While advances in data encryption have fuelled the progress in secure communication thus far, transmitted data is only one part of the information available to an observer. Even when transmitted data is encrypted, the transmission timing of nodes in a network can reveal critical information such as the source-destination pairs and the routes of information flow in a network . Modification of network protocols to minimize this inference however, results in a reduced network performance. Our goal is to understand fundamentally the extent to which such information retrieval from transmission timing can be prevented, and in the process, characterize the minimum "price" to be paid in terms of network performance.
The problem of hiding packet sources on the Internet was first addressed through the concept of Mixing by David Chaum . A mix is a proxy server/relay node that collects packets from multiple users and by re-encryption and random bit padding ensures that packet contents cannot be used to match incoming with outgoing packets at the mix. In addition, the mix waits until it has received packets from multiple users and transmits all of them in a single batch thus limiting an eavesdropper's ability to identify the sources of outgoing packets from the transmission timing. As expected, the anonymity of the sources of packets achieved through mixing comes at the cost of high packet latency. If, however, the packets are subjected to strict delay constraints, the capabilities of a mix are limited potentially resucing the achievable anonymity.
In the following papers, we analyze this fundamental tradeoff between anonymity and latency information theoretically. In particular, we quantify the anonymity of a network of mixes using the Shannon entropy of an eavesdroppers "estimate" of the packet sources, and propose delaying and reordering strategies for the mixes. We derive analytical lower and upper bounds on the maximum achievable anonymity as a function of the maximum packet latency, and characterize the optimal tradeoff under light traffic conditions.
"On the Anonymity of Chaum Mixes" abstract, pdf P. Venkitasubramaniam and V. Anantharam 2008 IEEE International Symposium on Information Theory, Toronto, Canada, July 2008.
"Anonymity under Light Traffic using a Network of Mixes" abstract, pdf P. Venkitasubramaniam and V. Anantharam 47th Annual Allerton Conference on Communications, Control, and Computing, Monticello, IL, Sep. 2008.
When long streams of packets are communicated between users, correlation between interarrival times at the source and relays can eaily reveal the path of data flow in a network. Under constraints on delay or memory, flow correlation based detection techniques showed that mixing strategies are ineffective for long streams . A key requirement for protecting long packet streams is the transmission of cover traffic, ie. dummy packets, to make otherwise correlated traffic flows appear independent. In wireless networks, owing to constraints on shared bandwidth and interference, dummy transmissions result in a tension between the anonymity that can be provided to routes and the achievable data rates. Our goal is to develop a theoretical framework and characterize the fundamental trade-offs between anonymity of routes and the achievable throughput in general multihop networks.
A basic building block for this framework is a multiple access relay which forwards streams of packets from multiple sources. In the following papers, we characterize inner and outer bounds on the achievable rate region of an anonymous multiaccess relay under orthogonal physical layer signaling strategies.
"Relay Secrecy in Wireless Networks with Eavesdropper" abstract, pdf P. Venkitasubramaniam, T. He and L. Tong Forty Fourth Annual Allerton Conference on Communication, Computing and Control, Monticello, IL, Sep. 2006.
"Networking with Secrecy Constraints" abstract, pdf P. Venkitasubramaniam, T. He and L. Tong 2006 IEEE MILCOM, Washington D.C., Oct. 2006.
In a general multihop network, if all relays generate cover traffic, an eavesdropper would obtain no information about any of the routes thus guaranteeing maximum anonymity. However, as shown in the following paper, such a strategy would result in an exponential reduction in throughput as the route lengths increase.
"Packet Scheduling against Stepping Stone Attacks with Chaff" abstract, pdf P. Venkitasubramaniam, T. He and L. Tong 2006 IEEE MILCOM, Washington D.C., Oct. 2006.
In order to design the transmission strategies in a general multihop network, we propose an information-theoretic metric of anonymity using the equivocation of routes given the eavesdroppers complete observation. We use the measure to optimize the set of relays to generate independent schedules so that the desired level of anonymity is achievable with minimum loss in network throughput. Under certain conditions, this optimization is shown to be equivalent to an information-theoretic rate-distortion optimization . The following papers detail the analysis, and characterize the tradeoffs between anonymity and network performance under different physical layer and delay models.
"Anonymous Networking amidst Eavesdroppers" abstract, pdf P. Venkitasubramaniam, T. He and L. Tong IEEE Transactions on Information Theory: Special Issue on Information-Theoretic Security, Vol. 54, No 6, pp. 2770-2784, June 2008.
"Anonymous Networking with Minimum Latency in Multihop Networks" abstract, pdf P. Venkitasubramaniam and L. Tong 2008 IEEE Symposium on Security and Privacy, Oakland, CA, May 2008.
"Throughput Anonymity Tradeoff in Wireless Networks under Average Latency Constraints" abstract, pdf P. Venkitasubramaniam and L. Tong 2008 IEEE INFOCOM, Phoenix, AZ, April 2008.
