RF Design (Smith Chart, S-Parameters, Impedance Matching)

1. At a glance

Radio-frequency (RF) design is the engineering discipline operating in the regime where the wavelength of the signal approaches the physical dimensions of the circuit — the boundary where lumped-element circuit theory breaks down and distributed-element transmission-line + electromagnetic-wave behaviour takes over. By convention RF starts where you can no longer pretend a wire is “just a wire” — practically from a few MHz on a long cable to several hundred GHz on a chip — and ends where free-space optics and quantum-electrodynamic effects dominate.

In 2026 the engineered RF spectrum spans 100 kHz (LF inductive coupling, contactless cards) to 300 GHz (D-band 6G research, sub-THz imaging), with the workhorse bands sitting between 400 MHz (LoRa / sub-GHz IoT) and 80 GHz (automotive radar). Applications: cellular (LTE, 5G NR sub-6 + mm-wave, early 6G prototypes), Wi-Fi 6E / 7 (6 GHz band), Bluetooth + UWB, automotive radar (24, 77, 79 GHz), GPS / GNSS (1.176–1.602 GHz), satellite (Starlink user-terminal K_u 12–18 GHz, gateway K_a 26–40 GHz, V-band 40–75 GHz for inter-satellite), broadcast (FM, DVB-T/T2, DAB), ISM (433, 868, 915 MHz, 2.4, 5.8, 24 GHz), mm-wave imaging (security, medical), and microwave control lines for superconducting-qubit and trapped-ion quantum computers (typically 4–8 GHz).

This note builds on [[Engineering/electromagnetics-engineering]] (Maxwell, wave equation, Γ, VSWR, η₀, microstrip impedance) and [[Engineering/pcb-design]] (controlled impedance, stack-up, return paths). It documents the three load-bearing tools of RF design — the Smith chart, the S-parameter formalism, and impedance-matching networks — together with the components, worked examples, simulators, and instruments needed to take a design from block diagram to compliant product.

RF design also occupies a unique pedagogical position: it sits at the intersection of physics (Maxwell, transmission lines, antennas), circuit theory (matching, stability, gain), signal-processing (modulation, equalisation, DPD), and regulatory engineering (FCC, ETSI, 3GPP). A practising RF engineer must be comfortable moving between the field domain (HFSS solving for current distribution on a copper patch), the circuit domain (an S-parameter cascade in ADS), the system domain (link-budget spreadsheet in dB), and the regulatory domain (translating spurious-emission masks into bandstop-filter specifications). This breadth — and the fact that physical layout is part of the design, not separable from it — is why RF teams remain specialised even at companies that have otherwise consolidated electrical engineering roles.

2. Why it matters — first principles

Every wireless device is an RF system, and the cost of failure scales with how late the failure is discovered:

Wrong matching → lost range and lost battery. A 2:1 VSWR mismatch reflects 11% of incident power; for a 1 W handset PA, that is 110 mW dissipated in the PA package as heat, not radiated. Range scales with √(P_radiated) in free space, so a poor match easily costs 1–2 dB of link budget — the difference between 2 bars and 4 bars of cellular service.
Wrong filter → adjacent-channel interference and FCC violation. Insufficient out-of-band rejection means a Wi-Fi transmitter splatters into LTE bands; an LTE handset desensitises a GPS receiver three centimetres away on the same PCB; an unfiltered VCO harmonic at 5.6 GHz lands directly on a Wi-Fi receive band. None of these are recoverable in firmware; they require hardware respin.
Wrong layout → parasitic resonance and self-oscillation. A PA with 25 dB of gain at 2.4 GHz will oscillate if there is 0.005 of stray coupling from output to input. A poorly-bypassed bias network introduces a low-frequency parasitic loop that turns the amplifier into a relaxation oscillator at 1 MHz.
Wrong test → certification failure. FCC, ETSI, and CISPR pre-compliance scans are pass/fail at fixed limits; a 1 dB exceedance is the same as 20 dB.
Wrong placement → desense. Two radios on the same PCB — Wi-Fi and LTE, Bluetooth and GPS — that pass standalone receiver-sensitivity tests will desense each other if antennas, LO frequencies, and harmonics overlap. Coexistence analysis is its own subspecialty.

The classic empirical rule: the cost of a debug late in cycle is 100× the cost of catching it early. A simulator catch costs an hour; a board-spin catch costs a week; a tooling-respin catch on a phone in tooling validation costs a quarter and a million dollars in nonrecurring engineering. The “three-board rule” is realistic for a first-time RF design at an unfamiliar frequency band — three PCB spins from concept to compliance is normal, not a sign of failure. The first board reveals what the simulator missed; the second incorporates the layout corrections; the third polishes for production.

3. First principles

From Maxwell to transmission lines

Maxwell’s equations (see [[Engineering/electromagnetics-engineering]] §2) reduce to the telegrapher’s equations for a transmission line with distributed series inductance L’ (H/m), series resistance R’ (Ω/m), shunt capacitance C’ (F/m), and shunt conductance G’ (S/m):

∂V/∂z = −(R’ + jωL’) · I
∂I/∂z = −(G’ + jωC’) · V

Solving for the lossless case (R’ = G’ = 0) gives:

Characteristic impedance Z₀ = √(L’/C’)
Phase velocity v_p = 1/√(L’·C’)
Wave equation ∂²V/∂z² = L’C’ · ∂²V/∂t²

In the lossy case the propagation constant γ = α + jβ describes both attenuation α (Np/m) and phase β (rad/m); the characteristic impedance becomes complex Z₀ = √((R’ + jωL’)/(G’ + jωC’)). At RF where ωL’ ≫ R’ and ωC’ ≫ G’ the approximation Z₀ ≈ √(L’/C’) (real) holds within a fraction of a percent and is universally used; the loss term is then handled by adding the cumulative dB/m attenuation to the link budget.

For non-magnetic dielectrics v_p = c/√ε_r. Velocity factor is v_p/c, typically 0.66 for RG-58 polyethylene coax, 0.70 for FR-4 microstrip (ε_eff ≈ 3.0), 0.85 for foam-dielectric coax, 0.95 for air-line. The input impedance of a line of length ℓ terminated in Z_L is:

Z_in(ℓ) = Z₀ · (Z_L + jZ₀·tan(βℓ)) / (Z₀ + jZ_L·tan(βℓ))

Two cases that fall out and matter constantly: at ℓ = λ/4 the line inverts the impedance (Z_in = Z₀²/Z_L — the basis of the quarter-wave transformer), and at ℓ = λ/2 the line repeats the impedance unchanged (Z_in = Z_L — explaining why connector-launch sections are often half-wavelength multiples to defer impedance transformation to the launch itself).

Engineering values of Z₀: 50 Ω is the global RF standard (a compromise between minimum-loss 77 Ω air-coax and maximum-power 30 Ω air-coax); 75 Ω is the broadcast / CATV standard chosen for minimum attenuation in a fixed-size cable; 600 Ω is the historical telecom audio impedance (open-wire pairs); 300 Ω is the TV ribbon-line value; 377 Ω is the impedance of free space (η₀, the matching target for an antenna).

Reflection coefficient, VSWR, and return loss

A transmission line of impedance Z₀ terminated in load Z_L produces a reflected wave with complex amplitude:

Γ = (Z_L − Z₀) / (Z_L + Z₀)

The reflection coefficient is a complex number; on the Smith chart it lives in the unit disc |Γ| ≤ 1. Open-circuit Γ = +1, short-circuit Γ = −1, matched load Γ = 0. The voltage standing-wave ratio:

VSWR = (1 + |Γ|) / (1 − |Γ|)

VSWR = 1.0 means perfect match (no reflection); VSWR = ∞ means total reflection (open or short). Practical specifications: VSWR < 2.0 is “acceptable” for a consumer antenna, < 1.5 is good, < 1.2 is excellent.

Return loss in dB:

RL = −20·log₁₀|Γ|

RL > 10 dB (Γ < 0.32, VSWR < 1.93) is the loose acceptable; > 14 dB (Γ < 0.20, VSWR < 1.5) is the common datasheet spec; > 20 dB (Γ < 0.1, VSWR < 1.22) is excellent.

Mismatch loss (power lost relative to a matched load):

ML = −10·log₁₀(1 − |Γ|²) dB

For VSWR = 2 (Γ = 0.33), ML = 0.51 dB; for VSWR = 1.5 (Γ = 0.20), ML = 0.18 dB; for VSWR = 3 (Γ = 0.50), ML = 1.25 dB. Mismatch uncertainty in a measurement chain adds an additional ± term: when two components with |Γ₁| and |Γ₂| are cascaded, the measured insertion-loss uncertainty is ± 20·log₁₀(1 ± |Γ₁|·|Γ₂|) dB, which is why precision RF measurement chains insist on VSWR < 1.05 isolators and matched terminations between elements.

Power and voltage on a line

Forward and reverse travelling waves V⁺(z) and V⁻(z) carry power P⁺ = |V⁺|²/(2·Z₀) and P⁻ = |V⁻|²/(2·Z₀); net delivered power is P_net = P⁺(1 − |Γ|²). On a line with standing waves, the voltage envelope ranges from V_max = |V⁺|·(1 + |Γ|) to V_min = |V⁺|·(1 − |Γ|), separated by λ/4 along the line. Peak-voltage stress is the design driver for power amplifier matching networks — a 100 W PA at 50 Ω drives V_peak = √(2·50·100) = 100 V into a matched load, but on a VSWR = 3 mismatch (open or short), the standing-wave peak rises to 200 V — and capacitor / connector voltage ratings must cover the mismatch case.

S-parameters (scattering parameters)

The S-matrix, introduced by Kurokawa (1965, IEEE Trans. MTT), describes a linear N-port network in terms of normalised incident and reflected travelling waves a_n and b_n at each port, all referenced to a common impedance Z₀ (almost always 50 Ω):

b = S · a

For a two-port:

b₁ = S₁₁·a₁ + S₁₂·a₂
b₂ = S₂₁·a₁ + S₂₂·a₂

Reading the entries:

S₁₁ = b₁/a₁ with a₂ = 0 → input reflection coefficient with output matched. Same as Γ_in.
S₂₂ = b₂/a₂ with a₁ = 0 → output reflection coefficient with input matched.
S₂₁ = b₂/a₁ with a₂ = 0 → forward transmission coefficient (gain or loss).
S₁₂ = b₁/a₂ with a₁ = 0 → reverse transmission (isolation, leakage).

