VQ Monitor: Real-time VoIP Call Quality Monitoring

VQ Monitor is Sevana’s server for real-time VoIP call quality monitoring on live networks. Crucially, it measures quality without sitting in the media path or touching your endpoints. The ITU-T G.107 E-model scores every stream, while Sevana’s PVQA perceptual analysis runs on the calls you select. As a result, you can finally tell network problems apart from audio problems instead of guessing.

Request a demoTalk to us about licensing

VoIP call quality monitoring without touching your endpoints

VQ Monitor reads a mirrored (SPAN) copy of your production VoIP traffic, or HEP feeds from your SBCs and proxies. From there, it reports per-stream quality continuously. The E-model scores every stream it sees; however, perceptual scoring stays reserved for the calls you choose, because that is where the CPU cost lives.

How VQ Monitor works

The deployment splits into two cooperating processes. First, vq-core captures and decodes traffic, runs PVQA analysis, and publishes per-stream and per-interval reports on a ZeroMQ bus. Then the optional vq-db subscribes to one or more cores, stores the results, evaluates alarms, exports CSV, and serves the embedded web dashboard. Both ship as single static binaries with no runtime dependencies, so installation amounts to copying a file and a config. Only vq-core is mandatory; the bus speaks a fully documented Protocol Buffers schema and vq-db exposes a documented REST API (see Documentation), so you can replace or complement vq-db with your own consumer at any time.

Two quality scores that answer different questions

Network MOS (ITU-T G.107 E-model)

This simplified ITU-T G.107 E-model uses per-codec impairment values from G.113. Because it works from packet statistics alone, such as loss, jitter and round-trip time, it never depends on a successful audio decode, and it keeps working even when the media stays encrypted. Consequently, the CPU cost stays negligible, so VQ Monitor runs it on every stream.

Sevana MOS (PVQA)

Sevana’s PVQA algorithm works on the decoded PCM, locates intra-audio impairments such as clipping, echo, noise and dead air, and then predicts a MOS on the ITU-T P.800 scale. In short, Network MOS tells you what the transport delivered, whereas Sevana MOS tells you what the caller heard. On a clean call the two rarely disagree; however, when they do, the gap itself becomes the diagnosis.

Capture, signalling and codecs

The SIP capturer reconstructs call setup and correlates RTP addresses and SSRC values with SIP transactions. As a result, it tags each stream with its caller and callee URIs. For the streams you select, the decoder then turns the RTP payload into PCM and hands it to PVQA.

Supported codecs

G.711 (PCMA/PCMU)G.722G.729OPUSiLBCAMR-NBAMR-WBEVSGSM FRGSM HRGSM EFR

Carriers sometimes wrap AMR-NB, AMR-WB and EVS in IuUP frames inside RTP; fortunately, the decoder parses both layouts. In addition, you can switch off patent-licensed codecs at runtime and therefore avoid licensing exposure. For AMR and EVS streams the RTP payload headers also yield DTX/SID silence statistics, even when the audio itself is never decoded.

SIP call analytics, not just media metrics

Beyond per-stream quality, VQ Monitor reconstructs the SIP call lifecycle itself. Every established call is recorded with its post-dial delay (INVITE to answer), every failed call with its outcome — rejected, canceled or timed out — together with the final SIP response code, and every mid-call reINVITE or UPDATE (hold/resume, codec renegotiation) with the renegotiated media details. Because each RTP stream is tagged with its SIP Call-ID, the dashboard and the API let you move in both directions: from a call to all of its media streams, and from a poorly scoring stream back to the signalling timeline that produced it. Consequently, “the audio was bad” and “the call never set up” finally live in one tool.

Dashboard, API and alarms

For day-to-day VoIP call quality monitoring, the vq-db dashboard lists active and finished streams together with their Sevana MOS, Sevana R-factor, Network MOS, jitter and SIP addresses. From any row, you can drill down to single-stream and per-chunk PVQA detail. Each stream also carries round-trip time, min/max/average envelopes for jitter and both MOS scores, detected echo peaks, and DTX/SID silence counters. Moreover, everything the dashboard shows — streams, SIP calls, live and historical summaries — is also available as JSON or CSV through a fully documented REST API (see Documentation), so it feeds straight into your own tooling.

Finally, threshold alarms run a shell command of your choice whenever R-factor, Sevana MOS, Network MOS, call duration, packet loss, jitter or free disk space crosses a limit you set. Data retention is equally configurable: stream records and decoded audio age out on independent lifetimes, or the database can run entirely in memory with bounded retention when you only care about the live picture.

Tuning for high-volume probes

These knobs matter most on a busy link, in roughly the order you should reach for them:

• Choose which calls deserve PVQA. A runtime-controlled “track list” of SIP addresses decides which calls get full perceptual analysis, and a process-wide cap bounds the audio workload; everything else still gets Network MOS. Because audio decoding dominates CPU, this single lever drives most of your budget.
• Optionally filter at the capture stage. UDP port-range and illegal-stream filters keep spurious traffic off the worker threads — the biggest saving on a busy 10G mirror. Plain libpcap is all a probe needs; for multi-million-packet-per-second links, optional Intel DPDK capture takes over at line rate, and per-NIC drop counters prove whether any loss happened before the probe ever saw the packet.
• Optionally separate storage from capture. Run vq-db on its own host and let several probes publish into it; the ZeroMQ feed then scales close to linearly.

Requirements and deployment

Documentation

The full product documentation is available as PDF — including the two open integration specifications, so your own tooling can consume the same data the bundled dashboard does:

• VQ Monitor — product documentation (PDF) — architecture, configuration reference, deployment and tuning
• vq-core ZeroMQ bus API (PDF) — the Protocol Buffers event schema every subscriber can consume
• vq-db REST API (PDF) — endpoints, filter grammar and response shapes

See VoIP call quality monitoring on your own traffic

Want to watch quality on your own network in real time? Simply book a guided demo of VQ Monitor, or talk to us about licensing.

Request a demoContact Sevana