Why modern radios modulate the PA supply rail to conserve battery — and what it takes to do it well.
The RF power amplifier (PA) is the single largest power consumer in a wireless transmitter, and in a smartphone it is one of the top consumers of battery energy overall. Modern high-throughput waveforms (OFDM in LTE/5G-NR and Wi-Fi, CDMA) have high peak-to-average power ratios (PAPR), forcing a conventionally supplied PA to operate many dB backed off from its efficiency sweet spot. The energy gap between the fixed supply rail and the instantaneous signal envelope is dissipated as heat.
Envelope tracking (ET) closes this gap by dynamically modulating the PA supply voltage to follow the envelope of the RF signal, keeping the transistor near compression — and near peak efficiency — at every instant. This document explains the motivation, the system architecture (envelope detection, shaping tables, the envelope-tracking power supply, and time alignment), quantifies the battery savings, compares ET against average power tracking (APT) and other efficiency-enhancement techniques, and discusses the practical engineering challenges: modulator bandwidth, alignment sensitivity, memory effects, and the interplay with digital predistortion (DPD).
- The Problem: PAs, PAPR, and Wasted Battery
- PA Efficiency Fundamentals
- The Envelope Tracking Concept
- ET System Architecture
- Envelope Shaping Tables
- ET vs. APT vs. Fixed Supply
- Time Alignment: The Make-or-Break Requirement
- Quantifying the Battery Savings
- Design Challenges and Trade-offs
- ET and Digital Predistortion
- ET in Modern Standards: LTE, 5G NR, Wi-Fi
- Comparison with Other Efficiency Techniques
- Simulation: Reproducing the Delay-Sensitivity and Efficiency Results
- Conclusion
- References
In a conventional transmitter, the PA is fed from a fixed DC supply, typically the battery rail or a fixed regulator output:
This design served early cellular systems well. GSM/GPRS use GMSK, a constant-envelope modulation: the RF signal's amplitude never changes, only its phase. The PA can therefore be biased hard into saturation (class C/E-like operation) and run at peak efficiency continuously — there is no envelope to track.
The pursuit of spectral efficiency broke this comfortable arrangement. OFDM (LTE, 5G NR, Wi-Fi) and CDMA waveforms carry information in both amplitude and phase, so the signal envelope varies continuously and dramatically:
| Waveform | Typical PAPR |
|---|---|
| GSM (GMSK) | 0 dB (constant envelope) |
| WCDMA uplink | 3–4 dB |
| LTE uplink (SC-FDMA) | 6–7 dB |
| LTE downlink / Wi-Fi OFDM | 8–12 dB |
| 5G NR CP-OFDM uplink | 9–12 dB |
With a fixed supply sized for the signal peaks, the PA spends nearly all of its time transmitting well below those peaks. The voltage headroom between the envelope and the rail is dropped across the transistor and dissipated as heat:
The consequences are concrete and user-visible:
- Battery drain — the PA can consume 3–4× more DC power than the RF power it actually delivers.
- Thermal load — hotter phones, thermal throttling, tighter industrial-design constraints.
- Operating cost at scale — for base stations and broadcast networks, wasted PA energy is a direct, recurring electricity bill.
Drain (or collector) efficiency is defined as:
For a classical linear PA (class B is the convenient idealization), efficiency falls with the square root of output power as the signal backs off from the supply-limited peak:
An ideal class-B PA peaks at η = π/4 ≈ 78.5 % (real PAs: 50–65 %). At 8.5 dB backoff — the average power of a typical OFDM signal — the ideal class-B efficiency has already dropped to roughly 37 % of its peak value, and real-world numbers of 20–30 % are typical for a fixed-supply LTE handset PA.
The key insight behind ET: the PA is only inefficient because its supply is too high for the instantaneous signal level. If the supply itself scaled with the envelope, the PA would sit near its compression point — and near peak efficiency — at every envelope level:
The dashed red curve is the PA alone under ET; the green curve includes a realistic ~85 % efficient supply modulator. Even after paying the modulator tax, ET roughly doubles system efficiency at typical OFDM average power.