It is well known that  peer review in many journals suffers from large delays, and the primary reason attributed is the lack of real incentive for reviewing. Our goal is to study a system where fast reviews are incentivized using quantifiable public reputation for researchers. Since the reputation is used to prioritize a researchers submitted papers, increasing ones score can motivate expedient reviews by a researcher. However, the availability of the public scores can potentially compromise the anonymity of the reviewers. In the following paper, we model such a system mathematically, and study its viability in incentivizing fast reviews and protecting reviewer anonymity from timing analysis.
"Incentivizing Anonymous 'Peer-to-Peer' Reviews" abstract, pdf P. Venkitasubramaniam and A. Sahai 47th Annual Allerton Conference on Communications, Control, and Computing, Monticello, IL, Sep. 2008.
While statistical inference is a well-studied subject in literature, the advent of networked sensing has sparked a renewed investigation due to critical properties such as low-energy communication and distributed control. Mathematically, the task of a network of sensors- collectively sensing phenomena such as the location of an object, presence of a chemical, atmospheric condition etc,- is one of distributed statistical inference; sensors measure observations, and transmit data to a fusion center (such as a base station or a collection agent) which processes the gathered data to infer the underlying phenomenon.
In such networks, many questions arise from a network designers perspective: How to communicate data from sensors to the fusion center in a decentralized manner? How does the limited power available at sensor nodes affect the network performance? How to best compress the sensor data with minimum effect in the global statistical inference? My thesis work addresses some of these questions by using mathematical abstractions to develop communication and compression protocols and characterizing their fundamental performance limits.
Compression of sensor observations prior to collection is different from classical data compression as the goal in a sensor network is to infer an underlying phenomenon from the gathered data rather than reconstruct the original observations. When the inference is modeled as distributed detection problem, the optimality of Likelihood-Ratio Quantizers (LRQ) has been established under general performance criteria . In the following paper, we obtain a parallel result for the distributed estimation of an underlying parameter by developing the notion of a Score function quantizer.
"Quantization for Distributed Estimation in Large Scale Sensor Networks" abstract, pdf P. Venkitasubramaniam, G. Mergen, L. Tong and A. Swami Third International Conference on Intelligent Sensing and Information Processing, Bangalore, India, Dec. 2005.
Score-function quantizers can be used to provide tight benchmarks for performance in general maximum likelihood estimation problems. However, unlike LRQs, they vary with the underlying parameter and do not directly lend themselves to practical quantizer design. In the following papers, we present an adaptation of SFQs to a large scale distributed estimation problem, and study their performance under different criteria.
"Score-Function Quantization for Distributed Estimation"abstract, pdf P. Venkitasubramaniam, L. Tong and A. Swami 40th Annual Conference on Information Systems and Sciences, Princeton, NJ, March 2006.
"Quantization for Maximin ARE in Distributed Estimation" abstract, pdf P. Venkitasubramaniam, L. Tong and A. Swami IEEE Transactions on Signal Processing, Vol. 55, Issue 7, pp. 3596-3605, July 2007.
The conventional approach to protocol design in networks is based on the separation of layers in the standard OSI model , with independent design of tasks in each layer. We, however, demonstrate that a joint design that combines channel estimation (PHY layer), random access (MAC layer) and coding (DATA LINK layer) can achieve optimal network performance in a decentralized manner with as little per-node power as possible. We consider Opportunistic ALOHA , a random access protocol, wherein the transmission policy of each sensor is a function of its own local information. In the following papers, we optimize the decentralized transmission controls and design coding schemes for sensors in a large network, and demonstrate the asymptotic optimality, reliability and robustness of the cross-layer design.
"Sensor Networks with Mobile Access Points: Optimal Random Access and Coding" abstract, pdf P. Venkitasubramaniam, S. Adireddy and L. Tong IEEE Journal for Selected Areas in Communication, Vol. 22, Issue 6, pp. 1058-1068, Aug. 2004.
"Sensitivity and Coding of Oppertunistic ALOHA in Sensor Networks with Mobile Access" abstract, pdf P. Venkitasubramaniam, S. Adireddy and L. Tong IEEE Journal for VLSI Signal Processing, Vol. 41, Issue 3, pp. 329-344, Nov. 2004.
When the sets of sensors activated by multiple collection agents overlap, then sensors in the overlapping region have a higher chance of successful transmission due to receiver diversity. As a result, it may be more efficient to use inter-sensor routing to transfer a portion of the data to the overlapping region prior to transmitting to the access points. In the following papers, we study the distributed data collection problem at multiple access points, and derive conditions under which inter-sensor routing and direct multiaccess transmission strategies are optimal.
"Energy Efficient Data Collection in Sensor Networks with Multiple Mobile Access Points"abstract, pdf P. Venkitasubramaniam and L. Tong 39th Annual Conference on Information Systems and Sciences, Princeton, NJ, March 2005.
"Sensor Networks with Multiple Mobile Access Points"abstract, pdf P. Venkitasubramaniam, Q. Zhao and L. Tong 38th Annual Conference on Information Systems and Sciences, Princeton, NJ, March 2004.