VoIP Jitter and Latency: What Causes Poor Call Quality on a Business Network

Bad voice quality on VoIP is one of the hardest problems in business telephony to diagnose, because the symptoms look identical from the user's side — choppy, delayed, echoing, breaking up — while the causes live in completely different layers of the network. People describe it as "the line is bad" and then spend weeks replacing equipment that was never the problem.

The way through is to separate four things that all present as "the call sounds bad": latency, jitter, packet loss and echo. Each has a different cause, a different measurement and a different fix.

Latency: delay in one direction

Latency is how long a packet takes to reach the other end, and it is measured in milliseconds. Human conversation tolerates delay reasonably well up to about 150 milliseconds one way, and becomes noticeably awkward beyond 300 — people start talking over each other because they cannot tell whether the other person has finished.

What causes it in a business network is usually distance and congestion. Distance is fixed: your carrier's point of presence may be in another city, and the round trip to the far end of a call is beyond your control. Congestion is not — a saturated uplink adds queueing delay to every packet that crosses it, including voice.

Latency is usually a carrier distance problem and congestion is usually an internal network problem. Telling them apart early saves a lot of unnecessary equipment replacement.

Jitter: variation in delay, which the ear hates more

Jitter is the variation in inter-arrival times of packets. A steady 100 milliseconds of delay is a fixed cost you can design around. A delay that swings between 20 and 100 milliseconds produces packets that arrive out of order relative to each other, and the receiving buffer has to reorder or drop them.

Jitter is almost always caused by queuing somewhere in the path — a link where packets wait behind other traffic, a WiFi segment, or an overloaded uplink. This is why jitter is the more common complaint in practice even when average latency looks fine, and why a jitter buffer on the receiving side is an important part of any VoIP system.

Packet loss: the silent killer

Lost packets produce the audible artefacts people recognise immediately: clipped words, a robotic or garbled voice, complete silence for a fraction of a second. A small amount of loss is audible; a larger amount makes the call unusable.

Loss comes from physical problems — a faulty cable, a failing connector, a switch port with an error, a WiFi signal problem — and from congestion that causes queue overflow. Physical causes are worth checking first because they are easy to find and easy to fix, and because they tend to affect a specific set of devices rather than everything at once.

Echo: the one that is usually not the network

Echo is distinctive because it comes back to the talker. It is caused by audio being picked up by the microphone and returned through the loudspeaker, which happens when the acoustic environment reflects a lot and the echo canceller cannot keep up — a bare-walled room, a handset off the cradle, or a speakerphone in a reflective space.

Network delay makes echo cancellation harder, because the canceller has more time in which the returning audio might be genuine. So echo that appears intermittently on long-distance calls is often a delay problem presenting as an acoustic one. The voice quality and QoS guide covers how the two interact.

How to measure rather than guess

Before changing anything, measure. Two steps separate the layers quickly.

1. Measure the network path from the affected device to your carrier — latency, jitter and packet loss, at busy hour
2. Measure the same path from a different device in the same building — if the second is clean, the problem is the first device or its cabling
3. Test a call on a mobile network from the same location — if it is clean, the problem is inside the building network

That third step is the one people skip and it is the fastest way to separate "our network" from "the carrier's network" from "this one desk". A tool that reports jitter as an average is not much use; what matters is the variation, and the worst-case figures over a busy period rather than the average.

The fixes that work, in order

Newrock's OM series IP-PBX supports QoS marking and voice VLAN separation, which covers the two most common internal causes. Where the delay is carrier-side or the endpoint is genuinely remote, the SX100 SBC and the OM1000-SBC place media control at the edge of the network, where it can smooth jitter and terminate the session closer to the far end.

Set up monitoring so you stop guessing

The lasting improvement is instrumentation. Call quality monitoring means you know a problem exists before the users tell you, and — more usefully — you know which segment of the network it is in. The SX100 includes voice quality monitoring among its capabilities, and the call centre and reporting guide covers the reporting side of the same problem.

Frequently asked questions

What is an acceptable jitter value?

Jitter is judged by variation rather than absolute value, and the tolerable figure depends on the jitter buffer in the receiving equipment. As a rough guide, single-digit milliseconds is comfortable, tens of milliseconds starts to be audible on a poor connection, and beyond that buffered packets start being dropped.

Does call quality get worse with more users?

It can, but only if the users are consuming shared uplink capacity. Voice itself is not demanding. The common pattern is that a site was fine until file sharing, backup or video traffic began competing for the same uplink — at which point QoS or more bandwidth addresses it directly.

Why does the call sound fine on my mobile and bad on the desk phone?

That points to the desk phone's path: its cabling, its switch port, or its WiFi segment. Testing a second device on the same desk, and the same handset on a different desk, separates the cable from the phone quickly.

Can echo be a network problem?

It can contribute to it. Echo cancellation depends on delay, so higher delay makes cancellation harder, which makes an echo that only appears on long or congested calls a delay symptom rather than an acoustic one.

Related reading on the Newrock site

• OM20G all-in-one IP-PBX
• SX100 session border controller for SMB
• OM1000-SBC (SX3000) session border controller
• VoIP Security: Where Attacks Actually Happen in a Business Network

Conclusion

Poor voice quality is diagnosable, but only if you separate the four causes before you touch anything. Measure latency, jitter and loss, test from a second device and a second network, and you will know whether you are looking at congestion, a cable, a WiFi segment or the carrier. Most cases are congestion or a physical fault, and both are fixed by a specific, known action rather than by replacing the phone system.