Elza Erkip - US grants

There is a tremendous market for tetherless and ubiquitous network
access, and a great demand for nomadic computation and
communication. A major component of such a system is a high speed
robust wireless local area network (WLAN) that provides services for a
limited geographical area. This limited geographical area enables a
large density of access points as well as high power tranmissions and
opens up the possibility of high rate robust multimedia communications
comparable to wired systems. Even though the bandwidth is still
scarce, the spatial dimension can now be exploited: The size of
typical WLAN devices, such as laptops and hand-held computers, enable
proper spacing of multiple antennas to provide diversity and coding
gains.

This research is aimed at a complete system design for multiple
antenna methods in the WLAN environment including coding, modulation,
channel estimation and power control strategies specifically tailored
for the slow fading environment of a WLAN channel. In particular,
high-rate multiple antenna coded modulation techniques and multiple
transmit antenna joint source and channel coding strategies are
developed. The issue of channel estimation along with feedback is
investigated and power control based on limited feedback is adressed.
A general approach to multiple transmit antenna methods is explored by
providing a theoretical foundation for spatially separated antennas in
WLAN environments.

The education plan is closely aligned with the research work and
emphasises the use of internet and computers in education by making
use of Polytechnic University's newly installed WLAN system on
campus. Curriculum development addresses specific needs of
Polytechnic's professional masters program and is based on an
integrated, end-to-end approach to communication systems.

EIA 0224387
Czarkowski, Dariusz
Erkip, Elza
Goodman, Davis J.
Karri, Ramesh.
Wang, Yao
Polytechnic University of New York

Title: CISE RR: Instrumentation for Research on Energy Aware Multimedia Information Terminals

This project, performing studies related to minimizing the energy consumed by mobile cellular and wireless networking hardware, builds a framework to understand, model and measure power consumption, as well as dynamic allocation of resources to affect power consumption in portable multimedia communication devices. Three platforms constitute the focus for implementing signal processing algorithms (general microprocessors, e.g., Pentium and ARM), digital signal processors (DSP), and reconfigurable hardwarefine grain field programmable gate arrays (FPGA) and coarse grain Application Specific Programmable Processors (ASPP). The work considers two communications systems that will carry the majority of traffic in wireless Internets: cellular networks and wireless local area networks (WLAN). Because battery life depends not only on total energy drain but also on the voltages employed in the terminal and the temporal profile of energy consumption, the research includes experiments with lithium batteries to determine the effects on battery life of continuous discharge and pulsed discharge with various peak-to-average ratios. The results of these experiments are then merged with the results of signal processing and communications studies in an energy-management testbed that verifies predictions and demonstrates the effects of the new adaptation techniques. Hence, the following four components form the research plan:
Theoretical and simulation studies exploring power allocation among source coding, channel coding, encryption, and transmission in a multimedia, multiuser setting
Measurement and design study examining the power and energy requirements of signal processing algorithms and producing a software library of software modules, FPGA configurations, and ASPPs that can be dynamically selected by a portable device
Measurement and design study creating an intelligent power supply system that takes into account effects of battery discharge characteristics and demands of algorithms implemented on microprocessors, DSPs, FPGAs, and ASPPs.
Creation of experimental testbed running in the WLAN network from the results of the first three studies
The experimental testbed addresses three areas of research:
Power allocation among different components of the processing pipeline (source coding, channel coding, encryption, transmission);
Power and energy requirements of signal processing algorithms, producing a library of various granularity modules (software, FPGA, ASPP) to be selected dynamically; and
Power supply system that takes into account both battery characteristics and algorithms needs.
On the educational side, specific plans involve undergraduate students hands-on work, organization of workshop, and summer school. Industrial partners and international collaboration also form part of the research.

