CafeTeleRadio Optimization SRPO-600 Enroll now ↗
‹ Day 4 · Throughput Course / Day 4 / 4.1 / 5G SA Throughput
SA
5G
4.1 · 5G SA · Option 2

5G SA Throughput — One Playbook, Four Load Bands, Every Lever Named

The SA deck's masterpiece is a decision engine: read TotDLUsedPrbScheAvg, place the cell in one of four load bands, and each band hands you its own counters-to-check and its own verbatim parameter pack — from "SU-MIMO preferred, dl-target-bler 5–8%, pf-alpha-dl 0.9–1.0" at low load to "very aggressive MU-MIMO, bler 18%, pf-alpha 0.4, HARQ 3" under congestion. Beneath it: the burst-method UE rate (5192), the PRB anatomy (5041 — unicast vs broadcast vs SIB/paging), CQI/MCS/RI distributions, the CA action chain, the MIMO families and the FGR-RC/RS feature stack. Band → evidence → pack — every value verbatim.

5192UE thpt SA (burst)
5041PRB anatomy
5018·21CQI · MCS dist
5019HARQ trials + BLER
5008-11CA action chain
4load-band packs
The SA throughput KPI set — burst method + the PRB anatomy
KPIs (counters verbatim)
KPI / setCounters (verbatim)Read it as
SA UE throughputDLUeThru_SA = DLUeThruVol_SA ÷ DLUeThruTime_SA (Fam 5192; per PLMN·S-NSSAI: 5452) · UL: ULUeThru (5144)The burst method — what a user feels mid-download; slice view included
PRB anatomyTotDLUnicastPrbScheAvg · TotDLBroadcastPrbScheAvg · TotDLSibPrb · TotDLPagingPrb · TotDLCcchPrb/DcchPrb · TotDLAvailablePrb + UL twins (5041)Not all PRBs earn revenue — broadcast/SIB/paging overhead is measured separately; unicast-sched-avg is the load truth
Quality distributionsDLReceivedCQI0…15 (5018) · DLTransmittedMcs0…31 (5021) · MCS-BLER (5165/5166) · DLReceivedRi1/Ri2… (rank)The CQI→MCS→rank chain; gaps between them name the loss
HARQ endgameNrof1st…8thTxTrial_DL · NrofResidualError_DL · BLERof1st…TxTrial (5019/5055; VoNR twins 5402/5403)Deck bands: BLER <5% normal · 5–10% warn · >10% critical
CA action chainCaAddAtt/Succ/Fail/CaRel (5008) · act/deact (5009/5084) · CaAddTime (5010) · capability (5011) · CA_ACT_UE_NUM (5417) · SCell CQI (5418)The same capable→configured→activated→used audit as LTE
Air delayAirIfDelayTot_SA ÷ AirIfDelayCnt_SA + RLC SDU delay (5194)The latency twin of throughput; compare with the NSA variant (5193)
The PRB anatomy is an SA upgrade — LTE counted used-vs-available; NR counts what the PRBs did: unicast (revenue), broadcast/SSB, SIB, paging, CCCH/DCCH. A "70% loaded" cell whose overhead PRBs eat 15 points is a different problem from one at 70% unicast. Always read TotDLUnicastPrbScheAvg, not the total.
The four-band PRB playbook (deck-verbatim) — as a decision engine

The deck's DL workflow: PRB band → the analysis table for that band → the parameter pack. Try it live — set the evidence, get the band's verbatim pack:

