CafeTeleRadio Optimization SRPO-600 Enroll now ↗
‹ Day 2 · Retainability Course / Day 2 / 2.3 / LTE Retainability
RET
4G
2.3 · LTE · Retainability

LTE Retainability — Call Drop · E-RAB Release · RLF, in Full

Accessibility asks "did the call get in?" — retainability asks "did it stay in?" Samsung measures it as drops per session-time: ErabRetainability = SumRelActive ÷ RetainSessionTimeUE × 100, then decomposes every abnormal end across six families — Call Drop (24/106 per QCI), E-RAB Release (15 + by-eNB/by-MME 422/421), RRC Release (4), Re-establishment — the drop rescue path (3), RLF Detection (105) and MRO RLF Classification (26). Every drop counter carries the exact reason in its name. Per KPI: formula → counters → drop causes → optimization — every name verbatim from the Samsung eNB Counter Description.

24CALL_DROP (+106 QCI)
15ERAB_REL · RelActive
4RRC_RELEASE
3RRC_REESTAB rescue
105RLF_DETECTION
26MRO_RLF classify
The retainability KPI set — drops per session-time, then four decompositions

The headline is a rate over time, not a ratio of procedures: how often does a user abnormally lose a bearer per unit of holding time. Below it, four counter-level rates localize the loss. All formulas verbatim — the first from the Samsung eNB Counter Description (Retainability family, ch. 2 p. 1228), the rest from the operator reporting workbook.

The Samsung retainability KPIs — counters & formulas (verbatim)
KPIFamilyFormula (Samsung counters)Alert (deck)
ErabRetainability (official)RetainabilitySumRelActive ÷ RetainSessionTimeUE × 100 · per QCI: RetainRelActive ÷ RetainSessionTimeQci × 100<95% Critical
Call Drop Rate (%)24 CALL_DROPCallDrop(all types Σ) ÷ (EstabInitSuccNbr + InterX2InSucc + InterS1InSucc) × 100>2% Major
E-RAB Drop Rate (%)15 ERAB_RELRelActive ÷ (EstabInitSuccNbr + EstabAddSuccNbr) × 100abnormal >3% Major
RRC Abnormal Release Rate (%)4 RRC_RELEASE(ConnRelease_RrcHcTo + ConnRelease_RrcSigFail + ConnRelease_RrcSigTo) ÷ SumRrcConnEstabSucc × 100>5% Minor
RRC Retention Rate (%)100 − RRC Abnormal Release Rate
RRC Reconfig Timeout Rate (%)cross-family(CallDrop_EccbTmoutRrcConnectionReconfig + ErabInfo_CallDrop_TmoutRrcConnectionReconfig + RelActive_EccbTmoutRrcConnectionReconfig) ÷ RrcRrcConnectionReconfiguration × 100
Per-QCI alert bands (deck-verbatim) — Call Drop per QCI (Fam 106): QCI-1 > 1% (VoLTE), others > 3% Major · Packet Drop (Fam 33 PDCP_DROP) > 1% Minor. The denominator of Call Drop Rate includes incoming handovers (InterX2InSucc + InterS1InSucc) — a cell that receives traffic it then drops is charged for it.
Where a connected call dies — the drop chain & its peg points

Every counter in this page pegs after INITIAL CONTEXT SETUP RESPONSE — the call was successfully connected, then lost. The chain: the radio degrades → RLF detection (Family 105) → T310 runs → the UE attempts re-establishment (Family 3, the rescue) → rescued, or the eNB books the loss simultaneously in three ledgers: Call Drop (24), E-RAB Release (15) and RRC Release (4).

Connected → RLF 105 → T310 → re-estab 3 → saved / dropped (24 + 15 + 4)peg points
Connected
after Initial Ctx Setup Rsp
RLF detect · 105
OutOfSync from MAC
ArqMaxReTx from RLC
T310 running
n310 → T310 → RLF
recover if n311 in-sync
Re-estab · Fam 3
Att @ ReestabRequest
T311 cell-search window
SAVED
ConnReEstabSucc
DROPPED
CallDrop_* (24)
RelActive (15) · ConnRelease (4)
MRO classification · 26
too-late / too-early / wrong-cell / ping-pong
one radio event books into three ledgers at onceFam 24 counts the call, Fam 15 the bearers, Fam 4 the RRC connection — same drop, three views.
Read the counter name — the LTE drop-cause decoder

