Departments Wireless Communication and Networks Advanced Methods in Signal Processing, Communication and Networking
Queueing networks are established in many disciplines and queueing theory in combination with control and information theory is an extremely useful tool. In wireless systems queuing of data not only has the effect of using opportunistic transmission schemes but has particular design and theoretical effect since there is a canonical routing and resource scheduling strategy, called MaxWeight policy.
The standard MaxWeight policy invented by Tassiulas et al in 1992, is a beautiful and simple concept to achieve the entire throughput region even if it is not known a priori. Very recently developed ideas on policy design called Generalized MaxWeight can be used to improve on network performance including potentially network coding. This research area also addresses the inherent resource allocation problem for multiple antennas typically involving non-convex rate regions. Outcome of this research is a new philosophy for policy design aiming to assess the impact of network coding on routing, scheduling, and even buffering in networks. These features in combination with suitable flow control algorithms can be used to design networks from scratch.
"Improving the MaxWeight Delay Behaviour"
As a major drawback, MaxWeight is known to have a poor delay performance. That is, with decreasing load, the average delay is increasing. Even worse it can behave completely irrationally under low load, as illustrated in the picture.
Here, a situation is shown, where a packet is routed infinitely in a loop.
"Advanced Proof Techniques"
Proof sufficient stability conditions for broadcast queuing systems involves
showing that weight vector can be arbitrarily closely approximated by some integrable function. Left the construction of a 3-dimensional example is illustrated.
A major challenge is to generalize the intuition obtained from broadcast queuing networks to general network topologies. A suitable general model for stochastic queuing networks is the Controlled Random Walk model. |  |
In the contex of broadcast queuing networks there are several examples of throughput optimal scheduling policies. Well known policies are maximum weight matching strategy (first illustration) and the queue-proportional scheduling (second illustration). How can we know if a particular policy is throughput-optimal? For broadcast queuing networks we obtained sufficient conditions for stability.
|   |
