AI-based digital predistortion (DPoD) has moved out of the seminar room and into working engineering laboratories, where teams now use it for 5G uplink testing and, in parallel, as a research instrument for 6G, a report by EDN, Voice of the Engineer, details.

The technique targets one of the oldest problems in radio engineering. Power amplifiers in user equipment and network gear are nonlinear; they distort signals, waste battery and spill energy into adjacent channels. Digital predistortion applies an inverse distortion before the amplifier so the combined chain comes out linear. Traditional DPD depends on hand-built mathematical models of amplifier behaviour. The AI variant learns that behaviour directly from measurement data.

That difference matters most on the uplink. Handset amplifiers operate at low power with tight energy budgets, and 5G uplink waveforms — with high peak-to-average power ratios from OFDM and wide channel bandwidths in mid-band spectrum — punish imperfect linearization. EDN's account places AI DPoD squarely in uplink test setups, where learning-based predistortion can track amplifier quirks that fixed polynomial models miss, including memory effects and device-to-device variation across temperature and ageing.

What does this change for test engineers?

In practical terms, it shifts part of the linearization burden from the design table to software. A lab that once characterized an amplifier exhaustively and then froze a DPD table can instead train a model against live measurements, iterate and retrain as silicon lots vary. For 5G uplink conformance and verification work, that means faster correction of adjacent-channel leakage and error-vector-magnitude problems late in the device cycle, when a hardware respin is no longer an option.

The same infrastructure serves the 6G side of the lab. Standards bodies, including 3GPP and regional research programmes, are still years from freezing 6G waveform decisions, and candidate waveforms are expected to be wider-bandwidth and more peaky than today's. Any of them will make amplifier linearity harder, not easier. Research groups therefore need linearization tools that do not presuppose a particular modulation scheme, and AI DPoD — being model-agnostic and data-driven — fits that requirement.

The regulatory context is thin but relevant. Spectrum regulators have begun consulting on spectrum above 7 GHz, notably in ITU discussions on bands for IMT-2030. Wider bands at higher frequencies push amplifiers toward less efficient operating points, which makes predistortion research a precondition for viable 6G handset economics rather than an optional refinement.

Vendor claims versus lab reality

The commercial pitch for AI-driven test and linearization tools follows a familiar arc: faster setup, better correction, less expert dependency. Lab practice, as described by engineers working with the technique, is more sober. Training data quality governs results; a model trained on one amplifier population does not transfer cleanly to another; and validation against 3GPP-defined metrics — ACLR, EVM, spectral emission masks — remains the binding constraint on what any AI approach can claim.

In other words, AI DPoD is a tool that compresses engineering time, not a method that removes the engineer. The distortion model gets smarter. The measurement discipline does not.

The forward path is legible from the work itself: teams now validating AI DPoD on 5G uplink chains are building the datasets and workflows they will carry into 6G prototyping as candidate bands and waveforms firm up over the coming standards cycle.