Samsung drop counters are CallDrop_Eccb<Cause> / RelActive_Eccb<Cause> / ConnRelease_<Domain><Cause>. The cause token names the failing layer — and therefore the team and the first place to look:

Cause token → layer → first place to look (deck-verbatim layer mapping)
TokenLayer / domainWhat actually happenedFirst place to look
RadioLinkFailurePHY / MACHARQ ACK never came back after max retransmissionsCoverage, UL interference, UL/DL imbalance
ArqMaxReTransmissionRLCARQ ACK not received after max RLC retransmissionsAir quality; enb/ue-max-retx-threshold
TmoutRrcConnectionReconfigRRCReconfigurationComplete timed out (often mid-handover)Mobility params (a3-offset/TTT), t304
TmoutRrcConnectionReestablishRRCReestablishmentComplete timed out — the rescue failedt311, neighbour health, RLF-triggered HO
DspAudit(Rlc/Mac)CallReleaseeNB internalECCB↔MAC/RLC audit mismatch — software/board stateBoard alarms, SW version — not a radio parameter
RcvResetRequestFromEcmb / RcvCellReleaseIndSystem / cell opsECMB reset · cell state → disabledOperations calendar — deliberate, exclude from tuning
S1SctpOutOfServiceS1 / transportS1AP link left service — every call on it books a dropBackhaul / MME SCTP — one event, mass drops
Cp* / Up* / RrcSig* / S1ap* / X2ap* (Fam 4)accessibility decoderSame prefix grammar as the accessibility page — call control / protocol blocks / air / interfacesReuse the Day-2.1 decoder on release counters
!
The retainability superpower — the same cause token repeats across the three ledgers: CallDrop_EccbRadioLinkFailure (24) + RelActive_EccbRadioLinkFailure (15) + CallDropQci_EccbRadioLinkFailure (106, per QCI) are one radio event read at call, bearer and QCI granularity. Diverging trends between ledgers = mix shift (which QCI is dying), not a new fault.
Ledger ① — Call Drop (Family 24 · CALL_DROP, + 106 per QCI)

The family (verbatim): "statistics for the call drop. Call drop statistics are calculated for successfully connected calls that abnormally dropped due to eNB failure or by the MME." Ten cause counters, indexed by CallDropEstabCause; Family 106 repeats all ten per QCI (plus the incoming-HO denominators ErabIncomingHoSuccNbr/AttemptNbr). All peg after INITIAL CONTEXT SETUP RESPONSE.

①-a · The 10 drop causes (deck-verbatim reason · layer)
CounterLayerDrop reason (verbatim)
CallDrop_EccbRadioLinkFailurePHY/MACHARQ ACK not received after max retransmissions — the radio drop
CallDrop_EccbArqMaxReTransmissionRLCARQ ACK not received after max retransmissions
CallDrop_EccbTmoutRrcConnectionReconfigRRCRRCConnectionReconfigurationComplete timeout
CallDrop_EccbTmoutRrcConnectionReestablishRRCRRCConnectionReestablishmentComplete timeout — the rescue itself timed out
CallDrop_EccbS1SctpOutOfServiceS1APS1AP link status changed to Out of Service
CallDrop_EccbDspAuditRlcMacCallReleaseRRC/MAC/RLCMismatch between ECCB and MAC/RLC during UE audit
CallDrop_EccbDspAuditMacCallReleaseMACMismatch between ECCB and MAC during UE audit
CallDrop_EccbDspAuditRlcCallReleaseRLCMismatch between ECCB and RLC during UE audit
CallDrop_EccbRcvResetRequestFromEcmbSystemECMB reset received
CallDrop_EccbRcvCellReleaseIndFromEcmbCell mgmtCell operational state changed to disabled
①-b · Optimization playbook — dominant cause → diagnosis → lever
Dominant counterDiagnosisTechnique / lever
…RadioLinkFailureCoverage / interference / UL imbalanceRF first (RSRP/SINR maps, tilt/power); then RLF timers t310 (1000→2000 ms deep-fade) + n310 (n1→n2), HARQ dl-max-harq-transmission 3→4 — use the T310 simulator below
…ArqMaxReTransmissionRLC exhausts in poor-but-recoverable radioenb-max-retx-threshold / ue-max-retx-threshold t8→t16 (adds latency — not for QCI-1); check UL interference
…TmoutRrcConnectionReconfigReconfig lost mid-air — usually a mobility eventRead Family 26 first (too-late vs too-early vs wrong-cell); then a3-offset / time-to-trigger / hysteresis, t304 1000→2000 ms if execution-phase
…TmoutRrcConnectionReestablishThe rescue is failing — recovery path brokent311 5000→10000 ms, enable LTE-SW1014 RLF Triggered Handover, verify X2 to re-estab targets (see InterEnbConnReEstabReject_RlfHoNotPossible)
…S1SctpOutOfServiceTransport event — mass simultaneous dropsS1/backhaul health, SCTP paths, MME redundancy; s1-relocation-overall only if HO-release related — not a radio parameter
…DspAudit* / RcvReset*eNB software/board state, or planned opsBoard alarms, SW version, restart history; exclude maintenance windows from KPI baselines
Ledger ② — E-RAB Release (Families 15 · 14 · 422 by-eNB · 421 by-MME)

