VoNR field testing 2026: how to validate voice quality on 5G Standalone

A senior RF integrator drives the Sheikh Zayed Road on a Tuesday morning in March 2026. The handset is locked to 5G SA on a UAE commercial network. Every VoNR call placed at high speed survives the first sector, then collapses to LTE the moment the car crosses into the next gNB footprint. POLQA scores on the recorded audio look fine when the call stays on 5G, and barely acceptable the moment EPS Fallback kicks in. The customer wants to know whether voice on 5G SA is production-grade or whether the rollout should wait another quarter. The answer is not in a vendor dashboard. It is in the NAS and SIP traces sitting on the phone.

This is what VoNR field testing looks like in 2026.

VoNR is no longer a roadmap item: it is live in 2026

For five years, Voice over New Radio was treated as a “next year” topic. That is no longer true. In Q1 2026, Verizon and Ericsson announced the first nationwide VoNR roaming in the United States. In February 2026, du in the United Arab Emirates went live with the country’s first commercial VoNR launch. Bell Canada has been carrying VoNR on iPhone in the Greater Toronto Area since August 2025. T-Mobile US and Vodafone Germany have moved Voice over 5G into commercial production. The list is short but no longer experimental.

The macro picture confirms the shift. Per Opensignal’s February 2026 5G SA State of Play, around 17.6% of global users have access to a 5G Standalone network, and the GSA Standalone tracker reports 89 commercial SA networks launched, with 181 operators investing in SA across 73 countries. Standalone has tipped from a deck slide into a deployed reality.

The catch is that voice has lagged the architecture. SA radio works. SA core works. What still breaks, in the wild, is the IMS-on-5GC integration: the cell-by-cell behaviour of EPS Fallback, the codec negotiation between the UE and the IMS Application Server, and the QoS treatment of the voice flow once the PDU session is up. These are the things a customer experiences as “calls drop” or “voice sounds robotic”, and they are the reason VoNR validation in 2026 is a field testing problem, not a lab problem.

What changes between VoLTE and VoNR, and why your VoLTE test method will not work

VoLTE field methods do not transfer cleanly to VoNR. The signalling path, the QoS marker, and the codec defaults all change. Reusing a VoLTE drive-test plan on a VoNR rollout is the single most common reason teams miss the actual failure modes.

In VoLTE, the voice path runs over an LTE PDN connection terminated on the PGW. The IMS signalling rides on a default bearer; the voice itself rides on a dedicated bearer marked QCI=1, with AMR-WB as the default codec and EVS optional in profiles where the network and the device both support it. Service Request, RRC Reconfiguration, and SIP/SDP negotiation are observed through MME, S-GW, and P-GW touchpoints.

In VoNR, the entire transport pipeline shifts to the 5G Core. The voice flow is carried inside a PDU session anchored at the UPF, with QoS handled per QoS flow rather than per bearer. The marker becomes 5QI=1, attached to a dedicated QFI signalled through the SMF. The S-NSSAI is selected at registration (typically the eMBB slice or a dedicated voice slice when the operator has provisioned one), and EVS Super-Wideband becomes the preferred codec, with AMR-WB acceptable and AMR-NB no longer acceptable for commercial 5G voice. The IMS layer itself does not change in principle (P-CSCF, I-CSCF, S-CSCF, IMS-AS), but it now sits behind an N6 interface from the UPF rather than behind the PGW.

The harder shift is fallback behaviour. EPS Fallback (EPS-FB) is not a degraded mode: it is a designed feature. When a UE attempts to set up a voice call on a cell that does not support full VoNR or on a UE not provisioned for it, the AMF triggers a redirection to LTE during call setup, and the call completes as VoLTE on the LTE leg. From a user’s perspective, the call works. From a measurement perspective, the call setup time you record is now dominated by the redirection, not by the SIP INVITE round-trip. Apply the same MOS principles validated on VoLTE to VoNR, but expect setup time and fallback rate to behave very differently.

The 6 field KPIs that prove a VoNR call works (or does not)

Vendor labs and conformance reports are not enough. Once VoNR goes commercial, the field engineer needs a small set of measurable KPIs that map directly to user experience and to 3GPP requirements. Six of them carry the signal.

KPI 1: VoNR call setup time. Target on a mature SA cell is below 4 seconds end-to-end, measured from the UE-initiated Service Request to the SIP 200 OK on the answering leg. The VoLTE benchmark sits around 2 seconds. On early VoNR commercial loads, setup time is the single most visible friction point: SIP INVITE round-trips ride the new 5GC path, and any bottleneck in AMF-SMF-UPF coordination shows up here first.

KPI 2: POLQA MOS measured per ITU-T P.863.2 (Super-Wideband mode). The accepted thresholds are MOS ≥ 4.0 on EVS Super-Wideband and MOS ≥ 3.5 on AMR-WB. Below 3.5 in either codec, the call is degraded enough that customers will complain.

KPI 3: One-way latency on the 5QI=1 QoS flow. 3GPP TS 23.501 Table 5.7.4-1 specifies a packet delay budget of 100 ms for 5QI=1 (Conversational Voice). Field-measured one-way latency above 100 ms is a sign that the QFI is being mapped to the wrong DRB on the radio side or that the UPF is not on a low-latency path.

KPI 4: EPS Fallback rate. In a mature SA zone with VoNR provisioned, EPS-FB rate should sit below 5% of voice call attempts. Above that figure, the issue is almost never radio: it is provisioning (UDM/AMF subscriber profile), IMS-on-5GC enablement, or a cell that simply lacks VoNR support flagged in the SIB.

