Research Projects

Broadband to the bush: Polarization as a new resource in wireless cross-layer design

Collaborators: Prof. Robert Calderbank, Princeton, USA
                       Prof. Linda Davis, Macquarie University, NSW
                       Prof. Bill Moran, EE, Melbourne University
                       Dr. Songsri Sirianunpiboon, DSTO, SA

The advertised research fellow position is associated with this, and the following, project. In this project, we consider the impact of polarization diversity on wireless mesh networking. Polarization diversity can be exploited in rural environments when other forms of diversity are not present. In Australia, there is particular interest in the problem of "broadband to the bush" and we believe that multihop wireless mesh networks may play an important role in helping with this problem. A key research problem is to determine the optimal point on the rate diversity tradeoff curve for each link in the network to provide reliable service from an end-end perspective. Scheduling, routing and flow control are clearly important issues to be addressed. The project involves collaboration with physical layer researchers working on polarization diversity. A full description of the project can be found here

 

Routing, Scheduling, and Flow Control in Wireless Mesh Networks

Collaborators: Professor P.R. Kumar, University of Illinois at Urbana-Champaign, USA
                       Vivek Raghunathan
                       Min Cao

The advertised research fellow position is also associated with this project. Flow control is a mechanism for resource allocation, but in wireless networks we also have to consider power control and scheduling as part of the overall resource allocation problem. In fact, since optimal routes depend on interference, all layers interact, and a cross-layer design process is required. In this project (very related to the above project) we take a joint approach to flow control, routing, scheduling and power control using methods from optimization theory, including linear programming. The key insight is that time-sharing is a fundamental feature of the problem, even when posed in terms of the physical layer problem of maximizing information-theoretic rates. We have a linear programming formulation, but the programs are typically very large, and hence there is a need for heuristic methods to improve the convergence rate. Earlier work considered the problem of routing in multi-hop cellular networks, in which all traffic is destined for the base station. Using spatial point process analysis, we found two optimal transmission range distances to use in routing: one distance for the ad-hoc part of the network, and the other for the final hop to the base station. A full description of the project can be found here

 

Routing and control of dense sensor networks

Collaborators: Prof. Chen Tham, NUS, Singapore
                       Zhenqing Hu

This is a new project in which we consider routing in dense sensor networks in which nodes spend most of their time asleep. To avoid synchronous sleep cycles, we take a random approach to routing, in which a sending node is unaware, a priori, who he is going to send to, but simply makes a request to send to the network. A clever contention resolution algorithm selects the next hop node from those nodes that heard the transmission. We have tested the protocol against recently proposed protocols with a similar flavour (eg Geraf) , and against more traditional protocols that require signalling to determine complete paths to the sink, with promising results. Energy consumption is an important factor, along with delay and packet loss. Further analysis is required to truly understand the potential of this decentralized algorithm.

 

Power and Rate Control in OFDM Cellular Networks

Collaborators: Dr. Lachlan Andrew, Caltech, USA
                       Thaya Thanabalasigham 

In OFDM wireless networks there are two intimately connected allocation problems: power allocation and subcarrier allocation. Since there are multiple mobiles in each cell, one needs to allocate the subcarriers to the mobiles in an equitable manner, perhaps related to their rate targets, but one can also allocate different power levels for each subcarrier. Given fixed rate targets, we consider the problem of minimizing the maximum outage probability across the mobiles in the cell. We provide a layered approach in which power levels are first found, and then a discrete subcarrier allocation algorithm is run independently in each cell. One characteristic of the solution is that the base station uses a flat transmit power spectrum, and varies the powers to each user by varying the subcarrier allocation. A link to a paper on this topic is: here . New work is focusing on rate control for elastic traffic, where power levels and subcarrier allocations are controlled to maximize overall network utility.

