Picture a busy highway intersection without traffic lights. The road can handle one car at a time through the intersection safely. But when two drivers approach from perpendicular streets and both try to cross simultaneously—collision. The result isn't just inconvenient; it's destructive. Both vehicles are damaged, both drivers lose time, and the intersection becomes blocked for others.

Collisions in computer networks follow the same principle. When two network devices transmit simultaneously on a shared medium, their signals overlap and interfere with each other. Neither transmission succeeds. The data is corrupted, the channel capacity is wasted, and both devices must start over.

This page explores collisions in exhaustive depth: the physics of signal interference, how collisions are detected or avoided, the performance consequences, and the mathematical models that help us understand and design around this fundamental limitation of shared channels.

A collision is not merely an abstract protocol concept—it's a physical phenomenon where signals interfere destructively. Understanding the physics helps explain why collisions are problematic and why different media require different handling strategies.

Collisions on Electrical Media (Copper Cables)

In traditional Ethernet using coaxial cable or early twisted-pair hub networks, data is transmitted as voltage levels. A logical '1' might be represented by +2.5V and a logical '0' by -2.5V (simplified example).

Normal transmission: Station A transmits: +2.5V, -2.5V, +2.5V (representing 101) Receiver sees: +2.5V, -2.5V, +2.5V and decodes: 101

Collision scenario: Station A transmits: +2.5V (1) Station B transmits: -2.5V (0) at the same instant The voltages add algebraically on the wire: +2.5V + (-2.5V) = 0V Receiver sees: 0V — an undefined, uninterpretable signal

Worse, different bit combinations create varying interference:

• Both send 1 (+2.5V each): Wire shows +5V (out of normal range)
• Both send 0 (-2.5V each): Wire shows -5V (out of normal range)
• One sends 1, one sends 0: Wire shows 0V (undefined)

The result: The receive circuitry cannot interpret these corrupted voltage levels. The frame is destroyed.

Collisions on Wireless Media (Radio Waves)

Wireless transmissions use radio frequency electromagnetic waves. When two devices transmit simultaneously on the same frequency, the waves combine in the air through electromagnetic superposition:

Constructive interference: When wave peaks align, amplitude doubles Destructive interference: When peak meets trough, signals cancel Complex patterns: In practice, constantly shifting phase relationships create unpredictable combinations

The result at the receiver is a garbled signal that cannot be decoded. The original information is lost in the noise.

Why wireless is worse:

• Signals propagate in all directions (omnidirectional)
• No clear boundary to the "channel" — interference can come from anywhere
• Multipath propagation creates self-interference
• Hidden terminals create collisions the sender cannot detect

Collisions on Optical Media (Fiber)

While single-mode point-to-point fiber links don't experience collisions (dedicated paths), some fiber topologies with shared segments can:

Passive optical networks (PON): Multiple terminals share fiber to a central splitter. Simultaneous upstream transmissions collide at the splitter.

Legacy fiber bus networks: Historic shared fiber designs faced collision issues similar to electrical buses.

Modern fiber networks avoid this through careful scheduling (TDMA in GPON) or wavelength separation (WDM), but understanding that optical signals can also collide is important.

Collision detection (CD) is the ability to recognize that a collision is occurring while transmission is still in progress. This is the key innovation that makes CSMA/CD efficient: if you can detect a collision early, you can stop transmitting immediately rather than wasting time sending the rest of a doomed frame.

How Collision Detection Works in Ethernet

In classic Ethernet (10BASE5, 10BASE2, and hub-based 10BASE-T), collision detection works through simultaneous transmission and reception monitoring:

Step 1: Station begins transmitting its frame onto the cable.

Step 2: While transmitting, the station also listens to what's actually on the wire.

Step 3: If the signal being received differs from what's being transmitted, something else must be transmitting too—a collision is occurring.

Specifically, the transceiver monitors the DC voltage level on the cable. During normal single-station transmission, the voltage follows expected patterns. During collision, voltages exceed normal ranges due to superposition of multiple signals.

Step 4: Upon detecting a collision, the station:

• Stops transmitting the data frame immediately
• Transmits a jam signal (32-48 bits of specific pattern) to ensure all stations recognize the collision
• Enters the backoff procedure to schedule retransmission

The Minimum Frame Size Requirement

For collision detection to work, a critical constraint must be met: the transmitter must still be transmitting when collision evidence propagates back.

