Optimization method for maximizing energy efficiency in cognitive drone edge computing networks

Abstract

The present invention belongs to the field of radio communication technology, specifically, it is an optimization method for maximizing energy efficiency in a cognitive drone edge computing network, which uses drones as both cognitive users and source nodes for joint spectrum perception, and as relay nodes for auxiliary computing and data transmission. Under information causality constraints, interference power constraints, and shared data bit constraints, the energy efficiency in the cognitive drone edge computing network is taken as the optimization goal, and a particle swarm algorithm is used to update and iterate the speed and position, optimize the perception time, computing resources, and communication resources, and obtain the optimal perception time, data bit amount, and central processing unit frequency, so as to maximize the energy efficiency in the cognitive drone edge computing network. The present invention can improve the energy efficiency of the edge computing network system.

Description

Technical Field

[0001] The present invention belongs to the field of radio communication technology. Specifically, it is an optimization method for maximizing energy efficiency in a cognitive drone edge computing network. A particle swarm algorithm is used to search for the best allocation of communication resources, computing resources and perception time, so that the energy efficiency of the cognitive drone edge computing network reaches the maximum value. Background Art

[0002] Mobile Edge Computing (MEC) technology is one of the key technologies of 5G and B5G. Its core principle is that users can offload computing tasks to edge servers with powerful computing capabilities through wireless links. Due to the scarcity of spectrum resources and the explosive growth of mobile devices, it is unrealistic to allocate independent spectrum resources to all users. Therefore, spectrum resources are crucial in edge computing networks.

[0003] Cognitive radio is a dynamic spectrum sharing technology that can effectively alleviate the problem of scarce spectrum resources in edge computing networks. In cognitive radio networks, secondary users (unauthorized users) can access the primary user frequency band under the condition that the interference temperature constraint of the primary user (authorized user) is met (that is, the interference power to the primary user cannot exceed a preset threshold). Obviously, cognitive radio technology can be applied to mobile edge computing to provide secondary users with spectrum access opportunities for task offloading. Energy detection is easy to implement and does not require any prior information about the primary user. It has become a widely used spectrum sensing method, but its sensing performance will be affected by hidden terminal problems such as shadow effects and multipath fading. To overcome the hidden terminal problem, collaborative spectrum sensing effectively improves the system spectrum sensing performance by fusing the local sensing information of multiple secondary users in different sensing environments and making a comprehensive decision.

[0004] Generally speaking, the deployment location of MEC servers is fixed, and the communication link between the server and the user is usually non-line-of-sight, with a small coverage range. Drones are highly maneuverable and flexible to deploy. There is usually a line-of-sight link between the MEC server mounted on a drone and the user. Therefore, deploying MEC servers on drones can provide ground users with flexible and wide-coverage services. In addition, they have certain computing and processing capabilities, which can realize energy-efficient, low-latency auxiliary processing of computing tasks for terminal devices.

[0005] The particle swarm optimization algorithm simulates the process of a flock of birds searching for food based on swarm intelligence. Through the information sharing characteristics within the swarm, the birds cooperate with each other and finally move to the location of the food. The so-called particles are each individual in the flock of birds in the search space, and they all propose a feasible path to find food. During the operation of the algorithm, the particles constantly compare their current positions with their own historical optimal positions (individual optimal values) and the historical optimal positions of the swarm (group optimal values), and then use the results as the basis for adjusting their speed and position. The state of each particle in the swarm is brought into the fitness function, and multiple iterations are performed to eventually make the particles reach the optimal position, that is, to obtain the optimal solution to the optimization problem.

[0006] The operation procedure of the particle swarm algorithm mainly includes the following steps: 1) Initialization, setting the size, initial position, and initial speed of the particles; 2) Calculating the objective function of each particle, finding the current extreme value of each particle, and finding the current global optimal solution of the entire particle swarm; 3) Updating the speed and position of each particle; 4) Before reaching the maximum number of iterations, it is necessary to loop the above steps 2) and 3) to finally obtain the optimal solution.

[0007] Intelligent applications are often computationally intensive and sensitive to latency. However, most mobile terminal devices have limited computing power and power reserves and cannot independently meet current needs. MEC solves this problem well. By deploying cloud computing and information technology services to the edge of the network and providing auxiliary computing, it can effectively reduce task processing latency, avoid network congestion, and increase terminal battery life. At the same time, there is a problem of scarce spectrum resources in edge computing networks, which can also be effectively alleviated by cognitive radio technology. By allowing secondary users to access the unused legally authorized frequency bands of primary users to improve the utilization rate of existing spectrum resources. Obviously, cognitive radio technology can be applied to mobile edge computing to provide secondary users with spectrum access opportunities for task offloading. By deploying edge computing servers on highly maneuverable drones, flexible MEC services can be provided to secondary users in cognitive radio networks when they perform task offloading or data transmission. Optimizing the energy efficiency of cognitive drone edge computing networks, that is, finding the best allocation of communication resources, computing resources, and perception time for cognitive drone edge computing networks, the particle swarm algorithm can directly search for the optimal data bit amount, central processing unit frequency, and perception time value, solving the energy efficiency optimization problem that belongs to non-convex optimization. Summary of the invention

[0008] The present invention discloses an optimization method for maximizing energy efficiency in a cognitive drone edge computing network. First, the distribution of communication resources, computing resources, and perception time in the cognitive drone edge computing network is regarded as a population. Then, based on boundary conditions, an initialized population is obtained. Then, the particle swarm is iteratively updated until the maximum number of iterations is reached. In this way, the optimal communication resources, computing resources, and perception time distribution can be found, thereby maximizing the energy efficiency of the system.

