VoNR Control-Plane KPIs in 5G NR

Introduction

Voice over New Radio (VoNR) is native 5G voice: the call is carried as IP packets over the NR air interface, using the IMS for call control and a dedicated 5G QoS flow for the media. Nothing about it is circuit-switched — it is signalling stacked on signalling, and every step can succeed or fail on its own.

Because a voice call is a chain of independent events, operators never judge it with a single number. They watch a family of Key Performance Indicators (KPIs), each answering one narrow question about one phase of the call: can it start, does it stay up, does it survive movement, does it sound good. This page covers the first three — the control-plane KPIs, computed from signalling events at the gNB and the IMS core. The fourth, MOS, is a user-plane metric named only for contrast.

You meet these KPIs the moment a network turns on VoNR: they are what an optimisation engineer stares at on the OSS dashboard, what an interviewer asks you to define and decompose, and what tells you where voice is breaking rather than merely that it is. The measurement framework is TS 28.554, resting on the QoS model of TS 23.501 and the fallback procedures of TS 23.502.

Why voice KPIs are needed

A raw "voice works / voice is broken" verdict is useless to the engineer who has to fix it. The whole point of the KPI framework is localisation: decompose the call into measurable phases so a bad number points at one domain — the IMS core, the RAN admission-control path, the air interface, or the mobility/neighbour plan. Each KPI is deliberately a simple ratio of a success counter to an attempt counter (or a timed interval), so it can be sampled cheaply per cell, per hour, and compared across a network of thousands of cells.

They are needed for a second reason too: voice is the service subscribers judge a network by. A dropped call generates a complaint in a way a slow web page rarely does, and regulators and enterprise SLAs are written against exactly these accessibility, retainability and mobility figures. Getting the definitions right — which numerator, which denominator, which plane — is therefore not academic; it is what makes the dashboard trustworthy.

The KPI hierarchy for voice

Operators think about voice as a layered story, and every layer answers one blunt question. You climb the ladder in the order a real call experiences it, and — crucially — you diagnose in that same order, because a lower rung is meaningless until the rung beneath it is solid.

The funnel shape is deliberate. Every attempt that clears accessibility becomes an established call that retainability then judges; every established call that survives becomes a candidate for the mobility events that only happen mid-call. Each layer operates on the survivors of the layer above, which is why a headline number in isolation can lie: a cell with a superb drop rate might simply be establishing very few calls in the first place. Read the families together, in order, and the story becomes honest.

Accessibility — can the call start?

Accessibility groups every KPI between "user presses dial" and "the call is up with media flowing both ways". For VoNR that spans two domains that must both succeed: the IMS layer (SIP signalling in the core) and the RAN layer (creating the radio bearer that actually carries the voice media). A break in either domain shows up as an accessibility failure, so the family is deliberately decomposed into sub-KPIs that localise the fault.

IMS Registration Success Rate

Before any call can happen, the UE must register with the IMS using SIP, carried on the default IMS signalling flow (5QI=5, a non-GBR flow set up when the IMS PDU session is established). If registration fails, the subscriber simply cannot originate or terminate a VoNR call at all — so this is the true root of accessibility, sitting logically before every other rate on the page.

A depressed value here almost always points at the core or the signalling path — P-CSCF discovery, authentication (IMS-AKA), or the 5QI=5 flow not coming up — rather than at the radio serving an individual call. It is the first counter you check, because it caps everything downstream.

VoNR Call Setup Success Rate (CSSR)

CSSR is the headline accessibility KPI: of all voice call attempts by registered subscribers, how many reached a fully established, media-capable call? It is really the product of several sub-successes, and decomposing it is precisely how you find where setup breaks rather than merely that it broke.

