Antennas in Real Products

Failure: the open-board antenna collapses after final assembly

The same antenna and matching parts are fitted. Why did the installed link change?

Our illustrative engineering case is a battery-powered condition-monitoring node near 2.45 GHz. The open board looks promising. Final assembly adds plastic, a battery, fasteners and a mounting assembly. A technician leaves a programming cable attached during a repeat test. The product now has several physical configurations, not one context-free “antenna result.”

Before interpreting a graph, say the unit and mechanical revision, the feed/calibration plane, tune, cable route, surrounding object, orientation and environment. A lower reflection can mean more accepted power. It cannot tell us how much is dissipated or where the radiation goes.

All rows use IFA-like, ground 60 × 40 mm, clearance and object distance 5 mm, bypass, orientation A, matched real 50 Ω source and 1 mW incident at R2. b = 0.1 except the cable row, b = 0.6. Temperature validity is unknown. The wall row uses the battery/conductor proxy for an invented mounting assembly. It is not a wall-material prediction. Full per-state metadata appears in the evidence map.

The enclosed row improves S11 relative to open board while total efficiency falls slightly. The hand-like row has ideal match yet radiates only 40%. In explanatory notation its signed S11 is −∞ dB, ideal model; data uses “ideal match” and a null logarithm rather than a serialized nonfinite number.

The case is frozen as p06-m03-integration-v1. It does not continue measured performance of 06.1’s candidates or select an IFA/flex winner from 06.2. To interpret these independent teaching numbers, first expand the antenna boundary.

The antenna includes the product current distribution

If RF current flows on the board, why would the board be outside the antenna?

The designated antenna conductor launches one current distribution together with its return. An IFA-like arm can excite the finite ground and chassis. The feed location determines how strongly it couples to those available current distributions. Battery metal, a shield bond or a mounting plate can redirect or extend them. “Ground” names a circuit reference; it does not make every point on a finite GHz structure the same potential.

Think of the drawing as a boundary-condition record. If a ground edge is cut, a bond moves, or the chassis becomes connected through a screw, the allowable current paths change. This can change radiation resistance, stored energy and pattern without simply adding heat. A balanced two-arm radiator still interacts with its surroundings even when its intended return is on the other arm.

MIT’s Electromagnetics and Applications, Chapter 10, §§10.2–10.3 develops current-to-radiation and radiation-resistance foundations. It does not assign the product sensitivities in this lesson. Once the boundary is named, locate the plane at which its port is being described.

Feed placement, reference plane, matching, and return current

Was the improved match measured ahead of a loss, or at the antenna itself?

R1 is the component/package RF port. R2 is the product antenna-feed reference plane. In this lesson R2 sits immediately before an optional series teaching tuner; R2-A is its bare-antenna side. State this local mapping whenever reporting efficiency. A physical feed between R1 and R2 needs its own characterized transformation.

Place the tuner with a defined return and a documented antenna-side plane. Moving it along an electrically significant feed changes the impedance it sees. A fixed compensation that cancels open-board reactance may reinforce the error in another state. Matching-network topology and physical realization remain the subject of 03.6.

TI AN058, §6.2 describes feed-point connection and cable calibration in its development-kit context. Its magnitude-preservation shortcut must be restricted to a lossless matched line. IEEE 370-2020’s public scope covers interconnect data quality; no inaccessible de-embedding clause is used here. A correct plane still does not prove a balanced excitation.

Balanced antennas on unbalanced lines

Can the port look matched while the outside of the coax radiates?

Yes. Define both conductor currents in the same longitudinal direction. Opposite signs then describe the intended differential mode. Their shared part describes common mode. Reversing one current’s reference arrow without changing the equations would erase the meaning of this comparison.

In a supplied algebra example, I1 = +12 mA and I2 = −8 mA give Idiff = 10 mA and Icm = 2 mA. With +10 and −10 mA, Icm = 0. These are invented current inputs, entirely separate from the explorer’s dimensionless C. No equation converts its C to these milliamperes.

The intended coax mode uses the inner conductor and shield’s inner surface. Shield outer-surface current is a different path. A balun provides a balanced/unbalanced transition; a suitably placed choke can impede an unwanted common-mode path. Their frequency response, loss, impedance, orientation and physical placement require evidence. A choke can also change the impedance environment it was meant to stabilize.

TI AN058, §§6.2.1–6.2.3 explicitly treats cable attachment, fixed routing and ferrite placement as setup variables. Next, apply that same configuration discipline to conductors inside the product.

Batteries, displays, shields, cables, fasteners, and mounting metal

A metal mounting plate is added. Which result should we expect to change first?

