Digital Signal Processing Reference
In-Depth Information
5.5 Scan-Based Approaches
In [19, 65, 70], we have shown that it is possible to outperform RFS, since the con-
sidered IEEE 802.11 interference does not vary every IEEE 802.15.4 period. Here,
we explain how scan-based approaches relying on techniques similar to simulated
annealing, can be elegantly implemented in the considered network setup.
In Sect. 7.3.5, we use simulated annealing can be applied to the distributed op-
timization of IEEE 802.11 networks. In this section, we use a similar technique,
that only considers the potential energy as a variable. Hence, we don't anneal. In
Sect.
5.6.3.2
, we explain in detail why it is very difficult to anneal in this context.
When optimizing the frequency allocation over a large sensor network affected
by dynamic interference, terminals need to keep looking for another channel. Every
period, next to its current frequency channel,
f
i
, terminal
i
scans
n
s
other channels
(grouped in
S
i
) and assesses their performance. This is done according to a reward
function
R
w
.
We use the following definition of
R
w
(f )
as an approximation of the real target
function (also see Sect.
5.4
):
(
beacons heard
)
+
1
when no interference detected,
R
w
(f )
=
(5.2)
0
when interference is detected.
When IEEE 802.11 interference is present, the reward for using that channel is equal
to zero (worst-case). When no IEEE 802.11 interference is present, the reward func-
tion is assumed to be proportional to the number of heard beacons, augmented by
one to distinguish from the worst-case.
In simulated annealing, exploration is embedded in the algorithm to allow the
system to jump out of a local optimum. This means that a scanned channel can be
accepted even if it is measured to be worse than the current channel according to
the observed reward. Usually, this happens with a certain probability that should be
decreased each time interval to eventually converge to the optimal solution.
The group of candidates for possible selection,
C
i
is defined as follows:
C
i
=
(f
i
∪
S
i
)
\
I
i
,
(5.3)
where
f
i
is the current channel,
I
i
are the
channels where terminal
i
detected interference. The probability to select a channel,
f
, from the group of candidates is then equal to:
S
i
is the group of scanned channels and
e
R
w
(f )
T
∀
f
∈
C
i
e
R
w
(f )
p(f )
=
,f
∈
C
i
,
(5.4)
T
where
T
is a constant temperature.
As far as energy is concerned, the proposed algorithm requires to scan the current
channel and
n
s
. Scanning cost is, hence, multiplied with a factor
n
s
as compared to
RFS, where only the current channel is scanned.
