VoIP Bandwidth & SIP Concurrent Call Calculator

Engineering Theory: VoIP Protocol Encapsulation & WAN Sizing

1. The Anatomy of Packetized Voice: Why a 64 kbps Codec Consumes 87+ kbps

A frequent source of capacity miscalculation in enterprise wide-area network (WAN) provisioning is the assumption that 24 channels of uncompressed G.711 voice consume exactly 1.536 Mbps (24 × 64 kbps). While 64 kbps accurately reflects the raw pulse-code modulation (PCM) sampling rate (8,000 samples/sec × 8 bits/sample according to Nyquist's theorem), audio data cannot traverse modern packet-switched IP networks without being encapsulated into discrete protocol envelopes at Layers 4, 3, and 2.

To transmit conversational audio, real-time voice packets incur protocol header taxes at each architectural layer:

• Audio Voice Payload (160 Bytes at 20 ms ptime): Represents 20 milliseconds of 64 kbps uncompressed digitized speech (G.711 µ-law or A-law).
• Real-Time Transport Protocol (RTP — RFC 3550, 12 Bytes): Positioned on top of UDP, RTP provides sequential packet numbering (16-bit sequence number), presentation timestamps (32-bit timestamp), synchronization source identifiers (32-bit SSRC), and dynamic payload type identification to enable jitter buffering and lost-packet concealment at the receiver.
• User Datagram Protocol (UDP — RFC 768, 8 Bytes): Voice communication strictly utilizes connectionless UDP rather than TCP. Because voice is time-critical, retransmitting a dropped packet via TCP acknowledgments would introduce hundreds of milliseconds of latency, creating acoustic disorientation far more destructive than a momentary speech clip. UDP provides lightweight port-based multiplexing and an optional checksum.
• Internet Protocol (IPv4 RFC 791, 20 Bytes / IPv6 RFC 8200, 40 Bytes): Handles packet routing, time-to-live (TTL) counters, and Quality of Service (QoS) markings via Differentiated Services Code Points (DSCP EF — Expedited Forwarding, binary 101110, decimal 46). Under modern IPv6 networks, the base header expands from 20 to 40 bytes, introducing a 100% layer-3 header penalty.
• Data Link Layer (Ethernet IEEE 802.3, 18 Bytes): Composed of a 14-byte MAC header (destination MAC, source MAC, EtherType) and a 4-byte Frame Check Sequence (FCS/CRC). If 802.1Q VLAN trunking or MPLS label switching is enabled, an additional 4 bytes per tag is appended.
• Physical Layer Framing (Ethernet L1, 20 Bytes): Includes the 7-byte Ethernet Preamble, 1-byte Start Frame Delimiter (SFD), and the mandatory 12-byte Inter-Packet Gap (IPG / 96 bit times) required on physical wire links.

2. The Impact of Packetization Interval (ptime: 10ms vs. 20ms vs. 30ms)

The packetization interval (abbreviated as ptime in SIP Session Description Protocol / SDP negotiations, RFC 4566) defines the duration of analog acoustic audio encapsulated within each individual IP packet. The packet transmission rate is governed by the inverse frequency relationship:

Packets Per Second (PPS) = 1,000 / ptime (ms)

Varying the packetization interval presents a fundamental trade-off between acoustic latency and bandwidth consumption:

• 10 ms Packetization (100 PPS): Encapsulates only 80 bytes of G.711 audio per datagram. Halving the frame size reduces algorithmic packetization delay from 20ms down to 10ms, delivering crisp acoustic responsiveness for ultra-low latency contact centers and financial trading floors. However, because the fixed 40-byte IP/UDP/RTP header must be transmitted 100 times per second instead of 50, per-call Ethernet bandwidth surges from 87.2 kbps up to 110.4 kbps (+26.6% bandwidth penalty). Furthermore, edge routers must process double the packet interrupt load, creating CPU exhaustion on branch gateways.
• 20 ms Packetization (50 PPS): The global enterprise standard (ITU-T recommendation). Transmits 50 packets per second, carrying 160 bytes of G.711 audio. Achieves the optimal equilibrium between perceptual mouth-to-ear latency (remaining well under the ITU-T G.114 150 ms one-way budget) and wire efficiency.
• 30 ms Packetization (33.3 PPS): Transmits 240 bytes of G.711 audio per packet. Lowers per-call Ethernet bandwidth to 79.5 kbps. While highly beneficial on constrained satellite, cellular backhaul, or sub-rate leased circuits, the loss of a single 30ms packet results in a noticeable speech dropout that challenges DSP packet loss concealment (PLC) algorithms.

3. Compressed RTP (cRTP RFC 2508) & Low-Speed WAN Links

On low-bitrate codecs such as G.729 (8 kbps), a standard 20ms packet contains only 20 bytes of audio payload. Placing a 40-byte IP/UDP/RTP header onto a 20-byte payload yields an alarming 200% protocol tax.

To resolve this inefficiency over low-bandwidth serial, frame relay, or satellite links, the IETF developed Compressed RTP (cRTP — RFC 2508). cRTP recognizes that throughout a persistent SIP media session, the majority of header fields remain entirely static (Source IP, Destination IP, UDP Ports, SSRC, and Payload Type), while sequence numbers and timestamps increment by predictable, constant deltas. By maintaining first-order difference tables at both ends of a point-to-point hop, cRTP compresses the 40-byte header down to:

• 2 Bytes: When UDP checksums are disabled or omitted on trusted point-to-point links.
• 4 Bytes: When the 2-byte UDP checksum is retained for end-to-end data integrity.

Enabling cRTP on a 20ms G.729 call slashes single-call bandwidth from 31.2 kbps down to 11.2 kbps, unlocking up to 64% bandwidth savings and allowing enterprise WAN designers to pack triple the concurrent calls onto fractional T1 or sub-rate radio links.

4. Voice Activity Detection (VAD) & Silence Suppression

Human telephonic dialogue is inherently half-duplex: while one party speaks, the other listens. Empirical conversational analysis conducted by Bell Labs and the ITU indicates that a typical voice channel is active only 35% to 45% of the time, with the remaining 55% to 65% consumed by listening pauses, conversational turn-taking, and sentence hesitations.

Voice Activity Detection (VAD) capitalizes on this natural pause structure. When an endpoint digital signal processor (DSP) detects signal energy dropping below an ambient noise threshold:

1. The media engine ceases transmission of full 50 PPS audio RTP frames.
2. The transmitter emits sporadic, lightweight Silence Insertion Descriptor (SID — RFC 3389) packets (typically 1 to 2 packets per second).
3. The receiver's DSP reads the background spectral signature inside the SID frame and synthesizes matching, non-distracting Comfort Noise Generation (CNG). This prevents callers from experiencing the disconcerting "dead-air" phenomenon where callers assume the connection was abruptly dropped.