There is no universal ordering. Induced currents can couple to the radiator, alter the return, change stored energy or redirect radiation. A conductor may serve as a useful reflector or counterpoise in one arrangement and produce a poor coverage direction in another. Its conductivity, spacing, size, bonds and orientation matter.

In our synthetic machine state Rrad = 20 Ω, Rloss = 10 Ω and X = −40 Ω. These supplied changes identify a teaching problem; they do not predict a plate’s electromagnetic behavior. A ground-size edit changes two coefficients in this toy, while a real ground edit may also change the pattern and loss in ways the toy omits.

Conductors change boundary currents. Materials that appear electrically insulating can change the fields just as consequentially.

Enclosures, adhesives, coatings, tissue/water, and tolerances

Can adding loss make S11 look better?

A dielectric changes electric-field storage through its material response. A lossy dielectric also converts some accepted RF energy to heat. An enclosure, adhesive thickness, coating, moisture or hand-like object can therefore change reactance, dissipation and current distribution at once. The resonance shift is geometry/material specific; “plastic always shifts down” is not a rule.

The canonical hand-like preset is a different comparison: it changes Rrad, Rloss, X and the pattern coefficient together. One menu selection is not one physical-factor experiment. Its perfect match comes with 0.6 mW antenna dissipation and 0.4 mW radiation.

Luomaniemi and colleagues, IEEE TAP 69(7), 2021, include a phone and hand together in their characteristic-mode analysis and distinguish radiated, reactive and loss power. That study supports including the user in the design problem. Its handset geometry, material model and excitation cannot calibrate this node’s coefficients or guarantee performance for a different grip.

A candidate clearance must also protect the receiver from fields produced inside its own product.

Clearance, keepout, and coupling to noisy circuits

The passive antenna evidence looks stable. Why does reception worsen when the display turns on?

The antenna is also a coupling path into a receiver. A clock, display interconnect or switching supply can couple through fields or a shared return, raising self-noise or creating a spur near the wanted channel. Passive S11 says little about that operating-state disturbance. Keepouts reserve a controlled electromagnetic region; they are not decorative empty rectangles.

As a first-party implementation example, Antenova’s flexiiANT Integration Guide (2 September 2021) makes cable placement, orientation and surrounding circuitry part of FPC integration. Use the selected antenna’s own evaluation conditions; its routing advice is not a universal keepout specification. In this lesson the clearance coefficient is an arbitrary 5 Ω/mm teaching slope, not a PCB rule.

Record placement intent now; defer detailed stack-up, via, routing and grounding execution to Path 10.3. Next we organize all these changes into states a tuning policy can actually support.

Multiple user/mechanical states need a tuning strategy

Should one tune cover every state, or should the product recognize different states?

First decide which states are normal use, which are corners and which belong only to the test setup. Open board is a correlation reference; enclosed, wall and machine are installation states; hand-like proximity is a use/corner hypothesis; a programming cable is a debug state unless the product ships with it. None of these labels establishes how often a user encounters a state.

Exact machine fractions: bypass reflection 1/4, ideal cancellation 1/16, lossy cancellation 81/1681. Tuner heat in the lossy row is 100/1681 mW, antenna heat 500/1681 mW and radiation 1000/1681 mW. Add reflected 81/1681 mW to recover exactly 1 mW.

Follow the integration experiment

Predict first, then compare open/enclosed → hand-like proximity → machine with ideal/lossy cancellation → cable on open board → rigid orientation B. Finally select a suspected mechanism and reveal its evidence request. Return to the named open-board reference after each comparison; never silently replace it with the last edited state.

The complete teaching model

Illustrative circuit/pattern teaching model — no product prediction. Model antenna-integration-perturbation/2.0, fixture p06-m03-integration-v1. IFA-like baseline (Rrad0, Rloss0, X0) = (30, 5, 20) Ω; flex-like = (60, 5, 20) Ω. These are hypothetical circuit parameters, not typical antenna-family values.

Permitted controls: ground length/width 20–100 mm in 1 mm steps; clearance 1–20 mm and distance 0.5–50 mm in 0.5 mm steps; b = 0–1 in 0.05 steps. Default is 60 × 40 mm, clearance/distance 5 mm, b = 0.1, plastic, IFA-like, bypass, A. A rigid orientation and the key-direction view are different controls. Object “none” has zero coefficients, so distance cannot change its result.

Tuner choices are bypass (Xt = Rt = 0); fixed open-board compensation (Xt = −20 Ω, Rt = 0); ideal cancellation (Xt = −X, Rt = 0); and lossy cancellation (Xt = −X, Rt = 2 Ω). These are series demonstrations at one frequency, not synthesized L/C or adaptive networks. Cancelling X leaves real mismatch whenever Rrad + Rloss + Rt differs from 50 Ω.

