Cognitive allocation method for multi-user channels in wireless networks

Abstract

The present invention provides a cognitive allocation method for multi-user channels in a wireless network, and belongs to the field of cognitive wireless sensor networks. First, the cognitive wireless sensor network model is initialized; then, node clustering based on adaptive PU number routing is performed, including routing establishment and node clustering; then, cognitive detection of the channel by the node and establishment of the network throughput problem are performed; finally, spectrum allocation is performed based on the frog leaping game optimization algorithm. Experimental verification shows that the algorithm of the present invention is not only applicable to the ideal free space model, but also to the normal logarithm model of large-scale fading and the Rice model of small-scale fading. Since different conditions correspond to different channel indices, the global benefit after the algorithm is not positively correlated with .

Description

Technical Field

[0001] The present invention belongs to the field of cognitive wireless sensor networks, relates to a cognitive allocation method for multi-user channels in a wireless network, and can be used in related fields involved in cognitive sensor networks. Background Art

[0002] Cognitive Radio Sensor Networks (CRSN) introduce cognitive radio (CR) technology into wireless sensor networks, so that each sensor node has the ability to sense spectrum and dynamically access spectrum. With the increasingly limited development of spectrum resources, in order to ensure that the primary user (PU) uses the spectrum first, it is necessary to impose necessary spectrum restrictions on network nodes as secondary users (SU). Based on the CR technology, CRSN can effectively reduce the delay and energy consumption caused by channel collision and competition compared to traditional wireless sensor networks (WSN) under spectrum constraints. SU has the right to occupy the spectrum only when it recognizes that the spectrum is idle. At present, spectrum cognition can use energy detection (ED), matched filter detection (MFD) and cyclic feature detection (CFD).

[0003] The low system throughput is a reflection of CRSN being limited by spectrum. A reasonable spectrum allocation scheme can significantly increase the system throughput. According to different computing entities, spectrum allocation methods in CRSN can be divided into centralized and distributed. The centralized scheme relies on the base station to collect data from each node to allocate the global spectrum. The distributed allocation scheme generally takes a cluster or node as the main body, exchanges the collected data with the neighboring cluster or node and selects the spectrum by itself. Based on the difference in PU spectrum occupancy, spectrum allocation can also be divided into two forms: fixed and dynamic. The former is mainly used in networks with rich spectrum resources. The spectrum is allocated to nodes that are unlikely to interfere with the primary user. The nodes that have been allocated the spectrum will not change in the short term. At present, many researchers have conducted research on it based on game theory, graph coloring, group optimization algorithms and other directions. The latter is a dynamic spectrum allocation scheme for scenarios with extremely limited spectrum resources. Usually, several spectrum segments are allocated to a node, and various intelligent algorithms are used to calculate the time usage segments of the spectrum. For scenarios with fixed PU positions, in order to control the global spectrum allocation without loss of generality and improve network throughput, we focus on centralized fixed spectrum allocation schemes.

[0004] CRSN is limited by the number of spectrum and routing. The limited number of spectrum reduces the flexibility of SU communication and also brings hidden dangers of network hot spots and routing multi-branch. Routing is the basis of fixed spectrum allocation, and traditional WSN routing is affected by the random distribution of nodes and spectrum, and cannot meet the throughput and delay requirements of CRSN network. Therefore, traditional WSN routing is not suitable for direct use in CRSN. Therefore, it is crucial to build routing and spectrum allocation solutions while taking into account network throughput and communication restrictions.

[0005] In view of the adverse effects of spectrum limitations and routing difficulties on the throughput of CRSN systems, a cognitive spectrum allocation scheme based on the Improve Swarm optimization method based on Leapfrog Game (ISLG) is proposed. Adaptive adjustment routing is used to optimize network load and solve the network hotspot and local multi-branch problems caused by unreasonable node distribution. On the basis of this routing, ISLG is used to search for the best spectrum allocation scheme, and the allocation group of the basic optimization algorithm is re-moved through the Leapfrog Game (LG) to find the best spectrum allocation scheme. Through simulation comparative analysis, LG can significantly enhance the search performance of various group optimization algorithms. ISLG also has higher performance in solving spectrum allocation problems than the unimproved algorithm, which can effectively improve the total throughput of the system. Summary of the invention

[0006] In view of the above technical problem, the present invention provides a cognitive allocation method for multi-user channels in a wireless network.

