Pick a codec and packet size and see the bandwidth per direction, per call and per site, with the header overhead broken down byte by byte.
Estimate
| Ethernet frame | 18 B |
| IPv4 header | 20 B |
| UDP header | 8 B |
| RTP header | 12 B |
| G.711 (PCMU / PCMA) payload (20 ms) | 160 B |
| Total per packet | 218 B |
Assumptions: constant bitrate at the codec's nominal rate, no silence suppression, no RTCP (adds about 5 percent), no VPN or PPPoE overhead, no retransmission. Opus is variable bitrate; real usage differs. Wi-Fi and cellular add their own framing. Plan headroom.
Bandwidth per direction equals bytes per packet times eight bits times packets per second. Bytes per packet are the codec payload for the chosen packetization plus the RTP header (12 bytes), the UDP header (8), the IP header (20 for IPv4, 40 for IPv6), optionally the SRTP authentication tag (10) and optionally Ethernet framing (14-byte header plus 4-byte frame check). Packets per second equal 1000 divided by the packetization in milliseconds.
Each 20 ms packet carries 160 bytes of audio plus 12 bytes RTP, 8 bytes UDP, 20 bytes IPv4 and 18 bytes Ethernet framing: 218 bytes, 50 times per second, is 87.2 kbit/s per direction.
Yes. Headers are paid per packet, so 40 ms packets halve the header overhead compared with 20 ms. The cost is more latency and a bigger gap when a packet is lost. 20 ms is the common default.
SRTP encrypts the payload in place and appends an authentication tag, 10 bytes by default. For G.711 at 20 ms that is about 4 kbit/s per direction.
Divide the uplink by the per-direction figure and keep headroom. With G.711 at 87 kbit/s and a 70 percent ceiling, roughly 80 calls. With Opus at 24 kbit/s the number is several times higher. Upload is usually the limit.
Opus and G.711 ulaw and alaw. G.722 and G.729 are included in the calculator for comparison but are not offered by SIPdia.