The large number and diversity of quality of service (QoS) measures present a significant challenge to the design and operation of multimedia wireless communications systems. Each information source (data, audio, video) has its own metrics of signal quality and the network has an aggregate measure of capacity for each source (for example, Erlangs of telephone traffic and data throughput). In addition to signal quality and network capacity, the battery life of a portable terminal has a major effect on the value of mobile information services to consumers. Existing knowledge of quality optimization is generally confined to studies of individual sources. In the 1990s, a significant body of knowledge was created on radio resource management for cellular telephone communications. More recently, the research community has turned its attention to wireless data and video. Each study focuses on one or two quality measures for one type of information: for example, power and distortion for video; power and throughput for data.

This project takes a more comprehensive view by considering the collection of QoS measures to be a point in a multidimensional space. Given a system design and a set of operating conditions, the achievable points constitute a feasibility volume, with optimum points on the surface. Within this formulation, we study simultaneous transmission of data and video by analyzing projections of the volume onto various combinations of QoS dimensions including:
video distortion;
data throughput for each source;
data utility for each source;
total power dissipation (signal processing power and transmission power) in a terminal;
number of simultaneous video transmissions at a base station;
aggregate base station throughput;
aggregate base station utility.

The emphasis is on power efficient communications and the results provide guidance on joint adaptation of the following properties of terminals that transmit signals to the same base station: transmission power and rate in data terminals and transmission power, compression, and channel coding in video terminals.

Cooperative Source and Channel Coding



Current and next generations of wireless devices and services are substantially different than the original cellular phones which could only carry voice signals. Third/fourth generation cellular and wireless local area networks are designed to support data services, image and video communications as well as voice. Multimedia signals require higher data rates and larger bandwidths than their voice counterparts. This necessitates a more efficient use of already scarce radio resources. Furthermore, guaranteeing a desired level of signal quality for image and video, measured in terms of overall distortion, is especially difficult given that the wireless channel is unreliable and most efficient compression algorithms involve error propagation.

In order to provide robust wireless multimedia communications, this research uses cooperative communication techniques along with jointly optimized source compression and channel coding strategies. Cooperation of wireless terminals is achieved by overhearing other terminal's signals and retransmitting towards the desired destination. This provides signal diversity and enables robust source-to-destination routes which can adapt to changes in the wireless environment. In order to establish the theory and practice of cooperative source and channel coding, the research plan consists of three interrelated components: Information theory of source channel cooperation; design of cooperative source and channel coding techniques with numerical/simulation studies to jointly optimize the parameters; and application of these techniques to wireless video transmission.

This project investigates the concept of cooperation between nodes in a wireless network to increase its traffic capacity and reduce radio interference. Currently, transmissions between two nodes overheard by a third node are discarded. It has been shown that this overheard information can be used by the third "helper", through additional transmissions, to increase the capacity of the network. Alternatively, the cooperating nodes can also reduce the power needed to transmit the same amount of information as a network without cooperation. The notion of cooperation will be investigated both for the actual bits being transmitted, as well as in the medium access and routing protocols that facilitate and exploit it. The research will consider ad hoc networks in addition to infrastructure-based networks.

This work shows how to increase the capacity of wireless networks, and allow for a better spatial reuse of a limited amount of bandwidth by reducing the interference to other nodes in the vicinity. Analysis, simulations and proof of concept implementations are used in this research. Broader impacts, besides the research itself, include the positive outcomes of joint work with industry partners, and involving a wide group of students in research through the implementation-related projects.

Proposal #: CNS 07-08989
PI(s): Erkip, Elza
Bertoni, Henry L.; Panwar, Shivendra S.; Wang, Yao
Institution: Polytechnic University of New York
Brooklyn, NY 11201-3840
Title: IAD: Cooperative Networking Testbed Amount Rec: $ 300,000

Project Proposed:

