VoLTE voice quality: how to measure MOS automatically. HiCellTek

VoLTE voice quality is one of the most critical QoE indicators for mobile operators. A degraded voice MOS generates subscriber complaints, churn, and regulatory complications. This guide explains how to automatically and objectively measure VoLTE voice quality in the field, without a human listening panel.

Background: what is voice MOS?

The MOS (Mean Opinion Score) is the international voice quality measurement scale defined by ITU-T (recommendation P.800):

MOS 5: excellent quality (transparent, comparable to physical presence)
MOS 4-4.5: good quality, slight perceptible attenuation
MOS 3.5-4: fair quality, acceptable for professional use
MOS 3-3.5: audible degradation but comprehensible
MOS < 3: unacceptable quality for commercial use

Historically, MOS was obtained through panels of human evaluators (listening to standardized phrases, scoring). This method is costly and not reproducible in the field. Modern perceptual algorithms allow computing an objective MOS automatically.

VoLTE codecs and their MOS ceilings

The codec used determines the theoretical MOS ceiling achievable, regardless of network quality:

Codec	Band	Max theoretical MOS	Usage
AMR-NB 12.2 kbps	0.3-3.4 kHz	4.1	Basic VoLTE
AMR-WB 12.65 kbps	0.05-7 kHz	4.5	HD Voice
EVS 13.2 kbps	0.05-14.4 kHz	4.5+	Super HD Voice
EVS 32 kbps SWB	0.05-14.4 kHz	4.7+	Premium usage

Critical point: if a VoLTE call uses AMR-NB when both terminals support AMR-WB, the codec negotiation has failed or the network configuration is suboptimal. Check the SIP messages (INVITE + SDP) to confirm the negotiated codec.

Automated MOS calculation methods

PESQ (Perceptual Evaluation of Speech Quality — ITU-T P.862)

PESQ is the first standardized automated MOS calculation algorithm. It compares a degraded audio signal to a reference signal and computes a perceptual score. It is still used in legacy systems but is considered outdated for modern codecs (EVS, Opus).

PESQ limitations: does not account for wideband codecs (AMR-WB, EVS), nor the perceptual characteristics of non-PSTN codecs.

POLQA (Perceptual Objective Listening Quality Analysis — ITU-T P.863)

POLQA is the successor to PESQ, standardized in 2011. It supports:

Narrowband (NB: 0.3-3.4 kHz)
Wideband (WB: 0.05-7 kHz)
Super-wideband (SWB: 0.05-14.4 kHz)

POLQA is the reference for VoLTE compliance testing in regulatory reports and operator audits. Limitation: costly commercial license (Opticom/SwissQual), not suitable for embedded use in lightweight field tools.

ViSQOL (Virtual Speech Quality Objective Listener — Google Research)

ViSQOL is an open-source algorithm developed by Google Research (available on GitHub). It uses a spectro-temporal approach based on NSIM (Neurogram Similarity Index Measure) to compare the degraded signal to the reference signal.

Strengths for field use:

Open source (Apache 2.0)
High performance on AMR-WB and EVS (superior to POLQA in some cases)
Two modes: speech (16 kHz) and audio (48 kHz, for musical/multimedia content)
Natively embeddable in Android applications via JNI (libvisqol_jni.so)

This is the algorithm now preferred for automated VoLTE measurements in the field.

VoLTE MOS field test protocol

Recommended setup

Minimum configuration:

Measurement terminal (rooted Qualcomm smartphone with measurement suite)
Active SIM on the tested operator network
Call termination point: test IVR server or second reference terminal

Advanced configuration (end-to-end measurement):

Terminal A (measurement, field): instrumented Qualcomm smartphone
Terminal B (reference, fixed): smartphone or POTS via SIP gateway
Reference audio source: standardized ITU-T P.501 WAV file (test phrases)
MOS computation server: on-device or cloud-side

Measurement sequence

Step 1: VoLTE call establishment

Dial from Terminal A to Terminal B
Verify in SIP messages: 180 Ringing -> 200 OK -> ACK
Verify the negotiated codec in the SDP (AMR-NB / AMR-WB / EVS)
Verify QCI bearers (QCI 1 for VoLTE voice)