Each entry is a complex number at a single frequency; a real S-parameter file (Touchstone .s2p) is a table of these complex values across a frequency sweep. Why S-parameters instead of Y or Z parameters? Because at GHz frequencies it is impossible to enforce a true open or short circuit for Z and Y measurement; whereas a 50 Ω matched termination is trivial to build. S-parameters are the only RF network description that is physically measurable above ~1 GHz.

Useful derived quantities:

|S₂₁|² in dB = gain (or insertion loss if negative)
|S₁₁|² in dB = return loss at port 1
K-factor (Rollett, 1962) = (1 − |S₁₁|² − |S₂₂|² + |Δ|²) / (2·|S₁₂·S₂₁|), with Δ = S₁₁·S₂₂ − S₁₂·S₂₁. Unconditional stability requires K > 1 and |Δ| < 1 (equivalently B₁ = 1 + |S₁₁|² − |S₂₂|² − |Δ|² > 0).
MAG (Maximum Available Gain) = |S₂₁/S₁₂| · (K − √(K² − 1)), valid only when K > 1.
MSG (Maximum Stable Gain) = |S₂₁/S₁₂|, the gain achievable when neutralised to the K = 1 stability boundary.
Source / load stability circles — on the Smith chart, the loci of Γ_S and Γ_L that produce |Γ_in| = 1 or |Γ_out| = 1; these bound the unstable region.

S-parameters are inherently small-signal and frequency-domain. Large-signal behaviour (compression, IP3, harmonic distortion) requires extended formats — load-pull data, X-parameters (Verspecht et al., Keysight), Cardiff behavioural model, or measured AM-AM / AM-PM characteristics — and these dominate PA design where the device is run within 1–3 dB of compression for efficiency.

Touchstone file format

The de facto interchange format is Touchstone (.sNp where N = number of ports), originated by EEsof in the 1980s and now governed by IBIS-ATM. Plain text, one frequency per line, with options for magnitude/phase or real/imaginary, dB/degrees, etc. A header line such as # GHZ S MA R 50 declares frequency unit, parameter type, format, and reference impedance. Vendor S-parameter files (Qorvo, Skyworks, Mini-Circuits, Analog Devices) are downloaded and dropped directly into ADS, AWR, or scikit-rf.

Loss mechanisms in RF conductors and dielectrics

Three loss mechanisms dominate above 100 MHz:

Conductor loss — current crowds into a surface layer of depth δ = √(2/(ωµσ)). For Cu at 1 GHz δ ≈ 2 µm; surface roughness comparable to δ increases AC resistance by 2× or more. Smooth electrodeposited or rolled copper matters above 5 GHz; specifications for VLP (very-low-profile) and HVLP foils exist for this reason.
Dielectric loss — proportional to ε_r · tan δ · f. FR-4 tan δ ≈ 0.02 at 1 GHz, rising to 0.03 at 10 GHz. Rogers RO4350B tan δ ≈ 0.0037, Megtron 6 ≈ 0.004, PTFE ≈ 0.0009. Loss in dB/m scales as f·√ε_r·tan δ / 27.3 for microstrip.
Radiation loss — uncontrolled if traces approach λ/2 between vias; mitigated by stitching vias along controlled-impedance lines at < λ/10 spacing.

Skin depth in copper

Frequency	δ (Cu, 20 °C)	Engineering implication
1 MHz	66 µm	RF conductors plated, not bulk
100 MHz	6.6 µm	trace surface finish matters
1 GHz	2.1 µm	rolled Cu preferred for low loss
10 GHz	0.66 µm	HVLP / VLP foils mandatory
30 GHz	0.4 µm	mm-wave laminates only
77 GHz	0.24 µm	automotive radar substrate spec

4. The Smith chart

The Smith chart, invented by Phillip H. Smith at Bell Labs (1939, Electronics magazine), is a conformal map of the complex impedance plane onto the unit disc of the complex reflection-coefficient plane. It is the single most-used graphical tool in RF design and remains taught today because no purely numeric tool gives the same instantaneous visual feedback for impedance behaviour, matching topology, and stability.

Construction

Normalise impedance to Z₀: z = Z/Z₀ = r + jx. Then Γ = (z − 1)/(z + 1) maps the right-half z-plane (r ≥ 0, passive load) onto the unit disc |Γ| ≤ 1. Two families of curves carry over:

Constant-resistance circles — circles tangent to Γ = +1, indexed by r ∈ [0, ∞]. The r = 1 circle passes through the chart centre (Γ = 0).
Constant-reactance arcs — arcs tangent to Γ = +1, indexed by x ∈ (−∞, +∞). x > 0 (inductive) is the upper half-disc; x < 0 (capacitive) is the lower half.

The admittance chart is the Smith chart rotated 180°. YZ Smith chart overlays both, so admittance + impedance manipulations on the same diagram switch between series (impedance) and shunt (admittance) elements.

Operations on the chart

Transmission-line propagation rotates Γ clockwise (toward generator) by 2β·ℓ radians, where β = 2π/λ. One full revolution = λ/2 of line.
Series L moves clockwise along a constant-r circle (x increases).
Series C moves counterclockwise along a constant-r circle (x decreases).
Shunt L moves counterclockwise along a constant-g circle on the admittance overlay (b decreases).
Shunt C moves clockwise along a constant-g circle (b increases).
Loss moves Γ inward along a radial; gain moves outward (e.g. negative-resistance device).

Matching procedures graphically

L-match (2 components): from Z_L, move along a constant-r circle with one element to reach the g = 1/Z₀ admittance circle, then jump to the chart centre with a shunt element. Always possible if r < 1 or g < 1.
π-match (3 components): C-shunt → L-series → C-shunt; high Q, harmonic filtering.
T-match (3 components): L-series → C-shunt → L-series; suited to low-impedance loads.
Stub matching: a single open-stub or short-stub of length d_s placed at distance d from the load can match any complex Z_L → Z₀. Single-stub is narrowband; double-stub allows partial broadband response.

Constant-Q contours, noise circles, gain circles

Three families of overlays make the Smith chart load-bearing for amplifier design:

Constant-Q contours — arcs of constant |x/r|; following one keeps the network Q (and hence the −3 dB bandwidth) constant. Used to bound the Q of a matching network: stay inside the Q = 3 contour and you guarantee BW > f₀/3.
Constant-noise-figure circles — loci of Γ_S that produce a given NF for an active device, centred on the noise-optimal source impedance Γ_opt (typically given on the LNA datasheet at one frequency). The minimum-NF point is rarely the minimum-S₁₁ point — a real LNA design picks a Γ_S that compromises between NF and input return loss.
Constant-gain circles — loci of Γ_S (or Γ_L) producing a given transducer gain G_T; useful when stability constraints prevent maximum-gain matching.

The classic LNA design exercise is: place Γ_S on a noise circle 0.2 dB above NF_min, on a gain circle 1 dB below MAG, outside the source stability circle, with the resulting Γ_in close enough to 50 Ω for acceptable input return loss. Four-objective optimisation on a 2-D chart; this is why the Smith chart endures.

Computer-aided Smith charts

The chart is still drawn and manipulated in modern tools:

Keysight ADS Smith Chart Utility — drag-and-drop matching design with real component models, frequency sweep, and Q-circle overlay.
RFSim99 / QucsStudio — free schematic + Smith chart.
Online JavaScript Smith charts — Will Kelsey, smithchart.org.
Python scikit-rf — programmable Smith plots and network manipulation; the modern reproducible workflow. A typical idiom: net = rf.Network('lna.s2p'); net.plot_s_smith(m=1, n=1).
MATLAB RF Toolbox — smithplot with full noise-circle / stability-circle overlays.

5. Impedance matching techniques

Topology	Components	Q (bandwidth)	Best for	Notes
L-network	2 (L + C)	Low Q (wide BW)	One-frequency point match	Always possible; Q = √(R_high/R_low − 1)
π-network	3 (C-L-C)	High Q (narrow BW)	Harmonic filtering output of PA	Tunable; standard PA output matching
T-network	3 (L-C-L)	High Q	Low-Z loads	Used at PA bias networks
Quarter-wave transformer	λ/4 transmission line	Narrow BW (~10%)	Real-to-real impedance match	Z_match = √(Z_S · Z_L)
Single-stub	Stub + line length	Narrow BW	Microstrip integration	Open stub if x < 0 capacitive needed
Double-stub	Two stubs spaced λ/8	Moderate BW	Microstrip, no line-length tuning	Forbidden region for some Z_L
Tapered line (Klopfenstein)	Continuously-varying Z₀	Octave or more	Broadband transformer	Optimal ripple/length trade
Multi-section transformer	N λ/4 sections	Broadband	Antenna feeds	Chebyshev (equiripple) or binomial (maximally flat)
Balun	Transformer or transmission-line	Octave-decade	Single-ended ↔ differential	Coupled-line, Marchand, ferrite

L-network design equations

For matching source resistance R_S to load resistance R_L (assume R_S > R_L; swap orientation otherwise):

Q = √(R_S/R_L − 1)
Series reactance X_s = Q · R_L (sign chosen for the element type)
Shunt reactance X_p = R_S / Q (sign chosen on the high-Z side)

Bandwidth ≈ f₀ / Q. For Q = 3, BW ≈ 33% of f₀.

Quarter-wave transformer

For matching two real impedances Z_S and Z_L over a transmission line:

Z_match = √(Z_S · Z_L), with line length ℓ = λ/4

Bandwidth is set by the worst tolerable VSWR; for VSWR_max = 1.5, fractional BW ≈ 30%. Cascading multiple λ/4 sections at intermediate impedances widens the band: Chebyshev sections give equiripple |Γ| across the band; binomial (maximally flat) sections give monotonic |Γ| with no ripple.

Single-stub matching

The single-stub procedure finds a stub length ℓ_s and a position d (from the load) such that the line impedance + stub reactance matches Z₀:

Plot Z_L (or Y_L) on the Smith chart, normalised to Z₀.
Rotate along the line toward generator until Γ lands on the unit-conductance circle (g = 1).
The required stub admittance is the negative of the residual susceptance b at that point.
Stub length ℓ_s is the line length needed to produce −b looking into a short (or open) at the stub end.

