CN111953730B - NOMA-based server cooperation edge computing method - Google Patents
NOMA-based server cooperation edge computing method Download PDFInfo
- Publication number
- CN111953730B CN111953730B CN202010633925.5A CN202010633925A CN111953730B CN 111953730 B CN111953730 B CN 111953730B CN 202010633925 A CN202010633925 A CN 202010633925A CN 111953730 B CN111953730 B CN 111953730B
- Authority
- CN
- China
- Prior art keywords
- server
- user
- far
- end server
- stage
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000004364 calculation method Methods 0.000 title claims abstract description 40
- 206010042135 Stomatitis necrotising Diseases 0.000 title claims abstract description 34
- 201000008585 noma Diseases 0.000 title claims abstract description 34
- 238000005516 engineering process Methods 0.000 claims abstract description 7
- 238000000034 method Methods 0.000 claims description 34
- 238000005265 energy consumption Methods 0.000 claims description 31
- 238000004891 communication Methods 0.000 claims description 15
- 230000005540 biological transmission Effects 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 12
- 230000009365 direct transmission Effects 0.000 claims description 5
- 230000014509 gene expression Effects 0.000 claims description 5
- 239000000654 additive Substances 0.000 claims description 2
- 230000000996 additive effect Effects 0.000 claims description 2
- 230000001934 delay Effects 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 230000008030 elimination Effects 0.000 claims 1
- 238000003379 elimination reaction Methods 0.000 claims 1
- 238000005457 optimization Methods 0.000 description 44
- 238000004422 calculation algorithm Methods 0.000 description 12
- 238000000638 solvent extraction Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000009795 derivation Methods 0.000 description 4
- 230000002829 reductive effect Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000009977 dual effect Effects 0.000 description 2
- 230000036961 partial effect Effects 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000001010 compromised effect Effects 0.000 description 1
- 239000013256 coordination polymer Substances 0.000 description 1
- 238000012888 cubic function Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 229920003087 methylethyl cellulose Polymers 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
- H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
- H04L67/1004—Server selection for load balancing
- H04L67/1008—Server selection for load balancing based on parameters of servers, e.g. available memory or workload
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Abstract
The embodiment of the application provides a server cooperation edge computing method based on NOMA, wherein a user acquires computing resources on each server, and according to the acquired computing resources on each server, the user adopts the NOMA access technology and a preset unloading strategy to unload part of tasks to each server. Because the application considers the calculation resources on each server when the user unloads the task amount to each server, for example, the user unloads the task amount to two servers, namely, a near-end server and a far-end server, when the calculation resources on the near-end server are less and the calculation resources on the far-end server are more, the application adopts a preset unloading strategy to unload less tasks to the near-end server and more tasks to the far-end server, thereby effectively balancing the resources on the near-end server and the far-end server.
Description
Technical Field
The invention relates to the field of communication, in particular to a server cooperation edge computing method based on NOMA.
Background
Mobile edge computing (Mobile Edge Computing, MEC) is a new computing paradigm introduced by 3GPP in the 5G architecture, i.e. a MEC server with computing, storage and network resources is equipped at a base station or a wireless access point, and mobile users can upload their own urgent intensive computing tasks to the cloud to reduce the computing delay of the tasks and save their own power. Through reasonable communication resources (such as load power, occupied frequency bandwidth and load time), computing resources (such as task allocation and computing resource blocks) and allocation and scheduling of computing modes, mobile edge computing meets the expansion requirement of computing capacity of terminal equipment and overcomes the defect of long cloud computing time delay.
Non-orthogonal multiple access techniques (Non-orthogonal Multiple Access, NOMA) take up an extra advantage due to their efficient spectrum utilization. Non-orthogonal multiple access further increases the information rate of the network and increases the number of mobile users that can be served by the network as compared to conventional orthogonal multiple access techniques. NOMA is used as a user access mode in the MEC system, which is an important innovation for exerting the utilization advantages of spectrum resources and computing resources of the NOMA and the MEC system, and various optimization frameworks show that the NOMA-MEC system is a very promising design scheme, and the NOMA is used as an access mode, so that multiple users can be unloaded simultaneously, and the overall time delay is reduced.
Although MECs play an important role in saving time delay in the computational power consumption of mobile users, some challenges still remain.
Disclosure of Invention
The inventor finds that the computing resources and communication resources of each server are generally unevenly distributed in time and space, which can reduce the unloading execution efficiency of the MEC network to a certain extent, and how to effectively balance the resources on each MEC server is a problem to be solved.
In order to solve the technical problem, the application provides a method for computing edges of server collaboration based on NOMA, when a user unloads task amount to each server, computing resources on each server are considered, for example, the user unloads the task amount to two servers, a near-end server and a far-end server, when the computing resources on the near-end server are fewer and the computing resources on the far-end server are more, a preset unloading strategy is adopted, fewer tasks are unloaded to the near-end server, and more tasks are unloaded to the far-end server, so that the resources on the near-end server and the far-end server are effectively balanced.
The specific technical scheme provided by the application is as follows:
a NOMA-based server collaboration edge computing method, the method comprising:
the user obtains the computing resources on each server;
according to the acquired computing resources on each server, a user adopts NOMA access technology and a preset unloading strategy to unload part of tasks to each server, wherein the preset unloading strategy is obtained by calculation according to preset conditions, and the preset conditions comprise the computing resources on the servers;
when a user unloads tasks to a far-end server, the tasks unloaded to the far-end server are simultaneously transmitted to a near-end server, the near-end server decodes task signals of the far-end server by using the SIC principle and then sends the task signals to the far-end server, the far-end server receives the tasks sent by the user and the tasks forwarded by the near-end server, the two paths of signals are identical, and the far-end server combines the two paths of identical signals by using a maximum ratio combining mode.
