Communications Research Colloquium and Informal Seminar

School of Electrical and Computer Engineering -- Cornell University

Spring 2002 -- Meetings: Wednesdays, 11am, 310 Rhodes (coffee and cookies available before)
If you would like to give a talk, please get in touch with Sergio Servetto.

Date Speaker Title
Wed 1/23 Sergio Servetto Quantization with Side Information: Lattice Codes, Asymptotics, and Applications in Wireless Networks
Wed 1/30 Gokhan Mergen Capacity of Regular Networks with Multipacket Reception
Wed 2/06 Anna Scaglione Random Matrices and the Statistics of the Capacity of MIMO Frequency Selective Fading Channels with Application to Space-Time Coding and Multiple-Access Channels
Wed 2/13 Clint Kelly Network on a Chip: Modeling Wireless Networks with Asynchronous VLSI
Wed 2/20 Edwin C. Kan Hardware Considerations for Extremely Low Power (uW) and Low Bit Rate (100b/s) Submillimeter Autonomous Microsystems
Wed 2/27 empty No seminar
Wed 3/06 Jaiganesh Balakrishnan Time-Reversal in Decision Feedback Equalization
Wed 3/13 empty No seminar
Wed 3/20 Spring break No seminar
Wed 3/27 Toby Berger Information in Living Systems
Wed 4/03 Allen MacKenzie Games of Population Transition and Medium Access Control
Wed 4/10 Zygmunt Haas CANCELLED
Wed 4/17 Wen Xu Spectrally Efficient Partitioning of Video Streams for Robust Transmission Over Multiple Channels
Wed 4/24 Richard Martin A Blind, Adaptive TEQ for Multicarrier Systems
Wed 5/01 Sergio Servetto Queues under Feedback Control


Speaker:
Sergio Servetto.

Title:
Quantization with Side Information: Lattice Codes, Asymptotics, and Applications in Wireless Networks.

Abstract:
We consider the problem of rate/distortion with side information available only at the decoder. For the case of jointly-Gaussian source $X$ and side information $Y$, and mean-squared error distortion, Wyner proved in 1976 that the rate/distortion function for this problem is identical to the conditional rate/distortion function $R_{X|Y}$, assuming the side information $Y$ is available at the encoder. In this talk we present a construction of asymptotically optimal lattice quantizers for this problem: under the assumption of high correlation between source $X$ and side information $Y$, and for any finite rate, we prove there exist quantizers within our class whose performance comes arbitrarily close to Wyner's bound. We also explore the use of these quantizers in the context of a problem of data compression for wireless networks, which involves a large number of devices collecting highly correlated measurements within a confined area. An important feature of our formulation is that, although the per-node throughput of the network tends to zero as the network size increases, so does the amount of information generated by each transmitter. This is a situation likely to be encountered often in practice, which allows us to cast under new---and more "optimistic"---light recent results on the transport capacity of large-scale wireless networks.

Speaker:
Gokhan Mergen.

Title:
Capacity of Regular Networks with Multipacket Reception.

Abstract:
In the previous work on the capacity of wireless networks, source burtiness and advanced physical layer options are not considered. Moreover, the effect of distributed random access on network capacity is not well understood.

In this talk, I will talk about our recent work (jointly with Prof. Tong) on the capacity of wireless networks. We will look at the effects of multipacket reception (MPR) capability on the capacity of regular wireless networks. We have obtained the maximum stable packet generation rate for the MPR Manhattan networks under uniform traffic and minimum connectivity. For a network of size N, it is shown that the capacity is K_1 / sqrt{N}, and adding MPR affects the coefficient K_1 of a non-MPR network by no more than 1.6 times. For the same network the stability region of the slotted ALOHA random access protocol with MPR is shown to be K_2 / sqrt{N}, for some constant K_2 < K_1.

For increased connectivity, it is shown that the capacity can be increased further by adding MPR. In the limiting case, in a fully connected network, MPR increases the capacity linearly. These results indicate that minimum connectivity for MPR networks is not necessarily optimal which is in constrast to the results obtained by Gupta and Kumar for the conventional collision channel.

For more information, download our paper.


Speaker:
Anna Scaglione.

Title:
Random Matrices and the Statistics of the Capacity of MIMO Frequency Selective Fading Channels with Application to Space-Time Coding and Multiple-Access Channels.

Abstract:
The classic problem of maximizing the information rate over parallel Gaussian independent sub-channels with a limit on the total power leads to the elegant closed form water-filling solution, which has inspired practical systems such as DMT. In the case of MIMO frequency selective fading Space-Time DMT solution requires the derivation of the eigenvalue decomposition of channel matrices for every frequency bin which have a generalized Wishart distribution. This paper present the general methodology used to derive the statistics of eigenvalues and eigenvectors and derive the closed form expressions of the probability density function (p.d.f) of the channel Capacity for the case of uncorrelated Rayleigh fading and a finite number of transmit and receive antennas. Other applications of Random Matrices in communication systems will also be presented.

