Demo: iJam with Channel Randomization
Jordan L. Melcher, Yao Zheng, Dylan Anthony, Matthew Troglia, Yanjun Pan, Ming Li, Thomas Yang, Alvin Yang, Samson Aggelopoulos
DDemo: iJam with Channel Randomization
Jordan L. Melcher, Yao ZhengDylan AnthonyMatthew Troglia Thomas Yang, Alvin Yang,Samson Aggelopoulos
University of Hawai’i at MÄĄnoaHonolulu, Hawaii
Yanjun PanMing Li
University of ArizonaTucson, Arizona
ABSTRACT
Physical-layer key generation methods utilize the variations of thecommunication channel to achieve a secure key agreement betweentwo parties with no prior security association. Their secrecy rate(bit generation rate) depends heavily on the randomness of thechannel, which may reduce significantly in a stable environment.Existing methods seek to improve the secrecy rate by injecting arti-ficial noise into the channel. Unfortunately, noise injection cannotalter the underlying channel state, which depends on the multipathenvironment between the transmitter and receiver. Consequently,these methods are known to leak key bits toward multi-antennaeavesdroppers, which is capable of filtering the noise through thedifferential of multiple signal receptions. This work demonstratesan improved approach to reinforce physical-layer key generationschemes, e.g., channel randomization. The channel randomizationapproach leverages a reconfigurable antenna to rapidly change thechannel state during transmission, and an angle-of-departure (AoD)based channel estimation algorithm to cancel the changing effectsfor the intended receiver. The combined result is a communicationchannel stable in the eyes of the intended receiver but randomlychanging from the viewpoint of the eavesdropper. We augmentedan existing physical-layer key generation protocol, iJam, with theproposed approach and developed a full-fledged remote instrumen-tation platform to demonstrate its performance. Our evaluationsshow that augmentation does not affect the bit error rate (BER)of the intended receiver during key establishment but reduces theeavesdropper’s BER to the level of random guessing, regardless ofthe number of antennas it equips.
CCS CONCEPTS • Security and privacy → Key management ; Mobile and wire-less security . ACM Reference Format:
Jordan L. Melcher, Yao Zheng, Dylan Anthony, Matthew Troglia, ThomasYang, Alvin Yang, Samson Aggelopoulos, Yanjun Pan, and Ming Li. 2020.Demo: iJam with Channel Randomization. In
ACM, New York, NY, USA, 3 pages.https://doi.org/10.1145/3395351.3401705
Physical layer key generation schemes aim to establish a sharedkey between two parties through an open channel eavesdropped by
WiSec ’20, July 8–10, 2020, Linz (Virtual Event), Austria © 2020 Copyright held by the owner/author(s).This is the author’s version of the work. It is posted here for your personal use. Not forredistribution. The definitive Version of Record was published in , https://doi.org/10.1145/3395351.3401705. an adversary. The majority of the schemes leverage the changingwireless channel to generate the key bits, with a generation rateproportional to the entropy of the channel. The more random thechannel the faster the key is generated. A few other designs seek tomake key generation rate independent from the channel variation,by injecting artificial noise into the wireless channel to aid security[2, 3, 9, 10, 14]. The combination of the transmission and jammingsignal introduces uncertainty to an eavesdropper in the form ofnoise [7, 8, 13], and prevents the eavesdropper from obtaining theunderlying key bits.Gollakota et al. developed iJam, a robust friendly-jamming sys-tem which improved the physical-layer key generation for station-ary wireless networks [4]. The scheme lets the transmitter (Alice)interleave two identical sequences of bits while the intended re-ceiver (Bob) jams one at random. Since Bob knows which bits arejammed and which ones are clear, he can select the clear bits and re-construct the key, whereas the eavesdropper (Eve), being unawareof Bob’s jamming targets, cannot recreate the key. To increase therandomness of the channel, the key is repetitively transmitted untilBob reconstructs the original message.Although iJam can achieve secure key exchange in a static chan-nel, Steinmetzer et al. identified a vulnerability by implementing amulti-antenna adversarial model designed to take advantage of thespatial variance to discern between the jammed and clean signals[12]. The root of this vulnerability is due to the fact that the channelstates between Alice and Eve remain unchanged in spite of Bob’sjamming signal, which allows Eve, equipped with multiple anten-nas, to exploit the pilot or known symbols in Alice’s transmissionto estimate the channel and cancel its effect. Once Eve equalizes thechannel, she may evaluate the signal divergence among multipleantennas to identify the clear symbol. Specifically, a symbol with alarge divergence among multiple antennas is likely to be jammedand thus ignored. After a few iterations, clean transmission can bespliced together, and the key can be known.We propose a defense mechanism against such an attack bycombining channel randomization with prediction-based channelequalization. Channel