Electronics Guide

Why Conducted Measurement Runs Out

Three forces push millimeter-wave characterization off the cable and into the chamber, and they act together rather than in isolation.

The first is integration. Antenna-in-package construction places the radiating elements on the same laminate or redistribution layers that carry the die, and antenna-on-chip designs put them on the silicon itself. A 28 GHz phased array with sixteen elements has sixteen feed points, each a fraction of a millimeter across, all buried under the mold compound. Even where a probe pad exists, landing a probe on it perturbs the very structure being measured. The array is not an accessory to the radio; it is part of the radio, and the specification describes their combined behavior.

The second is loss. Coaxial attenuation climbs steadily with frequency, the conductor term in proportion to the square root of frequency and the dielectric term in proportion to frequency itself, so a meter of high-performance flexible microwave cable that costs a fraction of a decibel at 1 GHz costs several decibels at 40 GHz, and costs more again after every flexure. Connector families shrink to keep higher-order modes out: the 2.92 mm interface reaches 40 GHz, the 2.4 mm interface 50 GHz, the 1.85 mm interface 67 GHz, and the 1.0 mm interface 110 GHz. Each is more fragile and more torque-sensitive than the last, and each mating event is a fresh opportunity for error. Threading a cable to every port of a production device is impractical even when the ports exist.

The third is the specification itself. Regulators and standards bodies write millimeter-wave requirements in radiated terms because the user experiences a radiated field. A conducted power measurement at a single element tells almost nothing about what an eight-beam array delivers toward a base station. The compliance question is what leaves the device in a given direction, and only a radiated measurement answers it.

Far-Field Conditions and Range Length

An antenna radiates a field whose character depends on distance. Close to the aperture lies the reactive near field, where stored energy dominates and the field does not propagate. Beyond it lies the radiating near field, or Fresnel region, where the field propagates but the pattern still changes with distance because different parts of the aperture arrive with different phases. Only in the far field, or Fraunhofer region, does the angular pattern become independent of range, so that a measurement made at one distance predicts behavior at any greater distance.

The Far-Field Criteria

Far-field measurement conventionally requires that the range length R satisfy several conditions at once. The dominant one for electrically large apertures is the Fraunhofer criterion, R greater than or equal to 2D squared divided by lambda, where D is the largest dimension of the smallest sphere enclosing the radiating structure and lambda is the free-space wavelength. At that separation the path from the aperture edge to the probe is longer than the path from the aperture center by one-sixteenth of a wavelength, which is a phase error of 22.5 degrees; the Third Generation Partnership Project, hereafter 3GPP, states the same figure directly in its uncertainty analysis, defining the phase curvature at a distance of 2D squared over lambda as 22.5 degrees. Auxiliary conditions guard against the reactive near field and against amplitude taper across the test volume, requiring the range to be long compared with the wavelength and long compared with the aperture itself. Standards state those secondary limits differently. The 3GPP radiated two-stage method, for instance, takes the minimum range length as the largest of three quantities: the test-zone radius plus twice the wavelength, for the reactive near field; three times the test-zone diameter, for amplitude taper; and the test-zone radius plus twice the squared radiating aperture divided by the wavelength, for phase curvature. For a small element at low frequency the auxiliary conditions govern; for a handset-sized aperture at millimeter-wave frequencies the Fraunhofer term dominates overwhelmingly.

What the Arithmetic Demands

The quadratic dependence on D and the inverse dependence on lambda make range length grow explosively. At 28 GHz the free-space wavelength is 10.7 mm. A five-centimeter array module therefore requires about 0.47 m of range, which is comfortable. Measure the same array as installed in a fifteen-centimeter handset at 39 GHz, where lambda is 7.7 mm, and the requirement rises to roughly 5.9 m. Qualify a thirty-centimeter test volume at 43.5 GHz, the upper edge of band n259, and the direct far-field distance approaches 26 m. That is nowhere near the ceiling: FR2-1 runs to 52.6 GHz and FR2-2 to 71 GHz, where the same test volume would demand 32 m and 43 m respectively. A chamber of that size, lined with absorber and equipped with a precision positioner, is an expensive building rather than a piece of laboratory equipment, and the free-space path loss over that distance consumes the dynamic range needed to measure sensitivity.

