The 802.11 standard, known also as WiFi, was designed for wireless local area networks (WLAN) for homes and enterprises. The huge success of WiFi in these markets and the economy of scale it created, and its further successful penetration into mobile devices such as notebooks, PDAs, phones, etc, created an opportunity to use it for broadband wide-area coverage in outdoors environments. This opportunity is further driven by the fact that the WiFi spectrum is unlicensed and therefore free for use in most parts of the world.

Yet, the requirements and the challenges of outdoors WiFi are very different from the indoors WiFi of the homes and enterprises. The ranges required to be covered are larger, the nature of the multipath environment it faces is different, and more critically, the level of interference it needs to cope with is significantly larger.

This talk will present a WiFi multiple-antenna base station based on Beamforming and SDMA technologies that copes effectively with these challenges. The Beamforming and SDMA technologies are completely transparent to the clients and are therefore applicable to all existing off-the-shelf WiFi clients. The challenges in applying these technologies to the WiFi protocol will be described, as well as the benefits and advantages they provide in terms of performance and deployment cost.

Direction-of-arrival estimation using an array of sensors has been an active research area for several decades. Model-based approaches started to appear in the 60's, and became popular with the introduction of MUSIC and other subspace-based methods in the late 70's. In contrast to conventional beamforming based approaches, these techniques are sensitive to errors in the assumed reception model of the array. In the signal processing literature, Very simplistic models based on geometry only are usually applied. In contrast, real sensor arrays involve many more physical phenomena, such as mutual coupling between sensors, mismatch, and other imperfections in the receiver hardware. The exact sensor locations and orientations may also be uncertain under real experimental conditions. In practice, these effects must be taken into account by using more accurate physical models and/or by using calibration measurements.

The purpose of this talk is to give some background into sensor models, focusing on electromagnetic antenna arrays, and to give an overview of software calibration techniques. Some more recent results on error interpolation in one and more dimension will also be presented. We also look at the effect of modeling errors and calibration on the direction estimation performance, and how this can be predicted from data.

Accurate time or clock synchronization is critical in applications such as range finding for target tracking and localization, intrusion detection, time correlation of telemetry data, sensor fusion, slot assignment in TDMA, duty cycling protocols, and so on. We review the synchronization problem, focusing on key issues such as estimation of clock offset and skew, and performance bounds on these estimates. We next consider the distributed synchronization problem in a consensus type setting and consider the impact of asymmetry and asynchronous updates due to link failures and node failures. We will consider the impact of network topology on convergence of these estimators, and will provide comparisons with centralized estimators.

Optimization of communication networks has recently witnessed an impressive growth of research activities. In addition to viewing networks as objects to be optimized, some of these works also view networks as optimizers themselves. In addition to "Design by Optimization", some recent results also demonstrate the principle of "Design for Optimizability". Indeed, more than a tool to solve for optimal resource allocation, optimization theory provides to networking applications all of the following: a modeling language for design, a reverse-engineering methodology for analysis, a theoretical foundation for architectural decisions, a quantitative basis for fairness and robustness, and even an indicator of flaws in engineering assumptions. Many of these new uses of optimization actually do not involve solving any problem optimally. Reflecting upon the history of optimization-based solutions to congestion, collision, and interference in the last 15 years, this talk discusses the reach and limitation of network optimization. Then, drawing from recent results on open problems in stochastic utility maximization and Internet routing, this talk surveys the emerging trends that give many new meanings to the phrase "Optimization of Networks and by Networks".

Designing sensor networks with the capability of taking decisions autonomously and in a decentralized fashion is a problem that has received considerable attention in the last years. The distributed approach is well motivated from fundamental information theoretic results as well as from practical considerations concerning the network vulnerability to node failures and congestions around the sink nodes. However, the inherent iterative nature of distributed algorithms makes them prone to an energy consumption that depends on the algorithms convergence time and on the transmit power of each node. Furthermore, the interaction among sensors in a realistic environment is inevitably corrupted by a communication noise that affects the final decision. Finally, the iterated interaction raises a complexity issue in the implementation of the medium access control mechanism. In this presentation, we start showing how a globally optimal operation like projection of the whole set of data gathered by the network onto the useful signal subspace can be implemented in a totally distributed way, under a constraint on the coverage radius of each node dictated by the transmit power. In particular, we show how to choose the iterative algorithm parameters in order to minimize the overall energy necessary to achieve the desired decision with the required accuracy. We evaluate the effect of communication noise on the interaction mechanism and propose ways to keep the noise variance bounded. Finally, we concentrate on a particular example of projection represented by consensus algorithms and we show how, building on the analogy between consensus algorithms and diffusion processes, the speed of the consensus algorithms can be increased by emulating a stirring mechanism inspired by fast fluid mixing techniques.

