PaperlayerView exact revision on arXiv

Source-linked AI summary

Is Collision-Free Backoff Worth It in Wi-Fi?

Mohammad Yousefi, Francesc Wilhelmi, Boris Bellalta

arXiv:2609.03817v1cs.NI

TL;DR

Whether collision elimination justifies rigid deterministic access in modern Wi-Fi remains uncertain, particularly with variable traffic and inexpensive collisions. The paper compares BEB and three collision-free schemes in system-level simulations across full-buffer, non-full-buffer, and coexistence settings. Gains are modest, delay tails track CW policy, and downlink-heavy deterministic scheduling can degrade performance.

  • Problem

    It is uncertain whether eliminating collisions justifies deterministic access in modern Wi-Fi, where rigid schedules can be disrupted by variable traffic and collisions may be inexpensive.

    Source: §I · p. 1

  • Method

    System-level simulations compare standard and single-stage BEB with ECA, E2CA, and DetBO under full-buffer, non-full-buffer uplink/downlink, and legacy-BEB coexistence conditions.

    Source: §I · p. 2

  • Results

    Across traffic settings, gains are modest, reaching at most 6.3% throughput under full-buffer traffic; non-full-buffer delay tails follow CW policy, and downlink-heavy deterministic scheduling can degrade performance.

    Source: §V · p. 7

  • Takeaways & Limitations

    Collision probability alone is a poor performance indicator, and stations fall back to random backoff at least 93% of the time, so schedules rarely form outside the AP.

    Source: §V · p. 7

Abstract

from arXiv · show

The Distributed Coordination Function (DCF)---the underlying channel access protocol in Wi-Fi, based on Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) and Binary Exponential Backoff (BEB)---is simple and effective, but its performance can degrade with increasing contention due to collisions. Collision-free backoff algorithms replace randomization with deterministic channel access, potentially improving efficiency and predictability. But are collision-free methods, with all their implications in terms of protocol design, sufficient to improve Wi-Fi performance under realistic non-full-buffer traffic with both uplink and downlink transmissions? We investigate this question by comparing three collision-free algorithms, CSMA/ECA (ECA), CSMA/E2CA (E2CA), and deterministic backoff (DetBO), with standard and single-stage BEB. Using system-level simulations, we evaluate full-buffer traffic, non-full-buffer ON/OFF traffic with both uplink and downlink transmissions, and coexistence with legacy BEB stations. Under full-buffer traffic, collision-free operation provides only modest throughput gains, up to 6.3\%. With non-full-buffer traffic, delay is governed primarily by the contention window (CW) rather than by collision avoidance, and collision-free access can even degrade performance under downlink-heavy traffic. Stations fall back to a random backoff at least 93\% of the time, so a collision-free schedule rarely forms outside the AP.

I. INTRODUCTION

Wi-Fi’s randomized DCF access supports uncoordinated coexistence, but collisions can increase delay. The paper asks whether deterministic backoff improves performance under realistic traffic and compares several such schemes with BEB.

I. INTRODUCTION

  • DCF combines carrier sensing with random backoff, which avoids coordination and supports long-term fairness but can cause collisions and longer retransmission delays.After collisions, BEB doubles the contention window, potentially creating a long tail in packet delay.Source: §I · p. 1
  • Interactive cloud virtual reality targets 3–10 ms delays with reliability above 99.9%, while tactile and haptic traffic targets 1–5 ms with reliability above 99.999%.Meeting these requirements remains challenging, particularly with randomized channel access.Source: §I · p. 1
  • ECA, E2CA, and DetBO replace randomization with locally determined backoffs that can form repeating collision-free schedules.Unlike centralized schemes such as PCF and R-TWT, these methods require neither a coordinator nor explicit signaling.Source: §I · p. 1
  • Rigid per-station schedules can be disrupted when traffic becomes idle, while RTS/CTS and frame aggregation can make collisions relatively inexpensive.These trade-offs leave uncertain whether eliminating collisions justifies deterministic access under non-full-buffer traffic.Source: §I · p. 1
  • The study compares standard and single-stage BEB with ECA, E2CA, and DetBO across full-buffer and realistic traffic, uplink/downlink, and legacy-station coexistence.The evaluation uses an IEEE 802.11 simulator with a common set of assumptions.Source: §I · pp. 1–2

II. BACKOFF SCHEMES

The schemes share DCF’s basic countdown and differ in how they choose backoffs after transmissions. BEB randomizes every backoff, while ECA and E2CA use deterministic post-success access with different collision responses.

