MA enhanced mobile edge cloud computing network performance optimization method

Abstract

The invention belongs to the technical field of intelligent power grid and vehicle network interaction, and discloses an MA-enhanced mobile edge cloud computing network performance optimization method, which constructs an MA-enhanced WPB-MEC system, improves the communication performance in the vehicle network interaction scheduling optimization process, and improves the network performance. Calculating the total calculation bit of the WPB-MEC system according to the transmitting covariance matrix at the AP, the network time resource allocation, the transmitting power and the calculation frequency of the WDs and the position of the MAs; a block coordinate descent algorithm is designed, a time allocation scheme in a specified transmission block, a transmission covariance matrix at an AP, the calculation frequency of WDs, the transmission power and the position of MAs are optimized, an optimization problem is established, and the total calculation bit of the system is maximized. The computing capability of the network is remarkably improved, the communication capability of vehicle-network interaction scheduling optimization is enhanced, and the communication performance during vehicle-network interaction is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grid and vehicle-grid interaction, and specifically relates to a MA-enhanced mobile edge cloud computing network performance optimization method. Background Art

[0002] In recent years, new energy vehicles such as electric vehicles have attracted more and more attention from the world for their energy-saving and environmentally friendly advantages. Under the supportive policies implemented by various countries, the electric vehicle-related industry has developed rapidly. However, due to the high randomness and strong volatility of the charging load of electric vehicles in the temporal and spatial distribution, it will have a significant impact on the safety and stability of the distribution network operation. Therefore, the vehicle-grid interaction technology based on orderly charging and discharging came into being, bringing changes to the operation of smart grids. While reducing the adverse effects of disordered charging through vehicle-grid interaction, electric vehicles can be used as mobile energy storage to achieve load transfer, peak load regulation, frequency regulation, standby and other services. In the process of vehicle-grid interaction, the intelligent terminal in the substation sends control instructions to regulate the charging piles under it. Therefore, on the one hand, the charging piles must have reliable communication capabilities for information exchange with the intelligent terminal in the substation; on the other hand, the efficient electric vehicle control strategy (control instructions sent by the intelligent terminal in the substation) must be issued based on the perception of the substation operation status and make full use of the computing capabilities of the intelligent terminal in the substation in edge computing, cloud computing and other aspects. Therefore, optimizing the information transmission performance during vehicle-grid interaction is of great significance for improving the accuracy and reliability of the orderly charging and discharging control strategy of electric vehicles and supporting the realization of vehicle-grid interaction.

[0003] However, the wireless devices (WDs) deployed in the process of realizing vehicle-grid interaction usually have limited battery capacity and computing power, which affects the performance of the vehicle-grid interaction system. In order to solve the above problems, the concept of wireless powered and backscattering mobile edge computing (WPB-MEC) came into being. It combines wireless power transfer (WPT) and backscattering enhanced mobile edge computing (MEC) technology. Specifically, in the WPB-MEC system, the MEC server performs data collection, transmission and simple data processing for WDs, and transmits data with low latency requirements to the cloud platform. WDs can realize passive offloading of tasks by backscattering signals from access points (APs), or they can realize active offloading of tasks by using a collection-first-then-offload protocol. AP provides wireless energy transmission for WDs to support the operation of WDs. When one WD passively offloads its tasks, other WDs can collect energy at the same time. However, in the current existing technology, the fixed position antennas (FPAs) equipped on APs and MEC servers cannot fully utilize their spatial degrees of freedom (DoF), resulting in low efficiency of wireless energy transmission and active and passive task offloading, which affects the performance of the network.

[0004] Recently, a movable antenna (MA) has been proposed as a new antenna paradigm, which can be connected to its radio frequency (RF) chain via a flexible cable, enabling it to flexibly adjust its position in a given spatial area with the help of a stepper motor; however, there is still no report on how to build a MA-enhanced WPB-MEC network. In addition, the performance optimization method for MA-enhanced WPB-MEC networks is still unknown. Summary of the invention

[0005] To solve the above technical problems, the present invention provides a MA-enhanced mobile edge cloud computing network performance optimization method. Based on MAs, favorable channel conditions can be obtained, thereby making full use of DoFs to improve communication performance. During the vehicle-grid interaction process, the effective regulation of the orderly charging and discharging of electric vehicles depends on a reliable signal transmission network. The present invention takes enhancing the total computing bits of the network as a design goal, thereby improving the efficiency of global information interaction during vehicle-grid interaction.

[0006] The MA-enhanced mobile edge cloud computing network performance optimization method of the present invention comprises the following steps:

[0007] Step 1: construct an MA-enhanced WPB-MEC system, wherein the WPB-MEC system includes an access point AP with N movable antennas MAs, an MEC server with M movable antennas MAs, and K wireless devices WDs;

[0008] The movable antennas MAs of the AP and MEC server are connected to their respective radio frequency RF chains through flexible cables, so that the MAs are in their respective two-dimensional areas C of size A×A. t and C r The position in can be moved;

[0009] The tasks of each WD can be divided. Each WD is equipped with a single fixed-position antenna FPA. Based on the backscatter communication circuit and active RF circuit, each WD uses passive task offloading or active task offloading to offload the corresponding computing tasks to the MEC server.

[0010] Step 2: Based on the MA-enhanced WPB-MEC system, the total calculation bits of the WPB-MEC system are calculated according to the transmit covariance matrix at the AP, the network time resource allocation, the transmit power and calculation frequency of the WDs, and the position of the MAs;

[0011] Step 3: Design a block coordinate descent algorithm to optimize the time allocation scheme within a specified transmission block, the transmit covariance matrix at the AP, the calculation frequency of WDs, the transmit power, and the location of MAs, and establish an optimization problem to maximize the total computational bits of the system.

