5G NSA Throughput — Two Radios, One PDCP, and the Split That Decides Everything

EN-DC throughput is addition with a referee: one PDCP entity at the SgNB splits each bearer's bytes between the NR leg and the LTE leg, and Samsung counts the split verbatim — Family 5126: PdcpSduVolumeSplitNrDL / PdcpSduVolumeSplitLteDL / PdcpSduVolumeNonSplitNrDL (+ UL twins and averages). The user's rate is 5016 DLUeThru (burst method: volume ÷ active time), the UL per EN-DC cell type in 909, and the air delay in 5193. When NSA throughput disappoints, the split shares tell you which leg — or the referee — to fix. Every name verbatim.

5126split volume anatomy

5016DLUeThru burst method

909UL IP thpt per NR type

5193air delay (NSA)

5124data forwarding NSA

943X2 link time

The NSA throughput KPI set — the burst method and the split shares

KPIs (counters verbatim)

KPI	Formula (Samsung counters)	Read it as
NSA DL UE throughput	`DLUeThruVol ÷ DLUeThruTime` (Fam 5016 — the triple `DLUeThru/Vol/Time`)	The burst method: bytes while transmitting ÷ time spent transmitting — idle time excluded, the honest per-UE rate
NR-leg share (DL)	`(PdcpSduVolumeSplitNrDL + PdcpSduVolumeNonSplitNrDL) ÷ PdcpSduVolumeDL × 100` (Fam 5126)	How much of the bearer actually rode 5G
Split efficiency	`PdcpSduVolumeSplitLteDL` vs `SplitNrDL` within split bearers	The referee's decision quality — LTE-heavy splits under good NR = flow-control/X2 problem
UL path truth	Fam `909 UL_IP_THROUGHPUT_ENDC` — UL IP throughput per EN-DC NR cell type + `PdcpSduVolume*UL` split twins	UL usually rides one leg — which one, and at what rate
Air delay (NSA)	`AirIfDelayTot ÷ AirIfDelayCnt` (Fam 5193)	The latency the split adds — compare against the SA twin (5194)
Session PS volume	Operator OMC report: 5G Data Volume [GB], DL/UL MB per cell	The commercial truth the shares must explain

≡

Why the burst method matters — DLUeThruVol ÷ DLUeThruTime ignores idle gaps, so it measures what a user feels during a download, not cell utilisation. A cell can be nearly idle and still show poor UE throughput — that's a per-UE problem (split, MCS, leg quality), not capacity. The LTE page's effective-vs-average lesson, upgraded.

The split-bearer anatomy — one PDCP, two legs, three volume buckets

S1-U → PDCP@SgNB → {NR RLC/MAC · X2-U → LTE RLC/MAC} → UE recombinesFam 5126 buckets

the LTE leg pays an X2 tax on every bytea slow X2 makes the referee LTE-shy — or floods the UE's reordering window. Both show in the 5126 shares.

Reading Family 5126 — the four signatures

Split-share signatures → diagnosis → fix

Signature (5126 shares)	Diagnosis	Fix surface
NonSplitNr dominant, healthy rate	Flow control keeps bearers on NR alone — the ideal when NR covers well	Nothing — this is the target state
SplitLte heavy while NR CQI is good	The referee distrusts the NR leg it shouldn't — X2-U flow-control feedback or NR BLER misread	X2 latency/capacity (Fam 943 link time), flow-control config, NR BLER targets (SA page values: dl 8%)
Split volumes low, NonSplit low too	Legs aren't being added or bearers aren't SCG/split type	Accessibility page (SgNB addition), bearer-type policy (719 per endc_bearerType — retainability page)
UL on LTE while NR UL configured	UL path selection conservative — power-split or UL-outage history	Fam 909 per NR cell type; the en-dc UL machinery (retainability page); UE TX power sharing

The X2 tax is measurable — Family 943 ENDC_X2_LINK_TIME times the X2 the split rides. Rising link time + falling SplitLte share = the flow control is correctly retreating from a degrading X2. Fix the transport; don't force the split.

