A traditional phone call reserved a continuous electrical path from caller to callee for the duration of the conversation. That model, circuit switching, is how the public switched telephone network grew through the 20th century. VoIP does the opposite: it samples the microphone, compresses the samples with a codec, and ships them as discrete packets that share the same pipes as web traffic, email, and video.
The idea is older than broadband. Experimental packet voice ran on early ARPANET and Internet links in the 1970s and 1980s. Commercial VoIP became practical once access bandwidth, cheap digital signal processors, and interoperable signaling (chiefly Session Initiation Protocol, IETF RFC 3261) were widely available. Today most 'phone' calls that touch a business already spend part of their life as IP, even when one party still uses a copper pair.
From circuits to packets
Circuit switching guarantees a fixed bit rate for the call and keeps delay low and stable. The cost is inefficiency: silence still occupies the circuit, and a transcontinental call might lock capacity on many tandem switches. Packet switching multiplexes many conversations onto shared links. Each packet carries a destination address, and each router forwards it independently. Nothing is reserved end to end unless a quality-of-service (QoS) policy says otherwise.
Audio does not tolerate that freedom as well as a file download does. Real-time media uses the Real-time Transport Protocol (RTP, IETF RFC 3550) over UDP in almost every deployment. RTP adds sequence numbers and timestamps so the receiver can reorder packets, detect loss, and play samples at a steady rate. A companion protocol, RTCP, carries quality reports. The trade-off is explicit: if the network is congested, packets are delayed or dropped rather than waiting in a reliable TCP queue.
- Capture: analog speech is sampled (8 kHz for narrowband telephony, 16 kHz or higher for HD Voice).
- Encode: a codec such as G.711 or Opus produces a bitstream.
- Packetize: typically 20 ms of audio per RTP packet, plus IP/UDP/RTP headers.
- Transport: the packets traverse LANs, WANs, and often a provider core before reconstruction.
How a VoIP call is set up
RTP only moves media. Something else has to find the other party, negotiate codecs, and tear the session down. That control plane is usually SIP. A User Agent sends an INVITE, the far side answers with SDP (Session Description Protocol) that lists IP addresses, ports, and codecs, and both sides then send RTP to those addresses. WebRTC uses the same media stack (ICE, DTLS-SRTP, RTP) with a different signaling channel, often JSON over HTTPS or a proprietary socket.
Because signaling and media take different paths, a call can succeed at the SIP layer and still sound poor if the RTP path is lossy, or fail to ring if a firewall blocks SIP but would have passed media. Diagnosing VoIP means looking at both planes.
Typical architecture
A business VoIP path is a chain, not a single hop. A desk phone, softphone, or browser client registers with a provider or private PBX. That system applies routing rules (IVR, hunt groups, queues, forwarding). If the destination is not another IP endpoint on the same network, a media gateway or SIP trunk interconnects with the PSTN so the call can reach a mobile or landline number in E.164 format.
- Device: SIP phone, mobile app, or WebRTC browser tab captures and plays audio.
- Access network: LAN, Wi-Fi, or cellular data with optional QoS (DSCP EF for voice).
- Provider or PBX: registration, dial plan, recording, and business logic. Many of these platforms are sold as CPaaS.
- Interconnect: SIP trunks or TDM gateways toward other carriers.
- PSTN or peering: the last hop to a number that is not native IP.
Carriers bill that last hop as origination or termination depending on direction. The two terms are used inconsistently: a wholesale voice contract and a CPaaS product catalog may use the same word for opposite legs of a call. See the origination and termination guide before you compare quotes.
VoIP compared with landline
| Aspect | Classic landline (PSTN access) | VoIP |
|---|---|---|
| Switching model | Circuit: a path is reserved for the call | Packet: speech shares IP capacity with other traffic |
| Signaling | SS7 / analog loop signaling | SIP, or WebRTC signaling plus RTP |
| Audio bandwidth | Typically 300-3400 Hz (narrowband) | Narrowband or wideband / HD when both ends agree |
| Numbering | E.164 via the PSTN | Same E.164 numbers via DIDs, plus URI addressing |
| Power and access | Line power from the central office (legacy copper) | Depends on local power and the data connection |
| Emergency calling | Native E911 association to a wireline address | Requires registered location and a compliant provider |
| Typical failure mode | Cable cut or switch outage | Packet loss, NAT, DNS, or congestion (see call quality) |
Benefits and limits
The engineering case for VoIP is multiplexing and software. One broadband circuit can carry many simultaneous calls. Features that required a Class 5 switch (voicemail, recording, menus, presence) become application logic. International calling becomes a matter of routing and rates rather than a dedicated foreign-exchange line. HD Voice is possible when the path never downsamples to G.711.
The limits are physical and operational. Voice is sensitive to one-way delay, jitter, and loss (ITU-T G.114 and related guidance; see call quality). A power cut or ISP outage takes the phones with it unless there is backup power and a secondary path. NAT and firewalls still break signaling or media if STUN/TURN or a Session Border Controller is missing. Interoperability with the PSTN reintroduces analog impairments, DTMF transport quirks, and Caller ID rules that IP-only calling can ignore.
Where VoIP meets the PSTN
Pure IP calling (app to app, or SIP URI to SIP URI) never needs the PSTN. Most business calls still do, because customers dial numbers. The handoff is a gateway that translates SIP/RTP to SS7 and PCM, or a SIP peering session with a carrier that performs that translation further downstream. Codec choice often collapses to G.711 at that border. That is why an HD call between two VoIP users can fall back to narrowband the moment one side is a mobile number on a network that does not pass wideband.
Understanding VoIP in practice means following a call across those borders: device, provider, interconnect, and PSTN, then the same chain in reverse for the return audio.