Consider this scenario:

• Station A and Station B are at opposite ends of a 2500-meter cable
• Signal propagation takes approximately 12.5 microseconds one way (25μs round-trip)
• At 10 Mbps, a 64-byte frame takes 51.2 microseconds to transmit

If Station A sends a very short frame (say, 10 bytes = 8 microseconds), A would finish transmitting before the collision signal from B could return. A would believe its transmission succeeded when it actually collided.

Solution: Ethernet mandates a minimum frame size of 64 bytes (512 bits). At 10 Mbps, 512 bits take 51.2 microseconds to transmit—longer than the maximum round-trip time. This ensures the sender is still transmitting when any collision evidence could arrive.

Minimum Frame Time ≥ 2 × Maximum Propagation Delay

For 10 Mbps Ethernet:
- Maximum network diameter: 2500 meters
- Propagation delay: ~5 μs per kilometer
- Round-trip time: 2 × 2500m × 5μs/km = 25μs
- Slot time (with safety margin): 51.2μs
- Minimum frame: 51.2μs × 10 Mbps = 512 bits = 64 bytes

Collision Detection in Higher-Speed Ethernet

As Ethernet speeds increased, the minimum frame size became problematic:

At 100 Mbps: 64 bytes transmit in 5.12 μs, but propagation times don't shrink. To maintain collision detection, maximum cable lengths were reduced.

At 1 Gbps: 64 bytes transmit in 512 nanoseconds—far too short. Gigabit Ethernet introduced carrier extension (padding to 512 bytes) for half-duplex mode.

Modern solution: Full-duplex switching eliminates collision domains entirely. Each switch port is a separate collision domain with only one device. Collision detection became irrelevant because collisions became impossible.

This is why modern Ethernet NICs don't implement CSMA/CD in practice—switched networks don't need it. But understanding the mechanism remains important for wireless protocols and legacy compatibility.

When collision detection is impossible or impractical, the alternative strategy is collision avoidance (CA)—taking proactive steps to reduce the probability of collisions occurring in the first place.

Why Collision Avoidance?

In wireless networks, stations cannot detect collisions because:

• The transmitted signal (at high power) overwhelms the receiver (expecting weak signals)
• By the time a collision is recognized (no ACK received), the entire frame has been transmitted and wasted
• The collision occurs at the receiver, not the sender, due to hidden terminals

Since detection doesn't work, avoidance becomes essential. The philosophy shifts from "transmit and detect problems" to "take precautions and hope they work."

CSMA/CA Mechanisms

Carrier Sensing (the CS in CSMA): Before transmitting, a station listens to determine if the channel is busy. Unlike CSMA/CD where this is just the first step before potential collision detection, in CSMA/CA this sensing must be more conservative—the goal is to never transmit into a busy channel.

Inter-Frame Spacing: Wi-Fi mandates waiting periods between transmissions:

• DIFS (Distributed Inter-Frame Space): Wait time before a station may attempt transmission—gives priority to ongoing exchanges
• SIFS (Short Inter-Frame Space): Shorter wait for immediate responses (ACKs)—gives priority to completion of exchanges

Random Backoff: Even if the channel appears idle after DIFS, a station doesn't transmit immediately. It waits a random number of slot times (the contention window). If the channel becomes busy during this wait, the counter pauses. This randomization reduces the probability that two stations waiting for the same transmission to end will collide.

The RTS/CTS Mechanism

The hidden terminal problem occurs when stations that cannot sense each other's transmissions both transmit to a common station, causing collision at that receiver.

Scenario:

• Station A cannot hear Station C (blocked by walls/distance)
• Both A and C are within range of Station B (access point)
• A begins transmitting to B
• C senses idle channel (doesn't hear A) and also transmits to B
• Collision occurs at B; both frames destroyed

RTS/CTS Solution:

1. Station A wants to transmit to B
2. A sends short RTS (Request to Send) frame: "I want to send X bytes to B"
3. B responds with short CTS (Clear to Send) frame: "OK, go ahead. Channel reserved for X bytes."
4. Critically: All stations that can hear B also hear the CTS—including C
5. C sees the CTS and sets its NAV timer (Network Allocation Vector) to defer transmission
6. A transmits its data; C stays silent even though it can't hear A

The RTS and CTS frames are small (20 bytes), so if they collide, the waste is minimal. The large data frame transmits collision-free.

