← course  ·  MODULE 2 · THEORY + KPI OPTIMIZATION SCENARIOS

Getting In

Call-setup & accessibility optimization, written as a book chapter: the access door, the admission lobby, the fragile first seconds, the resurrection, the search, and the sacred path — closing with six KPI-optimization case studies you can run on a real network.

No fake things. Every FAJ/CXC, MO attribute, threshold, and feature state below is from the Ericsson CPI (LTE RAN 25.Q4.4), the live NYC node kget (25.Q3), the operator golden file, and the LTE PM counter inventory. 3GPP citations name exact specifications.

CH 1Access barring

Every SIM holds a random access class 0–9; AC 10 is emergency, AC 11–15 are priority users (TS 22.011). SIB2 carries, per traffic type, an ac-BarringFactor and ac-BarringTime (TS 36.331 §5.3.3.2): a UE draws rand(), proceeds if below the factor, else waits (0.7 + 0.6·rand) × barringTime and retries — probabilistic crowd control.

On the live cells: acBarringForMoData/MoSignalling/Csfb all staged at factor 95, time 4 s; the 17-member acBarringPresence struct runs AUTO on MoData/MoSign/CSFB — automation owned by Load-Based Access Barring (FAJ 121 3093 / CXC4011807, ACTIVATED), which starts barring from the RRC reject:attempt ratio, recalculates periodically, and — per the FD — withdraws the cell from inter-frequency load-balancing candidacy until barring probability returns to 0. Progressive Access Barring (FAJ 121 4746) adds three priority tiers over five traffic classes; this operator parks it (golden: Deactivated) and maps all five classes to PRIORITY0 with special ACs at PRIORITY4.

The VoLTE skip — Prioritization of VoLTE in Access Barring (FAJ 121 4329 / CXC4011942, High Load Handling FAJ 801 0415): three cell booleans (acBarringSkipForMmtelVoice/Video/Sms) inject Rel-12 SIB2 IEs (ac-BarringSkipForMMTELVoice-r12…) that let voice originations bypass the barring check. Two FD footnotes: skips broadcast only while barring broadcasts; only Rel-12 UEs read them (Rel-9/10/11: SSAC only; Rel-8: neither). On this network the skip is off and golden-compliant — a deliberate engineering position: skipped attempts still cost PRACH/PDCCH/RRC and fail deeper, where failures are expensive; barring is the only tool protecting already-connected calls. Emergency is never barred: acBarringForEmergency=false, presence MANUAL.

CH 2Seeing voice early

Historically a VoLTE origination arrived as mo-Data — invisible at the moment of rejection. Rel-12 (TS 36.331 V12.8.0) added the mo-VoiceCall-v1280 establishment cause, broadcast-enabled by SIB2 voiceServiceCauseIndication-r12, and copied into the S1AP InitialUEMessage (core must support TS 36.413 ≥ V13.2.0).

Dynamic UE Admission Control (FAJ 121 4301) defends against connection storms with three states — Admit All → Initial Blocking → Block All — crossed at a differentiation and an overload threshold. Optimized VoLTE Access at High Load (FAJ 121 5062 / CXC4012368) makes exactly one move, quoted from the FD: “a Mobile Originating voice request is given the same priority level as Reestablishment and Handover requests” — admitted in Initial Blocking, rejected only in Block All. New counters split voice out (pmRrcConnEstabFailDynUeAdmCtrlMoVoice), and the FD warns the Initial E-RAB Establishment SR formula must incorporate the new moVoice counters. On this network the whole DUAC chain is parked by documented policy (both CXCs DEACTIVATED; golden: Deactivated, “Not GA'd as off”) — this operator defends with reservations and barring instead. Two schools; mixing them carelessly multiplies state machines.

CH 3The admission family

LayerIdentityAddsThis node
BasicFAJ 121 1857 · unlicensedPA reservations + transport NW admission (dl/ulTransNwBandwidth=1000)base behavior
DifferentiatedFAJ 121 1787 / CXC4011062ARP blocking + pre-emption (optionally SPID)ACTIVATED
Dynamic GBRFAJ 121 1748 / CXC4011060 · VoLTE Performance VPadmission on measured utilization (MSR), headroom for fading/mobilityACTIVATED

The thresholds form a corridor: dlAdmDifferentiationThr=800 (low-ARP GBR refused) → dlAdmOverloadThr=850 (hard wall; MOM default 950 — this operator stops earlier); UL mirrors both; MOM rule Overload ≥ Diff. Pre-emption is armed (arpBasedPreEmptionState=ACTIVATED) with two refinements: preemptInactTimerMin=15 s (default 4 — don't evict the briefly silent) and recalcPreemptableUeTimer=1 s. Rejected high-ARP requests cascade into pre-emption of admitted low-ARP bearers. Costs land in named counters: pmErabEstabFailGbrDlEnb/UlEnb, pmErabEstabAttRejAdmMsr…Arp (per-ARP bins), and victims as abnormal releases — when pre-emption is armed, some drop rate is policy, not fault.

