Latency, Jitter, and Packet Loss in Voice Networks

Networking · Last updated · About 5 min read

Overview

Three network impairments determine whether live voice over IP — telephony VoIP (Voice over IP) or radio RoIP (Radio over IP) — sounds good, tolerable, or broken: latency (delay), jitter (variation in delay), and packet loss. Data applications barely notice all three; a file transfer just retransmits and finishes a moment later. Live voice cannot retransmit — a packet that misses its playout moment is gone. Understanding what each impairment is, where it comes from, and how it sounds is the foundation of voice-network troubleshooting.

Latency

Latency is the one-way delay from sound entering the microphone to sound leaving the far speaker ("mouth-to-ear"). It accumulates from: codec framing and processing, packetization (waiting to fill each 20 ms packet), network transit (propagation over distance plus queuing in every router and switch), jitter buffering, and playout.

ITU-T G.114 one-way delay bands A scale of one-way mouth-to-ear delay. Zero to 150 milliseconds is the preferred range and is essentially transparent. 150 to 400 milliseconds is acceptable but degrades as it rises. Above 400 milliseconds is unacceptable for general planning, because talkers start stepping on each other. How much one-way delay a conversation can takeITU-T G.114 bands for mouth-to-ear delayPreferrednobody noticesAcceptabledegrades as it risesUnacceptabletalkers step on each other0 ms150 ms400 ms600 ms +one-way delay, microphone to far speakerDelay is additive: the codec, the 20 ms it takes to fill each packet, every router and switch along theway, the jitter buffer and the playout all contribute. Each stage is small; the total is what the userhears.
These are one-way figures. A round trip that feels sluggish on a radio patch is usually two of these added together.

The authoritative targets come from ITU-T Recommendation G.114 (One-way transmission time), which defines three bands for one-way mouth-to-ear delay:

One-way delay G.114 assessment
0–150 ms Preferred range — essentially transparent interactivity for most applications, speech and non-speech.
150–400 ms Acceptable, but with increasing degradation as delay rises.
Above 400 ms Unacceptable for general network planning (exceptions exist, e.g. an unavoidable double satellite hop).

Practically, once delay climbs well into the middle band, talkers begin stepping on each other because neither can tell the other has started speaking. Note these are one-way figures — a round trip measured with ping is the sum of both directions and excludes all endpoint processing, so ping time is a floor, not the full story.

Latency's defining property: it is the one impairment no buffer can fix — buffering only adds more of it.

Jitter

Jitter is variation in packet arrival spacing. The sender emits voice packets on a metronome (typically every 20 ms); queuing along the path makes them arrive bunched and gapped. The receiver's jitter buffer absorbs this by holding packets briefly and playing them out on a steady clock — trading added delay for smoothness. Packets arriving later than the buffer can wait for are discarded, becoming effective packet loss.

A jitter buffer smoothing uneven packet arrival Ten voice packets are sent at an even 20 millisecond spacing. They arrive at uneven times because network delay varies. A jitter buffer holds each one briefly and releases it on the original beat, so playout is even again. One packet arrives after its playout moment and is dropped, heard as a short gap. What a jitter buffer doesPackets leave evenly, arrive unevenly, and must be played evenly againSentevery 20 ms, like clockworkArrivednetwork delay varies packet by packetPlayed outheld briefly, then released on the beat11122233344455566677788899—a gap in the audio101010deadlineA bigger buffer rides out more jitter — and adds that much delay to every single packet.That trade is the whole design problem: too small and audio breaks up, too large and the conversation drags.
A lost packet and a late packet sound the same. Live voice cannot wait for a retransmission, so a packet that misses its slot is simply gone.

Consequences worth internalizing:

  • Jitter within the buffer's reach is inaudible; it just costs delay.
  • Jitter beyond the buffer's reach sounds exactly like packet loss (choppiness), even though every packet eventually arrived.
  • Adaptive jitter buffers grow during bad spells — which is why some calls' delay creeps upward as they proceed.

Packet Loss

Packet loss is packets that never arrive (or arrive too late to use). Codecs apply packet loss concealment — interpolating over small gaps — which keeps speech natural up to roughly one percent loss; beyond that, audio turns choppy, syllables vanish, and low-bit-rate codecs (whose packets each carry more irreplaceable information) degrade fastest. Loss in voice networks is usually burst loss from momentary congestion, which is far more audible than the same average loss spread evenly.

Where Each Comes From

  • Distance: propagation delay — pure latency, no jitter or loss.
  • Queuing in congested links: the dominant source of jitter and burst loss — voice packets stuck behind bulk data in router buffers. Worst at the LAN-to-WAN choke point.
  • Wi-Fi: retransmissions, contention, and roaming events create jitter and loss spikes that wired Ethernet simply doesn't have. Wireless links are the first suspects for intermittent choppiness.
  • Congested or undersized WAN circuits: sustained high utilization turns every busy hour into a bad-audio hour.
  • VPN (Virtual Private Network) overhead: encapsulation adds a little latency, and MTU (Maximum Transmission Unit) fragmentation issues add loss-like symptoms; tunnels over the public internet inherit whatever jitter the internet path has.
  • Endpoint problems: an overloaded PC or gateway produces jitter and gaps that look like network faults but originate in the host.

Measuring

  • ping gives round-trip latency and coarse loss; run it sustained and watch the spread between minimum and maximum times — a wide spread is jitter, even though ping doesn't label it. Traceroute locates which hop the delay appears at.
  • RTCP (RTP Control Protocol) reports — the companion protocol to RTP (Real-time Transport Protocol) — carry the endpoints' own measured loss and interarrival jitter; many products surface these as call-quality statistics.
  • Packet capture analysis is the definitive method: Wireshark's RTP stream analysis reports per-stream loss and jitter from sequence numbers and timestamps — see Wireshark and Packet Capture.

PTT-Specific Concerns

Push-to-talk voice adds failure modes telephony doesn't have:

  • Late talk-permit: on systems where the user must wait for a grant/talk-permit tone, network delay stretches the wait, and users start talking before the channel is actually open.
  • Clipped first syllables: any delay in detecting and propagating the start of a transmission (slow COR — Carrier Operated Relay — detection, VOX — voice-operated switch — attack time, buffer fill time) cuts the beginning off words. Users report "I have to key up and wait a beat before talking."
  • Doubling: high mouth-to-ear delay makes two parties key simultaneously because neither hears the other start — the radio equivalent of talking over each other on a satellite call.

Remedies

1. 2. Wired over wireless: move consoles and gateways off Wi-Fi onto Ethernet. 3. Capacity: relieve chronically saturated links; QoS manages contention but cannot create bandwidth. 4. Jitter-buffer tuning: where products expose it, size buffers to the measured jitter — larger for rough WAN paths (accepting delay), smaller on clean LANs (minimizing delay). 5. Path improvements: prefer private/deterministic WAN paths over best-effort internet for critical voice, and keep the hop count and tunnel nesting down.

References

Latency bands verified against ITU-T G.114, August 2026. The ~1% packet-loss guidance is a widely-used rule of thumb rather than a figure fixed by standard — actual tolerance depends heavily on the codec and on whether loss is bursty or evenly distributed.

This article is part of the CTI Products Knowledge Base. Questions about RadioPro or MCN products? Call +1 (513) 595-5900 or email info@ctiproducts.com.

Informational only. This article is general background and is not engineering, safety, legal or regulatory advice. Verify requirements with the equipment manufacturer and the applicable authority before relying on it. Product names and trademarks belong to their respective owners.