Digital Signal Processing Reference
In-Depth Information
Fig. 7.1
General framework
how the environment reacts to the selected actions and adapts his actions accord-
ingly. As mentioned by Simon Haykin in his seminal paper: “The CR is therefore,
by necessity, an example of a feedback communication system”. Hence, feedback
is an essential part of the CR and, hence, of Spatial Learning.
In our current IEEE 802.11 scenario, the feedback from the environment is the
observed throughput. As mentioned above, this limited information assumption ex-
cludes the guaranteed convergence to a Nash Equilibrium, since the exact actions
taken by all players cannot be known [125]. As shown in Fig.
7.1
, the learner is as-
sisted by heuristics, i.e., DT procedures. Heuristic recommendations for the learner
are based on a scenario identification technique, that relies on local observations.
This in effect also increases the feedback information quality from the environ-
ment, which speeds up convergence, while increasing the performance during con-
vergence.
Below, we show how we instantiated the general framework in this case study
chapter. We also refer to the specific sections, where our selections are discussed in
more detail. As stated in Sect.
7.1
, the considered control dimensions in this case
study chapter are the transmission power, transmission rate and carrier sense thresh-
old. In Sect.
7.3.2
, we introduce the possible knob settings for each control dimen-
sion in the discrete state space. As defined in Chap. 3, actions are the relative change
in knob settings. In this case study chapter, the actions are to increase or decrease
each knob if possible. When decreasing the rate, the carrier sense threshold is also
reset. Further details can be found in Sects.
7.3.4
and
7.3.5
. The system scenarios
we consider in this instantiation are the hidden terminal starvation, neighborhood
starvation and asymmetric starvation and starvation-free scenario. The exposed ter-
minal problem cannot be monitored by local observation of the IEEE 802.11 ter-
minal so it cannot be considered as a system scenario. Instead, it is clustered in the
starvation-free scenario. Further details are presented in Sect.
7.3.3
. We detect the
different system scenario using simple energy detection and collision detection. The
exact procedure of scenario identification, based on these detection mechanism is
described in Sect.
7.3.3
. To calibrate the DT policies, we use the theoretical sat-