A crucial subtlety concerns the choice of D. If the requirement is to characterize only the array, D is the array dimension. If the requirement is to characterize the device as a user holds it, including the influence of the chassis, display, and hand, then D is the dimension of the whole device. 3GPP distinguishes the two as the black-box and grey-box approaches. The black-box view treats the entire device as the radiating aperture and is the conservative basis for conformance. The grey-box view rests on a vendor declaration of where the active antennas sit, so that only that part of the device must lie inside the quiet zone; it permits a shorter range and a smaller chamber, and it is useful during development. Choosing the wrong one is a common and expensive error, because a range built for the array alone will not certify the product.

Range Architectures

3GPP studied millimeter-wave test methods in Technical Report 38.810, and three test methodologies are now permitted for the FR2 user-equipment radio-frequency test cases of the conformance specification TS 38.521-2: direct far field, indirect far field, and near-field to far-field transformation. Each solves the range-length problem differently, and each carries a distinct set of strengths and limitations.

Direct Far Field

The direct far field method is the literal implementation: place the device at a true far-field distance from a probe antenna inside an anechoic chamber, and rotate it to sample the sphere. Nothing intervenes between device and probe, so the method is conceptually transparent and the error budget is short. It measures transmitter and receiver performance, modulated signals, and protocol-level behavior without qualification. Its weakness is size. The chamber must accommodate the full Fraunhofer distance, and the free-space path loss over that distance, which reaches about 61 dB over one meter at 28 GHz and about 75 dB over five meters, directly erodes the sensitivity floor available for receiver testing. Direct far field ranges are therefore practical for small modules and for the lower part of the band, and increasingly impractical as either aperture or frequency grows.

Indirect Far Field: the Compact Antenna Test Range

The indirect far field method, implemented almost universally as a compact antenna test range, or CATR, synthesizes far-field conditions in a small chamber. A feed horn illuminates an offset parabolic reflector from the reflector focus. The reflector collimates the spherical wave from the feed into a locally planar wave, and the device sits in that planar region. Because a plane wave has, by definition, uniform amplitude and constant phase across its cross section, the device experiences the same field it would see at infinite distance while occupying a chamber only a few meters across.

The offset geometry keeps the feed and its support structure out of the collimated beam, which would otherwise block and scatter it. Diffraction from the reflector rim is the dominant residual error, and manufacturers control it by serrating or rolling the reflector edges so that diffracted energy scatters away from the test volume rather than into it. Reflector surface accuracy is the other governing tolerance: surface deviations must remain a small fraction of a wavelength, and at 40 GHz a wavelength is only 7.5 mm, so the tolerance falls to tens of micrometers over an aperture of a meter or more.

The compact range is more than merely popular. 3GPP designates the indirect far field method the reference methodology for FR2 conformance and derives from it a threshold measurement uncertainty for each test case; a direct far field or near-field system may be used only where its own total expanded uncertainty falls at or below that threshold. The practical reasons run the same way, since a compact range delivers a large usable test volume, a modest path loss, and a chamber footprint that fits in a laboratory.

Near-Field to Far-Field Transformation

The third method measures amplitude and phase on a surface close to the device and computes the far-field pattern numerically. Sampling on a sphere, a cylinder, or a plane yields a set of modal coefficients from which the far field follows by a transformation. Spherical scanning suits devices that radiate in all directions; planar scanning suits directive apertures that radiate into a half space.

Near-field transformation is efficient and compact, and it is the established technique in antenna engineering. Its constraints matter, however. It requires accurate phase data, which demands a coherent receiver and a phase reference routed to the probe, and phase accuracy at millimeter-wave frequencies is difficult when cables move. It requires a probe positioner whose mechanical accuracy is a small fraction of a wavelength. It also assumes the device transmits a stable, repeatable signal throughout the scan, which is awkward when the device is an active radio whose transmitter power control and beam selection may drift. For these reasons near-field methods excel at pattern and gain work and are used more cautiously for radiated transmit power and sensitivity conformance.