Compared with the prior art, the technical scheme has the following advantages:
the embodiment of the application provides a method for calculating edges of server cooperation based on NOMA, wherein a user acquires computing resources on each server, and according to the acquired computing resources on each server, the user adopts the NOMA access technology and a preset unloading strategy to unload part of tasks to each server. Because the application considers the calculation resources on each server when the user unloads the task amount to each server, for example, the user unloads the task amount to two servers, namely, a near-end server and a far-end server, when the calculation resources on the near-end server are less and the calculation resources on the far-end server are more, the application adopts a preset unloading strategy to unload less tasks to the near-end server and more tasks to the far-end server, thereby effectively balancing the resources on the near-end server and the far-end server.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of a method for computing a collaboration edge of a server based on NOMA according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a NOMA-MEC system model based on server collaboration according to one embodiment of the present invention;
FIG. 3 is a timing diagram of server collaboration based offload computation according to an embodiment of the present invention.
Detailed Description
The inventor finds that the computing resources and communication resources of each server are generally unevenly distributed in time and space, which can reduce the unloading execution efficiency of the MEC network to a certain extent, and how to effectively balance the resources on each MEC server is a problem to be solved.
In order to effectively balance resources on each MEC server, the application provides an edge computing method based on NOMA server cooperation, a user obtains computing resources on each server, and according to the obtained computing resources on each server, the user uses NOMA access technology and a preset unloading strategy to unload part of tasks to each server. Because the application considers the calculation resources on each server when the user unloads the task amount to each server, for example, the user unloads the task amount to two servers, namely, a near-end server and a far-end server, when the calculation resources on the near-end server are less and the calculation resources on the far-end server are more, the application adopts a preset unloading strategy to unload less tasks to the near-end server and more tasks to the far-end server, thereby effectively balancing the resources on the near-end server and the far-end server.
As shown in fig. 1, the present application proposes a NOMA-based server collaboration edge computing method, including:
the user obtains the computing resources on each server;
according to the acquired computing resources on each server, a user adopts NOMA access technology and a preset unloading strategy to unload part of tasks to each server, wherein the preset unloading strategy is obtained by calculation according to preset conditions, and the preset conditions comprise the computing resources on the servers;
when a user unloads tasks to a far-end server, the tasks unloaded to the far-end server are simultaneously transmitted to a near-end server, the near-end server decodes task signals of the far-end server by using the SIC principle and then sends the task signals to the far-end server, the far-end server receives the tasks sent by the user and the tasks forwarded by the near-end server, the two paths of signals are identical, and the far-end server combines the two paths of identical signals by using a maximum ratio combining mode.
Therefore, when the user unloads the task amount to each server, the computing resources on each server are considered, for example, the user unloads to two servers, namely, a near-end server and a far-end server, when the computing resources on the near-end server are less and the computing resources on the far-end server are more, the preset unloading strategy is adopted, so that fewer tasks are unloaded to the near-end server, more tasks are unloaded to the far-end server, and the computing resources on the near-end server and the far-end server are effectively balanced.
The inventors have also found that when a user offloads to a remote server, the user may use a larger transmission power to ensure the reliability of the task transmission, which requires the user to consume more power, since the offloaded channel quality is affected by the distance.
In order to solve the problem that when a user unloads a task to a far-end server, the reliability of task transmission needs to be ensured, and larger transmission power is needed to cause the user to consume more energy, the method and the device enable a near-end server to decode task signals of the far-end server by utilizing the SIC principle, then forward the task signals, and enable the far-end server to receive two paths of identical signals from the user and the near-end server to combine and decode in a maximum ratio combining mode. Thus, the user can increase the received signal-to-noise ratio at the remote server without increasing the transmission power, and the reliability of the signal is enhanced. That is, when the user uninstalls the task to the remote server, the task uninstalled to the remote server by the user is forwarded to the remote server by the near-end server without increasing the transmission power of the user, and the remote server combines the received signal from the user and the received signal forwarded by the near-end server (i.e. combines the two paths of the same signals) to ensure the reliability of task transmission.
Further, the present application allows for the near-end server to operate in full duplex mode, completing the reception and transmission of signals in the same time slot, in view of enabling the user tasks to be completed as soon as possible. Thus, the whole cooperative unloading process can be completed in the same time slot, and the defect that the forwarding process occupies another time slot resource in the half duplex mode in the prior art is avoided.
Specifically, the local computing frequency and power allocation of the user, the forwarding power allocation of the near-end server, the allocation decision of the task amount, and the allocation of the offloading and computing time are all factors affecting the system performance (such as energy consumption and time delay). Therefore, the preset unloading strategy is to perform joint optimization on variables related to the factors under the constraint of user time delay to obtain the best system performance, namely, under the constraint of ensuring the completion time delay of the task, an optimization problem model with the minimum energy consumption for task allocation, time allocation, power allocation and local frequency joint optimization is established.
