Fax Over IP: T.38 vs G.711, What Survives VoIP

Fax survives VoIP in exactly two ways: T.38 fax relay, which converts the fax into data packets built for IP networks, or G.711 pass-through, which carries the raw fax tones inside an uncompressed audio stream with every voice-optimization feature turned off. Everything else — compressed codecs, default ATA settings, “it worked on the old line” wishful thinking — produces the familiar symptoms: faxes that die at page two, endless retrains, and machines that report “no answer” on a line that rings fine for voice.

This guide explains why a protocol designed in the analog era breaks on packet networks, how each survival path works, what the ATA actually has to do, and the honest cases where keeping fax on an analog-style line is the better engineering decision. It is an informational guide from a carrier that spends a lot of time on the trunk side of these problems.

Why fax breaks on VoIP

A fax machine is a modem. It encodes image data as precisely timed audio tones and expects the phone network to deliver those tones continuously, in order, without gaps. The T.30 protocol that governs Group 3 fax was designed for circuit-switched analog lines, and it enforces strict timers — command sequences that must complete within a few seconds, with a limited number of retries before one side abandons the call.

IP networks violate those assumptions in three ways. First, timing: packets arrive with variable delay (jitter), and T.30’s timers do not tolerate late responses the way a human conversation does. Second, compression: low-bitrate voice codecs like G.729 are engineered to reproduce speech intelligibly, not to preserve modem tones — a compressed codec anywhere in the path destroys the fax signal. Our G.711 vs G.729 comparison covers why the two codecs behave so differently.

Third, packet loss: a voice call shrugs off a lost packet as a click; a fax interprets the gap as line corruption. Lost packets force speed retrains, truncate pages, or kill the session outright when a T.30 command and its retries all disappear. If your network drops even a fraction of a percent of packets, fix that first — our guide to diagnosing and fixing packet loss walks through the process.

There is a fourth, quieter saboteur: voice-optimization features. Echo cancellation and voice activity detection (silence suppression) are helpful for speech and hostile to modems. Echo cancellers distort tone phases; VAD interprets the pauses in a fax handshake as silence and stops sending audio, cutting the signal mid-negotiation.

Path one: T.38 fax relay

T.38 is the ITU-T recommendation for real-time Group 3 fax over IP networks, first approved in 1998. Instead of carrying fax tones as audio, the gateway or ATA at each end demodulates the fax signal back into its underlying data, transports that data as T.38 packets, and remodulates it into tones for the receiving fax machine. The fax machines still think they are talking T.30 over an analog line; the IP network in the middle carries data, not sound.

This solves the codec problem completely — there is no audio stream to compress — and it blunts the packet-loss problem, because T.38 over UDPTL transport supports redundancy: each packet can carry copies of previous data, so a single lost packet does not lose information. Timing is also managed, since the endpoints can spoof T.30 keepalives locally while the network catches up.

A typical T.38 call starts as a normal voice call. When the receiving side hears fax tones (the CED answer tone), the endpoints negotiate a mid-call switch — a SIP re-INVITE — from the voice codec to T.38. That re-negotiation is itself a failure point: every device in the path (ATA, PBX, SBC, carrier gateway) must support T.38 and agree to the switch. One device that answers the re-INVITE incorrectly, or a router with a broken SIP ALG rewriting the signaling, and the call falls back to audio or drops.

The speed caveat: classic T.38 relays fax at up to 14,400 bps (V.17). “Super G3” machines that negotiate 33,600 bps over V.34 need T.38 version 3, published in 2007 — and both endpoints plus the carrier path must support it. In practice, many deployments simply let Super G3 machines fall back to 14.4 kbps, which adds transmission time but works reliably.

Path two: G.711 pass-through

G.711 pass-through (sometimes called fax over G.711, or informally “G.711u fallback”) skips the relay entirely. The fax tones travel inside a standard G.711 audio stream — the uncompressed 64 kbps codec that reproduces the analog waveform faithfully enough for a modem to survive, if the path is clean.

“If the path is clean” is doing heavy lifting. For pass-through to work, every device in the chain must cooperate:

• Codec locked to G.711 end to end — no transcoding to a compressed codec anywhere, including inside the carrier network.
• Echo cancellation disabled for the call, or the canceller mangles the modem tones.
• VAD / silence suppression disabled, or the handshake pauses get clipped.
• Jitter buffer set to a fixed (static) mode where the equipment supports it — adaptive buffers resize mid-call and disturb timing.
• Near-zero packet loss, because there is no redundancy layer; every lost packet is lost signal.

The appeal of pass-through is simplicity: no T.38 negotiation, no version mismatches, nothing for a misbehaving SBC to break mid-call. On a low-latency, low-loss connection — wired business internet, sane router, no Wi-Fi hops — it is often the more reliable option in practice, precisely because there is less machinery to go wrong. On a lossy or congested connection it is the less reliable option, because it has no defenses at all.

T.38 vs G.711: which one should you run?

There is no universal winner — the right answer depends on your network and your carrier path. A fair summary of the trade:

Factor	T.38 relay	G.711 pass-through
Packet loss tolerance	Better (redundancy)	Poor (none)
Codec/transcoding risk	Immune	Fatal if present
Negotiation complexity	High (re-INVITE, versions)	Low
Super G3 33.6 kbps	Needs T.38 v3 support	Possible on a very clean path
Best on	Lossy or long paths	Clean, low-latency paths

A pragmatic configuration order: ask your trunk provider what they actually support and test both. If T.38 is supported end to end, prefer it — it is the purpose-built tool. If T.38 negotiation proves flaky across your path, lock the ATA to G.711, disable echo cancellation and VAD, and verify your network is clean. Many ATAs can be set to attempt T.38 and fall back to pass-through, which covers both cases at the cost of some negotiation complexity.

