How Business VoIP Actually Works
Voice is the least forgiving traffic on your network. A web page that arrives a second late is unnoticed; a voice packet a second late is useless and gets discarded. That single difference explains why VoIP needs treatment other traffic does not, and why "the network is fine" is not an answer when calls sound bad.
The short version
- SIP sets up the call. The audio itself travels separately, which is why signalling can work while audio does not.
- Codecs trade bandwidth for quality. Voice uses very little bandwidth by modern standards.
- Jitter and packet loss ruin calls, not bandwidth. This is the part people get wrong.
- QoS and a voice VLAN are what make voice reliable on a shared network.
Signalling and audio are separate
SIP is the protocol that sets up, modifies, and ends a call — the equivalent of dialling and ringing. Once the call connects, the actual audio flows as a separate stream directly between the endpoints where possible.
This split explains a symptom that otherwise makes no sense: a call that connects, rings, and is answered, but where one or both parties hear nothing. Signalling worked and audio did not, which usually means a firewall or NAT problem affecting the media stream rather than anything wrong with the phone.
Many consumer and small business routers include a SIP ALG feature intended to help VoIP traverse NAT. In practice it frequently rewrites signalling incorrectly and causes one-way audio, calls dropping after a set time, or phones failing to register. If you are troubleshooting VoIP on a network with a general-purpose router, disabling SIP ALG is one of the first things to try — it fixes a surprising proportion of cases.
Codecs
| Codec | Character |
|---|---|
| G.711 | Uncompressed, the traditional standard. Best quality of the classic codecs, highest bandwidth — still modest by modern standards. The safe default on a LAN. |
| G.729 | Compressed to roughly an eighth of G.711. Noticeably lower quality. Made sense when bandwidth was scarce; rarely necessary now. |
| G.722 | Wideband — a wider audio range, so voices sound clearer and more natural. Requires both ends to support it, which usually means internal calls. |
| Opus | Modern, adaptive, and resilient. Adjusts to network conditions and handles packet loss better than the older codecs. Increasingly the default where supported. |
Voice bandwidth is small. A dozen simultaneous calls on G.711 consume a fraction of a typical business internet connection. Bandwidth is almost never why calls sound bad — which is why upgrading the connection so often fails to fix it.
What actually determines call quality
| Factor | Effect on the call |
|---|---|
| Latency | Delay between speaking and being heard. Past roughly 150 milliseconds one way, people start talking over each other. Not distortion — just an awkward conversation. |
| Jitter | Variation in packet arrival timing. The most common cause of choppy or robotic audio. Phones buffer to absorb it, but only so much. |
| Packet loss | Missing audio. Even low single-digit loss is audible as clipping or dropouts, because there is no time to retransmit — a late packet is a useless packet. |
| Bandwidth | Rarely the limit. Matters only when a link is saturated by other traffic, which is a queuing problem rather than a capacity one. |
| Echo | Usually acoustic — a speakerphone or headset picking up its own output — rather than a network issue. |
| Wi-Fi | Introduces jitter and loss that wired connections do not. Wired is strongly preferable for desk phones. |
The instinct when calls sound bad is to buy a faster internet connection. Voice uses very little bandwidth, so this rarely helps. Choppy or robotic audio almost always means jitter or packet loss — usually because voice is queued behind a large file transfer somewhere. The fix is prioritisation, not capacity.
QoS and the voice VLAN
Quality of service means the network recognises voice packets and sends them ahead of everything else. Voice is small and time-critical; letting it jump the queue costs other traffic almost nothing and fixes most quality problems.
- Put phones on their own VLAN. Separates voice from data, makes prioritisation straightforward, and simplifies troubleshooting.
- Mark voice traffic and honour the marking across every switch in the path. A single device ignoring the marking undoes the rest of the chain.
- Prioritise on the internet link too, where the constraint usually is. Upstream is typically slower than downstream and is where a large upload will crush call quality.
- Do not rely on Wi-Fi for desk phones. Use the pass-through port on the phone for the PC where available — the phone carries tagged voice and untagged data on one cable.
- Check PoE budget. Phones draw modestly each, but a floor of them adds up — see choosing switches and access points.
IP phones are powered by the switch. When the power goes out, the switch stops, and every phone in the building goes dead — including for emergency calls. Put the switches and the internet equipment on a UPS, and make sure staff know an alternative way to call emergency services during an outage. This is a genuine safety consideration, not a convenience one — see sizing a UPS.
On-premises, hosted, or hybrid
| Model | Trade-offs |
|---|---|
| Hosted / cloud PBX | Provider runs the system; you provide phones and network. Lowest setup effort, monthly per-user cost, and your internet link becomes critical. |
| On-premises PBX | System lives in your building. More control and no per-user fee, but you own the maintenance, updates, and resilience. |
| Hybrid | Local system with cloud trunking. Common where an existing investment is being extended. |
Whichever model, the phones speak SIP, and the network requirements are the same. That is what makes phone hardware relatively portable between platforms — subject to provisioning support, which is covered in provisioning and deployment.
Official manufacturer and standards resources
Codec support, provisioning methods, and platform certification vary by phone model and by service provider. Confirm compatibility before purchasing.
Moving to VoIP and unsure what the network needs?
Tell us how many phones, what switches you have, and who your provider is. We will tell you what needs configuring — voice VLAN, QoS, PoE budget — before the phones arrive rather than after.
Still stuck? Talk to someone who works on this hardware daily.
Tell us the model, what you have already tried, and what you are seeing. We will tell you whether it is a setting, a consumable, or a part — and we will say so if you do not need to buy anything.