Conceptually, CSSR ≈ P(5QI=1 flow setup) × P(RRCReconfiguration success) × P(SIP session established). A failure in any one factor drags the whole rate down, so operators track the two RAN-side factors as first-class KPIs in their own right. The decomposition below shows which domain owns each factor and how to read a failure there:

Dedicated 5QI=1 QoS-flow setup success

The make-or-break RAN step is establishing the dedicated GBR flow that will carry the voice media. This is triggered from the core: the SMF, on seeing the SIP session negotiate, requests the addition of the voice flow to the existing PDU session, which reaches the gNB as a PDU Session Resource Modify carrying the 5QI=1 QoS profile. The gNB must admit that GBR flow and map it to a data radio bearer.

A low value here is a RAN admission-control or resource story: the cell could not guarantee the GBR, or the bearer configuration failed. It is the single most diagnostic sub-KPI of CSSR, because the GBR flow is what distinguishes a real voice call from a signalling-only session.

RRCReconfiguration success

Admitting the flow is not enough; the gNB must actually reconfigure the UE over the air to add the voice DRB mapped to that 5QI=1 flow. It does this with an RRCReconfiguration message, and the UE must answer RRCReconfigurationComplete. If that handshake is lost — a missed message on a fading link, a UE that cannot apply the configuration — the flow is admitted in the core but never usable on air.

The 5QIs behind the counters

The two flows that gate accessibility are not arbitrary — they carry standardized 5QI characteristics defined in TS 23.501. Knowing the exact profile explains why each KPI behaves as it does: the media flow is delay-bounded and loss-tolerant, the signalling flow is loss-intolerant.

Read the profiles against the KPIs: 5QI=5 has a punishing 10−6 error target because a lost SIP message stalls the whole call — which is why IMS-registration and SIP-establishment failures show up as hard accessibility failures. 5QI=1 tolerates 10−2 loss (a codec can conceal the odd frame) but enforces a 100 ms delay budget and a guaranteed bit rate, so its setup is an admission-control question — can the cell promise the GBR? — which is exactly what the 5QI=1 setup-success KPI measures.

How the media flow is scheduled on air

Once the DRB exists, the voice media rides the shared channels: downlink frames on PDSCH, uplink frames on PUSCH, each scheduled by a DCI on PDCCH scrambled with the UE's C-RNTI in the UE-specific search space. Because AMR voice frames are small and strictly periodic (typically one packet every 20 ms), a fresh PDCCH grant per packet is wasteful, so many deployments use SPS (Semi-Persistent Scheduling) on the downlink and a Configured Grant on the uplink: the periodic allocation is activated once by a DCI scrambled with CS-RNTI and then recurs without further signalling. This matters for the KPIs — a mis-tuned SPS period, a lost activation, or PDCCH blocking under load inflates both setup time and the frame-loss the codec must conceal, which surfaces as poor MOS even while every control-plane counter looks green. The RRCReconfiguration that adds the DRB is itself carried on SRB1 over DCCH.

Call setup time — how long until it rings?

Success rates say whether a call sets up; setup-time KPIs say how long it takes. A call that succeeds but takes six seconds to ring feels broken to the user, so latency is measured alongside the rates and belongs to the same accessibility family.

Retainability — does it stay up?

Once a call is established, retainability asks whether it survives to a normal hang-up. The signature KPI is the drop rate: calls that ended abnormally — not because a user hung up — as a fraction of all established calls. This is the number that most directly maps to a subscriber complaint.

An "abnormal release" is any teardown the user did not ask for. The usual culprits, and what each tells you when it dominates, are worth memorising because the drop rate on its own only tells you that calls are failing:

• Radio Link Failure (RLF) — the radio link to the serving cell collapsed (coverage hole, interference, sudden fading) while the call was active. RLF-driven drops point to coverage or interference problems.
• Handover failure — the call was mid-handover and the target cell was never reached, so the context was lost. Points to mobility and neighbour-planning issues.
• QoS flow release — the 5QI=1 GBR flow was lost or pre-empted mid-call (for example the GBR could no longer be guaranteed under load). Points to congestion or admission-control problems.