[0012] Furthermore, the set of K wireless devices WDs is {WD-1, WD-2, …, WD-K}, the duration of the transmission block in the system is T seconds, and TDMA is used to divide the passive task offloading and active task offloading into K sub-phases respectively. In the kth sub-phase of the passive offloading phase of 10 seconds, WD-k offloads part of the tasks to the MEC server in the form of passive offloading based on the backscattering circuit; The subscript k in the figure corresponds to the kth sub-stage and the kth WD, and the superscript b indicates the passive unloading stage. At the same time, the remaining WDs collect energy from the AP to increase the total amount of energy collected; The kth sub-phase of the active unloading phase of seconds, The superscript a in the middle indicates the active unloading stage. WD-k uses the collected energy to actively unload another part of the tasks to the MEC server; specifically:

[0013] 1) In the passive task offloading stage, the AP transmits an energy signal to achieve passive offloading and wireless energy transmission WPT, specifically:

[0014] Let W k for The AP transmits the covariance matrix during the period; for passive offloading, the calculation bit of WD-k offloading is expressed as:

[0015]

[0016] in, B is the channel bandwidth, ρ is the reflection coefficient, is the noise power spectral density, ξ represents the performance gap caused by using the actual modulation in the passive unloading phase;

[0017] In the passive unloading phase, WPT is performed on WD-k; for WD-k, part of its received power ρtr(B k W k ) is used to implement Passive unloading during the period, and the remaining power (1-ρ)tr(B k W k ) is converted into direct current and stored in the battery of WD-k;

[0018] During the entire passive unloading phase, except for Outside the corresponding sub-phase, WD-k keeps collecting energy from AP. According to the nonlinear energy harvesting model, the energy obtained by WD-k in the passive unloading phase is:

[0019]

[0020] in, represents the energy collected by WD-k in the kth sub-stage of the passive unloading phase, represents the duration of the kth sub-phase of the passive unloading phase, represents the duration of the i-th sub-phase of the passive unloading phase, i≠k; represents the received power of WD-k in the i-th sub-phase of the passive unloading phase, i≠k; represents the logic function in the energy harvesting circuit of WD-k, P sat is the saturation power, a k and b k are the control parameters of the energy harvesting circuit of WD-k;

[0021] 2) In the active task offloading phase, let p k is the transmission power of WD-k, WD-k is The calculation bits of active unloading during the period are expressed as:

[0022]

[0023] Assume E c,k represents a constant circuit energy consumption, then the energy consumption of WD-k in the active unloading stage is expressed as

[0024] 3) The calculation bits of WD-k calculated locally are:

[0025]

[0026] Among them, f k represents the calculation frequency of WD-k, represents the computational complexity of the task at WD-k; the energy required to perform local computation at WD-k is Indicates that, where ε k is the effective capacitance coefficient of the on-chip processor located at WD-k.

[0027] Furthermore, the channel model in the WPB-MEC system is as follows:

[0028] Assume t n =[x n ,y n ],r m =[x m ,y m ] represent the nth MA position of the AP and the mth MA position of the MEC server, respectively. x n ,y n is the horizontal and vertical coordinates of the nth MA position of AP, x m ,y m is the horizontal and vertical coordinates of the mth MA position of the MEC server; the set of MAs positions on the AP and MEC server is expressed as

[0029] The number of transmission channel paths between AP and WD-k and between WD-k and MEC server are respectively and To express it, then the nth MA position of WD-k and the lth reference point between the AP t The signal propagation phase difference of the paths is expressed as:

[0030]

[0031] In the formula, and Respectively represent the first t The elevation and azimuth angles of the transmission paths;

[0032] Channel vector from AP to WD-k It is expressed as:

[0033]

[0034] is the transmitting field response matrix, is the corresponding vector of the transmission field of the nth MA in the AP, λ is the wavelength of the carrier, j is the imaginary sign, is the path response vector between AP and WD-k;

[0035] Similarly, the channel vector from WD-k to the MEC server is It is expressed as:

[0036]

[0037] in represents the corresponding path response vector, is the receiving field response matrix of the MEC server, is the corresponding vector of the mth receiving field in the MEC server, represents the signal propagation phase difference of the lrth path between the mth MA location of the MEC server and the reference point, and They are the first r The elevation and azimuth angles of the receiving paths,

[0038] Furthermore, the optimization problem is constructed as follows:

[0039] According to the location of MAs in AP and MEC server and Channel vector from AP to WD-k Channel vector from WD-k to MEC server The transmit covariance matrix at AP is W = {W k ,k∈K}, the transmission power of WDs p={p k ,k∈K}, calculate the frequency f={f k , k∈K, we establish the problem P0 of maximizing the total computational bits of the network, which can be expressed as:

[0040] P0:

[0041] stC1:

[0042] C2:

[0043] C3:

[0044] C4:

[0045] C5:

[0046] C6:

[0047] C7:

[0048] C8: ||t n -t n′ ||2≥D,1≤n≠n′≤N

[0049] C9: ||r m -r m′ ||2≥D,1≤m≠m′≤N

[0050] Where P H is the maximum transmit power of the AP, is the maximum calculation frequency of WD-k, D is the minimum distance between any two MAs; ||·||2 represents the Euclidean norm.

