5G vs fibre: what actually decides VoIP call quality

Which numbers actually decide how a call sounds?

Not bandwidth. A single G.711 call needs about 80 kbps, headers included – any line sold in South Africa today clears that hundreds of times over. Voice lives or dies on three other numbers.

Latency is the one-way trip time from your mouth to the other ear. ITU-T G.114, the international standard for transmission time, puts the budget at 150 ms one way for most applications; beyond that, people start talking over each other because the conversational rhythm breaks. Every hop spends part of that budget, and the access line is only one of them.

Jitter is the variation in that trip time between packets. Cisco’s QoS design guidance targets average one-way jitter under 30 ms for voice. Your PBX hides jitter with a jitter buffer, but the buffer works by holding packets back, so every millisecond of jitter it absorbs becomes latency you pay elsewhere in the budget.

Packet loss is the killer. Voice packets are never retransmitted; a lost packet is a gap in the audio. Cisco’s guidance caps loss at 1 percent, and notes that compressed codecs like G.729 need loss far below that before errors become audible. This is where 5G and fibre genuinely differ: loss on a healthy fibre line is rare, while loss on a radio link tracks signal strength and cell congestion.

The counter-intuitive part: the radio leg itself is not the big cost. The deeper jitter buffer that a variable 5G link forces on your PBX is. Fix the variability and 5G voice gets most of the way to fibre.

What 5G does to your MOS score and codec

MOS, the mean opinion score, grades call quality from 1 to 5. The codec sets the ceiling; the line decides how far below it you land.

On a perfect connection, G.711 – the uncompressed 64 kbps codec that most South African SIP trunks default to – scores a MOS of 4.1. Compressed G.729 tops out at 3.92 on one eighth of the bandwidth. Those ceilings are fixed. Packet loss, jitter-buffer discards and delay all subtract from them, which is why the same trunk can sound like a 4.1 on fibre at 9am and a 3.2 on a congested cell at 5pm.

Transcoding makes it worse. Cisco’s codec testing shows a G.729 stream re-encoded three times falls from 3.92 to 2.68, well into “unacceptable” territory. A call that hops from a mobile network onto your 5G-fed PBX and out again through a compressed trunk can rack up those hops without anyone noticing until customers complain.

The modern escape hatch is Opus, the codec inside most softphone and browser-based clients. It scales from 6 to 510 kbps, switches sampling rates on the fly and has packet loss concealment built in, so it degrades gracefully exactly where a 5G link is weakest. If your team calls through apps rather than desk phones, 5G’s disadvantage shrinks further.

Fibre vs 5G, measured in South Africa

The theory says fibre should win on latency and stability. The measured data says the same, but by less than the marketing on either side suggests.

MyBroadband Insights’ 2025 analysis of over a million speed tests put South Africa’s mobile networks between 28.78 ms (Telkom) and 35.78 ms (Cell C) average latency, with Rain at 30.36 ms, MTN at 29.81 ms and Vodacom at 30.81 ms. Average jitter ran from 6.11 ms (Rain) to 8.66 ms (Vodacom) – all comfortably inside the 30 ms voice target. These are network-wide mobile averages rather than 5G-only figures, but they are the best like-for-like public measurement we have.

Fibre is in another bracket. The same lab’s Q1 2025 fibre report, built on 793,772 tests, clocked the top ten ISPs between 7.41 ms (Cybersmart) and 14.44 ms (MWEB). That is roughly a third of the mobile figure, and, more importantly for voice, it barely moves between 9am and 9pm because a glass line is not shared with every phone in the suburb.

The CGNAT row deserves a flag. Many cellular and fixed-wireless products put you behind carrier-grade NAT with no public IP, which ITWeb notes complicates VPNs, remote access and high-quality VoIP hosting. Registration-based SIP extensions mostly work fine behind CGNAT; hosting your own PBX, remote handsets or anything that needs an inbound connection gets painful.

Rand-for-rand pricing is a separate fight, and we have already run it: see our Rain 5G vs fibre head-to-head for what each option actually costs per month.

When 5G is genuinely fine for a phone system

The measured numbers back a position most fibre marketing will not: a small office on strong 5G signal will not notice the difference on the average call.

At 30 ms average latency and single-digit average jitter, SA mobile networks sit well inside the ITU and Cisco voice thresholds. If you run five extensions through a hosted PBX, call mostly during the day and have solid signal at the router, 5G will carry your voice acceptably, and it will do it this week rather than after an installation lead time. The same logic makes 5G the obvious choice for temporary sites, fibre-poor industrial parks and any office you plan to leave within a year.

Where 5G stops being fine: call-heavy operations. A contact centre stacks dozens of concurrent calls on the line all day, and the failure mode of a radio link – peak-hour congestion, exactly when your queue is longest – is the worst possible one for that business. Voice at scale needs the flat, symmetric, SLA-backed profile that business fibre sells.

Keep reading