Digital Signal Processing Reference
In-Depth Information
The variation in time results from the fact that the activity of the IEEE 802.11
networks is dependent on the time-varying demand of its users [66]. From this study
it can also be reasonably assumed that the IEEE 802.11 interference stays constant
over several (thousands) of IEEE 802.15.4 inter-beacon periods.
The spatial variations result from the IEEE 802.15.4 deployment, which usu-
ally covers a very large area (e.g., for monitoring purposes). As a result, the
IEEE 802.15.4 networks are expected to be large both in terms of the area they cover
and the number of terminals they consist of. The theoretical transmission, and hence
also interference, range of IEEE 802.11 networks is 100 m to even 250 m. Although
this is a significant range, sensor networks can cover larger areas since they consist
of a large number of terminals in a mesh topology. As a result, the IEEE 802.11
interference is assumed to affect large geographical subsets of the IEEE 802.15.4
network of terminals (Fig.
5.2
).
As a last step, we assume that an active or interfering IEEE 802.11 network will
always be detected by the IEEE 802.15.4 terminals [65].
The IEEE 802.11 interference can thus finally be modeled as a matrix
I
(n,F )
.
Each interfering network then corresponds to a submatrix of dimensions
(n
i
,
4
)
,
where
n
i
denotes the number of terminals that are interfered by network
i
(depend-
ing on its output power), and where it is taken into account that every IEEE 802.11
interference pattern has a width of four IEEE 802.15.4 channels. Networks can swap
frequency over time, disappear or appear, but this time variation is assumed to be
slow compared to the IEEE 802.15.4 frequency adaptation.
5.2.3 Performance and Energy Measures
We consider delay as a relevant performance metric (throughput requirements in
sensor networks are typically low). More precisely, assuming that sensors monitor a
variable that should be communicated to a central sink, we consider the number of
periods required to forward a measurement to a fixed central sink. This average is
computed over time and over the terminals in the network. The more the network is
affected by the interference, the more periods are required on average to reach the
sink. We assume that every packet is forwarded during each period, to the terminal
closest to the sink that can be reached in that period. As a result, packets travel each
period the largest possible distance.
For the energy cost, we can model the energy needed during every period inde-
pendently of the actual packets sent, received or beacons overheard. This is a valid
assumption as throughput in sensor network applications is typically very low, and,
moreover, the transmit power,
P
Tx
, is lower than the receive power,
P
Rx
[30]. Since
the full receive chain is required to be on for scanning, the power consumption in
that mode is the same as the power cost in the receive mode! During every super-
frame, each terminal is awake to listen at least to its current frequency channel.
The quality of a frequency channel can be assessed by counting the overheard bea-
cons of neighbors. If no beacons are heard, energy detection, which is part of the
