National Conference on Communications

Plenary Talks

• 1.Vijay Kumar Saraswat, Scientific Advisor to the Raksha Mantri; Secretary, Defence R&D
  Next Generation Military Networks
• 2.Subhasis Chaudhuri, IIT Bombay
  Retargeting of lecture video: Extending the outreach of distance education Abstract
  
  How do we retarget (repackage) instructional videos to deliver it without any significant loss in its pedagogic content and yet achieve a very high level of compression? We shall discuss methods that collectively contribute to developing technologies for distance education, so that its outreach could be extended even to mobile platforms, enabling a student to study on the move. The key idea is that a programmed playback of the (cleaned up) keyframes from the video along with the audio, according to a metadata derived from the original video will suffice to render an estimate of the original video, without much loss in instructional values. Further, we will demonstrate how to effect a legibility retentive display of instructional media on miniature display devices. Details of the technology developed at IIT Bombay will be explained during the talk.
• 3.K. Giridhar, IIT Madras
  Reality constrained capacity – the curious case of mobile broadband access systems Abstract
  
  Wireless cellular networks have changed the way we communicate, and have been able to offer high voice capacity (simultaneous voice calls that can be supported per cell site). The migration to 3G and later, 4G cellular networks, brings in several new challenges, one of them being the ability to support variable bit rate data services in a packet switched air interface. Here, it is important to serve a highly disparate set of wireless links by resorting to rate control and judicious resource allocation. With this, the meaning of system capacity needs to be clarified -- whether it indicates cell average spectral efficiency, or the ability to ensure throughput guarantees to various user/services, or both.
  
  The fact that these mobile broadband networks will also encourage the use of relays, micro, and pico base-stations, in order to multiply the bit-rate that can be carried per square kilometer, brings into question on what will be the best practices to enhance end-to-end performance for most users on such networks, even while keeping the costs low for the operator.
  
  In this talk, we discuss the roles played by various practical issues that affect the system capacity in MIMO-OFDM mobile broadband -- including reuse-1 deployment, constraints imposed in the standard by way of pilot designs, and backward compatibility. We also look how closed loop schemes (beamforming, multi-user MIMO, best-band feedback), unequal cell loading, and finite buffer traffic will impact interference limited scenarios, and then motivate the necessity to create stable zones where interference can be accurately measured. The good news and the bad news with having zones in the transmit frame will be delineated.
  
  We describe the importance of careful estimation (of channel, of covariances), interference aware symbol decoding, robust LLR estimation, and careful resource allocation between macro-picos. On top of this, the role of channel aware and interference aware scheduling is also key in getting good cell average spectral efficiencies even while meeting minimum throughput guarantees for cell edge users.
  
  Since mobile broadband will provide the basic broadband access to Internet in countries like India, one is curious to see how such high capacities will be achieved. In other words -- what modes will the MIMO-OFDM air-interface use, and what will be the deployment strategy of the operator? There is much to be learnt in the coming years.PDF of presentation
  

Invited Talks

Indian Languages Processing

• 1.A. G. Ramakrishnan, IISc Bangalore
  Enriching quality of life through language technology Abstract
  
  One of the commitments of Medical Intelligence and Language Engineering Laboratory is to develop technology that empowers a person with visual disability to access knowledge available in any printed material in Indian languages. We are working on all research and development issues leading to the fulfillment of this goal: mosaicing of coloured document images; text extraction from complex colour images, including camera captured images; document layout analysis; detection of broken and merged characters; OCR technology for Tamil and Kannada; text to speech conversion in Tamil and Kannada; pitch modification using discrete cosine transform in the source domain; automated parts of speech tagging; phrase prediction and prosody modeling. Our demo version of Tamil OCR is being used by Worth Trust, Chennai and Indian Association for the Blind, Madurai to convert printed books into computer readable text and Braille format. Our Tamil TTS is already being used for student assignments by some school teachers in Singapore.
  
