A digital predistortion framework developed in the group keeps a load-modulated balanced amplifier linear across every carrier configuration it is given, from a single trained model. The work was built for the 2025 IEEE International Microwave Symposium Student Design Competition 9 and is published in IEEE Microwave Magazine.
Load-modulated balanced amplifiers are efficient, but their behaviour shifts with the signal they carry, so a predistorter tuned for one carrier setting degrades on the next. The framework pairs a complexity-reduced Volterra model with a hybrid indirect–direct learning strategy, which converges quickly, and adds three signal-processing measures that matter when feedback bandwidth is limited: peak-to-average power ratio reduction, adaptive coefficient pruning and bandwidth-aware filtering of the basis functions.
The part that makes it practical is a configuration-robust training method: one model, trained once, holds its adjacent-channel power ratio and error-vector-magnitude improvements as the amplifier is driven with different carrier configurations, rather than needing a retrain for each. That suits adaptive transmitter architectures of the kind 5G and 6G equipment is moving towards.
The paper is by Mohammad Abdollah Chalaki, Emma Gu, Ahmed Ben Ayed, Patrick Mitran and Slim Boumaiza, published in IEEE Microwave Magazine, vol. 27, no. 6, pp. 113–121, June 2026. doi:10.1109/MMM.2026.3652159