The optimization problem is a non-convex optimization problem, i.e. it cannot be solved in a general way, due to the coupling involved in the individual optimization variables. Based on this, the present application proposes an algorithm based on taylor approximation and iterative joint communication resource and computing resource optimization to solve this non-convex problem. The algorithm is carried out in three steps: 1) Reducing the optimized variable through mathematical derivation to simplify the original problem, namely, firstly optimizing the local frequency to obtain an optimal expression of the local frequency relative to the local task variable; and then in each iteration step, 2) under the given task allocation variable and time allocation variable, simplifying the complex constraint expression firstly, and then converting the power optimization sub-problem into a solvable optimization problem based on a Taylor approximation method. Obtaining closed optimal solutions of three power variables by utilizing Lagrangian dual and a series of mathematical analysis; 3) And under a given power variable, obtaining a closed optimal solution of the task dividing variable and the time distribution variable. The algorithm is low in complexity and converges faster.
By solving the optimization problem through the provided algorithm, the optimal unloading decision can be finally obtained, so that the computing resources on each server are balanced, for example, when the computing resources of the near-end server are fewer and the computing resources of the far-end server are more, the computing resources on the far-end server are fully utilized by the method disclosed by the application, and the load on the near-end server is reduced. And the optimal power distribution of the user and the optimal value of the forwarding power of the near-end server can be obtained, so that the signal reliability at the far-end server is enhanced under the condition that the consumed power of the user is as small as possible.
The above system model proposed in the present application can be briefly described with reference to fig. 2. The method mainly relates to an edge user, two base stations with different distances are respectively provided with MEC servers, the near server is S-A, and the far server is S-B. A user adopts se:Sup>A NOMA mode to unload part of task amount to two servers at the same time, S-A can decode S-B signals and then extract own signals by utilizing serial interference cancellation (successive interference cancellation, SIC) after receiving NOMA signals, therefore, the near server is set to work in se:Sup>A relay mode of full duplex decoding and forwarding, namely, the near server is used for forwarding S-B signals to S-B after the near server decodes the S-B signals, and the S-B receives signals from the user and the S-A in se:Sup>A maximum ratio combining mode.
The optimization problem modeling process required for the system model is described in detail below.
In a first step, basic parameter settings are made for the collaborative NOMA-MEC system. Since the near server operates in full duplex relay mode, it is assumed that both the user and the far server are equipped with one antenna, and the near server is equipped with two antennas for receiving and transmitting signals, respectively. The task execution amount required by the user is recorded as gamma, the input bit number D is contained in the task execution amount, and the maximum completion time of the task is recorded as T. The coherence time of the channel is assumed to be greater than the maximum completion time block T, so the channel state can be unchanged within T. The user adopts a partial unloading mode, and the local server and the two servers can execute tasks in parallel, so the total task bit is divided into d u ,d A And d B And respectively corresponding to the local calculated quantity and the task quantity unloaded to the S-A and the S-B. In summary, the system comprises four wireless links, namely user-S-A, user-S-B, S-A-S-B and S-A-S-A, and the channel gains are respectively |h 1 | 2 ,|h 2 | 2 ,|h 3 | 2 ,|h s | 2 Wherein the channel S-A→S-A is se:Sup>A self-interference channel generated by the near server due to operating in full duplex mode, assuming it is subject to free fading and setting |h s | 2 The remaining three channels are assumed to follow the Rayleigh distribution, with two NOMA channels at |h, for a constant 2 | 2 ≤|h 1 | 2 And (5) sequencing. Based on the above settings, the offload computation process of the present application includes four parts: and the user local calculation, the full duplex cooperation unloading stage, the remote server calculation stage and the calculation result downloading stage. As shown in fig. 3.
And secondly, describing the whole unloading calculation process of the proposed system and the optimization quantity such as energy consumption time delay and the like generated by each stage.
(1) User local computing phase
Set beta u The number of CPU cycles required to calculate an input bit for the user, then d is performed u The total number of cycles required for a bit is beta u d u . And assuming that the user can adjust the operating frequency f per cycle n ∈(0,f u,max ]，n∈{1,…,β u d u To save local energy consumption, where f u,max The maximum CPU frequency for the user. The constraint expression of the local calculation time, the local energy consumption and the frequency can be obtained according to the maximum completion time delay requirement of the task:
wherein, kappa u The energy coefficient for a local server depends on its chip structure.
(2) Full duplex cooperative offloading phase
Since the full duplex collaborative communication computing mode is adopted, the user offloading process is at t up The time period is divided into a direct transmission stage and a cooperative communication stage.
First, the user uses NOMA technique to simultaneously make d in the direct transmission stage A ,d B Corresponding transmission signal
s 1 ,s 2 And unloading to two servers according to the superposition coding principle. After S-A receives the mixed signal, SIC is executed to realize S 2 And after the solution, transmitting the solution to the remote MEC server S-B in a cooperative communication stage. Because the near-end server operates in full duplex mode, server a is subject to residual self-interference from transmit antennas to receive antennas even when the receiver employs different combinations of self-interference cancellation techniques. The received signals at the two servers at the kth time are therefore:
wherein p is 1 ,p 2 Transmit power allocated to server a and server B for user and satisfying p 1 ≤p 2 .p 3 Transmit power z for server a forwarding i I epsilon {1,2} is additive Gaussian white noise at two servers, all obeying Gaussian distribution with mean value of zero and variance of 1; τ is the decoding processing delay of the near server and let τ < T.