[0007] The technical solution of the present invention:

[0008] A cognitive allocation method for multi-user channels in a wireless network, comprising the following steps:

[0009] Step 1: Initialize the cognitive wireless sensor network model;

[0010] Step 2: Node clustering based on adaptive PU number routing, including routing establishment and node clustering;

[0011] Step 3: Node recognition detection of the channel;

[0012] Step 4: Establishment of network throughput problem;

[0013] Step 5: Spectrum allocation based on leapfrog game optimization algorithm.

[0014] The specific process of step one is:

[0015] Given a cognitive wireless sensor network CRSN with M primary users PU and N secondary users SU randomly distributed, where PU = {pu m |m=1,2,…,M},SU={su n |n=1,2,…,N}. Based on the principle of frequency division multiplexing and PU equality, the spectrum resources in the network are evenly divided into M channels C={c m |m=1,2,…,M} and is allocated to each PU accordingly, where the PU signal power conforms to the normal distribution and has a mean of 0.

[0016] In order to meet the requirement of fixed node channels, all nodes in unidirectional multi-hop routing have to use the same designated channel, which reduces spectrum utilization and increases network conflicts. Therefore, SU nodes adopt a working mode that distinguishes between receiving and sending, and divides the sending and receiving work of a node to two antennas to achieve the expansion of fixed spectrum throughput and balance the mutual interference and conflict waste under a single channel.

[0017] Based on the known PU location and communication information, the behavior of each PU occupying the channel in the CRSN network can be compared to a Markov process.

[0018] For each time slot of length T, SU divides each time slot into two phases: cognition and transmission. The first τ time of each time slot is the cognition phase, and the remaining T-τ time is the transmission phase. m Occupied channels c m , and p m and q m Respectively represent c m The probability that the channel status changes from idle to occupied and from occupied to idle in the current time slot. Only when the channel is idle can the SU transmit opportunistically, so the channel c m Theoretical transmission ratio D m As shown in formula (1).

[0019]

[0020] The specific process of step 2 is:

[0021] (1) Establishment of triangulation hierarchical adaptive PU number routing APRT

[0022] In CRSN, network routing has a great impact on throughput. Therefore, Adaptive PU Quantity Routing Based on Triangulation Layering (APRT) is proposed. The generation of APRT routing can be divided into two steps: layering and allocation.

[0023] In the layering stage, all SU nodes in the network, including the base station, will be triangulated, and then the network will be layered with the base station as the center. The layering logic is that the nodes with direct edge connection to the base station are the first-layer nodes. Since the base station can receive all channel data at the same time, the hot zone nodes do not have to be limited to reuse one channel. Therefore, in order to maximize the throughput of the hot zone nodes, the following restrictions are made: the number of nodes in the first layer is limited to not more than the number of channels M, so the nodes in the first layer are sorted according to the Euclidean distance from the base station, and the nodes with long distances will be directly placed in the second layer. The nodes in the subsequent layers follow the principle that the nodes with direct edges to the nodes in the i-th layer are the i+1-th layer nodes, and a node will not be in multiple layers.

[0024] In the allocation phase, the up-hop node and down-hop node of each node will be determined. A node can only have one down-hop node, but can have multiple up-hop nodes, which is the generation of multiple branches in the route. Multiple branches will cause too many nodes to share a channel, which will significantly reduce the throughput of related nodes. In order to make the distribution of branches and leaves in the route more balanced, the APRT network is set to generate at most two branches, in exchange for the increase in the length of routing branches and leaves in exchange for the reduction of routing branches. The allocation process is described as follows: select the routing layers in order from small to large, traverse the i-th layer, and select up to two nodes in the i+1th layer as up-hop nodes for the nodes according to the Euclidean distance. If there are still nodes left in i+1 after the traversal, put the remaining nodes in i+1 into the i+2 layer.

[0025] Considering that triangulation has the characteristic of maximizing the minimum angle, it can be applied to the formation of network routing, which can reduce the generation of obtuse angles in the network and thus reduce the generation of long edges, and can reduce the probability of long-distance single-hop paths in routing. Therefore, an Adaptive PU Quantity Routing Based on Triangulation Layering (APRT) is proposed. Based on the characteristics of maximizing the minimum angle, it limits routing scheduling and increases the rationality of distance and space when nodes select links, thereby alleviating the generation of local conflicts in the network.