DSL networks have played a key role over the last decade in providing broadband connectivity and expanding the public's access to internet resources. Signal processing techniques have been instrumental in enabling high speed transmission over copper and meeting the growing bandwidth demand. A new bandwidth demand step is expected in the next five years as IP networks are taking over the transport of voice and entertainment signals. The next challenge for DSL is whether signal processing and physical layer technologies can further advance to address this new demand, or whether copper networks will simply not be able to support higher rates and will reach the end of their usefulness. In this talk, current performance bottlenecks will be discussed and promising new techniques will be reviewed. The problem of crosstalk mitigation will be discussed in further detail, which presents one of the major obstacles in reaching higher performance. This problem has spurred an active area of research involving the application of MIMO techniques, vectored transmission and dynamic spectral management to the DSL environment. A summary of the state of the art in this research area in industry and academia will be presented.

We consider a very general class of wide-sense stationary uncorrelated scattering (WSSUS) multiple-input multiple-output channels that allow for selectivity in time and frequency as well as for spatial correlation. Besides the underspread assumption (i.e., the product of the channel's delay spread and Doppler spread is much smaller than 1), we make no further simplifying assumptions. Virtually all channels in wireless communications are (highly) underspread. Starting from a continuous-time channel description, and exploiting the underspread property to obtain a suitable discretization, we derive (tight) bounds on noncoherent channel capacity under a peak constraint on the transmit signal. The bounds are explicit in the channel's scattering function, are useful for a large range of bandwidths, and allow to coarsely identify the capacity-optimal combination of bandwidth and number of transmit antennas. Furthermore, we obtain a closed-form expression for the first-order Taylor series expansion of capacity in the limit of infinite bandwidth. From this result, we conclude that in the wideband regime: (i) it is optimal to use only one transmit antenna when the channel is spatially uncorrelated; (ii) rank-one statistical beamforming is optimal if the channel is spatially correlated; and (iii) spatial correlation, be it at the transmitter, the receiver, or both, is beneficial. The results in this talk are based on the theory of underspread time-varying stochastic systems, the relation between mutual information and minimum mean-square error discovered recently by Guo, Shamai and Verdu, various flavors of Szeg's theorem for two-level Toeplitz matrices, and martingale theory.

The new discoveries of physics during the 20th century have revolutionized our understanding of the basic structure of the world. Whereas new physics models are developed, more and more questions remain unsolved. In order to answer some of the questions, new accelerators and experiments are being designed and built. Although statistical data analysis techniques are routinely employed in high energy physics (HEP), statistical signal processing (SSP) methods and experts are rarely involved in these experiments. This scenario is changing as faster tools become available and as high energy physicists discover the benefits of SSP. Data processing in HEP experiments is a multi-tiered process in which raw detector signals are first processed locally into physics objects, and then collated into event records which can be scrutinized by a fast online trigger system. The resulting selection of events passes through a number of software filters before the final offline analysis, in which hard physical constants are extracted. For both trigger and offline applications, a great number of challenges for SSP are possible. Some are already beginning to establish, such as the use of track reconstruction techniques. At this talk, an overview of the new experiments processing challenges will be given, and some of the popular track reconstruction methods will be reviewed. Then, novel techniques for track reconstruction based on statistical signal processing will be introduced. These techniques are examples of the potential benefits of using statistical signal processing for HEP.

Time delay estimation (TDE) has been studied extensively for several decades, with most of the attention focused on signals received over a single channel. In this talk, we study the problem of estimating a time delay (TD) parameter based on processing signals received on multiple, parallel channels. An example of TDE on parallel channels is a waveform containing multiple frequency subbands that is received through a frequency-selective channel, or a frequency-hopping waveform in which multiple hops are processed to estimate the TD. The channels may be modeled as deterministic (known or unknown) or randomly fading, and the objective is to jointly process the signals on each channel to estimate the (common) TD parameter. If the parallel channels include random fading, then interesting questions arise such as the effect of diversity gain on TDE accuracy and tradeoffs between signal energy per channel and the number of channels. This talk begins with a general model for TDE on parallel channels with several models for the channel gain, including deterministic (known and unknown) and random (Rayleigh fading). Each channel model case is analyzed to obtain the maximum-likelihood estimator (MLE), Cramer-Rao bound (CRB), and Ziv-Zakai bound (ZZB) for the TD parameter. The bounds facilitate an analysis of the effects of fading and diversity on TDE accuracy over parallel channels. Computer simulations of the mean-squared error of the MLEs confirm the validity of the bounds.