| Up
Further Reading:
P. Gupta and P. R.
Kumar, "The Capacity of Wireless Networks", IEEE
Transactions on Information Theory, vol. 46, pp. 388-404, March
2000.
|
This problem attempts to extend the results
of P. Gupta and P. R. Kumar, who in their pioneering paper (see link),
conclusively proved that the throughput of an ad-hoc network falls as the
square root of the number of wireless node, n. To overcome this
limitation, a certain fraction q of access points are included in
the network to form a wired backbone, and the effect of this on the system
throughput is studied. A practical application of this concept would be in
designing a hierarchical ad-hoc network with wired access points being
connected to the Internet.
Abstract
This treatise provides a solution for the
throughput of an ad-hoc network containing n nodes, a fraction q
of which have wired connections and can behave as access points. The
problem attempts to build on the results obtained by Gupta and Kumar for
an ad-hoc network containing n wireless nodes, namely, that the throughput
falls as the square root of n.
The complete text of
the technical paper may be accessed here (8 pages, 100 KB).
(Requires Adobe Reader)
The slides of the
presentation made at WINLAB on November 8, 2002, may be accessed here (25 slides,
175 KB).
(Requires Macromedia Flash)
Research Update
The expression for expected throughput that
was obtained in the analysis reduced to the original Gupta-Kumar result
when the fraction q of access points is considered to be zero.
However, the treatment of the four
transmission combinations as mutually exclusive was flawed, since a
typical transmission could comprise more than one of the possibilities. To
sidestep this error, a new cluster model has been investigated to solve the
throughput problem.
Abstract
This treatise aims at extending the results
obtained by Gupta and Kumar for the throughput of an ad-hoc network containing
n wireless nodes, by assuming that a fraction q
of the nodes have wired connections and can behave as access points. The
problem is modelled as a collection of qn clusters, each of which
has a wired node as its master. The net throughput of the network is
obtained by combining the throughput in each of the clusters with the
traffic that flows through the wired backbone. The behaviour of the
throughput with n and q are analysed and compared with the
results obtained by Gupta and Kumar, namely that the throughput falls as
the square root of n.
The complete text of
the technical paper may be accessed here (10 pages, 185 KB).
(Requires Adobe Reader)
|
|