[0026] The base station can receive data from all channels at the same time, so the hot zone nodes do not have to be limited to multiplexing one channel, that is, the channels that the hot zone nodes can use to reach the base station in a single hop can be different from each other. If the number of nodes in the APRT that can reach the base station in a single hop is strictly limited to be less than or equal to the number of channels in the network, the condition that the up-hop channels selected by the hot zone nodes are different from each other can be met, which ensures the throughput performance of each node in the hot zone and can also avoid hot zone conflicts. Since the number of channels is fixed, under the above conditions, only the number of hot zone nodes can be limited.

[0027] Thus, the rules for limiting the number of multi-branch nodes and hot zone nodes in APRT are shown in equations (3) and (4).

[0028]

[0029]

[0030] Among them, A={a k =|l k |-1|k=1,2,...K} is the matrix of the number of up-hop nodes in the cluster corresponding to L; |·| represents the number of elements in the set. After establishing the triangulation hierarchical adaptive PU number routing, node clustering can be performed.

[0031] (2) Node clustering

[0032] The node distinguishes between receiving and sending, which can remove the transmission channel selection restriction of its down-hop node while ensuring that the sending channel is the same as the receiving channel of its down-hop node. The receiving and sending channels of SU are independent channels that are different from each other. Therefore, the cognitive wireless sensor network CRSN as a whole presents a clustering structure with channels as distinguishing marks, which realizes the binding of some nodes and helps to allocate channels. The clusters formed can have the same nodes, but except for the cluster with the base station as the down-hop node, no two clusters have the same down-hop node.

[0033] After the nodes in the cluster are bound, the original node channel allocation problem can be converted into a cluster channel allocation problem. In order to reduce inter-cluster interference, different channels should be allocated as much as possible between adjacent clusters in the channel allocation. Obviously, after the route is determined, each cluster can also be determined accordingly. Then the set L consisting of the existing K clusters can be described as follows:

[0034]

[0035] Where, k = 1, 2, ..., K; l k (0) is cluster l k The only next hop node; l k (i)∈SU,1≤i≤K is cluster l k The up-hop node in , BS is the base station.

[0036] like Figure 1 Among them, su1, su2, and su3 form a cluster, which includes multiple up-hop nodes {su2, su3} and one down-hop node su1, where each up-hop node shares the channel based on the frequency division multiplexing principle.

[0037] The specific process of step three is:

[0038] Energy detection is the most common channel occupancy detection method. By counting the energy of the PU signal over a period of time and comparing it with a preset threshold, the PU's occupancy of the spectrum can be determined. In this patent, the PU's duty cycle can be obtained by integrating the cognitive information of each node on the PU at the base station.

[0039] The cognitive transmission model of nodes is as follows: Figure 2 As shown, SU samples the channel energy N times within the duration τ T Second, the PU’s transmission signal x PU If it obeys the normal distribution, then the signal x in the corresponding channel received at the SU also obeys the normal distribution. Since the SU has cognitive ability, it can calculate the signal strength variance δ of the received PU x 2 According to the known normal distribution of network noise variance δ x 2 , we can get n Cognition m Detection probability Q n,m As shown in formula (5).

[0040]

[0041] γ=βN T +(1-β)N T (SNR+1) (6)

[0042] Where Q(·) represents the right tail function of the normal distribution; β is the detection factor, and its value range is [0,1]; SNR = δ x 2 / δ w 2 .

[0043] It can be calculated that su n Using c m The theoretical transmission time ratio of the channel is g n,m As formula (7)

[0044] g n,m =D m Q n,m (7)

[0045] The specific process of step 4 is as follows:

[0046] When su n Detected pu m The closer the distance is, the higher the received signal strength is. n In c m The channel capacity e n,m And su n C m Cognitive success rate Qn,m The higher the transmission rate, the higher the transmission rate. Considering the time separation between [0,τ] and [τ,T] in each time slot, there will be no inter-node transmission interference in the [0,τ] time period. However, in the [τ,T] time period, nodes that choose the same channel will inevitably cause inter-node transmission interference, reducing system throughput.

