‹ Day 2 · Accessibility Course / Day 2 / 2.1 / 5G SA Accessibility

2.1 · 5G SA · Option 2

5G SA Accessibility — RRC · NG · F1 · PDU, in Full

Standalone 5G is native NR into the 5GC — and Samsung counts it as a four-stage chain, one counter family per stage: RRC Setup (Family 5213) → NG-Connection to the AMF (5223) → F1 UE Context across the CU↔DU split (5136/5137) → PDU Session (5222). Each family carries its own attempt, success and cause-split failure counters, and every failure suffix names the owning element. Per stage: KPI → counters → failure counters → optimization — every name verbatim from the Samsung 5G NR Counter Description.

4Stages: RRC·NG·F1·PDU

5213RRC_ESTAB

5223NG_ESTAB (AMF)

5136·7F1 UE Setup CU/DU

5222PDU_SESSION_SETUP

5243PRACH per beam

The SA accessibility chain — four families, one product

Each stage's success rate is a straight ratio of its family's counters; end-to-end session accessibility is their product — the same decomposition method as LTE's ErabAccessibilityInit, applied to the SA chain. A single headline number can't localize a fault; the four factors can.

The four stage KPIs — Samsung counters & formulas

Stage KPI	Family	Formula (Samsung counters)	Typical target
① RRC Setup SR	`5213 RRC_ESTAB`	`ConnEstabSucc ÷ ConnEstabAtt × 100`	≥ 99%
② NG-Connection SR	`5223 NG_ESTAB`	`NgConnEstabSucc ÷ NgConnEstabAtt × 100`	≥ 99.5%
③ F1 UE Context SR	`5136 F1AP_UE_SETUP_CU`	`CU_F1UeSetupSucc ÷ CU_F1UeSetupAtt × 100`	≥ 99.5%
④ PDU Session SR	`5222 PDU_SESSION_RESOURCE_SETUP`	`PduSetupSucc ÷ PduSetupAtt × 100`	≥ 99%
Session accessibility	chain	`① × ② × ③ × ④`	≥ 98%

≡

Companion families — 5214 RRC_SETUP_TIME times stage ①; 5224 NG_SETUP_TIME times stage ②; 5438 repeats the PDU family per PLMN per S-NSSAI (the slice/operator view); 5226 RRC_MSG and 5225 NGAP_MSG_UE hold the raw message evidence (Rx_RRCSetupRequest, Rx_NGInitialContextSetupRequest…); 5203 RRC_REESTAB covers recovery.

Where each family pegs — the SA call flow

The UE RACHes (per-beam Family 5243), completes NR RRC setup (①), the gNB opens the UE-associated NG connection to the AMF with the Initial UE Message (②), the CU builds the UE context at the DU over F1 (③), and the AMF's PDU Session Resource Setup completes the data path (④).

RACH → ① RRC 5213 → ② NG 5223 → ③ F1 5136/5137 → ④ PDU 5222peg points

a break at any stage lowers the session product③ F1 has two views: CU-side 5136 and DU-side 5137 — comparing them localizes CU-vs-DU faults.

Read the counter name — the SA failure-domain decoder

Every SA failure counter ends in a suffix that names the owning element. Six suffixes cover the whole chain:

Suffix → domain → first place to look

Suffix	Domain	Meaning	First place to look
`_CpFail` / `_Others`	gNB-CU-CP	Internal failure of the CU-CP blocks	CU-CP health, gNB load
`_RrcTimeout`	Air · UE	RRC message timeout — the UE never answered	UL/DL coverage, beam/RACH health (5243)
`_NgFail` / `_NgTimeout`	AMF · NG	Failure or timeout from the AMF over NG	NG-C transport, AMF capacity
`_E1Fail` / `_E1Timeout`	gNB-CU-UP · E1	Failure/timeout from the CU-UP or the E1 link	E1 transport, CU-UP health
`_F1Fail` / `_F1Timeout` / `_DUFailure` / `_DUTimeout`	gNB-DU · F1	Failure/timeout from the DU or the F1 link	F1 / fronthaul, DU health
`_MacFail` / `_RlcFail` (+timeouts)	DU lower layers	MacUeSetup / RlcUeSetup failed inside the DU	DU channel-card / lower-layer health
`Reject_CacFail` / `CacFail_*`	Admission	CAC rejected (call/DRB count, QoS-based)	Families 5147/5148 — capacity & policy

The SA superpower — the same suffixes repeat across all four families. _E1Fail in both NG ② and PDU ④ = one CU-UP/E1 event; _RrcTimeout in ① and ④ = the UE keeps dropping at the air; _NgFail only in ④ = the AMF accepts the UE but rejects the session (subscription / slice / UPF side). Cross-family patterns turn four counters into one diagnosis.