The family (verbatim): Family 15 counts "E-RABs released by the E-RAB Release Command message from MME to eNB, and the active E-RABs that are released abnormally"RelActive is the numerator of both official retainability KPIs. Families 422/421 split every release by initiator (eNB vs MME) and normal vs abnormal, per QCI — the cleanest view of who ended the bearer and whether data was still flowing (…ActQci = data remained in the buffer).

②-a · Family 15 — the RelActive split (verbatim descriptions)
CounterInitiatorVerbatim description
RelActiveeNBCall releases due to an abnormal internal operation of the eNB in the setup E-RAB — the retainability numerator
RelActive_EccbRadioLinkFailureeNBEstablished E-RABs released — HARQ ACK never returned after max retransmissions
RelActive_EccbArqMaxReTransmissioneNBEstablished E-RABs released — RLC ARQ exhausted
RelActive_EccbTmoutRrcConnectionReconfig / …ReestablisheNBEstablished E-RABs released — RRC reconfig / re-establishment timeout
RelActive_MmeInitErabReleaseMMEUE Context Release Command with abnormal CAUSE from MME, data still buffered
RelActive_ResetReleaseMME/eNBReset message from/to MME with buffered data — transport/core event
RelActive_ErabReleaseMMEE-RAB Release Command / Indication with abnormal cause
RelAttbyEnbNbr_CpCcTo / _S1apCuFail / _CpPreemption (Fam 14)eNB→MMEeNB-requested releases via E-RAB Release Indication — incl. preemption
②-b · Families 422/421 — initiator × normal/abnormal, per QCI (verbatim)
CounterInitiatorMeaning
RelNormalbyEnbQci_UserInactivityeNB · normalReleased by 'User Inactivity' cause — healthy behaviour, not a fault
RelNormalbyEnbQci_PreeNB · normalNormal release by preemption policy
RelAbnormalbyEnbActQcieNB · abnormalAbnormal release with data remaining in the buffer — the true user-visible drop
RelAbnormalbyEnbActQci_RadioLinkFail / _TimeOut / _S1LinkFail / _S1ReseteNB · abnormalThe abnormal-with-data split by cause — radio vs timers vs S1
RelAbnormalbyMmeActQci (+_S1Reset, _EutranGen)MME · abnormalMME-initiated abnormal with buffered data; _EutranGen = cause originated in E-UTRAN
i
Playbook — compute the abnormal-with-data share: RelAbnormalbyEnbActQci ÷ RelbyEnbQci per QCI. High on QCI-1 = VoLTE users hearing dead air → radio/mobility work. High RelAbnormalbyMme* = core-initiated (paging, S1, MME pool) → not a radio parameter. High RelNormal…UserInactivity with short session times = inactivity timer too aggressive → check ReleaseCntByTimer (Fam 333) and lengthen the inactivity profile before touching RF.
Ledger ③ — RRC Release (Family 4) & the rescue path (Family 3 · RRC_REESTAB)

Family 4 books every RRC connection release with a 25-way cause split using the same Cp/Up/RrcSig/S1ap/X2ap prefix grammar as accessibility. The abnormal-release KPI counts three of them: RrcHcTo (reconfig lost in intra-eNB handover), RrcSigFail, RrcSigTo. Family 3 is the counter-evidence of recovery: every re-establishment that succeeds is a drop that didn't happen.

