Intra-Net Cognitive Radio Intelligent Utility Maximization using Adaptive PSO-Gradient Algorithm

Artificial intelligence now days are mainly dependent on deep learning techniques as it is rapidly growing and capable to outperform other approaches and even human at various problems. Intelligently utilizing resources that meets the growing need of demanding services as well as user behavior is the future of wireless communication systems. Autonomous learning of wireless environment at run time by reconfiguring its operating mode that maximize its utility, cognitive radio (CR) can be programmed and configured dynamically and their utility maximization inside a building is a challenging task. Re-configurability and perception are the key features of cognitive radio while latest machine learning techniques like deep learning is used for system adaptation. In this paper an adaptive model to enhanced cognitive radio utilization to be maximized is proposed, that is, Particle swarm optimization (PSO) in combination with Gradient-method and intends to maximize the utility of CR. For this purpose the primary objective is allocation of optimum powers to base stations (BSs), which is achieved in an iterative manner keeping in view power constraints. A novel Distributed power PSOGradient Algorithm (DPPGA) is introduced, which assures utility maximization under network power constraints. The information regarding utility and interference of an individual BS is available to all of BSs, which is a key parameter, exploited in the proposed algorithm. Simulations are carried out by considering different scenarios and results are compared with existing algorithms. The performance of proposed algorithm is remarkable.


INTRODUCTION
The Cognitive Radio (CR) is the prime focus of researchers of current era of communication. Lot of research has been carried out for the problems pertaining to the outdoor wireless communication, whereas indoor setups however need further elaborations such as Intra-Net which is the prime focus of this work. In wireless communication, electromagnetic spectrum scarcity is being faced by the new systems. Features like spectrum availability, power consumption, and reserve level during operation are considered to be key features to detect and percept system capability. Intelligent wireless systems being a key component of CR, i.e., make it easy for the device to adapt environment as well as maximize its utility within the available spectrum resources [1]. Contemporary 3G and 4G technologies broadly deployed in cities uses IMT-2000 band, having in the frequency range of 2 GHz, which is unable to easily penetrate inside huge buildings. Cognitive Radio based internet of Thing (CR-IOT) is a leading step to smart world of technology in upcoming 5G networks [2]. The 5G, Machine to Machine (M2M) communication will be there which is beyond current mobile networks; CR networks therefore have the capacity to address key challenges of limited radio spectrum faced in 3G and 4G networks [3]. Two approaches used for efficient utilization of licensed spectrum, one is by using small cells (e.g. Femtocells) which extend licensed cellular services inside the building for users and other approach is CR to develop spectrum utilization and increase the efficiency of spectrum sharing systems. Josif Motila proposed the first concept of CRs in 1998 [4]. Spectrum holes are searched by CRs for communication, running applications, services and transporting data packets over the CRs nodes. These terminals are capable of sensing spectrum and communication environments. The CR adjusts and reconfigures itself according to well-read information with compliance to regulations. Downloading files from specific Access Point (AP), the CR mobile station sense the spectrum before communication. In order to communicate with other CRs, the CRs establish an ad-hoc network. The CR has three basic types, Interweave CR, Overlay and Underlay systems. In Interweave CR system; Primary Users (PUs) are active with CRs and able to transmit signal using unoccupied spectrum holes. The CR continuously monitors the transmission environment and search for vacant spectrum [5]. In Overlay system, PU provides their signal information including Codebooks and share massages with Secondary User (SU) and thus CR coordinate with PUs which further improve transmission efficiency [6]. In Underlay systems, the CRs are active using low powers in parallel with PU transmission under some threshold levels to avoid interference at PUs' side irrespective of PUs transmission. In License Shared Access (LSA) the SUs of CRs are allowed to access the shared spectrum subject to services of PUs are not distressed [7]. Alternative approach is centralized authority 13 which calculate and assigns a transmit powers to CR network; this approach is discussed in details by a researcher [8]. Distributed power allocation algorithm for industrial sensor network is already being addressed [9]. Distributed power allocations among the CR users protocol based using auction is also being researched [10], which increase the network performance of CRs. Pricing and bidding strategies were used to maximize the users weighted sum rates. Distributed power allocation in CR network under network power constraint is discussed in a research work [11]. Energy efficient power allocation in CRs network schemes have been proposed [12] and power allocation strategy in CR network in signal carrier is being described thoroughly [13], with transmit average and interference power constraints. A (PSO) algorithm was proposed which finds the optimal solution of power allocation [14].
Global Database server (GLDB) for specific area (building) calculates the powers and controls the transmit powers which must be accepted by CR before transmission [15]. The transmit powers of CRs are synchronized with predefine powers to avoid interferences in PU sides of license users [16]. Utility maximization, high reliable, outage probability and interference free are technical scenarios of Underlay CR network. As conferred different types of CRs, the prevalent type is Underlay network. These CRs are functioning in parallel with PUs without interfering by controlling its transmitting powers. The CRs are generally functioning inside the building as shown in Fig. 1. The SBs are installed in corridors of the building and communicating with their respective MSs [17]. for that specific CR network to avoid interference with PUs. The gradient based optimization algorithm, proposed in literature used for maximization of CR network utility under network power constraint may be stuck in local optima [18]. Therefore the required utility maximization in Underlay CR network cannot be achieved. A new algorithm is required for CR network to utilize the internal resources and overcome the gradient optimization imperfections. The proposed algorithm must be capable to find-out the optimum transmit powers for Cognitive Radio Base Stations (CR-BSs) comprehensively and eventually to be able to enhance the overall network utility. Heuristic based gradient optimization technique is proposed for maximization of network utility. The contribution of this paper is summarized as follows; This research work is further organized in the following structure: Section II describes in depth study of choosing Particle Swarm Optimization (PSO) and its working in general. In Section III material and methods; discussion of proposed technique in respect of mathematically derived proofs and optimization problem formulations, Section IV a working algorithm PSO Gradient Algorithm (DPPGA) and all path loss office model, where all path losses and shadowing, fading factors are incorporated to calculate the channel gains between CR-BS and CR users, is presented. Section V presented simulation results achievement and its relevant discussion over the proposed scenarios. Section VI concludes the presented work whereas Section VII highlights the limitation as suggested future work for improvement.