[0009] The specific technical solution adopted by the present invention is as follows:

[0010] An optimization method for maximizing energy efficiency in a cognitive drone edge computing network includes the following steps:

[0011] Step S1: Determine the spectrum hole probability

[0012] The collaborative spectrum sensing process is as follows: in the T1 time slot, the source node S and the UAV relay respectively perform local energy detection on the received signal from the main transmitter PT, and then the UAV relay transmits the local sensing result (PT existence or non-existence, i.e. 1 or 0) to the source node S through a dedicated control channel. The source node S uses the "OR" fusion rule to fuse the UAV relay's sensing result and the local sensing result and make the final decision. The sensing time slot can be divided into two sub-time slots t s and t r In the first sub-time slot, the source node S and the drone relay perform spectrum sensing respectively. In the second sub-time slot, the drone relay transmits the sensing result to the source node S, which integrates and makes the final decision. Since the sensing result reporting time is t r It is very short and can be ignored.

[0013] Assume that y(n) is the PT transmission signal received by the source node S or the drone relay in T1. According to the binary hypothesis, the received signal when the PT exists (represented by H1) or does not exist (represented by H0) is as follows:

[0014]

[0015] Among them, x p (t) is the power P p The primary user transmits a signal, n(t) is subject to N(0, σ 2 ) distributed Gaussian white noise, h ST is the channel gain of the PT-S link (if it is a PT-UAV link, the channel gain is h rT ). M=t s f s , is the number of sampling points, f sis the sampling frequency. Local sensing uses energy detection, and the energy statistics of the local sensing signal of S (or drone relay) can be expressed as

[0016]

[0017] The local false alarm probability P of the source node S (or drone relay) under energy detection f and the local detection probability P d It can be expressed as:

[0018]

[0019]

[0020] Among them, Q(@) is the standard Gaussian complementary distribution function: η is the decision threshold of the local energy detector of the source node S and the drone relay, and γ is the average signal-to-noise ratio of the authorized user signal received by S (or drone relay):

[0021] The source node S uses the “OR” fusion rule to fuse the local decision results of S and the drone relay. Therefore, the final cooperative false alarm probability P of the entire system is fa and cooperative detection probability P de as follows:

[0022] P fa =1-(1-P f ) 2 (1.5)

[0023] P de =1-(1-P d ) 2 (1.6)

[0024] From this we can get:

[0025]

[0026] There are two cases for the source node S (or drone relay) to transmit data:

[0027] (1) When the authorized user PT is in the H0 state, the cognitive system correctly detects that the PT is idle. The probability of this situation is P(H0)(1-P fa ).

[0028] (2) When the authorized user PT is in the H1 state, the cognitive system detects a false positive. The probability of this happening is P(H1)(1-P de ).

[0029] Where P(H0) and P(H1) represent the probabilities that the primary user is in the H0 state and the H1 state respectively.

[0030] Step S2: Information transmission and reception and task calculation

[0031] Assume that a partially unloaded edge computing mode is adopted and S can perform data transmission and local computing at the same time. At the same time, assume that a certain size of cache is configured in the drone relay to store the unloaded tasks waiting for computing. Since information transmission and reception and task computing are performed in different functional units, they can be performed simultaneously.

[0032] A three-dimensional Cartesian coordinate system is used, where the three-dimensional coordinates of the source node S, the destination node D, the drone relay r, the main transmitter PT, and the main receiver PR are (0,0,0), (x d ,0,0)、(x r ,y r ,H)、(x T ,y T ,0)、(x R ,y R ,0). For the convenience of exploration, the transmission time T2 is divided into N equal-length time slots, and the length of each time slot is

[0033] The wireless channel between the UAV relay and the ground communication node is mainly a line-of-sight channel. Therefore, the channel gains between S and UAV, UAV and D, S and PR, UAV and PR, S and PT, and UAV and PT can be expressed as

[0034]

[0035]

[0036]

[0037]

[0038]

[0039]

[0040] Where β0 represents the channel power gain when the reference distance d0=1, d sr d rd d sR d rR d sT d rT They represent the distances between S and UAV, UAV and D, S and PR, UAV and PR, S and PT, and UAV and PT respectively. Under the fast fading channel model, the wireless channel involved in the system remains stable within the time length T.

[0041] Assume that the raw data obtained by S is independent bit by bit and can be split in any proportion for parallel processing. Based on this, S can use the following two methods to collaboratively share the data processing results: 1) S completes the calculation and processing of part of the raw data locally, and then sends the calculation results to D with the assistance of the drone relay; 2) S unloads the remaining raw data to the drone, which performs auxiliary calculation processing and sends the results to D. Further, assume that the calculation delay of S and the decoding delay of the drone in method 1) are one time slot respectively, and the calculation preparation delay and calculation processing delay of the drone in method 2) are also one time slot respectively.

[0042] Step S2.1: Determine the amount of data bits and energy consumption in local calculation

[0043] After capturing the raw data, S performs local computation and offloading of the task synchronously. For local computation, let C represent the number of CPU cycles required to perform the unit bit computation task, and ρ∈(0,1) represents the data compression ratio. In order to efficiently utilize limited energy resources, S uses dynamic voltage and frequency scaling technology to adaptively control computational energy consumption. Let the CPU frequency of S at the nth moment be f s [n] cycles per second. Therefore, the task bit amount calculated by S at the nth time and the corresponding energy consumption are

[0044]

[0045]

[0046] Among them, γ s It indicates that S depends on the effective capacitance coefficient of the chip structure.