Step 2: Reference audio injection

Play the reference WAV file from Terminal B
The signal traverses the VoLTE network and is received on Terminal A
Capture the received signal on Terminal A (RECORD_AUDIO permission required)

Step 3: MOS computation

Temporally align the received signal with the reference signal
Compute the ViSQOL MOS (or POLQA if available)
Export the CSV report (MOS score, timestamps, aligned WAV files)

Step 4: Correlation with radio KPIs

Correlate the computed MOS with RF KPIs (RSRP, SINR, BLER) during the call duration
Identify correlations: MOS drops during radio degradations
Correlate with Layer 3 messages: check for Bearer Modification, SRVCC triggers

Test session duration

For a statistically representative MOS:

Minimum duration: 30 seconds of active speech (excluding silences)
Recommended duration: 90 seconds to 2 minutes per call
Number of repetitions: minimum 5 calls per test zone

Standardized reference files

ITU-T P.501 defines the reference files for perceptual tests. Phrases in multiple languages (French, English, Arabic, etc.) are available for multi-region testing.

Most frequent VoLTE problems and their diagnosis

Problem 1: MOS < 3.5 despite good RSRP

Probable cause: high RTP jitter or packet loss, despite good radio coverage.

Layer 3 diagnosis: analyze RTP flows (QCI 1 bearer). Jitter > 20 ms or loss > 1% significantly degrades MOS, even if RSRP is excellent.

Frequent secondary cause: QoS issue in the core network (misconfigured DPI, QCI 1 priority not respected under load).

Problem 2: Unexpected SRVCC handover

SRVCC (Single Radio Voice Call Continuity) is the mechanism that hands over a VoLTE call to 2G/3G CS when LTE coverage becomes insufficient. Each SRVCC introduces an audible interruption of 0.3 to 1.5 seconds.

Diagnosis: detect RRC Handover Command messages with target RAT = UTRAN/GERAN in Layer 3 logs. The associated GPS location reveals LTE coverage edge zones.

Problem 3: MOS degradation indoors

Radio attenuation inside buildings (typically 10-20 dB additional) can push RSRP below the VoLTE threshold. The Android suite then falls back to SRVCC or to a classic CS call.

Diagnosis: map MOS by zone (indoor/outdoor, floor) and correlate with indoor RSRP. Zones with MOS < 3.5 indoors reveal areas needing repeaters or small cells.

Problem 4: AMR-NB forced instead of AMR-WB

Symptom: MOS capped at 4.1 even on an excellent network.

Diagnosis: check the SIP INVITE / SDP message: the m=audio line must list the proposed codecs. If AMR-WB (payload type 106) is absent, the terminal is not offering HD Voice. If AMR-WB is present but the 200 OK responds with AMR-NB, the network or remote terminal does not support AMR-WB.

VoLTE MOS: typical operator targets and thresholds

Region / Operator	Target MOS (median)	Minimum MOS (P10)
Europe (ETSI TS 103 531)	>= 4.0	>= 3.5
Gulf (TRA/CTRA regulation)	>= 4.0	>= 3.5
France (ARCEP)	>= 3.8	>= 3.2
Francophone Africa (ARTEC)	>= 3.5	>= 3.0

These thresholds are often embedded in SLA contracts between operators and equipment vendors, and in deployment specifications.

VoNR MOS (Voice over NR)

For 5G SA deployments, voice traffic flows over VoNR (Voice over New Radio) instead of VoLTE. MOS measurement methods remain identical, but the protocol layers change:

5G QoS bearer (5QI 1) instead of QCI 1
5G NAS MM/SM protocols instead of EPS
IMS SIP remains identical

VoNR implementation is still being deployed in most networks as of 2025, with SRVCC to VoLTE (not to 2G/3G) as fallback.

Conclusion

Automated VoLTE MOS measurement is now accessible directly in the field, through perceptual algorithms like ViSQOL, integrated into Android measurement suites. This capability eliminates the need for human listening panels or dedicated test equipment to obtain voice QoE indicators comparable to laboratory tests.

For an RF or QoS engineer, the MOS + radio KPIs + Layer 3 messages correlation is the most powerful approach for diagnosing and resolving VoLTE voice quality issues on the operator network.

VoLTE voice quality: how to measure MOS automatically