The two legs' own evidence — no overlap, all cross-referenced

Where each leg's depth lives

Leg	Evidence here (NSA-specific)	Full depth
NR leg	`5016 DLUeThru` triple · `5193` air delay · DU PRB/MCS/CQI/HARQ for the NSA cells	The SA throughput page owns the NR radio levers (PRB playbook, MCS/BLER, CA, MIMO) — they apply to NSA cells identically
LTE leg	`PdcpSduVolumeSplitLte*` — what the anchor carries for 5G	The LTE throughput page owns the ladder and levers; remember the anchor also carries its own LTE traffic — the split adds load the LTE page's PRB bands must absorb
The referee	`5126` shares · `943` X2 time · `5124 DATA_FW_NSA` (forwarding traffic at mobility)	This page

Interactive — the split-bearer flow visualizer

Drag the leg conditions and watch the referee's split — and what the user actually gets. The model follows the 5126 accounting: total = SplitNr + SplitLte + NonSplitNr.

Split-Bearer Flow Visualizerlive

NR leg capacity: 300 Mbps

LTE leg spare capacity: 80 Mbps

X2-U health (latency/loss): good

Demand (UE bearer): 400 Mbps

User gets (recombined)

— Mbps

The EN-DC throughput levers (sourced set)

NSA-specific levers (radio levers → LTE/SA pages)

Lever	Evidence counter	Guidance
scg-band-selection-mode + thresholds	SgNB addition per band (5103)	LB deck verbatim: capacity mode, 70–80% — "better NR leg performance = better overall EN-DC throughput"
anchor-selection (idle) + cell-capacity-class per FA	Fam 793	An overloaded anchor throttles the LTE leg of every split bearer — balance anchors first (mobility page)
X2-U transport (capacity/latency)	`943 ENDC_X2_LINK_TIME`, GTP fw (Fam 39)	The split's ceiling; every SplitLte byte crosses it twice (data + feedback)
en-dc-ul-radio-link-outage/resume set	UL split shares (5126), 909	Owns UL-path confidence — full set on the retainability page
NR leg radio (BLER targets, MCS caps, CA, MIMO)	DU families 5018/5021/5019…	The SA throughput page's tables apply verbatim to NSA NR cells
LTE anchor health (PRB bands, scheduler)	LTE 54/55 + ladder	The LTE page's MOP — remember the split adds invisible load to the anchor

The 5-step NSA throughput MOP

Read the burst-method truth

5016 triple per cell (+ per PLMN 5482); the OMC volume report. Idle-but-slow = per-UE problem; busy-and-slow = capacity on one of the legs.

DLUeThruVol ÷ DLUeThruTime Skip this and you chase utilisation instead of experience.

Read the 5126 shares

NonSplitNr / SplitNr / SplitLte, DL and UL. Match against the four signatures — the shares name the suspect: referee, X2, NR leg, or LTE leg.

the split anatomy If wrong: NR radio tuning for an X2 transport problem.

Check the X2 tax

943 link time trend + GTP forwarding evidence. Rising tax = transport work; the flow control retreating from LTE is correct behaviour, not a bug.

943 + Fam 39 GTP If wrong: forcing splits over a dying X2.

Fix the guilty leg with its own page

NR leg → SA throughput page (PRB playbook, MCS/BLER, CA/MIMO). LTE leg → LTE ladder + PRB bands (with the split load added). UL → power-split + outage machinery.

no duplicate tuning — one owner per lever If wrong: two pages' levers moved at once.

Verify the recombined rate + guards

5016 up; 5193 delay not worse (reordering!); anchor LTE users unharmed (their PRB view); SgNB addition/retention untouched.

user rate + air delay + anchor guard If wrong: a split "gain" that doubles jitter.

NSA throughput is two legs, one referee, and a measurable tax.

The 5126 shares name the suspect; each leg has its own page of levers; the X2 tax is a counter, not a guess.

LTE Throughput ›5G SA Throughput ›

Sources: Samsung 5G NR Counter Description (CU-UP AUPF Family 5126 PDCP_VOLUME_PER_GNB — the split-volume counters verbatim: PdcpSduVolumeSplitNrDL/SplitLteDL/NonSplitNrDL + UL and Avg twins; 5125/5195/5196 packet/per-5QI twins; 5124 DATA_FW_NSA; DU Families 5016 DL_UE_THROUGHPUT, 5193 PACKET_DELAY NSA) · Samsung LTE eNB Counter Description (909 UL_IP_THROUGHPUT_ENDC — "UL IP Throughput per EN-DC NR Cell type" verbatim; 943 ENDC_X2_LINK_TIME; 39 GTP forwarding) · LB-deck EN-DC parameters (scg-band-selection, anchor selection — verbatim). The NR radio levers live on the SA throughput page; the LTE ladder on the LTE throughput page; UL-outage machinery on the NSA retainability page — cross-referenced, not duplicated. EN-DC user plane per 3GPP TS 37.340.