[0047] As local computation proceeds, S shares the computation results with D with the assistance of the drone relay. When the primary user is indeed idle and there is no false alarm in cooperative sensing, let represents the number of data bits sent out by S at the nth moment. Therefore, the corresponding information transmission energy consumption of S at the nth moment can be obtained as

[0048]

[0049] in, represents the information transmission power, B represents the channel bandwidth, σ 2 Represents the antenna noise power of the drone.

[0050] When the primary user is actually busy but cooperative sensing misses detection, the number of data bits sent by S at time n is expressed as

[0051]

[0052] It is easy to see from the analysis that at the nth moment, S can only send or share data that has been processed by local calculations. Therefore, the information causality constraint is

[0053]

[0054] Considering the existence of processing delay, S no longer transmits the calculation results in the first and last time slots, and no longer performs local calculations on the data in the last two time slots. Therefore,

[0055] and f s [N] = f s [N-1]=0.

[0056] After receiving the information from S, the UAV acts as a relay to decode the information and forward it to D. When the primary user is indeed idle and there is no false alarm in cooperative sensing, represents the amount of data bits forwarded by the drone at the nth moment, then the corresponding information forwarding energy consumption of the drone at the nth moment is

[0057]

[0058] in, Indicates the information forwarding power of the drone.

[0059] When the primary user is actually busy but cooperative sensing misses detection, the amount of data bits forwarded by the drone at time n is expressed as

[0060]

[0061] Through analysis, it is easy to see that the drone can only decode and forward the information data that has been sent from S. Therefore, the information causality constraint is

[0062]

[0063] Due to processing delay, the drone will not forward any information to D in the first two time slots. Therefore,

[0064] Based on the above analysis, the amount of data bits successfully shared by S and the corresponding energy consumption can be expressed as

[0065]

[0066]

[0067] Step S2.2: Determine the data bit amount and energy consumption in task offloading

[0068] In addition to local computing, S offloads the remaining computing tasks to the drone. When the primary user is indeed idle and there is no false alarm in cooperative sensing, let represents the amount of task data bits unloaded by S at the nth moment. Then the task unloading energy consumption of S at the nth moment is

[0069]

[0070] in, Indicates the transmit power.

[0071] When the primary user is actually busy but cooperative sensing misses detection, the amount of task data bits unloaded by S at time n is

[0072]

[0073] After receiving the task data sent by S, the UAV first calculates and processes it, and then forwards the calculation result to D. Assume that the UAV also uses dynamic voltage and frequency scaling technology to adaptively control its own calculation frequency, and let f r [n] represents the CPU frequency at the nth moment. Therefore, the amount of task data bits calculated by the drone at the nth moment and the corresponding energy consumption can be expressed as

[0074]

[0075]

[0076] Among them, γ r Indicates the effective capacitance coefficient of the drone that depends on the chip structure. When the primary user is indeed idle and there is no false alarm in cooperative sensing, represents the amount of data bits forwarded by the UAV to D at the nth moment, and the corresponding transmission energy consumption is

[0077]

[0078] in, Indicates the transmission power.

[0079] When the primary user is actually busy but cooperative sensing misses detection, the amount of data bits forwarded by the UAV to D at time n is

[0080]

[0081] From the analysis, we can see that in each time slot, the drone can only calculate the task data it has received from S, and the amount of data forwarded cannot be more than the amount of data generated by the drone through its own calculation and processing. Therefore, the information causality constraint is

[0082]

[0083]

[0084] Considering the existence of processing delay, S should not offload tasks to the UAV in the last two time slots. The UAV does not perform task calculations in the first and last time slots, and the UAV has no calculation results forwarded to D in the first two time slots. Therefore,

[0085] f r [1] = f r [N] = 0 and

[0086] Similarly, based on the above analysis, the amount of data bits successfully shared by S under this method and the corresponding energy consumption can be expressed as

[0087]

[0088]

[0089] Within a limited time T, the amount of data bits shared by S is not less than the minimum threshold I min , so there is a constraint

[0090]

[0091] In the process of data relay transmission between S and UAV, S and UAV will interfere with the normal communication of the main communication system. The interference generated in the local calculation stage is

[0092]

[0093] The interference generated during the task offloading phase is

[0094]

[0095] The average interference generated by S and UAVs to the main receiver PR is not greater than the interference upper limit Γ, so there is a constraint

[0096] P(H1)(1-P de )(I local +I offload )≤Γ (2.30)

[0097] Step S3: Determine the system average energy efficiency objective function

[0098] From the perspective of the entire system, the average energy efficiency of the system is maximized by optimizing the perception time, computing resources, and communication resources. The system energy efficiency is defined as the ratio of the total shared data bits to the total energy consumption of the system, that is,

[0099]

[0100] Among them, P v This is the user-perceived power loss. To consume energy while hovering, P UAV It is the power of the drone when it is in hovering state.

[0101] Step S4: Particle Swarm Optimization

[0102] Step S4.1: Set population parameters

[0103] Initialize the population sizepop and set the maximum number of iterationsger. In order to maximize the energy efficiency of the cognitive drone edge computing network, it is necessary to reasonably optimize computing resources, communication resources and perception time, that is, f s [n], f r [n], (n∈N) and t s The total dimension dim is 6N+1, and the variables to be optimized are divided into three groups: central processing unit frequency, data bit amount, and perception time, and their position and speed upper and lower limits are set respectively.