[0051] Furthermore, the total computation bit maximization problem (P0) is solved, including:

[0052] Step 3-1: Given Solve the problem to obtain the suboptimal solution t*, p of variables t, p, f * 、f * ;

[0053] Step 3-2, given {t,p,f} and By introducing the auxiliary variable β k 、z k，i and c k , obtain the suboptimal solution W* of W through continuous convex approximation;

[0054] Step 3-3, given {t, p, f} and W, use the GA-PSO algorithm to solve the problem and obtain and Suboptimal solution and

[0055] Step 3-4: Repeat steps 3-1 to 3-3 until the total computational bits of the network tend to a fixed value, and obtain the suboptimal solution to the final optimization problem. * 、p * 、f * , W* , and

[0056] Furthermore, step 3-1 is specifically as follows:

[0057] In a given After that, the total network computation bit maximization problem P0 is rewritten as problem P1, which is expressed as:

[0058] P1:

[0059] st.C1′

[0060] C5′:

[0061] C2:

[0062] C4:

[0063]

[0064] in as well as The maximum total computation bits are obtained when each WD exhausts its available energy, so C1 is updated as:

[0065]

[0066] Will Defined in Rewritten as:

[0067]

[0068] Based on the above formula, the P1 problem is rewritten as the convex optimization problem P1.1, which is expressed as:

[0069] P1.1:

[0070] stC2:

[0071] C5′:

[0072] Solve the convex optimization problem P1.1 using the CVX tool and obtain the corresponding suboptimal solution t * 、p * .

[0073] Furthermore, step 3-2 is specifically as follows:

[0074] Given {t,p,f}, by introducing auxiliary variables β k 、zk，i and c k , rewrite the total network computing bit maximization problem P0 into problem P2, problem P2 is expressed as, problem P2 is expressed as:

[0075] P2:

[0076] stC1:

[0077] C3:

[0078] C6:

[0079] C7:

[0080] C8: || t -t n′ ||2≥D,1≤n≠n′≤N,

[0081] C9: ||r m -r m′ ||2≥D, 1≤m≠m′≤N,

[0082] Among them, the variable β k =q k / (E k -E c，k ) represents the energy distribution between WD-k’s active offloading and local computation; β k (E k -E c，k ) and (1-β k )(E k -E c,k ) represent the energy allocated to active offloading and local computing, respectively;

[0083] Based on this, the number of bits calculated for active offloading is:

[0084]

[0085] The number of bits calculated locally is:

[0086]

[0087] By and Substituting into problem P2, problem P2 can be rewritten as problem P2.1, expressed as:

[0088] P2.1:

[0089] stC3:

[0090] C6:

[0091] C7:

[0092] C8: ||t n -t n′ ||2≥D,1≤n≠n′≤N

[0093] C9: ||r m -r m′ ||2≥D,1≤m≠m′≤N;

[0094] because The non-convexity of the above problem still exists due to the coupling of W and Two related sub-problems are solved separately;

[0095] Next, given {t,p,f} and And by introducing the auxiliary variable z k，i and c k , simplifying problem P2.1 to problem P2.2, expressed as:

[0096] P2.2:

[0097] stC3:

[0098] C10:z k，i ≥exp(-a k (tr(B k W k )-b k )),

[0099] C11: c k ≥exp(-a k ((1-ρ)tr(B k W k )-b k )),

[0100] where z k，i and c k It is an auxiliary variable introduced to solve the non-convexity of the objective function.

[0101] Based on this, Rewrite as This is with z k，i and c k Related non-convex problems;

[0102] By applying the continuous convex approximation SCA method, we get E k The approximate value of is:

[0103]

[0104] in, and are respectively z in the sth iteration k，i and c k Based on this, problem P2.2 becomes a convex problem. We use the CVX toolbox to solve it and obtain the corresponding suboptimal solution W * .

[0105] Furthermore, steps 3-4 are specifically as follows:

[0106] Given {t,p,f} and W, Problem P2.1 is simplified to Problem P2.3:

[0107] P2.3:

[0108] stC6:

[0109]

[0110] C8: ||t n -t n′ ||2≥D, 1≤n≠n′≤N,

[0111] C9: ||r m -r m′ ||2≥D, 1≤m≠m′≤N;

[0112] Use GA-PSO algorithm to solve the problem. First, initialize J particles, whose positions are Speed Specifically, the particles The position of represents a possible solution to Problem P2.3, where particle According to their personal best position and the global best position among all particles In the t-th iteration of the GA-PSO algorithm, the particle The speed and position are modified to:

[0113]

[0114] Among them, c1 and c2 are personal learning factor and global learning factor respectively; and are two parameters uniformly generated in [0,1]; ω is the inertia weight that controls the convergence performance and is updated as:

[0115]

[0116] where ω max and ω min It limits the range of ω; d1 and d2 are control factors; is a constant, indicating the maximum number of iterations; to ensure the constraints and That is, the solution obtained should be within the feasible region, and the following mapping operations are performed:

[0117]

[0118] in, express The i-th element of ;

[0119] In order to evaluate the impact of position optimization on the total computational bits, the fitness function is defined as:

[0120]

[0121] In the formula, use Represents the total computational bits; is the location set of MAs pairs that violate C8 and C9 on AP and MEC servers respectively;

[0122] express The cardinality of is a large penalty parameter that guarantees constraints on C8 and C9, otherwise will be less than zero;

[0123] By evaluating the particles The fitness of

[0124]

[0125] and Update the individual and global best positions of the particles until convergence.

[0126] Furthermore, the GA-PSO algorithm includes crossover and mutation operations. The crossover probability p in the operation c and mutation probability p m Greater than a random number , the crossover and mutation operations are re-executed as follows:

[0127]

[0128] in and Respectively represent The random number in crossover and mutation, and represent the jth and lth crossover particles randomly selected at fixed crossover positions, respectively. represents the mutation selected at the i-th MA position particle;

[0129] Check whether the updated solution satisfies constraints C6 and C7; if not, perform crossover and mutation operations again.