This project, building two complementary testbeds for cooperative networking, responds to the fading and multipath distortion, as well as interference caused by multiple users operating over a limited bandwidth, often suffered by wireless communication systems. Respectively, each testbeds will
. Use open source drivers for backward compatibility with the current WiFi technology based on the IEEE 802.11 standard. This approach uses a standard, non-cooperative physical layer since it is not possible to access the physical layer in commercial products.
. Be based on software defined radio, allowing maximum flexibility in the implementation of a cooperative physical layer, a cooperative medium access control (MAC) layer as well as cross-layer design.
Cooperative networking, where two or more active users in the network share their resources to jointly transmit their messages, provides resistance to fading, high throughput/low delay and reduced interference/low transmitted power. This infrastructure, leveraged in part by the WARP platform at Rice U, the ORBIT testbed at Rutgers U, and the CRAWDAD database at Dartmouth U, provides an open-access platform that will be used to build experimental deployable and scalable cooperative wireless networks, enabling current techniques to be moved beyond the current theoretical and simulation studies.
The testbeds validate the feasibility of cooperative networking, enable platforms where new algorithms can be tested, and lead to new theory founded on more realistic assumptions.

Broader Impacts: This first effort in implementing a fully cooperative network is expected to accelerate commercial developments in the field and impact current wireless standards. It facilitates a closer relationship with industry. Moreover, the infrastructure contributes to train students and service new courses.

Wireless network design poses several challenges that do not exist in wired networks. The wireless medium suffers from random time variations, the broadcast nature of wireless signals causes interference, the wireless spectrum is limited and mobile terminals have small batteries. These features become especially limiting for applications such as multimedia over wireless and low power sensor networks, where maintaining the end-to-end signal quality given the specific system resources is difficult.

This research outlines a cross-layer approach between the application layer and the physical layer to address these problems. The objective is to design joint source and channel coding techniques to minimize the end-to-end source distortion. A general source and channel separation theorem for wireless networks does not exist; optimality of Shannon's separate source and channel code design fails for non-ergodic fading channels or for multiuser communication systems. On the other hand, even when source and channel separation is not optimal, it is desirable to have only a loose coupling between the source and channel coders to simplify the designs.

The research addresses joint source and channel coding for the fundamental building blocks of a wireless network, namely single user (point-to-point), multiple access, broadcast and interference channels. The system model is general to encompass different communication scenarios: Correlation among the source signals is allowed, the receivers may have correlated side information, the channel can be time-invariant or fading may be present, links can have multiple degrees of freedom such as multiple antennas or multiple fading blocks. This project investigates the design of optimal joint source and channel coding strategies, and performance improvements when minimal interaction among the source and channel coders is allowed. The goal is to discover scenarios under which separation is optimal, or close to optimal. The information theoretic results will be complemented by analytical and experimental studies for video transmission over wireless channels.

Proposal #: CNS 07-22868
PI(s): Erkip, Elza
Panwar, Shivendra S.; Wang, Yao
Institution: Polytechnic University of New York
Brooklyn, NY 11201-3840
Title: MRI/Acq.: Experimental Platform for Wireless Multimedia Networking

Project Proposed:
This project, acquiring an experimental platform to support integrated research and educational activities on wireless multimedia networking, examines problems of limited bandwidth of the wireless channel, of
interference caused from multiple users operating in the same band, of rapid variations due to signal fading, of limited battery life of wireless devices, and of speed and reliability of wireless networking. The instrumentation consists of multiple radios based on software defined radio platforms, wireless nodes and open source drivers based on the IEEE 802.11 Wireless Local Area Network (WLAN) standard, DSP platforms enabling real time video encoding and wireless transmission, dynamic power scaling, and power measurement, and test equipment. The instrument supports research on:
. Cooperative wireless networking, Wireless video transmission, Energy efficient networking, and
. Integration of research and education through the Wireless Information Systems Lab.
The platform contributes to advance the state-of-the-art in multimedia wireless communications, not only enabling testing of new and existing algorithms, but also leading new theory founded on more realistic assumptions.
Broader Impacts: The instrument impacts new wireless technologies, serves to integrate research and education through the Wireless Information Systems Lab (WISL), enables the training of undergraduate and graduate students in wireless communication, and development of new courses for the curriculum. It contributes to the first efforts of implementing a fully cooperative network, whose benefits have been well established through theory and simulations. The experimental platform is expected to accelerate commercial developments in the field and impact current wireless standards. The impact on industry will be facilitated by the close relationship of the institution with Wireless Internet Center for Advanced Technology (WICAT) member companies.

