AI in Telecom Optimization — from raw 3GPP counters to closed-loop automation. Six production playbooks, 40 hands-on labs with a real 5G core, animated wire diagrams, and zero hand-waving.
Written for RF engineers, NOC engineers and telecom data scientists who are done with slideware. Chapter 1 even teaches AI, machine learning and deep learning from zero — with a neuron you drive yourself — so you need no ML background to start.
Catch degradations hours before customers feel them. z-score → MAD → STL → Isolation Forest, alert budgets, CM-aware suppression — with an interactive threshold you drag yourself.
P90 quantile forecasts that feed a defensible capex list. Includes a live traffic synthesizer and the honest rolling-backtest protocol (plus the two ways teams cheat it by accident).
TA-histogram overshooter hunts, MDT grid maps, PCI mod-3/mod-4 graph re-coloring, and the guarded tilt loop with automatic rollback — the full contract, in code.
The fastest ROI in telecom AI. Coverage-safe sleep policies, wake triggers, PEE-metered savings per TS 28.552/28.554 — and the interactive savings estimator for your own fleet.
400 alarms → 1 incident. The compression ladder, topology grouping, lift mining that discovers hidden dependencies, and root ranking that survives broken clocks.
VSWR creep, twin-unit deltas, leakage-proof labels (drag the gap, watch fake AUC evaporate), precision@k dispatch lists and survival analysis.
Not exercises — working systems. Lab 2 generates a realistic 200-cell network with planted incidents and ground truth; every detector you build gets scored, not admired. By Module L5 you're running a genuine Open5GS 5G core in Docker, decoding real NGAP with Wireshark, closing an A1-style policy loop with automatic rollback, and building a cite-or-abstain 3GPP RAG copilot on a local LLM. Lab 40 wires it all into one make all.
Every standardized claim carries its reference: TS 28.552 counters, TS 23.288 NWDAF, TS 28.104/105 MDA & AI/ML lifecycle, TR 37.817 & TR 38.843 RAN AI, TS 28.100 autonomy levels, O-RAN WG2/WG3. Appendix A is a full atlas of all 29 cited specs. If we can't cite it, we don't claim it.
Yes — Chapter 1 is completely free, no signup: the scale problem, the full AI/ML/deep-learning foundations (with the interactive neuron), the four-layer AI map, the KPI tree and the TS 28.100 autonomy ladder.
A laptop with Python 3.10+ — pandas, scikit-learn, LightGBM (all free). No cloud, no GPU. Labs 33–35 additionally use Docker to run a real Open5GS 5G core + UERANSIM. Every lab has exact commands, expected output and a verify checklist.
That's exactly who Chapter 1 and 3 are written for: AI, ML and deep learning taught from zero through telecom examples, with interactive widgets instead of math walls. The Python in the playbooks is commented line by line.
Chapter 2 is your missing manual: PM counters, granularity periods, KPI formulas, CM/FM/MDT/NWDAF — the domain knowledge that kills most telecom-ML projects, with the data-quality gate that saves them.
It's a book you can search, skim and return to — with course-grade interactivity: steppers that replay call flows and training loops, draggable widgets, shuffled quizzes, and labs with ground-truth scoring. Reading time ~7 h; the lab track ~40 h of real building.
If the book isn't what you expected, email [email protected] within 7 days — we'll sort it out.
See the animated diagrams, drive the neuron, walk the closed loop — then decide.
📖 Read Chapter 1 free ⚡ $3.99 — Unlock everything