After receiving the mixed signal, the server A decodes s according to SIC principle 2 Then s is taken 2 Removal of decoding s from mixed signal 1 . Thus, the SINR of S-B at S-A is:
the SINR of the S-A decoded own signal is:
the signal achievable rate for S-se:Sup>A is:
R 1 ＝Blog 2 (1+γ 1,1 ) (7)
where B is the system bandwidth. Server S-B will receive the mixed signal from the user directly and the signal forwarded by server S-se:Sup>A, assuming that the two signals from user and server S-se:Sup>A are fully solvable at server S-B, so they can be approximately combined by maximal ratio combining. S-B detects NOMA signals transmitted by users as weak users, will S 1 The received SINR is considered as noise decoding its own signal, so this received SINR is:
the received signal-to-noise ratio for detecting the information forwarded from server S-se:Sup>A is:
γ 2,A ＝|h 2 | 2 p 3 (9)
after maximum ratio combination, the total received signal to noise ratio at S-B is obtained as follows:
γ 2,MRC ＝γ 2,2 +γ 2,A (10)
thus, the achievable rate of S-B is:
R 2 ＝Bmin{log 2 (1+γ 1,2 ),log 2 (1+γ 2,MRC )} (11)
to sum up, it can be derived that this stage is at t up Energy consumed by the inner user and the near server:
E up ＝(p 1 +p 2 )t up (12)
E tr ＝p 3 t up (13)
(3) Server computing phase
Although MEC servers are typically connected to the grid, in order to more fully analyze the overall performance of the system, the execution power consumption and computation delay of both servers are also considered. But the transmission bits downloaded as a result are smaller and the transmission delay and energy consumption of this process are therefore negligible. Assuming constant frequency of execution of tasks by MEC servers S-A and S-B, denoted as f, respectively A ,f B The number of cycles required by the two servers to calculate the respective user data bits is denoted as beta A d A ,β B d B . The energy required to be consumed at the server is:
wherein,execution energy, κ, of S-A, S-B, respectively A ,κ B Is the power consumption coefficient of the two servers, depending on their chip structure.
The required computation delays at the two servers are:
in,the execution times at S-A and S-B, respectively.
Since the bit size of the calculation result is much smaller than the task bit of the uploading and the downloading rate is higher than the uploading rate, the time delay and the energy consumption of the stage are negligible.
Thirdly, according to the analysis of the model, the optimization problem of minimizing the total energy consumption of the system is presented. The total energy consumption of the system is mainly divided into two parts, namely transmission energy consumption consumed by the user and the S-A, and execution energy consumption consumed by the user and the two servers. To ensure that the task successfully offloads the computation at its maximum completion time, the local CP frequency fn needs to be jointly optimized, and the power allocation p= [ p ] of the user and S-se:Sup>A 1 ,p 2 ,p 3 ]Time allocation t= [ t ] loc ,t up ,t exe ]Division d= [ d ] of user tasks u ,d A ,d B ]. The optimization model for minimizing the energy consumption can be obtained by the method:
P1:
s.t.0≤t loc ≤T (16b)
0<f n ≤f u,max ,n∈{1,…,β u d u } (16c)
p 1 +p 2 ≤P u,max ,p i ≥0,i∈{1,2} (16d)
0≤p 3 ≤P A,max (16e)
t up +t exe ≤T,t up ≥0,t exe ≥0 (16f)
d A ≤R 1 t up (16g)
d B ≤R 2 t up (16h)
d u +d A +d B ＝D (16i)
d u ≥0,d A ≥0,d B ≥0 (16j)
the objective function (16 a) is the total energy consumption of the system, i.e. the sum of the user's local calculation energy consumption, the user's offloading energy consumption, the server a forwarding energy consumption and the execution energy consumption of both servers. Constraints (16 b) and (16 c) limit the local completion delay and its CPU frequency; (16d) Ensuring that the total power allocated to the two links by the user does not exceed the maximum power of the user; (16 e) limiting the forwarding power of server S-se:Sup>A; (16f) Is a constraint limit on the offload latency and the remote execution latency; (16g) And (16 h) ensuring that the offloading tasks are successfully completed within a time period T; (16i) And (16 j) represents a partition constraint on the total bits of the task.
And fourthly, solving the optimization problem P1 to obtain the optimal value of each optimization variable. Due to d A ,d B ,t up The tight coupling in the objective function, constraint (16 g) and constraint (16 h) of P1 results in that this optimization problem is not a convex optimization problem and therefore cannot be solved with general optimization algorithms. In this regard, we propose an algorithm for joint communication resource and computing resource optimization based on taylor approximation and iteration. Firstly, optimizing the local CPU frequency to obtain a simplified equivalent optimization problem of the original problem. The iterative split into two steps solves this equivalence problem: 1) On the basis of given time allocation and task division, decomposing the power allocation sub-problem into an inner layer problem and an outer layer problem; 2) At a given power allocation, the time allocation and task partitioning are jointly optimized for solution. The specific implementation steps of the algorithm are as follows:
1) Local frequency optimization
At a given local computing task d u The energy consumed by the local calculation is only related to the CPU frequencyThe sub-problem of local optimization is thus obtained:
P1.a:
s.t.(16b),(16c) (17b)
through some mathematical derivations, we can derive that the optimal local frequency should meet the following conditions:
f 1 ＝f 2 ＝…＝f n ＝β u d u /T (19)
i.e. the frequency used on each CPU cycle should be equal and the local execution time should be completed just within the maximum completion time to minimize the local energy consumption. Substituting (19) into (1) and (2) to obtain optimal local calculation energy consumption and calculation time delay:
t loc ＝T (21)
thus, P1 can be equivalently the following optimization problem:
P2:
s.t.(16d),(16e),(16f),(16g),(16h),(16i),(16j) (22b)
obviously, P2 is still not a convex optimization problem. Therefore, we next split the P2 iteration into two solutions.
