Victor Bahl
 |
Principal Researcher & Manager |
|
|
Completed
Projects (Consultant, Mobile Systems Division, Business Communications Networks Inc.) MediCard was a "smart card" designed and targeted toward the Insurance Industry. We designed this card to be fairly modular and versatile. Its dimensions were 3.5"(L) x 2"(B) x 0.375"(H), which are slightly more than that of a typical credit card. The MediCard could store personal information including medical records, prescription information, insurance records, and reminders in addition to the basic functions such as calculator, calendar, alarm clock etc. It had a built-in spread spectrum 900 MHz RF transceiver which allowed location tracking and paging. The card could be remotely programmed using existing phone-line infra-structure and featured a 16x1 LCD display, a two-key membrane keypad, an acoustic modem and a indicator LED. --- Product Shipped --- (Project Lead & Principal Engineer, Multimedia Engineering, Digital Equipment Corporation) SLIB was/is a software library designed to include highly optimized implementations of ISO, ITU-T, and industry de-facto audio/video compression algorithms along with Digital's proprietary video rendering algorithms. Applets layered on SLIB are able to encode and decode multiples of SIF/CIF size video bitstream at or near real-time rates (25-30 fps). The video rendering algorithms allow flexible and fast display of multi-spectral images on a variety of frame buffers. Our primary motivation behind this work (initially) was to highlight the power of the Alpha processors by showing-off real-time video coding/decoding operations on AXP based workstations and PCs. SLIB was demonstrated on several platforms including Unix, OpenVMS and NT. A fun side-project for me that got unexpectedly good coverage was a video screen saver application nicknamed Video Odyssey which would decode and display multiple video streams inside several bouncing windows on AXP systems. The decoding and rendering was all done in software. Additional information about the motivation, architecture and algorithms in SLIB is available in this paper. A significant amount of additional work and effort have gone into the library since this paper was written in 1994 --- Product Shipped --- (Principal Engineer, Multimedia Engineering, Digital Equipment Corporation) J300 was Digital’s flagship high-performance TURBOchannel-based feature-rich multimedia hardware adapter with Motion-JPEG compression, proprietary video rendering, analog video I/O and DSP based audio compression. I was primarily involved with the design and development of the software architecture for this board. The challenge for us was to extract real-time performance from a non-real time operating system (Unix). The architecture we developed for this product became the basis for follow-up products including FullVideo Supreme and FullVideo Supreme-JPEG, both of which were offered on Unix and NT. Details on the design philosophy behind the hardware architecture is available in this paper and behind the software architecture is available in this paper -- both of which were published in 1995 in the Digital Technical Journal - Special Issue on Audio/Video Technologies. --- Prototype Shipped to Universities--- (Principal Engineer, Multimedia Research, Digital Equipment Corporation) Started in early 1990, Jvideo was an A/D project for studying the feasibility of managing and manipulating digital audio and video on personal workstations. It was a TURBOchannel based h/w board that used two C-Cube’s CL550 chips for simultaneous JPEG compression and decompression (the prevailing standard at the time). We experimented with different software designs to identify and fix performance bottlenecks when operating over a non real-time operating systems (Ultrix). Also studied was the problem of audio-video synchronization and of providing an effective API for multimedia programming. The on-board DSP was used for audio compression and audio processing (sample rate conversions, gain control, format conversion etc.). J-Videos were used extensively both internally for research in high-speed (ATM) networking and externally for research on packet video (nv, ivs) in several projects including: Sequoia 2000, BERKOM, BAGnet and MBONE. Jvideo was possibly the first high-visibility project in the industry that proved that management and manipulation desktop video was feasible. Its achitecture was unique in the sense that it was independent of the system's frame buffer. Several follow-on products were built because of the success of Jvideo. --- Product Shipped --- (Senior Engineer, Image Processing Research, Digital Equipment Corporation) This was my first project out of school. Pictor was an advanced development effort within Digital's Imaging Research Group focussed on providing the ability to render (scale, quantize, dither, tone-adjust and sharpen) color and monochrome images in a device independent manner. As part of the effort, I studied and implemented several multi-bit image halftoning algorithms including void and cluster dithering, error diffusion, blue noise dithering, dispersed-dot ordered dithering and histogram based methods such as the median-cut algorithm. Working with Robert Ulichney, I developed a novel fast method for mapping YUV dithered points to RGB color cube while providing controls for brightness, saturation and contrast in multi-spectral images. This method was patented and subsequently used in several multimedia/graphics products marketed by Digital. |