CH 4Protecting the first seconds

Call setup (~2 s ladder) races mobility's continuous clock (A3/A5/B1/B2). LTE Delayed Mobility for Robust Call Setup (FAJ 121 5850 / CXC4012835) is selective deafness during setup, per direction, via the RobustCallSetup MO: intraLteMobilityDelayEnabled (discard A3/A5, block blind IF-HO), geranUtranMobilityDelayEnabled (discard B1/B2, block blind RwR to GSM), endcSetupDelayEnabled (discard EN-DC B1; FD recommends reportIntervalB1 ≥ MS_480). Side-effect catalog: ANR's PCI-conflict discovery and every coverage-triggered feature are delayed too — regression scope is the mobility KPIs.

Robust SRVCC Handover at Call Setup (FAJ 121 4855 / CXC4012240, VoLTE VP): when QCI-1 establishes and srvccDelayTimer>0, a B2-triggered SRVCC preparation is cached (newest kept), released three ways — A1 (coverage back: discard), A3/A5 (LTE solution wins), or timer expiry (proceed, late enough to survive). On this metro grid both features sit DEACTIVATED with srvccDelayTimer=0 — geography is a parameter; corridor/rural networks flip it.

CH 5Re-establishment

After radio link failure the UE sends RRCConnectionReestablishmentRequest (TS 36.331 §5.3.7) carrying C-RNTI + PCI of the failed cell + shortMAC-I — a fingerprint, valid for a seconds-long window. The basic feature (FAJ 121 3019, unlicensed) answers “yes” only where the context lives; calls die at borders onto foreign eNodeBs. Multi-Target (FAJ 121 3023 / CXC4011366, ACTIVATED here) extends it: cross-eNodeB context fetch over X2, basic inter-vendor support, and the EN-DC case with a clean Secondary Node Release first. FD requirement worth framing: emergency ARP must be uniform network-wide or X2→S1 handovers fail later — multi-node features audit your network for you, rudely. Every success subtracts directly from pmErabRelAbnormalEnbActQci[1].

CH 6Paging & Priority Paging

Idle UEs wake only at occasions computed from identity + defaultPagingCycle=128 (1.28 s) and nB=T (TS 36.304); one paging message carries ≤ maxNoOfPagingRecords=7; stale pages die at pagingDiscardTimer=3 — late delivery is worse than none, the MME re-pages fresh. A lost page is the stealthiest accessibility killer: the UE never attempts, so nothing steps in the radio counters — only the paging counters see it. Priority Paging (FAJ 121 3081 / CXC4011711, ACTIVATED) lets MME-marked voice pages (S1AP paging priority) jump the queue: it creates no capacity, it decides who gets the capacity that exists. 41 golden rows govern the Paging MO — the least casual tuning surface in the module: idle-mode dials move whole tracking areas.

CH 7Emergency

Layered and absolute: never barred (acBarringForEmergency=false, presence MANUAL); recognized by the Rel-8 emergency establishment cause above every ordinary class; privileged at admission via Emergency Call Prioritization (FAJ 121 1039 — unlicensed, hence no FeatureState object in the kget: expected) protecting initial access, E-RAB establishment, and incoming handover, with the PA seats and paArpOverrideHpa lane; emergency attach/PDN per TS 23.401, PSAP via TS 23.167; fallbacks defined across every boundary (CSFB emergency; Limited Service Mode FAJ 121 1743; LTE↔NR standalone). Emergency configuration is audited as compliance with evidence, not optimization.

CH 8Six KPI-optimization case studies

Each scenario follows the professional loop: KPI symptom → counter evidence → root cause → parameter/feature action (exact MO.attribute) → verification plan → rollback. All configuration named is real; the traffic narratives are the engineering situations these features exist for.