  We are also working on Online handwriting recognition (OHWR) in Tamil and Kannada. Our research consortium partners in IIT Madras, IIIT Hyderabad, ISI Kolkata and CDAC Pune are working on Telugu, Malayalam, Bangla and Hindi. Currently we have technology that recognizes unlimited vocabulary words with a character accuracy of about 85%. Three form-filling applications have been developed by our industry partners, integrating word-level recognition engines. Combining our OHWR engines and TTS, we have a preliminary demo of a handwritten word to speech demo in both Tamil and Kannada. In collaboration with St. Johns Medical College, we plan to explore the use of the same for persons with vocal disability. We are also working on porting our technologies to Android based mobile platform. The various databases required, recognition technology, coding in C and the natural language processing modules have all been fully developed by us. Technology created at the MILE Lab in the Department of Electrical Engineering will be demonstrated during the talk.

Communications

• 1.Gerhard Fettweis, Technische Universität Dresden, Vodafone Chair Mobile Communications Systems
  Challenges in next generation cellular communications, and possible ways forward Abstract
  
  As LTE, the 4th generation cellular technology is starting roll-out, the next update is being prepared in research and standardization: LTE-Advanced. Current day cellular standards have been created to maximize the data rate at given SINR. The next step in cellular is to manage and cancel interference. This includes advanced precoding for the downlink as well as MIMO CoMP (cooperative multi-point) for the uplink. To ensure that theoretical results match up with scientific measurements of the real world, an extensive test-bed has been setup in Dresden. An overview of theoretically achievable chances and gains, as well as testbed results will be provided. One main goal is to understand the limitations of CoMP which were uncovered within the testbed results, in particular when applying the technology for rural areas to enable large-scale broadband internet access to villages via cellular communications. A set of challenges has been identified when trying to cover rural areas with broadband coverage everywhere.PDF of presentation
  
• 2.Srikrishna Bhashyam, IIT Madras
  Coding schemes for bidirectional relaying Abstract
  
  In the bidirectional (or two-way) relaying problem, two nodes want to exchange messages through an intermediate relay node. All the nodes are assumed to be half-duplex. Communication happens in two phases: a multiple access phase and a broadcast phase. Several relaying schemes have been proposed for this problem in the last few years. Physical layer network coding, and more generally, compute and forward schemes perform well. In this talk, an overview of the achievable rate pairs for various relaying schemes and capacity outer bounds will be presented. Then, new code designs to achieve these rate pairs will be presented.
• 3.D. Manjunath, IIT Bombay
  Network flows for functions Abstract
  
  We consider in-network computation of an arbitrary function over an arbitrary communication network. A network with capacity constraints on the links is given. Some nodes in the network generate data, e.g., like sensor nodes in a sensor network. An arbitrary function of this distributed data is to be obtained at a terminal node. The structure of the function is described by a given computation schema, which in turn is represented by a directed tree.
  
  We introduce a novel flow-based technique for distributed computation. We design computing and communicating schemes to obtain the function at the terminal at the maximum rate. For this, we formulate linear programs to determine network flows that maximize the computation rate. We then develop a fast combinatorial primal-dual algorithm to obtain ε-approximate solutions to these linear programs. We then describe extensions of our techniques to the cases of multiple terminals wanting different functions, multiple computation schemas for a function, and some practical variants of the problem set-up.PDF of presentation
  

Networks

• 1.Ranjita Bhagwan, Microsoft Research India
  Automating problem diagnostics for Access Control Abstract
  
  Diagnosing the cause of networking problems and finding the correct fixes is a difficult problem. The problem is particularly acute in home networks because of the heterogeneity of the environment and, more importantly, the lack of a network administrator in this setting. Enterprise networks suffer from a similar issue as well since the interaction of various networking components such as firewalls, routers and end-hosts makes it difficult to diagnose and pinpoint the cause of problems. Consequently, there is a need for structured and automated techniques to facilitate problem-solving in these complex environments. In this talk, I will concentrate on two techniques towards diagnosing and fixing networking problems. The first involves using a decision tree classifier to find configuration fixes in a home networked environment, and the second uses learning techniques to deduce signatures of networking problems from packet traces which can then be used to match new problems to existing ones to which solutions are known.Powerpoint presentation
  
• 2.Vinay Ribeiro, IIT Delhi
  Wireless TDMA mesh networks Abstract
  
  Wireless TDMA mesh networks are multi-hop wireless networks which give various nodes access to the wireless medium based on a time-division schedule. Such networks have the potential to provide QoS for triple-play over a large coverage area. It is hard to give equivalent QoS with Wi-Fi based networks which are fraught with various hidden terminal problems.
  