The basic principle of envelope tracking is to constantly adjust the supply voltage of the PA according to the envelope of the RF input signal, so the PA always operates close to saturation, where it is most efficient [1].
Compare Figures 2 and 3: the orange "wasted headroom" region collapses to a thin sliver. Instead of burning the peak-to-envelope difference as heat inside the RF transistor, the system simply never delivers that energy to the PA in the first place — the supply modulator (a switching converter at its core) only sources what the instant requires.
Two properties make this more than a clever trick:
- The savings scale with PAPR. The higher the waveform's PAPR, the more headroom a fixed rail wastes — and the more ET recovers. ET is most valuable exactly for the modern waveforms that hurt the most.
- The PA stays in compression. A side effect is that the PA's gain becomes a function of the supply trajectory. This is both a challenge (Section 9) and an opportunity: with a well-designed shaping table, ET can actually flatten gain (iso-gain operation) and simplify linearization.
The signal chain, following the Keysight reference concept [1]:
The envelope is generated digitally from the baseband I/Q samples:
Doing this in baseband (rather than with an analog RF envelope detector) gives an exact, noise-free envelope and — critically — makes the envelope path deterministic, so it can be delay-matched against the RF path to sub-nanosecond precision.
The raw magnitude is not the ideal supply waveform. It is passed through a shaping table — a nonlinear mapping from instantaneous magnitude to required supply voltage:
The shaping table is where most of the system-level engineering lives; it determines the efficiency/linearity trade of the entire ET system, and designers spend significant effort optimizing it [1]. Section 5 covers this in depth.
The shaped envelope drives the ETPS, also called the supply modulator or ET modulator. This is a wideband, high-efficiency power converter — commonly a hybrid of:
- a switching buck converter that efficiently delivers the bulk, low-frequency energy, and
- a linear amplifier stage in parallel that supplies the fast, high-frequency corrections the switcher can't follow.
The hybrid achieves 80–90 % modulator efficiency while tracking envelope bandwidths of tens of MHz. The ETPS must source (and sometimes sink) amperes of current with a bandwidth of roughly 3–5× the RF channel bandwidth, because the magnitude operation
MediaTek's white paper [2] gives a useful view of how this architecture looks in shipping cellular silicon. Their ET digital baseband inserts a power scaling block between the envelope detector and the LUT: the envelope is first scaled according to the target antenna output power, and only then mapped through the Vpa look-up table to a supply voltage. This matters in practice because the network commands the handset's transmit power dynamically, so the magnitude-to-voltage mapping must be power-level aware, not a single static curve. The RF path in their diagram likewise carries an RF-compensation stage, with both paths upsampled and converted through separate DACs before meeting at the supply modulator and PA. Their platforms support external supply modulators driving either GaAs or CMOS PAs [2].
MediaTek also identifies the two objectives of ET factory calibration: configuring the Vpa LUT that characterizes the integrated amplifier against pre-defined design criteria, and detecting the delay difference between the ET path and the TX path so timing alignment is guaranteed — with sensitivity to timing mismatch growing as TX bandwidth widens [2]. Beyond the factory, their system performs self-recalibration during active transmission to track load mismatch, temperature shifts, VSWR changes, and device aging [2] — a production answer to the drift problems discussed in Section 9.
The RF I/Q samples are delayed to match the group delay of the envelope path (shaping, DAC, ETPS, board routing), then converted and upconverted as usual. Both paths converge at the PA: RF at the input, modulated
Why not just drive the supply with the raw scaled magnitude? Three reasons:
-
The PA must stay alive at zero-crossings. The envelope of an OFDM signal routinely touches zero. Dropping
$V_{CC}$ to 0 V would cut the PA's bias, collapse its gain, and create severe distortion. The table therefore enforces a minimum voltage floor$V_{min}$ (typically 0.5–1 V for a handset PA). - Bandwidth containment. Sharp troughs in the envelope are the highest-bandwidth features. "Detroughing" — smoothly rounding the bottom of the envelope — dramatically reduces the bandwidth the ETPS must track, at a tiny efficiency cost (the troughs carry little energy anyway).