KPI ① — NR RRC Connection Establishment (Family 5213 · RRC_ESTAB)

The KPI. ConnEstabSucc ÷ ConnEstabAtt × 100. The message evidence lives in Family 5226 RRC_MSG (Rx_RRCSetupRequest → Tx_RRCSetup → Rx_RRCSetupComplete, rejects in Tx_RRCReject); Family 5214 times the setup; Family 5203 RRC_REESTAB covers recovery.

①-a · Counters (verbatim descriptions)

Counter	Domain	Verbatim description
ConnEstabAtt	attempt	Number of RRC connection setup attempts.
ConnEstabSucc	success	Number of successful RRC connection setups.
ConnEstabFail_CpFail	CU-CP	Setup failures because of an internal failure of the gNB-CU-CP blocks.
ConnEstabFail_RrcTimeout	Air · UE	Setup failures because of RRC message timeout — SetupComplete never arrived.
ConnEstabReject_CpFail	CU-CP	gNB transmitted RRC Reject because of an internal CU-CP condition.
ConnEstabReject_CacFail	Admission	gNB transmitted RRC Reject because CAC rejected the call.

①-b · Optimization playbook

Dominant counter	Diagnosis	Technique / lever
`ConnEstabFail_RrcTimeout`	UE unreachable after Setup — coverage/beam	Per-beam RACH first (Family 5243: `SuccessRateCBRAPerBeam` — a weak beam shows here); then NR RRC timers (`t300` ms100…ms2000, default ms200) and RACH power levers (`preamble-trans-max` n3…n200 default n4, `power-ramping-step` dB0/2/4/6 default dB2)
`ConnEstabReject_CacFail`	Admission ceiling	Families 5147/5148: identify which cap fired (`CacFail_CellCallCount / GnbCallCount / …DrbCount / QosBasedCount`) and raise the matching limit or offload — see the CAC section below
`ConnEstabFail_CpFail / Reject_CpFail`	CU-CP internal	CU-CP health & alarms; if load-driven, gNB capacity — not a radio parameter

KPI ② — UE-associated logical NG-Connection (Family 5223 · NG_ESTAB)

The KPI. NgConnEstabSucc ÷ NgConnEstabAtt × 100 — the SA analogue of LTE's S1SIG: "statistics for logical NG connections" — attempt pegs with the Initial UE Message to the AMF (evidence: Tx_NGInitialUEMessage, Rx_NGInitialContextSetupRequest, Tx_NGInitialContextSetupResponse/Failure in Family 5225). This factor isolates "is the AMF answering the gNB?"

②-a · Counters (verbatim descriptions)

Counter	Domain	Verbatim description
NgConnEstabAtt	attempt	Number of UE-associated logical NG-connection setup attempts.
NgConnEstabSucc	success	Number of successful UE-associated logical NG-connection setups.
NgConnEstabFail_NgTimeout	AMF · NG	Failures because of NG message timeout — the AMF never answered.
NgConnEstabFail_NgFail	AMF · NG	Failures because of a failure from the AMF over NG.
NgConnEstabFail_CpFail	CU-CP	Failures because of an internal CU-CP failure.
NgConnEstabFail_E1Fail	CU-UP · E1	Failures because of a failure from the gNB-CU-UP / E1.
NgConnEstabFail_F1Fail	DU · F1	Failures because of a failure from the gNB-DU / F1.