(2) Power allocation optimization for given time allocation and task partitioning
At a given time allocation t and task division d, the power optimization sub-problem can be expressed in the form of:
P2.a:
s.t.(16d),(16e),(16g),(16h) (23b)
at this time, the problemThe objective function of p2.A is a convex function, whereas constraints (16 g) and (16 h) are due to R therein 1 ,R 2 The non-convex structure is still a non-convex constraint, so p2.a is also a non-convex problem. This problem is not solved by first deforming the constraints (16 g) and (16 h) to the following structure:
wherein,
γ s ＝|h s | 2 p 3 +1
γ 3 ＝|h 3 | 2 p 3 +1
next we decompose P2.a into an inner and outer problem resolution, i.e. at a given p 3 Downlink to downlink NOMA power p 1 ，p 2 Optimizing the distribution of (C) to obtain the p-containing product 3 Then the optimization problem can be considered as an optimal solution for p 3 Single variable optimization function problem. Based on this, it can be seen that at a given p 3 Thereafter, constraint (24) has become a convex constraint, but constraint (25) cannot be a convex constraint because its last term is non-convex. Thus, we will make a first order taylor expansion approximation to its last term and iterate the taylor expansion points until the original function is approached. After taylor approximation, the constraint (25) is transformed into the following form:
wherein,is p at the mth iteration 1 Value of->Which is a constant at each iteration that does not affect the concavity and convexity of the overall equation. And thus becomes a convex constraint after approximation (26).
Based on the above derivation, the inner layer problem with NOMA power allocation can be described as follows, with the constant term of the objective function of the problem p2.a omitted:
P2.b:
s.t.(16d),(16e),(24)and(26) (27b)
p2.b is a convex optimization problem that can be solved using lagrangian dual methods. After a series of derivations, the following equation can be obtained:
wherein,is p 1 ,p 2 The parameter expressions included in the solution are specifically:
A＝|h 1 | 2 |h 2 | 2 (γ s -θ)
obviouslyShould be the root of equation (28). To ensure->Equation (28) should have only one non-negative real root, namely:
to ensure the establishment of (30), p 3 The optimal solution of (2) should satisfy:
AB<0 (31)
Δ＝B 2 -4AC＝0 (32)
thus, P2.a is p-dependent 3 And is constrained to (16 e), (31), (32). To obtain the minimum value of p2.A, we first find p by three constraints 3 Is a feasible region of (2).
For constraint (31), its inequality can be left-hand as to gamma through a series of simplifications 3 Is a cubic function of (2), which is formed by gamma 3 Further find p 3 Is a feasible region of (2). This cubic equation is as follows:
in order to establish (31), f (r) is required 3 )<0 is gamma 3 Is established in the feasible domain of (1). Thus, it is necessary to pass f (r 3 ) To study f (r) 3 ) Further determining the monotonicity of f (r 3 )<Gamma of 0 being strictly true 3 Is a feasible region of (2). The derivatives are as follows:
f′(r 3 )＝3θ|h 2 | 2 γ 3 2 -2θ|h 2 | 2 γ 3 +ψ (34)
wherein the method comprises the steps of
Let f' (r 3 ) =0, but the number of roots cannot be determined due to the size of uncertainty ψ. We therefore discuss from the following two cases: psi is less than or equal to 0; psi phi type>0。
Case 1: when ψ is less than or equal to 0, the equation has a negative root and a positive root, which are respectively marked as x 0 ,x 1 . Due to gamma 3 ∈[1,γ 3,max ]，γ 3,max ＝P A,max |h 3 | 2 +1, so the negative root is truncated.
When x 1 F' (r) when less than or equal to 1 3 ) Not less than 0, so f (r 3 ) At gamma 3 ∈[1,γ 3,max ]Is a monotonically increasing function. If f (1) is not less than 0, no viable gamma is present 3 Establishing (31); if f (1)<0,f(γ 3,max ) Not less than 0, it is known from the zero point theorem that at least one point mE (1, gamma) 3,max ) So that f (r 3 ) =0, thus γ 3 The optimum value of (2) should satisfyIf f (1)<0,f(γ 3,max )<0,f(r 3 )<0 is [1, gamma ] 3,max ]Always do so.
When 1.ltoreq.x 1 <γ 3,max ,f(r 3 ) At [1, x 1 ]Monotonically decreasing in [ x ] 1 ,γ 3,max ]And monotonically increases. Due to f (0)<0, thus f (1)<0. If f (gamma) 3,max )<0, then f (r 3 )<0 is [1, gamma ] 3,max ]Always hold; if f (gamma) 3,max ) Not less than 0, then there is a point m 2 ∈(1,x 1 ) So that f (m 2 ) =0, at which time γ 3 The optimum value of (2) should satisfy
When x 1 ≥γ 3,max ，f(r 3 ) At [1, gamma ] 3,max ]Monotonically increasing, thus f (r 3 )<0 is [1, gamma ] 3,max ]Internal constant establishment
Case 2: when psi is>At 0, the number of equation roots depends on the magnitude of the function value corresponding to the symmetry axis, that isIf->The equation has at most one root, where f (r 3 ) At [1, gamma ] 3,max ]Is a monotonically increasing function; if it isThe equation has two positive roots, the root which is larger than the symmetry axis and is marked as x 2 . The following discussion is similar to case 1, so the discussion is not repeated here.