[0104] Step S4.2: Generate initial population

[0105] The population of each generation consists of sizepop individuals with a dimension of 6N+1, and each individual is set to pop_x j,g (j=1,2,……,sizepop). Where j represents the sequence of individuals in the population, and g represents the number of iterations. The initial population velocity and position are randomly selected based on the given boundary constraints. The randomly generated initial population position and velocity are expressed as

[0106]

[0107]

[0108] Where: i = 1, 2, ...; 6N + 1, j = 1, 2, ..., sizepop, and Respectively represent pop_x i The maximum and minimum values ​​of and Respectively represent pop_v i The maximum and minimum values ​​of rand[0,1] represent the generation of uniform random numbers between [0,1].

[0109] Through continuous searching, we find pop_x that satisfies equations (2.11), (2.14), (2.23), and (2.24) ij,0Under the constraints (2.27) and (2.30), the fitness of each particle is calculated according to the objective function (3.1), and the initial historical best position and historical best fitness of the individual as well as the initial historical best position and historical best fitness of the group are obtained.

[0110] Step S4.3: Particle swarm iteration

[0111] The velocity position evolution equation of the standard particle swarm algorithm is:

[0112]

[0113] pop_x j,g+1 =pop_x j,g +pop_v j,g+1 (4.4)

[0114] pop_v j,g represents the velocity of the jth particle in the gth generation, represents the optimal position of the jth particle in the gth generation, represents the optimal position of the g-th generation of the group, c1 represents the inheritance coefficient of the particle to the previous velocity, c2 is the influence coefficient of the particle's own behavior on the subsequent behavior, c3 represents the influence coefficient of the group behavior on each particle, r1 and r2 are random numbers distributed between 0 and 1.

[0115] According to formulas (4.3) and (4.4), the particle speed and position are updated and boundary processing is performed. Then, the particle position that satisfies formulas (2.11), (2.14), (2.23), and (2.24) is found again, the constraints (2.27) and (2.30) are judged, and the fitness of each individual position of the new population is calculated.

[0116] The new fitness is compared with the individual's best historical fitness, the individual's best historical position is updated, and the individual's best historical fitness is updated; the individual's best historical fitness is compared with the population's best historical fitness, the population's best historical position is updated, and the population's best historical fitness is updated.

[0117] When the algorithm runs to the maximum number of iterations set in advance, it stops calculating and derives the global optimal value and fitness value of the particle population at the current moment. If the maximum number of iterations is not reached, it returns to continue the iterative cycle.

[0118] Beneficial effects of the invention: The method proposed in the invention uses a particle swarm algorithm to optimize the data bit amount, central processing unit frequency and perception time in the cognitive drone edge computing network. The method obtains an optimal population through continuous iterative updates, thereby obtaining an optimal individual, thereby improving the energy efficiency of the cognitive drone edge computing network:

[0119] (1) It solves the problem of how cognitive users allocate perception time, communication resources and computing resources in the collaborative spectrum perception, task offloading and task data processing and computing stages. The present invention uses the particle swarm algorithm to define the system energy efficiency by taking the ratio of total shared data bits to total system energy consumption, and continuously cycles the algorithm to reach an optimal value of the population.

[0120] (2) The probability of spectrum holes being discovered and the energy efficiency of the system are improved, a trade-off is achieved between the total shared data bits and the total energy consumption of the system, and cognitive users achieve an optimal strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0121] Figure 1 This is the flow chart of the optimization algorithm of the present invention.

[0122] Figure 2 This is the edge computing and result sharing system assisted by the cognitive drone in the present invention.

[0123] Figure 3 It is a schematic diagram of the frame structure of the present invention. DETAILED DESCRIPTION

[0124] In order to deepen the understanding of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The embodiments are only used to explain the present invention and do not limit the protection scope of the present invention.

[0125] Example: Figure 1 , Figure 2 and Figure 3 As shown, an optimization method for maximizing energy efficiency in a cognitive drone edge computing network includes the following steps:

[0126] Step S1: Determine the spectrum hole probability

[0127] The collaborative spectrum sensing process is as follows: in the T1 time slot, the source node S and the UAV relay respectively perform local energy detection on the received signal from the main transmitter PT, and then the UAV relay transmits the local sensing result (PT existence or non-existence, i.e. 1 or 0) to the source node S through a dedicated control channel. The source node S uses the "OR" fusion rule to fuse the UAV relay's sensing result and the local sensing result and make the final decision. The sensing time slot can be divided into two sub-time slots t s and t r In the first sub-time slot, the source node S and the drone relay perform spectrum sensing respectively. In the second sub-time slot, the drone relay transmits the sensing result to the source node S, which integrates and makes the final decision. Since the sensing result reporting time is t r It is very short and can be ignored.

[0128] Assume that y(n) is the PT transmission signal received by the source node S or the drone relay in T1. According to the binary hypothesis, the received signal when the PT exists (represented by H1) or does not exist (represented by H0) is as follows:

[0129]

[0130] Among them, x p (t) is the power P p The primary user transmits a signal, n(t) is subject to N(0, σ 2 ) distributed Gaussian white noise, h ST is the channel gain of the PT-S link (if it is a PT-UAV link, the channel gain is h rT ). M=t s f s , is the number of sampling points, f s is the sampling frequency. Local sensing uses energy detection, and the energy statistics of the local sensing signal of S (or drone relay) can be expressed as

[0131]

[0132] The local false alarm probability P of the source node S (or drone relay) under energy detection f and the local detection probability P d It can be expressed as:

[0133]

[0134]

[0135] Where Q(·) is the standard Gaussian complementary distribution function: η is the decision threshold of the local energy detector of the source node S and the drone relay, and γ is the average signal-to-noise ratio of the authorized user signal received by S (or drone relay):

[0136] The source node S uses the “OR” fusion rule to fuse the local decision results of S and the drone relay. Therefore, the final cooperative false alarm probability P of the entire system is fa and cooperative detection probability P de as follows:

[0137] P fa =1-(1-P f ) 2 (1.5)

[0138] P de =1-(1-P d ) 2 (1.6)

[0139] From this we can get,

[0140]

[0141] There are two cases for the source node S (or drone relay) to transmit data:

[0142] (1) When the authorized user PT is in the H0 state, the cognitive system correctly detects that the PT is idle. The probability of this situation is P(H0)(1-P fa ).