The ATA is where fax over IP succeeds or fails

Almost every fax-over-IP deployment has an analog telephone adapter (ATA) between the fax machine and the network — the fax machine speaks analog, the ATA speaks SIP. The ATA’s fax-specific settings decide the outcome, and the factory defaults are usually tuned for voice, not fax.

The settings that matter, whatever the vendor calls them: fax mode or fax event handling (T.38 vs pass-through vs auto-detect), codec preference (pin G.711), echo cancellation (off for fax ports), VAD/silence suppression (off), jitter buffer (fixed, generously sized), and network jitter level. Plug the fax machine directly into the ATA — not through a splitter shared with other devices — and put the ATA on wired Ethernet, never Wi-Fi.

Two realities worth stating plainly. First, an ATA cannot fix a bad network; it can only stop making things worse. If jitter and latency are out of spec, fax fails no matter what the ATA does. Second, multifunction printers with built-in fax are often the least tolerant endpoints — some negotiate Super G3 aggressively and handle fallback badly. Setting the fax machine itself to a lower maximum speed (14.4 kbps) and disabling ECM (error correction mode) on stubborn machines are legitimate field fixes, not hacks.

When analog is still the right answer

Honest engineering includes knowing when not to convert. Some fax workloads should stay on an analog-style line — and with carriers retiring copper POTS lines, that increasingly means a managed POTS replacement service that presents an analog port to the device while handling transport behind the scenes, rather than a literal copper pair.

The cases where the analog-port route wins: fax machines wired into compliance-critical workflows where a failed transmission has real consequences; sites with unreliable internet where a fax-over-IP path would inherit every outage; and mixed analog device pools — fax plus alarm panels plus elevator phones — where one purpose-built gateway is simpler than converting each device separately. Our POTS replacement guide covers how those lines migrate and which transport suits which device.

Medical and dental offices are the canonical example: referral and records workflows still move by fax in much of US healthcare, and a fax that silently truncates page three of a patient chart is a workflow failure, not an inconvenience. Practices modernizing their phones — the subject of our medical office phone system overview — often move voice to SIP first and keep fax on a dedicated, properly configured path (T.38-tested ATA or POTS replacement) rather than treating it as an afterthought on the same box.

The decision is not VoIP-versus-fax loyalty. It is: does this fax volume justify engineering a clean T.38/G.711 path, or is a dedicated analog-port service the cheaper, more reliable answer for one machine that sends four faxes a week?

Frequently asked questions

What is T.38 fax relay and how does it work?

T.38 is the ITU-T standard for real-time Group 3 fax over IP networks, approved in 1998. Instead of sending fax tones as audio, the gateway or ATA demodulates the fax signal into data, transports it as T.38 packets (with redundancy over UDPTL transport), and remodulates tones for the receiving machine. The fax endpoints still speak T.30 as if on an analog line. Calls typically start as voice and switch to T.38 via a SIP re-INVITE when fax tones are detected — which means every device in the path must support the negotiation. Broken middleboxes are a common failure point; see our SIP ALG explainer for one frequent culprit.

Should I use T.38 or G.711 pass-through for faxing?

Prefer T.38 when your ATA, PBX, and trunk provider all support it end to end — its redundancy tolerates packet loss and it is immune to codec problems. Prefer G.711 pass-through when T.38 negotiation is flaky across your path and your network is clean: lock the codec to G.711, disable echo cancellation and silence suppression, use a fixed jitter buffer, and confirm near-zero packet loss (our packet loss guide shows how to measure it). Many ATAs can attempt T.38 and fall back to pass-through, covering both cases.

Why does my fax fail halfway through a page on VoIP?

Mid-page failures usually mean packet loss or jitter corrupted the image data stream after a successful handshake. The T.30 protocol retries failed commands a limited number of times within strict timers; when the retries are exhausted, the call drops and the receiving side keeps a truncated document. Check for packet loss first, then verify VAD/silence suppression and echo cancellation are disabled on the ATA’s fax port, and consider forcing the machines down to 14,400 bps — slower speeds are markedly more tolerant of imperfect lines. Adaptive jitter buffers resizing mid-call can also disturb the modem timing; use a fixed buffer where the ATA supports it.

Does Super G3 33.6 kbps fax work over IP?

Sometimes, and only with the right support. Super G3 machines negotiate V.34 modulation at up to 33,600 bps. Classic T.38 tops out at 14,400 bps (V.17); carrying V.34 over T.38 requires version 3 of the standard, published in 2007, supported by both endpoints and the carrier path. G.711 pass-through can carry V.34 on a very clean, low-latency connection, but it is fragile. The common field practice is to let Super G3 machines fall back to 14.4 kbps — transmissions take longer but complete reliably, which is the metric that actually matters.

Can I keep my fax machine when my copper phone line is retired?

Yes — the machine does not need copper, it needs an analog port with a transport behind it that preserves modem signals. Two paths work: a properly configured ATA on a SIP line running T.38 relay or G.711 pass-through, or a managed POTS replacement service that presents a standard analog jack and handles the fax-safe transport for you. The second option suits sites with several analog devices (fax, alarm panels, elevator phones) or low fax volume that does not justify per-device tuning. Our POTS replacement guide maps each legacy line type to its modern path.

---

Getting fax across an IP network starts with a trunk that treats it as an engineering problem, not an afterthought. SIPNEX provides business SIP trunking as an FCC-licensed carrier — G.711 end to end when your fax path needs it, clean routing, and real answers about what the network supports before you cut over. Talk through your fax and analog-line migration at (833) 665-2220 or contact our team.