Plane-Wave Generators

A fourth architecture, which is not among the three methodologies permitted for FR2 user-equipment radio-frequency conformance but is increasingly deployed for development and pre-compliance work, replaces the reflector with an array. A plane-wave generator, sometimes called a plane-wave converter or synthesizer, uses a two-dimensional array of probe elements whose amplitude and phase weights are computed to synthesize a planar wavefront in a test volume placed only a short distance away. Because the wavefront is formed electronically rather than optically, the system needs no large reflector and can be reconfigured, and the chamber becomes very compact. The cost is complexity: dozens to hundreds of coherent channels, each requiring calibration, and a synthesized field whose quality depends on the accuracy of those weights across the full frequency range.

The Quiet Zone and Its Qualification

Every range architecture defines a quiet zone: the volume in which the field is sufficiently close to an ideal plane wave that a measurement made there is trustworthy. The quiet zone, not the chamber, is the specification that matters. A device larger than the quiet zone cannot be measured correctly no matter how good the instrumentation.

Quiet-zone quality is expressed through four field-uniformity metrics. Amplitude taper is the systematic variation of field strength across the volume, arising from the illumination pattern of the feed. Amplitude ripple is the oscillatory component superimposed on the taper, produced by reflector edge diffraction, chamber reflections, and feed-reflector interaction. Phase curvature, or phase variation, measures the departure from constant phase across the volume. Cross-polarization measures the unwanted orthogonal field component, which corrupts polarization-sensitive measurements. 3GPP collapses the four into a single quality-of-quiet-zone contributor in its uncertainty budgets, and that contributor is not negligible: for a quiet zone of thirty centimeters or less it carries a standard uncertainty of 0.6 dB from 23.45 to 40.8 GHz and 0.7 dB from 40.8 to 44.3 GHz. The degradation toward the top of the band is written into the budget rather than left as a caveat.

Qualification is performed by field probing. A small probe antenna is scanned through the volume on precision axes while amplitude and phase are recorded, producing maps that reveal ripple structure and expose reflections from cables, positioner surfaces, and absorber discontinuities. The exercise must be repeated at multiple frequencies across each band, because ripple patterns are strongly frequency dependent, and repeated after any physical change to the chamber. Test plans specify a quiet-zone size to match the devices in scope. 3GPP now recognizes a family of FR2 quiet-zone sizes, at twenty, thirty, forty, and fifty-five centimeters, and maps the maximum permitted device size and the applicable uncertainty threshold onto them; a device larger than fifty-five centimeters has no applicable threshold at all, because no larger quiet zone has yet been defined. The test zone for standardized multiple-input multiple-output OTA work, in both FR1 and FR2, is twenty centimeters.

Absorber behaves differently at millimeter-wave frequencies than at the frequencies EMC engineers know best. Pyramidal absorber whose tips are many wavelengths long performs very well, so reflectivity is rarely the limiting factor, and thinner absorber suffices than a low-frequency chamber requires. The practical hazards shift to mechanical ones: broken or bent tips scatter energy, accumulated dust changes surface properties, and any exposed metal fastener, cable clamp, or positioner edge becomes a specular reflector at a wavelength of a few millimeters. Millimeter-wave chamber discipline is largely a matter of covering everything metallic and keeping the absorber intact.

Path-Loss Calibration

An over-the-air measurement is meaningful only after the entire path between the device position and the instrument has been characterized. That path includes the free-space or reflector-mediated propagation, the probe antenna, every cable and connector, any switch matrix, and any frequency converter. Its loss is large and frequency dependent, and it must be known to a fraction of a decibel if the result is to be traceable.

The standard technique is gain comparison, also called gain substitution. A reference antenna of known, calibrated gain is placed at the device position, and the received level is measured. The difference between the known radiated quantity and the measured level yields the total path loss, which is then applied as an offset to subsequent device measurements. The reference antenna's calibration is itself established by a traceable method, most commonly the three-antenna technique, which determines the absolute gain of three antennas from three pairwise transmission measurements without needing any of them to be known in advance.

