Research

Ravi Prakash's research interests include 360-degree video streaming (specifically Field-of-View prediction and mobile edge caching), wireless networking, sensor networks, and mobile ad hoc networks. He tries to develop distributed solutions for various networking problems, and evaluate their performance through simulation experiments and small scale implementations. The next few paragraphs provide additional information about the specific research problems. However, if you are the sensible type, not wasting your time reading self-serving descriptions, you can go directly to the Journal and Conference publications

360-degree Video Streaming

As network bandwidth increases, new applications are conceived that consume that increased bandwidth. 360-degree video streaming is one such application. Such videos need to be streamed at high resolution, especially when viewed on head-mounted displays. As viewers move their head the view changes, offering them an immersive experience. The viewports of head-mounted displays typically cover only a small fraction of the total 360-degree immersive space and limit viewers' Field-of-View. By some estimates, a viewport only covers a sixth of a 360-degree video frame. This observation can be exploited to reduce bandwidth consumption of 360-degree videos as follows: (i) divide the frame into multiple tiles, and (ii) only stream the subset of tiles that cover the viewer's Field-of-View. Additionally, collective viewing behavior of multiple viewers can be used to identify hot spot(s) in a video frame more likely to be viewed by future viewers of the same 360-degree video. Tiles corresponding to these hotspots could be cached/prefetched at edge-servers to reduce traffic in the core network and reduce download latency. We have found innovative solutions for Field-of-View prediction and caching by employing ideas from the field of voting algorithms, and continue to develop new algorithms.

Performance of Wireless Networks

Some wireless networks (for example, cellular networks) operate in licensed frequency bands and have exclusive rights to these bands. Having paid a handsome amount of money for such rights, the service providers wish to make the most of the spectrum they have. So, efficient spectrum utilization is of utmost importance to them. Other wireless networks operate in unlicensed frequency bands. It is a veritable jungle out there with multiple networks, based on different technologies, vying with each other to access the spectrum. An example is the 2.4 GHz ISM band where the IEEE 802.11 Wireless LAN, IEEE 802.15.4 WPAN/ZigBee, and Bluetooth networks operate. Just to make life a little more interesting, the good old microwave oven also offers interference in that band (so, wirelessly surfing the web while warming that stale coffee in the microwave oven is problematic in more ways than one!). Before we start developing new solutions to promote co-existence of these networks in the same band, it is important to fully understand how they interact with each other and how effective are the existing solutions. While analytical modeling and simulation experiments can be useful, they also have their limitations. So, we have built a wireless networking testbed to perform controlled and repeatable experiments and find answers to life's persistent questions (a la Guy Noir, Private Eye).

Sensor Networks

One model of sensor networking assumes that the sensor nodes are resource poor (limited amount of stored energy, limited processing power, low data rate transceivers, small communication range, etc.). These nodes form a multi-hop wireless network with a path from each node to the sink where sensor data is analyzed. In order to prolong the duration for which the network can operate it is imperative that energy consumption be kept to a minimum.

One approach is to reduce the energy spent in communication. In this context, contention-based channel access protocols are not desirable. So, we have proposed distributed algorithms for link scheduling in wireless sensor networks, based on Vizing's Theorem. A node communicates with each of its neighbors in a different time slot, avoiding collisions and offering more communication opportunities to nodes with a larger number of neighbors.

Another approach is to save energy by reducing the overall volume of traffic in the sensor network. A variety of data aggregation algorithms have been proposed for this purpose. The simplest aggregation approach utilizes a tree rooted at the sink, and sensor data is convergecast towards the sink. Each node aggregates its own data and the data received from its children and sends the result to its parent. While this has the potential to significantly reduce the volume of traffic, values reported by individual sensors are ususally filtered out. This can severely hamper the ability to answer complex queries and perform a posteriori analysis of data. So, we have proposed to logically organize the sensor field as a forest of trees, based on the gradient of sensor values. Only the summary data is communicated to the sink, while the sensor network maintains sufficient information to answer complex queries in a time- and bandwidth-efficient manner.

Mobile Ad Hoc Networks and Vehicular Ad Hoc Networks

While a considerable amount of initial research on routing in MANETs was focused on bidirectional wireless links, we felt that unidirectional wireless links are likely to be present in such networks. Routing protocols that implicitly assume all links to be bidirectional will not function efficiently in the presence of unidirectional links. So, we studied the impact of unidirectional wireless links on MANET routing and proposed solutions for the problem.

We also developed MANETconf, one of the first protocols for distributed, dynamic IP address assignment to nodes in a MANET. Subsequently, we proposed anothed algorithm that utilizes the buddy data structure.

Vehicular networks that rely solely on car-to-car communication are referred to as vehicular ad hoc networks (VANETs). For safety-critical applications in VANETs (like braking) it is important that messages transmitted by one vehicle reach other vehicles who may benefit from such messages in a timely manner. We have proposed channel access scheduling protocols for such scenarios.