RF design is the engineering of circuits and interconnects that generate, receive, translate, and measure signals whose physical wavelength and parasitics materially affect behavior. At radio and microwave frequencies, a conductor is no longer an ideal wire, a capacitor includes unwanted inductance, transistor gain changes with frequency, and the package becomes part of the circuit. Successful RF design therefore joins system budgets, transistor circuits, transmission-line theory, electromagnetic fields, layout, packaging, calibration, and measurement into one evidence-driven workflow.

The design begins with a communication or sensing requirement, not a transistor schematic. Frequency band, channel bandwidth, modulation, data rate, range, antenna gain, blocker environment, allowed error rate, output power, supply, area, and regulatory limits determine the signal chain. The receiver must detect the weakest useful signal without being corrupted by its own noise or nearby interferers. The transmitter must deliver enough clean power while limiting harmonics, spectral regrowth, leakage, heat, and battery drain.

<svg viewBox="0 0 960 400" xmlns="http://www.w3.org/2000/svg" font-family="-apple-system,Segoe UI,Roboto,sans-serif">
  <defs><marker id="rfArrow" markerWidth="8" markerHeight="8" refX="7" refY="4" orient="auto"><path d="M0 0 L8 4 L0 8 Z" fill="#94a3b8"/></marker></defs>
  <rect width="960" height="400" rx="16" fill="#111318"/>
  <text x="480" y="34" fill="#e5e7eb" font-size="22" font-weight="700" text-anchor="middle">Integrated RF transceiver: signal, frequency, and control paths</text>
  <text x="480" y="58" fill="#94a3b8" font-size="13" text-anchor="middle">Every block shares the antenna, reference clock, supplies, substrate, package, and thermal budget</text>
  <g text-anchor="middle" font-size="14" font-weight="700">
    <rect x="34" y="104" width="116" height="66" rx="10" fill="#132238" stroke="#60a5fa" stroke-width="2"/><text x="92" y="132" fill="#bfdbfe">Antenna</text><text x="92" y="153" fill="#94a3b8" font-size="11" font-weight="400">duplexer / switch</text>
    <rect x="194" y="104" width="116" height="66" rx="10" fill="#10251f" stroke="#34d399" stroke-width="2"/><text x="252" y="132" fill="#a7f3d0">LNA</text><text x="252" y="153" fill="#94a3b8" font-size="11" font-weight="400">low-noise gain</text>
    <rect x="354" y="104" width="116" height="66" rx="10" fill="#292013" stroke="#fbbf24" stroke-width="2"/><text x="412" y="132" fill="#fde68a">Mixer</text><text x="412" y="153" fill="#94a3b8" font-size="11" font-weight="400">RF → baseband</text>
    <rect x="514" y="104" width="116" height="66" rx="10" fill="#1c1b2f" stroke="#a78bfa" stroke-width="2"/><text x="572" y="132" fill="#ddd6fe">Filter / VGA</text><text x="572" y="153" fill="#94a3b8" font-size="11" font-weight="400">select + scale</text>
    <rect x="674" y="104" width="116" height="66" rx="10" fill="#251723" stroke="#f472b6" stroke-width="2"/><text x="732" y="132" fill="#fbcfe8">ADC</text><text x="732" y="153" fill="#94a3b8" font-size="11" font-weight="400">sample + quantize</text>
    <rect x="834" y="104" width="92" height="66" rx="10" fill="#15202b" stroke="#22d3ee" stroke-width="2"/><text x="880" y="132" fill="#a5f3fc">DSP</text><text x="880" y="153" fill="#94a3b8" font-size="11" font-weight="400">recover data</text>
  </g>
  <g stroke="#94a3b8" stroke-width="2" marker-end="url(#rfArrow)"><path d="M150 137h36"/><path d="M310 137h36"/><path d="M470 137h36"/><path d="M630 137h36"/><path d="M790 137h36"/></g>
  <rect x="194" y="236" width="276" height="96" rx="11" fill="#181a27" stroke="#a78bfa" stroke-width="2"/>
  <text x="332" y="265" fill="#ddd6fe" font-size="16" font-weight="700" text-anchor="middle">Frequency synthesis</text>
  <text x="332" y="290" fill="#cbd5e1" font-size="13" text-anchor="middle">reference → phase detector → loop filter → VCO</text>
  <text x="332" y="313" fill="#94a3b8" font-size="12" text-anchor="middle">phase noise and spurs enter through mixer LO ports</text>
  <rect x="514" y="236" width="276" height="96" rx="11" fill="#182018" stroke="#84cc16" stroke-width="2"/>
  <text x="652" y="265" fill="#d9f99d" font-size="16" font-weight="700" text-anchor="middle">Transmit path</text>
  <text x="652" y="290" fill="#cbd5e1" font-size="13" text-anchor="middle">DAC → modulator → driver → power amplifier</text>
  <text x="652" y="313" fill="#94a3b8" font-size="12" text-anchor="middle">linearity, efficiency, EVM, and spectral mask</text>
  <path d="M412 228v-50" stroke="#c4b5fd" stroke-width="2" stroke-dasharray="6 5" marker-end="url(#rfArrow)"/>
  <path d="M652 332v20H92V178" fill="none" stroke="#a3e635" stroke-width="2" stroke-dasharray="6 5" marker-end="url(#rfArrow)"/>
  <text x="480" y="374" fill="#94a3b8" font-size="13" text-anchor="middle">Co-design across schematic, extracted layout, EM model, package, board, antenna, and calibration</text>
</svg>