[0047] su n In c m The channel capacity e n,m As shown in formula (8).

[0048]

[0049] In the formula, l k It is node su n As the up-hop node corresponding to the cluster; a n =|l k |-1 is cluster l k The number of nodes in the middle jump; 1 / (a n a i ) is the node su n with su i The expected ratio of channel overlap; B is the channel bandwidth; r n,i is the node su in the same channel i To su n The interference can be caused by the node su n Pre-perceived.

[0050] Node su n In channel c m The throughput on b n,m It can be calculated as follows,

[0051] b n,m =g n,m ×e n,m (9)

[0052] In order to ensure the overall throughput balance of the network, the throughput requirements of nodes that forward more data must be met first. For this reason, each node is set to have a throughput weight w in the calculation of the global benefit. n , in order to improve the channel allocation priority of nodes with high data volume in the algorithm, and its value is proportional to its expected data transmission volume. n Characterized by the ratio of the amount of data to be sent to the amount of data itself, there is a node weight matrix W = [w1,w2...w n ... N ].

[0053] Due to the routing binding effect, the transmission channel allocation for a single SU can be converted into the transmission channel allocation for the up-hop node in the cluster, so the channel allocation matrix XX = {x k |1≤xk ≤M,k=1,2,...K}, where x k for l k The selected channel number, i.e. for l k Select a channel.

[0054] The problem of maximizing the global throughput benefit U under the channel selection constraints of each cluster can be transformed into equations (10) and (11).

[0055]

[0056]

[0057] The specific process of step five is:

[0058] Since the nodes are clustered, the channel allocation problem for nodes is converted into the channel allocation problem for clusters, which greatly reduces the complexity of the problem.

[0059] The Leapfrog Game (LG) inherits the strategies and ideas of the Shuffled Frog Leaping Algorithm (SFLA). It mainly emphasizes the pseudo-randomness of the players and strategies. The pseudo-randomness can increase the disorder of the game while ensuring that different initial solutions have a unique Nash equilibrium.

[0060] Since each strategy selection must be in accordance with the player's only strategy that is better for himself and does not make other players worse, then in a finite round of games, it will converge to the Pareto optimal. This allows LG to reduce the complexity of the spectrum allocation problem while obtaining better global search capabilities. LG based on Pareto optimality can limit the complexity of the algorithm while ensuring the effect.

[0061] When using LG for channel allocation, LG can treat each cluster as a frog, and the channel represents the stone that can be jumped on. The cluster throughput gain lb k,m The frog l k In stone m mood. Each frog has its own independent random jump sequence f k , this sequence can increase the randomness of frog jumping and the discreteness of the solution after frog jumping, which reduces the probability of channel selection conflict when frog jumps under the same jumping order. For a channel allocation, each cluster l k The random jump sequence f k The following jump matrix F can be formed:

[0062]

[0063] Among them, f k The internal elements must not be repeated, and rand(M) is a random integer in the range [1,M] corresponding to the channel number.

[0064] In each round of the game, the cluster with the worst benefits is selected according to the random jump sequence f k The judgment is made in sequence. When the selected channel meets the Pareto improvement condition, the current round stops and a cluster is selected for the cycle. The algorithm stops when it converges to the Pareto optimality.

[0065] If only LG is used for channel allocation, the effect of channel allocation will be limited to the initial solution. Therefore, an optimization algorithm based on Leapfrog Game (ISLG) is proposed, which optimizes LG by introducing a swarm optimization algorithm. Although the standard swarm optimization algorithm is difficult to find the global optimal solution in such high-dimensional discrete problems and is very likely to fall into a local optimal solution, it can complement each other's advantages when combined with LG.

[0066] In view of the above problems, an improved swarm optimization method (ISLG) based on leapfrog game is proposed to find the optimal spectrum allocation scheme. Through LG, each allocation set in the swarm optimization algorithm is re-moved to achieve low-complexity re-optimization in a single swarm optimization iteration. The re-movement only takes the benefits of each allocation set after executing LG as its benefit, and does not move the position of the allocation set. In the optimization process, if the discrete particle swarm algorithm DPSO is improved based on LG, the particle with the best benefit can update its position to the position after LG, so that the algorithm can achieve better results.