[0143] (2) When the authorized user PT is in the H1 state, the cognitive system detects a false positive. The probability of this happening is P(H1)(1-P de ).

[0144] Where P(H0) and P(H1) represent the probabilities that the primary user is in the H0 state and the H1 state respectively.

[0145] Step S2: Information transmission and reception and task calculation

[0146] Assume that a partially unloaded edge computing mode is adopted and S can perform data transmission and local computing at the same time. At the same time, assume that a certain size of cache is configured in the drone relay to store the unloaded tasks waiting for computing. Since information transmission and reception and task computing are performed in different functional units, they can be performed simultaneously.

[0147] A three-dimensional Cartesian coordinate system is used, where the three-dimensional coordinates of the source node S, the destination node D, the drone relay r, the main transmitter PT, and the main receiver PR are (0,0,0), (x d ,0,0)、(x r ,y r ,H)、(x T ,y T ,0)、(x R ,y R ,0). For the convenience of exploration, the transmission time T2 is divided into N equal-length time slots, and the length of each time slot is

[0148] The wireless channel between the UAV relay and the ground communication node is mainly a line-of-sight channel. Therefore, the channel gains between S and UAV, UAV and D, S and PR, UAV and PR, S and PT, and UAV and PT can be expressed as

[0149]

[0150]

[0151]

[0152]

[0153]

[0154]

[0155] Where β0 represents the channel power gain when the reference distance d0=1, d sr d rd d sR d rR d sT d rT They represent the distances between S and UAV, UAV and D, S and PR, UAV and PR, S and PT, and UAV and PT respectively. Under the fast fading channel model, the wireless channel involved in the system remains stable within the time length T.

[0156] Assume that the raw data obtained by S is independent bit by bit and can be split in any proportion for parallel processing. Based on this, S can use the following two methods to collaboratively share the data processing results: 1) S completes the calculation and processing of part of the raw data locally, and then sends the calculation results to D with the assistance of the drone relay; 2) S unloads the remaining raw data to the drone, which performs auxiliary calculation processing and sends the results to D. Further, assume that the calculation delay of S and the decoding delay of the drone in method 1) are one time slot respectively, and the calculation preparation delay and calculation processing delay of the drone in method 2) are also one time slot respectively.

[0157] Step S2.1: Determine the amount of data bits and energy consumption in local calculation

[0158] After capturing the raw data, S performs local computation and offloading of the task synchronously. For local computation, let C represent the number of CPU cycles required to perform the unit bit computation task, and ρ∈(0,1) represents the data compression ratio. In order to efficiently utilize limited energy resources, S uses dynamic voltage and frequency scaling technology to adaptively control computational energy consumption. Let the CPU frequency of S at the nth moment be f s [n] cycles per second. Therefore, the task bit amount calculated by S at the nth time and the corresponding energy consumption are

[0159]

[0160]

[0161] Among them, γ s It indicates that S depends on the effective capacitance coefficient of the chip structure.

[0162] As local computation proceeds, S shares the computation results with D with the assistance of the drone relay. When the primary user is indeed idle and there is no false alarm in cooperative sensing, let represents the number of data bits sent out by S at the nth moment. Therefore, the corresponding information transmission energy consumption of S at the nth moment can be obtained as

[0163]

[0164] in, represents the information transmission power, B represents the channel bandwidth, σ 2 Represents the antenna noise power of the drone.

[0165] When the primary user is actually busy but cooperative sensing misses detection, the number of data bits sent by S at time n is expressed as

[0166]

[0167] It is easy to see from the analysis that at the nth moment, S can only send or share data that has been processed by local calculations. Therefore, the information causality constraint is

[0168]

[0169] Considering the existence of processing delay, S no longer transmits the calculation results in the first and last time slots, and no longer performs local calculations on the data in the last two time slots. Therefore,

[0170] have and f s [N] = f s [N-1]=0.

[0171] After receiving the information from S, the UAV acts as a relay to decode the information and forward it to D. When the primary user is indeed idle and there is no false alarm in cooperative sensing, represents the amount of data bits forwarded by the drone at the nth moment, then the corresponding information forwarding energy consumption of the drone at the nth moment is

[0172]

[0173] in, Indicates the information forwarding power of the drone.

[0174] When the primary user is actually busy but cooperative sensing misses detection, the amount of data bits forwarded by the drone at time n is expressed as

[0175]

[0176] Through analysis, it is easy to know that the drone can only decode and forward the information data that has been sent from S. Therefore, the information causality constraint is:

[0177]

[0178] Due to processing delay, the drone will not forward any information to D in the first two time slots. Therefore,

[0179] Based on the above analysis, the amount of data bits successfully shared by S and the corresponding energy consumption can be expressed as

[0180]

[0181]

[0182] Step S2.2: Determine the data bit amount and energy consumption in task offloading

[0183] In addition to local computing, S offloads the remaining computing tasks to the drone. When the primary user is indeed idle and there is no false alarm in cooperative sensing, let represents the amount of task data bits unloaded by S at the nth moment. Then the task unloading energy consumption of S at the nth moment is

[0184]

[0185] in, Indicates the transmit power.

