VoIP Quality: Network Requirements for Crystal-Clear Calls

Voice over IP converts analogue speech into digital packets that traverse the same network as email, web browsing, and file transfers. Unlike those data applications, voice is real-time and unforgiving — a 200ms delay in loading a web page is imperceptible, but a 200ms delay in a phone call creates awkward conversation overlap. Packet loss that causes a web page to reload transparently causes audible clicks, gaps, or robotic distortion in a voice call. For VoIP to work well, the network must guarantee low latency, low jitter, and near-zero packet loss for voice traffic — even when the network is under load from other applications.

A voice codec compresses and encodes speech for transmission. The codec choice determines both call quality and bandwidth consumption. The most common codecs in modern VoIP deployments are G.711 (used in traditional telephony and many SIP trunks) and Opus (used by Microsoft Teams, Zoom, and other modern platforms). G.711 provides toll-quality audio at a fixed 64 kbps payload rate, but with IP, UDP, and RTP headers, each call consumes approximately 80-90 kbps per direction (or about 180 kbps bidirectional). G.729, a compressed codec, reduces this to about 30 kbps per direction at the cost of some quality. The Opus codec is adaptive — it dynamically adjusts bitrate from 6 kbps to 128 kbps based on network conditions, offering excellent quality even at lower bitrates.

Three network metrics determine VoIP call quality: latency (the one-way delay from sender to receiver), jitter (the variation in latency between packets), and packet loss (the percentage of packets that never arrive). For acceptable call quality, the ITU-T G.114 recommendation specifies one-way latency should not exceed 150 milliseconds. Beyond 150ms, speakers begin to talk over each other because the delay is perceptible. Jitter should be below 30 milliseconds — higher jitter causes the jitter buffer to overflow or underflow, resulting in audible distortion. Packet loss should be below 1%, and ideally below 0.5% — even 2-3% packet loss makes calls noticeably degraded.

QoS is the mechanism that prioritises voice traffic over less time-sensitive data. Without QoS, a large file download or a backup job can saturate a link and starve voice packets of bandwidth, causing quality degradation. QoS works by marking voice packets with a DSCP (Differentiated Services Code Point) value — typically EF (Expedited Forwarding, DSCP 46) for voice media and CS3 (DSCP 24) or AF31 for call signalling. Network switches and routers are then configured to place EF-marked traffic into a strict priority queue that is served ahead of all other traffic, ensuring voice packets experience minimal delay even during congestion.

Implementing QoS requires configuration at every hop in the path — access switches, distribution switches, core routers, and WAN links. If a single device in the chain does not honour DSCP markings, the prioritisation breaks for all downstream hops. For Microsoft Teams, configure QoS policies on client PCs via Group Policy to mark Teams audio traffic with EF (UDP ports 50000-50019), video with AF41 (UDP ports 50020-50039), and screen sharing with AF21 (UDP ports 50040-50059). On the network side, configure trust boundaries at the access layer to either trust DSCP markings from endpoints or remark traffic based on source port ranges.

Placing IP phones on a dedicated voice VLAN — separate from the data VLAN used by PCs — is a best practice that provides several benefits. First, it isolates voice traffic from broadcast storms and other data-plane issues on the PC VLAN. Second, it makes QoS configuration simpler because you can apply policies based on the VLAN or subnet rather than inspecting individual packet markings. Third, it enhances security by preventing PCs from directly accessing phone management interfaces. Most managed switches from Cisco, Aruba, and other vendors support an "auxiliary VLAN" or "voice VLAN" feature on access ports, which instructs the connected IP phone to tag its traffic with the voice VLAN while the PC behind the phone uses the untagged data VLAN.

When deploying softphones (Teams desktop or mobile client) rather than physical IP phones, the voice VLAN concept does not apply in the same way because the softphone runs on the same PC as data applications. In this scenario, QoS marking at the endpoint (via Group Policy for Windows) and policy-based queuing on the network become even more important. Ensure the access switch trusts DSCP markings from the PC port, or use port-based classification to identify and prioritise voice traffic by its UDP port range.

IP phones require electrical power, and Power over Ethernet (PoE) delivers it through the same Ethernet cable that carries data — eliminating the need for separate power adapters at every desk. Most modern IP phones require IEEE 802.3af (PoE), which delivers up to 15.4 watts per port, though phones with large colour touchscreens or built-in Wi-Fi may require 802.3at (PoE+) at up to 30 watts. When planning a deployment, calculate the total PoE power budget needed across all switch ports and ensure the switch's power supply can deliver it. A 48-port PoE+ switch fully loaded with phones may need 700-800 watts of PoE budget — confirm this does not exceed the switch's rated capacity, especially if wireless access points on the same switch also draw PoE power.

The Mean Opinion Score (MOS) is the standard metric for quantifying voice quality on a scale of 1 (unintelligible) to 5 (excellent). In practice, VoIP calls rarely exceed a MOS of 4.4 due to inherent codec limitations. A MOS of 4.0 or above is considered good quality, 3.5-4.0 is acceptable, and below 3.5 users will perceive noticeable quality issues. MOS can be estimated algorithmically using the E-model (ITU-T G.107), which calculates an R-factor from network metrics (latency, jitter, packet loss) and codec characteristics, then converts it to a MOS equivalent. Many monitoring tools report both R-factor and estimated MOS.

Before deploying VoIP, run a network assessment using tools designed to simulate voice traffic and measure quality. Microsoft's Network Assessment Tool tests connectivity to Teams media relays and reports latency, jitter, and packet loss. iPerf can generate UDP traffic patterns that simulate voice calls, allowing you to measure network behaviour under load. PathSolutions TotalView and Viavi Solutions provide enterprise-grade VoIP quality testing. For ongoing monitoring, platforms like IR Collaborate (formerly Prognosis) and Nectar DXP provide real-time MOS scoring and alerting across Teams, Zoom, and other UCaaS platforms, pinpointing quality issues to specific network segments or endpoints.

For cloud-based VoIP (Teams, Zoom, RingCentral), the internet link is the most critical segment because voice traffic must traverse it to reach the provider's data centres. Ensure sufficient bandwidth headroom — a site with 20 concurrent calls using the Opus codec needs approximately 2.4 Mbps dedicated to voice in each direction, but you should provision at least 3x this amount to account for data traffic bursts. SD-WAN solutions can help by steering voice traffic over the best-performing link in real time, applying QoS policies at the WAN edge, and failing over to a secondary link if the primary degrades. For sites with only a single internet connection, consider a 4G/5G failover link specifically for voice continuity during outages.

Before going live with any VoIP deployment, validate the following: internet bandwidth is sufficient with headroom for concurrent calls and data traffic; QoS is configured end-to-end from endpoint to WAN edge with DSCP EF for voice media; voice VLANs are configured on access switches with correct DHCP scopes; PoE budgets support all planned IP phones plus any wireless APs on the same switches; firewall rules permit the required UDP port ranges and destination IPs for the VoIP platform; a network assessment tool has confirmed latency below 100ms, jitter below 20ms, and packet loss below 0.5% to the provider's media endpoints; and UPS backup covers critical network infrastructure to maintain calling during power outages.

Never deploy VoIP without a pre-deployment network assessment — it is the single most important step in ensuring a successful rollout. Include network readiness as a mandatory phase in every voice project scope of work, and charge for it as a professional service. When issues are found — and they often are, from misconfigured QoS to under-provisioned WAN links — remediate them before go-live rather than troubleshooting after users are already complaining. Position yourself as the expert who delivers crystal-clear calls by ensuring the foundation is right, and your clients will trust you with their entire communications infrastructure.