Completed Research(Research Associate, Advanced Communications Technologies Lab., Boston University) Research Collaborator - Tong Liu (Tellabs Inc.) We explored the fundamental problem of providing lifetime connectivity to on-going sessions initiated by mobile users in a cell based wireless network. Employing well established techniques from signal processing, we were able to configure the base station to predict the future movements of the mobile user. The prediction algorithm was hierarchical and was carried out through a combination of local prediction -- using an adaptive self-tuning Kalman filter and global prediction -- using approximate pattern matching techniques. With time, the predictive algorithm would improve its performance by learning the repetitive (regular) movement patterns of individual users building a per-user mobility profile database. Once this database was created, the prediction problem was reduced to a classical pattern matching problem. Connection quality was greatly enhanced with accurate prediction as the system was able to improve handoffs thus reducing the call blocking probability. Additionally, end-to-end delays were reduced since optimal routes from future base stations could be determined in advance and intelligent hot-spot traffic congestion relief was provided through line-of-path channel borrowing. --- Paper available --- (Research Associate, Advanced Communications Technologies Lab., Boston University) Research Collaborator - Prof. Andras Farago (Technical University of Budapest, Hungary) We developed a novel reservation-random TDM-based WMAC protocol called ARMAP. ARMAP preserves the desirable features of current second generation and proposed third generation voice-centric protocols while adding support for real-time VBR video communications. Specifically, ARMAP provides optimum resource utilization and optimum mobile power consumption while allowing wireless (video) connections to make timely contention-free reservation requests. Through intelligent admission control, resource reservation and scheduling, a minimum negotiated quality of service is guaranteed. Efficient usage of the reserved resources is ensured by a novel intra-frame statistical multiplexing technique. A novel bandwidth partitioning scheme (Partial Sharing with Restrictions) ensures fairness to different traffic classes. The resource allocation algorithm (The Smart Allocate Algorithm) ensures that the maximum blocking probability is minimized. --- Technical Report available --- (Research Associate, Advanced Communications Technologies Lab., University of Massachusetts, Amherst) In most narrowband wireless video communications, video frame rates tend to less than 30 fps. Statistical models describing the video packet arrival distribution do not take into accounts the different frame rates and are therefore not accurate. We undertook a study to consider the combined effects of the application type, the compression algorithm, and frame capture rate on the traffic generation process. In doing so, we developed novel evaluation techniques for studying the trueness of such models and used our results to tackle the problems of capacity planning and resource allocation in wireless networks. --- Paper available --- (Research Associate, Advanced Communications Technologies Lab., University of Massachusetts, Amherst) Research Collaborator - Dr. Wei-Lin Hsu (DIGITAL, MA) Developed a joint source-channel low-bit rate spatially segmented, multi-scale, wavelet-based subband video codec suitable for transmission over bandwidth-limited, error-prone radio channels. A novel error concealment strategy (called Lego Reconstruction, after the Lego building blocks) was shown to provide excellent results in terms of temporal resolution. --- 3 papers available --- (Research Assistant, Vision and Image Analysis Research Lab., SUNY Buffalo, NY) As part of a multi-year research project on computer vision systems we designed a novel algorithm for automatic recognition of handwritten characters. We did this by modeling the english language as a Hidden Markov Model, tackling the recognition of non-cursive, segmented script. A variety of well known techniques were employed including Palvadis's classical thinning algorithm (for normalization), feature vector extraction (15 features, statistical and heuristic, were selected), vector quantization, probability determination and specification (using cryptographical analysis), maximum likelihood estimation using Viterbi's algorithm and a dictionary search. --- Paper available --- (Research Assistant, Vision and Image Analysis Research Lab., SUNY Buffalo, NY) We implemented enhanced versions of classical image/signal processing techniques such as image segmentation, edge detection, medial axis transformation etc. to model the growth patterns of dendritic ice crystals. Ice crystals are important as they are used for understanding properties of certain important metals. Images taken at different time instances during the growth of the crystals were digitized and then processed by our novel iterative procedure that attempted to best-fit geometric shapes (ellipses, hyperbolas, parabolas, and circles) to the crystals. |
©2009 Microsoft Corporation. All rights reserved.