wilamowski-b-m-irwin-j-d-industrial-communication-systems-2011

Ultralow-Power Wireless Communication 10-7
23
23
23
Room temperature in °C
22.5
22
21.5
21
20.5
Original value
Transmitted samples
Reconstructed signal
3 6 9 12
Time in min
Room temperature in °C
3 6 9 12
Time in min
Constant sampling period Adjusts interval of periodic sampling
All samples are transmitted All samples are transmitted
(a) (b) (c)
22.5
22
21.5
21
20.5
Original value
Transmitted samples
Reconstructed signal
Room temperature in °C
22.5
22
21.5
21
20.5
Original value
Transmitted samples
Reconstructed signal
3 6 9 12
Time in min
Bases on a periodic sampling
Sends only significant samples
May adjust sampling interval
Energy-efficiency
FIGURE 10.6 Step responses for different sampling approaches in a temperature control loop: (a) periodic
sampling every 30.s, (b) adaptive periodic sampling every 30.s for the first 3.min and 90.s afterward, and (c) eventbased
(send-on-delta) sampling using a sampling interval of 30.s and a delta of 0.25°C.
Adaptive sampling approaches adjust the sampling intervals based on different criteria, e.g., network
load [GDD04], round trip time, signal dynamics, sampling error, or operation mode. They can be generally
distinguished in adaptive periodic sampling and event-based sampling approaches.
Adaptive periodic sampling adjusts the sampling interval of periodic sampling, for example, in Figure
10.6b, where the sampling period is increased after the step response passed its peak after 6.min. The benefit
is that approaches for periodic systems can still be used, e.g., for signal reconstruction or fault-tolerance
mechanisms. The downside is that reaction times are limited by the constant sampling period.
Event-based sampling transmits only significant changes in sampling values. Different decision
criteria may be used like the difference between the actual and last transmitted value (send-on delta
sampling [NK04]) or the integral of this difference [VK07b]. Send-on-delta, also referred as absolute
deadband sampling [OMT02], is the most common approach. It samples with a constant interval T A , but
sends a sample y(t) only if it differs more than a threshold δ from the last transmitted sample y(t L ), hence
|y(t) − y(t L )| > δ. The benefit of these approaches is that the number of transmitted messages is reduced
based on the signal dynamic. As sampling and processing usually cost less energy than transmission,
event-based sampling is generally more energy efficient. Recent event-based sampling approaches adjust
also the period of wake-ups to the signal dynamics [PVK09] to remove the overhead of periodic sampling
and improve energy efficiency.
Model-based reconstruction may increase the energy efficiency of adaptive sampling approaches even
further by permitting larger sampling intervals, because they use a process model in the sender and
receiver to reconstruct the signal transition between samples. They were applied on adaptive periodic
sampling [MA02] and event-based sampling [LYT06]. However, they need precise process models—
otherwise the sampling quality decreases dramatically.
Also, closed-loop controls can be run in an ultralow-power network using adaptive sampling
approaches. However, the usage of standard, nonadapted control algorithms like proportional–integral–
derivative (PID) lead to a significant degradation of control loop performance [VK07b]. The reason is
that these control algorithms are based on periodic sampling and are not adapted to the rare, nonperiodic
update events. A uniform theory for such event-based controls does not exist and their properties
were analyzed only for some scenarios [A07].
The degradation of control loop performance is also visible in Figure 10.6. The maximum of the step
response for the event-based sampling in Figure 10.6c is slightly higher with 22.7°C than in Figure
10.6a and b with 22.55°C. Note that the sampling period of the event-based sampling is identical
© 2011 by Taylor and Francis Group, LLC