A short-circuit stub of any length 0 < ℓ_s < λ/4 generates inductive susceptance; open-circuit stubs generate capacitive susceptance. On a PCB, short stubs are awkward (a via to ground at a precise location); open stubs are trivial. Stub matching is ubiquitous in microstrip filter and matching networks because no lumped-element values are needed — the design is fully realisable in copper geometry.

Lumped-element bandwidth limits — Bode-Fano

For any lossless matching network terminating a load with a specific complex Q (such as an antenna), the fundamental bandwidth limit is set by Bode-Fano (Fano, 1950):

∫₀^∞ ln(1/|Γ(ω)|) dω ≤ π / (R·C) (for a parallel RC load)

The integral on the left is a fixed budget — trade depth of match (low |Γ|) against bandwidth, but cannot reduce both arbitrarily. A high-Q antenna (small antenna at low frequency, like a 50 mm whip at 433 MHz) has a hard physical limit on achievable match bandwidth; no amount of network cleverness improves it. This is why aperture-tuner ICs exist — they re-tune the matching network as the user holds the phone differently, rather than try to broadband-match a fundamentally narrowband resonator.

6. RF components

Passive components

Couplers and combiners. Directional couplers tap off a small known fraction of forward (or reverse) power for monitoring; tight specifications on directivity (20–40 dB) and coupling factor (typically 10–30 dB). Common topologies:

Wilkinson combiner — equal-split, 3-port, isolated; standard PA combining. 100 Ω resistor between output legs gives 20+ dB isolation.
Branch-line (quadrature) hybrid — 4-port, 90° phase split; used in balanced amplifiers, image-reject mixers.
Rat-race (ring) hybrid — 4-port, 0°/180° split; balun and combiner uses.
Lange coupler — interdigitated 4-bar coupled-line quadrature hybrid; broadband (octave+) at the cost of fabrication complexity.

RF filter design proceeds from a prototype low-pass response (Butterworth maximally flat, Chebyshev equiripple, Cauer / elliptic with finite-frequency zeros, or Bessel maximally-flat-group-delay), with order N set by the required out-of-band rejection. The prototype is frequency-transformed to band-pass via the impedance/frequency scaling rules. For narrow-band PCB filters above 1 GHz the lumped-element prototype is replaced with distributed coupled resonators: half-wave open-circuit microstrip resonators coupled by gap or λ/4 inverters (Cohn 1957, Matthaei/Young/Jones 1964). Modern flow: synthesise in AWR iFilter or ADS Filter DesignGuide, EM-tune in Sonnet or HFSS, build and measure on a VNA, iterate.

Filters — the technology depends on frequency and bandwidth:

Type	Range	Q	Use
LC ladder (discrete)	< 500 MHz	50–200	IF, baseband, sub-GHz radios
Microstrip (interdigital, hairpin, edge-coupled)	0.5–20 GHz	100–500	Bandpass between LNA and mixer
SAW (surface acoustic wave)	30–3000 MHz	1000+	Cellular RX/TX preselect, IF
BAW / FBAR	1.5–6 GHz	1500–3000	5G n78/n79, high-band Wi-Fi
Ceramic monoblock (Murata, TDK)	0.4–6 GHz	200–500	Compact, cheap, Bluetooth/Zigbee
Cavity	0.1–40 GHz	5000+	Base-station preselect
Waveguide iris	8–110 GHz	10000+	mm-wave, satellite uplink
YIG (yttrium-iron-garnet)	0.5–40 GHz	5000, electronically tunable	EW, instrumentation

Isolators and circulators use Faraday rotation in a ferrite biased by a permanent magnet; pass forward, attenuate reverse 20–30 dB. Typical from 0.4 to 40 GHz; circulators come in three-port “Y-junction” form, isolators are two-port circulators with the third port terminated.

Phase shifters — switched-line (digital, fast, narrow-band), varactor-tuned (analogue, fast, modest range), MEMS (low-loss, slow), liquid-crystal (low-cost mm-wave research). 5G phased-array beam-steering uses RFIC-integrated 6-bit phase shifters.

Attenuators — fixed π or T resistive networks (DC to 18 GHz, ±0.5 dB flat); variable using PIN-diode (continuous, fast) or MMIC digital step attenuators (6-bit, 0.5 dB step, 30 dB range; ADI HMC, Peregrine, Qorvo).

Active components

Low-Noise Amplifier (LNA). The first active stage on the receiver chain; dominates the system noise figure per Friis. Process choices:

GaAs pHEMT — Avago/Broadcom MGA, Qorvo TQP, Mini-Circuits PMA series. NF 0.4–1.0 dB up to 18 GHz; gain 15–25 dB; cost commodity in volume.
SiGe BiCMOS — Skyworks SKY, Infineon BGA, Analog Devices HMC. NF 0.8–1.5 dB up to 24 GHz; cheaper than GaAs in handset volumes; integrated DC blocks and matching.
GaN HEMT — Wolfspeed CGH, Qorvo TGA; high power-handling LNAs (for radar receivers that must survive a near hit), NF 1.5–2.5 dB.
CMOS — used in deeply-integrated transceivers (5G handset RFICs); NF 1.5–3 dB but the integration wins.

Power Amplifier (PA). The final transmit stage. Process choices follow output power:

CMOS / SOI — handset 2 W class-AB / class-F PAs (Qorvo, Skyworks, Broadcom); integrated with switch and LNA in a front-end module.
GaAs HBT — handset PA workhorse, 1–5 W, 0.7–6 GHz; Skyworks SKY77, Qorvo QPA.
GaN-on-SiC HEMT — base station and radar PAs, 10–500 W, 0.4–18 GHz; Wolfspeed, Qorvo Spatium, NXP. 5G massive-MIMO arrays put 40+ GaN PAs per antenna panel.
LDMOS — base-station legacy below 4 GHz, 50–1000 W, cheap but capped at sub-6 GHz; Ampleon, NXP.
Class definitions: A (linear, 25% η_max), AB (handset linear, ~40%), C (single-tone high-η, 70%, harmonics not OK), D/E/F (switching, 80–95%, harmonics shaped).

Mixers. Convert RF to/from IF by multiplication. Topologies:

Passive diode-ring (DBM) — Schottky-quad balanced mixer; LO drive +7 to +17 dBm; IP₃ +15 dBm typical; insertion loss 6–8 dB; Mini-Circuits ZX05, Marki ML1.
Active Gilbert cell — integrated, low LO drive, gain instead of loss; standard in transceiver ICs.
Image-reject (Hartley, Weaver) and single-sideband mixers — quadrature LO + I/Q combining to suppress unwanted image, 25–35 dB image rejection; essential in zero-IF / low-IF architectures.

VCO / PLL. A voltage-controlled oscillator (LC tank or ring) phase-locked to a crystal reference. Integer-N PLLs are simple and low-noise but coarse in frequency step; fractional-N PLLs (ΔΣ modulator on the divider) give fine step at the cost of close-in phase noise. Modern parts: Analog Devices ADF4xxx / HMC833 (wide-band, integrated VCO), Texas Instruments LMX2594 / LMX2820, Renesas (IDT) RC5C, Crystek free-running VCOs for low-noise references.

RF switches. PIN diodes (high power, ms switching), GaAs pHEMT FETs (ns switching, 0–18 GHz, low IL), CMOS SOI (highly integrated, used in handset antenna-aperture-tuner ICs). Insertion loss 0.3–1.5 dB, isolation 25–40 dB.

Detectors. Schottky-diode square-law detectors (low-cost RSSI), log detectors (Analog Devices AD8307, AD8318 — 80 dB log range), RMS detectors (true power for OFDM-modulated signals, AD8364).

Limiters. Antenna-input limiters (PIN-diode or back-to-back Schottky) prevent strong out-of-band signals from compressing or destroying the LNA. Recovery time 10 ns – 1 µs; insertion loss 0.3–1 dB; survival up to +30 dBm CW or 30 W peak pulse.

Modulators / demodulators. I/Q modulators (Analog Devices ADL537x, Texas Instruments TRF372xx) take baseband I/Q signals and generate a single-sideband modulated RF carrier; pair with an LO. Direct-conversion architecture in modern transceiver ICs (Analog Devices AD9361 / AD9371, Texas Instruments AFE7920, Xilinx RFSoC integrated DACs/ADCs at 4–12 GS/s) is replacing super-heterodyne for software-defined-radio applications.

Frequency multipliers and dividers. Diode-based ×2 / ×4 multipliers extend a low-frequency source up to mm-wave; commercial parts from Marki, Eravant, Virginia Diodes. ÷N prescalers (Hittite HMC366 to 13 GHz) feed PLLs at high frequency.

Antennas are covered in the planned companion note [[Engineering/antenna-theory]].

7. Worked examples

Example A — L-network match: 5 Ω to 50 Ω at 100 MHz

Match a low-impedance load (typical of a PA collector to a 50 Ω line) to 50 Ω at f₀ = 100 MHz.

Q = √(R_high/R_low − 1) = √(50/5 − 1) = √9 = 3.0
Series reactance on the low-Z side: X_s = Q · R_L = 3 · 5 = 15 Ω. Choosing an inductor (positive reactance): L = X_s/(2π·f) = 15 / (2π · 10⁸) = 23.9 nH.
Shunt reactance on the high-Z side: X_p = R_S / Q = 50 / 3 = 16.7 Ω. Choosing a capacitor (negative reactance to ground): C = 1/(2π·f·X_p) = 1/(2π · 10⁸ · 16.7) = 95.4 pF.
Bandwidth ≈ f₀/Q = 100 / 3 = 33 MHz. Acceptable for narrow-band ISM 100 MHz radios; insufficient for octave-bandwidth applications (where a Chebyshev multi-section matcher is needed).
Component selection: 24 nH air-core or wirewound inductor (Coilcraft 0805CS-240XJL) and 100 pF NP0 MLCC (Murata GRM21A5C2D101J or Johanson Technology high-Q variants). Avoid X7R / Y5V at RF — temperature-coefficient and voltage-coefficient nonlinearities create IMD.

Example B — Microstrip 50 Ω trace on FR-4 at 1 GHz

Compute trace width and electrical length for a 50 Ω line on FR-4 (ε_r = 4.3, tan δ = 0.02) with dielectric height h = 1.6 mm (standard two-layer board).

