A −35 dB match at 28 GHz can still be poor engineering. If network Q equals 20, resonant 3-dB bandwidth is only ~5%.

At 28 GHz, that corresponds to roughly:

BW ≈ 1.4 GHz

That relationship follows the familiar resonator approximation: Qₗ ≈ f₀ / BW₃dB

Higher Q creates sharper resonance and narrower 3-dB bandwidth, but there is an important distinction. Resonator Q is not universally identical to matching-network bandwidth. Return-loss bandwidth also depends on load impedance and network topology. [Ref. 1]

That distinction becomes essential in modern RF hardware.

It matters across:

• mm-Wave 5G Infrastructure

• Massive MIMO

• Open RAN radios

• Phased Arrays

• Radar

• Satellite Communications

• high-power PA networks

The goal is not the deepest Smith-chart crossing. The goal is controlled power transfer across the operating envelope.

First: “Q” Is Not One Single RF Quantity

RF engineers often say: “This matching network has high Q.”

But which Q?

There are several related definitions. Confusing them can produce incorrect bandwidth conclusions.

1. Component Q

For an RF inductor:

Qcomponent ≈ |Im{Z}| / Re{Z}

or, below self resonance:

Q ≈ ωL / Rs

Coilcraft defines inductor Q using imaginary-to-real impedance ratio. It represents stored reactive energy relative to dissipated energy. Component Q is strongly frequency dependent. [Ref. 2]

Therefore:

Q at 2.4 GHz is not Q at 28 GHz.

2. Loaded Resonator Q

For a narrowband, single-resonance system:

Qₗ ≈ f₀/BW₃dB

This describes resonance sharpness. It includes external loading and internal loss. A higher loaded Q produces narrower resonance. [Ref. 1]

3. Impedance-Transformation Q

Matching networks have another useful Q concept.

Consider an L-network matching two real resistances.

For: RH > RL, the minimum transformation Q becomes:

Qm = √(RH/RL − 1)

MACOM derives this relationship for two-reactance L-matching networks. [Ref. 3]

This quantity reveals how severe the impedance transformation becomes and that directly affects bandwidth sensitivity.

What Impedance Ratio Does to Matching Q

Consider several 50 Ω transformations.

100 Ω → 50 Ω

Qm = √(100/50 − 1)

Qm = 1;

200 Ω → 50 Ω

Qm = √3

Qm ≈ 1.73;

500 Ω → 50 Ω

Qm = √9

Qm = 3;

1000 Ω → 50 Ω

Qm = √19

Qm ≈ 4.36

The trend is fundamental.

Larger impedance transformation requires larger matching Q.

MACOM's L-network derivation shows this directly. [Ref. 3]

More reactive energy must circulate around the transformation. That generally makes broadband matching increasingly difficult.

Pain Point 1: A Perfect Match and a Broadband Match Are Different Objectives

A matching network can force Γ = 0 at one selected frequency. That does not guarantee acceptable Γ across bandwidth.

This limitation is deeper than poor component selection. It originates in passive-network physics.

The Bode–Fano Limit

R. M. Fano formalized fundamental broadband matching limits in 1950. The theory concerns passive reactive matching of complex loads. [Ref. 4]

For a canonical parallel RC load:

∫ ln[1/|Γ(ω)|] dω ≤ π/(RC)

The equation creates an unavoidable trade-off. You can demand better matching or more bandwidth, but not arbitrarily maximize both.

Fano's original paper establishes those bandwidth-versus-tolerance limitations. Modern RF references continue using Bode–Fano as the theoretical ceiling.

Translate Bode–Fano Into Something More Practical

Assume approximately constant reflection magnitude across bandwidth.

Then:

Δω · ln(1/Γmax) ≤ π/(RC)

For a parallel RC load:

Qload = ω₀RC

Therefore:

FBW ≤ π/[Qload · ln(1/Γmax)]

This is a theoretical upper bound. It assumes an ideal passive lossless matching network. Real networks usually perform worse.

Now Put Numbers Into It

Assume Qload = 20;

Target: RL ≥ 10 dB

Then: |Γ|max = 0.316

The theoretical upper fractional bandwidth becomes approximately: FBW ≤ 13.6%

At 28 GHz: BW ≤ 3.82 GHz

Target: RL ≥ 15 dB

Then: |Γ|max ≈ 0.178

The theoretical upper fractional bandwidth becomes approximately: FBW ≤ 9.1%

At 28 GHz: BW ≤ 2.55 GHz

Target: RL ≥ 20 dB

Then: |Γ|max = 0.1

The theoretical upper fractional bandwidth becomes approximately: FBW ≤ 6.8%

At 28 GHz: BW ≤ 1.91 GHz

The exact numbers apply to this canonical RC example. They are not universal matching-network specifications, but the engineering message is extremely important.

