power amplifier

**Power amplifier (PA)** is an electronic circuit that increases the power level of an RF or audio signal — converting DC supply power into output signal power. PAs are defined by the transistor technology, biasing class, matching networks, and the fundamental efficiency-linearity trade-off. **Operating classes** determine where the transistor is biased on its I-V curve. Class A (conduction angle 360°) is the most linear but least efficient (~50%); Class AB (180-360°) balances linearity and efficiency; Class B (180°) reaches ~78% theoretical efficiency; Class C (< 180°) is highly efficient but nonlinear. Switched-mode classes (D, E, F) can exceed 90% efficiency but require complex harmonic tuning. **Key specifications**: Power Added Efficiency (PAE = (Pout-Pin)/PDC) measures how well DC power converts to RF; P1dB is the output power where gain compresses by 1 dB; IP3 (third-order intercept) quantifies intermodulation distortion; Error Vector Magnitude (EVM) measures signal fidelity for modulated waveforms. **Technologies**: GaN HEMT dominates high-power (>10W) applications due to high breakdown voltage and electron velocity. GaAs pHEMT leads in high-frequency (>10 GHz) mobile and millimeter-wave. LDMOS silicon handles high-power cellular base stations. SiGe BiCMOS enables highly integrated mmWave arrays. **Linearization** is critical for modern wideband modulation (5G NR, 802.11ax). Digital Pre-Distortion (DPD) is the industry standard — a lookup table or polynomial model pre-warps the input to compensate for the PA's AM-AM and AM-PM transfer characteristics, enabling Class-AB efficiency with Class-A linearity. **AI/ML context**: AI is transforming PA design through neural-network behavioral models (replacing slow SPICE simulations), reinforcement-learning DPD, and automatic load-pull optimization. For AI chip systems, high-speed wireless interconnects (future mmWave NVLink) will require integrated GaN PAs; autonomous vehicle radar ASICs embed SiGe PA arrays alongside neural network inference engines. ```svg PA Circuit Topology VDD RFC Lm Cm RL Gate D S GaN HEMT GND IMN RF in Vbias→ RF out Key PA Parameters Gain (dB): power out / power in ratio P1dB: 1dB compression point PAE (%): (Pout-Pin)/PDC × 100 IP3: 3rd-order intercept Noise Fig: NF = 10log(SNR_in/SNR_out) VSWR: impedance match quality EVM: error vector magnitude (IQ) Load-Line + Amplifier Classes Vds Id Load line Sat. A (50%) AB B (78%) C Vq Idq Class Efficiency vs Linearity: A: ~50%, best linear AB: ~60%, good linear B: ~78%, moderate C: ~90%+, nonlinear PA Technologies + Topologies Transistor Technologies: GaN HEMT 30–100W, 1–40GHz, 70% PAE mmWave 5G base stations, radar GaAs pHEMT 0.5–5W, 1–100GHz, 45% PAE Mobile handsets, WiFi 6E/7 LDMOS Si 50–300W, <4GHz, 65% PAE Cellular macro BTS, broadcast SiGe BiCMOS 1mW–100mW, to 300GHz mmWave phased array, automotive Efficiency Topologies: Doherty PA carrier + peaking; 50-60% avg Class-E switch-mode; >90% peak Envelope Tracking VDD tracks envelope; +5-8% Outphasing (LINC) 2 PAs + combiner; >70% eff Matching Networks + Stability Impedance Matching: Source: 50Ω → low Ropt of transistor Output: Ropt → 50Ω load Techniques: L-net, Pi-net, T-net, coupled Bandwidth: Bode-Fano limit: BW × RL inversely linked Broadband: resistive match (lossy), filters Stability (K-factor): Rollett K > 1 + |Δ| < 1 = unconditionally stable Source/load terminations must avoid |S11|>1 Stabilize: series R at gate (lossy, NF penalty) Linearization Techniques DPD (Digital Pre-Distortion): Pre-warps input to cancel PA nonlinearity LUT or Volterra series model; adaptive Corrects AM-AM and AM-PM distortion Feedback methods: Cartesian feedback: IQ loop, <20MHz BW Polar: envelope + phase separate paths Feed-forward: Error amplifier cancels distortion; wideband High linearity but poor efficiency (aux amp) Used in BTS where power budget allows PA in AI Chip Systems AI accelerator connectivity: NVLink/PCIe: copper, no PA needed Wireless NVLink (future): mmWave PA required Compute-in-package RF: GaN on SiC flip-chip AI-driven PA design: RL-based DPD: 3dB better ACLR vs LUT DPD Neural network load-pull: predict Zopt GAN-based PA behavioral models (fast SPICE) Radar + sensing: Automotive radar 77GHz: SiGe PA, 20dBm AI inference for beam-steering (phased array) ```

Go deeper with CFSGPT

Get AI-powered deep-dives, save terms, and run advanced simulations — free account.

Create Free Account