The Design and Implementation of a Throughput Improvement Scheme Based on TOW Algorithm for Wireless LAN Hiroki KATO † , Song-Ju KIM †† , Kaori KURODA † , Makoto NARUSE ††† and Mikio HASEGAWA † † Department of Electrical Engineering, Tokyo University of Science 6-3-1 Niijuku, Katsushika-ku, Tokyo 125-0051, Japan ††National Institute for Materials Science, 1-1 Namiki, Tsukuba-shi, Ibaraki, 305-0044 Japan ††† Photonic Network Research Institute, National Institute of Information and Communications Techno logy, 4- 2-1 Nukui-kita, Koganei, Tokyo 184-8795, Japan. E-mail: †katou@haselab.ee.kagu.tus.ac.jp, {kuroda, hasegawa}@ee.kagu.tus.ac.jp †† KIM.Songju@nims.go.jp ††† naruse@nict.go.jp Abstract– Recently, various mobile communication systems are widespread and mobile traffic is increasing. However, the bandwidth of available frequency for mobile communications is limited. Depletion of radio resource is a serious problem. As an approach to solve this problem, a model of cognitive wireless communication has been proposed. This model is for efficient exploration of vacant time slot in multi-channel communication systems. It has been shown that the efficiency of frequency utilization can be improved by multi-armed bandit problem (MAB) solver algorithms. In this paper, we show effectiveness of the TOW algorithm, which is one of MAB algorithms, by comparing with conventional wireless LAN system using the implemented system. 1. Introduction Recently, various mobile communication devices have been widely deployed and the amount of the mobile traffic is significantly increasing. However, suitable frequency bands for mobile communications are limit and scarcity of spectrum bands is getting very serious problem. As one of the approach to solve this problem, the cognitive radio systems have been studied. The cognitive radio improves efficiency of radio resource usage by opportunistic spectrum access [1, 2]. In the spectrum sharing type of cognitive radio system, the cognitive users utilize vacant frequency bands to improve the radio resource usage. The vacant radio resource dynamically changes by change of other user usage of the radio resource. In order to efficiently detect the vacant radio resources, effective channel sensing scheme is necessary. In Ref [3, 4], Lai et al. modeled a cognitive radio as a multi-armed bandit problem (MAB). The MAB is the problem that how a player gets the biggest rewards from slot machines, which probabilistically provide rewards. In Ref [3, 4], each channel is assumed to be probabilistically vacant in time division. Such a cognitive radio system can be defined as the MABs. In this paper, we implement an access point of wireless LAN (CSMA/CA MAC protocol) based on the channel model [3, 4], and evaluate the effectiveness of our proposed scheme, which is called “TOW,” using the implemented system. 2. Cognitive Radio Model as a MAB Figure 1 shows the channel model of cognitive radio [3, 4]. This model has N channels, and each channel is separated time slot. In this model, we consider a primary network consisting of N non-overlapping channels. When the cognitive users communicate, cognitive user senses the channel, and utilizes the vacant channel. Figure 1 Channel model. 3. MAB The MAB introduced by Robbins in 1952 [5] is a simple machine learning problem, based on an analogy faced by a player trying to get the biggest rewards from many slot machines. The purpose of MAB is to detect which slot machine should be selected in order to maximize the amount of rewards obtained from slot machines with finite trials. It is assumed that the player has no information of the winner rate of each slot machine in advance. In the beginning, the player has to gather information on each slot machine. Specifically, the player tries as many slots as