II. PARTICLE SWARM OPTIMIZATION (PSO)
PSO is a stochastic optimization process used for global optimization. Its computational cost is high by carrying random searches compared to gradient-based optimization, used for local optimization. Deterministic Gradient based optimization may converge fast, however this method may held in local optima in multimodal problem. The research work presented in this paper is novel adaptive PSO-Gradient Algorithm which assures maximization of CR network utility. This Adaptive optimization technique maximizes Signal to Interference plus Noise Ratio (SINR) of CR network. The utility of Underlay CR network get maximized by using this Adaptive algorithm i.e. PSO-Gradient in comparison to independent gradient methods. We presented cooperated scheme which exchange prices / Interferences between CR-BSs and non-cooperative scheme where CR-BSs are not sharing their prices with other BSs. In this paper, we use both random and deterministic optimization that can search out combined benefit and avoid their limitation. PSO is used for global search and gradient-optimization used to get best local optimum. In order to formulate the problem using PSO we consider Eq. 10 and Eq. 11 as cost function to maximize the utility. The CR-BSs randomly initialized with transmit powers. We consider random powers for PSO, (1 x n) is called the particle (population) and compute its utility of every set of powers and search out the best utility as global best. These power particles are searched similar to a birds flock to get the optimum solution. After calculating its utilities each vectors of particle has its local best and among the local best select global best. To calculate the particles velocity we have,  are uniformly random number generated independently for each dimension, P is local best found by particle and G is the global best in entire population. V is bounded by constraint min P and max P to prevent divergence. Selecting

{M }
Network Power constraint provided by GLDB is supposed to be there i.e. is the maximum power limit that must not be exceeded by the sum of individual transmit powers of BSs. Accordingly, the set of powers is: and the corresponding power constraint is: The available resources in the network are equally distributed among the BSs, which are further uniformly distributed among the respective MSs [19]. The overall set of resources is given as: The objective is to maximize the sum utility of the CRs given as: (p, W U (p, w ) Where s u is the Shannon Rate capacity in (Nats/s/Hz). PF-rate share the common resources power and scheduling weights in fair distribution among the MSs served by BS i .
This logarithmic form of utility will improve the user's data rates, even those users that are experiencing low data rates due to high channel interference. SINR for each Mobile Station  (8), this equation is non-convex, which can be made convex if one variable is fixed and optimization performed on other variable briefly discussed by in a study [20]. We considered the decomposition method, which will eventually allow us to write pricing algorithm for finding solution of Eq. (8) in distributed way using PSO and gradient methods. Here we use single channel model of frequency flat channel. When there is more than one variable, the primal decomposition is a suitable method used for decomposition [21]. This Eq. (8) can be solved by using primal decomposition, by fixing powers and decouple the problem into sub-problems independent optimization problems of scheduling weights, so the problem divide into two stage optimization problems. At one stage fixing powers, then the problem decouple into scheduling weights optimization sub- Now Eq. (8) turns into convex problem and can be solved at each BS and updating coupling variable power. Optimization over "p" is perform using pricing algorithm, the master problem is , ii  is called the power benefit calculated by i th BS and j,i  is called power prices, which are calculated and shared by other BSs. ,    After getting this new population, cost function evaluated to find the utilities of each population and get the local and global best. Repeat the process until converge or maximum iteration is reached. The global best powers are used for gradient based optimization to get (maximum) utilities. Complete code DPPGA algorithm shown above.

V. RESULTS AND DISCUSSION
This research study analyzes the performance of proposed algorithm on the basis of results achieved with cooperative and non-cooperative schemes. The simulations schemes using Table II as parameter values are presented in Section IV then the simulation results are discussed in Section V. In this study low power small cell network consisting of low power BSs to provide services to its MSs in three-story building. As shown in Fig. 1. Cognitive Base stations placed in building corridors, rooms having multiple MSs communicating with their BSs. The size of the three-stories building is 100x50 meters with height of 10feet of each story. A top view of the building shown Black little circled the installations of low power BSs. The signal propagations from BS to MS are modeled by using WINNER-II office model [22]. Path loss model parameters are shown in Table I.
The path-loss calculated as, 10 10 ( In Table II we take parameters of simulations of WINNER-II model, we use 12 BSs and 12 MSs. Maximum transmit powers of BSs is set to 100-mw and 1-mw power minimum. Noise figure, shadowing and thermal noise are also incorporated. In order to make the simulation scenario simple, among each cognitive base station, only select one cognitive base station according to received signal strength. Assuming the backhaul connections are available between BSs to communicate and exchange prices. In order to solve our optimization problem, i.e., Eq. (10)  however not sharing the powers prices. The SINR at MSs significantly increased, by implementing DPPGA. PSO-cooperative scheme has more SINR than cooperative and non-cooperative schemes as shown in Fig. 3. It is to mention that by using non-cooperative scheme the possible cause that BSs does not share their pricing information among the neighboring BSs. The determination of the paper is to enhance the network utility by using DPPGA. The network utility in this case is increases by DPPGA. In DPPGA PSO algorithm is incorporated to find the optimum global maxima of powers allocated to Cognitive Base Stations and then using this optimum powers, calculate power prices at BSs then each BS share its   Fig. 3 shows that as the number of iterations has increased gradually so in comparison to the other techniques proposed PSO power technique provides high  Fig. 4 has shown relationship between Cumulative Distribution Function (CDF) with respect to network utilization underperformed simulations. In comparison to other cooperative and noncooperative technique the proposed PSO techniques utilizes network initially at start with 0 % but while non-cooperative started at -2 at start; proposed PSO technique has ended at 15%. While other two techniques ended at 7% and 12 %. So network utilization has increased gradually for all the techniques. Different schemes, PF-rate utility plots have been shown in Fig. 2, 3 and 4. In this case, the overall network utility is enhanced and maximized.

VI. CONCLUSION
In this research study an intelligent DPPGA pricing algorithm based on decomposition method is proposed that maximizes resource utilization under CRs network under network based on power constraint. PSO based gradient algorithm for PF-rate maximization to meet the network power constraint.
Simulations are carried out on small cell CRs network; different cooperative and non-cooperative scenarios are compared and results shows that the proposed intelligent PSO based gradient adaptive method is efficient then general gradient based optimization to maximize the network resources utilization.

VII. FUTURE WORK
In future, artificial intelligence latest techniques such as Deep Learning could be investigated so that human interaction could be minimized under Cognitive Radio Networks; by doing this more intelligence could be possible to utilize resources efficiently and with limited power utilization.

AUTHOR CONTRIBUTIONS
Javed Iqbal and Imranuallah Khan proposed idea and conceptualized the problem. Mukhtar Rana guided to formulate proposed methodology, Ahthasham Sajid write initial draft and revised draft, Imran Baig improved-original draft and revised draft to be submitted, Afia Zafar works on formatting of the paper according to Journal requirements and improved quality of figures.