- Linearity shaping. The supply trajectory sets the PA's instantaneous gain. A table can be tuned so that gain is constant across the envelope range — iso-gain shaping — making the composite ET-PA look nearly linear before DPD is even applied.
A classic detroughing function:
Shaping table design is empirical in practice: the PA is characterized across a
Average Power Tracking (APT) is ET's simpler sibling and an important stepping stone. Instead of following the instantaneous envelope, APT adjusts the supply rail per transmit slot (e.g., every 0.5 ms in LTE) to match the average output power commanded by the network. A plain buck converter suffices — no wideband modulator needed.
| Aspect | Fixed supply | APT | ET |
|---|---|---|---|
| What supply follows | Nothing (constant) | Average power per slot | Instantaneous envelope |
| Supply bandwidth | DC | ~kHz | Tens of MHz (3–5× channel BW) |
| Hardware | LDO / battery rail | Buck converter | Hybrid switcher + linear amp |
| Efficiency at max power, high PAPR | Poor | Poor–moderate | Best |
| Efficiency at low output power | Very poor | Good | Good, but modulator overhead |
| Calibration burden | None | Small | Large (shaping + timing) |
| Typical use today | Legacy | Low/mid power levels, very wide BW | High power levels, moderate BW |
Real handset front-ends use both: the tracker IC runs APT mode at low output powers (where ET's modulator overhead isn't worth it) and switches to ET mode near maximum power, where the PA burns the most energy and ET's gains are largest.
For an ET PA, the supply waveform and the RF signal must arrive at the PA strictly time-aligned. Even small timing deviations substantially degrade the RF output, deteriorating both ACLR (adjacent channel leakage) and EVM [1].
The intuition: if the supply arrives late, the PA is momentarily starved — the envelope demands more voltage than the rail provides — and the output clips hard. If it arrives early, the PA is over-supplied at the wrong instants, wasting energy and modulating the gain incorrectly:
Rules of thumb from published ET characterization work:
- For a 20 MHz LTE channel, delay error must typically be held within ±1–2 ns to avoid measurable ACLR degradation; wider channels are proportionally more sensitive.
- Alignment is a standard factory calibration: sweep the envelope-path delay, measure ACLR/EVM, and program the delay that minimizes distortion. Instrument-grade ET test equipment (e.g., the Keysight N7614 Signal Studio family that the reference concept page belongs to) automates exactly this sweep.
- Delay drifts with temperature and aging, so production designs leave margin or re-trim — MediaTek's system, for instance, self-recalibrates during active transmission to track exactly these drifts [2].
Section 13 reproduces this sensitivity quantitatively with an included simulation: on a 20 MHz channel, the modeled ACLR crosses the 3GPP limit at 10–20 ns of misalignment.
A worked, illustrative example for a handset transmitting LTE at +24 dBm (0.25 W) average output — a typical near-max uplink operating point:
| Supply scheme | PA system efficiency | DC power drawn | Power wasted as heat |
|---|---|---|---|
| Fixed supply | ~25 % | 1.00 W | 0.75 W |
| APT | ~33 % | 0.76 W | 0.51 W |
| ET | ~45 % | 0.56 W | 0.31 W |
ET cuts the PA's battery draw by ~45 % versus a fixed rail at this operating point, and cuts heat dissipation by more than half. During a sustained uplink-heavy use case (video call, hotspot, poor coverage forcing high Tx power), the PA can account for a large fraction of total platform power — making this one of the highest-leverage energy optimizations in the entire phone. The same arithmetic, multiplied across thousands of radios, is why ET (and its cousins) matter for base-station OPEX as well.