Cooperative networking exploits the broadcast nature of the wireless channel by effectively pooling the overheard information, which is traditionally treated as harmful interference. While there is a mature suite of tools at the physical (PHY) layer to harvest cooperative gains, it is still unclear how these tools can be employed to deliver significant network capacity gains. The goal of this project is to design and implement cross-layer mechanisms for cooperative networking. By integrating PHY layer cooperation with Medium Access Control (MAC) and application layers, the project will provide higher network capacity and improved multimedia quality.

The project has two interrelated components investigating basic architectures for next generation cooperative networks: (i) Cooperative data transmission, which focuses on a robust cooperative MAC-PHY incorporating multiple relays under mobility and loose requirements on synchronization and network topology. (ii) Cooperative video transmission, which exploits the synergy between cooperation and layered compression in providing unequal error protection, as well as differential quality in multicast.

Apart from potential impacts on the theory and practice of new wireless technologies, this project will train undergraduate and graduate students in all aspects of wireless communications. The impact on industry will be facilitated by the close relationship of NYU-Poly with WICAT member companies.


This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).

This proposal seeks funding for the Center for Wireless Internet Center for Advanced Technology (WICAT) studies conducted by the Polytechnic Institute of New York site (lead), the University of Virginia site and the site at Auburn University. Funding Requests for Fundamental Research are authorized by an NSF approved solicitation, NSF 10-507. The solicitation invites I/UCRCs to submit proposals for support of industry-defined fundamental research.

This proposal is a comprehensive research project for significantly improving the capacity of wireless networks. In past decades, there has been an exponential growth of wireless devices and wireless networks. While wireless networks have brought us the convenience of mobility and new applications, they are limited by bandwidth bottlenecks. The industry and spectrum regulators are trying to allocate more bandwidth, but they still fall behind the bandwidth increases in wireless networks. Thus it is critically important to use spectrum resources more efficiently. The proposed work aims to decrease interference, including cooperative and cognitive networking. This research will not only make significant contributions to the research community, but also be very valuable for the wireless industry and spectrum regulators.

This proposal aims to eliminate wireless system bottlenecks using cooperative and cognitive technologies, with the potential of enabling a broad spectrum of new wireless applications. The PIs will exploit existing programs, such as NSF REUs and the collaboration with Tuskegee and Northern New Mexico College, which focus on including students from underrepresented groups in the research. The student body at NYU-Poly is quite diverse. Most of the requested funding is going toward student support. The Pisa lo intend to incorporate this work into the new graduate course at NYU-Poly and the new ABET-accredited program at Auburn. Dissemination is clearly outlined: reports to NSF, reports to their IAB, journal and conference papers and making the resulting software packages open source.

The International Symposium on Information Theory (ISIT) is the premier annual information theory conference worldwide, and thus a valuable forum for graduate students in this discipline to present their work, receive feedback from other researchers working on similar or related topics, and to make professional acquaintances within the discipline. This award provides international travel funds to be awarded to students who have a paper accepted at this year's ISIT and who otherwise would not have sufficient travel funds to attend the symposium. As information theory is a core component of the Communications and Information Foundations cluster, the efforts to ensure participation of graduate students in this symposium are consistent with NSF's goal of promoting education and research of the next generation of scientists.

This project considers cellular systems based in the millimeter wave (mmWave) bands, between 30 and 300 GHz, where the available bandwidths are much wider than today?s cellular networks [4-6]. Indeed, available spectrum at these frequencies can be easily 200 times greater than all cellular allocations today under 3 GHz. Moreover,the very small wavelengths of mmWave signals combined with advances in low-power CMOS RF
circuits enable large numbers of miniaturized antennas to be placed in small dimensions. These multiple antenna systems can be used to form very high gain, electrically steerable arrays, fabricated at the base station, in the skin of a cellphone, or even within a chip. Due to the limited range of mmWave signals, the proposal envisions cellular systems based on large numbers of mmWave ?picocells? (100 to 200m radius), each using highly directional antennas for improved range and spatial separation. Combining dramatically increased bandwidths with spatial multiplexing gains from the high-dimensional multiple antenna transmissions such mmWave picocellular systems offer the possibility of of 1000 times more capacity than current commercial networks.

