A Feature Partitioning Approach to Casebased Reasoning in Cognitive Radios Daniel Ali Bradley Department of Electrical and Computer Engineering Virginia Tech danielali@vt.edu Jung-Min ”Jerry” Park Bradley Department of Electrical and Computer Engineering Virginia Tech jungmin@vt.edu Ashwin Amanna Bradley Department of Electrical and Computer Engineering Virginia Tech aamanna@vt.edu Abstract—Cognitive radios have applied various forms of artificial intelligence (AI) to wireless systems in order to solve the complex problems presented by proper link management, network traffic balance, and system efficiency. Case-based rea- soning (CBR) has seen attention as a prospective avenue for storing and organizing past information in order to allow the cognitive engine to learn from previous experience. CBR uses past information and observed outcome to form empirical relationships that may be difficult to model using theory. As wireless systems become more complex and more tightly time constrained, scalability becomes an apparent concern to store large amounts of information over multiple dimensions. This paper presents a quickly accessible data structure designed to reduce access time several orders of magnitude as opposed to traditional similarity calculation methods. A framework is presented for case representation, which provides the core of useful information contained within a case. By grouping possible similarity dimension values into distinct partitions called buckets, we develop a data structure with constant (O(1)) access time. I. I NTRODUCTION Cognitive Radio (CR) was first introduced as a technique to adapt a radio to its users’ needs, introduced by Mitola [1]. Since then, software defined radio (SDR) technology has enabled a wide variety of CR applications with different objectives in mind. With a diverse set of applications, CR research has cata- pulted into an active field of study and seen many theoretic and real world applications. Cognitive Engines (CEs) are the main driver behind decision making in a CR system. CEs are charged with managing more layers of the network stack, the number of parameters and their possible values grows the possible case size exponentially. For example, consider a simple example of all possible parameter values as 4 vectors, each able to take on one of 10 possible values. This results in 10 4 = 10000 possible combinations for a simple scenario. Allowing these possible values to be on a continuous dimension instead of a discrete one and the problem quickly gains additional complexity. As a CR grows in number of system responsibilities, the possible states of the radio grow even faster. Traditional Casebased Reasoning (CBR) research advocates an aggressive casebase pruning module [2] [3]. This is an attempt to effectively limit the size of the casebase in order to reduce the access time of information as a whole. This can be unnecessary as memory is becoming less of a concern and processing time remains a more precious resource, especially in the case of tightly timed networks such as IEEE 802.22 [4]. The strength in casebase reasoning is its ability to store past empirical knowledge covering the relevant dimensional space that provides the best performance. This provides motivation of a proper design for a scalable, yet efficient data structure for these relevant dimensions. A critically important aspect of proper CBR usage is the design of the case. Accurate and relevant information should be contained within the case and a framework is presented from [5] that provides adequate utility subspaces to accurately define parameters and their meaning to the system as a whole. Each space is considered a partition of its own with multiple vectors and can be mapped into other spaces to provide common grounds for relative comparison, such as usefulness (i.e. utility) and similarity (e.g. Euclidean distance). This paper’s contributions can be summarized as follows. First a renewed look at casebased reasoning is presented in the context of communications parameters as presented in [5]. Using these parameters this paper also contributes a new case design that separates the aspects to be considered in case similarity with the contents of the case. The parameters of this similarity dimension of the case design is then bucketed to provide a groundwork for the development of a casebased data structure. Finally, this data structure is presented and reduces the correlation between casebase size and access time as well as preserve good situational descriptions and orderly case retrieval. This is shown through several simulations. II. CASEBASED REASONING FOR CR When CBR was first introduced to CR it was under the guise of Case-base Decision Theory (CBDT), which has a natural correlation to CBR [6]. Here we consider the application of CBDT to be synonymous with CBR. CBR incorporates a group of case manipulation mechanisms to drive its reasoning including case representation and indexing, case retrieval, case adaptation, and case-base maintenance [2]. These case manipulation techniques work on the information available in the cases they act on, such as similarity for retrieval, or fitness for projection. In this section, we will lay out the ground work CROWNCOM 2013, July 08-10, Washington DC, United States Copyright © 2013 ICST DOI 10.4108/icst.crowncom.2013.252035