Value Packs / Telecom / Network Alarm & Fault Analytics / Wisdom

Network Alarm & Fault Analytics · Wisdom notebook

Ask the follow-up.

Agentic conversational BI · from the alarm storm to the band and sector

For the top sites by alarm volume this week, what is the category breakdown and how much of it is really one fault?

RK

De-duplicated alarms per element per window, clustered by signature, then compared raw counts against distinct faults for the ten busiest sites.

Volume and fault count are almost unrelated — the busiest site is not the worst problem:

Raw alarms by site — and how many faults they really are

New York · site 494 raw alarms · impact high · fallback 3.1×1 fault
Queens · site 271 raw alarms · impact high · fallback 2.8×1 fault
Staten Island · site 949 raw alarms · impact medium · fallback 1.9×1 fault
Chicago · site 1144 raw alarms · impact low · no fallback9 faults
Kansas City · site 637 raw alarms · impact none12 faults

Chicago and Kansas City have many small independent faults and no subscriber impact — the classic noise profile. The New York sites have one fault each and real impact.

Which of those actually correlate with customer complaints?

RK

Joined the alarm clusters to care contacts and 5G fallback on the same districts and windows, keeping only correlations that hold.

Three sites, and the correlation is tight enough to prioritise on:

005050100100normnormNew York · site 4Queens · site 2Staten Island · site 9Chicago · site 11Kansas City · site 65G fallback ratespeed complaints / 1k

The three New York sites sit together in the corner. The two noisy sites sit with the quiet ones, which is the point.

Is it whole sites, or one band? Break it by frequency band and sector.

RK

Split the 19 affected sites by frequency band and sector, and checked availability per carrier rather than per site, so a single-carrier fault separates from a site-wide one.

One band, and not every sector — n78 only, which is why the sites never went down and the fault stayed invisible to availability monitoring:

Availability by frequency band × sector · the 19 affected sites

Sector 1Sector 2Sector 3
n78 · 3.5 GHz (5G)61%58%94%
n1 · 2100 MHz (5G)99%99%100%
B3 · 1800 MHz (4G)100%100%100%
B7 · 2600 MHz (4G)99%100%99%

Site availability stays green because 4G and n1 are carrying the traffic — the degradation is confined to n78 on two of three sectors. That is exactly the shape that produces 5G→4G fallback and speed complaints while every site-level dashboard reads normal.