[0130] The beneficial effects described in the present invention are as follows: the present invention deploys MAs on APs and MEC servers to make full use of spatial DoFs, thereby improving the energy transmission efficiency from APs to WDs and the task offloading efficiency from WDs to MEC servers; based on the MA enhancement scheme, an efficient time slot scheduling scheme is constructed to make full use of the characteristics of active and passive task offloading, thereby improving the amount of energy collected and the number of offloaded tasks; further, a network performance optimization problem is defined, and the problem is solved using the designed block coordinate descent algorithm, thereby achieving network performance optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0131] Figure 1 is a WPB-MEC system model diagram supporting MA provided by an embodiment;

[0132] Figure 2 It is the convergence behavior curve of GA-PSO algorithm;

[0133] Figure 3 It is a graph of the total calculated bits versus AP transmit power;

[0134] Figure 4 It is a graph of the total computational bits versus the number of WDs;

[0135] Figure 5 It is a graph of the total computational bits versus the number of MAs;

[0136] Figure 6 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0137] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments in conjunction with the accompanying drawings.

[0138] like Figure 6As shown, in the field of smart grid and vehicle network interaction described in the present invention, a MA-enhanced mobile edge cloud computing network performance optimization method comprises the following steps:

[0139] Step 1: Build the MA-enhanced WPB-MEC system;

[0140] Step 2: Based on the MA-enhanced WPB-MEC system, the total calculation bits of the WPB-MEC system are calculated according to the transmit covariance matrix at the AP, the network time resource allocation, the transmit power and calculation frequency of the WDs, and the position of the MAs;

[0141] Step 3: Design a block coordinate descent algorithm to optimize the time allocation scheme within a specified transmission block, the transmit covariance matrix at the AP, the calculation frequency of WDs, the transmit power, and the location of MAs, and establish an optimization problem to maximize the total computational bits of the system.

[0142] The WPB-MEC system model supporting MA adopted in the embodiment of the present invention is shown in FIG. Figure 1 As shown. The WPB-MEC system includes an access point AP with N movable antennas MAs, a MEC server with M movable antennas MAs and K wireless devices WDs, wherein the set of K wireless devices WDs is {WD-1, WD-2, …, WD-K}, wherein k and K are pre-set values, k represents the index value of WDs, k≤K; each WD is an energy-constrained device and is equipped with a single fixed-position antenna FPA.

[0143] Each WD can backscatter the incident signal from the AP to achieve passive offloading from WDs to the MEC server. In addition, it can also implement a collection-and-re-offload protocol to actively offload its tasks to the MEC server using the previously collected energy. Considering the bit-by-bit independence and arbitrary split functions of the tasks of WDs, a partial offloading strategy is adopted. The MAs of the AP and MEC server are connected to their respective RF chains using flexible cables, so that the MAs are located in their respective two-dimensional (2D) areas C of size A×A. t and C r The location in the network can be moved to establish a more efficient transmission connection.

[0144] The duration of the transmission block in the system is T seconds, which includes four execution phases: passive unloading, active unloading, edge computing phase and result feedback phase. In order to avoid interference, TDMA is used to divide the passive unloading and active unloading phases into K sub-phases respectively. In the kth subphase of the passive offloading phase of 10 seconds, WD-k backscatters part of its tasks to the MEC server, where At the same time, the remaining WDs are able to harvest energy from the AP to increase the total amount of harvested energy. In the kth subphase of the active offloading phase of 10 seconds, WD-k actively offloads part of its tasks to the MEC server by utilizing its harvested energy.

[0145] The channel model in the system is as follows:

[0146] Assume t n =[x n ,y n ],r m =[x m ,y m ] represent the nth MA position of the AP and the mth MA position of the MEC server, respectively. x n ,y n is the horizontal and vertical coordinates of the nth MA position of AP, x m ,y m is the horizontal and vertical coordinates of the mth MA position of the MEC server. The set of MAs positions on the AP and MEC server is expressed as

[0147] The number of transmission channel paths between AP and WD-k and between WD-k and MEC server are respectively and To express it, WD-k is the first position between the nth MA position and the AP reference point. The signal propagation phase difference of the paths is expressed as:

[0148]

[0149] In the formula, and Respectively represent the first t The elevation and azimuth of the transmission path. The channel vector from AP to WD-k Expressed as

[0150]

[0151] is the transmitting field response matrix, λ is the carrier wavelength, is the path response vector between AP and WD-k.

[0152] Similarly, the channel vector from WD-k to the MEC server is Expressed as

[0153]

[0154] in represents the corresponding path response vector, is the receiving field response matrix of the MEC server, and They are the first r The elevation and azimuth angles of the receiving paths.

[0155] The propagation model in the system is as follows:

[0156] 1) Passive unloading stage: In this stage, the AP transmits energy signals to achieve passive unloading and WPT. k for During the period, the AP transmits the covariance matrix. For passive offloading, the calculation bit of WD-k offloading is expressed as:

[0157]

[0158] in, B is the channel bandwidth, ρ is the reflection coefficient, is the noise power spectral density and ξ represents the performance gap caused by using the actual modulation in the passive unloading phase.

[0159] In the passive unloading phase, WPT can be performed on WD-k. For WD-k, part of its received power (ρtr(B k W k )) is used to implement Passive unloading during the period, and the remaining power ((1-ρ)tr(B k W k )) is converted into direct current and stored in the battery of WD-k. During the entire passive unloading stage, except Except for the corresponding sub-phase, WD-k can always collect energy from AP. According to the nonlinear energy harvesting model, the energy obtained by WD-k in the passive unloading phase is

[0160]

[0161] in,

[0162] P sat is the saturation power, a k and b k are the control parameters of the energy harvesting circuit of WD-k;

[0163] 2) Active unloading stage: Let p kis the transmission power of WD-k. The calculation bits of active unloading during the period are expressed as:

[0164]

[0165] Assume E c,k represents a constant circuit energy consumption, then the energy consumption of WD-k in the active unloading stage is expressed as

[0166] 3) Local calculation: The calculation bits of WD-k calculated locally are:

[0167]

[0168] Among them, f k represents the calculation frequency of WD-k, represents the computational complexity of the task at WD-k. The energy required to perform local computation at WD-k is Indicates that, where ε k is the effective capacitance coefficient of the on-chip processor located at WD-k.