II. BACKOFF SCHEMES

  • When a station empties its buffer, it clears its backoff state and re-enters contention with a random value, regardless of scheme.The schemes differ in post-transmission backoff choice only while a station remains active, and in collision handling.Source: §II · p. 2

A. Binary Exponential Backoff

  • BEB draws random backoffs from a contention window that doubles after successive collisions and resets after success.The backoff stage is capped at six, so CW does not exceed 1023; randomization prevents persistent collisions but does not yield a collision-free schedule under saturation.Source: A. Binary Exponential Backoff · p. 2
  • Single-stage BEB disables exponential growth, keeping the backoff stage at zero and drawing each value from [0, CWmin].Multi-stage BEB is the standard configuration described for comparison.Source: A. Binary Exponential Backoff · p. 2

B. ECA

  • ECA uses a deterministic post-success backoff but retains BEB after collisions; with CWmin = 15, its cycle has eight slots and supports at most eight contenders.Once stations occupy different positions in the cycle, transmissions become collision-free.Source: B. ECA · p. 2
  • E2CA retains its deterministic backoff after the first collision following success and reverts to BEB only after a second consecutive collision.This is intended to improve robustness to transient collisions while preserving the schedule.Source: B. ECA · p. 2

C. Deterministic Backoff

DetBO adapts its deterministic backoff to observed channel activity and uses a separate collision-recovery rule. The evaluation compares schemes under shared assumptions, including ON/OFF traffic and a contender count within the ECA/E2CA cycle capacity.

C. Deterministic Backoff

  • DetBO sets the post-success backoff to b = α + i, where α = 10 and i counts interruptions during the preceding countdown.Consecutive busy periods without an intervening idle slot count as one interruption; unlike ECA, DetBO adapts to channel activity.Source: C. Deterministic Backoff · p. 2
  • After collisions, DetBO does not double CW: with m = 7 and τ = 3, it keeps its deterministic value through two consecutive collisions and draws randomly after the third.A station returning from an emptied-buffer state starts with a random backoff and switches to deterministic operation after its first successful transmission.Source: C. Deterministic Backoff · p. 2

III. EVALUATION SETUP

  • The evaluation uses a common single-BSS scenario and progressively adds realistic traffic and coexistence conditions to separate collision-avoidance effects from CW-policy effects.Results are averaged over multiple independent simulation runs using the same network configuration.Source: §III · p. 2

A. System and Traffic Model

  • Non-full-buffer traffic uses exponential ON/OFF periods averaging 20 ms and 80 ms, with Poisson arrivals during ON periods; the AP handles downlink traffic in one queue.With N = 7, at most eight contenders include the AP, matching the ECA/E2CA cycle capacity; using the same MCS isolates backoff effects.Source: A. System and Traffic Model · pp. 2–3

B. Scenarios

The scenarios cover continuous backlog, non-full-buffer uplink/downlink traffic, and coexistence with legacy BEB stations. Offered loads are swept or set relative to a threshold determined by the first scheme to saturate.

B. Scenarios

  • In the full-buffer scenario, every node remains continuously backlogged, contender count varies, and aggregate throughput is measured.The full-buffer case varies N from 2 to 30.Source: B. Scenarios · p. 3
  • The non-full-buffer ON/OFF scenario sweeps offered load for uplink-only, balanced 50% uplink/50% downlink, and downlink-heavy 20% uplink/80% downlink traffic.The offered-load levels are given in Mbps as fractions of saturation load S.Source: B. Scenarios · p. 3, Table II · p. 3
  • In coexistence tests, collision-free stations share the medium with legacy BEB stations under balanced medium-high load while their relative proportions vary.Reference load S is the highest offered load at which all schemes keep packet drops below 7% in both directions; it ranges from 490 Mbps for uplink-only to 160 Mbps for 80% downlink, and is set by the first scheme to reach the threshold.Source: B. Scenarios · p. 3, Table II · p. 3

C. Metrics

The section defines the reported collision, throughput, and packet-delay metrics, including how delivered payload and packet delay are measured.

