How a phone call works: from dial to audio, PSTN to VoIP

A phone call begins when a device sends a destination and call-setup request to a telephone network, which authenticates the origin, resolves the dialed number, selects a route, alerts the destination, and reports whether it answers. After answer, a separate bearer path carries audio in both directions until one party hangs up and signaling releases the resources. On older networks that bearer is a reserved circuit; on modern IP networks it is usually a timed stream of voice packets, even when the caller sees an ordinary phone number.

A call is signaling plus a bearer

The word “call” hides two coordinated systems. Signaling conveys intent and state: who is calling, which destination is requested, whether the far end is ringing or busy, whether the call was answered, and when it ended. The bearer carries the conversation itself. In traditional telephony the bearer is a circuit or timeslot; in VoIP it is an audio packet stream.

Keeping these planes separate explains common failures. A caller can hear ringing generated by an intermediate network even though no end-to-end audio path exists yet. A call can show as answered while one direction of media is blocked. Engineers diagnose phone calls by following signaling and media independently, then correlating them by timestamps and call identifiers.

Step 1: the caller enters a destination

The digits a person enters are not always the digits the network routes. A handset, PBX, or carrier applies a dial plan: it may add a country code, remove an outside-line prefix, expand an extension, or reject a prohibited destination. Good systems normalize public numbers to E.164 format, an ITU-T numbering structure that combines country code and national destination digits within a maximum of fifteen digits.

E.164 identifies the destination but does not prescribe a packet route. It is closer to a globally interpretable key. Networks maintain routing data that maps number ranges or individual ported numbers to the next responsible network. Internal extensions and SIP URIs can be resolved without entering the public number hierarchy at all.

On an analog line, dialing once produced pulses or combinations of tones that the local exchange interpreted. Modern devices generally collect the destination first and submit a complete setup request to a PBX or carrier. Mobile phones send call-control messages through the radio access and mobile core. SIP phones send an IP signaling message. The user action is similar, but the first network hop differs.

Step 2: the originating network authenticates and applies policy

Before routing, the originating service identifies the line, device, subscriber, or trunk making the request. It checks whether calling is allowed, whether the destination is valid, which calling identity may be presented, and what route class applies. A business PBX may rewrite digits and choose among multiple trunks. A mobile core may verify subscriber state and service permissions.

The presented calling identity is distinct from the authenticated origin. A business can legitimately present a main callback number while placing the call from another internal extension, but the originating provider should still know which customer submitted it. Modern caller-authentication frameworks can attach a signed assertion about the provider's relationship to the calling identity. That attestation travels in signaling and can inform downstream verification, but it does not decide by itself what name a screen displays or whether analytics will label the call.

Policy can stop the call here. An invalid number, blocked destination, missing permission, exhausted capacity, or authentication failure may cause an immediate error without any request reaching the called party.

Step 3: the network resolves and routes the number

Telephone routing is hop by hop. The originating switch or proxy chooses a next network based on the normalized destination, commercial interconnects, portability data, policy, and availability. That next network may route directly to the terminating provider or pass the call to another transit network.

Number portability makes simple prefix routing incomplete. A number can retain its familiar digits after moving between providers, so a routing lookup may be needed to learn its current network. Toll-free numbers introduce another mapping layer: the dialed number resolves to routing instructions that can vary by time, caller location, or service policy. Enterprise numbers may ultimately map to a SIP trunk, PBX, queue, or application.

Each hop creates its own operational view of the same attempt. One network's outgoing leg becomes the next network's incoming leg, and identifiers may be added or translated at boundaries. End-to-end troubleshooting therefore needs more than one local call ID. Timestamps, calling and called numbers, trunk identifiers, and propagated correlation values help operators join the legs.

Routing is not necessarily symmetric. The return audio packets may follow a different IP route, and a later call in the opposite direction may use entirely different carriers. “It works outbound” does not prove the inbound route is configured correctly.

Step 4: the destination is located and alerted

The terminating network resolves the public number to a subscriber or service. It might page a mobile device, ring an analog line, send a SIP invitation to a PBX, or invoke routing logic that tries several employees. The destination can accept, reject, redirect, or simply fail to respond.

Meanwhile, progress information travels back toward the caller. Common states include trying, ringing, busy, and unavailable. In SIP these are explicit responses such as 100 Trying, 180 Ringing, and a range of final status codes. Traditional telephone signaling carries equivalent call-progress information in another form.

What the caller hears during this phase is early media or locally generated progress audio. Ringback may be produced by the originating network based on a ringing indication; it is not necessarily the sound of the far handset. A downstream network can instead send early media containing an announcement such as a changed number or insufficient balance. If the signaling and early-media policies disagree, callers can hear silence or miss a useful announcement.

Answer is a billing and state boundary in many systems. A human lifting a handset, an automated attendant accepting the call, or a voicemail system answering can all produce the same fundamental network transition. Signaling alone may show that a destination answered, not whether it was the intended person.

Step 5: the networks establish the audio path

On the traditional digital PSTN, a call reserves a 64 kbit/s channel carrying pulse-code-modulated voice between switching points. Time-division multiplexing gives each call a repeating timeslot. Signaling tells switches which incoming and outgoing timeslots to connect, producing a temporary circuit through the network.

