Mobile RF Front-End: Power Amplifier, Filter and Antenna Switch Architecture

The RF front-end (RFFE) is where bits become waves. Between the modem and the antenna sits a dense chain of power amplifiers, acoustic filters, and antenna switches that determines your phone’s range, battery life, and whether it holds a 5G call at the cell edge. We tear down the architecture.

1. The Modern RF Chain: From Bits to Air

A smartphone radio is a brutal compromise. The baseband spits out I/Q samples, the RF transceiver mixes them to 600 MHz – 7 GHz, and the front-end must launch them from a sub-5mm-thick chassis while sharing antennas with Wi-Fi, GPS, UWB, and NFC — without desensing any receiver.

In a 2025 flagship teardown we counted 11 RF modules around the perimeter: 2x PAMiD (power amp + duplexer), 2x L-PAMiD, 3x diversity/MIMO FEMs, 1x mmWave IF module, 2x antenna tuners, and 1x SP10T switch. The modem itself — handling compute not far from on-device AI image generation workloads — only sees digital. Everything analog lives on the front-end.

Key specs that drive architecture:

• Regulatory mask: 3GPP demands -50 dBm/MHz spurious and adjacent channel leakage ratio (ACLR) >31 dB at +23 dBm. Your PA linearity budget is ~2% EVM.
• Insertion loss (IL): Every 0.5 dB loss after the PA costs ~11% battery current at cell edge. This is why filter IL and switch Ron dominate vendor selection.
• Carrier aggregation (CA) and ENDC: 4CA + 2x MIMO means up to 8 simultaneous receive paths. Filter isolation and switch linearity (IIP3 > +85 dBm) become system-level.

2. Power Amplifier: Efficiency vs Linearity

The PA is the hungriest block in the phone. At +26 dBm output, a Class-AB GaAs PA draws ~650 mA from 3.8V. Power-added efficiency (PAE) is everything.

Technology choice

Low-band (600–960 MHz) still prefers GaAs HBT for high breakdown and thermal conductivity. Mid/high-band (1.7–2.7 GHz) and 5G UHB (3.3–5 GHz) have shifted to SOI CMOS and SiGe in PAMiD modules because they integrate the switch and controller. GaAs PAE leads by 3–5 points, but SOI wins on integration and cost.

Supply modulation

Phones no longer run the PA at fixed Vcc. Two schemes dominate:

• Average Power Tracking (APT): DC-DC steps supply per slot (10–20 µs). Cheap, ~10% efficiency gain. Used for LTE and low-power NR.
• Envelope Tracking (ET): A fast tracker (e.g., Qualcomm QET5100) modulates Vcc at 100 MHz to follow the RF envelope. PAE jumps from ~32% to 43% on a 100-MHz NR carrier, cutting junction temp by 12–15°C. ET is mandatory above +22 dBm in n77/n78.

For 256-QAM and HPUE (power class 2, +26 dBm), vendors add digital predistortion (DPD) in the transceiver. DPD extends the P1dB by ~1.5 dB and keeps ACLR under -38 dBc without backing off, but costs current in the baseband — a trade-off analogous to Vulkan driver optimization where a small CPU overhead buys large GPU efficiency.

Notice the triggered write. All RFFE devices share a 26 MHz SCLK/SDATA bus; a broadcast trigger aligns PA, switch, and tuner in the same microsecond to avoid hot-switching 2W of RF.

3. Filters: Why BAW Won at 5G

Acoustic filters do the heavy lifting of duplexing Tx and Rx with only 10–400 MHz guard band. Two physics dominate:

• SAW (Surface Acoustic Wave): Interdigital transducers on LiTaO3/LiNbO3. Low cost, excellent for <2 GHz. Temperature drift ~ -30 ppm/°C, Q ~ 800–1500.
• BAW/FBAR (Bulk Acoustic Wave): Resonator stacked vertically (AlN/Mo/AlN). Q > 2000, temp drift ~ -18 ppm/°C, handles higher power and frequency. Cost 2–3× SAW.

For n7, n41, n77, and Wi-Fi coexistence at 5–6 GHz, SAW insertion loss above 2.5 dB and its thermal drift would miss the 3GPP blocking spec. That is why every flagship since 2023 uses BAW for bands above 2.3 GHz and TC-SAW or IHP-SAW (Murata’s temperature-compensated) for low-band.

In teardown, look for the duplexers: Murata, Broadcom (Avago), Taiyo Yuden, and Qorvo supply most BAW. A typical n78 PAMiD integrates 5 BAW resonators + 2 SAW for LB/MB in one 5.2 × 4.1 mm LGA. Bulk acoustic resonators are hermetically sealed — you can spot the cap wafer under X-ray.

Spec example — n77 (3.3–4.2 GHz) Tx BAW duplexer: Center 3.75 GHz, BW 900 MHz, IL 1.6 dB (Tx), 2.1 dB (Rx), Tx-Rx isolation >55 dB, rejection at Wi-Fi 5.5 GHz >45 dB. To hit that, designers use ladder + lattice topologies and add a notch for GPS L1 when needed.