Two practical concerns dominate. The first is drift. Millimeter-wave cables change loss and, more importantly, phase with temperature and flexure, and a calibration performed in the morning may not describe the path in the afternoon. Laboratories manage this by minimizing cable movement, stabilizing chamber temperature, routing the calibration through the same switch paths used for measurement, and re-verifying at defined intervals. The second is dynamic range. Because path loss is large, receiver-sensitivity testing must place enough power at the device to overcome it, and transmitter measurements must resolve signals well below the noise floor of an ordinary spectrum analyzer. Many chambers therefore place up-converters and down-converters immediately behind the probe or feed, carrying only an intermediate frequency down the long cable runs while distributing a common local-oscillator reference. This preserves signal-to-noise ratio but adds converter gain flatness and local-oscillator phase noise to the error budget.

Radiated Figures of Merit

Because no conducted reference plane exists, millimeter-wave specifications are written in radiated quantities. Four appear in nearly every test plan: effective isotropic radiated power, effective isotropic sensitivity, total radiated power, and spherical coverage.

Directional Quantities

Effective isotropic radiated power, or EIRP, is the power that an isotropic radiator would have to emit to produce the observed field strength in a particular direction. It is the natural transmit metric for a beamforming device, since the whole point of the array is to concentrate power in a chosen direction. Its receive counterpart is effective isotropic sensitivity, or EIS: the EIRP that a source at the measurement distance must radiate for the device to meet a defined throughput or error-rate threshold in a given direction. Expressing sensitivity as an equivalent isotropic power makes it directly comparable with the transmit figures, and, unlike them, a lower number is the better one.

Integrated Quantities

Total radiated power, or TRP, integrates radiated power over the complete sphere and over both orthogonal polarizations. It answers a different question from EIRP: not how well the device concentrates power, but how much power it emits in total, which is what regulators care about for exposure and interference. Total radiated sensitivity, sometimes written TRS or reported as an averaged EIS, is the corresponding receive integral.

Spherical Coverage

A phased array cannot steer everywhere. Elements at the edge of the array scan poorly, the chassis blocks some directions, and a handset may be held in any orientation. Reporting only peak EIRP would reward a device that is superb in one direction and useless everywhere else. Spherical coverage solves this by measuring EIRP over the whole sphere, forming the cumulative distribution function of the results, and reporting the value at a specified percentile. The requirement then states that the device must achieve at least a given EIRP over at least a given fraction of all directions.

The published cellular requirements illustrate the structure. For a power class 3 device, the handheld category and the default class, TS 38.101-2 sets the minimum peak EIRP at 22.4 dBm in bands n257, n258, and n261, which together span 24.25 to 29.5 GHz; at 20.6 dBm in n260, at 37 to 40 GHz; at 18.7 dBm in n259, at 39.5 to 43.5 GHz; at 16.0 dBm in n262, at 47.2 to 48.2 GHz; and at 14.1 dBm in the unlicensed band n263, at 57 to 71 GHz. The corresponding spherical-coverage figures, in the same order, are 11.5, 8.0, 5.8, 2.9, and 2.3 dBm, all defined at the 50th percentile of the cumulative distribution. The ceiling comes from regulation rather than from the array: maximum EIRP is 43 dBm and maximum total radiated power 23 dBm.

The percentile itself moves with the power class, and that is where the requirement reveals what it assumes about the product. Seven power classes are now defined: fixed wireless access terminals as classes 1 and 5, vehicular devices as class 2, handhelds as class 3, high-power non-handheld equipment as class 4, roof-mounted high-speed-train equipment as class 6, and reduced-capability devices as class 7. Spherical coverage is evaluated at the 85th percentile for the fixed wireless access classes, the 60th for vehicular, the 50th for handheld and reduced-capability, and the 20th for high-power non-handheld. A fixed terminal aimed once at a base station needs only a small part of the sphere to be good, so it is judged at a point where fifteen percent of directions must exceed the limit; a high-power non-handheld device is judged where eighty percent must.

Sampling the Sphere

Every integrated or distributional quantity depends on how the sphere is sampled. A constant-step-size grid takes equal angular increments in theta and phi, which oversamples the poles. Constant-density grids equalize solid angle per sample instead, either by varying the number of azimuth points with elevation or by placing points with a charged-particle repulsion model or a golden-spiral construction. The distinction is not cosmetic: 3GPP requires a mean-error correction when total radiated power is integrated by quadrature from constant-step-size data, and none when a constant-density grid is used. At each grid point the receiver must record two orthogonal polarization components, because total power is the sum of the theta and phi contributions. A fifteen-degree constant-step-size grid gives thirteen elevations and twenty-four azimuths, and because every azimuth coincides at the two poles it has 266 unique orientations; halving the step to 7.5 degrees raises that to 1,106, roughly a factor of four. Grid choice is thus a direct trade between accuracy and test time, and test plans specify it explicitly so that results from different laboratories remain comparable.