Speaker:
Clint Kelly.

Title:
Network on a Chip: Modeling Wireless Networks with Asynchronous VLSI.

Abstract:
We present the design of a network-on-a-chip: a flexible hardware network simulator tailored for wireless network simulation and implemented using asynchronous VLSI techniques. We present the architecture of the simulator and its components and show how it can be programmed to simulate different network protocols and topologies. We present simulation results for different protocols and topologies, and show that the simulator architecture is capable of simulating the behavior of networks faster than real-time.

Speaker:
Edwin C. Kan.

Title:
Hardware Considerations for Extremely Low Power (uW) and Low Bit Rate (100b/s) Submillimeter Autonomous Microsystems.

Abstract:
Integrated RF technology has brought a booming industry in wireless network in the last 15 years. However, when the system size constraint is below 1mm, on-chip power sources cannot support conventional RF system design. Alternative RF technology from near-field antenna faces entirely different coding and network issues. The wavelength is more than hundred times larger than the transmission distance and hence the phase information cannot be used. The bit rate is low due to the power constraints and has higher error rate. Nevertheless, if ad hoc submillimeter autonomous system can be put together, a new market revolution is at the door. Not only we can extend the present radio-tag security system, but also applications in implanted biomedical devices, non-scanning POS (point of sales), etc. can potentially change the way we live today around again.

Speaker:
Jaiganesh Balakrishnan.

Title:
Time-Reversal in Decision Feedback Equalization.

Abstract:
A major obstacle in reliable digital communication is inter-symbol interference (ISI), encountered in transmission over frequency selective channels. A decision feedback equalizer (DFE) offers an effective and a low complexity solution to combat ISI. However, the DFE is suboptimal and has a performance gap from the matched filter bound. In addition, the DFE suffers from error propagation caused by the feedback of incorrect decisions. The effectiveness of employing time-reversal of the received signal, in a DFE, to combat both of these drawbacks has been demonstrated in recent work by the speaker and McGahey et al.

In this talk, a bidirectional decision feedback equalizer (BiDFE) structure that combines the outputs of a normal mode DFE and a time-reversal mode DFE will be introduced. The performance gains achievable by the use of a BiDFE will be quantified and simulation results demonstrating BER improvements will be provided. The optimization of the tap coefficients of the BiDFE will be considered and closed form solutions provided. It will be shown that an MMSE optimized infinite length BiDFE, under the ideal feedback assumption, attains the matched filter bound.


Speaker:
Toby Berger.

Title:
Information in Living Systems.

Abstract:
The brain may be modeled (well?) as a discrete time homogeneous Markov chain that changes state once every 2 ms, the effective time width of a typical neural spike. During each time step, each neuron either produces a spike (henceforth denoted by 1) or doesn't (henceforth denoted by 0). The state space thus is {0,1}^N, where N is the number of neurons in the brain; N is widely believed to lie between 10^11 and 10^16. Clearly, state space is astronomical; the vast majority of possible states will never be visited during one's lifetime.

In response to a sensory stimulus, the brain chain attempts to "classify" or "identify" said stimulus by traversing a path in state space that converges in several probabilistic senses toward a (unique?) equilibrium distribution associated with the stimulus. This is done in a way that reduces the conditional entropy of the stimulus given the response (equivalently, increases the Shannon mutual information between the stimulus and the response) by as much as possible subject an energy constraint. Here, "energy" is metabolic energy plus perhaps certain other valuable resources. More importantly, during this process the local response of relatively densely connected regions of the brain (e.g., V1 in visual cortex) is characterized by maximization of the mutual information between the joint binary spiking processes of the region's afferent and efferent neurons subject to an energy consumption profile constraint.

Widely accepted models consider the spiking, or not, by each efferent neuron in each time slot in such a relatively dense neuronal coalition to be a consequence of whether or not its post-synaptic potential (PSP) exceeds a certain threshold during this slot. Said PSP is modeled as a fixed linear combination of r.v. associated with each of the neuron's synapses, with each of these r.v. multiplied in turn by two independent binary r.v. The first binary r.v. represents a spike, or not, on the axon afferent to the synapse; the second is a so-called "quantal failure" r.v. which, whenever it assumes the value 0, annihilates the would-be contribution to the PSP from a spiking afferent neuron. In a typical relatively dense coalition of neurons, roughly half the synapses are between pairs of neurons in the coalition, one quarter involve afferent neurons that are "bottom up" (i.e., feeding forward) and the reamining quarter involve afferent neurons that are "top down" (i.e., feeding back).

