ABSTRACT
Contents 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.1.1 Varying Channel Conditions Require Frequent Retraining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.1.2 Why Adaptation Searches Can Be Expensive . . . . . . . . . . . . . . . 126 5.1.2.1 Genetic Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 5.1.2.2 Statistical Methods: Design of Experiments and
Response Surface Methodology . . . . . . . . . . . . . . . . . . . . 128 5.1.2.3 Game Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 5.1.2.4 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.2 Reducing the Adaptation Search through Fractional Factorial Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 5.2.1 Factor Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 5.2.2 Factor Interactions in Wireless Networks . . . . . . . . . . . . . . . . . . . 132 5.2.3 Creating Sparse Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.3 Applicability of Fractional Factorial Designs for Cognitive Radio Networks: A Proof of Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 5.3.1 Estimating Throughput . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 5.3.2 Estimating Latency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.4 Automatic Design and Implementation of Fractional Factorial Designs: The Rapid Adaptation Architecture . . . . . . . . . . . . . . . . . . . . . . . 142 5.4.1 General System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 5.4.2 Enhancing the Adaptation Search . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
5.4.2.1 Feature 1: Directing the Adaptation Search . . . . . . . 144 5.4.2.2 Feature 2: Removing Configurations without
Information Gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 5.4.2.3 Feature 3: Expediting Promising
Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 5.5 Enhancing Current Control Algorithms through the Rapid
Adaptation Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 5.5.1 Guiding the Adaptation Search. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 5.5.2 Removing Redundancy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 5.5.3 Faster Learning and Higher Model Robustness . . . . . . . . . . . . . 154
5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
When placed in its operating environment, a cognitive radio commands over a large number of sensors and system parameters to find the configuration maximizing its operation. This flexibility, however, makes the search for the best configuration complex and highly expensive, as the system parameters may influence each other, thus leading to an exponential if not factorial search complexity.