Beam Management and the Test-Time Problem

A millimeter-wave device does not radiate one pattern; it radiates a codebook of them. A handset array may support a few dozen beam states, and a base station many more. Conformance is defined per beam or over a defined set of beams, so the spherical sweep must be repeated for each state of interest. The measurement count multiplies as the product of grid points, polarizations, beam states, frequencies, and channel bandwidths. Take the 266-point fifteen-degree grid, a dozen beam states, three channels, and three bandwidths, and the count passes twenty-eight thousand; at a second or two per point, before any allowance for positioner settling, that is eight to sixteen hours of chamber occupancy for a single configuration, and a conformance campaign covers many.

Three mechanisms keep the problem tractable. The first is beam locking. Test specifications define a mode in which the device holds a commanded beam rather than re-selecting it as the positioner rotates, which is essential because an adaptive device would otherwise chase the probe and make the measured pattern meaningless. The second is a two-stage search: a coarse grid locates the approximate beam peak, and a fine grid is applied only in its neighborhood. The third is hardware parallelism, including multiple probes that sample several directions at once, fast switching, and instruments that capture wide bandwidths in a single acquisition. Because measurement throughput determines the cost of every device that passes through the chamber, speed is treated as a first-order design requirement of the test system rather than a convenience.

Demodulation, Radio Resource Management, and MIMO

Radiated power and sensitivity are only part of conformance. Demodulation performance and radio resource management behavior must also be verified over the air, and these require the chamber to reproduce not just a signal but a propagation environment.

The reference method for millimeter-wave multiple-input multiple-output OTA testing is the three-dimensional multi-probe anechoic chamber, or MPAC. A channel emulator drives each probe with an appropriately faded, delayed, and Doppler-shifted component of a synthesized channel, and the superposition of those components inside the test zone recreates the angular, temporal, and polarimetric structure of a realistic multipath channel, so that the device experiences something resembling a street or an office rather than a single clean plane wave.

The FR2 geometry is not the enveloping ring of probes the name suggests. Surrounding a test zone with enough coherent millimeter-wave probes to reproduce a channel from every direction is prohibitive, so 3GPP permits a sectored arrangement: six dual-polarized probes placed on a sector at a range length of at least 0.75 m from the center of the test zone, with the device reoriented to bring other angles of arrival into that sector. The channel models come from the clustered delay line family, the indoor office CDL-A and urban microcellular CDL-C non-line-of-sight cases serving as the FR2 references, at device speeds of 3 km/h indoors and 12 km/h in the urban microcellular case. The test zone is twenty centimeters, and the black-box approach is adopted outright, the physical center of the device being placed at the center of the test zone.

Multi-probe systems are demanding. Every probe path must be amplitude- and phase-calibrated relative to the others, the probe geometry constrains which angular spreads can be reproduced, and the channel emulator must supply as many coherent output paths as there are probes and polarizations. The reward is that beam-management algorithms, beam-switching latency, and link adaptation can be exercised realistically, which no single-probe range can do.

Positioners, Fixtures, and Mechanical Accuracy

Mechanical error becomes electrical error very quickly at these frequencies. Two positioner topologies are common. A combined-axis system mounts the device on a two-axis positioner that provides both theta and phi rotation, keeping the probe fixed. A distributed-axis system rotates the device about one axis and moves the probe, or the reflector feed arm, about the other. Combined-axis systems are mechanically simpler; distributed-axis systems avoid tipping the device, which matters when the device must remain in a defined orientation or when a phantom is attached.

The device holder is an electromagnetic component whether the designer intends it or not. Holders are built from low-permittivity, low-loss foam or thin-walled plastic precisely because anything denser scatters and detunes. Cables running to the device for power or control must be routed along field nulls, ferrite-loaded or otherwise decoupled where possible, and kept identical between calibration and measurement; a control cable that moves between runs is a classic source of unexplained ripple.