Demanding deeper return loss consumes achievable bandwidth. Bode–Fano theory formalizes this compromise. [Ref. 5]

This Changes How We Should Read S₁₁

Imagine two designs.

Design A

Centre-frequency S₁₁: −35 dB

But band-edge S₁₁: −6 dB

Design B

Centre-frequency S₁₁: −17 dB

Worst-case S₁₁: −15 dB

across the complete required band.

Design A wins the screenshot.

Design B may win the system.

That depends on the actual RF requirement. A perfect centre-frequency match can therefore be misleading.

The Objective Is Not the Deepest Smith-Chart Null

The correct objective may instead be:

• worst-case transducer gain

• worst-case return loss

• usable fractional bandwidth

• PA efficiency

• noise figure

• EVM

• EIRP

• G/T

• scan-dependent active impedance

That is a fundamentally different optimization strategy.

Pain Point 2: At mm-Wave, the Matching Components Become the Parasitics

Consider a simple impedance transformation.

Assume: 50 Ω → 200 Ω

The minimum L-network transformation Q is: Qm = √3 ≈ 1.73

For one low-pass L-network realization: Xs ≈ Qm × 50 Ω

Therefore: Xs ≈ 86.6 Ω

The corresponding shunt reactance becomes: Xp ≈ 200/Qm

Therefore: Xp ≈ 115.5 Ω

Now Design That Network at 28 GHz

Suppose the series element is inductive: L = Xs/(2πf)

Therefore:

L ≈ 0.492 nH

Suppose the shunt element is capacitive: C = 1/(2πfXp)

Therefore:

C ≈ 49 fF

Stop there. A 49 fF matching capacitor is extraordinarily small.

At that point, the pad is part of the capacitor. The package is part of the capacitor. The via field is part of the network. The IC pad is part of the network.

Only 0.1 pF Already Looks Like 57 Ω

Consider: Cparasitic = 0.1 pF

At 28 GHz: |XC| = 1/(2πfC)

Therefore:

|XC| ≈ 56.8 Ω

That parasitic reactance is comparable to a 50 Ω system. It is also twice our illustrative 49 fF matching capacitance. This changes the design philosophy completely.

At mm-Wave, parasitics are no longer corrections around the matching network. They are matching-network elements.

A Real Component Benchmark Shows the Problem

Coilcraft's current 0201CT RF-inductor family reaches SRFs up to 35.2 GHz.

Its 0.6 nH component has typical SRF around 35.2 GHz.

But its 1.5 nH component has SRF around 20.3 GHz.

The 2.4 nH part falls further, near 15.8 GHz. [Ref. 6]

Therefore, at 28 GHz:

• 0.6 nH remains below nominal SRF.

• 1.5 nH is already above nominal SRF.

• 2.4 nH is significantly beyond nominal SRF.

That does not mean every above SRF component becomes useless instantly. It means the ideal inductor model has stopped being sufficient.

Q Must Also Be Known at the Actual Frequency

The same 0.6 nH Coilcraft device lists: Q ≈ 53 at 2.4 GHz. That value cannot simply be carried to 28 GHz.

Coilcraft explicitly states Q depends strongly on measurement frequency. It recommends evaluating inductors near their actual application frequency. [Ref. 2]

This is a critical procurement lesson.

A component marked “High-Q RF Inductor” is not a complete specification.

Ask: “What is its complex impedance at my operating frequency?”

Finite Component Q Converts Matching Into Dissipation

For an inductive element: Qcomponent ≈ |X|/Rs

Therefore:

Rs ≈ |X|/Qcomponent

Return to our illustrative: Xs = 86.6 Ω

If actual component Q were: Q = 20, then: Rs ≈ 4.33 Ω

If: Q = 50, then: Rs ≈ 1.73 Ω

These are illustrative values. They are not Coilcraft 28 GHz specifications.

What Does 4.33 Ω Mean in a 50 Ω Environment?

If treated only as a series resistance:

|S₂₁| ≈ 2Z₀/(2Z₀ + Rs)

For: Rs = 4.33 Ω, the simple attenuation becomes approximately: 0.37 dB

For: Rs = 1.73 Ω, it becomes approximately: 0.15 dB

Actual matching-network loss depends on current circulation. Topology and impedance transformation also matter, but fractions of a decibel are already expensive.

Why 0.3 dB Matters in Massive MIMO

Consider a simplified 256-channel RF aperture. Suppose every channel delivers 1 W.