Wheeler / Hammerstad approximation, iterating until Z₀ = 50 Ω:

Try W = 2.95 mm → W/h = 1.84 → ε_eff = (4.3 + 1)/2 + ((4.3 − 1)/2) · (1 + 12 · 1.6/2.95)^(−1/2) = 2.65 + 1.65 · 0.358 = 3.24
Denominator: 1.84 + 1.393 + 0.667 · ln(1.84 + 1.444) = 3.23 + 0.667 · 1.190 = 4.02
Z₀ = (120π / √3.24) / 4.02 = 209.4 / 4.02 = 52.1 Ω ≈ 50 Ω (within 5%; a numerical solver like Polar Si9000 or KiCad’s trace-calculator would tighten this to 49.5 Ω at W = 3.05 mm).

Phase velocity v_p = c / √ε_eff = 3·10⁸ / √3.24 = 1.67 × 10⁸ m/s (velocity factor 0.56).

Frequency	λ on this microstrip	λ/4 length	λ/10 transmission-line threshold
100 MHz	1.67 m	418 mm	167 mm
1 GHz	167 mm	41.8 mm	16.7 mm
2.4 GHz	70 mm	17.4 mm	7.0 mm
5 GHz	33 mm	8.4 mm	3.3 mm
24 GHz	7 mm	1.7 mm	0.7 mm

A standard 100 mm trace is a transmission line above ~150 MHz on FR-4. Loss in FR-4 microstrip at 1 GHz: ≈ 0.15 dB/cm conductor + 0.10 dB/cm dielectric = 0.25 dB/cm; at 5 GHz ≈ 0.7 dB/cm — which is why anything above 6 GHz migrates to Rogers RO4350B (0.08 dB/cm at 5 GHz).

Example C — LNA noise-figure cascade (Friis)

A receiver front end has three cascaded stages:

Stage	Gain (dB)	NF (dB)	F (linear)	G (linear)
LNA	+18	1.0	1.26	63.1
Mixer	−8 (loss)	8.0	6.31	0.158
IF amp	+20	3.0	2.00	100

The Friis formula (Friis, 1944, Proc. IRE):

F_total = F₁ + (F₂ − 1)/G₁ + (F₃ − 1)/(G₁·G₂) + …

F₁ = 1.26
(F₂ − 1)/G₁ = (6.31 − 1)/63.1 = 5.31/63.1 = 0.084
(F₃ − 1)/(G₁·G₂) = (2.00 − 1)/(63.1 · 0.158) = 1.00/9.97 = 0.100
F_total = 1.26 + 0.084 + 0.100 = 1.444
NF_total = 10·log₁₀(1.444) = 1.60 dB

The first stage dominates the noise figure if its gain is high (G₁ = 63× = 18 dB). That is why the LNA always goes first on the receive chain, immediately after the antenna and the band-select filter. Inserting a lossy filter (say, 3 dB) before the LNA would raise NF_total by exactly 3 dB; the same filter after the LNA contributes only (10^0.3 − 1)/63 = 0.016 to F, a negligible 0.05 dB hit.

System sensitivity at the receiver:

S_min = kT_B + NF + SNR_required

At T = 290 K, kT₀·B over 20 MHz Wi-Fi channel = −101 dBm; with NF = 1.6 dB and required SNR = 20 dB for 64-QAM: S_min = −101 + 1.6 + 20 = −79 dBm. A real Wi-Fi 6 STA is specified at −82 dBm sensitivity for MCS9 — consistent allowing for implementation margin.

Example D — Stability check on a GaAs pHEMT at 2 GHz

Given a measured S-parameter set at 2.0 GHz, Z₀ = 50 Ω:

S₁₁ = 0.65 ∠ −110°
S₁₂ = 0.05 ∠ +60°
S₂₁ = 4.0 ∠ +120° (|S₂₁|² = 16 → 12.0 dB raw gain)
S₂₂ = 0.45 ∠ −80°

Compute:

Δ = S₁₁·S₂₂ − S₁₂·S₂₁ = (0.65 ∠ −110°)(0.45 ∠ −80°) − (0.05 ∠ +60°)(4.0 ∠ +120°) = 0.293 ∠ −190° − 0.20 ∠ +180° = 0.293 ∠ +170° − 0.20 ∠ +180° ≈ 0.093 ∠ +148° → |Δ| = 0.093 ✓ (< 1)
|S₁₁|² = 0.423, |S₂₂|² = 0.203, |Δ|² = 0.0086
K = (1 − 0.423 − 0.203 + 0.0086) / (2 · 0.05 · 4.0) = 0.383 / 0.40 = 0.957 → K < 1 → conditionally stable.

The device can oscillate at 2 GHz for certain Γ_S, Γ_L combinations. Two design responses are common: (1) resistive stabilisation — add a small series resistor (5–10 Ω) at the input or a shunt RC across the output to push K above 1, accepting ~0.5 dB NF or gain hit; (2) plot stability circles on the Smith chart, choose source and load impedances outside both unstable regions, and live with the constraint. At the device’s intended bias and 2 GHz, MSG = |S₂₁/S₁₂| = 4.0/0.05 = 80 = 19.0 dB — but only achievable on the K = 1 boundary, where the device is on the edge of oscillation; practical designs back off to 14–16 dB transducer gain.

Example E — 2.4 GHz Wi-Fi PA π-match output network

Design the output matching network of a 100 mW (20 dBm) Wi-Fi PA on a GaAs HBT, with output load-pull data indicating the optimum load impedance at 2.4 GHz is Z_opt = 6 + j8 Ω (typical of a 0.5 W-class HBT operating at 3.3 V supply). Match to a 50 Ω antenna feed using a π-network and provide ≥ 20 dB suppression of the 2nd harmonic at 4.8 GHz.

Working in admittance, Y_opt = 1/Z_opt = 1/(6 + j8) = (6 − j8)/100 = 0.060 − j0.080 S. The Q of the load itself is |X/R| = 8/6 = 1.33.

Choose Q_net = 5 for the π-network (sets bandwidth ~ 480 MHz, more than the 100 MHz Wi-Fi channel plus margin). Working through the standard π-design (Bowick, RF Circuit Design, 2008):

Virtual intermediate resistance R_v = R_opt / (Q² + 1) = 6 / 26 = 0.231 Ω
Series-L between the two shunt caps: X_L = Q · R_v + |X_opt| · (Q_net/Q_L) intermediate algebra; numerical solver yields L_s = 1.5 nH.
Shunt-C at the device side: includes absorbing the +j8 Ω device reactance. C_dev = 1/(2π · 2.4e9 · X_total_dev). Numerically C_dev = 12 pF.
Shunt-C at the load side: C_load = 1.9 pF.

At the 2nd harmonic 4.8 GHz, the network presents a high impedance (shunt C ≈ 1/(2π · 4.8e9 · 1.9e-12) = 17.4 Ω going to ground in parallel with the line — high enough to reflect harmonic energy back into the device for class-F shaping). With a 2.4 GHz output bandpass filter (or harmonic-trap LC notch at 4.8 GHz on the PCB) the combined harmonic suppression hits the FCC Part 15.247 requirement of −20 dBc.

Real-world component selection: 1.5 nH high-Q wirewound or air-core inductor (Coilcraft 0908SQ, Q ≈ 90 at 2.4 GHz); 12 pF and 1.9 pF NP0 0402 capacitors (Murata GJM, AVX/Kyocera Accu-P). Layout: place the device-side capacitor within 1 mm of the PA collector pin, ground both shunt caps via two vias each to minimise ESL.

8. Design heuristics

Frequency-band classification

Band	Range	Primary uses
LF	30–300 kHz	LORAN-C (decommissioned), inductive contactless cards (125 kHz NFC)
MF	300 kHz – 3 MHz	AM broadcast (530–1700 kHz), maritime
HF	3–30 MHz	Shortwave broadcast, amateur, OTH radar, CB
VHF	30–300 MHz	FM broadcast, public-safety land mobile, DVB-T
UHF	300 MHz – 3 GHz	Cellular (LTE, 5G sub-6), Wi-Fi 2.4, TV (470–698 MHz), GPS L1 1.575, ISM 433/868/915
SHF	3–30 GHz	Wi-Fi 5/6/6E, 5G n77/n78/n79, Wi-Fi 7 (5.925–7.125 GHz), satellite K_u 12–18, K_a 26–40
EHF	30–300 GHz	mm-wave 5G n257/n258/n260/n261, automotive radar 77/79, V-band 60 GHz Wi-Gig, satellite V/W
THz	> 300 GHz	6G research (D-band 110–170, sub-THz), spectroscopy

The mm-wave bands (K_a, V, W) are aliases predating modern naming and persist in radar/satellite contexts: K_u 12–18, K 18–26.5, K_a 26.5–40, V 40–75, W 75–110, D 110–170 GHz.

System-level heuristics

System-level VSWR budget < 1.5 at every interface; tighter (1.2) between high-gain stages where standing-wave gain peaking would create ripple.
Stability check on every active stage: K-factor > 1 and |Δ| < 1 across the entire band where the device has gain (not just the design band — out-of-band oscillations are real). If conditionally stable, plot input and output stability circles on the Smith chart and choose source/load impedances outside the unstable region.
PCB stack-up matters: above 3 GHz, FR-4 is uneconomical; choose Rogers RO4350B / RO3003, Megtron 6/7, Isola I-Tera MT40, Panasonic Megtron, or PTFE composites depending on frequency, cost, and processability. Mixed stack-up (Rogers core for RF, FR-4 elsewhere) is common in cellular handsets.
Connectors are part of the design, not afterthought:

Connector	Frequency limit	Use
BNC	4 GHz	lab, oscilloscope, lower-frequency RF
TNC	11 GHz	threaded BNC, vibration-resistant
Type N	11 GHz (precision 18 GHz)	base-station outdoor, antenna feed
SMA	18 GHz (precision 26.5)	almost all GHz-range modules
3.5 mm	26.5 GHz	precision SMA-compatible
2.92 mm (K)	40 GHz	mm-wave automotive radar
2.4 mm	50 GHz	mm-wave 5G test
1.85 mm (V)	67 GHz	E-band backhaul
1.0 mm (W)	110 GHz	D-band, 6G research
MMCX / U.FL	6 GHz	board-to-cable, internal
WR-90 / WR-28 etc.	waveguide	high-power mm-wave, sat-com

IEC 61169 standardises mechanical and electrical interfaces; mixing precision and commercial grades — e.g. mating an SMA into a 3.5 mm — is mechanically possible but degrades performance and damages the precision part.