③-a · Family 4 — the release-cause map (25 counters, key rows verbatim)
CounterDomainVerbatim description
ConnRelease_CpCcNormalnormalNormal release — the healthy end of a call
ConnRelease_RrcHcToAir · mobilityReconfigurationComplete not received during the intra-eNB handover — KPI counter
ConnRelease_RrcSigTo / _RrcSigFailAir · UERRC response timeout / signalling failure after connection established — KPI counters
ConnRelease_UpMacUEInactnormal-ishReleased on inactivity timeout — policy, not fault
ConnRelease_UpMacFail / _UpRlcFail / _UpPdcpFail / _UpGtpFaileNB boardsFailure between protocol block and ECCB — board/transport health
ConnRelease_S1apLinkFail / _S1apSigTo / _S1apRoTo / _S1apCuFailS1 · MMES1 link / signalling timeouts — core-facing releases
ConnRelease_CpPreemptionAdmissionReleased by preemption after setup — ARP policy in action
ConnRelease_MmeOverloadCoreRelease sent because the MME is in overload state (LTE-SW0503 Overload Action)
③-b · Family 3 — re-establishment: the rescue ledger (verbatim)
CounterRoleVerbatim description
ConnReEstabAttattemptRRCConnectionRe-establishmentRequest messages received from the UE
ConnReEstabSuccsuccessReestablishmentComplete received — a drop prevented
ConnReEstabFail_UpMacFail / _UpPdcpFail / _UpRlcFail / _CpCcFail / _RrcSigTo / _S1ap*failThe rescue failed — same domain grammar; the failing layer names the fix
ConnReEstabReject_CpCcFail / _CpCapaCaccFail / _CpQosCacFailrejectThe eNB refused the rescue — CAC/capacity rejecting re-establishment
InterEnbConnReEstab* (12 counters)inter-eNBThe UE re-establishes at a different eNB — needs UE context via X2
InterEnbConnReEstabReject_RlfHoNotPossiblerejectRejected because RLF Triggered Handover is OFF or X2 is missing — the feature gap, verbatim
The rescue-rate lens — ConnReEstabSucc ÷ ConnReEstabAtt is your drop-recovery rate. Low with high Reject_Cp*CacFail = you are refusing rescues under load (review CAC for re-establishment). High InterEnbConnReEstabReject_RlfHoNotPossible = enable LTE-SW1014 + audit X2/NRT — the counter literally names the missing feature. Re-establishment success books the call as retained; every improvement here moves the drop KPI directly.
Ledger ④ — RLF Detection (Family 105) & MRO Classification (Family 26)

Family 105 counts the raw physics: RlfDetection_RadioLinkFailure"ECCB receives an OutOfSync notification from MAC" (verbatim) — and RlfDetection_ArqMaxReTransmission from RLC (plus EnDc_* twins for NSA anchors). Family 26 tells you why: the SON/MRO algorithm classifies each handover-related RLF.

④-a · MRO RLF classification — per neighbour relation (verbatim)
CounterMRO verdictVerbatim triggerStandard fix direction
TooLateHoRlfBeforeTriggering / …AfterTriggeringToo-late HORLF before the A3 report / after the HO command was transmittedEarlier trigger: ↓a3-offset, ↓time-to-trigger, CIO+
TooEarlyHoFailure / TooEarlyHoRlfAfterHoToo-early HOToo-early handover after completing the handover processLater trigger: ↑a3-offset, ↑TTT, CIO−
WrongCellRlfAfterTriggering / …AfterHo (+InterFreq)Wrong cellRLF received from a neighbour after HO completed (same-frequency target)NRT audit, PCI confusion check, per-pair CIO
PingpongHandoverPing-pongPing-pong conditions met by the detection algorithmhysteresis, ↑TTT on the oscillating pair
CoverageHole / CoverageHoleRsrpCoverage holeRLF/HO failure due to coverage hole based on SINR / RSRPRF design: tilt/power/new site — parameters won't fill a hole
i
Playbook — Family 26 is indexed per neighbour cell — it hands you the exact cell-pair to fix. If MRO is licensed, set mobility-robustness-enable (son-cell-info-func) to son-auto-apply and let it steer CIO per pair; the deck calls this "a significant optimization step". Voice twin: Family 353 MRO_RLF_VOICE repeats the classification for QCI-1 only — read it before touching VoLTE mobility.
Interactive — the T310/N310/N311 RLF machine, simulated

The entire RLF chapter compresses into four numbers. Drag the fade length and the timer values and watch the outcome flip between ride-through, rescued and dropped — then read which counters would peg. Model: out-of-sync indications arrive every 10 ms radio frame while the fade lasts (teaching model; deck-verbatim value sets).