An extra 0.3 dB loss corresponds to roughly 6.7% power reduction.

The ideal aggregate RF power is 256 W.

After 0.3 dB loss ≈ 239 W, approximately 17 W never reaches the intended RF load.

This calculation is intentionally simplified, but it shows how small channel losses scale.

Capacitors Face the Same Problem

Real capacitors include:

• ESR

• ESL

• pad capacitance

• package inductance

• resonant behaviour

KYOCERA AVX currently offers thin-film Accu-P capacitors for operation through 60 GHz. The 0201 family spans approximately 0.05–22 pF. [Ref. 7]

These components target low ESR and high-Q microwave applications. But even excellent components require correct mounting models.

At 28 GHz, the PCB transition may dominate component behaviour.

The Ground Connection Can Be Worse Than the Capacitor

A shunt matching capacitor needs an RF ground. That ground is not mathematically zero impedance.

Analog Devices reports approximately 0.75 nH for one 1.6-mm via example. At 2.5 GHz, that corresponds to roughly 12 Ω reactance. [Ref. 8]

The same lumped-inductance extrapolation gives approximately: 132 Ω at 28 GHz. However, that extrapolation becomes physically incomplete.

At 28 GHz, the via requires distributed electromagnetic modelling. That is exactly the point. A single grounding via cannot automatically represent “RF ground.”

Another Via Benchmark Shows Why Full-Wave Analysis Matters

Analog Devices gives another PCB example containing roughly 1.2 nH via inductance and 0.5 pF parasitic capacitance. [Ref. 9]

Their simple LC resonance occurs near 6.5 GHz.

Therefore, treating that structure as pure inductance at 28 GHz fails. The structure has become distributed and resonant. So has your matching network.

Pain Point 3: The Load You Matched Yesterday May Not Exist Today

Passive matching examples often assume one fixed impedance. Real RF hardware rarely provides one.

A PA output impedance changes with:

• frequency

• output power

• compression

• bias

• temperature

• harmonic termination

An LNA optimum impedance may also differ from conjugate power match. Noise matching and gain matching can require different impedances.

Antenna impedance changes through its electromagnetic environment and phased arrays make this problem even harder.

The Phased-Array Load Moves With Scan Angle

Mutual coupling alters active antenna impedance. That active impedance changes with scan direction.

IEEE research identifies scan-dependent active impedance as a major limitation. The variation can reduce gain and produce scan blindness. A matching network optimized only for broadside may therefore fail off-axis. [Ref. 10]

This changes what “optimal Q” means.

A 2025 PA Benchmark Shows the System Consequence

A 2025 IEEE ISSCC study examined mm-Wave phased-array PA loading. The work considered large-angle scanning beyond approximately ±60°. [Ref. 11]

Array active mismatch exceeded approximately: 4:1 VSWR under the reported wideband scanning conditions.

For a conventional common-source PA, power gain varied by over 7 dB across those changing load conditions. That is not merely an antenna problem.

It is an amplifier-to-antenna impedance-interface problem.

This Is Why a Narrow Deep Match Can Be Dangerous

Imagine optimizing the PA output for one impedance.

At broadside, Zactive ≈ Zopt. S₁₁ looks exceptional. Now scan to 60°.

Mutual coupling changes the antenna's active impedance. The carefully tuned high-Q network transforms the new load differently. PA gain and phase can then move strongly.

This can corrupt:

• EIRP

• EVM

• PAE

• channel phase

• DPD behaviour

• beam calibration

A deep centre-state match has not protected the system.

Wide-Angle Impedance Matching Is Now an Antenna-Level Discipline

A 2024 IEEE phased-array study used artificial dielectric sheets. Its goal was reducing active-reflection variation across scan angle. The compensated array maintained active VSWR below two over broad scans. [Ref. 12]

Reported ranges reached approximately:

• ±45° E-plane

• ±65° H-plane

• ±80° diagonal plane

That is a useful architectural lesson. Sometimes the best matching network is not beside the PA. It may be integrated into the electromagnetic aperture.

This Is Also a Bode–Fano Problem

High-Q antennas are intrinsically difficult to match broadly. Fano's limits cannot be defeated using clever optimization alone.

More matching sections can distribute reflection across frequency. They cannot remove the underlying physical bound.

Modern treatments summarize the result clearly: higher-Q reactive loads are harder to match over wide bandwidth. [Ref. 4]

That applies to:

• electrically small antennas

• resonant patches

• high-impedance devices

• narrowband PA loads

• filters

• tunable RF networks

Antenna Q Adds Another Important Layer

Antenna Q describes stored electromagnetic energy relative to radiated energy. It is widely used to quantify antenna bandwidth limitations. For sufficiently narrowband antennas, Q relates closely to achievable bandwidth. [Ref. 13]

Therefore, matching-network Q cannot be considered independently from antenna Q.