Shielding every gain block above 20 dB; faraday-cage cans on the PCB. For 30+ dB stage gain, sometimes two stages in separate cans with through-board feedthroughs.
Filter before amplification in the receive chain so the LNA is not driven into compression by an out-of-band blocker; filter after amplification in the transmit chain to scrub PA harmonics.
AGC (automatic gain control) on the receive chain handles 60–80 dB of dynamic range — wireless receivers always face this — usually split as 30 dB at the LNA stage (variable-gain LNA or attenuator) and the rest in the IF / baseband.
Decoupling networks at RF are cascaded: a small ceramic (1 nF NP0) right at the IC pin handles GHz decoupling, a 100 nF X7R handles MHz, a 10 µF tantalum / polymer handles kHz, all in parallel. Each cap has its own self-resonant frequency above which it looks inductive — staging values gives broadband low impedance.
Layout rules at RF:
- Ground stitching via spacing < λ/10 along controlled-impedance traces and around RF blocks.
- Minimise via stubs on signals above 10 GHz; back-drill where stubs exceed λ/20.
- Impedance-controlled traces with stack-up matched to target Z₀ ± 10% (5% on high-speed differential).
- Cut return-current ground splits under high-speed lines.
- Place sensitive RF blocks on a low-loss material island; isolate from digital with copper-pour barriers + stitching.

Anti-pattern: routing large copper pours parallel to controlled-impedance RF traces creates strong unwanted coupling and turns the pour into a slot antenna at the trace harmonics. Pours should be ground, stitched, and either continuous or kept > 3·W away from RF traces.

Linearity figures of merit

P1dB (1 dB compression point) — input or output power at which gain has dropped 1 dB from the small-signal value. Typical LNA OP1dB +10 to +20 dBm; PA OP1dB at the rated power minus 1 dB.
OIP3 / IIP3 (third-order intercept) — extrapolated power where the third-order intermodulation product equals the fundamental. IP3 ≈ P1dB + 10 dB for class-A amplifiers; this rule of thumb fails for class-AB and switching PAs.
ACLR / ACPR (Adjacent Channel Leakage / Power Ratio) — for digitally-modulated carriers, ratio of in-channel power to the adjacent channel; the regulatory limit is typically −30 to −45 dBc depending on the standard.
EVM (Error Vector Magnitude) — vector distance between actual and ideal constellation point, in % rms; 1024-QAM Wi-Fi 7 demands EVM < 1.5% (−36 dB).
Noise / linearity trade-off: dynamic range DR = OIP3 − NF − 10·log₁₀(BW) − kT₀. A receiver with NF = 3 dB, OIP3 = +15 dBm, and 1 MHz BW has SFDR (spurious-free dynamic range) ≈ 86 dB — sufficient for most narrowband ISM, marginal for cellular co-channel reception.

Phase noise

VCOs and oscillators are characterised by phase noise £(f) in dBc/Hz at an offset f from the carrier. Typical:

TCXO 10 MHz reference: −150 dBc/Hz at 10 kHz offset
Crystek CCHD-957 100 MHz OCXO: −175 dBc/Hz at 10 kHz
Integer-N PLL with VCO at 2.4 GHz: −110 to −120 dBc/Hz at 10 kHz
Free-running VCO at 5.8 GHz: −95 dBc/Hz at 100 kHz

Reciprocal mixing — a strong adjacent-channel signal mixed with the LO phase-noise skirt produces in-band interference, even if the channel filter is ideal. The dominant noise limit in dense urban cellular reception.

Power-amplifier class summary

Class	Conduction angle	Linearity	Theoretical η	Use
A	360°	Excellent	50%	Driver / instrumentation; small handhelds
AB	180 to 360°	Good	50–78%	Handset PA (linear OFDM)
B	180°	Marginal (push-pull)	78%	Audio + RF push-pull, rarely solo
C	< 180°	Non-linear	80–90%	CW carrier, FM, classic AM TX
D	50% duty switching	Switching (non-linear)	> 90%	Audio, low-frequency RF
E	Single-ended ZVS	Switching	> 90%	ISM bands, magnetron drive
F	Harmonic-shaped	Hybrid	80–90%	Modern handset PA, 2.4/5 GHz Wi-Fi
G/H	Supply switching	Linear	Variable	Envelope-tracking handset PA
J	Continuous-mode	Hybrid	70–80%	Wideband HF/VHF

Modern cellular handset PAs run envelope-tracking (ET) — the supply voltage tracks the OFDM signal envelope dynamically, recovering 8–12% absolute efficiency over a static-supply class-AB. Required external chip: ET supply modulator (Qorvo QM30100, Skyworks SKY58, Cirrus CS35Lxx) co-designed with the PA.

Decoupling-network design at RF

The classical “100 nF MLCC at every IC pin” rule is necessary but insufficient above 100 MHz. A 100 nF 0402 X7R has ESL ≈ 0.5 nH and self-resonates at ~22 MHz; above SRF it looks inductive. Build a decoupling cascade:

Frequency range	Capacitor	Package	Notes
< 1 MHz	10–100 µF tantalum or polymer	1206 / 2010	Bulk reservoir
1–30 MHz	1 µF X5R / X7R	0603	Energy reservoir
30–300 MHz	100 nF X7R	0402	Standard “supply bypass”
0.3–3 GHz	10 nF X7R or NP0	0402	Above XR’s SRF
3–10 GHz	100 pF – 1 nF NP0	0201 / 01005	Mounted at the pin, 1 via each side
> 10 GHz	On-chip capacitor + EBG	—	PCB caps too inductive

Each capacitor’s parallel combination must produce an impedance low enough (typically < 100 mΩ at the worst frequency) across the full band. Tools: ANSYS SIwave, Cadence Sigrity PowerDC + PowerSI for PDN impedance plots; rule-of-thumb hand-design is feasible up to about 1 GHz.

Coexistence rules of thumb

In a multi-radio product (phone, IoT module):

Antenna-to-antenna isolation ≥ 15 dB at the operating bands.
Frequency separation avoid harmonics: 2.4 GHz Wi-Fi 2nd = 4.8 GHz lands near 5 GHz Wi-Fi; LTE B7 (2.6 GHz) is the 2nd of 1.3 GHz which doesn’t exist in cellular but the 2nd of GPS L2 (1.227 GHz) lands at 2.454 GHz — Wi-Fi channels 1–6 desense GPS L2.
Filter the offending transmitter and add a low-rejection LNA-input filter to the victim receiver, before optimising layout.

Antenna placement on a small product

Place the antenna at the edge of the PCB ground plane, oriented with the feed pointing outward.
Ground-plane size of at least λ/4 by λ/4 around 2.4 GHz (31 × 31 mm) for acceptable radiation efficiency.
Keep components — especially shielded metal cans — at least 5 mm from the antenna keep-out region.
Use a small-form-factor antenna (chip, F-PIFA, IFA, monopole) from Pulse, Linx, Taoglas, Ignion, Antenova; load-pull tune the matching network on a representative product mockup.

Mixed-signal isolation rules

Separate RF and digital ground domains with single-point connection at a chosen reference node — and only where the return-current path demands it. Never split planes under high-speed signals.
Use guard rings around sensitive RF nodes (LNA input, oscillator tank) tied to ground every λ/20.
Shield-can the digital section at 30 dB+ if it sits within 50 mm of a high-gain RF block.
Clock-spreading (SSC) dithers the digital clock by ±0.5%, spreading harmonic energy across a broader band and reducing peak emissions by 6–10 dB.
Route DDR memory traces away from the antenna feed and any LNA input; 1 Gbps memory generates significant harmonic energy through 5 GHz.

Frequency-band engineering trade-offs

Band	Range typical	Loss in walls	Antenna size	Doppler at 60 km/h
Sub-GHz 868/915 MHz	1–10 km outdoor	3–8 dB / wall	λ/4 = 8 cm	50 Hz
2.4 GHz	30–100 m indoor	6–12 dB / wall	λ/4 = 3.1 cm	130 Hz
5–6 GHz	20–60 m indoor	10–18 dB / wall	λ/4 = 1.5 cm	300 Hz
28 GHz mm-wave	100 m line-of-sight	30–60 dB / wall	element 2 mm	1.5 kHz
77 GHz radar	250 m	n/a (outdoor)	1 mm patch	4 kHz

Doppler bandwidth scales linearly with carrier frequency — mm-wave channels coherence-time at vehicle speeds shrinks to sub-millisecond, demanding faster channel-tracking loops in the modem.

Useful constants and conversions

c = 2.998 × 10⁸ m/s
η₀ = 376.73 Ω
kT₀ (T = 290 K) = 4.00 × 10⁻²¹ J = −174 dBm/Hz noise spectral density
0 dBm = 1 mW = 0.224 V_rms into 50 Ω = 0.316 V_peak
+30 dBm = 1 W
+60 dBm = 1 kW
dBµV / dBm conversion at 50 Ω: dBµV = dBm + 107
dBµV/m / dBm/m² conversion: 0 dBm/m² = 145.8 dBµV/m
Wavelength λ (cm) = 30 / f (GHz)
λ/4 quarter-wave at 2.4 GHz on FR-4 microstrip ≈ 17 mm

Practical bench setup for new RF design

A new product RF lab needs (approximate 2026 commercial cost):

VNA 0.01–18 GHz (Anritsu MS46322B, ~ $60 k u se d; or P i co V N A 106$ 5k for hobbyist work).
Spectrum analyser 9 kHz – 26.5 GHz (Keysight N9020B MXA, ~ $50 k; S i g l e n tSS A 3000 X - R$ 5k for IoT bench).
Signal generator 9 kHz – 6 GHz (Keysight EXG N5172B, ~ $25 k; S D R - ba se d b e n c h so u rce$ 2k).
Power meter + thermal sensor 9 kHz – 6 GHz (Keysight U2001A, ~$3k).
Calibration kit (Anritsu 3653 SMA cal, ~ $3 k; ore - c a l Keys i g h t 85056 D$ 15k).
Calibrated noise source for NF measurement (Keysight 346B, ~$3k).
OTA chamber — closed-cavity reverberation chamber for compliance pre-scan (Ramsey REC-T, ~ $10 k f or h o bb y i s t; Bl u e t es tRTS 65$ 200k for production).
Component kit — 3 dB / 6 dB / 10 dB / 20 dB attenuators, 50 Ω terminations, splitters, bias-tees, DC blocks, isolators, LNAs (Mini-Circuits, Pasternack — budget $5k for a 5–10 unit kit).