In conclusion, from gamma 3 And p 3 We can get p from the relation of (2) 3 The feasible fields of (a) are:
wherein z is i I= {0, …,13} represents an event that is different in both cases. When psi is less than or equal to 0, z 1 Represents x 1 <1,f(1)<0,f(γ 3,max )；z 2 Represents x 1 <1,f(1)<0,f(γ 3,max )<0 or x 1 >γ 3,max Or 1.ltoreq.x 1 ≤γ 3,max ,f(γ 3,max )<0；z 3 Represents 1.ltoreq.x 1 ≤γ 3,max ,f(γ 3,max ) Not less than 0; when (when)ψ>0,z 4 Representation off(1)≤0,f(γ 3,max )≥0；z 5 Representation of
f(1)≤0,f(γ 3,max )<0；z 6 Representation->x 2 <1,f(1)≤0,f(γ 3,max )≥0；z 7 Representation->x 2 <1,f(1)≤0,f(γ 3,max )<0；z 8 Representation->1≤x 2 ≤γ 3,max ,f(1)>0,f(γ 3,max )≥0；z 9 Representation->1≤x 2 ≤γ 3,max ,f(1)>0,f(γ 3,max )<；z 10 Representation->1≤x 2 ≤γ 3,max ,f(1)≤0,f(γ 3,max )≥0；z 11 Representation->1≤x 2 ≤γ 3,max ,f(1)≤0,f(γ 3,max )<0；z 12 Representation->γ 3,max ,f(1)>0,f(γ 3,max )<0；z 13 Representation->f (1) is less than or equal to 0. For convenience we will represent the occurrence of an event as z i ＝1。
For constraint (32), a term p can be obtained by sorting the pair-wise equations 3 Is defined by the four equations:
B 1 p 3 4 +ap 3 3 +bp 3 2 +cp 3 +d＝0 (36)
wherein:
a＝2B 1 B 2 -4A 1 C 1
b＝2B 1 B 3 +B 2 -4(A 1 C 2 +A 2 C 1 )
c＝2B 2 B 3 -4(A 1 C 3 +A 2 C 2 )
d＝B 3 2 -4A 2 C 3
A 1 ＝|h 1 | 2 |h 2 | 2 |h 3 | 2
A 2 ＝A 1 -θ|h 1 | 2 |h 2 | 2
B 1 ＝θ|h 1 | 2 |h 2 | 4 |h 3 | 4
the solution of this fourth-order equation can be used to obtain the information about p 3 Is denoted as p 3,1 ,p 3,2 ,p 3,3 ,p 3,4 Then p is 3 Optimum value of (2)The establishment of (32) must be guaranteed for one of the roots.
By analysis of the two constraints, we can know that taking the objective function of p2.A to a minimum,should be such as to satisfy p 3 The smallest root of the four equation roots of the feasible region is:
thus, the optimum values can be obtained by (30) and (29), respectively
(3) Time allocation and task partitioning optimization for a given power allocation
At a given power allocation, the sub-problems with time allocation and task partitioning are as follows:
/>
(16f),(16g),(16h),(16i),(16j) (38b)
wherein p=p 1 +p 2 +p 3 。
By the anti-evidence method we can get the optimal strategy for time allocation:
t up +t exe ＝T,t up ≥0,t exe ≥0 (39)
(39) Indicating that the total time taken by the user to unload time and remote execution time may be just the maximum completion delay of the task is reasonable and efficient for efficient use of system energy. Thus, the user uninstalls the optimal value of timeThe method comprises the following steps:
similarly, for the remote execution time of two servers we can conclude that:
this means that the overall system power consumption is only minimized when the execution times on the two servers are equal. Thus, the optimal value t of the remote execution time can be obtained exe * ：
Thus, the problem P2.c becomes an optimization problem with respect to task division. It is assumed that the time per bit executed at server S-se:Sup>A is greater than the corresponding time generated at S-B, namely:combining constraints (16 i), let M (d) A ,d B ) The objective function representing p2.C is specifically:
to obtain a function M (d A ,d B ) We first determine d A And d A Is a feasible region of (2). From (40), (41) we can get the constraints (16 g) and (16 h) to deform into the following form:
due toTherefore, 0.ltoreq.d must be present A ≤d B So (44) can be rewritten as:
thus, d A And d A The feasible fields meeting the minimization are:
0≤d A ≤d B (47)
0≤d B ≤N (48)
wherein,
to determine M (d) A ,d B ) First, the minimum value of d and the corresponding minimum value point are calculated B The partial derivatives of (a), namely:
order theThe method can obtain the following steps:
combining (16 i) to obtain the optimal value d of the local calculated quantity u * ：
Due to uncertainty d B The' size relationship with its feasible region boundary points, needs to be discussed in case:
if 0<d B ′<N，M(d A ,d B ) At [0, d B ′]Monotonically decreasing in [ d ] B ′,N]And monotonically increases. So M (d) A ,d B ) Should be at least d B Obtained at' point, i.e. d B Is d B * ＝d B '. Substituting (50) into the function M (d A ,d B ) After that, it becomes about d A Univariate functions, namely:
M(d A ,d B )＝Qd A +Q 1 (52)
wherein,
when Q is>At 0, (52) is a reference to d A Increasing function of (d), thus when d A The function can obtain the minimum value when obtaining the minimum value in the feasible region, i.e. d A * ＝0；
When Q.ltoreq.0, (52) is a reference to d A And thus when d A The function can obtain the minimum value when obtaining the maximum value in the feasible domainI.e.
If d B '. Gtoreq.N, then d A >D-d u * N, which means d A >d B But violates constraint (47), thus this situation is compromised.
In summary, the optimal values for task partitioning are:
the best results are obtained from (40) and (42)And->
And fifthly, summarizing the algorithm proposed by the user based on the solution of the optimization problem. The proposed iterative optimization algorithm for joint communication resources and computing resources based on approximation and alternating iteration is shown in table 1.