A high-Q antenna and high-Q network compound sensitivity. That becomes especially important for compact mmWave apertures.

What Changes at mm-Wave?

At lower microwave frequencies, lumped matching remains practical. At mm-Wave, physical interconnects become electrically large.

The matching network increasingly becomes:

• transmission-line length

• package lead

• bondwire

• via field

• pad capacitance

• ground return

• antenna-feed transition

At this point, schematic and layout separate only conceptually. Electromagnetically, they are one structure.

This Is Where Altium and Allegro Stop Being Enough

Altium and Allegro remain critical implementation tools.

They enforce:

• geometry

• clearances

• stack-up

• component placement

• manufacturing constraints

But DRC cannot calculate matching-network Q accurately. It cannot determine distributed electromagnetic parasitics alone. That requires full-wave analysis.

HFSS and CST Must See the Actual Matching Structure

Include:

• component pads

• solder fillets

• signal vias

• ground vias

• antipads

• package models

• transmission lines

• nearby copper

• enclosure boundaries

At mm-Wave, those structures modify both reactance and loss. Extract their broadband S-parameters, then bring them into circuit-level optimization.

Circuit and EM Co-Simulation Is the Correct Workflow

A robust workflow becomes:

Device model → Circuit match → EM extraction → Co-simulation

Then iterate. Do not optimize ideal inductors first. Then “add layout” afterwards.

At mm-Wave: layout is part of the matching topology.

The Strategic Engineering Solution

Top-tier RF teams should design usable-band Q, not centre-frequency perfection.

1. Start With the Required Impedance Trajectory

Do not start with component values.

First understand: ZL(f, P, T, θ), where applicable.

For a PA, include:

• frequency

• output power

• bias

• temperature

For a Phased Array, include:

• scan angle

• polarization

• mutual coupling

For an LNA, include:

• noise optimum impedance

• gain optimum impedance

One Smith-chart point is not enough.

2. Calculate Transformation Q Before Choosing Components

For real-to-real L matching: Qm = √(RH/RL − 1).

Use this as an early difficulty indicator. [Ref. 3]

If Q becomes excessive, reconsider the architecture. Do not simply increase optimization iterations.

Possible alternatives include:

• transformer matching

• transmission-line transformation

• multi-section matching

• distributed matching

• impedance choice at device level

Architecture can reduce the transformation burden.

3. Check the Bode–Fano Ceiling Early

Before promising broadband S₁₁, estimate whether the target is physically reasonable.

Ask:

• What is the load's reactive Q?

• What return loss is truly required?

• What fractional bandwidth is required?

• Is a passive lossless network theoretically sufficient?

Bode–Fano provides the upper-bound framework. This can prevent weeks of impossible optimization. [Ref. 5]

4. Define Return Loss From System Requirements

Do not automatically demand S₁₁ < −20 dB across everything. Translate return loss into reflected power.

Remember:

−10 dB → 10% reflected power

−15 dB → 3.16%

−20 dB → 1%

Then decide what the system actually needs. A wider -15 dB solution may outperform a narrow -35 dB match.

5. Use Real Component Models

Never trust only L = 0.5 nH or C = 0.1 pF.

Request:

• S-parameter files

• Q versus frequency

• SRF

• tolerance

• temperature coefficient

• package geometry

Coilcraft explicitly recommends testing RF inductors near application frequency. SRF and Q are frequency-dependent parameters. [Ref. 2]

6. Prefer Distributed or Hybrid Matching at mm-Wave

A 50 fF intended capacitor invites parasitic domination. A 0.5 nH intended inductor approaches package-scale inductance.

Use transmission-line sections when appropriate. Use package parasitics intentionally where possible.

Hybrid distributed-lumped matching can improve model fidelity. The correct implementation depends on available area and bandwidth.

7. EM-Extract the Complete PCB Transition

Use HFSS or CST Microwave Studio. Include the matching network and return paths.

Also include:

• Rogers substrate properties

• copper roughness

• dielectric thickness

• connector launch

• via fencing

Do not extract only the signal trace. Extract the complete electromagnetic current path.

8. Run Process Corners, Not Only Nominal Optimization

Include:

• component tolerance

• Dk tolerance

• dielectric thickness

• etch variation

• copper roughness

• package parasitics

• solder variation

• temperature

For active devices, also include:

• bias

• compression

• device process

• load variation

The nominal S₁₁ minimum is only one statistical point.