C. Metrics

  • Metrics are averaged over 10 runs with 95% confidence intervals.All metrics cover the entire simulation, including its initial seconds, because each node starts in the stationary state of its ON/OFF process.Source: C. Metrics · p. 3
  • Conditional collision probability is the fraction of transmission attempts that collide.With RTS/CTS, an attempt corresponds to an RTS transmission; A_i and C_i denote node i’s attempts and collisions, respectively.Source: C. Metrics · p. 3
  • Aggregate throughput is delivered payload per unit time.The calculation uses the number of MPDUs in each successful TXOP, MPDU payload L, and run duration T.Source: C. Metrics · p. 3
  • Packet delay is measured from packet generation to successful delivery, with dropped packets excluded from delay statistics.A packet is dropped only when it arrives at a full buffer; frames are retransmitted until delivered because there is no retry limit.Source: C. Metrics · p. 3

IV. RESULTS

The results section compares five backoff schemes using throughput, conditional collision probability, and packet delay across the traffic and coexistence scenarios described earlier.

IV. RESULTS

  • The results compare five backoff schemes on throughput, conditional collision probability, and packet delay under the previously described traffic and coexistence scenarios.Source: §IV · p. 4

A. Full-buffer Traffic

Under full-buffer traffic, collision-free schemes provide only modest throughput gains over BEB. In non-full-buffer uplink, mean delay is similar across schemes, while tail delay and collision behavior vary.

A. Full-buffer Traffic

  • DetBO reaches fully collision-free operation for every tested N, whereas ECA and E2CA exceed their eight-slot cycle capacity beyond eight contenders, reaching collision probabilities of 0.37 and 0.10.E2CA preserves its schedule after an isolated collision and therefore re-converges faster.Source: A. Full-buffer Traffic · p. 4

B. Non-full-buffer Traffic

  • The non-full-buffer evaluation uses one AP and seven ON/OFF stations across three uplink/downlink configurations, with seven contenders for uplink-only traffic and eight when downlink is included.Source: B. Non-full-buffer Traffic · p. 4

A. Full-buffer Traffic

  • 519–554 Mbps is the aggregate throughput range for BEB, ECA, E2CA, and DetBO, while single-stage BEB falls from 536 to 262 Mbps across the contender range.At N = 30, DetBO and E2CA outperform BEB by about 6%; collision-free gains remain modest because RTS/CTS and A-MPDU aggregation make collisions inexpensive.Source: Figure 1 · p. 4, A. Full-buffer Traffic · p. 4

1) 100% Uplink:

  • DetBO’s conditional collision probability is 0.19 and 0.17 at the two lowest loads, then 0.081 at the highest load, while BEB and single-stage BEB rise to 0.143 and 0.193.At low load, interruptions are rare, so DetBO stations reuse the same fixed offset after collisions; at higher load, differing interruptions help colliders separate.Source: 1) 100% Uplink · p. 4
  • At the highest uplink load, mean delay is 18.8–21.2 ms across schemes and 99th-percentile delay ranges from 101 to 129 ms; E2CA and DetBO have the shortest tails at 101 and 106 ms.Mean delay is dominated by queueing; single-stage BEB’s tail is 116 ms despite its highest collision probability, while standard BEB has the largest tail at 129 ms.Source: 1) 100% Uplink · pp. 4–5, Figure 2 · p. 5

2) 50% Uplink – 50% Downlink:

With balanced uplink and downlink traffic, the AP becomes the bottleneck, and its delay depends more on contention-window behavior than collision probability. DetBO gives the AP the longest tail, while E2CA gives it the shortest.

2) 50% Uplink – 50% Downlink:

  • At the highest load, 99th-percentile downlink delay reaches 41–56 ms versus below 22 ms uplink, with DetBO worst at 56.4 ms.The AP serves half the traffic through one queue but gets channel access no more often than an individual station.Source: 2) 50% Uplink – 50% Downlink · p. 5
  • DetBO’s deterministic backoff averages 11.4 slots for the AP versus 10.1 for stations, leaving the heavily loaded AP waiting longer.The AP senses more busy periods while counting down; stations that empty their buffers rejoin with a shorter random backoff.Source: 2) 50% Uplink – 50% Downlink · p. 5
  • Under random schemes, the AP draws from the same contention window as stations, carries no systematic penalty, and clears its backlog more quickly.The AP’s persistent queue under balanced traffic contrasts with stations that can empty their buffers.Source: 2) 50% Uplink – 50% Downlink · p. 5
  • E2CA has the shortest downlink tail at 40.5 ms despite the same collision probability as BEB (0.098).E2CA delays CW growth after an isolated collision, avoiding repeated large backoff stages at the AP.Source: 2) 50% Uplink – 50% Downlink · p. 5

3) 20% Uplink – 80% Downlink:

With 20% uplink and 80% downlink traffic, downlink delay remains much higher than uplink delay, and DetBO again has the longest AP tail. Stations rarely use their deterministic backoffs, while the AP is the only node that consistently maintains a schedule.

