Defending Consensus-Based Cooperative Sensing Against Conformity-Driven Byzantine Attacks in 6G UAV Swarm Networks

I. INTRODUCTION

• ISAC combines sensing and communication, while UAV swarms exchange and fuse neighbor decisions to support high-reliability, low-latency cooperative sensing.Source: §I · p. 1
• Conventional accounts treat Byzantine attacks as isolated falsification, although compromised nodes can inject errors through the neighbor update rule.The introduction notes that shared-spectrum ISAC further enlarges this attack surface.Source: §I · p. 1
• Malicious nodes exploit consensus update rules to create false consensus, turning collaborative sensing into an error amplifier.The paper identifies conformity-driven Byzantine attacks as a novel threat class.Source: §I · pp. 1–2
• An evolutionary game framework models how adversarial inducement diverts honest nodes, and its resulting law enables dynamic calibration of their drifting error rates.Source: §I · p. 2
• The proposed scheme achieves near-perfect situational inference across varied attack intensities and network scales, while conventional independent fusion fails under the same conditions.The introduction reports this simulation outcome as the scheme's main result.Source: §I · p. 2

II. CONFORMITY: A DOUBLE-EDGED SWORD

• An honest UAV can shift toward erroneous neighbors, then supply a mistaken report that becomes misleading input for its own neighbors in the next round.Source: §II · p. 2
• False consensus arises when malicious nodes create a local illusion of agreement that conformity propagates, without infecting every UAV.The paper compares this layer-by-layer spread to a rumor spreading through social conformity.Source: §II · p. 2
• The danger shifts from isolated lies by bad nodes to the erroneous reports that conformity induces good nodes to produce collectively.Source: §II · p. 2

III. WHY CONVENTIONAL DEFENSES FAIL

• Mainstream Byzantine-resilient fusion methods assume normal-node error probability is fixed at roughly sensor accuracy and cross-node errors are mutually independent, enabling statistical anomaly screening and voting.Consensus-based sensing attacks can also disrupt convergence through malicious nodes that disobey update rules and manipulate their values [10].Source: §III · p. 2
• Under false consensus, a normal UAV’s true error probability varies with malicious influence and can rise far above its sensing error.A constant sensor-accuracy assumption therefore underestimates normal nodes’ true error level.Source: §III · p. 2
• Because today’s reports depend on neighbors’ earlier reports, conformity couples errors into a generation-to-generation chain reaction.Source: §III · pp. 2–3
• Conventional methods assume errors are fixed and independent, whereas conformity-driven attacks produce errors that drift and couple.Source: §III · p. 3

IV. EVOLUTIONARY-GAME MODELING AND ROBUST FUSION

• Evolutionary game theory models how honest UAVs balance their own sensing against neighbor conformity while malicious neighbors pressure them toward false consensus.The model describes the fraction of the population led astray converging toward a stable level.Source: §IV · p. 3
• The misled fraction is a time-varying trajectory that depends on sensing-channel quality, malicious-node fraction, and attack intensity.The fusion center can propagate the trajectory forward from an initial state in which no honest node has been misled.Source: §IV · p. 3
• The fusion center estimates normal nodes’ drifting error from the predicted misled fraction, adjusts report credibility, and makes a maximum-a-posteriori decision.The procedure uses the predicted misled fraction at the current instant to calibrate errors and credibility.Source: §IV · p. 3
• A fiercer inducement produces a more pronounced, predictable trajectory, allowing the fusion center to correct induced errors more reliably as attacks intensify.The paper presents this as the reason the defense grows more reliable under stronger attacks.Source: §IV · p. 3
• The defense models collective population drift rather than building a more elaborate bad-node detector.Its first defensive step is to characterize the evolution of normal nodes after inducement.Source: §IV · p. 3

V. SIMULATION VERIFICATION

• Simulations used regular networks of 300–500 UAVs with ten neighbors and found consistent trends on scale-free and random topologies.The reported data points were averaged over multiple Monte Carlo runs.Source: §V · p. 3
• Fusion accuracy stays close to 100% across most malicious-ratio and attack-probability combinations, weakening only in the most extreme corner.That corner combines malicious nodes reaching twice the number of normal nodes with a semi-random attack.Source: §V · p. 4, Figure 3 · p. 4
• About 60% of normal nodes become misled under strong attack, while conventional fusion collapses and the proposed mechanism maintains near-perfect accuracy.The proposed mechanism sustains near-perfect performance even after more than half of normal nodes have been induced.Source: Figure 4 · p. 4, §V · p. 4
• Even the shortest fusion window delivers near-perfect accuracy, supporting decisions without an extended observation horizon.The mechanism makes independent judgments at each time instant and predicts the instantaneous error level online.Source: §V · p. 4

From the sensing layer to the waveform layer

• The defense models attacks at the sensing-decision level; extending the evolutionary perspective to shared waveforms and signal processing remains open.Malicious nodes may contaminate echoes or forge sensing signals at the physical layer.Source: §VI · p. 4

Bringing latency and reliability into the loop

• Fusion latency and reliability under attack need to become quantifiable, optimizable metrics in the design loop.Source: §VI · p. 4

From static laws to adaptive adversaries

• If attackers dynamically adjust inducement, the defender’s evolution model must adapt online, motivating a dynamic attacker–defender game.Source: §VI · p. 4

Toward standardization and heterogeneous cooperation

• Deployment across heterogeneous 6G networks requires studying cross-domain conformity-driven attacks and standardizing corresponding defenses.The envisioned networks include terrestrial base stations, non-terrestrial platforms, and multiple sensing modalities.Source: §VI · p. 4

• Malicious nodes exploit neighbor updates to create false consensus, causing honest nodes’ errors to evolve rather than remain fixed and independent.Source: §VII · p. 5
• Evolutionary game theory models inducement, and its resulting law guides fusion-center correction of normal nodes’ drifting error levels.Source: §VII · p. 5
• Simulations show near-perfect situational inference across diverse attack intensities and network sizes, clearly outperforming independent fusion at low decision latency.Source: §VII · p. 5