RLF Timer Simulatorlive
The event
The four levers (timer-info)
Deck presets (verbatim)
Outcome
LTE Retainability Studio — three ledgers → grade → weakest link → fix

Enter the busy-hour counters; the studio computes the three drop KPIs plus the rescue rate, grades each against the deck's alert bands, finds the weakest ledger, and the engine below ranks every drop cause into an action plan.

Retainability Studiolive
Denominators (successes)
Drop causes (Fam 24) & ledgers
Call Drop Rate — Family 24 ÷ (init + incoming HO)
%
Recommended action
Diagnostic engine — drop Pareto & ranked action plan
Drop Pareto — who is eating the retainabilitycomputed live from the studio counters
Ranked action plan — fix in this order
The LTE drop-cause routing fabric — domain × ledger

Every drop counter routed from its root-cause domain to the ledger where it pegs. Click a domain for its blast-radius, a ledger for everything that books there, a scenario for a cross-ledger signature — or any counter chip for its full optimization card.

LTE Drop-Cause Routing Fabric
Scenarios
The retainability feature stack — prevent · rescue · move · protect

Ten Samsung features from the deck's "Feature for Improving Call Drop Rate" table, ordered by role in the drop chain: harden the link, rescue the failure, move the UE in time, and protect the cell under load. Click any node for its full card.

LTE Retainability Feature Pipelineclick a feature node to open its detail card
!
Order matters — features that prevent RLF (ICIC/IRC/MIMO/link-adaptation) buy the most; the rescue features (SW1014, re-establishment) convert the remaining failures into saves; mobility features stop drops from becoming inter-cell casualties; CAC/overload stop load from manufacturing drops. Don't tune rescue timers on a cell whose real problem is interference.
The LTE retainability parameter master table — 30+ levers (deck-verbatim)

Every retainability-relevant parameter from the Samsung LTE Retainability Optimization deck, grouped by function — MO source object, typical/default and the deck's recommended move with its caution. Ranges/defaults are the deck's stated typical values; always RTRV your release template first.