[0186] When the primary user is actually busy but cooperative sensing misses detection, the amount of task data bits unloaded by S at time n is

[0187]

[0188] After receiving the task data sent by S, the UAV first calculates and processes it, and then forwards the calculation result to D. Assume that the UAV also uses dynamic voltage and frequency scaling technology to adaptively control its own calculation frequency, and let f r [n] represents the CPU frequency at the nth moment. Therefore, the amount of task data bits calculated by the drone at the nth moment and the corresponding energy consumption can be expressed as

[0189]

[0190]

[0191] Among them, γ r Indicates the effective capacitance coefficient of the drone that depends on the chip structure. When the primary user is indeed idle and there is no false alarm in cooperative sensing, represents the amount of data bits forwarded by the UAV to D at the nth moment, and the corresponding transmission energy consumption is

[0192]

[0193] in, Indicates the transmission power.

[0194] When the primary user is actually busy but cooperative sensing misses detection, the amount of data bits forwarded by the UAV to D at time n is

[0195]

[0196] From the analysis, we can see that in each time slot, the drone can only calculate the task data it has received from S, and the amount of data forwarded cannot be more than the amount of data generated by the drone through its own calculation and processing. Therefore, the information causality constraint is

[0197]

[0198]

[0199] Considering the existence of processing delay, S should not offload tasks to the UAV in the last two time slots. The UAV does not perform task calculations in the first and last time slots, and the UAV has no calculation results forwarded to D in the first two time slots. Therefore, f r [1] = f r [N] = 0 and

[0200] Similarly, based on the above analysis, the amount of data bits successfully shared by S under this method and the corresponding energy consumption can be expressed as

[0201]

[0202]

[0203] Within a limited time T, the amount of data bits shared by S is not less than the minimum threshold I min , so there is a constraint

[0204]

[0205] In the process of data relay transmission between S and UAV, S and UAV will interfere with the normal communication of the main communication system. The interference generated in the local calculation stage is

[0206]

[0207] The interference generated during the task offloading phase is

[0208]

[0209] The average interference generated by S and UAVs to the main receiver PR is not greater than the interference upper limit Γ, so there is a constraint

[0210] P(H1)(1-P de )(I local +I offload )≤Γ (2.30)

[0211] Step S3: Determine the system average energy efficiency objective function

[0212] From the perspective of the entire system, the average energy efficiency of the system is maximized by optimizing the perception time, computing resources, and communication resources. The system energy efficiency is defined as the ratio of the total shared data bits to the total energy consumption of the system, that is,

[0213]

[0214] Among them, P v This is the user-perceived power loss. To consume energy while hovering, P UAV It is the power of the drone when it is in hovering state.

[0215] Step S4: Particle Swarm Optimization

[0216] Step S4.1: Set population parameters

[0217] Initialize the population sizepop and set the maximum number of iterationsger. In order to maximize the energy efficiency of the cognitive drone edge computing network, it is necessary to reasonably optimize computing resources, communication resources and perception time, that is, f s [n], f r [n], (n∈N) and t s The total dimension dim is 6N+1, and the variables to be optimized are divided into three groups: central processing unit frequency, data bit amount, and perception time, and their position and speed upper and lower limits are set respectively.

[0218] Step S4.2: Generate initial population

[0219] The population of each generation consists of sizepop individuals with a dimension of 6N+1, and each individual is set to pop_x j,g (j=1,2,……,sizepop). Where j represents the sequence of individuals in the population, and g represents the number of iterations. The initial population velocity and position are randomly selected based on the given boundary constraints. The randomly generated initial population position and velocity are expressed as

[0220]

[0221]

[0222] Where: i = 1, 2, ...; 6N + 1, j = 1, 2, ..., sizepop, and Respectively represent pop_x i The maximum and minimum values ​​of and Respectively represent pop_v i The maximum and minimum values ​​of rand[0,1] represent the generation of uniform random numbers between [0,1].

[0223] Through continuous searching, we find pop_x that satisfies equations (2.11), (2.14), (2.23), and (2.24) ij,0 Under the constraints (2.27) and (2.30), the fitness of each particle is calculated according to the objective function (3.1), and the initial historical best position and historical best fitness of the individual as well as the initial historical best position and historical best fitness of the group are obtained.

[0224] Step S4.3: Particle swarm iteration

[0225] The velocity position evolution equation of the standard particle swarm algorithm is:

[0226]

[0227] pop_x j,g+1 =pop_x j,g +pop_v j,g+1 (4.4)

[0228] pop_v j,g represents the velocity of the jth particle in the gth generation, represents the optimal position of the jth particle in the gth generation, represents the optimal position of the g-th generation of the group, c1 represents the inheritance coefficient of the particle to the previous velocity, c2 is the influence coefficient of the particle's own behavior on the subsequent behavior, c3 represents the influence coefficient of the group behavior on each particle, r1 and r2 are random numbers distributed between 0 and 1.

[0229] According to formulas (4.3) and (4.4), the particle speed and position are updated and boundary processing is performed. Then, the particle position that satisfies formulas (2.11), (2.14), (2.23), and (2.24) is found again, the constraints (2.27) and (2.30) are judged, and the fitness of each individual position of the new population is calculated.