  At IIT Delhi, we are building a rapidly deployable wireless TDMA network based on the WiMAX standard. Our network will be useful in disaster management and military battelefield scenarios which cannot depend on any prior existing network for communication. We have chosen a software-defined radio, called wireless open access research platform (WARP), to build our backbone mesh network.
  
  In my talk, I will explain our network design, provide preliminary experimental results, and describe applications we are building for disaster management. I will also highlight our experiments with commercial wireless TDMA mesh equipment which we plan to leverage as a second-tier in our own mesh network.PDF of presentation
  
• 3.Vikram Srinivasan, Bell Labs Research, Alcatel Lucent
  Designing wireless networks over large and fragmented spectrum Abstract
  
  There is tremendous interest in the research community in the design of frequency-agile wireless networks. By frequency-agile we mean that wireless nodes have the ability to tune across large swathes of spectrum. This is motivated by a confluence of emerging technology trends and new spectrum regulations. Recent advances in radio technology make it possible for radios to tune their center frequencies across large swathes of spectrum (100 MHz-8 GHz). Also, these radios can not only tune their center frequency, but these can also adapt the operating bandwidth. Secondly, a recent regulation by the FCC allowed the use of unutilized Digital TV Spectrum for unlicensed access. This implies that we now have unlicensed spectrum available from a few 100 MHz (512-698 MHz in DTV whitespaces) all the way up to a few GHz (2.4 GHz and 5.3 GHz ISM bands). A key observation is that different frequency bands have different propagation properties. The lower frequency bands propagate a lot farther than the higher frequency bands. While this implies that lower frequency bands provide better data rates at a fixed distance, it also implies that the lower frequency bands create higher interference to neighboring links. Designing practical networks that operate across such diverse spectrum and manage interference is a challenging problem. We will highlight some of the design challenges and outline some solutions that we have developed to address these challenges. PDF of presentation
  
  

Signal Processing

• 1.Ajit Rao, TI India
  Optimal systems architecture for high-density video Abstract
  
  The industry is currently witnessing a growing demand for high density video systems for multi-channel decode, encode, transcode, processing, composition, analytics, and ad insertion. We will present the toughest challenge in designing these systems - specifically the maximization of throughput given a power and cost budget. Our proposed solution relies on the use of multiple Systems-on-chips (SoC) on a PCI backplane. We will describe how we can achieve the target power and cost budgets by optimal hardware/software partitioning and through improvements in SoC systems architecture.PDF of presentation
  
• 2.Arjun Arunachalam, IIT Bombay
  Non-Fourier encoding for fast MRI Abstract
  
  Magnetic Resonance Imaging (MRI) is a powerful non invasive imaging modality that unlike computed tomography(CT), offers images with superior soft-tissue contrast and allows imaging planes in any arbitrary orientation. However the speed at which the data can be collected in MRI is limited by physical and physiological constraints which results in longer acquisition times making non stationary regions of the body difficult to image. This has led researchers seek for new methods to reduce the amount of acquired data without degrading the image quality. Parallel Imaging (PI) which has been the single biggest innovation in MRI in the last decade, is a technique which makes use of multiple receiver antennas(coils) to exploits the independence of the coil sensitivities to augment the time consuming Fourier encoding and thereby reduce acquisition times significantly. Further PI can be combined with non-Fourier(NF) spatial encoding in a complementary fashion as two spatial encoding mechanisms are physically and analytically separable.
  
  This talk will focus on our research (successes and failures) in combining non-Fourier signal encoding strategies with the MRI data acquisition process to accelerate MRI scans.PDF of presentation
  
• 3.Mahadeva Prasanna, IIT Guwahati
  Speaker-specific source information from linear prediction residual Abstract
  
  The speaker-specific information present in the speech signal is attributed mainly to the source and system components that involve in its production. Most of the existing speaker recognition systems exploit the speaker-specific system information. Alternatively, speaker-specific source information is less explored and possible reasons for the same are initially presented in the talk. Linear prediction residual is being used as the best approximation for the source component. The major focus of the talk will therefore be on presenting approaches for extracting and modelling speaker-specific source information from the linear prediction residual.PDF of presentation