Dr. Khan of Ericsson, Canada

Cell-edge rate is how next-generation networks get judged, and none of the utilities we actually optimize target it. Percentile programs do, and they are non-convex, non-smooth and strongly NP-hard. This talk follows that one problem across five papers, and at each step there is less of the algorithm left for a person to design.

It begins with sum-least-q-th-percentile rate maximization, where two fractional transforms buy block-concavity and cyclic algorithms orders of magnitude faster than SCA or SQP — in single-antenna, MU-MIMO, and now cell-free massive MIMO settings.

Then the solver goes away. A neural network trained directly on the non-smooth utility, differentiating through the sort itself, matches the optimum where one is known and runs three to four orders of magnitude faster.

Then the designer goes away too. Given an immutable evaluator and a page of instructions, an AI agent ran eighty-one experiments over twenty-six hours and returned a model within half a percent of our best solver at roughly 600× lower inference cost, with a small theorem attached.

And then the benchmarks are replaced by AI-designed ones too! Closed-form FP and WMMSE have been the default for the better part of two decades. They, and the two 2026 improvements on them, all maximize a surrogate that is linear in the interference power — the structure that decouples the links and buys closed form. That family has a unique tightest member, and an AI agent helped find it: the interference-conjugate transform, whose multiplier is the classical interference price, forced rather than chosen. Same cost per iteration, nine iterations against 1160 for FP and WMMSE in eight-antenna beamforming, the first solver-free percentile algorithm, and a proof that no interference-linear algorithm can do better.

The benchmarks that stood for two decades were ours. The ones that retire them are not. I will close on what that leaves for us to do.

Biography:

Ahmad Khan (M’IEEE) received the Ph.D. and M.A.Sc. degrees from the Department of Electrical and Computer Engineering at the University of Toronto, Canada, in 2023 and 2017, respectively, and the B.S. degree in Electrical Engineering with First Class Honours from King Fahd University of Petroleum & Minerals (KFUPM), Dhahran, Saudi Arabia, in 2015. During his graduate studies, he was awarded the Edward S. Rogers Fellowship, the Wallberg Memorial Fellowship, the Ontario Graduate Scholarship, and the Queen Elizabeth II Graduate Scholarship in Science and Technology.

In 2021, he joined Ericsson Ottawa as a Machine Learning Engineer, and since 2023 has held a dual role there as Project Manager and Research Scientist in AI RAN R&D. In 2024, he was a visiting researcher at Ericsson Research in Stockholm, Sweden, and is an alumnus of the company’s prestigious Early Career Program. He also holds a patent in wireless resource management.

His research interests include wireless resource management — specifically beamforming and scheduling — and agentic AI/machine learning for next-generation wireless networks. He has published in IEEE Transactions on Signal Processing, including a two-part paper, IEEE Transactions on Communications, IEEE Transactions on Wireless Communications, IEEE Signal Processing Letters, IEEE GLOBECOM, and IEEE ICASSP, among others.

Email:

Address:Canada