[0169] Optimize the network time allocation scheme, the transmit covariance matrix at the AP, the calculation frequency and transmit power of WDs, and the location of MAs. Since the optimal solution is difficult to obtain, the corresponding process of obtaining the suboptimal solution includes:

[0170] According to the location of MAs in AP and MEC server and Channel vector from AP to WD-k Channel vector from WD-k to MEC server The transmit covariance matrix at AP is W = {W k ,k∈K}, the transmission power of WDs p={p k ,k∈K}, calculate the frequency f={f k , k∈K, we establish the problem P0 of maximizing the total computational bits of the network, which can be expressed as:

[0171] P0:

[0172] stC1:

[0173] C2:

[0174] C3:

[0175] C4:

[0176] C5:

[0177] C6:

[0178] C7:

[0179] C8: ||t n -t n′ ||2≥D,1≤n≠n′≤N

[0180] C9: ||r m -r m′ ||2≥D,1≤m≠m′≤Nwhere P H is the maximum transmit power of the AP, is the maximum calculation frequency of WD-k, D is the minimum distance between any two MAs; ||·||2 represents the Euclidean norm;

[0181] Solving the total computational bit maximization problem (P0) comprises:

[0182] Step 3-1: Given Solve the problem to obtain the suboptimal solutions t*, p*, f* of variables t, p, f;

[0183] Step 3-2, given {t,p,f} and By introducing the auxiliary variable β k 、z k，i and c k , obtain the suboptimal solution W* of W through continuous convex approximation;

[0184] Step 3-3, given {t, p, f} and W, use the GA-PSO algorithm to solve the problem and obtain and Suboptimal solution and

[0185] Step 3-4: Repeat steps 3-1 to 3-3 until the total computational bits of the network tend to a fixed value, and obtain the suboptimal solution to the final optimization problem. * 、p * 、f * , W * , and

[0186] Among them, step 3-1 is specifically: optimize the calculation time, transmission power, and calculation frequency of WDs under the conditions of the given AP's transmission covariance matrix, the MAs position on the AP, and the MAs position on the MEC server, that is, fix Obtain the suboptimal solution t of variables t, p, and f by solving the problem* 、p * 、f * The specific process includes:

[0187] In a given After that, the total network computation bit maximization problem P0 is rewritten as problem P1, which is expressed as:

[0188] P1:

[0189] stC1′:

[0190] C5′:

[0191] C2:

[0192] C4:

[0193]

[0194] in as well as The maximum total computation bits are obtained when each WD exhausts its available energy. Therefore, C1 is updated as:

[0195]

[0196] Will Defined in Rewritten as:

[0197]

[0198] Based on the above formula, the P1 problem can be rewritten as the convex optimization problem P1.1. Problem P1.1 is expressed as:

[0199] P1.1:

[0200] stC2:

[0201] C5′:

[0202] Solve the convex optimization problem P1.1 using the CVX tool and obtain the corresponding suboptimal solution t * 、p * .

[0203] Step 3-2 is as follows: Optimize the AP's transmit covariance matrix under the given WDs calculation time, transmit power, calculation frequency, MAs location on the AP, and MAs location on the MEC server, that is, fix {t, p, f} and Optimize W. By introducing auxiliary variables β k 、z k，i and c k , rewrite the total network computation bit maximization problem P0 into problem P2, and problem P2 is expressed as:

[0204] P2:

[0205] stC1:

[0206] C3:

[0207] C6:

[0208] C7:

[0209] C8:||t n -t n′ ||2≥D,1≤n≠n′≤N,

[0210] C9: ||r m -r m′ ||2≥D,1≤m≠m′≤N,

[0211] Among them, the variable β k =q k / (E k -E c，k ) represents the energy distribution between WD-k’s active offloading and local computation; k (E k -E c,k ) and (1-β k )(E k -E c,k ) represent the energy allocated to active offloading and local computing, respectively.

[0212] Based on this, the number of bits calculated for active offloading is:

[0213]

[0214] The number of bits calculated locally is:

[0215]

[0216] By and Substituting into problem P2, problem P2 can be rewritten as problem P2.1, expressed as:

[0217] P2.1:

[0218] stC3:

[0219] C6:

[0220] C7:

[0221] C8: ||t n -t n′ ||2≥D,1≤n≠n′≤N

[0222] C9: ||r m -r m′ ||2≥D, 1≤m≠m′≤N;

[0223] because The non-convexity of the above problem still exists due to the coupling of W and Two related sub-problems are solved separately;

[0224] Next, given {t,p,f} and And by introducing the auxiliary variable z k，i and c k , simplifying problem P2.1 to problem P2.2, expressed as:

[0225] P2.2:

[0226] stC3:

[0227] C10:z k，i ≥exp(-a k (tr(B k W k )-b k )),

[0228] C11: c k ≥exp(-a k ((1-ρ)tr(B k W k )-b k )),

[0229] where z k，i and c k It is an auxiliary variable introduced to solve the non-convexity of the objective function.

[0230] Based on this, Can be rewritten as This is with z k，i and c kRelated non-convex problems. By applying SCA, we get E k The approximate value of in, and are respectively z in the sth iteration k，i and c k Based on this, problem P2.2 becomes a convex problem, which can be solved using the CVX toolbox to obtain the corresponding suboptimal solution W * .