These illustrative numbers are consistent with published silicon measurements. MediaTek's characterization of ET versus fixed-supply operation at +23 dBm across LTE bands 1, 3, 6, 8, and 41 (5–20 MHz channels, GaAs and CMOS PAs) found ET outperforming the fixed-supply configuration in both E-UTRA ACLR and overall efficiency in every band tested, with a maximum efficiency improvement of roughly 18.5 percentage points for an LTE 5 MHz signal in band 8 [2]. The commercial verdict came early: by 2015, ET had been adopted in flagship LTE smartphones from Apple, Samsung, LG, HTC, Sony, ZTE, and Amazon [2].
Secondary battery benefits compound the direct one:
- Lower thermal rise → less thermal throttling of Tx power and SoC, fewer "hot phone" duty-cycle penalties.
-
Reduced peak battery current → less
$I^2R$ loss in the battery and PMIC, and better voltage-droop margin at low state-of-charge.
ET's gains are real, but so is the engineering bill:
ETPS bandwidth and efficiency. The envelope's spectrum extends 3–5× beyond the channel bandwidth. Every doubling of tracked bandwidth pressures the modulator's efficiency and noise. This is the limiting factor for ET at 100–400 MHz 5G NR channel bandwidths — which is why very wide-band FR1 carriers and FR2 (mmWave) systems frequently fall back to APT.
Receive-band noise. The ETPS output rides directly on the PA drain; switcher ripple and linear-stage noise there translate into transmitted noise. FDD systems must keep ETPS noise in the paired receive band extremely low to avoid desensitizing their own receiver.
Memory effects. The supply network's finite impedance vs. frequency, thermal time constants, and trapping effects make the PA's behavior depend on envelope history, not just its instantaneous value. This complicates linearization (next section).
Calibration surface explosion. Shaping table, timing, and DPD coefficients vary per band, per channel bandwidth, per power level, and over temperature. Production calibration and non-volatile storage of this surface is a significant program cost.
Load sensitivity. Antenna VSWR changes the PA load line, shifting the optimal shaping table. Closed-loop schemes (feedback receivers, antenna tuners) mitigate this.
ET and DPD are partners, not alternatives. An ET PA held near compression is intrinsically nonlinear-ish, and the supply modulation itself introduces new distortion mechanisms (supply-dependent gain/phase, modulator artifacts, memory effects). Practical ET systems therefore run DPD around the composite ET-PA:
- Capture the PA output via a feedback receiver.
- Fit an inverse model — memory polynomial or generalized memory polynomial (GMP) are the workhorses — of the combined shaping-table + ETPS + PA response.
- Predistort the I/Q samples upstream of both the envelope and RF paths.
Two subtleties specific to ET:
- The DPD model must be trained with the shaping table and timing in their final state — change the table and the DPD solution moves.
- Iso-gain shaping and DPD trade against each other: a well-shaped ET system needs a milder DPD (fewer taps, less feedback bandwidth), which itself saves power. System-level co-optimization of {shaping table, delay, DPD} is where mature ET implementations differentiate.
LTE. ET reached commercial handsets in the LTE era (~2014 onward) and is now standard practice in premium and mid-tier phones for uplink at high power. LTE's ≤20 MHz channels sit comfortably inside modern tracker bandwidths.
5G NR. FR1 carriers up to 100 MHz stretch tracker technology; state-of-the-art ET ICs advertise 100–200 MHz tracking, and the industry has invested heavily because NR's higher PAPR and higher uplink duty cycles make the prize larger. Where tracking bandwidth runs out, designs revert to high-voltage APT, sometimes with fast slot-level transitions.
Wi-Fi. OFDM PAPR in Wi-Fi is as brutal as anywhere (11ax/11be with 1024/4096-QAM demand very high linearity). Interestingly, MediaTek's 2015 white paper predicted a greater ET benefit for 802.11ac than for 4G precisely because of Wi-Fi's higher PAPR, and noted ET's usefulness for the CMOS PAs common in Wi-Fi [2]. A decade on, that prediction has only partially materialized: the classic Wi-Fi answer has remained heavy backoff plus DPD rather than ET, because Wi-Fi channel bandwidths (80/160/320 MHz) exceed practical tracker bandwidths and the EVM floors required for 4K-QAM leave little tolerance for tracker noise. ET research for Wi-Fi continues, and APT-style supply optimization is common in Wi-Fi front-end modules.
