-------------------------------------------------------------------------------
                                       
                                       
		   B O S T O N   U N I V E R S I T Y
                                       
		      Computer Science Department
                                       
			  C O L L O Q U I U M
                                       
		       Wednesday October 1, 1997
                                       
			        3:00pm 
		       (Coffee served at 2:45pm)

                                       
			 Seminar Room / MCS 135
                                       

-------------------------------------------------------------------------------
	Dynamic Channel Allocation with distributed control in mobile
		  microcellular communication systems

			Serge Le Potier
CNET & IRISA, Campus de Beaulieu, 35042 Rennes cedex, France, name@irisa.fr


	In the future (ATDMA based) microcellular systems, existing
frequency planning and network control will be impractical. A solution
to the management problem of such networks is Dynamic Channel Allocation
with distributed control.
	Our goal is to develop and to analyse algorithms that achieve
efficiently to dynamic channel allocation. The decisions for the
allocation are taken at the base station level, and the control takes
into account the  quality of each channel which can be locally measured
on each cell.
        We modelize the interference constraints by means of a graph:
two cells which are bound by an edge can't use simultaneously a given
channel. When a call arrives, the algorithm can decide either to
allocate this customer on a free channel or to reject this call. The
problem is to find an channel assignment policy that minimizes the
blocking probability. The difficulty is that the underlying coloring
problem is NP-hard.
	However, the dynamic channel allocation problem can be
naturally formulated as an optimal control problem in a huge state space
in which the standard dynamic programming method is infeasible. D.P.
Bertsekas and J. Tsitsiklis propose (in Neuro-Dynamic  Programming,
Athena Scientific 1996) to look for an approximation of  the global cost
function by means of local parametric functions,  whose parameters can
be locally trained using a neural network. We choose a  different point
of view: we study how to distribute this control on each cell, so that
no global coordination is needed and the size of state space is thus
reduced.
       We also analyse an adaptative algorithm which provides a priority
list-based channel assignment technique that can automatically adapt to
traffic intensity. It gives a method to obtain an adaptative frequency
planning. Markov chains and stochastic approximation type techniques are
used for the analysis.

Host: Mark Crovella
-------------------------------------------------------------------------------
For colloquium info, including directions, see http://cs-www.bu.edu/colloquium
-------------------------------------------------------------------------------