Coexistence: TheWhole Is Greater than the Sum of Its Parts - Software Defined Radios - page 79

Fig. 5.3 The framework of the dynamic frequency selection algorithms relies on feedback from
the environment. The performance increase of the learning engine allows to decrease the quality
of the feedback (no out-of-band scanning), while maintaining similar performance (see Sect. 5.7 )
5.6 Distributed Learning and Exploration
In this section, we explain how we can design scanless approaches, using the generic
framework, presented in Chap. 3. The goal of these scanless approaches is to reach
similar or better performance as those presented in Sect. 5.5 , while removing the
extra hardware complexity and energy cost. Rather than extensively monitor the
scenario, which requires scanning overhead, in this section we learn at RT. As men-
tioned in Chap. 3, this is a vital element of Cognitive Spectrum Access.
5.6.1 General Framework
In this section, we give an overview of the proposed algorithm and how it relates
to the generic framework, presented in Chap. 3.InFig. 5.3 the framework of the
entire control algorithm is shown. The DT procedure used is simple RFS, presented
in Sect. 5.3 . From this DT procedure, a RT procedure is learned through interac-
tion with the environment. The framework relies heavily on the feedback from the
environment. The number of observed neighbors is used to learn an optimal config-
uration. In this case study chapter, we have selected a simple AR-filter as learning
engine. This is further detailed in Sect. 5.6.2 .
The exploration algorithm is discussed in Sect. 5.6.3 . We have already mentioned
earlier that it is very difficult to anneal in the current context. This is further ex-
plained in Sect. 5.6.3.2 . In Sect. 5.6.3.3 , we explain how a scenario-based approach
can overcome the obstacles faced when annealing. The considered scenarios are