[0231] Step 3-3 is as follows: Optimize the position of MAs on the AP and the position of MAs on the MEC server under the conditions of given WDs calculation time, transmission power, calculation frequency, and AP transmission covariance matrix, that is, fix {t, p, f} and W optimization Rewrite the total network computation bit maximization problem P2.1 into P2.3, and the problem P2.3 is expressed as:

[0232] P2.3:

[0233] st.C6:

[0234] C7:

[0235] C8: ||t n -t n′ ||2≥D,1≤n≠n′≤N,

[0236] C9: ||r m -r m′ ||2≥D,1≤m≠m′≤N;

[0237] Considering that the solution space of problem P2.3 is very large, the GA-PSO algorithm is used to solve the problem. In the GA-PSO algorithm, firstly, J particles are initialized, and their positions are Speed Specifically, the particles The position of represents a possible solution to Problem P2.3, where particle According to their personal best position and the global best position among all particles In the t-th iteration of the GA-PSO algorithm, the particle The speed and position are modified to:

[0238]

[0239] Among them, c1 and c2 are personal learning factor and global learning factor respectively; and are two parameters uniformly generated in [0,1]; ω is the inertia weight that controls the convergence performance, which is updated as:

[0240]

[0241] where ω max and ω min It limits the range of ω; d1 and d2 are control factors; is a constant, indicating the maximum number of iterations; to ensure the constraints and That is, the solution obtained should be within the feasible region, and the following mapping operations are performed:

[0242]

[0243] in, express The i-th element of ;

[0244] In order to evaluate the impact of position optimization on the total computational bits, the fitness function is defined as:

[0245]

[0246] In the formula, use Represents the total computational bits; is the location set of MAs pairs that violate C8 and C9 on AP and MEC servers respectively;

[0247] express The cardinality of is a large penalty parameter to ensure constraints on C8 and C9, otherwise will be less than zero.

[0248] By evaluating the particles The fitness of

[0249]

[0250] and Update the individual and global best positions of the particles until convergence.

[0251] The GA-PSO algorithm includes crossover and mutation operations. Specifically, when the The crossover probability p in the operation c and mutation probability p m Greater than a random number , the crossover and mutation operations are re-executed as follows:

[0252]

[0253] in and Respectively represent The random number in crossover and mutation, and represent the jth and lth crossover particles randomly selected at fixed crossover positions, respectively. represents the mutation selected at the i-th MA position After completing these operations, check whether the updated solution satisfies the constraints C6 and C7. If not, perform crossover and mutation operations again.

[0254] The method of the present invention is described below in conjunction with specific experiments.

[0255] Consider a two-dimensional simulation environment, where the locations of the AP and MEC server are set to (-15m, 0) and (15m, 0), respectively, and the WDs are evenly distributed in a circle with a center of (0, 0) and a radius of 10m. The mobile area of ​​the MAs on the AP and MEC server is set to a square area [-3λ / 2, 3λ / 2]×[-3λ / 2, 3λ / 2], where λ = 0.1m. The number of sending paths and receiving paths is set to be the same, that is, Assume q k and Each element in follows Where C0 = -10dBm represents large-scale fading with a reference distance of 1m, α = 2.8 is the path loss exponent, and d represents the distance between two nodes. Other parameters are expressed as follows: H =30dBm, T=1s, N=M=10, K=4, B=0.1MHz, ρ=0.8, ξ=0.0316, E c,k =10 -10 J, D = 0.5λ, A = 3λ, a k =1500, b k =0.0022, P sat =0.024, cycles / bits, ε k =10 -26 Watt / Hz 3 , J=300, c1=c2=1.4, d1=0.2, d2=7, ω max =0.9,ω min =0.4, p c =0.7, p m =0.3,

[0256] For performance comparison, the MA scheme based on the GA-PSO algorithm is compared with the following reference schemes: i) MA scheme based on the particle swarm optimization (PSO) algorithm: the positions of MAs in the MA scheme are optimized using the standard PSO algorithm; ii) FPA scheme: all antenna positions are fixed and necessary parameter optimization is considered; iii) Full offload scheme: for the invented MA scheme, WDs can only perform their tasks remotely through the MEC server; iv) Full local scheme: WDs can only use their local processors to compute their tasks.

[0257] Figure 2 The relationship between the total computational bits of the system and the number of iterations is shown. Figure 2 As shown in Figure 1, with the increase of the number of iterations, the overall computational bit performance obtained by the GA-PSO algorithm is significantly better than that of the standard PSO algorithm. This is because the GA-PSO algorithm implements crossover and mutation operations and has strong global search capabilities and particle diversity.

[0258] Figure 3 The relationship curve of the total computational bits of the system as a function of the maximum transmit power of the AP is shown. Obviously, increasing the transmit power can increase the total computational bits of all schemes. Compared with the MA scheme using the PSO algorithm, the MA scheme using the GA-PSO algorithm achieves a higher total computational bits, which confirms that the use of the GA-PSO algorithm can improve the global search capability and avoid falling into the local optimal solution. Compared with the FPA scheme, using MAs on the AP and MEC server can ensure better system performance. This is because MAs can create favorable channel conditions between APs and WDs and between WDs and MEC servers.

[0259] Figure 4 The relationship curve of the total computational bits of the system as a function of the number of WDs is shown. As K increases, the total computational bits of all schemes are improved. The performance gap between the schemes using GA-PSO and PSO algorithms widens with the increase of WDs, which again proves that the GA-PSO algorithm is superior to the PSO algorithm. In addition, compared with the full offloading scheme and the full local scheme, the hybrid offloading scheme using the GA-PSO algorithm can increase the total computational bits by 34.6% and 286.6%, respectively.

[0260] Figure 5The relationship between the total system computing bits and the number of MAs is shown. It is obvious that the number of MAs has a positive impact on the total system computing bits. This is because more MAs can achieve higher antenna gain from WDs to AP / MEC servers. Similar to Figure 3 and Figure 4 , the total computation bit achieved by the GA-PSO algorithm is the largest. The simulation results show that the use of MAs in the WPB-MEC system of the present invention can greatly increase the performance of the MEC network.

