引用本文
  • 马子龙,杨润河,张敏,孙瑶.基于神经网络的RC4密钥恢复攻击[J].信息安全学报,已采用    [点击复制]
  • mazilong,Yang Runhe,Zhang Min,SunYao.RC4 Key Recovery Attack Based on Neural Networks[J].Journal of Cyber Security,Accept   [点击复制]
【打印本页】 【下载PDF全文】 查看/发表评论下载PDF阅读器关闭

过刊浏览    高级检索

本文已被:浏览 680次   下载 0  
基于神经网络的RC4密钥恢复攻击
马子龙, 杨润河, 张敏, 孙瑶
0
(中国科学院信息工程研究所)
摘要:
RC4算法凭借其加解密速度快,易于实现等优点成为了一种应用广泛的流密码算法,被广泛应用到SSL/TLS、WEP 等多种信息安全协议中,虽然在部分协议中RC4因为安全漏洞被弃用,但其在信息传输、图形图像加密、计算机硬件和操作系统等领域仍有良好的应用,随着近年来数学理论以及计算机技术的发展,RC4面对着更多的安全威胁,因此RC4算法的攻击仍是重要的研究方向。本文通过引入神经网络改善RC4密钥恢复攻击的成功概率,利用多项实验对比了ResNet18,VGG11,LSTM全连接网络以及卷积神经网络的RC4密钥恢复攻击模型,最终提出了基于神经网络的RC4投票算法,该算法通过输入偏差投票数据训练神经网络模型,用于RC4的密钥恢复攻击。通过实验结果证明,相比于现有攻击方式,本模型具有更高的RC4密钥恢复成功率。我们首次将神经网络模型应用到RC4密钥恢复攻击当中,并给出了组合偏差数据的更好方式,为RC4密钥恢复攻击的投票预测方法提供了新思路。同时还找到了对输入数据新的处理方法,通过引入原始偏差数据防止有用信息丢失,使得不同情况下RC4密钥恢复攻击的成功率均有提高,单个字节使用30000个WEP数据包进行单轮投票的攻击成功率达到89.07%,较现有的最好攻击方法提升了1.27%。30000个WEP数据包进行两轮投票的攻击成功率达到94.351%,较现有的最好攻击方法提升1.44%。
关键词:  RC4  密钥恢复  神经网络  偏差
DOI:
投稿时间:2023-02-13修订日期:2023-03-30
基金项目:
RC4 Key Recovery Attack Based on Neural Networks
mazilong, Yang Runhe, Zhang Min, SunYao
(Institute of Information Engineering,Chinese Academy of Sciences)
Abstract:
Due to its advantages of fast encryption and decryption speed and easy implementation, RC4 has become a widely used stream cipher algorithm, which is widely used in SSL/TLS, WEP and other information security protocols. Alt-hough RC4 has been abandoned in some protocols due to security vulnerabilities, However, it still has a good ap-plication in the fields of information transmission, graphic image encryption, computer hardware and operating system. With the development of mathematical theory and computer technology in recent years, RC4 is facing more security threats, so the attack of RC4 algorithm is still an important research direction. By introducing neural net-work to improve the success probability of RC4 key recovery attack, this paper compares the RC4 key recovery attack model of ResNet18, VGG11, LSTM fully connected network and convolutional neural network with a number of experiments, and finally proposes the RC4 voting algorithm based on neural network. This algorithm trains the neural network model by inputting biased voting data and is used for key recovery attacks of RC4. The experi-mental results show that this model has a higher success rate of RC4 key recovery than the existing attack methods. For the first time, we apply the neural network model to RC4 key recovery attack, and give a better way to combine the deviation data, which provides a new idea for the vote prediction method of RC4 key recovery attack. At the same time, a new processing method for input data is also found. By introducing the original deviation data to pre-vent the loss of useful information, the success rate of RC4 key recovery attack is improved under different condi-tions. The success rate of single-round voting attack using 30,000 WEP packets in a single byte reaches 89.07%, which is 1.27% higher than the existing best attack method. The success rate of two-round voting by 30,000 WEP packets reaches 94.351%, which is 1.44% higher than the existing best attack method.
Key words:  RC4  key recover  the neural network  bias