We show that the information-rate-maximizing joint binary process on the afferent neurons produces a joint binary process on the efferent neurons that is effectively first-order Markov, though generally non-homogeneous. This elucidates how latency in neural processing is kept small despite the sizable number of coalitions along the bottom-up and top-down paths. Moreover, each coalition operates at a saddle point in the sense that, although its afferent-to-efferent information rate is maximized subject to the imposed mean energy consumption schedule, said information rate also simultaneously is minimized subject to the achievement of a specified level of fidelity with respect to the aforementioned global goal of equivocation reduction.


Speaker:
Allen MacKenzie.

Title:
Games of Population Transition and Medium Access Control.

Abstract:
Traditional communication network design assumes that the system designer has complete control over the algorithms running on communication nodes. In many modern communications systems this assumption is not satisfied. Communications nodes may be controled by users with the incentive and capability to modify their terminals to improve their own performance --- possibly to the detriment of overall network performance. We propose a new design paradigm which considers network users to be selfish agents. We use game theory as a tool to analyze the interaction of these agents.

Game theory has been developed primarily by the economics research community. As a result, it is optimized for modeling problems in the realm of economics. Applying game theory to telecommunications network design and analysis will require the development of a new set of game theoretic tools. We have developped one such tool which we denote "Games of Population Transition." Games of Population Transition are games in which the set of players is not fixed. New players enter the game via an exogenous stochastic process; the departure of active players is also stochastic, but depends on the actions of the players. This type of game models a communications network scenario in which players enter the network randomly and depart when they have finished their intended task.

In this talk, I will define a game of population transition, motivate and define an equilibrium of such a game, present sufficient conditions for the existence of an equilibrium, and define a notion of stability for these games. I will then apply this game model to a model of Aloha, proving conditions under which Aloha with selfish users is stable. I will then describe our ongoing work examining a more sophisticated Aloha model and developping a model for CSMA within the same framework. Finally I will describe the future directions of this research.


Speaker:
Wen Xu.

Title:
Spectrally Efficient Partitioning of Video Streams for Robust Transmission Over Multiple Channels.

Abstract:
Reliable transmission of video over wireless networks must address both the limited bandwidth and the possibility of loss. When bandwidth sufficient to transmit the video is unavailable on a single channel, the video can be partitioned over multiple channels with possibly unequal bandwidths and error characteristics at the expense of more complex channel coding (error correction).

In this talk, the problem of efficiently partitioning forward error protected, pre-encoded video data is introduced for transmission over multiple channels. The partitioning formulation will be clearly defined, which exploits the structure of MPEG video while taking into account the varying channel conditions. Considerations such as latency for video applications will be discussed and approximation algorithms are shown to be well feasible for the problem. Simulation results are to be presented for a variety of channel conditions and for a broad range of source material.


Speaker:
Richard Martin.

Title:
A Blind, Adaptive TEQ for Multicarrier Systems.

Abstract:
Multicarrier modulation is the accepted standard for ADSL, wireless LANs (IEEE 802.11a), and European Digital Television. Unlike the single carrier case, equalization is not needed for multicarrier receivers, provided that the length of the channel is shorter than the guard interval between symbols. However, the channel often exceeds this length. The standard solution is to use a channel shortening equalizer that forces the combined channel and equalizer to fit within the guard interval.

This talk presents a blind, adaptive channel shortening algorithm that is globally convergent, and has a complexity similar to LMS. We relate the cost function to that of the Shortening SNR solution of Melsa et al., and provide simulations to demonstrate the performance of the algorithm.


Speaker:
Sergio Servetto.

Title:
Queues under Feedback Control.

Abstract:
We consider a queuing problem in which the service rate of a queue is a function of a partially observed Markov chain, and in which the arrivals are controlled based on those partial observations so as to keep the system in a stable regime.

From standard results in the theory of controlled Markov chains, we know that the optimal controller for this problem satisfies a "separation" property: first compute a probability measure on the state space of the chain (the probability of the system being in any state given the entire history of observations and control actions), then use this measure as the new state based on which the best possible control action is chosen. This measure is sometimes referred to as an "information state". Note that the information state is a random quantity itself, since it is a function of past (random) observations and controls. In this talk, we show how the ergodic behavior of our queueing model is characterized by an invariant measure over all possible information states, and then construct that measure. This last step is not trivial: because of the feedback control loop, none of the standard theorems on the existence of invariant measures for Markov chains are applicable in our context.

To the best of our knowledge, not much is known about controlled queues. The problem is simple: control introduces dependencies that render classical queueing models (based on arrivals and services rates with independent increments) inappropriate. The significance of our results lies in the fact that, by conditioning on information states, we are able to break those dependencies, thus rendering our model analytically tractable.

If time permits, we will also discuss new ongoing work, on applications of these results in the derivation of minimum-delay controllers to stabilize the slotted Aloha protocol, on the design of verifiable MAC protocols for large scale sensor networks, and on performance bounds for TCP/AQM systems.

Joint work with Razvan Cristescu.