Playbook — NgTimeout dominant = AMF slow/silent → NG-C transport (SCTP) + AMF capacity/dimensioning; NgFail dominant = the AMF is actively failing the request → decode the NGAP cause (subscription, slice, security); Tx_NGInitialContextSetupFailure in Family 5225 gives the raw evidence. RRC ① healthy + NG ② degraded = pure core problem — no radio change will move it. Per FGR-CC3101 (SVR25A): on Error Indication with "Unknown local UE NGAP ID / Inconsistent remote UE NGAP ID" the gNB performs a local release — a rising 5391 SETUP_UECONTEXT_RELEASE alongside points at NGAP-ID inconsistencies.

KPI ③ — F1 UE Context Setup (Families 5136 CU-side · 5137 DU-side)

The KPI. CU_F1UeSetupSucc ÷ CU_F1UeSetupAtt × 100. Unique to the split gNB — no LTE equivalent. The same procedure is counted from both ends: the CU's view (5136) and the DU's view (5137, including the lower-layer causes). Comparing them localizes the fault to the link or the DU internals.

③-a · Both views (verbatim descriptions)

Counter	View	Verbatim description
CU_F1UeSetupAtt	CU	Number of F1 UE Context Setup Request messages transmitted to the gNB-DU.
CU_F1UeSetupSucc	CU	Number of F1 UE Context Setup Response messages received from the gNB-DU.
CU_F1UeSetupFail_DUFailure	CU	Failures because an F1 UE Context Setup Failure was received from the DU.
CU_F1UeSetupFail_DUTimeout	CU	Failures because of a timeout waiting for the DU.
DU_F1UeSetupFail_MacFail	DU	Failures because a MacUeSetup failed inside the DU.
DU_F1UeSetupFail_RlcFail	DU	Failures because an RlcUeSetup failed inside the DU.
DU_F1UeSetupFail_MacTimeout / _RlcTimeout	DU	Failures because of a timeout of the MAC / RLC setup inside the DU.

≡

The CU-vs-DU triangulation — CU_…DUTimeout high but the DU-side family clean → the F1/fronthaul link is eating messages. DU_…MacFail/RlcFail high → the DU itself (channel card / lower layers) is failing — same treatment as LTE's Up* counters: board health, not parameters. This two-view comparison is exactly why Samsung counts the procedure twice.

KPI ④ — PDU Session Resource Setup (Family 5222 · + 5438 per PLMN·S-NSSAI)

The KPI. PduSetupSucc ÷ PduSetupAtt × 100. The richest failure split of the chain — seven causes spanning every element. Family 5438 repeats it per PLMN and per S-NSSAI: the slice accessibility view.

④-a · The 7 failure causes (verbatim)

Counter	Domain	Verbatim description
PduSetupFail_NgFail	AMF · NG	Failures because of a failure from the AMF over NG.
PduSetupFail_E1Fail / _E1Timeout	CU-UP · E1	Failures because of a failure / message timeout from the gNB-CU-UP over E1.
PduSetupFail_F1Fail / _F1Timeout	DU · F1	Failures because of a failure / message timeout from the gNB-DU over F1.
PduSetupFail_RrcTimeout	Air · UE	Failures because of RRC message timeout — the UE lost at reconfiguration.
PduSetupFail_CpFail	CU-CP	Failures because of an internal CU-CP failure.

Playbook — read ④ per S-NSSAI (Family 5438) before acting: one slice failing = slice config / UPF for that slice; all slices failing with _E1Fail = CU-UP/E1; _NgFail only = AMF-side rejection (per FGR-CC3101, the gNB also answers a failure to the AMF when it cannot process the Setup — check both directions). VoNR: watch the SVR25A-added F1 DRB-setup failure counters for "Not supported QCI value" on 5QI-1.

Stage 0 — per-beam RACH (Family 5243) & admission (Families 5147/5148)

Before RRC there is the beam: NR RACH is counted per cell per beam — a single bad SSB beam hides inside a healthy cell average. And before the procedure there is admission: the CAC families tell you exactly which cap rejected.

RACH per beam (5243, verbatim) — the Msg1→Msg5 ladder

Counter	Verbatim description
RachPreambleAPerBeam	Detected preambles for CBRA per cell and per beam (CFRA: `RachPreambleACFRAPerBeam`).
NumRARPerBeam	Scheduled RARs per cell/beam corresponding to Msg1.
NumMSG3PerBeam / NumMSG5PerBeam	Successfully decoded Msg3 / Msg5 per cell and per beam.
SuccessRateCBRAPerBeam / SuccessRateCFRAPerBeam	The per-beam success rate of CBRA / CFRA.