Base stations. Macro base stations mostly use Doherty PAs (see below) for efficiency at backoff; ET appears in some small-cell and research contexts where Doherty's bandwidth limits bite.
| Technique | Principle | Strengths | Weaknesses |
|---|---|---|---|
| Envelope Tracking | Modulate supply with envelope | Large efficiency gain across wide power range; PA-technology agnostic | Modulator bandwidth/noise; timing calibration; complexity |
| APT | Adjust supply per slot to average power | Simple, robust, works at any bandwidth | No gain within a slot's envelope swing |
| Doherty | Carrier + peaking PA load modulation | Excellent at 6–8 dB backoff; no supply modulator; base-station standard | RF-bandwidth limited; layout-sensitive; less flexible across bands |
| EER (Kahn) | Amplitude via supply, phase via saturated PA | Theoretically highest efficiency | Extreme alignment/bandwidth demands; largely superseded by ET |
| Outphasing (LINC) | Two constant-envelope PAs + combining | Efficient saturated PAs | Combiner losses; bandwidth expansion |
| Backoff + DPD only | Linearize and accept backoff | Simple supply chain; best EVM floors | Poor efficiency for high PAPR |
ET's distinguishing feature is that it improves efficiency continuously across the envelope range rather than at one or two designed load points, and it does so around an otherwise conventional linear PA.
The claims in Sections 7 and 8 can be reproduced with the included simulation (scripts/et_simulation.py, NumPy + Matplotlib only). The model:
- Waveform — 64-QAM OFDM, 1200 occupied subcarriers (~18 MHz, LTE-20-like), 8× oversampled at 160 MS/s, with mild crest-factor reduction to 7.8 dB PAPR.
-
Supply — the detroughing shaping table of Section 5 (
$V_{min}/V_{max} = 0.3$ ). -
PA — a Rapp AM/AM model whose saturation amplitude is proportional to the instantaneous supply,
$A_{sat}(t) = k \cdot V_{CC}(t)$ , plus supply-coupled AM/PM (12° at compression) and a short memory branch — a minimal model that still captures the supply-dependence and memory that make ET linearization interesting. - Delay error — applied to the envelope path via FFT fractional delay, swept ±40 ns.
-
Metrics — E-UTRA-style ACLR integrated in ±20 MHz adjacent channels, waveform EVM against the best complex linear fit, and an ideal class-B drain-efficiency estimate (
$I_{DC} \propto |y|$ ,$P_{DC} = \overline{V_{CC} I_{DC}}$ ).
With the supply aligned, the ET output spectrum is essentially indistinguishable from the fixed-supply case; at +20 ns of envelope-path delay error, adjacent-channel regrowth is clearly visible.
The delay sweep reproduces the characteristic V-shaped sensitivity around zero delay. In this model, ACLR crosses the −33 dBc 3GPP E-UTRA limit between roughly 10 and 20 ns of misalignment on a 20 MHz channel:
| Delay error | ACLR (dBc) | Waveform EVM (%) |
|---|---|---|
| 0 ns | −37.5 | 5.3 |
| 5 ns | −36.6 | 5.7 |
| 10 ns | −33.6 | 7.5 |
| 20 ns | −28.5 | 12.2 |
| 40 ns | −23.0 | 21.4 |
Running the identical waveform through the identical PA model, drain efficiency improves from 38.6 % (fixed supply) to 64.9 % (aligned ET) — a 26-point gain from supply modulation alone. Caveats for honest reading: these are idealized class-B drain efficiencies excluding ETPS losses (apply the ~85 % modulator efficiency of Section 4 for a system number), the EVM figures are pre-DPD waveform EVM including the PA's static nonlinearity, and the absolute values depend on the chosen drive level (peaks just reaching saturation) and model coefficients. The relative behavior — flat efficiency gain, sharp delay sensitivity scaling with bandwidth — is the transferable result, and it matches both the Keysight guidance [1] and MediaTek's measured band data [2] in character.
