Abstract: High power radios rarely fail because they can’t make watts; they fail because they can’t make clean watts. This article unpacks the EVM ACLR trade-off, why it tightens as bandwidths and modulation orders increase, and which design levers (back-off, DPD, CFR, filtering, and calibration) actually move the needle in real transmitters. It closes with a practical workflow and where Novocomms Space helps teams land compliance without sacrificing efficiency.
If you’re architecting a broadband transmitter, EVM ACLR is the argument you’ll have with your power amplifier on every build: the modulation wants linearity, the thermal model wants efficiency, and the spectrum mask wants you to behave. The uncomfortable truth is that you can often “fix” EVM or ACLR in isolation—only to discover you’ve quietly broken the other, or burned half your DC budget doing it.
In high power radios—whether 5G/NR base stations, wideband point-to-point links, or satcom uplinks—this trade-off becomes the central system constraint because the PA is being asked to amplify signals with high crest factor, wide instantaneous bandwidth, and tight adjacent-channel emission limits. The engineering task is less about chasing a single golden metric and more about choosing the cheapest way (in watts, size, cost, and risk) to satisfy both.
EVM ACLR: what you’re really measuring (and why they fight)
Error Vector Magnitude (EVM) tells you how far the transmitted constellation points deviate from ideal. It’s a modulation fidelity metric, so it’s sensitive to AM/AM and AM/PM distortion, phase noise, IQ imbalance, memory effects, and even group delay ripple in the transmit chain.
Adjacent Channel Leakage Ratio (ACLR) tells you how much energy leaks into adjacent channels due to spectral regrowth, usually driven by non-linearities and imperfect filtering. ACLR is a “be a good neighbour” metric—your receiver performance might look fine, but your neighbours’ won’t if ACLR is poor.
They fight because the mitigation techniques don’t align perfectly. For example:
• Backing off the PA typically improves both EVM and ACLR, but efficiency collapses—painful in high power, thermally constrained radios.
• Tight RF filtering can make ACLR look brilliant while leaving EVM unchanged (or slightly worse if group delay is ugly). It also adds insertion loss, which forces more PA output power for the same EIRP—again hurting efficiency.
• Aggressive crest factor reduction (CFR) can improve average power capability and sometimes ACLR, but may degrade EVM if clipping noise isn’t carefully shaped and aligned with the modem’s tolerance.
A recent industry reminder of how formal this has become: the draft ETSI EN 301 908-18 V17.1.0 (2024-11) specifies NR base station ACLR test requirements and includes ACLR limits (for certain adjacent-channel definitions) called out at 44.2 dB and 49.2 dB depending on configuration. Whatever your exact product category, regulators and network operators increasingly treat adjacent-channel emissions as a first-order product risk, not a “we’ll tune it later” item.
Why high power makes the EVM/ACLR problem harder
At low to medium powers, you can sometimes buy your way out of trouble: a more linear driver, a higher-grade LO, a slightly better filter. At high power, the transmitter becomes a coupled mechanical–thermal–RF system.
1) Crest factor meets compression. Modern OFDM waveforms (NR, DVB-S2X, etc.) demand headroom. Pushing average output power close to compression is exactly how you create spectral regrowth (ACLR hit) and constellation warping (EVM hit).
2) Wide bandwidth exposes memory effects. As instantaneous bandwidth increases, the PA’s electrical and thermal memory becomes visible: bias networks, matching networks, traps, and package parasitics start “remembering” the waveform. That degrades EVM and creates asymmetric regrowth that is difficult to filter away.
3) GaN is powerful, but not magical. GaN devices make high power density practical, but the linearity/efficiency trade doesn’t disappear; it just shifts. A useful benchmark from recent Ka-band work: a 24–28.5 GHz GaN/SiC PA report shows operation with wideband CP-OFDM signals where measured ACLR can be driven below roughly -26.5 dBc with EVM values reported around the mid-to-high 20s dB (reported as dB metrics in the paper), but at the cost of average output power and efficiency. In other words: you can achieve cleanliness, but you will pay for it somewhere—usually in back-off and DC.
Design levers that actually move EVM ACLR in high power transmitters
There are only so many knobs that matter. The trick is knowing which one is cheapest for your architecture.
1) Output back-off: the blunt instrument (and still the baseline)
Back-off reduces non-linear distortion, so it usually improves both EVM and ACLR. But it is brutal on efficiency, especially with high PAPR signals. If your system is power-limited (battery, solar, payload DC budget) or heat-limited (sealed outdoor radio, space payload), you quickly reach a point where extra back-off costs more than adding complexity elsewhere.