A link budget converts range and environment into circuit requirements. In logarithmic units, received power is the transmitted power plus antenna and path gains minus path, mismatch, polarization, cable, filter, and implementation losses:

$$P_{RX}=P_{TX}+G_{TX}+G_{RX}-L_{path}-L_{misc}$$

All terms must use consistent dB or dBm conventions. Free-space path loss grows with distance and frequency, but real channels add obstruction, reflection, fading, atmospheric absorption, body loss, and antenna detuning. Required receiver sensitivity follows from thermal noise, bandwidth, receiver noise figure, and the signal-to-noise ratio needed by the modulation and coding scheme:

$$P_{sens}\approx -174\ \text{dBm/Hz}+10\log_{10}(B)+NF+SNR_{required}$$

The familiar −174 dBm/Hz value is the approximate available thermal-noise density at room temperature. It is a reference, not a guarantee: temperature, impedance, filtering, implementation margin, and interference change the actual floor.

S-parameters describe how traveling waves interact with an RF network. With all unused ports terminated in the reference impedance, (S_{11}) is input reflection, (S_{21}) is forward transmission or gain, (S_{12}) is reverse transmission, and (S_{22}) is output reflection. Return loss is commonly reported as

$$RL=-20\log_{10}|S_{11}|$$

A return loss of 10 dB means about 10% of incident power is reflected, not that the impedance is “10% wrong.” S-parameters depend on frequency, bias, power level, temperature, reference plane, and port impedance. Small-signal measurements do not predict compression or harmonic generation, so power sweeps and nonlinear characterization are separate requirements.

Impedance matching controls power transfer, noise, gain, and voltage swing. The reflection coefficient for load (Z_L) relative to reference (Z_0) is

$$\Gamma=\frac{Z_L-Z_0}{Z_L+Z_0}$$

The Smith chart maps complex impedance and admittance through (Gamma), making series and shunt transformations visually tractable. A conjugate power match is not always the correct target. An LNA may accept a different input impedance for minimum noise figure; a PA may use load-pull data to choose an impedance for output power, efficiency, or linearity; a broadband interface may favor moderate match across an octave over a narrow perfect match.

The LNA establishes receiver noise performance while surviving strong blockers. Friis’ cascade relationship shows why early gain is valuable:

$$F_{total}=F_1+\frac{F_2-1}{G_1}+\frac{F_3-1}{G_1G_2}+\cdots$$

Here noise factors (F) and gains (G) are linear ratios, not dB values. Low first-stage noise factor and adequate gain suppress later-stage noise contributions. Excessive gain, however, can overload the mixer when a nearby transmitter or interferer is present. LNAs therefore balance noise match, input return loss, gain, linearity, current, ESD parasitics, and stability.

