Abstract: High-power broadband transmitters live or die on two numbers that pull in opposite directions: EVM and ACLR. This post explains why the EVM ACLR trade-off tightens as you chase efficiency, how PA nonlinearity, memory effects, phase noise and filtering interact, and what to prioritise in architecture and test. We close with practical design tactics and where Novocomms Space can help de-risk delivery.
When a high power radio misses its spectral mask or mod quality target, the root cause is rarely “the PA is non-linear” in isolation. It’s the system-level balancing act between EVM ACLR: the in-band cleanliness required to demodulate dense constellations, and the out-of-band cleanliness required to co-exist with neighbouring channels and regulators. Push for more watts or more efficiency and you typically pay in spectral regrowth (ACLR) and constellation smearing (EVM). Push for more linearity and you often pay in DC power, thermal headroom, size, and cost.
This tension is getting sharper. 5G NR base-station requirements (TS 38.104) have kept EVM expectations tight (commonly referenced values include 3.5% for 256-QAM and 2.5% for 1024-QAM), while wideband carriers and multi-carrier operation make adjacent-channel compliance harder. Meanwhile, satcom waveforms (including DVB-S2X deployments) continue to demand aggressive throughput, forcing gateways and terminals to decide how close they dare run HPAs to compression without creating adjacent transponder interference.
Why EVM ACLR becomes the bottleneck in high-power radios
EVM measures how far the transmitted I/Q points land from their ideal locations after the full RF chain. It is sensitive to:
- Nonlinear AM/AM and AM/PM distortion (PA compression and phase rotation)
- Memory effects (thermal, bias network, trapping effects—especially in wideband GaN)
- LO phase noise, spurs, and vibration-induced modulation in harsh environments
- I/Q imbalance, DC offsets, sampling clock issues, and crest factor management
ACLR (Adjacent Channel Leakage Ratio) captures how much power leaks into adjacent channels. 3GPP defines it as a ratio of filtered mean power in the assigned channel to filtered mean power in an adjacent channel (typically using square/defined measurement filters). In practice, ACLR is dominated by spectral regrowth from nonlinearities and can worsen dramatically with:
- Higher PAPR waveforms (OFDM, wide channel bandwidths)
- Multi-carrier aggregation (including cumulative ACLR considerations)
- Operating closer to saturation
- Insufficient post-PA filtering or duplexer constraints
The uncomfortable truth for system architects: you can “fix” ACLR while making EVM worse (e.g., heavy filtering plus group delay ripple), and you can “fix” EVM while making ACLR worse (e.g., reducing equalisation/DPD aggressiveness to avoid noise enhancement).
EVM ACLR in standards: what’s actually tightening?
Two recent, practical pressures are shaping the design space:
- Higher-order modulation targets are unforgiving. Publicly cited interpretations of 5G NR specs commonly point to 3.5% EVM for 256-QAM and as low as 2.5% for 1024-QAM. That’s before you’ve added real-world impairments like temperature drift, antenna mismatch, and mechanical vibration in deployed systems.
- ACLR is no longer a single-number tick-box. Modern deployments care about ACLR across bandwidth, across carriers, and sometimes across sub-block gaps. Multi-carrier and non-contiguous spectrum bring “cumulative” behaviours to the forefront, and what looked fine on a single CW-like test can fail once the waveform is truly representative.
For satcom, DVB-S2X implementation guidance has long highlighted an operational reality: the HPA operating point must be optimised with adjacent-channel interference in mind, not just link budget. The “extra dB” you squeeze out on the carrier can become self-defeating if it raises interference into neighbouring carriers/transponders enough to force backing off elsewhere.
Where the trade-off really comes from: PA physics, memory, and PAPR
At high power, the PA is the main act. GaN devices give you excellent power density and bandwidth, but you still contend with:
- Compression knee behaviour: as you approach P1dB/saturation, odd-order intermods grow quickly, hitting ACLR hard.
- AM/PM conversion: phase distortion directly damages EVM, especially for dense QAM.
- Memory effects: wideband signals (tens to hundreds of MHz) expose bias-network dynamics, thermal time constants, and trapping. These make “static” linearisation models underperform.
- PAPR reality: OFDM and multi-carrier signals spend most of their time well below peak, but peaks force you to choose: clip, back-off, or linearise. All three have consequences.