The optimistic mirror image of the drop rate is the Voice Call Completion Rate — the share of established calls that reached a normal release. In a clean network the two are near-complements of each other, so a completion rate well below 100 − dropRate is a hint that some releases are being miscategorised.

Mobility & EPS Fallback

Voice is used while people move — walking, driving, riding a lift. Mobility KPIs measure whether an active call survives a cell change, and whether a call that cannot be served natively on NR is handed cleanly to LTE. This family only ever operates on already-established calls, so it sits below retainability in the funnel.

In-call handover success

For an active voice call, a failed handover usually means a dropped call, so handover success is tracked specifically for calls in progress — both intra-NR (NR cell to NR cell) and inter-RAT NR↔LTE. It is the mobility-side counterpart to the handover-failure component of the drop rate.

EPS Fallback Success Rate

Many networks do not run VoNR everywhere yet. When a voice call is attempted on 5G but the NR cell is not provisioned to serve it, the network invokes EPS Fallback: it moves the UE to LTE (by redirection or inter-RAT handover) and completes the call as VoLTE. This KPI measures how reliably that redirection itself works — a failed fallback is a failed call.

EPS Fallback Execution Time

Even a successful EPS Fallback costs time: the UE must be redirected or handed over to LTE, re-establish the bearer there, and only then continue the call. This KPI captures that added latency, measured from the fallback trigger to the point the LTE bearer is ready to carry voice.

Integrity — MOS, the user-plane counterpart

The fourth family, Integrity, answers the question none of the control-plane KPIs touch: once the call is up and staying up, does it actually sound good? Its signature metric is MOS (Mean Opinion Score), and it is named here purely so you know where it sits — it is out of scope for this page because it is a user-plane measurement, not a signalling one.

The distinction is sharp and worth internalising. Every KPI above is computed from control-plane events — a message sent, a response received, a flow admitted, a release classified. MOS, by contrast, is derived from the media itself: the audio quality perceived by the listener, estimated from packet-level behaviour of the voice stream (loss, jitter, delay, and the codec in use) or scored subjectively in testing. A call can post a perfect scorecard on IMS registration, CSSR, drop rate, and handover success and still deliver poor MOS if the media path is congested, the jitter buffer is starved, or the codec is transcoded down.

Putting it together — the KPI table

Here is the full set on one page, grouped by family. Remember the counter names are illustrative — your vendor's OSS will label them differently — but the numerator/denominator logic and the diagnostic meaning are standard across TS 28.554 deployments.

Read the table top to bottom and it retraces the funnel: registration gates CSSR, CSSR decomposes into the two RAN factors plus setup time, retainability judges the survivors, mobility judges the movers, and MOS sits apart on the user plane. A second small table makes the plane split explicit, because it is the distinction interviewers probe most often:

Summary

Voice on 5G is judged as a narrowing funnel of KPI families, diagnosed top-down because each layer operates only on the survivors of the one above. Accessibility asks whether a call can start — IMS Registration Success Rate, VoNR CSSR, and its two RAN factors, 5QI=1 GBR-flow setup (via PDU Session Resource Modify) and RRCReconfiguration success (adding the voice DRB on SRB1). Setup time lives in the same family and is split into an end-to-end SIP window (SIP INVITE→200 OK) and a RAN slice, so you can tell whether delay is core/SIP or radio — and spot a silent fallback.

Retainability asks whether an established call stays up to a normal release: the drop rate and its mirror, the completion rate, decomposed by abnormal-release cause (RLF, handover failure, 5QI=1 QoS-flow release). Mobility judges the movers — in-call handover success (intra-NR and inter-RAT) and, uniquely to VoNR, EPS Fallback success and execution time. Integrity (MOS) sits apart on the user plane, judging the media itself. The single most important discipline is to keep numerator, denominator and plane straight for every counter, and to read the families together and in order — a headline number in isolation lies.

Quick Q&A