Research
Current Interests: Wireless and Wired networks, Wireless Sensor Networks (WSN), Software Defined Networking (SDN), network protocols, Operating Systems and Security
Most of the WiFi deployment commonly faces low throughput problem on regular basis. Customers/Corporate employees often complain about non working of WiFi. Sysads usually have a tough time in figuring out the cause of poor performance specially when everything seems fine from theory prespective i.e. clean channel etc. To solve this problem many diagnostic frameworks have been developed but most of them requires some kind of client's help or additional infrastructure in place. But having extra infrastructure is too costly and infeasible to deploy in companies, hostels etc. Furthermore asking every client to take part in trace collection causes discomfort. So recently more work is focused on AP (access point) centric WiFi diagnosis system. Here AP collects traces and mostly analyze it at central server. Few of these works require additional radio at AP and others who do not require additional radio are not complete, they seldom give any reason for bad performance of WiFi.
We have developed a framework to diagnose WiFi with AP centric approach that does not require any extra infrastructure or client's help. We do not require any extra radio at AP. Our diagnosis algorithm outputs the problem and real cause of the problem in WiFi if problem really occurs at wireless network. Our framework requires little modifications at driver level and runs mostly at user level. Instead of analysing at some central location, we are using AP for analysis itself. Our software is light weight hence does not effect the high speed WiFi (802.11n) performance. We installed this modified AP (running our software) in both AP mode as well as sensor mode at various locations for various experiments (both controlled and real life). Our alogrithm diagnose WiFi in real time and we found that diagnosis output matches with the ground truth.
Smartphones and tablets gave us a new way of interacting with a class. Through these devices instructor can distribute some materials, can ask them to upload some files, can take quizzes etc. In general class consist of 50+ students and they try to access the server simultaneously. Since all of them are connected to server via WiFi, this puts a huge overload over WiFi access point. It has been figured out by a PhD student in IIT Bombay that throughput of 802.11g decrease, once it crosses the certain number of client connections. Me and my colleague initially helped her in analyzing the dense(multiple clients) WiFi network(802.11g) and then we moved on to 802.11n.
From the data collected over a single client experiments we deduce that there are discrepancies in expected throughput and measured throughput in 802.11n. By extensive range of experiments we concluded a conjecture that some of the frames from aggregated frames are dropped because client is not able to handle the interrupts at that pace (as interrupts are generated per frame). Furthermore, rate adaptation algorithm think of these non-wireless losses as wireless loss. Hence, we see drop in data rate as well as throughput.
Aim was to design a mechanism such that video transmission is smooth over a multiple input multiple output(MIMO) network and can guarantee certain level of video quality under all circumstances. We developed the scheme over MIMO wireless channel, using wavelet layering and scalable video coding. Under the scheme, TCP was modified at both ends to intelligently drop the packets to avoid jitter in the video arisen from packet transmission error. Most of the packets were retransmitted in case of error just like normal TCP, only some packets were dropped to avoid jitter at the receiver end. We implemented the scheme over ns2 to prove it's productivity. We implemented/modified following network layers:
- Developed new application layer compromising wavelet layering, scalable coding, data distribution.
- Modified transport layer to support packet drop as described in the scheme.
- Modified physical layer to support MIMO wireless systems.
Aim was to develop a new key management scheme for WSNs that can be resilient against node capture attack(as it is supposed to be deployed in hostile environment) while keeping the energy requirement low. We developed a scheme based on assumption that base station(BS) does not suffer from resource constraints. Our scheme consists of base station and normal WSNs. Periodic transmissions from BS acts as a core of the scheme. The scheme consists of two stages. In first stage, BS computes two matrices then deduce another matrix from these two and finally assigns key to each node. In second stage, every node who wants to communicate uses a periodic update from message to establish a secure communication link with each other. Even when one of nodes were captured and key is revealed, other keys will remain safe. Therefore communication can not be compromised. Our previous work on related topic involved comparison of some of the existing key management scheme. We implemented those algorithms in tossims and compared the results.