On an IP call, the endpoints or gateways negotiate a codec and exchange network addresses and ports. Audio samples are encoded into short frames, wrapped in transport packets, and sent at a steady cadence. RTP commonly supplies sequence numbers and timestamps so the receiver can detect missing packets, reorder limited out-of-order arrivals, and play audio at the correct rate.

Packet audio is not delivered like a file. Waiting for every missing packet would make live conversation stall, so late packets are usually discarded after their playout deadline. A jitter buffer delays playback slightly to absorb variation in arrival time. More buffering tolerates more variation but increases conversational delay.

The two parties usually send independent streams, allowing full-duplex conversation. A problem in one direction does not imply a problem in the other. Firewalls, network address translation, an incorrect media address, or asymmetric routing can permit A-to-B packets while blocking B-to-A packets.

Where codecs and gateways enter

A codec defines how audio samples become bits. G.711 is common on narrowband telephony and directly represents companded pulse-code samples. G.722 carries wideband speech. Opus supports a broad set of packet voice and audio conditions. Endpoints advertise compatible codecs during setup, and an intersection is selected for each IP media leg.

When a call crosses between circuit and packet networks, a media gateway converts timeslot audio into packetized audio and back. When two packet legs choose different codecs, a transcoder decodes one format and encodes the other. Transcoding enables interconnection but adds processing and can reduce quality, especially across repeated lossy conversions.

A wideband first leg can also become narrowband at a gateway. Once frequencies are removed, a later wideband leg cannot restore them. For this reason, the actual call quality depends on the complete media path rather than the capabilities of either endpoint alone.

Many present-day calls are packetized for most of their journey even when they enter or leave through familiar telephone interfaces. “PSTN versus VoIP” is not always a clean end-to-end distinction; it often describes a particular segment and its control model.

DTMF and other information during the call

Pressing a telephone key creates DTMF: a pair of audio frequencies identifying the digit. On an analog or transparent audio path, the tones can travel in-band. On packet networks, codecs and packet loss may distort them, so systems often represent digits as named RTP telephone events under RFC 4733. SIP INFO is another method used in some environments.

The endpoints must agree on the method. If a gateway expects RTP events while the sender produces only compressed in-band tones, speech can work perfectly while an IVR ignores every keypress. Troubleshooting requires examining both the negotiated DTMF capability and what packets were actually sent.

Other mid-call operations use signaling to modify the session. Hold can change the advertised media direction or connect the caller to a media source. A conference creates several call legs mixed or selectively forwarded by a bridge. A transfer may keep the original system in the path and bridge a new outbound leg, or ask another endpoint to establish a replacement session.

Call recording typically occurs at an endpoint, PBX, media server, or carrier feature that has access to the audio. The signaling route alone does not provide the recording; a component must receive, duplicate, or reconstruct the media streams. Recording rules and required notices depend on jurisdiction and use, so the technical ability to record does not establish permission.

Step 6: hangup and teardown

When either party hangs up, its serving system sends a release signal. In SIP, an established dialog normally ends with BYE followed by a success response. A call attempt cancelled before answer uses a different transaction. Traditional networks exchange release messages and free the reserved circuit.

Every intermediate system should stop media, release ports or timeslots, close call state, and produce a completion record. The final reason matters. Normal hangup, busy, no answer, rejection, network failure, and timeout should not collapse into one generic “ended” status.

Teardown is not instantaneous everywhere. If a release message is lost or a gateway fails, one side may continue sending media until a session timer or inactivity rule expires. Well-configured systems use bounded timers as a safety net, but timers should not cut off a healthy call merely because one endpoint failed to refresh the dialog as expected.

Reading a failed call from the evidence

A disciplined investigation follows the call in order:

1. Was the dialed input normalized to the intended destination?
2. Did originating policy accept the attempt and choose the expected route?
3. Which network produced the final progress or failure indication?
4. Did the destination answer, redirect, or time out?
5. What media addresses and codecs were established on each leg?
6. Did packets or timeslot audio flow in both directions after answer?
7. Did a mid-call change coincide with the problem?
8. Who initiated release, and what cause did each boundary record?

This method avoids using a downstream symptom to guess at an upstream cause. “No audio,” for example, should first be divided into no answer, no negotiated media, no inbound packets, no outbound packets, or packets present but undecodable.

FAQ

Is the ringing sound carried from the called phone?

Often it is not. The originating network can generate ringback after receiving a signaling indication that the destination is being alerted. Some calls instead provide early media from a downstream system.

Does a phone call reserve one physical wire end to end?

Modern calls generally do not. Traditional digital segments reserve timeslots, while IP segments share network links and distinguish calls by packet addresses, ports, and session state. Gateways connect the segments.

Why can a connected call have one-way audio?

Signaling and media take separate paths, and the two audio directions are independent. An unreachable advertised address, firewall rule, or network-address translation error can block one stream even though setup succeeded.

What determines when a call is considered answered?

The terminating side sends an answer indication through the signaling path. That event can represent a person, voicemail, an automated attendant, or another system accepting the session; signaling does not inherently distinguish among them.