9. Optimize Worst-Case System Metrics

For transmitters, optimize:

• transducer gain

• PAE

• EVM

• output power

• EIRP

For receivers, optimize:

• noise figure

• gain

• linearity

• G/T

For Phased Arrays, optimize:

• active VSWR

• realized gain

• channel phase

• scan performance

System-level performance should drive network Q.

10. Validate at the Correct RF Reference Plane

Use calibrated VNA measurements. Move the reference plane toward the actual matching network.

TRL calibration becomes extremely valuable for PCB structures.

Then compare:

• S₁₁

• S₂₁

• phase

• group delay

Do not tune the model before verifying measurement reference planes.

For PA Hardware, Add Load-Pull

S-parameters describe small-signal network behaviour. Power amplifiers are nonlinear. Their optimum load changes with drive level.

Load-pull maps power, efficiency, and gain across complex loads. A matching network should therefore transform the relevant load region, not merely hit 50 Ω under small-signal conditions. [Ref. 2]

For LNA Hardware, Power Match May Be Wrong

Maximum available gain uses conjugate matching. Minimum noise figure generally follows another optimum impedance.

The network must balance:

• noise

• gain

• stability

• bandwidth

A deep S₁₁ null is not automatically the optimal LNA design. This is another reason Q should follow system objectives.

For 5G Infrastructure, Linearity Changes the Decision

Open RAN radios handle broadband modulated signals. The matching network affects more than output power.

It influences:

• compression

• EVM

• ACLR

• DPD behaviour

• efficiency

High-Q output matching can produce frequency-dependent load modulation. That becomes increasingly undesirable across wide instantaneous bandwidth.

For Aerospace & Defense, Temperature Cannot Be Ignored

Radar hardware often spans large environmental ranges. A high-Q match amplifies component and material sensitivity. Small reactance changes can shift centre frequency substantially.

That creates:

• gain ripple

• phase ripple

• power variation

• calibration drift

MIL-STD-461 addresses EMC rather than matching-network Q. But narrow RF margin can indirectly reduce EMC design flexibility.

Rigid-Flex RF Makes Q Even Harder to Control

Rigid-Flex RF structures change geometry across mechanical boundaries. The matching environment can include:

• rigid laminate

• flex polyimide

• coverlay

• connector transitions

Those regions alter distributed capacitance and inductance. A schematic match can therefore change after mechanical integration.

For mm-Wave rigid-flex hardware, validate the final assembled geometry electromagnetically.

Satellite Communications Adds EIRP and G/T Consequences

A matching-network loss on transmit reduces EIRP. The same loss before an LNA damages G/T.

For Ground stations, several tenths of decibel already matter.

For LEO payloads, thermal and DC budgets are tighter.

High-Q networks also create manufacturing sensitivity. That can increase calibration and production complexity.

The B2B Procurement Implication

A procurement specification should not state only: “S₁₁ < −20 dB at centre frequency.” That is inadequate.

For advanced RF hardware, request:

• S₁₁ across required bandwidth

• transducer gain

• insertion loss

• loaded-Q or bandwidth definition

• component Q versus frequency

• SRF

• tolerance data

• temperature performance

• EM-extracted transition models

• production-yield evidence

For PA modules, also request:

• PAE

• EVM

• load-mismatch sensitivity

• output phase variation

For Phased Arrays, add:

• active S₁₁ versus scan

• EIRP versus scan

• channel-phase variation

• thermal calibration drift

For Satellite Communications, add:

• feed loss

• G/T impact

• EIRP impact

This changes supplier qualification dramatically.

The Summary

A matching network has several different Q mechanisms.

For a narrow resonance: Qₗ ≈ f₀/BW₃dB.

Therefore: Qₗ = 20 → ~5% 3-dB bandwidth

At 28 GHz: BW ≈ 1.4 GHz

But matching bandwidth is also constrained by load reactance. Bode–Fano places a fundamental ceiling on broadband passive matching. Impedance transformation adds another Q requirement.

For a real-to-real L-match: Qm = √(RH/RL − 1).

Then real hardware adds:

• finite component Q

• self resonance

• package parasitics

• PCB parasitics

• material tolerance

• active-load movement

At 28 GHz, even: 0.1 pF → ~56.8 Ω reactance, and a 50-to-200 Ω example needs only ~49 fF shunt capacitance.

That is why mm-Wave matching becomes electromagnetic co-design.

So the correct engineering question is not:

“How deep can I make S₁₁ at f₀?”

It is:

“What network Q delivers the required system performance everywhere?”