Tradeoff: RTS/CTS adds overhead (two extra frame exchanges). It's most beneficial for large frames or in environments with many hidden terminals. Wi-Fi allows configuring a threshold—only frames above a certain size use RTS/CTS.

Collisions don't just cause individual frame losses—they systematically degrade network performance in measurable ways. Understanding this impact is essential for network capacity planning and troubleshooting.

Direct Costs of Collisions

Channel Time Wasted: Every collision wastes the time spent transmitting the colliding frames. In CSMA/CD, this is limited (transmission stops upon detection), but still significant:

Wasted time per collision = Transmission time before detection + Jam signal time
                          ≈ Slot time (worst case: collision at far end of cable)

In CSMA/CA (wireless), the entire frame is transmitted before collision is recognized (via missing ACK), wasting:

Wasted time = Full frame transmission + SIFS + ACK timeout

Retransmission Overhead: Collided frames must be retransmitted, consuming additional channel time. With multiple collisions, some frames are transmitted multiple times.

Backoff Delays: After each collision, stations must wait through backoff periods before retrying. These waiting periods represent idle channel time that could carry data.

Collision Rate and Network Load

Collision probability increases non-linearly with network load:

Under light load: Stations rarely have frames waiting. Transmissions are sparse, and collisions are unlikely. The network operates near its maximum efficiency.

Under moderate load: Multiple stations often have frames queued. Collisions begin occurring regularly. Each collision consumes channel time and triggers backoff, reducing throughput. However, the backoff mechanism prevents complete collapse.

Under heavy load: Collisions become frequent. Backoff windows grow (binary exponential backoff doubles the window with each collision). Stations spend more time waiting than transmitting. The network can enter collision saturation where more time is spent recovering from collisions than carrying data.

Throughput Analysis for ALOHA (Illustrative)

The pure ALOHA protocol provides a clean mathematical example of collision impact:

Assumptions:

• N stations, each generating frames at rate G/N
• Combined load: G frames per frame time
• Successful transmission requires no collision during the frame's "vulnerable period"

For Pure ALOHA:

• Vulnerable period = 2× frame time (collision can start before or overlap)
• Probability of success: P(success) = e^(-2G)
• Throughput: S = G × e^(-2G)
• Maximum throughput: S_max = 1/(2e) ≈ 18.4% at G = 0.5

For Slotted ALOHA:

• Vulnerable period = 1× frame time (transmissions synchronized to slots)
• Probability of success: P(success) = e^(-G)
• Throughput: S = G × e^(-G)
• Maximum throughput: S_max = 1/e ≈ 36.8% at G = 1.0

These maximum throughputs are remarkably low—even under optimal conditions, most channel capacity is lost to collisions and empty slots.

The Binary Exponential Backoff Mechanism

Ethernet's backoff algorithm is designed to adapt to collision conditions dynamically:

After 1st collision: Wait 0 or 1 slot times (random choice) After 2nd collision: Wait 0, 1, 2, or 3 slot times (random choice) After nth collision: Wait 0 to (2^n - 1) slot times, for n ≤ 10 After 10th collision: Window stays at maximum (0 to 1023 slots) After 16th collision: Frame is dropped; transmission fails

Why exponential? The doubling window provides negative feedback. If collisions persist (indicating many contending stations), the growing backoff range spaces out retransmission attempts, reducing collision probability. As the network calms down, stations with low backoff counters quickly claim the channel.

Why binary? The exponential base of 2 is computationally cheap (bit shifting) and provides a reasonable balance between fast recovery under light load and sufficient spreading under heavy load.

A collision domain is the network segment where frames from any device can potentially collide with frames from any other device. Understanding collision domains is essential for network design and troubleshooting.

Definition and Scope

Collision Domain: The set of devices whose transmissions can interfere with each other because they share a common physical medium or are connected through devices that don't isolate traffic.

Key property: If Station A and Station B are in the same collision domain and both transmit simultaneously, their frames will collide.

