1.1 Inefficient usage of the spectrum
The demand of spectrum resources has been rapidly rising due to the increasing number of mobile device users. However the network has always been facing a problem of inefficient spectrum utilization. A spectrum owner (or primary user/service) subscribes to a band of a licensed spectrum. However the spectrum band is not always used and thus leaves holes in the spectrum, which causes inefficient usage of the spectrum. Fig.
1 describes the spectrum holes in the spectrum band. We can find out that spectrum holes appear in both time and frequency domain.
1.2 Cognitive radio network
In order to solve this problem, we applied cognitive radio (CR), which is defined as an intelligent wireless communication system that is aware of its environment and uses the
Fig. 1 Spectrum hole.
1
methodology of understanding-by-building to learn from the environment and to adapt to statistical variations in the input stimuli [1]. The CR network is imposed on the existing network without modifying the original network [2]. Utilizing the technique of dynamic spectrum access, the CR network is able to detect the unused spectrum bands [3] and distribute them to the CR users (or secondary users/services) who do not subscribe to the bands and have no permits to access the licensed spectrum resources.
The CR network architecture is shown in the figure below.
In Fig. 2, we can discover that the CR users have three access types to use the spectrum resources, either directly or indirectly.
CR network access: The CR users access their own CR base station, on both licensed and unlicensed spectrum bands.
Fig. 2 Cognitive radio network architecture[2].
2
CR ad hoc access: CR users communicate with other CR users through an ad hoc connection on both licensed and unlicensed spectrum bands.
Primary network access: CR users can also access the primary base station through the licensed band directly.
In this paper, we studied the CR network access architecture. In this architecture, the CR users access their own CR base stations (CRBSs). Here a CRBS forms a CR network. As several CR networks share one common spectrum band, a spectrum broker [4] will collect the operation information from all the networks and distribute the resources properly to achieve efficient and fair spectrum sharing. The CR users can then access their own CRBSs and utilize the spectrum resources. The advantage of this architecture is that the CR network is independent of the original primary network and thus can have its own policy of spectrum sharing. In addition, there is only one hop interaction between the CRBSs and the CR users.
1.3 Cournot game
Game theory for cognitive radio networks has been studied recently since the emergence of CR network technology [5]. In traditional spectrum sharing, the network controller will face a lot of communication overhead when a small change of the network occurs. CR network, as a non-cooperative network, therefore requires game theory to model and solve its system.
Niyato and Hossain [6] have discussed the spectrum trading between the primary and secondary networks and considered the whole system as an economical model where the primary network is the spectrum supplier and the secondary network demands
3
spectrum resources. Gao et al. [7] have investigated an auction-based approach for dynamic spectrum access. The spectrum resources are priced and bid for by the secondary users. We formulated the CR network access architecture as a Cournot game (Cournot competition) [8], which is an economical game theory model.
Cournot game model originally describes the situation which more than one firm compete on the amount of the same product they will produce. All the firms decide independently and have no information of other firms' decision. However the price of the product is affected by the total amount of producing. The firms decide their own strategy and compete to maximize the profit. Both the efficiency and incentive issue need to be considered.
We considered the CRBSs as the players in the game. These players demand spectrum resources from the spectrum broker. The residual spectrum resources provided by the primary network are priced and the price is dependent on two factors, the external state and the players' behavior. The external state is the amount of residual spectrum resources provided by the primary network. The less the residual spectrum resources, the higher the price becomes. The second factor is the total demand from the CRBSs.
The price increases with higher total demand. Since each CRBS acts as an individual and has no information of how many spectrum resources other CRBSs demand, the main issue of the game is how many spectrum resources each CRBS should demand from the spectrum broker in order to maximize the profit of itself and also the whole system.
1.4 Stochastic learning
We applied a stochastic learning (SL) solution for each CRBS to decide how much
4
amount of spectrum resources to demand and to adjust it according to the action-reward history. Many works [9]-[11] have studied SL in CR networks. However, they all focus on the architecture in which the CR users detect and utilize the residual spectrum resources directly from the primary network, where the channel selection is the main issue to be discussed. Our SL solution is with the following characteristics: (i) the CRBSs do not need to know the action of each other, (ii) the CRBSs do not have to know the availability of the residual spectrum resources. We proved that the SL-based algorithm converges toward a Nash Equilibrium (NE) point. Numerical results also show the convergence of the algorithm. We could also see that the algorithm performs quite well in the total utility comparing with two other schemes.
This paper is organized as follows. In Chapter 2, the system model for CR network access architecture is presented. We formulated the system as a Cournot game and proved that the model is an exact potential game (EPG), where the game possesses at least one Nash equilibrium (NE) point. Chapter 3 presents the SL procedure for each CRBSs. The proof shows that the SL procedure can make the system converge toward a NE point. The simulation settings are shown in Chapter 4. Finally, the numerical results are given in Chapter 5, followed by the conclusion drawn in Section Chapter 6.
5