CAC — which cap fired (5147 DU / 5148 CU-CP, verbatim)

Counter	Scope	What was exhausted
CacFail_CellCallCount / _SystemCallCount / _ChannelCardCallCount	DU (5147)	Call-count cap per cell / system / channel card
CacFail_CellDrbCount / _SystemDrbCount / _ChannelCardDrbCount	DU (5147)	DRB-count cap per cell / system / channel card
CacFail_QosBasedCount	DU (5147)	QoS-based admission
CacFail_GnbCallCount / _AcpfCallCount	CU-CP (5148)	Call-count cap per gNB / per ACPF

≡

Golden RACH/timer levers — doc-exact defaults (Samsung DU Parameter Description, prach-config / ue-timer-constants) — preamble-trans-max n3…n200 default n10 · power-ramping-step dB0/2/4/6 default dB2 · ra-contention-resolution-timer sf8…sf64 default sf64 · msg1-fdm n1/n2 default n1 · ssb-per-rach-occasion-choice …four default one · cb-preambles-per-ssb 1…64 default 56 · zero-correlation-zone-config 0…15 default 12 · RRC timers: t300/t301 default ms200, t304 ms1000, t310 ms2000, t311 ms5000, t319 ms600, n310 n10, n311 n1. Tune the beam→RO mapping where one beam's SuccessRateCBRAPerBeam lags the cell. Full 50+ parameter master table below ↓

SA Optimization Studio — four stages → grade → weakest link → fix

Enter the four families' counters; the studio computes each stage SR and the session product, grades against targets, finds the weakest stage, and the diagnostic engine below ranks every failure counter into an action plan.

SA Optimization Studiolive

Stage counters (Att / Succ)

ConnEstabAtt ①

ConnEstabSucc ①

NgConnEstabAtt ②

NgConnEstabSucc ②

CU_F1UeSetupAtt ③

CU_F1UeSetupSucc ③

PduSetupAtt ④

PduSetupSucc ④

Top failure counters

…Fail_RrcTimeout ①

…Reject_CacFail ①

…Fail_NgTimeout+NgFail ②

…Fail_DUTimeout+DUFailure ③

PduSetupFail_E1Fail+Timeout ④

PduSetupFail_NgFail ④

Session accessibility (① × ② × ③ × ④)

—%

Recommended action

Diagnostic engine — failure Pareto & ranked action plan

Failure Pareto — who is eating the session SRcomputed live from the studio counters

Ranked action plan — fix in this order

The SA failure-routing fabric — domain × stage

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

SA Failure-Routing Fabric

Scenarios

The SA feature pipeline — every accessibility feature (SVR25B)

An SA UE passes four gates — access control → interfaces → procedure → admission. Each node is a real Samsung FGR feature; click any node for its full card: what it is, benefit, how it works, recommended parameters, technique and counters.

SA Accessibility Feature Pipelineclick a feature node to open its detail card

Order matters — a request passes UAC/barring → RACH → RRC/CAC → NG → F1 → PDU. Rule out the deliberate gates (UAC category restrictions? cell reserved? redcap barred? shutting-down?) before tuning capacity — rejected-by-design is not a fault.

The SA accessibility parameter master table — 50+ levers (doc-exact defaults)

Every accessibility-relevant parameter, grouped by function — MO path, range and default verbatim from the Samsung CU ACPF Parameter Description (SVR25A) and DU Parameter Description (SVR22B). The Optimization guidance column tells you when and which way to move each lever.