A · RLF timers & recovery — timer-info (the T310 machine)
ParameterTypicalDeck recommendationCaution
t3101000 msDeep fades where UEs could recover → 2000 ms. Deck profiles: Rural/Highway 2000 · Urban dense 1000 · Indoor/Stadium 1500 · High-speed 500Too long delays true-failure detection and re-establishment
n310n1Temporary fades → n2 (less sensitive RLF detection)Delayed RLF detection can mask other issues
n311n1Ping-pong RLF (recovers then fails again) → n2: require steadier recoverySlower recovery declaration
t3115000 msRe-estab failing on slow cell-search → 10000 msVery long keeps UE trying a bad frequency, delaying a fresh connection
t304ms1000HO-execution drops (target not ready) → 2000 msToo long delays recovery when HO is genuinely stuck
enb-max-retx-threshold / ue-max-retx-threshold (pld-rlc-info-func2)t8RLF by RLC-max-retx in recoverable radio → t16Adds latency — not ideal for QCI-1/VoLTE
B · Mobility — Event A3/A5, idle reselection & relocation guards
ParameterTypicalDeck recommendationCaution
a3-offset (eutra-event-a3-info-func)1–6 dBToo-late HO → decrease (3→2/1 dB); ping-pong → increase (2→3/4 dB)Monitor MRO Fam 26 after every move
time-to-trigger320/480/640 msToo-late HO → decrease (640→480/320); ping-pong → increaseBalance with a3-offset — short TTT + conservative offset often beats the reverse
hysteresis1–3 dBPing-pong → increase (1→2 dB); sticky serving cell → decreaseWorks jointly with a3-offset and TTT
a5-threshold1-rsrp / a5-threshold2-rsrp−110…−115 / −105…−110 dBmCell-edge drops before inter-freq HO → Th1 more aggressive (−110→−108); weak targets → Th2 more stringent (−110→−105)Keep a 5–10 dB gap or the thresholds oscillate
s-measure (mobility-common-info-func)≈ −100 dBmLate neighbour measurement on degrading serving → lower it (−100→−105/−110)More measuring = more UE battery
q-hyst (cell-reselection-general-info-func)2 dBDrops right after idle reselection + call start → 4 dB for stabler campingSlower idle reselection
mobility-robustness-enable (son-cell-info-func)son-auto-apply — MRO fixes too-early/too-late/wrong-cell automaticallyVerify MRO statistics are collected and acted on
x2/s1-relocation-overall (timer-info)2–5 s / 2–10 s"HO overall expiry" releases → increase slightlyTies up source-eNB resources longer
no-handover (eutran-neighbor-cell-relation-info-func) · administrative-statefalse / unlockedChecks, not tuning: a barred pair or a locked neighbour = drops with no parameter cureCritical after NRT updates / maintenance
C · HARQ / MAC / UL power — link robustness
ParameterTypicalDeck recommendationCaution
dl-max-harq-transmission (adaptive-modulation-coding-info-func)3–4RLF after DL HARQ NACK storms → 3→4Latency + resource cost per retransmission
max-harq-tx (ul-mac-config-info-func)n4UL-driven RLF → n4→n5UL latency grows
p0-nominal-pusch (ul-common-phy-info-func)−100…−80 dBmCell-edge UL SINR drops → +2…3 dB (e.g. −96→−93)UL interference to neighbours, battery
alpha0.7/0.8/1.0Edge-UE drops → al1 (full pathloss compensation)Fractional (0.7/0.8) manages interference better
time-alignment-timer-commonsf1280Drops preceded by UL-sync loss / RACH storms → sf2560Drift risk; with infinity ensure periodic TA works
tti-bundling (ul-mac-config-info-func)falseCell-edge VoLTE UL drops → enable (adaptive via ttib-in/out-tbs)Capacity cost; VoLTE-focused
periodic-bsr-timer / retx-bsr-timersf640 / sf2560Slow UL grants at talk-spurt start → halve periodic (640→320)More BSR = more UL signalling overhead
dl-target-bler / volte-dl/ul-target-bler~10%Voice robustness → VoLTE BLER targets 1–5% (lower than data)Costs capacity; >15–20% general BLER breeds RLF
D · VoLTE-specific — QCI-1 keeps its own rules
ParameterTypicalDeck recommendationCaution
discard-timer (pld-pdcp-info-func, QCI-1)100–500 msDrops via PDCP discard in congestion → 150→300 msEnd-to-end delay — conversational quality
scheduling-type (qci-resource-info-func, QCI-1)sps-schedulingIf dynamic → switch to SPS: regular low-latency voice allocation, deck calls it fundamental
implicit-release-after (sps-config-info-func)e2/e4Glitches after silence periods → e2→e4 (keep SPS alive longer)Wastes grants on long silences
rohc-context-continueenableKeep enable — avoids ROHC reset hiccups at intra-eNB HO
E · Admission, preemption & system checks
ParameterTypicalDeck recommendationCaution
preemption-enable / low-call-release-option (cell-cac-info-func)use / longest-callVoLTE preempted under load → ARP hygiene first; consider arp-based release policyPreemption books into ConnRelease_CpPreemption — by design
qos-cac-option / qos-policy-optionqos-cac-use / option0option1 for granular GBR admission so bulk GBR can't starve QCI-1Align with ARP/preemption settings
check-ue-id-enableuseDrops on S-TMSI duplication (existing call released) → consider no-useMME-pool duplication handling shifts to the core
crsic-support-cell-conf (cell-phy-config-info-func)HW-dependentDense grids with CRS interference → use (CRS-IC on)Re-evaluate if a specific cell degrades after enabling
si-window (sib-control-info-func)20 msTransition-time drops with SIB acquisition suspicion → 40 msLonger full SI cycle
ra-response-window-size / contention-resolution-timer (rach-config-logic)sf6 / sf48Recovery-RACH failing under load → sf8–10 / sf56–64Slower failure detection on genuinely missed preambles
!
Method, not magic numbers — values above are the deck's typical/recommended figures — releases differ, so RTRV the live MO first, back up (*.bak convention), move one lever per soak, and verify on the owning family (24/15/4/3/105/26) plus the guards: re-establishment success (Fam 3) and HO success — a "retainability gain" that ships failures into mobility is a net loss.
The 7-step LTE retainability optimization MOP
1
Baseline — pull all six families per cell

Busy hour: Families 24/106 (drops), 15/421/422 (E-RAB), 4 (releases), 3 (re-estab), 105/26 (RLF/MRO). Compute Call Drop Rate, E-RAB Drop Rate, RRC Abnormal Release Rate and the rescue rate.