This bench is sufficient for development through Wi-Fi 6 / Bluetooth / LoRa / sub-6 5G. mm-wave development (5G n257/n258/n260, automotive radar, 6G research) typically requires a contracted lab session at NIST, an accredited compliance house, or a university chamber.

Recent and ongoing developments (2024–2026)

Wi-Fi 7 (802.11be) ratification 2024; 320 MHz channels, 4096-QAM, multi-link operation (MLO); single-radio operating across 2.4 + 5 + 6 GHz simultaneously.
5G Advanced (3GPP Release 18/19, 2024–2026) — sub-band full duplex, network-controlled repeaters, AI-driven beam management.
6G research at D-band (110–170 GHz) and sub-THz — academic and consortia demonstrations at 100 Gbps over 100 m line-of-sight; commercial standardisation targeted 2028–2030.
Satellite direct-to-handset — Starlink + T-Mobile / AST SpaceMobile / Globalstar + Apple service launching in 2024–2025; new low-band (sub-1 GHz) PA + antenna requirements on phones.
Open RAN (O-RAN) — disaggregated base-station RF / DU / CU components driving demand for vendor-neutral RU hardware.
GaN-on-silicon for handset PAs — Wolfspeed, EpiGaN, MACOM commercialising GaN substrates that can be processed on standard 200 mm CMOS fabs.
Reconfigurable Intelligent Surfaces (RIS) — passive metasurfaces with controllable reflection characteristics, used as beam-steering “repeaters” for 5G coverage in indoor / NLOS scenarios; multiple commercial trials.

9. Edge cases and gotchas

Self-oscillation in PAs. A high-gain PA with finite output-input isolation oscillates if loop gain ≥ 1. The K-factor test catches the linear case; the load-pull test at the worst-case mismatch over the full output power range catches the large-signal case. Adding a low-Q series resistor or RC parallel feedback at the gate kills oscillation at the cost of 0.5–1 dB of small-signal gain.
Squegging in oscillators — slow modulation of the oscillator amplitude due to bias-network instability. Caused by a low-frequency parasitic loop through the supply or gate-leak resistor.
PIM (Passive Intermodulation). Rusty connector joints, ferromagnetic contamination in solder, or non-linear oxides at metal interfaces generate intermodulation products at high RF power (typical onset at +30 dBm). PIM is invisible at low power and ruins base-station receivers when high TX power excites it. Specifications PIM3 < −150 dBc at 2 × 43 dBm carriers.
Filter loading. A filter designed for a 50 Ω termination delivers its catalogue response only when terminated in 50 Ω. The input impedance of the next stage (LNA, mixer) is rarely exactly 50 Ω, so the filter’s actual response shifts. Datasheet curves are measured into 50 Ω; real-board response needs simulation with the actual next-stage S-parameters.
Group-delay variation distorts wideband signals (OFDM in particular). A filter’s amplitude response can be flat while its group delay varies by ±2 ns near the band edge; for an OFDM signal with 30 ns symbol cyclic prefix, this is significant.
Memory effects in PAs. Electrothermal time constants and bias-network charge storage make the PA’s response depend on its recent history, not just instantaneous input. Digital Pre-Distortion (DPD) in baseband corrects this — a polynomial or Volterra model of the PA is inverted and applied to the input signal, recovering linearity at the cost of computation and bandwidth (5× signal BW for the DPD path).
Temperature drift of varactor C-V and of ceramic capacitor C-T characteristics — NP0 / C0G are flat (±30 ppm/°C); X7R drifts ±15% over −55 to +125 °C; X5R ±15% but narrower temperature range; Y5V drops to 20% of nominal at the extremes. Never use X7R, X5R, or Y5V in an RF matching network — only NP0 / C0G.
Antenna detuning by the human body. A handset antenna design that achieves −10 dB return loss in free space typically degrades to −5 dB when the user grips the device (the “death grip” effect). Aperture tuning with digitally-tunable matching networks (Cavendish Kinetics MEMS, Qorvo RFFE-controlled tuners) compensates dynamically.
EMC at handset antenna near display. Display drivers radiate harmonics into the 700 MHz – 6 GHz cellular bands; a thin OLED panel with poor shielding can desense a cellular receiver by 5–10 dB. Mitigation: shield-can the display flex, route ground returns close to display signal lines, filter the display interface.
High-power RF safety. FCC and IEEE C95.1 set specific absorption rate (SAR) limits for handsets (1.6 W/kg averaged over 1 g of tissue, US; 2.0 W/kg over 10 g, EU). Beam-forming 5G mm-wave handsets must add MPE (maximum permissible exposure) compliance because mm-wave absorbs in skin, not bulk tissue.
Wi-Fi 6E / 7 in the 6 GHz band (5.925–7.125 GHz). Coexists with incumbent fixed-service users; FCC requires Automated Frequency Coordination (AFC) for standard-power devices, geofencing low-power-indoor (LPI) deployments to indoor use only. ETSI rules differ; products certified for both regions need region-aware firmware.
5G mm-wave reality. Beam-steering arrays + line-of-sight required, atmospheric absorption peaks near 60 GHz (O₂) and 183 GHz (H₂O); useful path lengths 200 m line-of-sight, 50 m through a window, 0 m through walls. mm-wave 5G has not displaced sub-6 GHz; it adds capacity in specific dense scenarios (stadia, airports, urban hot spots).
Test-fixture de-embedding. Measuring an IC’s S-parameters requires removing the contribution of the PCB launch, connectors, and probe pads. TRL (Through-Reflect-Line) and LRRM calibrations + de-embedding networks recover the device-under-test response. Failure to de-embed leads to 1–3 dB error in NF and gain measurements and ±5° phase error in matching design.
Cable flex and connector wear. Lab cables are graded — Sucoflex 104/106, Megaphase TM67, Pasternack — and a ±0.05 dB amplitude / ±2° phase stability over flex requires a real instrument-grade cable. Worn SMA centre pins go out of spec around 500–1000 mate cycles; precision 3.5 mm / 2.92 mm at 5000+. Always check connector pin depth with a gauge before high-precision measurement.
Bias-tee inductor self-resonance. The DC-feed inductor in a bias tee has a self-resonant frequency above which it looks capacitive. Below f_SRF/3 it acts as the intended choke; above, RF leaks into the DC supply. Cascade multiple inductor values + ferrite beads to cover from RF down to DC.
DC blocking capacitor low-frequency cutoff. A 100 pF DC block has X_C = 1.6 kΩ at 1 MHz — fine at GHz, but blocks the low-frequency content of broadband signals (pulse / OFDM). For wideband ESD pulse handling, choose DC blocks based on the lowest carrier frequency.
Crystal pulling and reference drift. A 10 MHz reference oscillator at ±1 ppm appears as ±2.4 kHz at 2.4 GHz after multiplication; ±10 kHz at 5G mid-band 3.5 GHz. Cellular handsets use AFC (automatic frequency control) loops that pull the TCXO from the base-station downlink to compensate.
EM coupling into bondwires. At mm-wave, the bondwires of a wirebonded IC act as inductors with 0.5–1 nH each and as antennas; the package becomes part of the matching network. Flip-chip or wafer-level chip-scale packages avoid this; modern 5G mm-wave RFICs almost always use FOWLP (fan-out wafer-level packaging).
Heat-induced gain drop. A PA at full output dissipates 60–70% of DC as heat; junction temperature rises 50–80 °C above ambient on a typical heatsink. GaN gain drops 0.01 dB/°C, GaAs HBT 0.02 dB/°C — measurable on long pulses. Drives the choice of pulse-width / duty-cycle in radar and the size of phone-PA thermal pads.
HBM / ESD damage to LNA gates. GaAs pHEMT gate-source breakdown is ~6 V; one human-body-model ESD event into an exposed antenna line kills the part. PIN-diode or anti-parallel-diode ESD clamps + DC return inductor on the antenna pin are mandatory.
VNA dynamic range cliff. A typical VNA noise floor at 1 GHz with 1 kHz IF bandwidth is ~−110 dBm; measurements of high-rejection filters (60–80 dB stopband) need narrower IFBW (10 Hz, slow) or external amplifiers + de-embedding to reach the floor.
Reference-plane shift over temperature. TRL calibration standards on FR-4 drift several picoseconds over a 40 °C span — a 60° phase shift at 60 GHz. Production test fixtures use Rogers or alumina substrates with low CTE for repeatable calibration.
OTA chamber dimension errors. Far-field distance d_ff = 2D²/λ; an array antenna of D = 0.3 m at 28 GHz has d_ff = 16.8 m — most anechoic chambers are smaller than that and rely on compact-range reflectors or near-field-to-far-field transformation. Misapplying free-space measurements at less than d_ff produces multi-dB pattern errors.
Battery-impedance modulation of antenna match. A small handheld product (medical wearable, asset tag) with a battery near the antenna sees the antenna match shift as the battery state-of-charge changes the chemistry’s dielectric constant by a few percent. Verified by VNA sweeps at full and depleted battery.
3D-printed enclosures and resonance. Plastic enclosures with metal-flake coatings appear at RF as partial Faraday cages; a 100 mm × 60 mm × 20 mm cavity has resonance modes at 1.5 GHz, 2.5 GHz, 4 GHz that can trap and re-radiate antenna or PA leakage. Real product enclosures need either fully conductive (with proper bond paths) or fully insulating designs — partial conductivity is the worst case.

9b. Radar and link-budget extensions

Radar range equation

For a monostatic radar at range R observing a target of radar cross-section σ:

P_r = (P_t · G_t · G_r · σ · λ²) / ((4π)³ · R⁴ · L)

with L the round-trip system loss. Range scales as R⁴ instead of R² — radar suffers a 12 dB penalty per range doubling vs. communication’s 6 dB. Automotive 77 GHz long-range radar (Continental ARS548, Bosch SRR / LRR, Aptiv ESR): P_t ≈ +12 dBm, G_t = G_r = 25 dBi, σ_car ≈ 10 m², λ = 3.9 mm; max range to a car ≈ 250 m at 13 dB SNR (sufficient for Doppler + range FFT integration).