A · UE RRC timers & constants — [DU] ue-timer-constants / ue-timer-const-info (9)

Parameter	Range	Default	Optimization guidance
t300	ms100 … ms2000	ms200	RRC setup supervision at the UE — lengthen only where RACH→Setup RTT is genuinely slow; shorter = faster retry
t301	ms100 … ms2000	ms200	Re-establishment request timer — pair with t311 when tuning recovery
t304	… ms10000	ms1000	Reconfig-with-sync (HO/PSCell) completion — widen on large cells / marginal targets
t310	… ms6000	ms2000	Out-of-sync → RLF declaration; shorter = faster recovery but drops recoverable links (drives 5085-style RLF)
t311	… ms30000	ms5000	Re-establishment cell-search window — widen where neighbour search is slow (deep indoor)
t319	… ms2000	ms600	RRCResumeRequest supervision (INACTIVE) — tune with the RRC-Inactive feature (group D)
n310	… n20	n10	Consecutive out-of-sync before t310 starts — lower = more sensitive RLF detection
n311	… n10	n1	In-sync count that cancels t310 — raise to require a steadier recovery
rrc-setup / rrc-resume / rrc-reestablish (gNB · rrc-sa-timer-profile)	10 … 65535 ms	1000 / 3000 / 2000	gNB-side supervision of the three procedures — the levers behind `ConnEstabFail_RrcTimeout`; shorten for faster failure detection once RF is clean

B · RACH / PRACH — [DU] ul-physical-resource-config/prach-config (8)

Parameter	Range	Default	Optimization guidance
preamble-trans-max	n3 … n200	n10	Raise where access is coverage-limited (weak-beam cells); more attempts before the UE gives up
power-ramping-step	dB0/2/4/6	dB2	dB4 converges faster at the edge at the cost of interference; keep dB2 in dense grids
ra-contention-resolution-timer	sf8 … sf64	sf64	Msg4 window — shorten (sf48) for faster failure/retry on low-latency networks
msg1-fdm	n1 / n2	n1	n2 doubles frequency-domain RACH occasions — use under PRACH congestion
prach-configuration-index	0 … 255	(plan)	PRACH format/periodicity (TS 38.331 §6.3.2) — denser index = more RACH opportunities; cell-lock needed to change
ssb-per-rach-occasion-choice	… four	one	Beam→RO mapping density — match to the SSB beam count; wrong mapping starves specific beams (check 5243 per beam)
cb-preambles-per-ssb	1 … 64	56	Contention preambles per SSB — lower it only when reserving CFRA preambles; too low = collisions
zero-correlation-zone-config	0 … 15	12	Preamble cyclic-shift spacing — must match cell radius; wrong value = ghost/missed preambles

C · Admission control (CAC) — [ACPF] acpf-cac / gnb-cac · [DU] qos-based-cac / cell-slice-group-cac (7)

Parameter	Range	Default	Optimization guidance
call-count-cac-usage (ACPF & gNB)	not-use / use	use	Keep on — the master call-count gate behind `CacFail_AcpfCallCount / GnbCallCount` (Fam 5148)
call-cac-threshold	0.00 … 100.00	100.00	Lower to 80–90% to intervene before hard overload — same early-intervention logic as the LTE guide's 90→85% move
qos-based-cac-usage	not-use / use	use	PRB-aware GBR admission — keep on where GBR/VoNR runs; behind `CacFail_QosBasedCount` (5147)
dl-normal-threshold (per QCI)	0.00 … 100.00	100.00	DL GBR PRB budget — lower per 5QI to protect voice from best-effort GBR load
ul-normal-threshold (per QCI)	0.00 … 100.00	100.00	UL twin of the above — UL is usually the tighter budget
slice-cac-usage	not-use / use	not-use	Enable when slicing is sold — per-slice-group admission (FGR-CC0303); read results per S-NSSAI in Fam 5438
slice-group-0…15-portion	0.00 … 100.00	0.00	Per-slice-group resource portion — size to each slice's SLA before enabling slice CAC

D · RRC Inactive & state control — [ACPF] rrc-inactive-info / release-preference / inactivity-data (9)