This project will provide a basis for bringing mmWave technologies to a multiuser, multi-cellular setting and enable mmWave systems for wide-area networks with mobility. With partnerships from leading device vendors in this space, the PIs aim to drive the innovation forward in terms of device and protocol development, and train a new generation of students at the graduate and undergraduate level in this emerging area of wireless communications. The enhanced throughput gains that could approach 1000x current wireless cellular network throughputs would revolutionize the wireless broadband industry and lead to improved delivery of network services to users.

Traditional radio transceivers are generally not able to receive and transmit on the same frequency band because of the crosstalk between the transmitter and the receiver circuits. Given that the received signal over the air is one million times or more weaker than the transmitted signal, it is very difficult, if not impossible, to detect the received signal under internal interference from the transmitter. Thus, to avoid this interference, today's communication systems rely on half duplex transmission, typically transmitting and receiving at different times or in different frequency bands. The full duplex technology developed in this project allows for transceivers to receive and transmit in the same frequency band, at the same time, potentially doubling data rates. Given the enormous market size of the wireless industry, combined with the pressing need for solutions to solve the 'spectrum crunch', full duplex technology thus offers tremendous societal and commercial impact.

This project addresses the fundamental challenges when incorporating full duplex radios in a cellular network to unlock the full potential of this technology. Specifically, the project consists of three inter-related components: (i) a radio front-end design that allows multiple antenna gains in conjunction with full duplex operation; (ii) a smart scheduler for full duplex base stations that coordinates the maximum set of transmissions with least mutual interference, while maintaining fairness among the mobiles; (iii) a joint routing and scheduling algorithm that fully utilizes full duplex enabled relay terminals. The technologies developed span physical, medium access control, and network layers, and jointly coordinate the air-interface and the backhaul. The theory and algorithms will be validated by over-the-air and end-to-end experiments.

With the severe spectrum shortage in conventional cellular bands, millimeter wave (mmWave) frequencies between 30 and 300 GHz have been attracting growing attention as a possible candidate for next-generation micro- and pico-cellular wireless networks. The mmWave bands offer orders of magnitude greater spectrum than current cellular allocations and enable very high-dimensional antenna arrays for further gains via beam-forming and spatial multiplexing. However, due to unique nature of propagation in these frequencies, cellular systems will need to be significantly redesigned to fully exploit the potential of these frequencies. This project investigates one key dimension in mmWave network design: resource sharing, involving the sharing of spectrum and network infrastructure resources by multiple operators. It is argued that sharing of this form will be essential to fully exploit the tremendous bandwidth and antenna degrees of freedom offered by these bands and also provide statistical multiplexing to accommodate the highly variable nature of traffic.

This inter-disciplinary project explores both the engineering and economic aspects of mmWave resource sharing. On the engineering side, the key challenge is that the spectrum must be shared not only in time and frequency, but also space. In addition, new decentralized mechanisms are required to rapidly reallocate backhaul and core network elements across multiple parties. The project develops signal processing methods, protocols and network optimization algorithms to facilitate directional sharing of spectrum along with dynamic re-routing and load balancing. On the economic and policy side, several shared ownership models under varying forms of licensing agreements between the operators, regulatory agencies and third-party spectrum and resource brokers are investigated. Based on the capacity estimates obtained from the engineering side, economic models are developed to assess key indicators such as operating and capital costs and coalition formation and market concentration.

Providing services such as voice, web, cloud computing, social networking, video on demand, live streaming and augmented reality requires the ability to globally handle massive amounts of data, which must be efficiently processed, stored, and delivered to end users. The medium of delivery is shifting from wireline to wireless; it is predicted that mobile traffic will account for nearly two-thirds of the total data traffic by 2018 and nearly three fourths of the mobile data traffic will be video by 2019. This puts a tremendous pressure on the limited wireless bandwidth and necessitates revolutionary approaches that exploit alternative resources available in the network. Motivated by the decreasing cost and abundance of storage capacity, this project considers joint design of storage and transmission schemes for the efficient delivery of video-based services over next generation heterogeneous wireless networks, by specifically taking into account the unique nature of video content and the wireless channel.

