Track · 12 modules
RAN fundamentals for AI engineers
Everything a machine-learning engineer needs to be useful in a telecom optimisation team
For: ML / AI engineers entering telecom. Assumes Python + linear algebra; assumes nothing about cellular networks.
- 01
Why RAN data is different
ROP windows, multi-vendor schemas, why a typical ML pipeline fails on PM/CM data.
~12 min○ planned - 02
Cell identifiers: NCI, ECGI, ECI, NCGI
The four ways operators count cells and why your join keys keep returning empty.
~15 min○ planned - 03
PM counters — what they are, what they aren't
TS 32.401 ROP, granularity periods, why counters are sums and how they roll up.
~20 min○ planned - 04
CM and the MO hierarchy
gNB → CU/DU → cell → relation. Reading vendor RAML and ENM exports.
~20 min○ planned - 05
FM alarms and the X.733 model
Severity, perceivedSeverity, alarm lifecycle, why FM counts lie.
~15 min○ planned - 06
KPIs vs counters — the formula layer
How TS 28.554 KPIs are computed from TS 32.450/28.552 counters, with worked examples.
~25 min○ planned - 07
Multi-vendor reality
Nokia NRCELL vs Ericsson NRCellDU, why your detector fails on the second operator.
~20 min○ planned - 08
Time, timezones, and ROP alignment
Why every chart should be in UTC, granularity drift, ROP boundary edge cases.
~12 min○ planned - 09
Mobility and handovers
A1/A3/A5 events, MRO classification, why handover-attribution is hard.
~25 min○ planned - 10
Energy, sleep states, and SON
Cell sleep, capacity vs coverage tradeoff, ANR / MLB / MRO basics.
~20 min○ planned - 11
Lab environments and digital twins
Why you can't ML on prod data and how O-RAN NONRTRIC + arano-lab give you a sandbox.
~18 min○ planned - 12
What makes a good detector
Latency, false-positive budget, baseline drift, operator UX.
~22 min○ planned