Engineers often describe this as “choose two of: efficiency, EVM, ACLR”. It’s not quite that simple, but it’s close enough to be useful in early architecture decisions.
Engineering the EVM ACLR trade-off: back-off, filtering, and DPD
There are three primary levers, and each has a signature failure mode:
1) Output back-off (OBO)
Backing off reduces nonlinearity, improving both EVM and ACLR. The price is DC power and thermal design. In high duty-cycle broadband transmitters, OBO quickly becomes a mechanical problem: heatsinking, airflow, or conduction paths.
2) Post-PA filtering
Filtering can clean up ACLR, but it won’t undo in-band distortion. Worse, real filters introduce group delay variation and amplitude ripple across bandwidth, which can inflate EVM unless you equalise and maintain calibration over temperature and production spread. In multi-band or compact platforms, cavity size, insertion loss, and power handling can become the limiting factor.
3) Digital predistortion (DPD)
DPD is the workhorse approach because it can give you “apparent back-off” without giving away as much efficiency. Recent published GaN PA + DPD results show what’s possible in practice: ACLR improvements on the order of ~15–17 dB, with reported adjacent leakage levels reaching roughly −50 dB in some setups when the model and training are well matched to the hardware and waveform. That’s the difference between “almost compliant” and “comfortably compliant” on many broadband masks.
The catch: DPD is a system feature, not a software feature. It pulls in converter linearity, crest factor reduction strategy, observation receiver design, FPGA/DSP throughput, and calibration stability. And it can raise the in-band noise floor if implemented without proper attention to quantisation, bandwidth, and loop dynamics—hurting EVM just as you celebrate ACLR.
Measurement pitfalls: when you think you’re passing, but you’re not
Most EVM/ACLR surprises arrive at verification because the lab setup is not representative. A few repeat offenders:
- Wrong waveform statistics: your “test OFDM” has different PAPR or resource mapping than the deployed signal, so the PA is effectively being exercised differently.
- Insufficient oversampling / capture bandwidth: satcom guidance explicitly warns that modelling/representing nonlinear behaviour needs high sampling rates to avoid aliasing. The same is true in test: if your capture chain truncates regrowth, ACLR looks artificially good.
- Phase noise and environment: industry presentations have demonstrated how vibration and LO phase noise can dominate EVM even when the PA is well linearised. If your product moves (airborne, maritime, vehicle-mounted), you must budget EVM for mechanics as well as electronics.
- Thermal drift: DPD coefficients that look brilliant at 25 °C can drift at +70 °C junction, and production variance means “one golden unit” proves very little.
A practical rule: validate EVM and ACLR across power, bandwidth, temperature, and mismatch (VSWR). If you don’t, your compliance margin is imaginary.
How Novocomms Space approaches high-power broadband transmitters
At Novocomms Space, we tend to start with the uncomfortable questions early:
- What is the real operational waveform (bandwidth, PAPR, duty cycle, multi-carrier plan)?
- Is the compliance risk driven by EVM, ACLR, or both—and at what temperature and power?
- Do we need efficiency (battery, platform power limit) or absolute linearity (dense spectrum, high-order modulation) more?
From there, we support programmes with a blend of RF power amplifier design (including GaN), wideband RF front-end engineering, and test and manufacturability thinking so the performance you see on the bench survives volume build and field conditions. Use-cases we commonly see include satcom uplinks and gateways, defence broadband radios, and high-throughput data links where both spectral compliance and delivered EIRP matter.
If the project calls for linearisation, we help frame the entire DPD story—PA characterisation, observation path requirements, calibration strategy, and what “good enough” looks like when the platform is hot, moving, and mismatched.
Conclusion: design for margin, not miracles
The EVM ACLR trade-off in high power radios is not a single knob; it’s a coupled system of PA physics, waveform statistics, linearisation, filtering, and measurement discipline. The winning architectures are the ones that:
- choose an operating point with real thermal and efficiency constraints in mind,
- use DPD and filtering deliberately (not as last-minute patches), and
- prove performance across the full operating envelope, not just one lab condition.
If you’re architecting a broadband transmitter and want a second set of eyes on the EVM/ACLR budget, PA line-up, linearisation plan, or verification approach, speak to the team at Novocomms Space. Contact us here: https://novocomms.space/contact-us/.
