56.58 Ω0.000%10.9412.64 dB/m

These exact endpoints are deterministic local sensitivity, not measurements, simultaneous corners, probability, or production yield.

If connector access disappears or the route must be shielded between planes, stripline can become the better system choice despite its via transitions. If the surface route remains simple and area is tight, microstrip may win. The structure name is the beginning of the decision, not the conclusion.

Ungraded review

Check your understanding

Answer each question in your own words, then reveal the model answer.

  1. 01Why can a trace built to the nominal width miss 50 Ω?
    Model answer

    Width is only one part of the cross-section. Finished dielectric height, copper thickness and etch shape, material Dk and construction, solder mask, nearby copper, return-path geometry, and transitions can all change the field solution. The evidence must describe finished geometry and its reference conductor, not only an artwork width.

  2. 02What makes a transmission-line mode more than a voltage trace?
    Model answer

    A mode is a repeatable electromagnetic field pattern with both signal and return current. Its impedance, velocity, and loss follow from that complete pattern and its boundaries. Removing or interrupting the intended return conductor changes the structure.

  3. 03Why does adding ground copper beside a trace not automatically create good GCPW?
    Model answer

    The coplanar grounds must be tied to the backing reference often enough and close enough to discontinuities to behave as one return system. Floating or sparsely connected rails can support slotline-like behavior and make the real mode different from the symmetric closed-form model.

  4. 04How are odd-mode, even-mode, differential, and common-mode impedance related here?
    Model answer

    For the declared symmetric two-line definitions, Zdiff = 2Zoo and Zcommon = Zoe/2. These relationships depend on the modal voltage/current normalization; the lesson reports the definitions with the numbers rather than treating one modal value as a universal single-ended impedance.

  5. 05What does the explorer’s dB/m output establish?
    Model answer

    Only a smooth-copper plus dielectric small-loss trend under explicit approximations. It helps compare direction with frequency, width, thickness, Dk, and loss tangent. It does not include roughness, current crowding, radiation, launches, packages, connectors, or measured laminate dispersion, so it is not a predicted insertion-loss budget.

  6. 06What evidence is still required before releasing the node feed?
    Model answer

    Fabricator-confirmed finished stackup and tolerances; frequency- and method-specific laminate data; a field-solved cross-section and 3D discontinuity/launch model; connector and package data; a coupon and calibration-plane plan; and measured correlation with acceptance limits.

Sources and model provenance

Normative standards are separated from formula provenance and application guidance. Accessed 5 September 2026. The calculator outputs are labeled Derived; the sensor-node decision is Illustrative.

Formula and material sources

Launch and measurement context