randomization has been used to strengthenphysical-layer security schemes, such as orthogonal blinding, thatare known to be vulnerable against multi-antenna eavesdropper[1, 5, 6, 11]. The method leverages a reconfigurable or movingantenna to create artificial changes in a wireless channel, result-ing in unstable channel state information (CSI) between the trans-mitter and receivers. The prediction-based channel equalizationcancels the randomizing effect for Bob by implementing an angle-of-departure (AoD) estimation algorithm to predict the CSI for anygiven antenna configuration. The combined results are that thechannel state appears stable for Bob but continuously changing for a r X i v : . [ ee ss . SP ] J u l iSec ’20, July 8–10, 2020, Linz (Virtual Event), AustriaJordan L. Melcher, Yao Zheng, Dylan Anthony, Matthew Troglia, Thomas Yang, Alvin Yang, Samson Aggelopoulos, Yanjun Pan, and Ming Li Eve. Furthermore, the channel prediction eliminates the need for pi-lot based channel measurements, which denies Eve the opportunityto measure and cancel the changing effects.In this demonstration, we developed and implemented the chan-nel randomizing iJam system with a custom reconfigurable antennaand real-time AoD based channel prediction algorithm, to enhancethe security of the key generation protocol. The audience is grantedaccess to our remotely accessible platform via live-stream to controland observe the real-time effects of channel randomization on Boband Eve. The increase in entropy contributes to both the key gener-ation rate and the overall security. Our results indicate the CSI ofthe intended receiver bit error rate (BER) does not fluctuate whenexposed to channel randomization while simultaneously worseninga multi-antennae adversary’s BER to the level of random guessing.
Consider an OFDM system shown in figure 1, where there are threeactive parties; Alice, who wants to establish a secret key with anintended receiver, Bob, and a passive eavesdropper, Eve. Alice, ingreen, equips a rotating antenna to randomize the channel, anda digital RF chain consisting of a compressed sensing-based AoDestimation algorithm and a precoding filter to predict and cancelthe channel randomization effect for the intended receiver. Bob, inblue, comprises of two stationary antennas, one for receiving, andthe other for jamming; while Eve, in red, uses two antennas and anadaptive filter to exploit the spatial variance of the jammed signal.
The channel randomization is implemented by rotating a log-periodicdipole array (LPDA) antenna using a stepper motor. The half-widthbeam angle of the LDPA antenna’s main lobe is approximately 60degrees. Hence, rotating the antenna by 60 degrees can significantlychange the channel state. The rotating speed of the step motor canvary from 1 RPM to 5 RPM. The rotation is carried out in syncwith the data transmission, which prevents Eve from obtaining achannel equalization filter to correct the randomization effects.
In our scheme, Alice predicts and cancels the channel randomizationeffect for Bob with a compressed sensing-based AoD estimationalgorithm. The CSI between Alice and Bob (or Eve) is due to thecombined result of multipath components and antenna patterns.The compressed sensing algorithm allows Alice to estimate the360-degree AoD vector, which defines the multipath components.Given that Alice knows the antenna pattern for a specific antennamode, the CSI can be computed as the inner product of the AoDdistribution and the antenna pattern vector.To predict the wireless channel, Alice selects every antenna modethen transmits pilot symbols to Bob at each mode. Bob measuresthe corresponding CSI and sends it back to Alice using implicitfeedback. Alice then estimates the AoD vector with the compressedsensing algorithm. Once known, Alice can predict the CSIs for allunused antenna modes and computes the corresponding precodingfilter to cancel the changing channel effects for Bob.
With the addition of channel randomization and prediction we cre-ated a less complicated physical layer key generation method. Inthe original scheme, iJam, Alice and Bob switched roles betweentransmitting and jamming to prevent adversaries who were unaf-fected by the jamming signals. Channel randomization replacesthis method. In our scheme, Alice sends clean OFDM symbols toBob. Simultaneously, Bob transmits jamming signal at randomly se-lected time intervals. Alice repeats the same message until Bob hassuccessfully jammed every bit. During this process, Eve comparesthe captured waveforms from both of her antennas to differentiatebetween clean and jammed samples.The bandwidth and frequency are carefully selected to optimizethe trade-off between the speed and effect of randomization. Therotating antenna has a limit of 5 RPM with five unique channelmodes. To put less strain on the AoD algorithm while providingample security, 16-QAM is selected. To successfully transmit therepeated key before a new channel mode is selected, the transmis-sion bandwidth is limited to 3.4 KHz. The key bits are randomlygenerated at Alice’s side and shared with Bob through the softwaredetailed in the appendix.