Phase-center alignment deserves particular attention. The measurement assumes the device rotates about the phase center of its radiating aperture. If the aperture sits off the axis of rotation, it traces a circle during the sweep, introducing a path-length modulation and a corresponding amplitude and phase error that masquerades as pattern structure. At 39 GHz a five-millimeter offset is nearly two-thirds of a wavelength, which is more than enough to distort a pattern measurement. Careful alignment, and in some cases an analytical offset correction, is part of range setup rather than an optional refinement.

Measurement Uncertainty

Because an over-the-air result cannot be checked against a conducted one, the uncertainty budget carries the entire burden of credibility. 3GPP tabulates the contributors and the assessment format for every FR2 test case in Technical Report 38.903, and an accredited laboratory publishes an expanded uncertainty alongside every reported figure.

The budget divides naturally into three groups. Instrumentation contributors include receiver level accuracy and linearity, signal-generator level accuracy, mismatch at every interface, and the stability of any frequency converter. Range contributors include quiet-zone ripple and taper, absorber reflectivity, probe antenna gain calibration and its own polarization purity, path-loss calibration accuracy, and the gain uncertainty of the reference antenna used in the substitution. Geometry and procedure contributors include positioner angular accuracy and repeatability, phase-center alignment error, the coarseness of the sampling grid, the influence of the device holder and cabling, and, for sensitivity work, the statistical uncertainty of the throughput or error-rate criterion.

Individual terms are classified as type A, evaluated statistically from repeated observations, or type B, evaluated from calibration certificates, manufacturer specifications, or assumed distributions. They are converted to standard uncertainties, combined by root-sum-square for uncorrelated terms, and multiplied by a coverage factor to give the expanded uncertainty. Two is the conventional factor; 3GPP's FR2 budgets use 1.96 standard deviations, for a 95 percent confidence interval.

The resulting figures are far larger than conducted practice would lead an engineer to expect, and they are worth stating plainly, because anyone anticipating tenths of a decibel will misread every FR2 result. For a power class 3 handset on a compact range with a quiet zone of thirty centimeters or less, under normal temperature conditions, the 3GPP reference budget gives a total expanded uncertainty for EIRP of 5.08 dB from 23.45 to 32.125 GHz, 5.28 dB from 32.125 to 40.8 GHz, and 6.64 dB from 40.8 to 44.3 GHz. Total radiated power comes out slightly lower across the same three ranges, at 4.61, 4.81, and 6.16 dB, because integrating over the sphere averages several directional terms away. Reference sensitivity sits alongside EIRP, between about 5.4 and 6.5 dB.

Two features of that budget are instructive. The largest single contributions are mismatch and the accuracy of the radio-frequency power measurement equipment, with quiet-zone quality worth 0.6 dB and the geometric terms, positioning misalignment and measurement distance, evaluated at zero for a range that genuinely meets the far-field criterion; the chamber is a necessary condition rather than the dominant error. And everything degrades sharply above 40.8 GHz, where the power-measurement contribution alone rises from 2.16 dB to 3.6 dB. Conformance limits and internal design margins are set with that spread in mind.

Inter-laboratory comparison is the practical check on all of this. Circulating a stable reference device among laboratories and comparing reported values exposes systematic errors that no internal budget will reveal, and such round-robin exercises have repeatedly driven refinement of both the test plans and the ranges themselves.

From the Laboratory to the Production Line

Conformance ranges and production testers answer different questions and are built accordingly. A conformance range establishes, once, that a design meets a written requirement, and it optimizes for accuracy and traceability at almost any cost in time. A production tester establishes, millions of times, that a particular unit resembles units already known to be good, and it optimizes for cycle time, footprint, and cost.

Production millimeter-wave testers therefore abandon the full spherical sweep. A single probe, or a small cluster of probes, is placed at a short fixed distance, usually well inside the Fraunhofer distance and therefore in the radiating near field, and the measurement is treated as a relative one. Calibration is performed against golden units characterized in the conformance range, and the tester verifies a limited set of quantities: per-element amplitude and phase, beam peak power in a few directions, and gross functional health of the array. The transfer function between the near-field production measurement and the far-field laboratory result is established during characterization and monitored over time.