In order to accomplish this ambitious task, the work is organized into the three research thrusts that develop next generation of wireless caching networks for video delivery:

-Thrust 1: Fundamental limits and practical schemes for cache-aided video delivery:
This thrust takes into account properties of video applications and their associated requirements for efficient and robust caching and delivery techniques particularly suited for such demands.

-Thrust 2: Video delivery over future cache-aided heterogeneous wireless networks:
This thrust builds upon the foundations developed in Thrust 1 and expands the cache-aided video delivery schemes to consider and take advantage of heterogeneous wireless channel conditions and network topologies.

- Thrust 3: Technology validation and experimentation: This thrust includes proof-of-concept prototyping efforts to validate the designs developed in this project.

The broader impacts resulting from the activity in this project will include significant enhancement of video delivery mechanisms over wireless channels beyond the current state of the art. This of key importance since the total global mobile traffic was about 885 petabytes per month at the end of 2012, and is expected to keep increasing. As the storage capacity becomes cheaper, the techniques developed in this project will be able to alleviate the spectrum crunch by trading off memory for bandwidth. Industry outreach and dissemination will be done through Bell Labs Nokia and other industrial partners at NYU. The PIs are committed to and have an excellent track record in increasing participation of as well as mentoring women and underrepresented minorities in the STEM fields. This grant will help to support these efforts.

The millimeter wave (mmWave) frequencies and other bands above 6 GHz are a new and promising frontier for cellular wireless communications and the focus of the fifth-generation (5G) standardization efforts. Due to the massive available bandwidths, the mmWave frequencies offer the possibility of orders of magnitude greater capacity than current cellular systems in the highly congested bands below 3 GHz. However, a key challenge in realizing mmWave mobile cellular networks is energy consumption. Mobile devices and small-cell access points must operate under highly-constrained power budgets, and developing mmWave systems within these power limits is a formidable task. This project will investigate energy consumption from a system's perspective to address the issues in an integrated and coherent manner. The project will develop mathematical models for understanding energy and performance tradeoffs. New technologies for the radio frequency (RF) circuits, antenna array system, signal processing, and network protocols will be jointly developed to optimize performance and deliver energy-efficient mmWave devices.

The research work will pursue four key thrusts: Thrust 1 will seek to understand the fundamental relation between throughput, processing power, and spectral emission constraints, and optimizes this tradeoff by dynamically controlling key energy drivers including bandwidth, quantization resolution, antenna architecture, and waveform selection. This is combined with novel energy-aware scheduling policies and signal processing techniques. Thrust 2 will focus on idle mode power savings and associated problems of energy-efficient directional channel estimation and delay in dynamic environments. Thrust 3 will develop fundamental circuits technologies that enable Thrusts 1 and 2, in particular the novel techniques for both high-efficiency power amplifiers and low-power fully digital RF transceivers. Thrust 4 will obtain the necessary channel models and build circuit prototypes to validate the concepts in Thrusts 1 to 3.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

As the autonomy of robots performing manipulation or legged locomotion tasks increases, they require an ever-growing amount of computational resources to successfully perceive their environment and make decisions. Yet, mobile robots are fundamentally constrained by weight, shape and power autonomy which impose important limits on the computational capabilities they can carry. As a possible answer to such dilemma, cloud robotics aims to move computation to remote servers, but it has thus far remained an elusive approach for tasks that necessitate low delay communication and high data bandwidth between the robot and the cloud. Such tasks include control, planning and perception algorithms for object manipulation and legged locomotion. The 5th generation of cellular network technology (5G) could revolutionize cloud robotics as it promises unprecedented access to high bandwidth and low latency wireless communication. Yet, formidable challenges remain to ensure communication reliability, safety of robotic operation under communication degradation, and scalability to multi-robot systems. This project aims to fully incorporate 5G technology into robotics systems performing complex manipulation and locomotion tasks. It will develop novel perception, control and planning algorithms that optimally distribute computations between robots and the cloud for guaranteed safe robotic operation. Ultimately, these algorithms will accelerate the ubiquitous deployment of untethered 5G-enabled robots in human environments and unlock a large range of applications for healthcare, service and industrial robotics.