Parameter	Range	Default	Optimization guidance
feature-enabled (rrc-inactive-info)	disable / enable	disable	Enable to use RRC_INACTIVE (FGR-CC0103) — cuts attach/service-request latency by keeping the UE context in RAN; costs gNB memory
rrc-inactive-timer	1 … 7200 s	10	CONNECTED→INACTIVE after silence — shorter frees radio faster; too short churns resume signalling
rrc-idle-timer	1 … 7200 s	10	INACTIVE→IDLE — longer keeps the fast-resume window open (better re-accessibility)
resume-wait-timer	1 … 10 s	3	Wait for RRCResumeRequest after RAN paging before releasing the inactive context
notification-level	ue-activity / drb-activity	ue-activity	Granularity of activity notification driving the state machine
release-preference-on-off	off / on	off	On honours UE-preferred release (FGR-CC0108, UEAssistanceInformation) — battery-friendly, fewer stale CONNECTED UEs
release-preference-prohibit-timer	… infinity	S5	Prohibits repeat preference reports — protects the gNB from assistance-info storms
ue-inactivity-timer	1 … 7200 s	10	UE-level inactivity to release (FGR-CC0504) — coordinate with rrc-inactive-timer so states cascade sensibly
drb-inactivity-timer	1 … 7200 s	10	Per-DRB inactivity — frees individual bearers without dropping the UE

E · Preemption & emergency — [DU] cell-cac-info / call-preemption-priority · [ACPF] emergency-arp-info (7)

Parameter	Range	Default	Optimization guidance
call-preemption-usage	not-use / use	use	Keep on so ARP means something under congestion (FGR-CC0401)
drb-preemption-usage	not-use / use	not-use	Enable to allow bearer-level preemption — finer than dropping whole calls
preemption-trigger-arp-priority	1 … 15	3	ARP at/above which a new call may preempt — lower value = more aggressive protection of priority users
preemption-with-redirection	not-use / use	not-use	Enable so the victim is redirected instead of dropped — protects retainability while preemption protects accessibility
emergency-priority	1 … 16	1	Standalone preemption priority for emergency — keep at the top
emergency-arp-value	1 … 15	2	ARP assigned to emergency calls (FGR-CC0601) — must beat `preemption-trigger-arp-priority`
ecall-over-ims-support	off / on	—	Broadcasts eCall-over-IMS support — regulatory; verify per market

F · Overload protection — [ACPF] overload-protect-info (6) · FGR-LC0303

Parameter	Range	Default	Optimization guidance
call-protect-usage	not-use / use	not-use	Enable — caps RRCSetupRequest/ResumeRequest + SgNB-Add per unit time so a signalling storm can't topple the CPU
paging-protect-usage	not-use / use	not-use	Enable — pairs with the paging thresholds; denies land in Fams 5180/5185
monitoring-period	1 … 15 s	2	Window over which the request rate is measured — shorter reacts faster, burstier
threshold-for-paging	0 … 65535	2400	Pagings per window before shedding — size to the AMF paging profile
scaling-factor-for-normal	0.00 … 2.00	1.00	Multiplier at Normal CPU state — with the minor/major/critical factors it shapes the graceful-degradation curve
scaling-factor-for-minor/major/critical	0.00 … 2.00	(per level)	Progressively stricter shedding as CPU load rises — tighten critical first

G · Access, capability, security & interfaces (9)

Parameter	Range	Default	Optimization guidance
cell-barred ([DU] cell-access-info)	barred / not-barred	not-barred	Deliberate gate (FGR-CC3102) — commissioning/tests only; SVR25A: calls accepted until "barred" actually broadcasts in MIB
ue-capability-enquiry-3step	off / on	off	3-step UECapabilityEnquiry (FGR-CC3103) — enable where large capability containers slow setup
cipher-preferred-algorithm	nea0 … nea2	nea2	AS ciphering preference (FGR-BC0323) — align with the AMF list to avoid security-mode failures
integrity-preferred-algorithm	nia0 … nia2	nia2	AS integrity preference — mismatches surface as setup failures at security mode
sctp-association-heartbeat-interval	100 … 60000 ms	1500	E1/NG/F1 SCTP heartbeat — faster detection of dead links (feeds the *Timeout counters); don't set so low it false-triggers
e1-interface-delete-time	1 … 1440 min	5	Grace before deleting a failed E1 — longer rides out flaps without tearing bearer contexts
f1-interface-delete-time	10 … 1440 min	0	Same for F1 — 0 = immediate; consider a grace period on flappy fronthaul
rrc-inactive-info-redcap · rrc-inactive-timer	1 … 7200 s	10	RedCap twin of the Inactive timer (FGR-CC2000/CC0103-SVR25A) — tune separately for IoT profiles

Method, not magic numbers — defaults above are verbatim from the SVR25A ACPF / SVR22B DU Parameter Descriptions (your release template may differ — always RTRV first). Move one lever per soak, verify on the owning family's counters, and guard with 5391/5203 (releases/re-establishments) so a "gain" doesn't ship its failure downstream.