[0261] The above description is only a preferred embodiment of the present invention and is not intended to be a further limitation of the present invention. All equivalent changes made using the contents of the present specification and drawings are within the protection scope of the present invention.

Claims

1. A method for optimizing network performance of mobile edge cloud computing enhanced by MA, characterized in that: The following steps are involved: Step 1: construct an MA-enhanced WPB-MEC system, wherein the WPB-MEC system includes an access point AP with N movable antennas MAs, an MEC server with M movable antennas MAs, and K wireless devices WDs; The movable antennas MAs of the AP and MEC server are connected to their respective radio frequency RF chains through flexible cables, so that the MAs are in their respective two-dimensional areas C of size A×A. t and C r The position in can be moved; The tasks of each WD can be divided. Each WD is equipped with a single fixed-position antenna FPA. Based on the backscatter communication circuit and active RF circuit, each WD uses passive task offloading or active task offloading to offload the corresponding computing tasks to the MEC server. Step 2: Based on the MA-enhanced WPB-MEC system, the total calculation bits of the WPB-MEC system are calculated according to the transmit covariance matrix at the AP, the network time resource allocation, the transmit power and calculation frequency of the WDs, and the position of the MAs; Step 3: Design a block coordinate descent algorithm to optimize the time allocation scheme within a specified transmission block, the transmit covariance matrix at the AP, the calculation frequency of WDs, the transmit power, and the location of MAs, and establish an optimization problem to maximize the total computational bits of the system.

2. According to claim 1, a method for optimizing network performance of mobile edge cloud computing enhanced by MA is characterized in that: The set of K wireless devices WDs is {WD-1, WD-2, …, WD-K}. The duration of the transmission block in the system is T seconds. TDMA is used to divide the passive task offloading and active task offloading into K sub-phases respectively. In the kth sub-phase of the passive offloading phase of 10 seconds, WD-k offloads part of the tasks to the MEC server in the form of passive offloading based on the backscattering circuit; The subscript k in the figure corresponds to the kth sub-stage and the kth WD, and the superscript b indicates the passive unloading stage. At the same time, the remaining WDs collect energy from the AP to increase the total amount of energy collected; The kth sub-phase of the active unloading phase of seconds, The superscript a in the middle indicates the active unloading stage. WD-k uses the collected energy to actively unload another part of the tasks to the MEC server; specifically: 1) In the passive task offloading stage, the AP transmits an energy signal to achieve passive offloading and wireless energy transmission WPT, specifically: Let W k for The AP transmits the covariance matrix during the period; for passive offloading, the calculation bit of WD-k offloading is expressed as: in, B is the channel bandwidth, ρ is the reflection coefficient, is the noise power spectral density, ξ represents the performance gap caused by using the actual modulation in the passive unloading phase; In the passive unloading phase, WPT is performed on WD-k; for WD-k, part of its received power ρtr(B k W k ) is used to implement Passive unloading during the period, and the remaining power (1-ρ)tr(B k W k ) is converted into direct current and stored in the battery of WD-k; During the entire passive unloading phase, except for Outside the corresponding sub-phase, WD-k keeps collecting energy from AP. According to the nonlinear energy harvesting model, the energy obtained by WD-k in the passive unloading phase is: in, represents the energy collected by WD-k in the kth sub-stage of the passive unloading phase, represents the duration of the kth sub-phase of the passive unloading phase, represents the duration of the i-th sub-phase of the passive unloading phase, i≠k; represents the received power of WD-k in the i-th sub-phase of the passive unloading phase, i≠k; represents the logic function in the energy harvesting circuit of WD-k, P sat is the saturation power, a k and b k are the control parameters of the energy harvesting circuit of WD-k; 2) In the active task offloading phase, let p k is the transmission power of WD-k, WD-k is The calculation bits of active unloading during the period are expressed as: Assume E c,k represents a constant circuit energy consumption, then the energy consumption of WD-k in the active unloading stage is expressed as 3) The calculation bits of WD-k calculated locally are: Among them, f k represents the calculation frequency of WD-k, represents the computational complexity of the task at WD-k; the energy required to perform local computation at WD-k is Indicates that, where ε k is the effective capacitance coefficient of the on-chip processor located at WD-k.

3. According to claim 2, a method for optimizing network performance of mobile edge cloud computing enhanced by MA is characterized in that: The channel model in the WPB-MEC system is as follows: Assume t n =[x n ,y n ],r m =[x m ,y m ] represent the nth MA position of the AP and the mth MA position of the MEC server, respectively. x n ,y n is the horizontal and vertical coordinates of the nth MA position of AP, x m ,y m is the horizontal and vertical coordinates of the mth MA position of the MEC server; the set of MAs positions on the AP and MEC server is expressed as The number of transmission channel paths between AP and WD-k and between WD-k and MEC server are respectively and To express it, then the nth MA position of WD-k and the lth reference point between the AP t The signal propagation phase difference of the paths is expressed as: In the formula, and Respectively represent the first t The elevation and azimuth angles of the transmission paths; Channel vector from AP to WD-k It is expressed as: is the transmitting field response matrix, is the corresponding vector of the transmission field of the nth MA in the AP, λ is the wavelength of the carrier, j is the imaginary sign, is the path response vector between AP and WD-k; Similarly, the channel vector from WD-k to the MEC server is It is expressed as: in represents the corresponding path response vector, is the receiving field response matrix of the MEC server, is the corresponding vector of the mth receiving field in the MEC server, represents the lth distance between the mth MA position of the MEC server and the reference point r The signal propagation phase difference of the paths is and They are the first r The elevation and azimuth angles of the receiving paths, 4. According to claim 3, a method for optimizing network performance of mobile edge cloud computing enhanced by MA is characterized in that: Construct the optimization problem, specifically: Based on the location of MAs in AP and MEC server and Channel vector from AP to WD-k Channel vector from WD-k to MEC server The transmit covariance matrix at AP is W = {W k ,k∈K}, the transmission power of WDs p={p k ,k∈K}, calculate the frequency f={f k , k∈K, we establish the problem P0 of maximizing the total computational bits of the network, which can be expressed as: AFTER: C8:||t n -t n’ ||2≥D,1≤n≠n'≤N C9:||r m -r m’ ||2≥D,1≤m≠m'≤N Where P H is the maximum transmit power of the AP, is the maximum calculation frequency of WD-k, D is the minimum distance between any two MAs; ||·||2 represents the Euclidean norm.

