Infrastructure

SIP trunking

SIP trunking delivers telephone calls over IP using SIP signaling and RTP media, in place of a dedicated TDM circuit such as a PRI. Capacity is counted in concurrent call paths, not in copper channels.

Updated August 27, 2026

A trunk is a shared path that carries many calls between a PBX (or contact center) and the public network. For decades that path was TDM: a Primary Rate Interface (PRI) on a T1 (23 bearer channels + 1 signaling, North America) or E1 (30+1, much of the rest of the world). A SIP trunk is the same job on IP. The PBX sends SIP INVITEs to a provider. The provider interconnects to the PSTN and to other IP networks. DID numbers still exist, but they are no longer tied to a physical circuit; they are inventory on the trunk group or a linked number range.

What replacing PRI actually changes

ISDN PRI still exists at some [[pstn|PSTN]] edges. Many "SIP trunks" still TDM-handoff inside the carrier. The customer-facing interface is SIP.
PRI (T1 example)SIP trunk
PhysicalT1/E1 copper or TDM over fiber into a gateway/PBX cardEthernet / internet / private IP. No channel bank required at the site.
Capacity unit23 (T1) or 30 (E1) simultaneous calls per circuit. More calls means more PRIs.Concurrent call paths (CCPs) licensed in software. Burst is a contract term, not a new span.
NumbersDIDs mapped onto that circuit's D-channel / provider orderDIDs assigned in the provider portal, portable via LNP, independent of call-path count.
SignalingISDN Q.931 on the D channelSIP (RFC 3261) plus SDP. Media is RTP, with the same codec choices as any VoIP call.
Failure domainCut span takes 23/30 calls downIP outage, NAT, or SBC misconfig. Failover is a second trunk or a DNS / SRV pair, not a second T1 by default.

Concurrent call paths

A call path is one active session (inbound or outbound, unless the contract splits them). A company with 50 users rarely needs 50 paths. Erlang tables and measured peak CCS still apply. Oversubscription is normal. Under-buying shows up as 486 Busy or a provider SIP 503 on the Nth call. Paths are not the same as minutes: per-minute origination and termination billing applies on top of the path limit.

  • G.711 (μ-law/A-law) is the PSTN-interop default. HD codecs (G.722, Opus) work only where both the trunk and the far end agree.
  • Fax and alarm panels still fail on aggressive transcoding. T.38 or a leftover analog gateway is the usual fix.
  • Emergency calling: a SIP trunk to a hosted PBX does not by itself satisfy E911 / MLTS duties. Location is a PBX and user-data problem.

BYOC (Bring Your Own Carrier)

BYOC means the collaboration or contact-center vendor does not have to be the PSTN carrier. The customer points the UCaaS or CPaaS at a SIP trunk they buy elsewhere (or already own). The cloud app handles registrations, WebRTC clients, recording, and routing. The trunk handles origination and termination to the PSTN. Split responsibility is the point: number portability, rates, and STIR signing stay with the carrier you chose. The application vendor is not a party to your carrier contract.

What BYOC changes operationally

Identity
STIR/SHAKEN attestation is applied by the originating carrier on that trunk, not by the UCaaS brand.
Numbers
DIDs live with the trunk provider unless you host numbers in the app and hairpin. Mixing both is a common source of wrong caller ID.
Support boundary
One-way audio is usually a NAT or SBC problem on the customer side, not an outage at the hosted PBX.

The session border controller

An SBC is the SIP firewall and interworking element at the edge. It sits between the PBX (or UCaaS BYOC port) and the SIP trunk provider. Typical jobs:

  • TLS/SRTP or at least topology hiding so internal IP addresses never appear in SDP toward the internet.
  • NAT traversal for RTP through media latching. The SBC replaces the router's SIP ALG, which should stay off on consumer routers.
  • Codec interworking (transcoding) when the PBX offers Opus and the trunk only allows G.711/G.729.
  • Registration or static IP authentication toward the provider. Rate limits and malformed-message protection.
  • Emergency and duplicate-dialog hygiene (REFER, 3xx, forked INVITEs).

Hosted PBX users often consume a provider-managed SBC they never see. On-premises PBX plus internet SIP trunk without an SBC is how credential-stuffing toll fraud happens: a guessed REGISTER becomes outbound international traffic.

Typical architecture

  1. User device: desk phone (SIP REGISTER) or browser (WebRTC).
  2. PBX or UCaaS: dial plan, IVR, hunts, queues.
  3. SBC (customer or vendor): security, NAT, transcoding.
  4. SIP trunk provider: originate and terminate on the PSTN, number hosting, often SHAKEN signing.
  5. PSTN or peer IP networks: the far-end phone.

Quality and identity on a trunk

Call quality on a SIP trunk is IP quality (latency, jitter, loss) plus codec choice. QoS marking (DSCP EF for RTP) matters on managed WANs and does nothing on an unmarked residential cable hop. Caller ID and STIR/SHAKEN are only as good as the originating provider's policy for the numbers you send. A trunk that lets you assert arbitrary NANP numbers is a spoofing and analytics risk, not a feature.