Mixers multiply waveforms and deliberately create sum and difference frequencies. Passive switching mixers can provide high linearity and low flicker-noise contribution but have conversion loss and require LO drive. Active Gilbert-cell mixers provide conversion gain and isolation at the cost of noise, headroom, and current. Mixer specifications include conversion gain, noise figure, input compression, IIP2/IIP3, port isolation, LO feedthrough, and spurious responses.

Frequency synthesizers determine tuning accuracy and close-in spectral purity. A phase-locked loop compares a divided oscillator phase with a reference and corrects error through a loop filter. Inside loop bandwidth, reference, divider, detector, and charge-pump noise are shaped differently from VCO noise outside the bandwidth. Wider bandwidth can suppress VCO noise and shorten settling, but passes more reference-path noise and may worsen spurs or stability.

Power amplifiers convert DC power into controlled RF power. Drain efficiency is RF output power divided by DC input power. Power-added efficiency accounts for RF drive:

$$PAE=\frac{P_{out}-P_{in}}{P_{DC}}$$

Saturated and switched-mode operation can be efficient for constant-envelope signals, but modern high-order modulation has a varying envelope and demands linear amplification or linearization. Output back-off preserves error-vector magnitude and adjacent-channel leakage at an efficiency penalty. Envelope tracking, digital predistortion, Doherty combining, and load modulation recover some efficiency over realistic power distributions.

Linearity predicts how strong signals create unwanted signals. The 1 dB compression point marks significant gain reduction. Third-order intercept is an extrapolated figure for weakly nonlinear behavior; it is not a safe operating power. Two tones at (f_1) and (f_2) produce third-order products at (2f_1-f_2) and (2f_2-f_1), which can fall close to the desired band and resist filtering. Even-order distortion is especially important in direct-conversion receivers because it can create baseband interference.

Stability must be demonstrated inside and outside the intended band. Feedback through transistor capacitance, substrate, supply, package, or shared ground can create oscillation. Rollet’s (K), determinant (Delta), stability circles, loop gain, and pole-zero analysis provide complementary evidence. A two-port that is unconditionally stable under a small-signal model can still show large-signal, odd-mode, bias, or low-frequency instability.

Physical layout is part of the RF schematic. Metal width, spacing, thickness, return-current path, via arrays, ground shields, device fingers, and component orientation determine inductance, resistance, capacitance, coupling, and current density. Differential paths require electromagnetic symmetry, not merely equal drawn length. Inductors need controlled spacing from noisy metals and neighboring magnetic structures. Guard rings and isolated wells reduce substrate coupling while adding parasitics that must be modeled.

Critical passive structures are extracted with a planar or three-dimensional EM solver. Ports and reference planes must correspond to the circuit model. The EM result is connected to transistor models for harmonic-balance, periodic steady-state, transient, noise, and modulated simulations. Partitioning errors—double-counted interconnect, missing return paths, ideal grounds, or poorly placed ports—can matter more than solver precision.

Package, board, and antenna co-design prevent boundary surprises. Bond wires, bumps, redistribution layers, substrate traces, vias, balls, connectors, and board launches form a cascaded network. Package resonance or ground inductance can undo an excellent on-die match. Antenna impedance changes with enclosure, battery, hand position, nearby radios, and manufacturing tolerance. Co-simulation and measured fixtures are required to allocate margin across these boundaries.

Verification uses the correct analysis for each operating regime. DC operating points confirm device regions and headroom. S-parameter analysis covers small-signal gain, match, isolation, group delay, and stability. Noise analysis computes source contributions. Harmonic balance and periodic steady state handle mixers, oscillators, compression, and frequency translation. Transient and envelope simulation capture startup, settling, modulation, AGC, and memory effects.

A robust RF design closes one continuous evidence loop. Translate the use case into link, noise, linearity, phase-noise, power, and thermal budgets; select an architecture and frequency plan; characterize devices and passives; design and extract the physical network; co-simulate die, package, board, and antenna; measure with calibrated reference planes; and feed discrepancies back into models. At RF, every boundary is an electrical component. Treating those boundaries explicitly is what turns a promising schematic into a manufacturable radio.