4. Antenna Switch and Aperture Tuning

The antenna switch module (ASM) looks simple — it is a large SOI CMOS matrix of FETs — but its linearity sets your intermodulation floor. An SP10T switch in the HB path must handle +26 dBm with second harmonic < -45 dBm and IMD3 < -110 dBm when +20 dBm blockers are present.

Three switch functions exist in every phone:

1. Band switch (ASM): Routes one of 8–12 Tx/Rx filter outputs to the main antenna. Ron < 0.8 Ω per throw, Coff ~ 65 fF. Isolation 30–38 dB is critical for CA.
2. Antenna swap switch: Flips between top and bottom antennas for SAR and hand-grip. Insertion loss budget < 0.4 dB or you lose the PA efficiency you paid for.
3. Aperture/impedance tuner: A digitally tunable capacitor (DTC) + SP4T that shifts antenna resonance. Qorvo QM13048 or Skyworks SKY59272 provides 32 states, 0.6–3.8 pF, Q > 80 @ 1 GHz. Firmware runs a closed loop with the transceiver’s VSWR detector, tuning every 20 ms.

Modern tuners use MIPI RFFE 3.0 with 19-bit addressing. A single USID controls 3 tuners; calibration tables map frequency + SAR sensor + grip state to capacitance. Without tuning, antenna efficiency at band edges drops 3–4 dB — equivalent to halving battery life at cell edge.

For forensic context, the antenna path also leaves traces. Investigators doing mobile forensics data extraction often check RFFE non-volatile logs (MIPI history buffers, PA temperature flags) to confirm radio state during an incident, though user data remains in the AP domain.

5. Integration: PAMiD, L-PAMiD and the mmWave Split

Integration is relentless. 2018 phones had 4 discrete PAs; 2026 flagships have 1–2 highly integrated modules:

• PAMiD (PA + Duplexer): For low/mid-band FDD. Example: Qorvo QM77048 — 3 PA dies + 10 duplexers + SOI switch in 6.0 × 5.3 mm.
• L-PAMiD / L-PAMiF: Adds low-noise amplifier (LNA) for the Rx chain. Saves 30 mm² and 0.6 dB trace loss versus discrete LBAR.
• MMMB PAM: Multi-mode multi-band PA covering 11 4G + 7 5G bands with one ET power supply.

mmWave (n257/n258/n260) breaks the model. At 28 GHz, PCB loss is ~1.8 dB/cm. So vendors move the entire front-end into an AiP (Antenna-in-Package) — Qualcomm QTM565 — beamformer + 4×4 patch array in an 8 × 6 mm package on the phone edge. The IF (4–6 GHz) travels over coax or waveguide to the transceiver. No acoustic filter can work at 28 GHz; instead, SIW/cavity filters on the package handle 800 MHz channels.

Trade-off: AiP reduces conducted loss but adds EIRP calibration complexity. Each module is factory trimmed for phase and gain across 64 beam states, stored in eFuse and loaded via RFFE at boot — similar to how GPU tables are calibrated in driver builds.

6. PCB, Coexistence and Thermal Reality

The RFFE is not just silicon. The PCB stackup makes or breaks it.

• Stackup: 8–10 layer HDI, 35 µm Cu on L1/L2 for PA power, ground cutouts under BAW to hit Q. Keep Tx traces 50 Ω ±7% — a 0.2 mm width error at 3.5 GHz adds 0.25 dB mismatch.
• Isolation: Tx leakage into GNSS (-130 dBm sensitivity) requires >90 dB isolation. That means via fences, separate LDOs, and a SAW notch on the GPS feed. We measure 8–10 mA desense if the shield can is not soldered continuously.
• Thermal: A PA at 42% PAE delivers 400 mW RF but dissipates ~550 mW heat. PAMiD junction hits 105°C in 30 seconds at +26 dBm. Phones throttle HPUE after 8–10 s continuous, falling back to power class 3 (+23 dBm). Vapor chamber placement matters more than the PA itself.

Test tip: During validation, sweep VCC vs Pout and capture ACLR vs temperature. A PA that passes at 25°C often fails ACLR at 55°C by 2 dB due to gain compression drift. Add 0.3 dB headroom or enable DPD table switching by temp.

Table 1 — Acoustic and switch technologies in today’s sub-6 GHz vs mmWave front-ends. Values measured at 25°C, 50Ω.

In practice, a BOM-optimized phone uses TC-SAW for LB, BAW for HB/UHB, and a single SOI switch matrix for both. A performance-first flagship goes all-BAW in UHB + dedicated BAW duplexers for n1/n3 to squeeze 0.7 dB better sensitivity — roughly 12% more range in weak signal.

Closing insight: The RFFE is now the gating factor for battery life in 5G standalone. When reviewers quote modem efficiency, 60% of the system loss is after the modem — in the PA supply, the filter ladder, and the antenna you are holding. Optimize the chain, not the die.