The project takes a holistic approach to control, perception and communication to establish the foundations of edge-based wireless real-time action-perception loops for autonomous robots. It is organized along four main thrusts of research. First, it will investigate novel optimal control and planning algorithms distributed between the network edge and the robot with performance guarantees under communication degradation. Second, it will propose efficient computational partitioning techniques for real-time perception using multi-modal sensing with high data rates. Third, it will characterize 5G specific communication channels in a robotics environment via experiments and simulations and design new mmWave communication protocols tailored for real-time robotic action-perception loops. Finally, extensive experiments on single and multi-robot systems, including fixed and mobile manipulators and a quadruped robot, will demonstrate the unique capabilities of 5G-enabled robotic systems. The outreach activities of the project will contribute to lowering barriers to entry for scientists and industries that seek to exploit 5G-enabled robotics through open-source distribution of algorithms and dissemination of results via NYU WIRELESS. This effort will contribute to the education of undergraduate and graduate students, leveraging its outcomes for curriculum development and offering supervised projects with the possibility to work directly on state-of-the-art experimental platforms.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Increasing amounts of data are being collected on mobile telephones and internet-of-things (IoT) devices. Users are interested in analyzing this data to extract actionable information, for example, identifying objects of interest from high-resolution mobile phone pictures. The state-of-the-art technique for such data analysis is via deep learning which makes use of sophisticated software algorithms modeled on the functioning of the human brain. Deep learning algorithms are, however, too complex to run on small, battery constrained mobile devices. The alternative, i.e., transmitting data to the mobile base station where the deep learning algorithm can be executed on a powerful server, consumes too much bandwidth. This project seeks to devise new methods to compress data before transmission, thus reducing bandwidth costs while still allowing for the data to be analyzed at the base station. Departing from existing data compression methods optimized for reproducing the original images, the project will use deep learning itself to compress the data in a fashion that only keeps the critical parts of data necessary for subsequent analysis. The resulting deep learning based compression algorithms will be simple enough to run on mobile devices while drastically reducing the amount of data that needs to be transmitted to mobile base stations for analysis, without significantly compromising the analysis performance. The proposed research will provide greater capability and functionality to mobile device users, enable extended battery lifetimes, and more efficient sharing of the wireless spectrum for analytics tasks. The project also envisions a multi-pronged effort aimed at outreach to communities of interest, educating and training the next generation of machine learning and wireless professionals at the K-12, undergraduate and graduate levels, and broadening participation of under-represented minority groups.

The project seeks to learn ?analytics-aware? compression schemes from data by training low-complexity compressor deep neural networks (DNNs) that execute on mobile devices and achieve a range of transmission rate and analytics accuracy targets. As a first step, efficient DNN pruning techniques will be developed to minimize the DNN complexity, while maintaining the rate-accuracy efficiency for one or a collection of analytics tasks. Next, to efficiently adapt to varying wireless channel conditions, the project will seek to design adaptive DNN architectures that can operate at variable transmission rates and computational complexities. For instance, when the wireless channel quality drops, the proposed compression scheme will be able to quickly reduce transmission rate in response while ensuring the same analytics accuracy, but at the cost of greater computational power on the mobile device. Further, wireless channel allocation and scheduling policies that leverage the proposed adaptive DNN architectures will be developed to optimize the overall analytics accuracy at the server. The benefits of the proposed approach in terms of total battery life savings for the mobile device will be demonstrated using detailed simulation studies of various wireless protocols including those used for LTE (Long Term Evolution) and mmWave channels.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.