CALL_DROP · ERAB_REL · RRC_RELEASE · RRC_REESTAB · RLF · MRO Skip this and the cell average hides the one dying QCI / neighbour pair.
2
Separate normal from abnormal

RelNormal…UserInactivity, ConnRelease_CpCcNormal/_UpMacUEInact, ReleaseCntByTimer, preemption and maintenance releases are policy, not faults. Compute the abnormal-with-data share (RelAbnormal…ActQci).

Fam 422/421 normal vs abnormal · Fam 333 If wrong: you "optimize" a KPI inflated by healthy inactivity releases.
3
Rank the drop causes — the Pareto

Within Family 24 (and 106 for QCI-1), rank the ten causes (the engine above does this). The token names the layer: RLF=air · ArqMax=RLC · Tmout*=RRC/mobility · S1Sctp=transport · DspAudit=eNB internal.

cause token = layer = team If wrong: tuning timers on a cell whose top cause is S1 transport.
4
Read the MRO verdict per neighbour pair

If Tmout…Reconfig or RLF dominates, open Family 26: too-late vs too-early vs wrong-cell vs ping-pong vs coverage-hole — per neighbour relation. Coverage holes get RF design, not parameters.

Fam 26 per NeighborCell · Fam 353 for voice If wrong: a3-offset moved the wrong way doubles the ping-pong.
5
Apply ONE lever

Air → t310/n310 + HARQ/RLC depth (+ TTI bundling for VoLTE edge). Mobility → a3-offset/TTT/hysteresis or son-auto-apply MRO. Rescue → t311 + LTE-SW1014 + X2 audit. Transport/core → S1/MME with the evidence, not radio params.

one change per soak — attribute the gain If wrong: two levers at once = unattributable results.
6
Soak & verify — same counters, same hour

The target cause must fall, its KPI must improve, and the guards must hold: rescue rate (Fam 3) up or stable, HO success stable, accessibility untouched. Check the same cause in ledgers 15 and 106 moved with it.

before/after on 24+15+4 · guards: 3, HO SR, Fam 1/8 If wrong: longer T310 "improves" drops while re-estab delay ruins user experience.
7
Lock in — the VoLTE view

Re-read Family 106 for QCI-1 (>1% alert) and Family 353 MRO_RLF_VOICE; confirm SPS/discard-timer/BLER-target hygiene. Document the change and the counter evidence.

106 QCI-1 · 353 · SPS/discard-timer If wrong: the aggregate looks fixed while voice still drops.

Retainability is one radio event, three ledgers, one diagnosis.

Drops per session-time, decomposed: Call Drop 24 · E-RAB 15 · RRC 4 — with Family 3 as the rescue and Family 26 as the witness. Decode the cause token, fix the owning layer. Then meet the same discipline in 5G.

Sources: Samsung eNB Counter Description (SLR 4.5 v7.0 — Retainability family formula p. 1228 verbatim; SLR 6.0) and the operator KPI workbook (LTE Samsung KPI.xlsx, "eNB Counters" sheet — Families 24 CALL_DROP, 106 CALL_DROP_QCI, 15 ERAB_REL, 14 ERAB_REL_ENB, 421/422 ERAB_REL by MME/eNB per QCI, 4 RRC_RELEASE, 3 RRC_REESTAB, 105 RLF_DETECTION, 26 MRO_RLF, 333 TIMER_RELEASE — counter names, descriptions and triggers verbatim) · Samsung LTE Retainability Optimization deck (alert thresholds, counter→reason→layer tables, parameter recommendations, feature table) · Troubleshooting Call Drop Failures deck · eNB (LTE) Feature Description (SVR24B — MO/parameter names). The SVR25A LTE Counter/Parameter Description originals are NASCA-DRM-locked; readable SLR equivalents were used. RLF procedure per 3GPP TS 36.331 §5.3.11. Targets marked "typical" are operator practice; counters, families and formulas are vendor-exact.