This structure works because production test is a screening problem rather than a characterization problem. Its dominant failure modes are drift in the tester and divergence between the golden units and current production, both of which are managed by statistical process control rather than by tightening the measurement itself.

Beyond Cellular

Cellular conformance drives most of the published methodology, but the same physics governs other millimeter-wave products, and each adds a wrinkle.

Automotive radar at 76 to 81 GHz must be measured through the bumper fascia, emblem, or radome that will cover it in service, because those layers alter both the pattern and the apparent range. Ranges for radar therefore combine conventional pattern measurement with radar target simulation, in which an instrument receives the sensor's transmission, applies a controlled delay, Doppler shift, and attenuation, and retransmits it so that the sensor perceives a synthetic object at a chosen range and velocity. Verifying detection, resolution, and angular accuracy over the air is the only way to test the sensor as the vehicle will use it.

Unlicensed 60 GHz equipment built to the WiGig amendments faces the same absence of connectors and the same far-field arithmetic, with the added complication of very wide channels whose flatness must hold across the beam. The band is no longer theirs alone: 3GPP has defined 57 to 71 GHz as NR band n263 within FR2-2, with a power class 3 minimum peak EIRP of 14.1 dBm, which brings the full cellular conformance apparatus to bear on the same spectrum. Fixed point-to-point links in the E-band, at 71 to 76 and 81 to 86 GHz, use highly directive antennas whose large apertures push Fraunhofer distances into the tens of meters, which makes compact ranges and near-field transformation not merely convenient but necessary. Satellite user terminals, imaging systems, and sub-terahertz research hardware extend the same techniques upward in frequency, where waveguide bands replace coaxial interfaces throughout the instrumentation chain.

Practices That Separate Good Results from Plausible Ones

Several habits distinguish laboratories whose radiated numbers survive scrutiny.

Fix the aperture definition before building the range. Deciding late that conformance requires the black-box aperture rather than the array aperture can invalidate a chamber. Establish D, and therefore the quiet-zone and range requirement, at the start.

Qualify the quiet zone at every band, and requalify after every change. Ripple maps are frequency specific. A chamber qualified at 28 GHz says little about its behavior at 43 GHz, and a new cable route or an added fixture can undo both.

Calibrate through the measurement path, not around it. The reference measurement must traverse the same switches, converters, and cables in the same positions as the device measurement. Any path the calibration skips becomes an unquantified offset.

Treat everything metallic in the chamber as an antenna. Fasteners, cable shields, positioner surfaces, and instrument faces all scatter at millimeter wavelengths. Cover them, or measure the consequence.

Lock the beam, and record which beam was locked. A pattern measured while the device re-selects beams describes neither the beam nor the device. Beam state belongs in the measurement record alongside frequency and orientation.

Report uncertainty with every number. A radiated result without its expanded uncertainty is not a measurement; it is an observation. The budget is the only evidence the reader has that the result means what it claims.

Design for testability. Provide a test mode that holds a beam, exposes element-level control, and reports internal state. Reserve a mechanical reference feature that lets a fixture register the array to the axis of rotation. These cost almost nothing in the product and save a great deal in the chamber.

Conclusion

Millimeter-wave over-the-air testing exists because integration removed the connector. Once the antenna is part of the package, the radiated field is the only interface, and every familiar element of the measurement discipline must be rebuilt around it. Range length follows from aperture size and wavelength through an unforgiving quadratic, and the architectures that tame it, compact ranges, near-field transformation, and synthesized plane waves, each trade chamber size for a new class of error. Calibration becomes an end-to-end path-loss determination anchored to a reference antenna. Specifications move to EIRP, EIS, total radiated power, and spherical coverage, each dependent on how the sphere is sampled and which beam is active. Uncertainty, counted in whole decibels rather than in tenths of one, becomes the measure of whether a laboratory can be believed.

The engineering lesson mirrors the one that runs through millimeter-wave signal integrity generally. Effects that lower-frequency practice treats as second order, a few millimeters of misalignment, a moved cable, an uncovered bracket, become first-order limits on what can be measured and therefore on what can be designed. Planning the radiated characterization strategy alongside the product, rather than after it, is what makes a millimeter-wave design verifiable at all.

Related Topics