TELCOMA Certified AI-RAN Specialist

What's assessed

6 domains, 42 questions in total — drawn from the course lessons.

• Why AI in the RAN

  7 questions

  The radio's real tension and the frames for reasoning about it: beams, channel, scheduling and position; where the RAN computes across RU, DU and CU; and AI-RAN as three strands in one discipline.

• ML mechanics for RAN engineers

  7 questions

  Exactly the mechanics the rest of the course spends, taught on radio data: a model as a fitted function, the channel as a time-series, one-sided versus two-sided models, life-cycle management, and the fallback floor.

• AI on the air interface — the 3GPP strand

  7 questions

  The 3GPP strand with its provenance: Rel-19 making one-sided models normative, beam and CSI prediction, AI positioning, life-cycle signalling to activate, monitor and fall back, and why two-sided CSI compression waited.

• RIC-hosted intelligence — the O-RAN strand

  7 questions

  Near-real-time and non-real-time as two clocks with real loop budgets, the E2 and A1 interfaces, where models are trained and shipped, rApps in production, and one energy rApp traced end to end.

• AI-RAN infrastructure — the shared-compute strand

  7 questions

  The shared-compute promise kept honest: AI-for-RAN, AI-on-RAN and AI-and-RAN; running L1 on GPUs; what has actually run versus been announced; the purpose-built silicon counter-argument; and whether sharing the machine pays.

• Energy savings & operating models in production

  7 questions

  The network-side track that went normative a release earlier: AI/ML for NG-RAN, cell sleep by level through DTX and DRX, energy savings quoted with their denominators, and where the loop actually runs.