CASE 1

Accessibility “drops” during a stadium event — and is not a fault

KPI symptom
Added E-RAB Establishment SR (bin 1) dips during the event window; management asks for an emergency parameter change.
Evidence
pmErabEstabAttAddedQci[1] itself fell — fewer attempts ever reached the eNodeB. SIB2 captures show Load-Based barring active (factors recalculated from the 95/4 staging); the cell withdrew from IFLB per the FD.
Root cause
Barring working upstream of the KPI tree. With the VoLTE skip deliberately off (golden-compliant), voice thins with data at the door.
Action
None to admission. If policy review wants voice protected at events: trial EUtranCellFDD.acBarringSkipForMmtelVoice=true + FAJ 121 4329 activation on event cells only (precondition: barring presence active).
Verification
Rel-12 IE visible in SIB2; bin-1 att/succ vs control cells during the next event; cost side on pmErabEstabFailGbrDlEnb + admission rejects.
Rollback
One boolean per cell.
CASE 2

Voice survives a festival; a subpopulation still fails

KPI symptom
On a network running the DUAC chain: RRC attempts ×3, bin-1 SR holds 99% — but complaint tickets cluster on specific handsets.
Evidence
pmRrcConnEstabFailDynUeAdmCtrlMoVoice ≈ 0 while plain DUAC rejects climb steeply — the voice lane worked. Ticketed IMEIs are Rel-10/11.
Root cause
Pre-Rel-12 phones cannot send mo-VoiceCall-v1280; their calls arrived as mo-Data and were triaged with the crowd.
Action
No parameter. Handset capability report; optionally lower the differentiation threshold margin before the next event to widen Admit-All.
Verification
Failure share by UE release in the next event window.
CASE 3

The lunch-hour GBR ceiling

KPI symptom
One business-district cell: bin-1 SR 97% every weekday 12:00–13:00; neighbors hold 99%.
Evidence
pmErabEstabFailGbrDlEnb steps only in that hour; measured (MSR) utilization peaks ~87% — above dlAdmOverloadThr=850.
Root cause
A capacity ceiling with a calendar — the corridor (800/850) behaving as designed.
Action (in order)
① structural: carrier add / sector split; ② steering: move data load via IFLB thresholds; ③ only then: trial AdmissionControl.dlAdmOverloadThr 850→900 on that cell, knowing it spends the fading/mobility headroom Dynamic GBR exists to protect.
Verification
Bin-1 SR in the lunch hour and pmErabRelAbnormalEnbActQci[1] (drops must not rise); next-cell-unlock change window per MOM.
Rollback
Restore 850; takes effect next cell unlock.
CASE 4

“Pre-emption broke our retainability” — a KPI literacy case

KPI symptom
After arming differentiation on a venue cluster: accessibility up as designed, total drop rate up, ticket claims the feature is faulty.
Evidence
Per-ARP release split: voice (bin 1) drops unchanged; the increase concentrates in low-ARP data bearers — the designed eviction cost, precisely placed.
Root cause
The dashboard reads retainability as one number.
Action
No parameter. Split the KPI by ARP/QCI on the dashboard; document the policy trade; close the ticket.
Guard rails already set
preemptInactTimerMin=15 (no eviction of briefly-silent users), recalcPreemptableUeTimer=1.
CASE 5

The highway corridor — SRVCC kills calls at birth

KPI symptom
Bin-1 SR two points below city average on a corridor cluster only.
Evidence
Setup failures with handover-related causes; UE traces show B2 reports inside the setup ladder, SRVCC preparations firing while QCI-1 builds. Coverage is genuinely thin.
Root cause
Mobility racing call setup at the 3G boundary.
Action (in order)
① B2/IRAT threshold hygiene (module 6); ② activate FAJ 121 4855 on the cluster, EUtranCellFDD.srvccDelayTimer = a few seconds; ③ if traces also show A5 storms: FAJ 121 5850 with intraLteMobilityDelayEnabled=true only — keep the IRAT escape live.
Verification
Two counters, opposite directions: setup-phase failures with HO causes ↓ while SRVCC success rate stays flat (handovers still happen, later). A week each side vs control.
Rollback
Timer to 0; switches to false.
CASE 6

The invisible paging outage

KPI symptom
Voice complaints spike (rings → voicemail) on an app-update morning; every radio KPI from Module 1 is clean.
Evidence
Paging records sent flat at the ceiling (7/occasion, cycle 128, nB=T); discarded pages climbing at pagingDiscardTimer=3; discards uniformly distributed — voice dying with notifications.
Root cause
MT chain failing upstream of all radio counters; no page priority in force.
Action
Activate FAJ 121 3081 (here: already ACTIVATED — the case is why); core team confirms the MME marks IMS voice pages with S1AP paging priority — an end-to-end checkbox the eNodeB cannot supply alone.
Verification
Next event: discards continue but concentrate on re-pageable background traffic; voicemail complaints collapse. Standing action: discarded-pages-per-cause joins the voice dashboard.
Do not
Improvise on cycle/nB/records — 41 golden rows, tracking-area-wide blast radius.
The method behind all six

Diagnosis before medication: name the KPI, walk its tree (Module 1, lesson 1-5), find the counter that moved, read the config that governs it, and only then touch a parameter — on a trial, with a two-sided verification plan and a one-line rollback. Features whose golden state is Deactivated are arguments to understand, not defaults to flip.

REFReferences