[0230] The new fitness is compared with the individual's best historical fitness, the individual's best historical position is updated, and the individual's best historical fitness is updated; the individual's best historical fitness is compared with the population's best historical fitness, the population's best historical position is updated, and the population's best historical fitness is updated.

[0231] When the algorithm runs to the maximum number of iterations set in advance, it stops calculating and derives the global optimal value and fitness value of the particle population at the current moment. If the maximum number of iterations is not reached, it returns to continue the iterative cycle.

[0232] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An optimization method for maximizing energy efficiency in a cognitive drone edge computing network, characterized in that: The steps include: Step S1, determining the spectrum hole probability; Step S2, information sending and receiving and task calculation; Step S3, determining the system average energy efficiency objective function; Step S4, optimizing using particle swarm algorithm; In step S1, the collaborative spectrum sensing process is: in the T1 time slot, the source node S and the UAV relay respectively perform local energy detection on the signal received from the main transmitter PT, and then the UAV relay transmits the local sensing result to the source node S through a dedicated control channel. The source node S uses the "OR" fusion rule to fuse the sensing result of the UAV relay and the local sensing result and make a final decision; the sensing time slot can be divided into two sub-time slots t s and t r In the first sub-time slot, the source node S and the drone relay perform spectrum sensing respectively. In the second sub-time slot, the drone relay transmits the sensing result to the source node S, which integrates and makes the final decision. When the sensing result report time t r Very short, ignore it; Assume that y(n) is the PT transmission signal received by the source node S or the drone relay in T1. According to the binary hypothesis, the received signal when the PT exists or does not exist is as follows: Among them, H1 means that PT exists, H0 means that PT does not exist, and x p (t) is the power P p The primary user transmits a signal, n(t) is subject to N(0, σ 2 ) distributed Gaussian white noise, h ST is the channel gain of the PT-S link; M = t s f s is the number of sampling points, f s is the sampling frequency; local sensing uses energy detection, and the energy statistics of the local sensing signal S is expressed as The local false alarm probability P of the source node S under energy detection f and the local detection probability P d It is expressed as: Where Q(·) is the standard Gaussian complementary distribution function: η is the decision threshold of the local energy detector of the source node S and the drone relay, and γ is the average signal-to-noise ratio of the authorized user signal received by S (or drone relay): The source node S uses the "OR" fusion rule to fuse the local judgment results of S and the drone relay. Therefore, the final cooperative false alarm probability P of the entire system is fa and cooperative detection probability P de as follows: P fa =1-(1-P f ) 2 (1.5) P de =1-(1-P d ) 2 (1.6) From this we can get: In step S2, a three-dimensional Cartesian coordinate system is used, in which the three-dimensional coordinates of the source node S, the destination node D, the drone relay r, the main transmitter PT, and the main receiver PR are (0,0,0), (x d ,0,0)、(x r ,y r ,H)、(x T ,y T ,0)、(x R ,y R ,0); Divide the transmission time T2 into N equal-length time slots, the length of each time slot is The wireless channel between the UAV relay and the ground communication node is mainly a line-of-sight channel. Therefore, the channel gains between S and UAV, UAV and D, S and PR, UAV and PR, S and PT, and UAV and PT are expressed as Where β0 represents the channel power gain when the reference distance d0=1, d sr d rd d sR d rR d sT d rT They represent the distances between S and UAV, UAV and D, S and PR, UAV and PR, S and PT, and UAV and PT respectively; under the fast fading channel model, the wireless channel involved in the system remains stable within the time length T; The step S2 includes the following process: Process S2.1: Determine the amount of data bits and energy consumption in local computing After capturing the raw data, S synchronously performs local computation and offloading of the task. For local computation, let C represent the number of CPU cycles required to perform the unit bit computation task, and ρ∈(0,1) represents the data compression ratio. To efficiently utilize limited energy resources, S uses dynamic voltage and frequency scaling technology to adaptively control computational energy consumption. The CPU frequency of S at the nth moment is represented as f s [n] cycles per second; the task bit amount calculated at time n and the corresponding energy consumption are Among them, γ s It indicates that S depends on the effective capacitance coefficient of the chip structure; As local computation proceeds, S shares the computation results with D with the assistance of the drone relay. When the primary user is indeed idle and there is no false alarm in the collaborative sensing, represents the number of data bits sent out by S at the nth moment, and the corresponding information transmission energy consumption of S at the nth moment is in, represents the information transmission power, B represents the channel bandwidth, σ 2 represents the antenna noise power of the UAV; When the primary user is actually busy but cooperative sensing misses detection, the number of data bits sent by S at time n is expressed as Through analysis, it is easy to see that at the nth moment, S can only send or share data that has been processed by local calculations, and the information causality constraint is Considering the existence of processing delay, S no longer transmits the calculation results in the first and last time slots, and no longer performs local calculations on the data in the last two time slots; and f s [N] = f s [N-1] = 0; After receiving the information from S, the UAV acts as a relay to decode the information and forward it to D. When the primary user is indeed idle and there is no false alarm in cooperative sensing, represents the amount of data bits forwarded by the drone at the nth moment, then the corresponding information forwarding energy consumption of the drone at the nth moment is in, Indicates the information forwarding power of the drone; When the primary user is actually busy but cooperative sensing misses detection, the amount of data bits forwarded by the drone at time n is expressed as Through analysis, it is easy to see that the drone can only decode and forward the information data that has been sent from S; The information causality constraint is When processing delay, the drone will not forward any information to D in the first two time slots. have: Based on the above analysis, the amount of data bits successfully shared by S and the corresponding energy consumption are expressed as Process S2.2: Determine the data bit amount and energy consumption in task offloading make represents the amount of task data bits unloaded by S at the nth moment, then the task unloading energy consumption of S at the nth moment is: in, Indicates the transmit power; When the primary user is actually busy but cooperative sensing misses detection, the amount of task data bits unloaded by S at the nth moment is: After receiving the task data sent by S, the UAV first calculates and processes it, and then forwards the calculation result to D. It is assumed that the UAV also uses dynamic voltage and frequency scaling technology to adaptively control its own calculation frequency, and let f r [n] represents the CPU frequency at the nth moment. The bit amount of mission data calculated by the drone at the nth moment and the corresponding energy consumption can be expressed as: Among them, γ r It indicates that the effective capacitance coefficient of the drone depends on the chip structure. When the primary user is indeed idle and there is no false alarm in the cooperative perception, represents the amount of data bits forwarded by the UAV to D at time n, and the corresponding transmission energy consumption is: in, Indicates the transmission power; When the primary user is actually busy but cooperative sensing misses detection, the amount of data bits forwarded by the UAV to D at time n is In each time slot, the drone can only calculate the task data it has received from S, and the amount of data forwarded cannot be more than the amount of data generated by the drone through its own calculation and processing; the information causality constraint is Considering the existence of processing delay, S should not offload tasks to the drone in the last two time slots. The drone does not perform task calculations in the first and last time slots, and the drone has no calculation results forwarded to D in the first two time slots. Therefore, f r [1] = f r [N] = 0 and Based on the above analysis, the amount of data bits successfully shared by S under this method and the corresponding energy consumption are expressed as: Within a limited time T, the amount of data bits shared by S is not less than the minimum threshold I min , with constraints In the process of data relay transmission between S and UAV, S and UAV will interfere with the normal communication of the main communication system. The interference generated in the local calculation stage is: The interference generated during the task offloading phase is The average interference generated by S and UAVs to the main receiver PR is not greater than the interference upper limit Γ, so there is a constraint P(H1)(1-P de )(AND local +I offload )≤Γ(2.30)。 2. The optimization method for maximizing energy efficiency in a cognitive drone edge computing network according to claim 1, characterized in that: The specific steps of step S3 are as follows: From the perspective of the entire system, the average energy efficiency of the system is maximized by optimizing the perception time, computing resources, and communication resources. The system energy efficiency is defined as the ratio of the total shared data bits to the total energy consumption of the system, that is: Among them, P v This is the user-perceived power loss. To consume energy while hovering, P UAV It is the power of the drone when it is in hovering state.