ACKNOWLEDGMENTS
This work is partly supported by NSF grants CNS-1948568, DGE-1662487, ARO grant W911NF-19-1-0050, and Naval InformationWarfare Center Pacific.
REFERENCES [1] Tomoyuki Aono, Keisuke Higuchi, Takashi Ohira, Bokuji Komiyama, and HideichiSasaoka. 2005. Wireless secret key generation exploiting reactance-domain scalarresponse of multipath fading channels.
IEEE Transactions on Antennas andPropagation
53, 11 (2005), 3776–3784.[2] Dennis Goeckel, Sudarshan Vasudevan, Don Towsley, Stephan Adams, ZhiguoDing, and Kin Leung. 2011. Artificial noise generation from cooperative relaysfor everlasting secrecy in two-hop wireless networks.
IEEE Journal on SelectedAreas in Communications
29, 10 (2011), 2067–2076.[3] Satashu Goel and Rohit Negi. 2008. Guaranteeing secrecy using artificial noise.
IEEE transactions on wireless communications
7, 6 (2008), 2180–2189.[4] Shyamnath Gollakota and Dina Katabi. 2011. Physical layer wireless securitymade fast and channel independent. In
In 2011 Proceedings IEEE INFOCOM . IEEE,1125–1133.[5] Haitham Hassanieh, Jue Wang, Dina Katabi, and Tadayoshi Kohno. 2015. SecuringRFIDs by randomizing the modulation and channel.
In 12th USENIX Symposiumon Networked Systems Design and Implementation (NSDI
15 (2015), 235–249.[6] Yantian Hou, Ming Li, Ruchir Chauhan, Ryan M. Gerdes, and Kai Zeng. 2015.Message integrity protection over wireless channel by countering signal can-cellation: Theory and practice. In
In Proceedings of the 10th ACM Symposium onInformation, Computer and Communications Security . ACM, 261–272.[7] Jinsong Hu, Shihao Yan, Feng Shu, Jiangzhou Wang, Jun Li, and Yijin Zhang.2017. Artificial-noise-aided secure transmission with directional modulationbased on random frequency diverse arrays.
IEEE Access
Wireless Networks
21, 8 (2015), 2631–2647.[9] Qiang Li and Wing-Kin Ma. 2011. A robust artificial noise aided transmit designfor MISO secrecy. In
In 2011 IEEE International Conference on Acoustics, Speechand Signal Processing (ICASSP) . IEEE, 3436–3439.[10] Wei-Cheng Liao, Tsung-Hui Chang, Wing-Kin Ma, and Chong-Yung Chi. 2010.QoS-based transmit beamforming in the presence of eavesdroppers: An optimizedartificial-noise-aided approach.
IEEE Transactions on Signal Processing
59, 3 (2010),1202–1216.[11] Yanjun Pan, Yao Zheng, and Ming Li. 2020. ROBin: Known-Plaintext AttackResistant Orthogonal Blinding via Channel Randomization. arXiv:2002.07355 [cs,math] (Feb. 2020). http://arxiv.org/abs/2002.07355 arXiv: 2002.07355.[12] Daniel Steinmetzer, Matthias Schulz, and Matthias Hollick. 2015. Lockpick-ing physical layer key exchange: weak adversary models invite the thief. In
Proceedings of the 8th ACM Conference on Security & Privacy in Wireless and emo: iJam with Channel Randomization WiSec ’20, July 8–10, 2020, Linz (Virtual Event), Austria
Figure 1: System Diagram (a) (b) (c)(d) (e) (f)
Figure 2: From top to bottom, left to right: (a) Network setup for remote demonstration. (b) Physical setup with Alice on theright, Bob and Eve on the left. (c) The radiation pattern of the LDPA antenna. (d-f) LabView panels to visualize CSIs and datatransmissions.
Mobile Networks - WiSec ’15 . ACM Press, New York, New York, 1–11. https://doi.org/10.1145/2766498.2766514[13] Joao P. Vilela, Matthieu Bloch, Joao Barros, and Steven W. McLaughlin. 2011.Wireless secrecy regions with friendly jamming.
IEEE Transactions on InformationForensics and Security
6, 2 (2011), 256–266.[14] Xiangyun Zhou and Matthew R. McKay. 2010. Secure transmission with artificialnoise over fading channels: Achievable rate and optimal power allocation.
IEEETransactions on Vehicular Technology
59, 8 (2010), 3831–3842.