Link-budget table for a 2.4 GHz LoRa-mesh sensor

Term	Value	Notes
TX power	+14 dBm	EU EN 300 220 ISM 868 max (10 mW)
TX antenna gain	+2 dBi	small chip antenna with ground plane
Cable / connector loss	−0.5 dB	board to antenna
Free-space path loss at 868 MHz, 1 km	−91.2 dB	FSPL
Building / vegetation penetration loss	−15 dB	typical for IoT-mesh urban
Fade margin	−10 dB	Rayleigh fading allowance
RX antenna gain	+2 dBi	identical sensor
RX cable loss	−0.5 dB
Received power	−99.2 dBm
RX sensitivity at SF12 / BW 125 kHz	−137 dBm	LoRa modulation
Link margin	+37.8 dB	ample for 1 km urban

LoRa achieves its remarkable −137 dBm sensitivity by trading data rate (~300 bps at SF12) for processing gain via chirp-spread-spectrum (CSS) modulation — Shannon’s theorem operating point.

Implementation losses and budget margin

Real transceivers have additional losses not captured in textbook formulas: I/Q imbalance (0.2–0.5 dB), phase-noise EVM contribution (0.3 dB), DAC/ADC quantisation (0.2 dB), filter passband ripple (0.5–1 dB), cable / connector tolerances (0.3 dB), temperature derating (0.5–1 dB). Aggregate implementation loss of 2–3 dB is a standard budget line — never assume the simulator gain is what the product delivers.

10. Tools and software

Simulators

Tool	Type	Strengths	Vendor
Keysight ADS	RF + microwave circuit + 2.5D EM (Momentum)	Industry default for RFIC, modules, PA, transceivers	Keysight
Cadence AWR Microwave Office	RF / microwave circuit + EM	Strong filter / PA synthesis	Cadence
Cadence Virtuoso RF	RFIC layout + simulation	Tape-out-grade RFIC flow	Cadence
ANSYS HFSS	3D full-wave FEM	Antennas, packages, complex structures	ANSYS
ANSYS Designer / Nexxim	Circuit + co-simulation	Pulls HFSS into circuit context	ANSYS
CST Studio Suite	3D FIT/FDTD time-domain	Broadband, transient (ESD, lightning, EMI)	Dassault
Sonnet	Planar 3D EM (MoM)	Microstrip, stripline, filters	Sonnet Software
Qucs-S, openEMS	Open-source	Pedagogical and lower-end production	FOSS
simNEC / NEC2 / 4NEC2	Wire-antenna MoM	Wire antennas, HF/VHF arrays	FOSS / Government
scikit-rf	Python S-parameter	Automated test, scripting, Touchstone manipulation	FOSS
LTspice / NI Multisim	Lumped circuit	LO/PLL, bias networks, low-frequency RF	Analog Devices / NI

PCB layout (RF-aware)

Cadence Allegro / OrCAD X — high-end commercial, RF-aware constraint manager, length matching, impedance-controlled stack-up.
Mentor (Siemens) Xpedition / PADS — competitor to Allegro.
Altium Designer — mid-tier, popular for RF/mixed-signal modules.
KiCad 8+ — FOSS, has improved differential routing and impedance calculators; common in academic and small-volume RF work.

Measurement instruments

Vector Network Analyzer (VNA) — measures complex S-parameters across a swept frequency. Lab-grade Keysight PNA-X / PXI, R&S ZNB / ZNA, Anritsu ShockLine; field/portable Keysight FieldFox, R&S ZNL; hobbyist NanoVNA (0.05–3 GHz, useful but noisy). 4-port models for differential measurements. Calibration kits (SOLT — Short-Open-Load-Through, or TRL — Through-Reflect-Line, or LRM / LRRM) define the measurement reference plane.
Spectrum analyzer — frequency-domain power vs frequency. Keysight N9020 / UXA, R&S FSW / FSV, Tektronix RSA / SignalVu. Add real-time bandwidth (160 MHz – 1.6 GHz spans captured live) for transient and frequency-hopping signal characterisation.
Signal generator — calibrated CW or modulated source. Keysight EXG / MXG / UXG, R&S SMW / SMB.
Noise-figure analyzer or Y-factor with calibrated noise source — Keysight U7227 / SNS, R&S FSV-K30. Modern approach: cold-source measurement on a VNA with NF option.
Power meter and sensor — Keysight U2000 series USB power sensors, R&S NRP, Mini-Circuits PWR series. Calibrated below 1% uncertainty.
Attenuators, terminations, DC blocks, bias-tees — Mini-Circuits (commodity), Marki Microwave (high-IP3), Krytar (broadband couplers), Pasternack, Fairview.

Production test and screening

Beyond bench characterisation, RF products undergo:

In-circuit test (ICT) — bed-of-nails probing for opens / shorts on the production PCB; not RF-specific but the first gate.
Functional RF test — golden-unit comparison on a calibrated production fixture; typical 1–5 second test time for a Wi-Fi / BT module.
Coupled-cavity OTA test — sealed RF chamber with a known coupling factor between DUT antenna and reference antenna; cheaper than a full anechoic chamber, accuracy 1–2 dB. Litepoint IQxel, Anritsu MT8870A, R&S CMW500 fronted with conducted-equivalent OTA fixture.
Compliance pre-scan — partial CISPR / FCC emissions in an in-house semi-anechoic chamber, before the final $50–200K accredited-lab session.

Calibration techniques

SOLT (Short-Open-Load-Through) — most common, requires precision standards, vendor-specific kits.
TRL (Through-Reflect-Line) — uses transmission-line standards, easier to fabricate in-house on test PCBs.
LRM / LRRM — line-reflect-match variants, used for on-wafer probing.
De-embedding — removes fixture / launch from a measurement; 2x-Thru / IEEE 370 standardised for high-speed serial channels.

Industrial chips (representative 2026 parts)

Function	Vendor / part	Notes
Wi-Fi 6E / 7 + BT	Qualcomm FastConnect 7800, Broadcom BCM4389 / 4395, Espressif ESP32-C6 / H2 / S3	FastConnect 7800 = Wi-Fi 7 320 MHz, BT 5.4
Cellular modem	Qualcomm Snapdragon X75 / X80 (5G + 6G research), MediaTek Dimensity 9300/9400 modem	mm-wave + sub-6, integrated front-end
GPS / GNSS	u-blox NEO-M9N (commercial), ZED-F9P (multi-band RTK), STMicroelectronics Teseo-V	F9P does cm-accuracy RTK
RF front-end module	Skyworks SKY77 / SKY13, Qorvo QM7 / QPF	Filter + LNA + PA + switch integrated
5G FEM	Qorvo QM77000, Skyworks SKY58, Murata 5G modules	Integrated PA / LNA / filter / switch for n77/n78/n79
LNA discrete	Qorvo TQP3M9009, Mini-Circuits PMA3-83LN+, Avago MGA-13116, Analog Devices HMC8410	NF 0.5–1.5 dB, sub-6 GHz
PA discrete	Qorvo TGA2598-CP (X-band GaN), Wolfspeed CGHV1J070D, NXP MMRF2014H	5–100 W ranges
Mixer	Marki ML1-0220LS, Analog Devices LTC5552, Mini-Circuits MAC-65LH+	DC – 18 GHz
VCO / PLL	Analog Devices ADF4371 / HMC833, TI LMX2820, Crystek CVCO55	Fractional-N, integrated VCO
RF switch	Analog Devices ADRF5040, Qorvo QPC6614, Peregrine PE42525, MACOM MASW	SP4T to SP10T, sub-6 GHz
Digital attenuator	Analog Devices HMC624A, Peregrine PE43712	6-bit, 0.5 dB step, DC – 6 GHz
5G cellular module	Quectel RG500Q, Sierra Wireless EM9191, Murata Type 1XK	Drop-in module with antenna interfaces
Software-defined radio	Analog Devices AD9361 / AD9371 / AD9081, Texas Instruments AFE7920, Xilinx Zynq UltraScale+ RFSoC	Direct RF sampling at 4–12 GS/s
BAW / SAW filter	Qorvo (Triquint) BAW, Skyworks SAW / TC-SAW, Murata SAW	Cellular RX/TX preselect
RF SoC (BLE / Thread)	Nordic nRF52840 / nRF54L15, Silicon Labs EFR32, Texas Instruments CC2652	Integrated transceiver + Cortex-M MCU
UWB	NXP Trimension SR040 / SR150 / SR250, Qorvo DW3000, Apple U1 / U2	6.5 / 8.0 GHz channels, time-of-flight ranging
Automotive radar	NXP TEF82xx, Infineon RXS8160PL / AURIX TC4xx, TI AWR2243 / AWR2944	76–81 GHz FMCW
5G mm-wave RFIC	Anokiwave AWMF-0156 / AWMF-0192, Renesas Sivers IMA, Movandi BeamX	24/28/39 GHz beamforming arrays

Modulation schemes and their RF impact

Modulation	PAPR (dB)	Spectral efficiency (bit/s/Hz)	Used by	PA back-off
FSK / GFSK	0	1	Bluetooth Classic, sub-GHz IoT	0 dB (saturated)
BPSK	3	1	LoRa, satellite, BPSK-FHSS	1–2 dB
QPSK / π/4-DQPSK	3.5	2	LTE control, GSM EDGE	2–3 dB
16-QAM	6.5	4	Wi-Fi MCS 3–4, LTE	4–5 dB
64-QAM	7.7	6	Wi-Fi MCS 5–7, LTE 256-QAM/cat.	5–6 dB
256-QAM	8.5	8	Wi-Fi 6 MCS 8/9, LTE PRO	7 dB
1024-QAM	8.7	10	Wi-Fi 6 MCS 10/11	8 dB
4096-QAM	9.0	12	Wi-Fi 7 MCS 12/13	9 dB
OFDM (multi-carrier)	10–12 (uncoded)	depends on modulation	Wi-Fi, LTE, 5G, DVB-T	8–10 dB

The peak-to-average power ratio (PAPR) sets the PA back-off — the operating point below saturation that the PA must maintain to keep the peak below P1dB. A handset PA rated 27 dBm peak at the antenna delivers only 17 dBm average for 256-QAM OFDM, with the corresponding efficiency drop. Envelope tracking and DPD recover much of this loss; crest-factor reduction (CFR) in baseband clips and reshapes peaks at a small EVM cost.