3. The optimization method for maximizing energy efficiency in a cognitive drone edge computing network according to claim 2, characterized in that: The specific process of step S4 is as follows: Process S4.1: Setting population parameters Initialize the population sizepop and set the maximum number of iterationsger. To maximize the energy efficiency of the cognitive drone edge computing network, it is necessary to reasonably optimize computing resources, communication resources and perception time, that is, f s [n], f r [n], and t s , the total dimension dim is 6N+1, and the variables to be optimized are divided into three groups: central processing unit frequency, data bit amount, and perception time, and their position and speed upper and lower limits are set respectively; Step S4.2: Generate initial population The population of each generation consists of sizepop individuals with a dimension of 6N+1, and each individual is set to pop_x j,g (j=1,2,……,sizepop), where j represents the sequence of individuals in the population, g represents the number of iterations, and the initial population speed and position are randomly selected according to the given boundary constraints. The randomly generated initial population position and speed are expressed as: Among them, i=1,2,……;6N+1,j=1,2,……,sizepop, and Respectively represent pop_x i The maximum and minimum values ​​of and Respectively represent pop_v i The maximum and minimum values ​​of rand[0,1] represent the generation of uniform random numbers between [0,1]; Through continuous searching, we find pop_x that satisfies equations (2.11), (2.14), (2.23), and (2.24) ij,0 ; Under the constraints (2.27) and (2.30), the fitness of each particle is calculated according to the objective function (3.1), and the initial historical best position and historical best fitness of the individual and the initial historical best position and historical best fitness of the group are obtained; Process S4.3, particle swarm iteration The velocity position evolution equation of the standard particle swarm algorithm is: pop_x j,g+1 =pop_x j,g +pop_v j,g+1 (4.4) pop_v j,g represents the velocity of the jth particle in the gth generation, represents the optimal position of the jth particle in the gth generation, represents the optimal position of the g-th generation of the group, c1 represents the inheritance coefficient of the particle to the previous velocity, c2 represents the influence coefficient of the particle's own behavior on the subsequent behavior, c3 represents the influence coefficient of the group behavior on each particle, r1 and r2 are random numbers distributed between 0 and 1, According to formula (4.3) and (4.4), the speed and position of the particles are updated and the boundary processing is performed. Then, the particle positions that satisfy formula (2.11), (2.14), (2.23) and (2.24) are found again, and the constraints (2.27) and (2.30) are judged and the fitness of each individual position of the new population is calculated. The new fitness is compared with the individual's best historical fitness, the individual's best historical position is updated, and the individual's best historical fitness is updated; the individual's best historical fitness is compared with the population's best historical fitness, the population's best historical position is updated, and the population's best historical fitness is updated; When the algorithm reaches the maximum number of iterations set in advance, it stops calculating and derives the global optimal value and fitness value of the particle population at the current moment. If the maximum number of iterations is not reached, it returns to continue the iterative cycle.

Images