For the scalar pattern, x and z are components of a unit observation direction in the product frame. θ is measured from +z and φ from +x toward +y. t below is a dimensionless mixing variable, not time. Both shape terms integrate to 8π/3, so the same normalization applies to every mixture.

∫D dΩ = 4π and q ≥ 0. This is a mixture of power shapes, not coherent vector-field addition. It cannot produce polarization, axial ratio, cable-current radiation, SAR, or a measured efficiency. For B apply the inverse +90°-about-y rotation to a world direction before evaluating q: (xproduct, yproduct, zproduct) = (−zworld, yworld, xworld). The integral and power ledger stay unchanged.

The map tells you which observables to request. A reproducible build and a correlation plan turn that request into an engineering experiment.

Coupons, prototypes, simulation correlation, and configuration control

What does a simulation-to-measurement discrepancy mean if the cable or mechanical build changed?

It may mean the two sides describe different experiments. Freeze a reference build before fitting parameters. A coupon can isolate a feed launch, material stack or matching part, but it does not establish the final chassis current distribution. A representative prototype contains the physical boundaries needed for the question. Keep both kinds of evidence and label which one they provide.

Correlate complex impedance, efficiency and absolute directional gain; matching a resonance point alone is weak evidence. Use held-out states or units to test a model adjustment. If several material or geometric parameters can explain one trace, preserve that ambiguity and choose another observable. An uncalibrated field image can suggest a current path, but it does not become an ampere reading.

Now choose the smallest set of paired observations that can distinguish these competing explanations.

Define node integration variants and acceptance evidence

Which comparison would make you change the integration decision?

Write the hypothesis before the test. Preserve one complete reference and change an identified factor or a deliberately named bundle. Record mechanical revisions, ground and return, battery/display/shield/fastener/adhesive configuration, feed/tune/cable/choke, object geometry/material/distance, orientation, frequency grid, unit, temperature, instrument/method owner, uncertainty and intended comparison.

One “observation” here is the complete planned frequency/angular evidence set for a unit/state/orientation, not one independent population sample. The 36-row screen is six mechanical/use states × A/B × U01/U02/U03. The conditional 48-row follow-up explicitly selects wall assembly and machine mount as the two actual installation finalists, bypass and one physically documented fixed tune, 0/50 °C, A/B and the same three units. If those installations cease to be finalists, revise the plan and count before testing.

The local requested band is 2.401–2.499 GHz, matching 06.2’s 4% symmetric request; it does not replace 06.1’s separate 2.400–2.500 GHz and five-unit proposal. The toy calculates only 2.450 GHz. A temperature such as 0 °C or 50 °C is a requested test condition, never a hidden model input or a measured pass.

Pair cable-attached and cable-free open-board states on the same unit. For a hand-like phantom, record material properties and their source, shape, spacing, side, fixture and contact pressure if applicable. For wall/machine fixtures, record dimensions, electrical bonds, fasteners and whether orientation rotates the whole assembly. Do not treat A/B relative to a stationary mounting plate as rigid free-space rotation.

All thresholds are proposed engineering requirements, not achieved values or regulatory limits. Report expanded uncertainty with k = 2 and its justified coverage. If the relevant bound fails, investigate/redesign; if uncertainty, material validity or measurement coverage is missing, the decision remains unknown. A numeric toy output never passes a physical requirement. Common-mode current and receiver self-noise need application-specific limits agreed by their owners.

Keep the experiment identifiable

Repeat REF-M3 at run start/end and after reconnecting; add these checks beyond the 36/48 counts. Randomize unit/state order within practical blocks. Block temperature when thermal settling makes random switching impractical, then counterbalance unit and tuner order within each block. Record order and dwell so temperature drift is not mistaken for tuner benefit. Cable changes, mechanical reassembly and tuner changes made together are confounded unless separately paired.

Compare repeated readings and reconnects on one unit for setup repeatability; compare U01–U03 for unit variation. A conditional fixed tuning strategy survives only if the uncertainty-bounded match, radiation and coverage requirements all hold in the named states. If good match hides loss, reopen material/placement/current paths. If only a repeatable state-dependent mismatch remains, investigate switched/adaptive tuning with its detection and loss burden.

Antenna and channel evidence map

This is a complete local p06-evidence-map-v1 snapshot, available without an earlier visit or saved state. It preserves M01-INSTALL-UNKNOWN and adds six local synthetic rows plus the proposed plan. R2-TX and R2-RX, when used elsewhere, map to portfolio R2; R3 stays the detector/decision boundary. Incident, accepted, available receive, delivered receive, directional EIRP and integrated TRP are different quantities. No receive power, channel success or product certification is inferred here.