Devices and Collision Domain Boundaries

Different network devices have different effects on collision domains:

Historical vs. Modern Networks

Classic Ethernet (1980s-1990s):

• All devices on a segment shared one collision domain
• Entire floors or buildings might be a single collision domain
• Collision rates of 10-30% were common under load
• Network performance degraded significantly as device counts increased

Bridged Ethernet (1990s):

• Bridges divided networks into smaller collision domains
• Each segment still experienced internal collisions
• Improved but didn't eliminate collision overhead

Switched Ethernet (Late 1990s-Present):

• Each switch port is its own collision domain
• Full-duplex connections eliminate collisions entirely
• Stations transmit whenever they have data—the switch handles forwarding
• "Collision domain" becomes a single wire from device to switch

Collision Domains in Wireless Networks

The challenge: Radio waves don't respect physical barriers the way cables do. Collision domains in wireless are defined by signal reach, not physical topology.

Wi-Fi collision domain: All devices within range of each other that operate on the same channel. This includes:

• Devices associated with the same access point
• Devices associated with nearby access points on the same channel
• Non-Wi-Fi devices using the same frequency (Bluetooth, microwave ovens in 2.4GHz)

Co-channel interference: Even "separate" Wi-Fi networks on the same channel must contend with each other. The collision domain spans all of them.

Mitigation strategies:

• Use non-overlapping channels (1, 6, 11 in 2.4GHz)
• Reduce transmit power to limit domain size
• Use 5GHz where more channels are available
• Deploy in a cellular pattern with careful channel planning

Network protocols can be broadly categorized by their approach to collisions: accepting them as a reality to manage (contention-based) or preventing them entirely through coordination (collision-free). Each approach has distinct characteristics and appropriate use cases.

Contention-Based Protocols (Random Access)

Philosophy: Stations transmit when they want, accepting that collisions will sometimes occur. Efficiency comes from minimal coordination overhead under light load.

Examples: Pure ALOHA, Slotted ALOHA, CSMA, CSMA/CD, CSMA/CA

Characteristics:

• Low latency under light load (transmit immediately)
• Simple implementation (no global state needed)
• Graceful handling of variable station counts
• Performance degrades under heavy load (collision spiral)
• Non-deterministic access delay (can't guarantee when transmission will succeed)

Best suited for: Bursty traffic patterns with many idle periods, unpredictable numbers of active stations, applications tolerant of occasional delays.

Collision-Free Protocols (Controlled Access)

Philosophy: Coordinate access so that only one station transmits at any time. Zero collisions means zero wasted channel time—but coordination has its own costs.

Examples: Token Ring, Token Bus, Polling, TDMA, FDMA

Characteristics:

• Zero collision overhead
• Bounded worst-case delay (important for real-time systems)
• Fair access (each station gets regular opportunity)
• Higher overhead under light load (waiting for permission, passing tokens)
• More complex failure handling (lost token, failed station)
• Fixed station count often required

Best suited for: Industrial control systems, real-time audio/video, situations requiring guaranteed bandwidth or maximum delay bounds.

Hybrid Approaches

Some protocols combine elements of both approaches:

CSMA/CA with RTS/CTS: Contention-based for channel access, but RTS/CTS provides reservation that prevents collision for the data frame.

Reservation ALOHA: Contention-based for reservation requests, scheduled access for data frames.

Polling with contention: Stations contend during a contention period; winners get scheduled transmission slots.

PCF in Wi-Fi: Point Coordination Function allows the access point to grant collision-free transmission opportunities.

These hybrids attempt to capture advantages of both approaches: low delay under light load (contention for setup) and collision-free efficiency under heavy load (scheduled data transmission).

We've explored collisions comprehensively—from physical signal interference to protocol design philosophies. Let's consolidate the essential knowledge:

What's Next

With a thorough understanding of collisions, we're prepared to study how MAC protocols are categorized. The next page examines the major MAC protocol categories: random access, controlled access, and channelization. You'll understand the design philosophy, advantages, and limitations of each category—essential knowledge before studying specific protocols like ALOHA, CSMA, token passing, and CDMA.

The final page of this module will cover channel allocation—the theoretical and practical considerations for dividing channel capacity among competing stations.