TABLE 1 iterative algorithm for joint computation and communication optimization
In summary, the present application has the following advantages:
(1) In consideration of the problems of uneven server resource distribution and reliability when users unload to a far server in the current MEC network, a scheme for cooperatively unloading a near server with insufficient computing resources and a far server with sufficient computing resources is provided, and the near server decodes and forwards far server signals by using the SIC principle so as to strengthen the reliability of the far server signals.
(2) The near server is designed to work in a full duplex relay mode, so that the cooperation unloading process works in the same time period, and the calculation time delay of a user is further reduced.
(3) In order to minimize the total energy consumption of the system, the local and unloading energy consumption of the user and the forwarding and calculating energy consumption of the server are considered, and the energy consumption of the system is optimized for minimizing the energy consumption under the frequency, time constraint, task constraint and power constraint of a local CPU.
(3) An optimization iterative algorithm framework of joint calculation and communication resources based on approximation and alternate iteration is provided for the proposed optimization problem.
(4) In the optimization process, a method for decomposing and optimizing the power distribution is provided in the given time distribution and task division, namely the problem of power distribution is decomposed into an inner-layer NOMA power distribution problem and an outer-layer near-server forwarding power optimization problem, and a corresponding closed optimal solution is obtained.
(5) In the optimization process, a joint optimization method of time distribution and task division is provided under given power distribution, and a corresponding closed optimal solution is obtained.
(6) The MEC network gradually becomes a technology for improving the experience of the edge user 5G, and the MEC system based on the server cooperation and utilizing the NOMA access mode, which is provided by the application, namely, the servers considering the capability of different computing resources in the far and near directions cooperate, so that a certain heuristic can be provided for improving the unloading experience of the user and balancing the computing resources of the MEC network.
(7) In the cooperation model, a full duplex decoding forwarding relay mode is adopted at the near server side, so that the signal quality of a user when unloading to the far server can be further and effectively improved, and the calculation time delay can be reduced.
(8) The provided optimization algorithm framework for minimizing the energy consumption of the system has certain reference significance for solving the complex non-convex optimization problem.
In the present description, each part is described in a progressive manner, and each part is mainly described as different from other parts, and identical and similar parts between the parts are mutually referred.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (1)
1. A NOMA-based server collaboration edge computing method, the method comprising:
the user obtains the computing resources on each server;
according to the acquired computing resources on each server, a user adopts NOMA access technology and a preset unloading strategy to unload part of tasks to each server, wherein the preset unloading strategy is obtained by calculation according to preset conditions, and the preset conditions comprise the computing resources on the servers;
when a user unloads tasks to a far-end server, the tasks unloaded to the far-end server are transmitted to a near-end server at the same time, the near-end server decodes task signals of the far-end server by using the SIC serial interference elimination principle and then sends the task signals to the far-end server, the far-end server receives the tasks sent by the user and the tasks forwarded by the near-end server, the two paths of signals are identical, and the far-end server combines the two paths of identical signals by using a maximum ratio combining mode;
the preset unloading strategy comprises the following steps: a user local calculation, a full duplex cooperation unloading stage, a remote server calculation stage and a calculation result downloading stage;
the local calculation of the user specifically comprises the following steps:
set beta u The number of CPU cycles required to calculate an input bit for the user, then d is performed u The total number of cycles required for a bit is beta u d u And assuming that the user can adjust the operating frequency f per cycle n ∈(0，f u，max ]，n∈{1，...，β u d u To save local energy consumption, where f u，max For the maximum CPU frequency of the user, a constraint expression of local calculation time, local energy consumption and frequency can be obtained according to the maximum completion time delay requirement of the task:
wherein, kappa u The energy coefficient is the energy coefficient of the local server;
the full duplex cooperative unloading stage comprises the following steps:
since the full duplex collaborative communication computing mode is adopted, the user offloading process is at t up The time is divided into a direct transmission stage and a cooperative communication stage;
first, the user uses NOMA technique to simultaneously make d in the direct transmission stage A ，d B Corresponding transmission signal s 1 ，s 2 Unloading to two servers according to superposition coding principle, after receiving mixed signal, near end MEC server S-A executes SIC to carry out S 2 After the solution, the signals are sent to a remote MEC server S-B in a cooperative communication stage, and the received signals at the two servers at the kth moment are as follows:
wherein p is 1 ，p 2 Transmitting power allocated to the near-end MEC server S-A and the far-end MEC server S-B for the user and satisfying p 1 ≤p 2 .p 3 Transmit power z for near-end MEC server S-se:Sup>A forwarding i I.e {1,2} is additive high at both serversThe white noise is subjected to Gaussian distribution with mean value of zero and variance of 1; τ is decoding processing delay of the near server, and τ is assumed to be less than T;
after receiving the mixed signal, the near-end MEC server S-A decodes S according to SIC principle 2 Then s is taken 2 Removal of decoding s from mixed signal 1 Thus, the SINR of S-B at S-A is:
the SINR of the S-A decoded own signal is:
