In the Data Link Layer, specifically the MAC sublayer, CSMA/CD was designed for classic shared, half-duplex Ethernet where multiple stations contend for one wired medium.2 Basic CSMA reduces collisions by listening first, but it cannot eliminate them because propagation delay means two distant stations may both sense an idle channel and begin transmitting almost simultaneously.2 CSMA/CD extends CSMA with a crucial idea: stations must keep listening while transmitting so they can detect a collision early, abort, send a jam indication, and retry later using randomized backoff.2

This mechanism is historically central to half-duplex Ethernet, hubs, and shared collision domains, though modern switched full-duplex Ethernet generally does not use collision detection in normal operation.2 Understanding CSMA/CD remains essential because it explains why Ethernet adopted a minimum frame size of 64 bytes, why slot time is 512 bit-times in 10/100 Mb/s half-duplex Ethernet, and how binary exponential backoff stabilizes repeated contention.3

Footnotes

1. Carrier-sense multiple access with collision detection - Wikipedia - Overview of CSMA/CD, jam signal, slot time, and half-duplex Ethernet behavior. ↩ ↩² ↩³ ↩⁴

2. [PDF] CSMA/CD - Explains Ethernet slot time, minimum frame size, late collisions, and retransmission behavior. ↩ ↩²

3. [PDF] Lecture 19: Ethernet / MAC - Alan Mislove - Summarizes Ethernet collision detection, slot time, jam signaling, and exponential backoff. ↩ ↩²

4. [PDF] 4. Media Access Control - IEEE 802 - IEEE-oriented description of collision handling, jam, and truncated binary exponential backoff. ↩ ↩²

5. [PDF] Tutorial for CSMA-CD - Provides slot-time interpretation and relation to round-trip propagation. ↩

Why basic CSMA is not enough on wired media

Basic CSMA says: listen first, and if the channel is idle, transmit. The limitation is that the channel state observed at one end of a cable is always slightly outdated at another end because signals travel at finite speed.2 Suppose station A starts transmitting; before A's first bit reaches distant station B, station B may still hear silence and begin transmitting as well. The overlap on the cable creates a collision domain, so both frames are corrupted.2

This is why collision avoidance by sensing alone is insufficient on long shared wires. The key engineering improvement is collision detection during transmission itself.2 A transmitting station compares what it expects on the line with what it actually senses. If interference is detected, the sender knows another transmitter is active and invokes the collision procedure.2

A useful abstraction is the vulnerable period of length approximately $2T_p$, where $T_p$ is the one-way end-to-end propagation delay of the collision domain.2 If another station starts within this interval, a collision can still occur.

$$\text{Collision window} \approx 2T_p$$

The factor of $2$ appears because a sender at one extreme must allow enough time for its signal to reach the far end and, in the worst case, for the evidence of a collision to propagate back.2

Footnotes

1. [PDF] CSMA/CD - Explains Ethernet slot time, minimum frame size, late collisions, and retransmission behavior. ↩ ↩² ↩³

2. [PDF] Tutorial for CSMA-CD - Provides slot-time interpretation and relation to round-trip propagation. ↩ ↩² ↩³

3. Carrier-sense multiple access with collision detection - Wikipedia - Overview of CSMA/CD, jam signal, slot time, and half-duplex Ethernet behavior. ↩ ↩² ↩³

4. [PDF] 4. Media Access Control - IEEE 802 - IEEE-oriented description of collision handling, jam, and truncated binary exponential backoff. ↩ ↩²

5. [PDF] Lecture 19: Ethernet / MAC - Alan Mislove - Summarizes Ethernet collision detection, slot time, jam signaling, and exponential backoff. ↩

Collision window, slot time, and the minimum frame size

In Ethernet, slot time is the time used for collision handling and retransmission scheduling.2 For classic 10 Mb/s and 100 Mb/s half-duplex Ethernet, the slot time is 512 bit-times; at 10 Mb/s, this equals $51.2\,\mu s$.2 This value is not arbitrary: it is chosen so that a station can still be transmitting when a worst-case collision returns from the farthest point in the permitted collision domain.2