PRB-Band Decision Enginelive
Playbook branch
The four packs, side by side (deck-verbatim values & impact claims)
Parameter<20% low load20–50% medium50–70% mod-high>90% congestion
dl-mu-su-mimo-modeSU-MIMO preferred (+15-20% per-user)Balanced (+10-15% overall)MU-MIMO preferred (+10-15% capacity)Very aggressive MU-MIMO (+15-20%)
dl-target-bler5–8%10%12%18% (reliability trade-off)
pf-alpha-dl0.9–1.0 (fairness)0.7–0.80.6–0.70.4 (favour top users)
dl-mu-mimo-entering-sinr-threshold15 dB13 dBaggressive
othersdl-la-step-size-db 1.0 dB · max-num-harq-transmission-dl 6snr-threshold-64q-to-256q 25→23 dBip-tput-enhanced-scheduling-enable Enabled (+5-10% TCP)max-num-harq-transmission-dl 3 · non-gbr-sch-mode-dl throughput-maximized
!
The packs point in opposite directions on purpose — low load optimizes the individual (conservative BLER, fairness, SU-MIMO); congestion optimizes the sector (aggressive BLER, capacity scheduling, MU-MIMO). Applying a congestion pack to an idle cell — or vice versa — makes both metrics worse. The band decides; that's the whole method.
The band analysis tables (deck-verbatim counters → parameters)
DL analysis by band (verbatim rows)
Band / questionCounter to checkCritical valueParameter → recommended
20–70% + low throughput
(RF quality)
DLReceivedCQIAvg< 7dl-target-bler → 5–8% (conservative)
DLTransmittedMcsAvg< 15snr-threshold-64q-to-256q → 25 dB (conservative)
PdschBlerMcs0-31> 20% any MCSdl-la-step-size-db → 1.0 dB
RF good, throughput low
(configuration)
SCellActSucc/SCellActFail< 80% successmax-dl-ca-cc-num → per UE capabilities; ca-scell-activation-deactivation-mode → more aggressive
DLReceivedRi1 vs Ri2mostly rank 1number-of-maximum-layer → review MIMO config
DLNumMuMimoUEs_A6below expecteddl-mu-mimo-entering-sinr-threshold → reduce 2–3 dB
>70% load
(scheduler)
RatioSlotwBlockDL> 10%pf-alpha-dl → 0.6–0.7
RatioBlockedPDCCHDL> 10%non-gbr-sch-mode-dl → throughput-optimized
UsedCCEperSlotRatio> 80%ip-tput-enhanced-scheduling-enable → Enabled
>90% congestionAirMacDLThruAvg vs PRBsaturateddl-mu-su-mimo-mode → aggressive MU-MIMO; pf-alpha-dl → 0.5
RatioBlockedPDCCHDL> 15%dl-target-bler → 15%; nr-bandwidth-dl → maximum available
The SA throughput feature stack — FGR-RC/RS catalogue, full detail (deck chapters)
SA Throughput Feature Pipelineclick a feature node to open its detail card
The 5-step SA throughput MOP (the deck workflow, ordered)
1
Burst-method truth + PRB anatomy

5192 triple (and 5452 per slice); 5041 unicast-vs-overhead split. Idle-but-slow = per-UE; busy-and-slow = the band playbook.

DLUeThru_SA + TotDLUnicastPrbScheAvg Skip this and overhead PRBs masquerade as user load.
2
Place the band, run its analysis table

<20 / 20–70 / >70 / >90 — each with its deck counters (CQI/MCS/RI · SCellAct · Blocked-PDCCH/CCE · saturation).

the decision engine above If wrong: a congestion pack on an interference problem.
3
Audit the multipliers

CA chain 5008-5011/5417 (capable→added→activated→used, SCell CQI 5418); MIMO rank (Ri1 vs Ri2, 5082/5022); 256QAM threshold; VoNR HARQ twins protected (5402/5403).

the multiplier audit If wrong: peak features idle while basics starve.
4
Apply the band's pack — one lever per soak

The four packs are starting points (deck's own impact claims noted). Move the pack's levers one at a time; UL has its own bands (>70/>90) and levers.

deck packs, one lever each If wrong: pack-mixing = unattributable results.
5
Verify on the burst rate + guards

5192 up; 5194 delay held; BLER inside the deck bands; retainability families untouched (aggressive BLER packs trade reliability — watch the drop pages' counters).

burst rate + air delay + Day-2 guards If wrong: throughput bought with drops is a refund waiting.

SA throughput is a band, a table, a pack — in that order.

The deck's decision engine turns "slow 5G" into a mechanical walk: band → evidence → verbatim pack. Compare with LTE's ladder and NSA's split to complete Day 4.1.

Sources: Samsung 5G NR Counter Description (DU + CU-UP — Families 5192/5452 UE_THROUGHPUT_SA, 5016/5482/5144 UE throughput, 5041 PRB_TOTAL (unicast/broadcast/SIB/paging anatomy), 5086/5145/5241/5411 PRB per QCI/5QI/PLMN/S-NSSAI, 5018 DL_CQI, 5021 DL_MCS_TRANSMISSION, 5165/5166 MCS-BLER, 5019/5055 HARQ trials, 5402/5403 VoNR HARQ, 5008/5009/5010/5011/5084/5167/5417/5418 CA set, 5082/5022/5057 MIMO, 5194 PACKET_DELAY_SA, 5126/5196 PDCP volumes — counters verbatim) · 5G NR Throughput Optimization SA deck (the four-band PRB playbook with analysis tables and parameter packs — all values and impact claims verbatim; feature chapters FGR-RC0202 DL SU-MIMO, RC0203 DL MU-MIMO, RC0402 DL CoMP, RC1000 TDD DL Massive MIMO, RS0252/0253 UL SU/MU-MIMO, RS0703 QoS Scheduling, RS0601 DL Link Adaptation, RS7003 DL Prescheduling). The TDD/VoNR/cell-edge scenario packs live on the SA retainability page; the split-bearer view on the NSA throughput page — cross-referenced, not duplicated.