5. According to claim 4, a method for optimizing network performance of mobile edge cloud computing enhanced by MA is characterized in that: Solve the total computational bit maximization problem (P0), including: Step 3-1: Given Solve the problem to obtain the suboptimal solution t of variables t, p, and f * 、p * 、f * ; Step 3-2, given {t,p,f} and By introducing the auxiliary variable β k 、z k,i and c k , obtain the suboptimal solution W of W through continuous convex approximation * ; Step 3-3: Given {t, p, f} and W, use the GA-PSO algorithm to solve the problem and obtain the suboptimal solution of t% and r% and Step 3-4: Repeat steps 3-1 to 3-3 until the total computational bits of the network tend to a fixed value, and obtain the suboptimal solution to the final optimization problem. * 、p * 、f * , W * , and 6. The method for optimizing network performance of mobile edge cloud computing enhanced by MA according to claim 5, characterized in that: Step 3-1 is as follows: In a given After that, the total network computation bit maximization problem P0 is rewritten as problem P1, which is expressed as: P1： in as well as The maximum total computation bits are obtained when each WD exhausts its available energy, so C1 is updated as: Will Defined in Rewritten as: Based on the above formula, the P1 problem is rewritten as the convex optimization problem P1.1, which is expressed as: P1.1： Solve the convex optimization problem P1.1 using the CVX tool and obtain the corresponding suboptimal solution t * 、p * .

7. The method for optimizing network performance of mobile edge cloud computing enhanced by MA according to claim 6, characterized in that: Step 3-2 is as follows: Given {t,p,f}, by introducing auxiliary variables β k 、z k，i and c k , rewrite the total network computing bit maximization problem P0 into problem P2, problem P2 is expressed as, problem P2 is expressed as: P2： C8:||t n -t n’ ||2≥D,1≤n≠n'≤N, C9:||r m -r m’ ||2≥D,1≤m≠m'≤N, Among them, the variable β k =q k / (E k -E c,k ) represents the energy distribution between WD-k’s active offloading and local computation; β k (E k -E c,k ) and (1-β k )(E k -E c,k ) represent the energy allocated to active offloading and local computing, respectively; Based on this, the number of bits calculated for active offloading is: The number of bits calculated locally is: By and Substituting into problem P2, problem P2 can be rewritten as problem P2.1, expressed as: P2.1： C8:||t n -t n’ ||2≥D,1≤n≠n'≤N C9: ||r m -r m’ ||2≥D, 1≤m≠m'≤N; due to The non-convexity of the above problem still exists due to the coupling of W and Two related sub-problems are solved separately; Next, given {t,p,f} and And by introducing the auxiliary variable z k，i and c k , simplifying problem P2.1 to problem P2.2, expressed as: P2.2： where z k,i and c k It is an auxiliary variable introduced to solve the non-convexity of the objective function. Based on this, Rewrite as This is with z k,i and c k Related non-convex problems; By applying the continuous convex approximation SCA method, we get E k The approximate value of is: in, and are respectively z in the sth iteration k,i and c k Based on this, problem P2.2 becomes a convex problem. We use the CVX toolbox to solve it and obtain the corresponding suboptimal solution W * .

8. The method for optimizing network performance of mobile edge cloud computing enhanced by MA according to claim 7, characterized in that: Steps 3-4 are as follows: Given {t,p,f} and W, Problem P2.1 is simplified to Problem P2.3: P2.3: C8:||t n -t n’ ||2≥D,1≤n≠n'≤N, C9:||r m -r m’ ||2≥D,1≤m≠m'≤N; Use GA-PSO algorithm to solve the problem. First, initialize J particles, whose positions are Speed Specifically, the particles The position of represents a possible solution to Problem P2.3, where particle According to their personal best position and the global best position among all particles In the t-th iteration of the GA-PSO algorithm, the particle The speed and position are modified to: Among them, c1 and c2 are personal learning factor and global learning factor respectively; and are two parameters uniformly generated in [0,1]; ω is the inertia weight that controls the convergence performance and is updated as: where ω max and ω min It limits the range of ω; d1 and d2 are control factors; is a constant, indicating the maximum number of iterations; to ensure the constraints and That is, the solution obtained should be within the feasible region, and the following mapping operations are performed: in, express The i-th element of ; In order to evaluate the impact of position optimization on the total computational bits, the fitness function is defined as: In the formula, use Represents the total computational bits; is the location set of MAs pairs that violate C8 and C9 on AP and MEC servers respectively; express The cardinality of is a large penalty parameter to ensure constraints on C8 and C9, otherwise will be less than zero; By evaluating the particles The fitness of and Update the individual and global best positions of the particles until convergence.

9. The method for optimizing network performance of mobile edge cloud computing enhanced by MA according to claim 8, characterized in that: The GA-PSO algorithm includes crossover and mutation operations. The crossover probability p in the operation c and mutation probability p m Greater than a random number , the crossover and mutation operations are re-executed as follows: in and Respectively represent The random number in crossover and mutation, and represent the jth and lth crossover particles randomly selected at fixed crossover positions, respectively. represents the mutation selected at the i-th MA position particle; Check whether the updated solution satisfies constraints C6 and C7; if not, perform crossover and mutation operations again.

Images