[0067] Beneficial effects of the invention: This patent provides a cognitive allocation method for multi-user channels in a wireless network. The patent generally includes APRT route generation and ISLG spectrum allocation algorithm. In CRSN, the initialization of network routing based on APRT can significantly increase the throughput of hot zone nodes. At the same time, since APRT limits the problem of multi-branch nodes compared to flooding routing and hierarchical routing, it also improves the throughput of some nodes. The spectrum allocation problem in CRSN can be regarded as a graph coloring problem. Since the complexity of the spectrum allocation problem in the network increases exponentially with the number of M, ordinary group optimization algorithms such as discrete particle swarm optimization and discrete gray wolf algorithm are difficult to obtain good results when the number of M in the network is large. LG can obtain a better solution with less complexity when the number of M is large. Therefore, ISLG improves various group optimization algorithms based on LG, which can complement the advantages of the two. When the number of M is small, the group optimization algorithm is dominant, and when the number of M is large, the algorithm is dominated by LG. Therefore, in the spectrum allocation problem, the ISLG algorithm can obtain better results under any M than various group optimization algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is the frequency division multiplexing diagram under the cluster, where su is the secondary user node in the network, and circles with different borders represent the channels used by different clusters.

[0069] Figure 2 This is a diagram of the SU cognitive transmission model, where the state transition model of the PU is a Markov model, and the cognitive state model of the SU is generated based on the state transition model of the PU.

[0070] Figure 3 The following is a comparison chart of the throughput benefits of APRT routing and other common routing.

[0071] Figure 4 The performance comparison chart of LG and common group optimization algorithms under different numbers of PUs.

[0072] Figure 5 The performance comparison chart of common group optimization algorithms before and after using LG improvement under different PU numbers.

[0073] Figure 6 The figure is a comparison chart of the performance of different algorithm optimization schemes under different channel models. DETAILED DESCRIPTION

[0074] This embodiment provides a method for cognitive allocation of multi-user channels in a wireless network.

[0075] In order to evaluate the performance of the scheme, a simulation experiment was established using MATLAB. The cognitive spectrum allocation scheme based on the leapfrog game optimization algorithm was analyzed. First, the adaptability of APRT and the other three commonly used wireless sensor routings was analyzed. Then, various ISLG algorithms were compared with their standard algorithms, and the advantages and disadvantages of each algorithm were compared by analyzing the network throughput benefits.

[0076] Assume that the network area is a square, the BS is in the center, the SU and PU are randomly scattered in the area, and the channel adopts a free space model. In the comparison of ISLG, the following basic swarm optimization algorithms are selected for optimization: discrete particle swarm algorithm DPSO (see J. Kennedy and RCEberhart. "Particle swarm optimization." In IEEE international Conference on Neural Networks, volume 4, IEEE Press, 1995, pp. 1942–1948.) and discrete grey wolf algorithm DGWO (see Mirjalili S, Mirjalili SM, Lewis A. Grey Wolf Optimizer [J]. Advances in Engineering Software, 2014, 69: 46-61.). The optimized ISLG algorithms are the improved DPSO based on LG (DPLG) and the improved DGWO based on LG (DGLG).

[0077] First, in order to verify the effectiveness of APRT routing, the number of PUs M is used as a variable. Under the same M, different network snapshots are set, and DPLG is used to allocate channels. Each round of the algorithm is run 50 times to obtain the average value.

[0078] The simulation results are as follows Figure 3 As shown in the figure, it can be seen that when M gradually increases, the global throughput benefits of the four routing methods all show an upward trend and gradually slow down, among which the effect of ARPT is always better than the other three types. In the hot zone restricted segment of 4≤M≤5, it can be clearly seen that the benefits of the four types of routing algorithms are extremely poor compared to the global ones. Although APRT is also limited to the selection of the number of channels, since there is no problem of hot zone conflict, its curve has a gentler rising gradient in 4≤M≤5 than THR, and the starting point is also higher.

[0079] To analyze the performance of LG in channel allocation, the LG, DPSO and DGWO algorithms are used to allocate channels on the same network snapshot and the performance is compared with M as a variable. The initial particles and wolf pack numbers of DPSO and DGWO are set to 1000, and LG, DPSO and DGWO are simulated 50 times under the same M and the average value is taken.

