Dynamic Spectrum Access
Artificial intelligence (AI) techniques for learning and decision-making may be applied to the notion of effective cognitive radio (CR) systems. The concept of machine learning may be applied to CR to maximize its ability for dynamic spectrum access. The architecture of the proposed system is shown in Figure 4.1.
Here, the knowledge base maintains the states of the system and the available actions. The reasoning engine uses the knowledge base to choose the best action. The learning engine realizes the manipulation of knowledge based on the information observed (e.g. information on channel availability and error rate in the channel).
In the knowledge base, two data structures, the predicate and the action, are defined.
The predicate (inference rule) is used to represent the state of the environment. Based on this state, an action can be performed to change the state so that the system objectives are achieved. For example, a predicate can be defined as “modulation==QPSK AND SNR == 5 dB”, whereas the action can be defined as “decrease modulation mode” with precondition “SNR ≤ 8 dB” and postcondition “modulation == BPSK”.
Given the input (which is obtained from measurement), the reasoning engine corresponds to the current state (modulation and signal-to-noise ratio (SNR) in this case) with the predicates and determines the underlying results (true or false). Then, from the set of predicate results, an appropriate action is taken.