Envelope tracking attacks the largest single source of wasted energy in a modern transmitter: the headroom between a fixed supply rail and a high-PAPR signal's instantaneous envelope. By deriving the envelope digitally as
The engineering price is paid in modulator bandwidth, receive-band noise management, nanosecond-class timing calibration, shaping-table optimization, and DPD co-design. As channel bandwidths grow through 5G-Advanced and Wi-Fi 8, the frontier of ET research is exactly there: trackers that stay efficient and quiet at hundreds of MHz, and calibration/adaptation schemes (increasingly ML-assisted) that keep the shaping table, timing, and DPD jointly optimal over temperature, aging, and antenna load.
- Keysight Technologies, "Envelope Tracking (ET) Concept," N7614 Signal Studio for Power Amplifier Test — Help. https://helpfiles.keysight.com/csg/n7614/Content/Main/Envelope%20Tracking%20Concept.htm
- MediaTek Inc., "Envelope Tracking Technology," White Paper, January 2015. https://www.mediatek.com/hubfs/MediaTek%20Assets/Pdfs/White_Papers/Envelope-Tracking.pdf
- Z. Wang, Envelope Tracking Power Amplifiers for Wireless Communications, Artech House, 2014.
- P. Asbeck and Z. Popović, "ET Comes of Age: Envelope Tracking for Higher-Efficiency Power Amplifiers," IEEE Microwave Magazine, vol. 17, no. 3, 2016.
- S. C. Cripps, RF Power Amplifiers for Wireless Communications, 2nd ed., Artech House, 2006.
- B. Kim et al., "Envelope Tracking Technique for Multimode PA Operation," Proc. European Microwave Conference, 2009.
- J. Hoversten et al., "Codesign of PA, Supply, and Signal Processing for Linear Supply-Modulated RF Transmitters," IEEE Trans. Microwave Theory and Techniques, vol. 60, no. 6, 2012.
.
├── README.md ← this paper
├── CITATION.cff ← machine-readable citation metadata
├── .gitignore
├── LICENSE ← MIT (code); text/figures are CC BY 4.0
├── figures/ ← all diagrams (PNG, 150 dpi)
└── scripts/
├── make_waveform_figures.py ← regenerates Figures 2, 3, 5, 6, 7, 8, 9
├── make_block_diagrams.py ← regenerates Figures 1, 4
└── et_simulation.py ← runs the Section 13 simulation (Figures 10–12)
All figures are generated programmatically with matplotlib/NumPy — clone and run the scripts to regenerate or restyle them. Figures 2–9 are conceptual illustrations (representative shapes and magnitudes); Figures 10–12 are simulated from the model in Section 13. Neither represents measurements of any specific device.
Prasuna is a seasoned Systems and Wireless Engineer specializing in Wi-Fi algorithms, RF systems, and embedded architecture. With extensive expertise spanning Wi-Fi 6, 7, and Next-Gen Wi-Fi 8 chipset development, Prasuna excels in RF front-end integration and physical-layer (PHY) system design, bridging hardware constraints with high-performance wireless algorithms, drives innovation in next-generation wireless connectivity.
- GitHub: @pchatla-arch
If you reference this work, please cite it as:
Prasuna Chatla, "Envelope Tracking for RF Power Amplifiers: Principles, System Design, and Efficiency Analysis," GitHub, 2026. https://github.com/pchatla-arch/Envelop-Tracking
BibTeX:
@misc{chatla2026envelopetracking,
author = {Chatla, Prasuna},
title = {Envelope Tracking for {RF} Power Amplifiers:
Principles, System Design, and Efficiency Analysis},
year = {2026},
howpublished = {\url{https://github.com/pchatla-arch/Envelop-Tracking}},
note = {Accessed: [date]}
}© 2026 Prasuna Chatla. Text and figures licensed under CC BY 4.0 — you may share and adapt with attribution. Simulation code licensed under the MIT License.