RF-aware DFM (Design for Manufacture)

Stencil thickness and aperture design for tight-tolerance RF caps (0201, 01005) — too much solder forms a fillet that detunes the package by 1–2 GHz at 28 GHz.
Component placement keep-outs around RF blocks — no test points in the keep-out, no silkscreen overlapping ground vias, no glue dots near matching networks.
ESD-clamp diode placement — outside the impedance-controlled trace, on a separate stub, to avoid loading the antenna feed.
Plated edge half-vias for shield-can soldering — keep RF traces 0.5 mm clear of the can wall to prevent solder bridging.
Panel-edge breakout tabs — placed so as not to act as antennas at 2.4 / 5 GHz; mouse-bite count and pitch matter for resonance.

11. Cross-references

[[Engineering/electromagnetics-engineering]] — Tier 1 foundation: Maxwell, wave equation, Γ, VSWR, η₀, microstrip impedance, skin depth.
[[Engineering/pcb-design]] — controlled impedance, stack-up, return paths, decoupling networks; the layout substrate on which every RF circuit sits.
[[Engineering/semiconductor-devices]] — physics of HEMT, HBT, BJT, MOSFET; small-signal models that go into S-parameter device files.
[[Engineering/op-amps]] — low-frequency sensor signal-conditioning that feeds RF chains in instrumentation receivers.
[[Engineering/signal-processing-dsp]] — IF / baseband digital filtering, DPD, channel equalisation, OFDM demodulation.
[[Engineering/antenna-theory]] — planned companion in this batch; dipoles, arrays, beam steering, ground-plane integration.
[[Engineering/photonics]] — planned; optical-frequency analogue (the RF / optical regime boundary).
[[Engineering/microcontrollers]] — planned; modern MCUs include RF SoCs (nRF52/53, ESP32, STM32WB), and RF design surrounds the digital core.
[[Engineering/rf-design]] — distributed-parameter analysis, Smith chart, S-parameters in their general form.
[[Engineering/power-electronics]] — class-D/E/F resonant converters share the same matching-network mathematics as PA output stages.
[[Robotics/sensors-pose-motion]] — GNSS-RTK, UWB, radar — the RF sensing modalities of mobile robots.
[[Languages/Tier3/network-protocol-dsls]] — IEEE 802.11 / 802.15 / Bluetooth / cellular protocol stacks that ride on the RF physical layer.
[[Languages/Tier3/network-protocol-dsls]] — if/when added, the radio-protocol descriptor families (LoRa, Zigbee, Thread, Matter).
[[Languages/Tier3/scientific]] — the standard environment for RF/communications system modelling.

12. Citations

Pozar, D. M. (2011). Microwave Engineering (4th ed.). Wiley. The canonical textbook for transmission lines, S-parameters, Smith chart, matching networks, filters, and microwave amplifiers.
Razavi, B. (2011). RF Microelectronics (2nd ed.). Prentice Hall. Standard reference for integrated transceiver design, noise analysis, mixer topologies, frequency synthesis.
Lee, T. H. (2004). The Design of CMOS Radio-Frequency Integrated Circuits (2nd ed.). Cambridge University Press. Influential treatment of CMOS-process RF design with deep coverage of LNAs, mixers, oscillators.
Maas, S. A. (2003). Nonlinear Microwave and RF Circuits (2nd ed.). Artech House. Mixers, harmonic balance, intermodulation, PA non-linearity.
Maas, S. A. (1992). Microwave Mixers (2nd ed.). Artech House. The reference on mixer design and analysis.
Rogers, J. & Plett, C. (2010). Radio Frequency Integrated Circuit Design (2nd ed.). Artech House. Industry-oriented complement to Razavi.
Steer, M. (2019). Microwave and RF Design (3rd ed., 5-volume open-source set). NC State University Press. Available free under CC license; volumes cover transmission lines, networks, modules, amplifiers, and radio systems.
Gonzalez, G. (1996). Microwave Transistor Amplifiers: Analysis and Design (2nd ed.). Prentice Hall. The classical reference on small-signal microwave amplifier design with S-parameters and stability circles.
Smith, P. H. (1939). “Transmission Line Calculator.” Electronics 12(1): 29–31. The original Smith-chart paper.
Smith, P. H. (1944). “An Improved Transmission Line Calculator.” Electronics 17(1): 130–133. The expanded immittance (impedance + admittance) chart.
Friis, H. T. (1944). “Noise Figures of Radio Receivers.” Proceedings of the IRE 32(7): 419–422. The cascaded noise-figure formula.
Friis, H. T. (1946). “A Note on a Simple Transmission Formula.” Proceedings of the IRE 34(5): 254–256. The Friis free-space-path-loss equation.
Kurokawa, K. (1965). “Power Waves and the Scattering Matrix.” IEEE Transactions on Microwave Theory and Techniques 13(2): 194–202. The original power-wave (S-parameter) formalism.
Maxwell, J. C. (1865). “A Dynamical Theory of the Electromagnetic Field.” Philosophical Transactions of the Royal Society 155: 459–512. Foundational; the wave equation predicting light as electromagnetism.
Rollett, J. M. (1962). “Stability and Power-Gain Invariants of Linear Twoports.” IRE Transactions on Circuit Theory 9(1): 29–32. The K-factor stability criterion.
Hammerstad, E. & Jensen, Ø. (1980). “Accurate Models for Microstrip Computer-Aided Design.” IEEE MTT-S Digest 80: 407–409. The microstrip-impedance closed-form approximation used in most CAD tools.
Klopfenstein, R. W. (1956). “A Transmission Line Taper of Improved Design.” Proceedings of the IRE 44(1): 31–35. The optimal (minimum-length-for-given-reflection) tapered-line impedance transformer.
Fano, R. M. (1950). “Theoretical Limitations on the Broadband Matching of Arbitrary Impedances.” Journal of the Franklin Institute 249(1–2): 57–83, 139–154. The Bode-Fano integral bandwidth-match bound.
Cohn, S. B. (1957). “Direct-Coupled-Resonator Filters.” Proceedings of the IRE 45(2): 187–196. The classical coupled-resonator microwave filter synthesis.
Matthaei, G., Young, L. & Jones, E. M. T. (1964). Microwave Filters, Impedance-Matching Networks, and Coupling Structures. McGraw-Hill (reprinted Artech 1980). The canonical microwave-filter design reference.
Bowick, C. (2008). RF Circuit Design (2nd ed.). Newnes. Practical handbook for matching networks and discrete RF design.
Cripps, S. C. (2006). RF Power Amplifiers for Wireless Communications (2nd ed.). Artech House. The standard PA design reference; load-pull, class definitions, linearisation.
Verspecht, J. & Root, D. E. (2006). “Polyharmonic Distortion Modeling.” IEEE Microwave Magazine 7(3): 44–57. The X-parameters formalism extending S-parameters to non-linear devices.
Shannon, C. E. (1948). “A Mathematical Theory of Communication.” Bell System Technical Journal 27: 379–423, 623–656. The Shannon-Hartley capacity bound on which every modern link budget rests.
IEEE Std 802.11ax-2021 / 802.11be-2024. Wireless LAN Medium Access Control and Physical Layer Specifications, Amendments for Enhancements for High Efficiency / Extremely High Throughput (Wi-Fi 6 / 6E / 7).
3GPP TS 38.101 / TS 38.104 / TS 38.141. NR; User Equipment / Base Station radio transmission and reception. The 5G NR radio specifications (frequency bands, EVM, ACLR, spurious limits).
3GPP TS 38.211 / 38.212 / 38.213. NR; Physical channels and modulation / Multiplexing and channel coding / Physical layer procedures.
Bluetooth SIG. Bluetooth Core Specification 5.4 (2023). RF specifications for classic and LE radios.
FCC 47 CFR Part 15. Radio frequency devices. Subpart B (unintentional radiators), Subpart C (intentional radiators including ISM and Wi-Fi), Subpart E (U-NII bands).
FCC 47 CFR Part 96. Citizens Broadband Radio Service (3.55–3.7 GHz CBRS spectrum sharing).
ETSI EN 300 328 v2.2.2. Wideband transmission systems; Data transmission equipment operating in the 2.4 GHz band.
ETSI EN 301 893 v2.1.1. 5 GHz RLAN; Harmonised Standard covering the essential requirements of article 3.2 of Directive 2014/53/EU.
ETSI TS 138 series. 5G NR (ETSI republication of 3GPP TS 38.x).
IEC 61169 series. Radio-frequency connectors. Multi-part standard on mechanical and electrical interfaces (e.g. 61169-15 SMA, 61169-16 Type N, 61169-17 BNC).
IEEE Std C95.1-2019. IEEE Standard for Safety Levels with Respect to Human Exposure to Electric, Magnetic, and Electromagnetic Fields, 0 Hz to 300 GHz. SAR / MPE basis.
ITU-R Recommendation P.676-13 (2022). Attenuation by atmospheric gases and related effects. The reference for mm-wave atmospheric absorption (O₂ at 60 GHz, H₂O at 183 GHz).
Keysight Technologies. ADS Documentation (version 2023). Smith Chart Utility, harmonic balance, envelope simulation.
ANSYS. HFSS User’s Guide (2024 R1). Mesh adaptation, port definitions, near-/far-field post-processing.
Dassault Systèmes. CST Studio Suite Documentation (2024).
IEEE 370-2020. Standard for Electrical Characterization of Printed Circuit Board and Related Interconnects at Frequencies up to 50 GHz. 2x-Thru de-embedding methodology.
IBIS Open Forum. Touchstone File Format Specification, Version 2.1 (2024). The standard interchange format for N-port S-parameter data.
IEEE 1597.1-2008. Standard for Validation of Computational Electromagnetics Computer Modeling and Simulations. Verification methodology for full-wave solvers.
ARRL. The ARRL Handbook for Radio Communications (annual). Comprehensive practical RF reference; the amateur-radio canon and a useful entry point into practical design.

Compendium

Explorer

RF Design (Smith Chart, S-Parameters, Impedance Matching) — Engineering Reference