[0080] The simulation results are as follows Figure 4 As shown in the figure, it can be seen that the three algorithms generally show an upward trend with the increase of M, but the upward trend of DPSO and DGWO is relatively slow, and even declines to varying degrees. LG is more stable, and its throughput benefit U is better than DPSO and DGWO after M≥9. At this time, the limitation of frog-leaping channel selection in LG is reduced, so the higher the number of M, the higher its benefit.

[0081] In order to analyze the performance of the ISLG algorithm in channel allocation, the following simulation is performed: the ISLG-type algorithms DPLG and DGLG are simulated with the DPSO and DGWO algorithms before improvement. Under the same network snapshot, different M are set to simulate and compare the two swarm optimization algorithms before and after optimization. In order to reflect the fairness of time complexity, under the premise of setting the number of particles of the ISLG-type algorithm to num=10, the initial number of particles of the algorithm before optimization is set to be equal to the product of the number of LG calculations and num, and the maximum number of iterations is 200. It is worth noting that all four algorithms have achieved convergence within 200 iterations. The two algorithms before and after improvement were simulated 50 times under different M and the average value was taken. The results are shown as follows Figure 5 shown.

[0082] Figure 5 In the figure, although the overall effects of the four algorithms increase with the increase of M, DPSO and DGWO are obviously not as stable as ISLG algorithms, but increase in a fluctuating manner. This proves that ISLG can inherit the advantages of LG to make up for the shortcomings of group optimization algorithms.

[0083] In general, when M>4, the ISLG algorithm performs better than the algorithm before improvement. As can be seen from the figure, DPSO performs better than DGWO, and DPLG also performs better than DGWO overall. This is consistent with the expected effect, that is, the basic group optimization algorithm inherited by ISLG will also affect its effect. When the number of PUs is high, the effects of DPLG and DWLG have tended to be synchronized, because the dominant role in the ISLG algorithm has changed from the basic group optimization algorithm to LG. In this way, frog jumping can improve the poor performance of the group optimization algorithm in high-dimensional situations, and the group optimization algorithm can indirectly improve the ability of the frog jumping game to jump out of the local optimum. Therefore, ISLG can ensure that the complexity is low while also having better performance in channel allocation.

[0084] In order to analyze the effect of the algorithm in different channels without losing fairness, the Ricean fading model and the Logmormal fading model are simulated and analyzed. The k factor in the Ricean fading model is set to 10, and the mean and variance of the Logmormal fading model are set to 1 and 2, respectively.

[0085] For the two channel models, OSPF with DPSO and APRT with DPSO were used for simulation. For different M, the channel gain was recalculated each time. The algorithm was simulated 50 times and the average value was taken. The results are shown in the figure. Figure 6 shown.

[0086] The results show that the algorithm of the present invention is not only applicable to the ideal free space model, but also to the normal logarithmic model of large-scale fading and the Rice model of small-scale fading. Since different M correspond to different channel indices, the global benefit after the algorithm is not positively correlated with M.