Let:

• $R$ = link bit rate in bits/s
• $T_p$ = one-way maximum propagation delay across the collision domain
• $L_{\min}$ = minimum frame length in bits

For collision detection to work correctly, the sender's transmission time must be at least the round-trip propagation delay:

$$\frac{L_{\min}}{R} \ge 2T_p$$

Multiplying both sides by $R$ gives the fundamental design inequality:

$$L_{\min} \ge 2RT_p$$

This is the mathematical justification for Ethernet's minimum frame size requirement.3 If a frame were shorter than $2RT_p$ bits, a far-end collision could occur and the sender could finish transmission before the collision signal returned, making detection impossible.2

For classic Ethernet, the standardized minimum frame length is 64 bytes, or:

$$L_{\min} = 64 \times 8 = 512 \text{ bits}$$

Hence the transmission time of the smallest legal frame equals one slot time:

$$T_{\text{slot}} = \frac{512}{R}$$

Examples:

A collision detected after the first slot time is called a late collision and generally indicates a misconfigured or faulty network, such as excessive cable distance or timing problems.

Footnotes

1. [PDF] Lecture 19: Ethernet / MAC - Alan Mislove - Summarizes Ethernet collision detection, slot time, jam signaling, and exponential backoff. ↩ ↩² ↩³

2. [PDF] 4. Media Access Control - IEEE 802 - IEEE-oriented description of collision handling, jam, and truncated binary exponential backoff. ↩

3. [PDF] CSMA/CD - Explains Ethernet slot time, minimum frame size, late collisions, and retransmission behavior. ↩ ↩² ↩³ ↩⁴ ↩⁵

4. [PDF] Tutorial for CSMA-CD - Provides slot-time interpretation and relation to round-trip propagation. ↩ ↩² ↩³

Mathematical justification of the $2T_p$ rule

Consider two stations A and B at opposite ends of the maximum allowed collision domain.2

1. A begins transmission at time $t=0$.
2. A's first bit reaches B after one-way propagation delay $T_p$.
3. In the worst case, B starts transmitting just before A's signal arrives, so B still believes the medium is idle.2
4. The collision effect must then travel back toward A, taking nearly another $T_p$.
5. Therefore A may not detect the collision until almost $2T_p$ after starting.2

So if A is to detect the collision before finishing transmission, the frame transmission time $T_f$ must satisfy:

$$T_f \ge 2T_p$$

Since $T_f = \frac{L}{R}$:

$$\frac{L}{R} \ge 2T_p \quad \Rightarrow \quad L \ge 2RT_p$$

This simple inequality connects physical distance, propagation speed, and MAC-layer frame design.2

If propagation speed is approximately $v \approx 2 \times 10^8 \text{ m/s}$, then for cable length $d$:

$$T_p \approx \frac{d}{v}$$

Substituting into the minimum-frame inequality:

$$L_{\min} \ge 2R\frac{d}{v}$$

or equivalently:

$$d \le \frac{L_{\min}v}{2R}$$

Thus, fixing $L_{\min}=512$ bits constrains the maximum collision-domain diameter for a given bit rate.2 This is why faster shared Ethernet requires tighter diameter limits or altered slot-time engineering.

Footnotes

1. [PDF] CSMA/CD - Explains Ethernet slot time, minimum frame size, late collisions, and retransmission behavior. ↩ ↩² ↩³ ↩⁴

2. [PDF] Tutorial for CSMA-CD - Provides slot-time interpretation and relation to round-trip propagation. ↩ ↩² ↩³ ↩⁴ ↩⁵

3. [PDF] Lecture 19: Ethernet / MAC - Alan Mislove - Summarizes Ethernet collision detection, slot time, jam signaling, and exponential backoff. ↩ ↩²

Binary exponential backoff: resolving repeated collisions

After a collision, immediate retransmission would likely cause another collision, especially if the same stations are contending.2 Ethernet therefore uses binary exponential backoff to spread retransmission attempts over time.