Practical approach: treat back-off as a budgeted resource. Decide how many dB you can afford thermally, then use other techniques to close the remaining EVM/ACLR gap.
2) Digital predistortion (DPD): the scalpel—with a calibration bill
DPD is often the most effective way to improve ACLR at a given average output power. It can also improve EVM, but only if your observation receiver, alignment, and model are good enough. Recent RAN4 discussions around FR2 multi-band base stations highlight what many teams are learning the hard way: as you move to wider bandwidths and multi-band operation, DPD must cope with stronger memory effects and cross-band interactions, not just static AM/AM correction.
DPD also shifts complexity into manufacturing and field support: you need repeatable calibration, stable temperature behaviour, and a plan for drift over life. If you don’t engineer the calibration flow early, DPD becomes a late-stage schedule hazard.
3) Crest factor reduction (CFR): helpful, but mind the EVM floor
CFR can let you run the PA closer to saturation for the same EVM—if done properly. Done badly, it raises the in-band noise floor and trashes EVM while giving you an ACLR graph that looks acceptable. The best results come when CFR and DPD are co-designed rather than bolted together, with clear limits on allowable clipping noise relative to the modulation and coding scheme (MCS).
4) RF filtering and frequency plan: the underappreciated ACLR lever
Filtering is still the most deterministic way to reduce out-of-band emissions. But it’s not free: insertion loss demands more PA power for the same EIRP, and group delay ripple can degrade EVM, particularly on wider bandwidths. A clean frequency plan (LO placement, spur management, fractional-N choices) is equally important; phase noise and spurs don’t always show up as “non-linearity”, but they absolutely show up in EVM and adjacent channel measurements.
How to translate requirements into a buildable spec (without wishful thinking)
Product requirements usually arrive as a bundle: target modulation order, throughput, channel bandwidth, maximum output power, and a regulatory/standards envelope. A sensible workflow is to convert those into two linked budgets: an EVM budget and an ACLR (and SEM) budget.
One practical insight echoed in recent small-cell design literature (for example, Richardson RFPD’s 2024 guidance) is that EVM and ACLR are not niche lab metrics—they directly map to modulation order capability and interference compliance. Treat them as system-level KPIs from day one, not PA-only parameters.
In practice:
• Allocate EVM contributors (baseband quantisation, DAC SFDR, LO phase noise, IQ imbalance, PA AM/PM, group delay, clocking, and calibration residuals).
• Allocate ACLR contributors (PA non-linearity, DPD residuals, CFR noise shaping, filter attenuation, and any leakage paths).
• Decide what is “tunable” (DPD coefficients, adaptive bias, temperature compensation) versus what must be “designed out” (spur plan, filtering, layout coupling).
Where Novocomms Space fits: making clean watts repeatable
Getting one golden prototype to pass EVM and ACLR is not the finish line. The hard part is making it pass across temperature, across device variation, and across production volumes.
At Novocomms Space, we work with teams building broadband transmitters and payload RF hardware where power, mass, thermal headroom, and compliance all collide. Typical engagement points include:
• High power RF and microwave amplifier design (including GaN-based architectures), with attention to linearity, stability, and thermal behaviour.
• Frequency conversion and RF front-end engineering where LO plan, filtering, and isolation are treated as first-class citizens in the EVM ACLR story.
• Design-for-manufacture and test strategy so that linearisation techniques (including calibration hooks and observation paths where applicable) are feasible at scale, not just in a lab setup.
Whether you’re building a satcom terminal uplink chain, a wideband point-to-point radio, or a high power transmitter module that must hold performance over life, the same rule applies: you can’t spreadsheet your way out of non-linear physics, but you can choose the architecture that makes compliance the cheapest.
Conclusion: treat EVM and ACLR as a coupled design problem
The fastest route to a clean, efficient high power radio is to stop treating EVM and ACLR as separate late-stage acceptance tests. They’re coupled symptoms of the same underlying behaviours: non-linearity, memory, noise, and imperfect filtering. Back-off, DPD, CFR, and filtering each improve the picture—but each comes with a cost in watts, complexity, or manufacturability.
If you’d like to review an EVM/ACLR budget, sanity-check a linearisation strategy, or de-risk a high power transmit chain before it becomes an expensive thermal experiment, speak to Novocomms Space. Contact us here: https://novocomms.space/contact-us/