Claims

1. A method for cognitive allocation of multi-user channels in a wireless network, characterized in that: Here are the steps: Step 1: Initialize the cognitive wireless sensor network model; Step 2: Node clustering based on adaptive PU number routing, including routing establishment and node clustering; Suppose there are M primary users PU and N secondary users SU randomly distributed in a cognitive wireless sensor network CRSN, where PU = {pu m |m=1,2,…,M},SU={su n |n=1,2,…,N}; When LG is used for channel allocation, LG treats each cluster as a frog l k ; k = 1, 2, ..., K, K clusters constitute the set L; The establishment of triangulation hierarchical adaptive PU number routing APRT is: In CRSN, network routing has a great impact on throughput, so triangulation hierarchical adaptive PU number routing APRT is proposed. The generation of APRT routing can be divided into two steps: hierarchical and allocation: Layering stage: All SU nodes in the network, including base stations, will be triangulated, and then the network will be layered with the base station as the center; the layering logic is that the nodes with direct edge connection to the base station are the first-layer nodes. Since the base station can receive all channel data at the same time, the hot zone nodes do not have to be limited to reuse one channel. Therefore, in order to maximize the throughput of the hot zone nodes, the following restrictions are made: the number of nodes in the first layer is limited to not more than the number of channels M, so the nodes in the first layer are sorted according to the Euclidean distance from the base station, and the nodes with long distances will be directly placed in the second layer; the nodes in the subsequent layers follow the principle that the nodes with direct edges to the nodes in the i-th layer are the i+1-th layer nodes, and a node will not be in multiple layers; Allocation phase: Select routing layers in ascending order, traverse the i-th layer, and select at most two i+1-th layer nodes as up-jump nodes for the nodes according to the Euclidean distance; The rules for limiting the number of multi-branch nodes and hot zone nodes in APRT are shown in equations (3) and (4): Among them, A={a k =|l k |-1|k=1,2,...K} is the matrix of the number of up-hop nodes in the cluster corresponding to L; |·| represents the number of elements in the taken set. After establishing the triangulated hierarchical adaptive PU number routing, node clustering can be performed; Step 3: Node recognition detection of the channel; Step 4: Establishment of network throughput problem; Step 5: Spectrum allocation based on leapfrog game optimization algorithm; The specific process is as follows: Since the nodes are clustered, the channel allocation problem for nodes is transformed into the channel allocation problem for clusters; Each strategy selection must be in accordance with the principle that the player can only make a strategy that is better for himself and does not make other players worse. Then, under a finite round of games, it will converge to the Pareto optimality. This allows the leapfrog game LG to reduce the complexity of the spectrum allocation problem while obtaining good global search capabilities. The LG based on the Pareto optimal theory can limit the complexity of the algorithm while ensuring the effect. When using LG for channel allocation, LG treats each cluster as a frog, and the channel represents the stone it can jump on. The cluster throughput gain lb k,m The frog l k In stone m mood; each frog has its own independent random jump sequence f k , this sequence can increase the randomness of frog jumping and the discreteness of the solution after frog jumping, thereby reducing the probability of channel selection conflict when frogs jump under the same jumping order; for a channel allocation, each cluster l k The random jump sequence f k The following jump matrix F can be formed: Among them, f k The internal elements must not be repeated, and rand(M) is a random integer from [1,M] to correspond to the channel number; In each round of the game, the cluster with the worst benefits is selected according to the random jump sequence f k The judgment is made in sequence. When it is judged that the selected channel meets the conditions of Pareto improvement, the current round stops and a cluster is selected for circulation. When it converges to Pareto optimality, the algorithm stops. If only LG is used for channel allocation, the effect of channel allocation will be limited to the initial solution. Therefore, an optimization algorithm ISLG based on frog leaping game is proposed, which optimizes LG by introducing a group optimization algorithm. Although the standard group optimization algorithm is difficult to find the global optimal solution in such high-dimensional discrete problems and is very likely to fall into a local optimal solution, it can complement each other with LG. ISLG re-moves each allocation set in the swarm optimization algorithm through LG to achieve low-complexity re-optimization in a single swarm optimization iteration; the re-movement only takes the benefits of each allocation set after executing LG as its benefit, and does not move the position of the allocation set; during the optimization process, when the discrete particle swarm algorithm DPSO is improved based on LG, the position of the particle with the best benefit can be updated to the position after LG, so that the algorithm can achieve better results.

2. The cognitive allocation method for multi-user channels in a wireless network according to claim 1, characterized in that: The specific process of step one is: Suppose there are M primary users PU and N secondary users SU randomly distributed in a cognitive wireless sensor network CRSN, where PU = {pu m |m=1,2,…,M},SU={su n |n=1,2,…,N}; Based on the principle of frequency division multiplexing and PU equality, the spectrum resources in the network are evenly divided into M channels C={c m |m=1,2,…,M}, and is allocated to each PU accordingly, where the PU signal power conforms to the normal distribution and the mean is 0; the SU node adopts a working mode that distinguishes between reception and transmission, and divides the transmission and reception work of a node to two antennas; Based on the known PU location and communication information, the behavior of each PU occupying the channel in the CRSN network is analogized to a Markov process; For each time slot of length T, SU divides each time slot into two phases: cognition and transmission. The first τ time of each time slot is the cognition phase, and the remaining T-τ time is the transmission phase. m Occupied channel c m , and p m and q m Respectively represent c m The probability that the channel status changes from idle to occupied and from occupied to idle in the current time slot; when the channel is idle, the SU can transmit opportunistically, so the channel c m Theoretical transmission ratio D m As shown in formula (1):