3) 20% Uplink – 80% Downlink:

  • In this downlink-heavy scenario, the AP carries 115 Mbps against 4 Mbps per station, resembling typical Wi-Fi deployments.The scenario tests how a sharper traffic asymmetry affects delay and persistence of deterministic schedules.Source: 3) 20% Uplink – 80% Downlink · p. 5
  • At the highest load, 99th-percentile downlink delay is 32.9–41.4 ms, about three times uplink delay, which stays below 14 ms for every scheme.E2CA has the shortest downlink tails (16.6 and 32.9 ms at the two highest loads); DetBO has the longest (30 and 41.4 ms).Source: 3) 20% Uplink – 80% Downlink · p. 6, Figure 4 · p. 6
  • DetBO drops 2.5% of packets versus 1.2% for E2CA, all at the AP, whose slower queue occasionally fills.Dropped packets do not enter the delay statistics, so DetBO’s downlink delay is worse than the figure shows.Source: 3) 20% Uplink – 80% Downlink · p. 6
  • Stations lose deterministic schedules before use at least 93% of the time, while the AP retains 71–81% of its deterministic backoffs.ECA and E2CA stations lose 100% of their deterministic backoffs; DetBO stations retain only 6.5%.Source: 3) 20% Uplink – 80% Downlink · pp. 6–7, Table III · p. 7
  • Stations lose schedules mainly when their buffers empty and they rejoin contention with a random backoff, rather than because of collisions.The AP goes idle less often, making it the only node that consistently maintains a schedule.Source: 3) 20% Uplink – 80% Downlink · pp. 6–7

C. Coexistence with Legacy Stations

Collision-free schemes coexist with legacy BEB without degrading legacy-station performance. DetBO gives stations shorter delay despite more collisions, while AP outcomes vary with the scheme’s contention-window behavior.

C. Coexistence with Legacy Stations

  • ECA and E2CA stations are statistically indistinguishable from co-channel legacy stations in collision probability and delay.The AP is excluded from Fig. 5; its results are reported separately.Source: C. Coexistence with Legacy Stations · p. 7
  • DetBO stations have 99th-percentile delay of 5.0–7.1 ms versus 15.9–19.5 ms for BEB stations, while legacy-station delay remains statistically unchanged across coexistence ratios.DetBO station collision probability falls from 0.206 to 0.101 as its proportion decreases, approaching the all-BEB value of 0.090.Source: C. Coexistence with Legacy Stations · p. 7
  • For the AP, E2CA yields 24.4 ms 99th-percentile delay versus 32.3 ms under all-BEB, while ECA yields 34.0 ms.ECA and E2CA AP collision probabilities are 0.112 and 0.108, respectively; the all-BEB baseline is 0.112.Source: C. Coexistence with Legacy Stations · p. 7
  • DetBO raises AP collision probability to 0.178 and 99th-percentile delay to 40.6 ms, compared with 0.112 and 32.3 ms under all-BEB.The AP’s backoff grows with sensed channel activity, so the busiest node waits longest.Source: C. Coexistence with Legacy Stations · p. 7

V. CONCLUSIONS

Collision-free backoff offers modest throughput gains under full-buffer traffic, while under non-full-buffer traffic delay depends more on contention-window policy than collision avoidance. Deterministic schedules rarely form at stations, and can degrade performance under downlink-heavy traffic.

  • Under downlink-heavy traffic, deterministic scheduling can degrade performance.Source: §V · p. 7
  • 6.3% is the maximum throughput gain from collision-free operation under full-buffer traffic, where RTS/CTS and A-MPDU aggregation make collisions inexpensive.Source: §V · p. 7
  • Under non-full-buffer traffic, delay tails depend primarily on contention-window policy; single-stage BEB has the highest collision probability but not the longest tail.Collision probability alone is therefore a poor performance indicator.Source: §V · p. 7
  • In coexistence with legacy BEB, collision-free stations cause no measurable degradation, and their own benefit follows contention-window policy rather than collision avoidance.Source: §V · p. 7, Figure 5 · p. 8
  • Stations use random backoff at least 93% of the time, usually because buffers empty before deterministic backoff is used; only the AP sustains a schedule, which penalizes it most.Source: §V · p. 7
Loading 2609.03817v1…