 

TCP, congestion signals, and Internet pricing

Collaborators: Dr. Lachlan Andrew, Caltech, USA

TCP currently uses loss as a signal of congestion, but this is problematic for the future, as wireless channels are notoriously lossy. New versions of TCP, notably Vegas, and Fast, use delay as a signal of congestion, but this is also problematic, given wireless networks are also notoriously delay prone, with highly variable delays. Another approach may have more long-term potential, but requires endorsement by router manufacturers; namely, explicit congestion signaling. The price can still be delay, but the router can do the requisite averaging, on the appropriate time-scale (see CLAMP paper ). We have developed a new packet marking technique, that we call ADPM (adaptive, deterministic, packet marking), which builds on the seminal work of Thommes and Coates. A link to our analysis of ADPM, which studies its price-tracking abilities, is here , but much further work can be done in this area. We are testing ADPM on Caltech's Wan-in-Lab, using it to signal congestion prices to sources using the newly proposed MaxNet flow control algorithm.

Another area we would like to explore is real Internet pricing; surely the Internet would become more efficient (and perhaps there would be a faster rate of innovation?) if market forces were in play! The problem is to convey real congestion prices from the routers to the ISP's, so that usage based pricing can be implemented. Perhaps ADPM can play a role in this area?

TCP flow control over wireless links

Collaborators: Dr. Lachlan Andrew, Caltech, USA

                       Dr. Darryl Veitch
                       Catherine Davey

A prominent transport layer flow control protocol in the internet today is TCP. This protocol is appropriate for elastic data traffic that can tolerate some delay. The aim is to regulate the flow of data in the network to ensure that no queue is overloaded, and this is done by the method of window flow control. The window represents how many un-acknowledged packets from the source are allowed in the network at any time, and the rate of flow of packets from the source is therefore controlled via the window size. TCP currently exists in a number of variants: Reno, NewReno, Bic, and others. All these protocols were developed for wireline networks, and only now is attention shifting to their performance over lossy, time-varying links that are characteristic of wireless channels. By far the most popular protocol in the internet today is NewReno, and NewReno operates under the assumption that packet loss implies congestion at a buffer in the network. However, with wireless links, packet loss is very likely due to impairments in the underlying physical channel, and the fact that the link layer may not be well matched to handle these impairments.  It is not clear at the moment whether NewReno will work satisfactorily over wireless links.

There are many challenging research problems in this area. Currently, we are trying to find good models to study the various protocols mentioned above,  as well as to come up with new protocols that will work better. There is scope for very practical research on the CUBIN testbed, implementing our ideas in routers, either for use with our wireless emulator, or our small WaveLan wireless network. At the other extreme, there are deep, theoretical issues to be explored, including optimal control, queueing theory, and asymptotics.

We have developed a receiver-based window flow control algorithm called CLAMP that, in contrast to NewReno, provides fair resource allocation in wireless networks. This includes a very precise model for window flow control that captures the queue dynamics and self-clocking beaviour. Open issues include a complete stability analysis. We are currently studying its behaviour over WLAN 802.11.
 

Multi-user signal processing/smart antennas/information theory

Collaborators: Prof. Rudolf Mathar, RWTH Aachen, Germany
                       Prof. Jamie Evans
                       Prof. Defne Aktas, Bilkent University, Turkey
                       Prof. Emre Aktas, Hacettepe University, Turkey
                       Prof. Alex Grant (University of South Australia)

In recent work we have studied wireless networks from an information-theoretic point of view. Multi-access wireless channels can be shared amongst many users , and signal processing techniques can be applied, such as the linear MMSE receiver. These approaches tend to suppress the interference, as well as enhance the signal, in order to maximize the SIR of the desired user. Similar techniques can be used with antenna arrays (smart antennas). We have studied the asymptotic limiting performance of these receivers, as bandwidth, and number of users grow large. The key theory required is that of the eigenvalue spectra of large random matrices.

Even more advanced techniques are possible if one can exploit the information content of the interferer's signal. The ultimate limits are provided by Shannon's theory of information, and we have applied this theory to not only characterize the limiting performance, but also the optimal allocation of resources (bandwidth and power) as a function of the channel state.

Currently, we are trying to combine ideas from signal processing and information theory. The information-theoretic bounds apply to a single cell, where the base station can utilize the information content of all data streams. This is not always possible in a multi-cell environment, where interfering signals may come from other cells. They may be strong enough to cause interference, but weak enough to be undecodable. The information theory of multi-cell, wireless channels, that mutually interfere with each other, is a murky area where not much is known. Some of the problems are open, and very hard, but we have some simpler problems that are tractable and will provide valuable insight into important issues associated with interference suppression, coding, and the use of smart antennas.