If a frame has experienced $i$ collisions, the station chooses a random integer $K$ from:

$$K \in \{0,1,\dots,2^i - 1\}$$

for early collisions, with Ethernet applying truncation so the exponent does not grow indefinitely.2 The retransmission delay is then:

$$T_{\text{backoff}} = K \cdot T_{\text{slot}}$$

where $T_{\text{slot}} = 512$ bit-times for 10/100 Mb/s half-duplex Ethernet.2

So:

• After the 1st collision: $K \in \{0,1\}$
• After the 2nd collision: $K \in \{0,1,2,3\}$
• After the 3rd collision: $K \in \{0,\dots,7\}$

The window doubles each time, reducing the probability that the same stations pick identical retry times again.2 This adaptive behavior is especially effective under bursty contention because it enlarges the contention window only when repeated collisions suggest high load.

The expected backoff after $i$ collisions, assuming uniform selection over $\{0,\dots,2^i-1\}$, is:

$$\mathbb{E}[K] = \frac{2^i - 1}{2}$$

Hence the expected retransmission delay is:

$$\mathbb{E}[T_{\text{backoff}}] = \frac{2^i - 1}{2} \, T_{\text{slot}}$$

This shows mathematically why waiting grows rapidly with repeated collisions.

Footnotes

1. [PDF] Lecture 19: Ethernet / MAC - Alan Mislove - Summarizes Ethernet collision detection, slot time, jam signaling, and exponential backoff. ↩ ↩² ↩³ ↩⁴

2. [PDF] 4. Media Access Control - IEEE 802 - IEEE-oriented description of collision handling, jam, and truncated binary exponential backoff. ↩ ↩² ↩³

3. [PDF] Tutorial for CSMA-CD - Provides slot-time interpretation and relation to round-trip propagation. ↩

Practical interpretation for Network Theory

From a theoretical perspective, CSMA/CD links the physical properties of a medium to MAC-layer correctness.2 The protocol is not merely a rule set for contention; it is an example of cross-layer constraint where propagation delay, transmission rate, and frame format must be jointly engineered.2

The most important takeaways for the Data Link Layer and Medium Access Sub Layer are:

1. Basic CSMA lowers collision probability but cannot prevent collisions on a finite-speed shared medium.2
2. CSMA/CD solves this by requiring stations to detect collisions while transmitting.2
3. The legal collision window is governed by approximately $2T_p$, motivating Ethernet slot time.2
4. The minimum frame size follows directly from the requirement $\frac{L_{\min}}{R}\ge 2T_p$.2
5. Repeated collisions are resolved dynamically by truncated binary exponential backoff.2

These ideas remain pedagogically significant even though switched full-duplex Ethernet largely removed shared-medium collisions from everyday practice.2

Footnotes

1. [PDF] CSMA/CD - Explains Ethernet slot time, minimum frame size, late collisions, and retransmission behavior. ↩ ↩² ↩³ ↩⁴

2. [PDF] 4. Media Access Control - IEEE 802 - IEEE-oriented description of collision handling, jam, and truncated binary exponential backoff. ↩ ↩² ↩³ ↩⁴

3. [PDF] Lecture 19: Ethernet / MAC - Alan Mislove - Summarizes Ethernet collision detection, slot time, jam signaling, and exponential backoff. ↩ ↩² ↩³

4. Carrier-sense multiple access with collision detection - Wikipedia - Overview of CSMA/CD, jam signal, slot time, and half-duplex Ethernet behavior. ↩ ↩² ↩³

5. [PDF] Tutorial for CSMA-CD - Provides slot-time interpretation and relation to round-trip propagation. ↩ ↩²