Ethernet Collisions
The past
Collisions occurred mainly in now‑obsolete Ethernet variants (10BASE5, 10BASE2, 10BASE‑T and 100BASE‑TX) that used coaxial media and repeaters, or twisted‑pair networks connected through hubs.
Switched Ethernet
Modern Ethernet networks, built around switches with full‑duplex links, rarely produce collisions, since every Ethernet segment is a point‑to‑point connection between a switch and a host, or between two switches — each direction having its own separate channel.
Today, a collision domain only becomes active when a faulty cable still falls back to half‑duplex, or when a switch is directly connected to a hub or a coax transceiver.
What is a collision?
If two packets appear on a medium (coax or UTP cable) at the same time, we speak of a collision.
During a collision, neither destination receives a complete frame.
Collisions are automatically detected and corrected in three steps:
Collision detection
On coax networks (10BASE5 or 10BASE2), collisions are detected by comparing transmitted data with received data, or by detecting higher signal amplitudes (the mixed signals can momentarily produce double amplitudes).
Jamming signal
A jamming signal is a medium‑filler: a 32‑bit pattern sent by transmitters after detecting a collision.
Its purpose is to inform other stations that a collision has occurred.
From Wikipedia (jam signal):
A station detecting a collision sends a 4‑ to 6‑byte pattern composed of sixteen 1‑0 bit combinations. The goal is to ensure that any node currently receiving a frame will receive the jam signal instead of the correct 32‑bit MAC CRC, causing the receiver to discard the frame due to a CRC error.
Backoff algorithm
Both senders need a method to transmit their packet without interfering with each other. This is done using random number generators.
Each sender chooses a random number. The typical inter‑frame gap is then multiplied by this number before attempting to access the medium again.
Since both participants will likely generate different numbers, one frame will be sent before the other, avoiding a new collision.
If a second collision occurs, the upper bound of the random generator is doubled, and the process repeats — now with twice as many chances of success. This doubling can occur several times in a row if needed.