KPI 5: Service Request reject rate. For voice attempts, this should remain below 1%. Above 1%, you are looking at AMF-side authentication, slice-selection, or PCF policy issues. The full set of NAS reject codes and how to read them is covered in our 5G SA registration reject diagnosis guide.

KPI 6: Codec negotiation outcome. Read it from the SDP answer in the 200 OK or 183 Session Progress. EVS-SWB is the preferred outcome. AMR-WB is acceptable. AMR-NB is not acceptable for 5G commercial voice in 2026, and its presence on a VoNR call indicates a misconfigured IMS-AS profile or a degraded interconnect.

End-to-end QoS observability for voice flows is what ties these KPIs into a coherent picture, and it is what a QoS-QoE field measurement approach is built for.

Diagnosing a failing VoNR call: 4 root causes and how to spot them on Android

When a VoNR call fails or sounds wrong on a commercial network, the underlying cause is almost always one of four recurring patterns. The right field tooling on Android, capturing NAS 5GMM/5GSM and SIP/SDP from the modem chipset, makes each one visible without lab equipment.

A. Unintended EPS Fallback. The signature is a NAS Service Reject during call setup with cause 5GS_services_not_allowed, or a systematic redirection to LTE on every voice attempt regardless of cell quality. Root cause: IMS-on-5GC is not provisioned on the AMF/UDM for that subscriber, or the cell does not advertise VoNR support. The fallback is “working as designed”, but the deployment is incomplete.

B. SIP 488 Not Acceptable Here. The IMS-AS rejects the INVITE because the codec offered by the UE is not in its acceptable list. Inspect the SDP offer/answer pair: if the UE offers EVS Super-Wideband but the IMS-AS profile only allows AMR-WB on this slice, the negotiation fails on 488. The fix is on the IMS side, but the diagnosis is on the device.

C. PDU Session Establishment reject from the SMF. Look at the PDU Session Establishment Accept/Reject in NAS 5GSM. Two recurring sub-cases: wrong DNN routing for the IMS APN (the operator pushed a 5G DNN that does not map to the legacy IMS APN in the HSS+UDM mapping table), or the SMF returns 5QI=5 (default best-effort) instead of 5QI=1, so the voice flow gets best-effort treatment and degrades on any congested cell. The DNN routing change between 4G APN and 5G DNN is covered in our APN-to-DNN migration guide.

D. Beam mismatch on Xn handover. The VoNR call survives steady-state but drops on the first inter-gNB handover. Root cause: the SCG configuration on the target gNB lacks the right beam set, and SS-RSRP on the target falls below -110 dBm at the moment of handover. The MeasurementReport sent before the handover shows the failure pattern; the missing beam set on the target is visible in the RRCReconfiguration that prepares the handover.

A practical VoNR field test protocol (Android only, no lab equipment)

You do not need a TEMS or Nemo licence to validate VoNR in 2026. The legacy drive-test stacks are the alternative being moved away from, not the destination. A modern Android-based protocol, captured on a rooted Qualcomm chipset, is enough to qualify a VoNR cell or a VoNR roaming corridor.

Setup. A rooted Qualcomm smartphone (Snapdragon 8 Gen 2 or later), 5G SA forced via the modem AT interface, IMS active and registered, a valid 5G SA SIM with VoNR provisioned in the UDM. A second device for the answering leg, ideally on the same network. POLQA reference clips loaded for objective MOS scoring on the recorded audio.

Method. 50 calls A→B inside a validated SA zone where coverage is dominant and stable, and 50 calls in a coverage edge area where Xn handovers occur during the call. Each call held for 60 seconds minimum. Audio recorded on both ends. NAS 5GMM/5GSM, RRC NR, and SIP/SDP captured in QMDL throughout.

Measure. POLQA MOS computed locally on the recorded audio per ITU-T P.863.2 SWB. Setup time extracted from the timestamp delta between Service Request and SIP 200 OK. EPS Fallback flagged whenever the call sets up on LTE despite SA registration. 5QI value read from the QoS Flow Description IE in the PDU Session Establishment Accept. Codec read from the SDP answer.

Reporting. A small markdown table is enough for the first iteration:

The same checker workflow that operators use to validate VoLTE behaviour can be extended to capture VoNR signalling on the same handset: the VoLTE checker workflow is the foundation, with VoNR-aware mode on the roadmap. For a complete view of voice quality of experience across LTE and 5G, the VoLTE QoE measurement approach gives the audio and signalling cross-correlation needed to defend the result with the operator.

Bottom line

VoNR in 2026 is a deployed reality, not a roadmap. Verizon, du, Bell, T-Mobile, and Vodafone Germany are all running it commercially, but only a fraction of SA networks deliver consistent voice quality across cells, sectors, and handovers. The validation work is no longer a lab exercise: it is a field exercise built on six measurable KPIs (per 3GPP TS 23.501, TS 24.501, TS 26.114, TS 23.228 and ITU-T P.863.2), four recurring root causes, and a protocol that fits on an Android phone.

The teams that win the next VoNR rollout will be the ones that capture NAS 5GMM/5GSM and SIP/SDP on every call, score every recording with POLQA, and turn the result into a single defensible number per cluster. The teams that wait for vendor dashboards will keep guessing.

What is your current VoNR fallback rate on commercial sites, and how much of it is design versus misprovisioning?