基于流关联的匿名通信识别研究
首发时间:2025-02-26
摘要:匿名通信技术广泛应用于隐私保护场景,Tor网络作为最流行的匿名通信系统之一,通过多跳代理和加密技术提供匿名性。然而,流量分析技术的进步使其匿名性受到挑战,其中流关联攻击通过对比入口和出口流量的统计特征,实现匿名通信关系的识别,威胁Tor的隐私保护。现有方法如DeepCorr和DeepCoFFEA在特定环境下表现优越,但在高噪声、时延抖动等复杂网络条件下,仍面临鲁棒性不足和时序特征建模受限的问题。针对上述问题,本文提出DeepRoFA(Deep Robust Flow Association)模型,在DeepCoFFEA基础上优化:(1)引入累计流量特征,以增强对流量模式的描述能力,提高模型在复杂网络环境中的稳定性;(2)融合CNN、LSTM和Attention机制,以更好地提取时序特征,并强化关键时间窗口的关注度,提高流量匹配的精度。实验基于DeepCoFFEA数据集,在标准环境和模拟复杂环境下评估,并与现有方法对比。实验结果表明,在标准测试环境下,DeepRoFA与DeepCoFFEA性能相当,而在高噪声和时延波动环境中,DeepRoFA保持较高的流量匹配准确率(TPR)并显著降低误报率(FPR),展现出更强的鲁棒性和适应性。研究表明,该模型可有效提升匿名通信流量分析的准确性,为Tor网络的安全性评估和流量分析对抗提供技术支持。
For information in English, please click here
Research on Anonymous Communication Identification Based on Flow Correlation
Abstract:Anonymous communication networks, particularly Tor, ensure privacy through multi-hop relays and encryption. However, advances in traffic analysis pose severe threats, with flow correlation attacks exploiting statistical similarities between ingress and egress traffic to infer communication links, compromising anonymity. Existing deep learning-based methods, such as DeepCorr and DeepCoFFEA, achieve notable performance under controlled conditions but exhibit limited robustness against noise, latency variations, and packet loss.This study proposes DeepRoFA (Deep Robust Flow Association), an enhanced flow correlation model addressing these limitations. Key innovations include cumulative flow features, improving traffic pattern characterization under adverse conditions, and a hybrid CNN-LSTM-Attention framework, which captures temporal dependencies while emphasizing critical time windows for higher correlation accuracy. Evaluations on the DeepCoFFEA dataset under standard and simulated high-noise, high-latency environments demonstrate DeepRoFA\'s superiority in maintaining a high true positive rate (TPR) and significantly reducing false positives (FPR).Results indicate that DeepRoFA surpasses existing methods in robustness and adaptability, offering a more reliable approach for traffic analysis in anonymous communication. This work provides valuable insights for Tor security assessment and countermeasures against flow correlation attacks.
Keywords: Tor Anonymous Communication Flow Correlation Attack Deep Learning
基金:
引用
No.****
同行评议
勘误表
基于流关联的匿名通信识别研究
评论
全部评论