The 7-step SA accessibility optimization MOP

Baseline — pull all four families per cell

Collect Families 5213 / 5223 / 5136(+5137) / 5222 for the busy hour; compute the four stage SRs and the session product. Add 5243 per-beam RACH and 5147/5148 CAC.

RRC_ESTAB · NG_ESTAB · F1AP_UE_SETUP · PDU_SESSION_RESOURCE_SETUP Skip this and the cell average hides the one bad beam / slice.

Find the weakest stage

The lowest of the four SRs owns the loss. ① = radio/CU-CP/CAC · ② = AMF/NG · ③ = DU/F1 · ④ = the element its dominant suffix names.

stage split — the first cut of every SA investigation If wrong: teams chase the wrong element for weeks.

Rank the failure counters — the Pareto

Within the losing stage, rank the failure counters (the engine above does this). The suffix names the owner: _RrcTimeout=air · _NgFail/NgTimeout=AMF · _E1*=CU-UP · _F1*/_DU*/_Mac*/_Rlc*=DU · _CpFail=CU-CP · CacFail=admission.

suffix = domain = team If wrong: fixing the #3 counter first wastes the window.

Rule out deliberate gates

UAC category restrictions active? Cell reserved / barred (Access Identity 11/15 exempt)? RedCap barring on? Cell shutting down (accepts calls until 'barred' hits the MIB — SVR25A behaviour)? CAC caps as designed?

FGR-CC3102 UAC/barring · Families 5147/5148 If wrong: you "fix" RF on a cell that is barred by policy.

Apply ONE lever

Air loss → per-beam RACH (5243) + preamble-trans-max / power-ramping-step / ssb-per-rach-occasion + t300. CAC loss → the specific fired cap. Transport loss → the owning link (NG/E1/F1). Core loss → AMF with the 5225 message evidence.

one change per soak — attribute the gain If wrong: two levers at once = unattributable results.

Soak & verify — same counters, same hour

Re-measure all four families over the same busy hour: the target counter must fall, its stage SR must rise, and the other three stages must hold. Check 5391 SETUP_UECONTEXT_RELEASE and 5203 RRC_REESTAB didn't rise — a fix that ships releases/re-establishments is a net loss.

before/after on 5213/5223/5136/5222 + guards 5391/5203 If wrong: a stage gain that moves the failure downstream fools the KPI.

Lock in — slice & VoNR view

Re-read ④ per PLMN·S-NSSAI (Family 5438) so no slice regressed, and watch the SVR25A F1 DRB-setup failures for "Not supported QCI value" on 5QI-1 (VoNR). Document the change.

Fam 5438 per-slice · 5QI-1 DRB failures If wrong: the aggregate looks fixed while one slice (or voice) is broken.

SA is the full chain — four families, one product, one diagnosis.

RRC 5213 × NG 5223 × F1 5136 × PDU 5222 — decompose, decode the suffix, fix the owning element. Compare with the LTE three-factor chain and the NSA SgNB-Addition split to master all three generations.

LTE Accessibility ›5G NSA Accessibility ›

Sources: Samsung 5G NR Counter Description (CU-CP ACPF + DU, SVR25A/B — Families 5213 RRC_ESTAB, 5223 NG_ESTAB, 5224, 5136/5137 F1AP_UE_SETUP, 5222/5438 PDU_SESSION_RESOURCE_SETUP, 5147/5148 ADMISSION_CONTROL_FAIL, 5243 NR_PRACH_BEAM, 5225/5226 message families, 5391, 5203 — counter names & descriptions verbatim, via the operator NR KPI workbook) · 5G RAN Feature Description SVR25B (FGR-CC3101 Basic Operation for NR SA, FGR-CC3102 Access control for SA incl. UAC & RedCap barring, FGR-CC1104/1105/1106 F1/E1/NG interface management) · NR golden parameters from the Samsung DU Parameter Description. SA procedures per 3GPP TS 38.331 / 38.413 / 38.473. Targets are typical operator values; counters, families and mechanics are vendor-exact.