the signal achievable rate for S-se:Sup>A is:
R 1 ＝Blog 2 (1+γ 1，1 ) (7)
where B is the system bandwidth, the far-end MEC server S-B will receive the mixed signal from the direct transmission of the user and the signal forwarded by the near-end MEC server S-A, and if the two signals from the user and the near-end MEC server S-A are completely resolvable at the far-end MEC server S-B, the approximate combination is performed by the maximum ratio combination, and the S-B will act as se:Sup>A weak user when detecting NOMA signals transmitted by the user 1 The received SINR is considered as noise decoding its own signal, so this received SINR is:
the received signal-to-noise ratio for detecting the information forwarded from the near-end MEC server S-se:Sup>A is:
γ 2，A ＝|h 2 | 2 p 3 (9)
after maximum ratio combination, the total received signal to noise ratio at S-B is obtained as follows:
γ 2，MRC ＝γ 2，2 +γ 2，A (10)
thus, the achievable rate of S-B is:
R 2 ＝Bmin{log 2 (1+γ 1，2 )，log 2 (1+γ 2，MRC )} (11)
to sum up, it can be derived that this stage is at t up Energy consumed by the inner user and the near server:
E up ＝(p 1 +p 2 )t up (12)
E tr ＝p 3 t up (13)
the remote server computing stage and the computing result downloading stage comprise the following steps:
the execution energy consumption and the calculation time delay of the two servers are also considered, but the transmission bit of the result download is smaller, so the transmission time delay and the energy consumption of the process are ignored, and the frequency of the MEC servers S-A and S-B when the user executes the task is assumed to be constant and respectively marked as f A ，f B The number of cycles required by two person servers to calculate the respective user data bits is denoted as beta A d A ，β B d B The energy required to be consumed at the server is:
wherein,execution energy, κ, of S-A, S-B, respectively A ，κ B Is the energy consumption coefficient of two servers;
the computation delays required at the two servers are:
in,the execution time at S-A and S-B, respectively;
since the bit size of the calculation result is much smaller than the task bit of the uploading and the downloading rate is higher than the uploading rate, the time delay and the energy consumption of the stage are negligible.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010633925.5A CN111953730B (en) | 2020-07-02 | 2020-07-02 | NOMA-based server cooperation edge computing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010633925.5A CN111953730B (en) | 2020-07-02 | 2020-07-02 | NOMA-based server cooperation edge computing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111953730A CN111953730A (en) | 2020-11-17 |
| CN111953730B true CN111953730B (en) | 2024-01-09 |
Family
ID=73337452
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010633925.5A Active CN111953730B (en) | 2020-07-02 | 2020-07-02 | NOMA-based server cooperation edge computing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111953730B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113613260B (en) * | 2021-08-12 | 2022-08-19 | 西北工业大学 | Method and system for optimizing distance-distance cooperative perception delay moving edge calculation |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108920280B (en) * | 2018-07-13 | 2021-06-11 | 哈尔滨工业大学 | Mobile edge computing task unloading method under single-user scene |
| KR102598084B1 (en) * | 2018-11-06 | 2023-11-03 | 삼성전자주식회사 | Method and apparatus for scheduling computing tasks to servers based on task dependency |
| CN110944335B (en) * | 2019-12-12 | 2022-04-12 | 北京邮电大学 | Resource allocation method and device for virtual reality service |
-
2020
- 2020-07-02 CN CN202010633925.5A patent/CN111953730B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111953730A (en) | 2020-11-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hu et al. | Wireless powered cooperation-assisted mobile edge computing | |
| CN111988806B (en) | Mobile edge computing network resource allocation method based on full-duplex relay | |
| CN111314935B (en) | Method for minimizing downlink transmission delay based on NOMA-MEC system | |
| Wu et al. | Joint user pairing and resource allocation in a SWIPT-enabled cooperative NOMA system | |
| Muñoz et al. | Joint allocation of radio and computational resources in wireless application offloading | |
| CN111405596B (en) | Resource optimization method for large-scale antenna wireless energy-carrying communication system under Rice channel | |
| CN110545584A (en) | Communication processing method of full-duplex mobile edge computing communication system | |
| Bashar et al. | On the energy efficiency of limited-backhaul cell-free massive MIMO | |
| Xu et al. | Energy-efficient design for IRS-assisted NOMA-based mobile edge computing | |
| Wei et al. | Spectral-energy efficiency trade-off based design for hybrid TDMA-NOMA system | |
| Azimi et al. | Ultra-reliable cloud mobile computing with service composition and superposition coding | |
| Nguyen et al. | Computation offloading and resource allocation for backhaul limited cooperative MEC systems | |
| CN111953730B (en) | NOMA-based server cooperation edge computing method | |
| Nguyen et al. | Joint computation offloading, SFC placement, and resource allocation for multi-site MEC systems | |
| US20230337131A1 (en) | Method and apparatus for asynchronously allocating resources in mobile edge computing system | |
| Nezhadmohammad et al. | SWIPT-enabled multiple access channel: Effects of decoding cost and non-linear EH model | |
| Eliodorou et al. | Energy efficiency for MEC offloading with NOMA through coalitional games | |
| CN112788764A (en) | Method and system for task unloading and resource allocation of NOMA ultra-dense network | |
| Altin et al. | Novel OMA and hybrid NOMA schemes for MEC offloading | |
| CN116405979A (en) | Millimeter wave mobile edge computing networking resource allocation method | |
| Duan et al. | Energy consumption minimization for near-far server cooperation in NOMA-assisted mobile edge computing system | |
| Yan et al. | Joint relay selection and power allocation in underwater cognitive acoustic cooperative system with limited feedback | |
| CN115811793A (en) | Energy collection moving edge calculation method based on assistance of intelligent reflecting surface | |
| Yang et al. | Resource allocation in full-duplex mobile-edge computing systems with NOMA and energy harvesting | |
| CN108809379B (en) | User terminal and MIMO data energy simultaneous transmission system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |