An energy-optimized game algorithm for wireless sensor networks

The ongoing development of wireless sensor networks has been greatly restricted because of scarce spectrum resources, limited battery power, and ineffective topologic structures. Thus, how to construct a suitable topological structure and allocate the appropriate node channels has become an urgent problem. In this study, we built a game model that took into account the influence of channel allocation and topology structure on network performance. The game model considered the nodes in wireless sensor networks to be players and took transmitting power, node channels, and node rest energy into account to establish the income function. Then, the model certified that it has Nash equilibrium. Next, we propose an energy-optimized game algorithm joint topology and channel for wireless sensor network (CETGA) in accordance with the game model. The CETGA algorithm improved each node’s income by changing the transmitting power and node channel gradually, assuming that the network retained connectivity. Then, we demonstrated that the algorithm could converge to a Pareto optimal. Finally, we used MATLAB software to verify the simulation. The results show that the topology created by CETGA is with low interference and long lifetime. In addition, the nodes’ average residual energy is more balanced and the network robustness and real time are improved.


Introduction
Lots of tiny sensor nodes communicate with each other and constitute a multihop self-organizing network which is named wireless sensor network (WSN). 1 WSN is widely used, not only to measure various physical parameters, but also in many monitoring fields, 2,3 such as agricultural surveying, 4 smart traffic monitoring, 5 environmental monitoring, 6 medical applications, 7 and many other fields. The topological structure of a WSN greatly influences the reduction of interference and energy consumption, the enhancement of network robustness in real time, and the prolonging of the network's lifetime. 8 In addition, channel allocation (CA) provides the foundation for further reducing interference and increasing network capacity. 9 Therefore, an urgent problem for WSN has become how to design a proper topological control algorithm to improve network reliability and many other issues based on network connectivity.
Currently, the two types of WSNs, single channel and multichannel, influence its topology control.
In a single-channel WSN, power control is the main method and is likely to reduce transmitting power to prolong the network's lifetime. Basing their design on the local minimum spanning tree, Li et al. 10 proposed a local minimum spanning tree algorithm to devise a connective topology. But the network's robustness was poor because the structure was sparse. Zhou et al. 11 proposed a control algorithm for network topological connection optimization based on network efficiency and average connection degree to increase the network's throughput. But the algorithm is not suitable for a WSN because the node energy is limited. Li et al. 12 provided a distributed topology control algorithm that 1 State Key Laboratory of Mechanical Behavior and System Safety of Traffic Engineering Structures, Shijiazhuang Tiedao University, Shijiazhuang, China 2 School of Mechanical Engineering, Shijiazhuang Tiedao University, Shijiazhuang, China 3 He Bei Extension and Experiment Center for Water Technology, Shijiazhuang, China considered the transmitting and receiving powers of nodes synthetically. The algorithm provided a method to extend the network life cycle by analyzing the effects of node communication distance, circuit loss, and node load on the network. Chen et al. 13 constructed a load balancing evaluation mode by researching the relationship between neighbor-residual energy, transmission power, and load. And then proposed a distributed topology control algorithm to show that the model can balance network load and energy consumption.
However, limited spectrum resources have greatly restricted WSN's further application. Thus, the combination of multichannel technology and topology control is utilized to optimize the performance In Thomas et al., 14 the algorithm was divided into two parts: power control was used to construct the topology structure, and CA was used to optimize network performance. But the method was just a simple superposition of the two parts and did not take into account the internal relations between them, so the topology structure obtained did not result in the best performance. Li et al. 15 used dynamic adjustment of power and channel to increase antidisturbance ability of the network, which improved throughput and enhanced stability. The paper explored the combined optimization of power and channel, however the topology was dynamic. Because the consumption of node energy was large, that method is not suited for WSN, whose available energy is limited. Hao et al. 16 proposed a distributed topology control and CA algorithm to establish an optimized network. The network can possess the lower inference and many other attractive network performances such as the stronger robustness. But the physical interference model is used in their model, which is too complex to quickly finish the calculation when the node number is much large. By studying the relationship between k-connected logical topology and the maximum number of assigned channels, Bao et al. 17 presented the lower and upper bounds of the maximum allowable number of assigned channels for k-connected logical topology and a static channel assignment algorithm according to minimum spanning tree search. Derdouri et al. 18 studied multicast features in wireless mesh networks by taking advantage of shared links to reach simultaneously multiple users and then evaluate the performance of two algorithms, one named Node Joining the Multicast tree and the other named Node Disjoining the Multicast tree to improve the throughput, delay, and robustness.
Basing our work on the analysis above, we first established an energy-optimized game model joint topology and channel for a WSN. Then, we designed an energy-optimized game algorithm joint topology and channel Joint channel and energy topology game algorithm, (CETGA) based on the model. The unique features are as follows: 1. The CETGA takes node energy consumption and network interference into account to formulate the game model's utility function. In addition, if a node's residual energy decreases, the node is more likely to retain its transmitting power to increase the utility value. 2. In the CETGA, a node, whose residual energy is low, is more likely to choose the minimum power to reduce consumption itself. Meanwhile, the immediately adjacent nodes, whose residual energy is higher, must use more power to keep the network connected. Thus, the energy consumption is balanced. 3. In the CETGA, every node adaptively adjusts its channel when the power changes. Then, the network is settled finally.

Establishment and analysis of the model
In a WSN, sensor nodes consume energy when they communicate with each other. Because the energy of nodes is limited, most nodes are likely to be ''selfish.'' In other words, a sensor node is inclined to optimize its own income as much as possible instead of collaborating with other nodes to improve the performance of the entire network. The status of the WSN completely matches the application of game theory. Thus, how to determine the network structure of WSN could be attributed to the optimal solution for a combinatorial algorithm based on game theory. [19][20][21] In section ''The game model,'' a CETGA for a WSN is established.

The game model
To establish a game model for a WSN, the definition of node interference, named the interference model, which is based on the neighbor node number, 22 is as follows.
Definition 1 (node interference). The interference I (c) k (p (c) k ) of any node k using channel c is the sum of node numbers. A node's number contains two parts: one is the interference generated by node k (i.e. the node number using channel c contained in the scope of node k using p (c) k ), the other is the interference suffered by k (i.e. the times k contained by any other node with c).
According to Fudenberg and Tirole, 23 a game model comprises three factors: player set, policy set, and income function. Here, players are the decision-makers in the gaming process, while policy set is the collection of all policies. Income function is the income of each player when the policy changes.
In our model, we designated an energy-optimized game model joint topology and channel as G = fN, S, ug, of which N, S, u respectively represents the factor motioned in a game model. The player set shown as N = f1, :::, k, :::, ng, n is total number of all nodes and k is any one node in WSN. In addition, we use Àk to denote nodes in WSN that are not k. The policy set S is S = (S k , S Àk ), where S k is the policy of k and S Àk is other node's policy vectors instead of k. The policy set of any node k is S k = f(p k , c k )g, p k 2 A k and c k 2 C k . Here, p k refers to node k's transmitting power, A k is an ordered policy set, and A k = fp max = p 0 , p 1 , p 2 , :::, p v = p min g, where p v \ p vÀ1 . p min is the minimum power, whereas p max is the maximum in WSN. c k is k's channel, and C k is the policy set. Let C k = f1, 2, :::, Cg, where C is the maximum channel number that can be available in our model and the channels that we use in this paper are all orthogonal channels. Any node k's income under the policy combinations (S k , S Àk ) can be expressed as u k (S k , S Àk ), and the income function u = fu 1 , u 2 , :::, u n g consists of all nodes' incomes. The income function is as follows In equation (1), f k (p, c) is the connectivity of network and it is the monotonic increasing function of p. If the transmitting power p is large enough, the network is connected and f k (p, c) = 1, else f k (p, c) = 0. p max is the maximum transmitting power. Take Definition 1 as a reference, I k (p max ) can be considered as node k's interference using p max and it also comprises two parts: one is the number of nodes covered by k using p max , and the other is the times that k contained by any other node with p max . Obviously, the inequality I k (p max )5I (c) k (p (c) k ) always holds. a and b are non-negative weight factors.
k represents k's present transmitting power at c, here p k = p (c) k . The income function can reflect many aspects of network performance; the major aspects are as follows: 1. Network connectivity is the basic requirement of the topology construction for WSNs. Information can be transmitted effectively in the network only if the conductivity of the WSN is guaranteed. In equation (1), f k (p, c) is the connectivity of the network. f k (p, c) = 1 indicates that the network is connected; f k (p, c) = 0 indicates that it is disconnected. If the network is disconnected, the value of u is negative because a and b are non-negative. In this way, the connectivity of WSN can be maintained during the process of building the topological structure. 2. For an energy-constrained WSN, the node's energy supply is an important factor that affects the lifetime of the WSN. The lower the residual energy E r (k) of any node k is, the greater 1=E r (k) is. In this circumstance, the change of node k's transmitting power p k has a greater effect on utility. In equation (1), the value of weight factor b is non-negative. When the residual energy E r (k) is lower and the transmitting power p k is greater, the node k's income of utility function becomes smaller. Thus, if node k's residual energy is relatively lower, the node tends to choose lower transmission power to reduce its own energy consumption.

Network interference is an important indicator
for evaluating the performance of a WSN. The amount of network interference directly affects the quality and efficiency of communication. In equation (1), ÀaI (c) k (p (c) k ) refers to the node k's interference. Because the weight factor a is nonnegative, the greater the network interference, the lower the node's income value. Besides, based on the definition of I (c) k (p (c) k ) and the illus-

Model analysis
Nash equilibrium (NE) is the solution of the game model and all players want to achieve this status. Its definition is given as follows.

Definition 2 (NE).
For i 2 N, s i 2 S i , the policy combination There are many ways to prove the existence of NE, and the ordinal potential game (OPG) is one of them.

Definition 3 (OPG).
When an ordinal potential function (OPF) V : S ! R can be established by equation (3), G = fN, S, ug is an OPG 24 where s Ài 2 S Ài , s i , t i 2 S i , and i 2 N. From Monderer and Shapley, 24 if a model is found to be an OPG, there must be at least one NE. Thus, Lemma 1 is given first.

Lemma 1.
For any node k, if 2k's states stay the same, the change of k's interference is the same as the sum of 2k's interference changes.
.Based on Definition 1, the description is below.
The interference created by k is the sum of j's interference subjected by k (here j is the other node except k), while the interference subjected by k is the sum of j's interference created by k. Thus, the interference change created by k is the same as the sum of the interference change subjected by j, and the interference change subjected by k is the same as the sum of the interference change created by j. Thus, equation (4) is always true Theorem 1. The energy-optimized game model joint topology and channel G = fN, S, ug is an OPG. Its OPF is Proof. Assume that 2k's states stay the same, k's income change is shown in equation (6) if k changes from (p k , c k ) to (q k , c k 0 ) By this time, the change of ordinalpotentialfunction V is f is the monotonic increasing function of p, and four situations are now discussed to prove Theorem 1.
( always holds. Moreover, DI k = P j2N, j6 ¼k DI j is true according to equation (4). So the equation above can be simplified as follows Thus, we can conclude that DV = 2Du k . Clearly, we have sgn(DV) = sgn(Du k ). ( Thus, we can conclude that DV = 2Du k . Clearly, we have sgn(DV) = sgn(Du k ).
(3) f k (p, c) = 1 and f k 0 (q, c 0 ) = 0 f k (p, c) = 1 represents the network is connected when node k uses transmission power p. Since 2k's states stay the same, when k's power changes from p to q, f 0 k (q, c 0 ) = 0 (i.e. the network is no longer connected), which means that k's power decreases, then we can get p . q p max À p k + q k ð Þ a and b are non-negative. And the inequalities I k (p max )5I (c) k (p (c) k ) and p max 5p k always stand up, then Du k . 0 So we can conclude that DV . 0 because Du k . 0. Clearly, we have sgn(DV) = sgn(Du k ).
(4) f k (p, c) = 0 and f 0 k (q, c 0 ) = 1 f k (p, c) = 0 represents the network is disconnected when node k uses transmission power p. Since 2k's states stay the same, when k's power changes from p to q, f 0 k (q, c 0 ) = 1 (i.e. the network is connected), which means that k's power increases, then we can get p \ q a and b are non-negative. And the inequalities I k (p max )5I (c) k (p (c) k ) and p max 5p k always stand up, then So we can conclude that DV \ 0 because Du k \ 0. Clearly, we have sgn(DV) = sgn(Du k ).
In conclusion, sgn(DV) = sgn(Du k ). So the energyoptimized game model joint topology and channel G = fN, S, ug is OPG.
Theorem 2. An NE must exist in CETGA G = fN, S, ug.
Proof. Since CETGA G = fN, S, ug is an OPG, the policy combination which can maximize its OPF is an NE.
In G = fN, S, ug, the policy combination is limited, so there must exist one policy which can make the OPF V maximum. That is, G = fN, S, ug has an NE.

CETGA
Considering the relation between topology and channel, a CETGA is established to solve the model and this algorithm considers not only the node residual energy but also the convergence.

CETGA algorithm
In CETGA, each node first gathers other nodes' information in its achievable region, and then decides its policy. The process is as follows: (1) Let all nodes' power be the maximum power and their channel the same channel c (0) . (2) m = 0, round = 0.
(3) k's current power is p k = p (m) 2 A k and current channel is c k = c (round) 2 C k . (4) Based on equation (1), any node k's income value u k is obtained. Judge whether every node reach an NE. In other words, if all nodes' income values are the same as that in previous cycle, CETGA ends, or it goes to (5). (5) m = m + 1 (6) k = 1 (7) Judge whether the network is connected: if it is, CETGA goes to (8); if not, k remains its power and channel in last cycle, that is, , :::, u (0) k ) and channel be c Ã k = arg max (u k (p Ã , 1), :::, u k (p Ã , C)) (9) If k\n, k=k+1, CETGA turns to (7). If k5n, round=round+1 and CETGA turns to (4).
The flow diagram of algorithm is shown as follows.

CETGA analysis
To energy-constrained WSNs, whether the algorithm can converge is a most important feature. Theorem 3 verifies the ability to converge.  Among the situations mentioned above, we can easily conclude that u k (new)5u k (old) ) P j2N, j6 ¼k u j (new) 5 P j2N, j6 ¼k u j (old). Because the function is monotonic and the power level and channel number are limited in the policy set, the CETGA can converge finally.
In a game model, the states of NE can be divided into three situations: none, one, or numerous. From Theorem 1, CETGA can converge to NE. But there is no guarantee that CETGA can converge to optimal NE. To achieve the optimal solution, another important concept is required.
Definition 4 (pareto optimality, PO). A policy vector s Ã is PO if no such policy vector exists to make u i (s Ã )4u i (s) for any i 2 N, and u j (s Ã ) \ u j (s) for at least one j 2 N. 25 Theorem 4. CETGA can converge to PO.
Proof. Nodes in the network obtained by CETGA could be separated into two main categories: (1) nodes using minimum power are to maintain the network nonseparation and (2) nodes with non-minimum power are to guarantee network robustness and real time.
If k's residual energy reduces, k would increase its income by decreasing its power to the minimum power. In this case, the structure would not be connected if k's power were continuously reduced after the algorithm ceased. At this time, if the network requires connectivity, nodes other than k should increase their power but will decrease their incomes.
If a node k's residual energy is relatively large, k has a larger power to guarantee the income itself. Right now, no matter reduce k's power or change others' station will decrease k's income.
Overall, no one can increase its own income without decreasing that of others. Based on Definition 4, CETGA can converge to PO.

Simulation verification
The network performance is studied by MATLAB. Assuming that nodes' residual energy in WSN is subject to Poisson distribution. Every node's maximal transmission range is d max = 150 m and the power threshold is p th = 7310 À10 w.

Weight value determination
In equation (1), a and b are unknown. Thus, we studied the influence of weight value change on network performance. We placed 30 nodes in the 300 m3300 m region randomly. Given b = 1, Figure 1 shows the network performance when a is 0.1, 0.5, 1, 3, and 5 separately.
From Figure 2, when a = 0:1 or a = 0:5, it is easier to waste node energy during information transmission because the average transmission power is higher. At the same time, the average interference of nodes is greater as well. Thus, transmitting information affects each node, which increases the packet loss rate and reduces the transmission accuracy of the network. Moreover, the greater channel variance represents an unbalanced CA. When a = 3 or a = 5, the power reduces, so is the average residual energy. Then, the network robustness is relatively poor and once a node dies, the whole network is paralyzed. What is more, the network real time is reduced. That augments the forward number during information transmission, which increases the node energy consumption to a certain degree.
In conclusion, decide that the weight value as a = 1 and b = 1.

Network structure
Randomly selected 30 nodes to place in a 300 m3300 m region and assumed that the maximum channel number was 5. Then, we compared the CETGA with a combination of the distance to ideal control (DIA) topology control algorithm 26 and the CA algorithm. 27 In Figure 3, the different colors indicate that the nodes used different orthogonal channels. In each node mark (x, y), x is the node number and y is the residual energy of the node x.
In Figure 3(a), the topology of DIA + CA is relatively simple. One or more nodes in the WSN undertook too many forwarding tasks; for example, nodes 3, 5, and 11. Because the algorithm did not take node energy into consideration, some nodes with lower residual energy were placed into the forward position and died prematurely, which would reduce the network's lifetime. In Figure 3(b), at the cost of improving node transmission power appropriately, some nodes added other transmission paths to reduce forward tasking and enhance the network's robustness. In this way, the real-time performance was improved and the network's lifetime was prolonged.

Topology performance
We raised 20-40 nodes in 300 m3300 m, and let the maximum number of orthogonal channel to be 5. Figure 4 compares the performances of the CETGA and the DIA + CA. The average transmission power is the mean value of nodes's transmission power in the network, and the less the average transmission power is,the less energy consumption will be. Then the average node degree directly affects the robustness performance of the network. As shown in Figure 4(a) and (b), the topology obtained by the CETGA had a slightly higher power and node degree than those of DIA + CA. This clearly shows that the proposed algorithm can create a topology structure that can reduce node power and energy consumption as well as enhance robustness and real-time performance based on guaranteeing network connectivity.
The average node interference represents the degree of interaction between nodes in the transmission process and the average residual energy within the node transmission range directly affects the lifetime of the network. In Figure 4(c) and (d), the average node interference in the CETGA's topology was less and the average residual energy within the node transmission range was greater than those of the DIA + CA. Thus, the CETGA can reduce inference and extend life of WSN. The channel variance is the level of equalization using the channel. In Figure 4(e), the relatively small channel variance further illustrates the superiority of the proposed algorithm.
The average-hop of the shortest path between two nodes represents the real-time performance of network transmission. In Figure 4(f), the topology obtained by CETGA has better real-time performance. Moreover, the algorithm can reduce network energy consumption by reducing the number of forwarding tasks.

Conclusion
Basing our work on game theory, we studied the problem of topology and channel. We first built an energyoptimized game model joint topology and channel. We then designed a utility function considering node transmission power, node channels, and residual energy. The proposed model was determined to be an OPG with an NE.
Second, in accordance with the model we designed an energy-optimized game algorithm joint topology and channel (CETGA). The algorithm increased the node income by gradually reducing the node transmission power and adjusting the node channel. The algorithm could then be certified to converge to Pareto optimality in guaranteeing network connectivity.
Finally, MATLAB was used to make simulation. Simulation result verified that the topology obtained by the CETGA could result in a more balanced structure and better robustness and real-time performance.
Moreover, the average node interference is relatively small, which can ensure the accuracy of the information transmission. In addition, the average residual energy of the nodes is higher, which can effectively prolong the lifetime of the network.

Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.