Multiagent Consensus Subject to Communication and Privacy Constraints

Dipankar Maity, Panagiotis Tsiotras

Abstract

We consider a multiagent consensus problem in the presence of adversarial agents. The adversaries are able to listen to the interagent communications and try to estimate the state of the agents. The agents have a limited bit rate for communication and are required to quantize the transmitted signal in order to meet the bit rate constraint of the communication channel. We propose a consensus protocol that is protected against the adversaries, i.e., the expected mean-square error of the adversary state estimate is lower bounded. In order to deal with the bit rate constraint, we propose a dynamic quantization scheme that guarantees protected consensus.

Introduction

In modern multi-agent systems, such as swarms of autonomous drones or fleets of self-driving vehicles, achieving "consensus"—an agreement on a shared value like heading or position—is fundamental. Traditionally, agents reach consensus by iteratively broadcasting their current states to neighbors. However, this open communication presents two significant challenges: privacy and bandwidth.

Passive adversaries (eavesdroppers) can intercept these broadcasts to reconstruct the internal states of the agents, leading to security vulnerabilities. Simultaneously, real-world communication channels have finite capacities. Agents cannot transmit high-precision real numbers; they must quantize their data into a limited number of bits. This quantization often introduces noise that can degrade the accuracy of the final consensus. This paper introduces a novel framework to achieve protected consensus under strict communication constraints.

The Innovation-Communication Consensus (ICC) Protocol

To address the privacy risk, the authors propose the Innovation-Communication Consensus (ICC) protocol. Unlike standard protocols where agents share their full state $x_i(k)$, the ICC protocol requires agents to transmit only an "innovation" signal $\xi_i(k)$. This signal represents the difference between an agent's current state and its neighbor's estimate of that state.

Algorithm 1: The Innovation-Communication Consensus (ICC) protocol steps.

By sharing innovations rather than raw states, the protocol creates a dependency chain. If an eavesdropper misses even a single packet, they lose the necessary context to accurately estimate the agent's state. Furthermore, as the system approaches consensus, the innovation signal naturally decays toward zero. This means that late-stage interceptions provide almost no information about the initial private states, effectively "locking" the privacy of the agents.

Dynamic Quantization for Bandwidth Constraints

To handle limited bit-rates, the authors extend ICC into the Bit-rate Constrained ICC (BICC). This version employs a dynamic quantization scheme. Static quantizers, which use fixed intervals, often fail in consensus problems because they either saturate early or lack the precision needed for final convergence.

The proposed dynamic quantizer adjusts its range over time. As agents converge and the innovation signals shrink, the quantizer "zooms in," increasing precision without requiring more bits. The authors derived a mathematical lower bound for the bit-rate; if the channel capacity exceeds this threshold, the system is guaranteed to reach consensus while maintaining privacy protections.

Simulation and Results

The theoretical framework was validated using numerical simulations of a 25-agent network. The agents were arranged in a ring-structure digraph, a common topology for testing decentralized algorithms.

The 25-agent ring-structure digraph used for numerical validation.

In these simulations, initial states were randomly selected, and the probability of a successful eavesdrop by an adversary was set to 50% ($\gamma = 0.5$). The results demonstrated a stark contrast between standard protocols and ICC. In standard consensus, the adversary's estimation error dropped to zero almost instantly. In contrast, under the ICC protocol, the adversary's error remained high and stable, confirming the protocol's "protected" status. Crucially, the ICC protocol reached the exact same consensus value as the unprotected version, proving that privacy does not necessitate a sacrifice in cooperative performance.

Conclusion

This research provides a robust solution for multi-agent systems operating in hostile or resource-constrained environments. By shifting the communication paradigm from state-sharing to innovation-sharing and utilizing adaptive quantization, the authors prove that consensus can be both private and precise.

Future work may explore active adversaries who attempt to inject malicious data, as well as time-varying network topologies where connectivity is intermittent. Nevertheless, the ICC and BICC protocols represent a significant step toward secure and efficient decentralized coordination.

Summary updated

Abstract

We consider a multiagent consensus problem in the presence of adversarial agents. The adversaries are able to listen to the interagent communications and try to estimate the state of the agents. The agents have a limited bit rate for communication and are required to quantize the transmitted signal in order to meet the bit rate constraint of the communication channel. We propose a consensus protocol that is <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">protected</i> against the adversaries, i.e., the expected mean-square error of the adversary state estimate is lower bounded. In order to deal with the bit rate constraint, we propose a dynamic quantization scheme that guarantees <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">protected consensus</i> .
BibTeX
@article{maity2022multiagent,
  author = {Dipankar Maity and Panagiotis Tsiotras},
  title = {{Multiagent Consensus Subject to Communication and Privacy Constraints}},
  journal = {IEEE Trans. Control. Netw. Syst},
  year = {2022},
  doi = {10.1109/tcns.2021.3124926},
}
Figure