3. The cognitive allocation method for multi-user channels in a wireless network according to claim 2, characterized in that: The node clustering process is: The node distinguishes between receiving and sending, while ensuring that the sending channel is the same as the receiving channel of its down-hop node, and removes the restriction on the selection of the sending channel of its down-hop node; the receiving and sending channels of SU are independent channels that are different from each other, so the cognitive wireless sensor network CRSN as a whole presents a clustering structure with channels as distinguishing marks, realizing the binding of some nodes, and the clusters formed can have the same nodes, but except for the cluster with the base station as the down-hop node, no two clusters have the same down-hop node; After the nodes in the cluster are bound, the original node channel allocation problem is converted into a cluster channel allocation problem. In the channel allocation, different channels should be allocated between adjacent clusters as much as possible. After the route is determined, each cluster is determined accordingly. The set L consisting of the existing K clusters can be described as follows: Where k = 1, 2, ..., K; l k (0) is cluster l k The only next hop node; l k (i)∈SU,1≤i≤K is cluster l k The up-hop node in the BS is the base station; each up-hop node shares the channel based on the frequency division multiplexing principle.

4. The cognitive allocation method for multi-user channels in a wireless network according to claim 3, characterized in that: The specific process of step three is: The duty cycle of PU is obtained by integrating the cognitive information of each node on PU at the base station; Node cognitive transmission model: SU samples the channel energy N times within the duration τ T Second, the PU’s transmission signal x PU If it obeys the normal distribution, then the signal x in the corresponding channel received at the SU also obeys the normal distribution. Since the SU has cognitive ability, the SU can calculate the signal strength variance δ of the received PU x 2 :According to the known normal distribution of network noise variance δ w 2. Get su n Cognition m Detection probability Q n,m As shown in formula (5): γ=βN T +(1-β)N T (SNR+1) (6) Where Q(·) represents the right tail function of the normal distribution; β is the detection factor, and its value range is [0,1]; SNR = v x 2 / δ w 2 ; Calculate su n Using c m The theoretical transmission time ratio of the channel is g n,m As shown in formula (7): g n,m =D m Q n,m (7)。 5. The cognitive allocation method for multi-user channels in a wireless network according to claim 4, characterized in that: The specific process of step 4 is as follows: When su n Detected pu m The closer the distance is, the higher the received signal strength is. n In c m The channel capacity e n,m And su n C m Cognitive success rate Q n,m The higher the value, the higher the transmission speed. Considering the time separation between [0,τ] and [τ,T] in each time slot, there will be no inter-node transmission interference in the [0,τ] time period. However, in the [τ,T] time period, there will inevitably be inter-node transmission interference between nodes that choose the same channel, which will reduce the system throughput. su n In c m The channel capacity e n,m As shown in formula (8): In the formula, l k It is node su n As the up-hop node corresponding to the cluster; a n =|l k |-1 is cluster l k The number of nodes in the middle jump; 1 / (a n a i ) is the node su n with su i The expected ratio of channel overlap; B is the channel bandwidth; r n,i is the node su in the same channel i To su n The interference can be caused by the node su n Pre-perceived; Node su n In channel c m The throughput on b n,m It can be calculated as follows, b n,m =g n,m ×e n,m (9) In order to ensure the overall throughput balance of the network, the throughput requirements of multiple nodes forwarding data must be met first; for this purpose, each node is set to have a throughput weight w in the calculation of the global benefit. n , improve the channel allocation priority of nodes with high data volume, w n The value is proportional to w n The expected data transmission volume; w n Characterized by the ratio of the amount of data to be sent to the amount of data itself, there is a node weight matrix W = [w1,w2...w n ... N ]; Due to the binding effect of routing, the transmission channel allocation for a single SU is converted into the transmission channel allocation for the up-hop node in the cluster, and the channel allocation matrix X is: X = {x k |1≤x k ≤M,k=1,2,...K}, where x k for l k The selected channel number, i.e. for l k Select channel; The problem of maximizing the global throughput benefit U under the channel selection constraints of each cluster can be transformed into equations (10) and (11):

Images