CN110011922B - RSA method adopting service prediction and spectrum conversion in elastic optical network - Google Patents

RSA method adopting service prediction and spectrum conversion in elastic optical network Download PDF

Info

Publication number
CN110011922B
CN110011922B CN201910273876.6A CN201910273876A CN110011922B CN 110011922 B CN110011922 B CN 110011922B CN 201910273876 A CN201910273876 A CN 201910273876A CN 110011922 B CN110011922 B CN 110011922B
Authority
CN
China
Prior art keywords
service request
rbfnn
spectrum
duration
executing
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
Application number
CN201910273876.6A
Other languages
Chinese (zh)
Other versions
CN110011922A (en
Inventor
张水艳
徐展琦
贾文彬
吴杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN201910273876.6A priority Critical patent/CN110011922B/en
Publication of CN110011922A publication Critical patent/CN110011922A/en
Application granted granted Critical
Publication of CN110011922B publication Critical patent/CN110011922B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/147—Network analysis or design for predicting network behaviour
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L45/00—Routing or path finding of packets in data switching networks
    • H04L45/02—Topology update or discovery
    • H04L45/08—Learning-based routing, e.g. using neural networks or artificial intelligence
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L45/00—Routing or path finding of packets in data switching networks
    • H04L45/12—Shortest path evaluation
    • H04L45/124—Shortest path evaluation using a combination of metrics
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04Q—SELECTING
    • H04Q11/00—Selecting arrangements for multiplex systems
    • H04Q11/0001—Selecting arrangements for multiplex systems using optical switching
    • H04Q11/0062—Network aspects
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04Q—SELECTING
    • H04Q11/00—Selecting arrangements for multiplex systems
    • H04Q11/0001—Selecting arrangements for multiplex systems using optical switching
    • H04Q11/0062—Network aspects
    • H04Q2011/0086—Network resource allocation, dimensioning or optimisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The invention provides an RSA method combining service prediction and spectrum conversion in an elastic optical network, which is used for solving the problems of dynamic routing and spectrum allocation in the elastic optical network and comprises the following implementation steps: firstly, selecting K candidate paths for all node pairs of a network; then, predicting the service request change information on the link, and calculating the comprehensive weight of each candidate path; secondly, judging whether the service request can be processed according to the frequency spectrum constraint condition; and finally, when all the candidate paths do not meet the spectrum constraint condition, adding a spectrum converter to process the service request is considered. The invention uses the RBFNN prediction technology of the radial basis function neural network, not only considers the path length and the spectrum occupation condition during the path selection, but also considers the competition degree of the service request duration on the candidate path, and combines the spectrum conversion, thereby effectively improving the network performance and providing an efficient resource optimization allocation scheme for the requirements of interconnection of operators or data centers and the like.

Description

RSA method adopting service prediction and spectrum conversion in elastic optical network
Technical Field
The invention belongs to the technical field of communication, and further relates to an RSA (routing and spectrum assignment) method adopting service prediction and spectrum conversion in elastic Optical network EONs (elastic Optical networks) in the technical field of network communication. The invention can complete routing selection and spectrum allocation aiming at each dynamic service request which arrives randomly in the elastic optical network.
Background
Elastic optical network EONs (optical-Orthogonal-frequency division Multiplexing) based on Coded optical Orthogonal frequency division Multiplexing (CO-OFDM) can flexibly allocate appropriate spectrum resources according to user request bandwidth, and become one of the development trends of future optical networks. In the elastic optical network, the routing and spectrum allocation RSA problem is divided into a routing sub-problem and a spectrum allocation sub-problem. The network operator or the network resource distributor establishes an end-to-end optical path according to the service request of the user, and distributes corresponding spectrum resources on the path. The optimization goal of the dynamic routing and spectrum allocation RSA problem is usually to reduce the traffic blocking rate or to increase the network resource utilization.
In the patent document "resource-aware routing and spectrum resource allocation method and system in elastic optical network" (publication No. CN 103051547B, application No. 201210568557.6), the beijing post and telecommunications university discloses a resource-aware routing and spectrum resource allocation RSA method. Aiming at an arriving service request, the method firstly judges whether an idle optical path established in the current network can bear the request, if so, the idle optical path is directly utilized for communication, otherwise, a candidate path is recalculated, a working path is selected according to the link hop number of the path and the available frequency slot number, then frequency spectrum distribution is carried out on the working path, and a new optical path is established to complete the communication between source and destination nodes; after the communication is finished, the new optical path is not dismantled for the moment, the timer is started, and the arriving service request is continuously processed within the time of the timer. The method has the disadvantages that two factors considered in the routing process are based on the current network state information, the influence of a specific source and destination node on future service requests on the current service routing is ignored, and the network resources are not fully utilized from the global view.
Wenbin Jia, Zhanqi Xu, Zhe Ding and Kai Wang disclose an RSA method using predictive Routing and spectrum allocation in the published paper "Routing and spectrum Assignment Algorithm with Prediction for Elastic Optical Networks under the book" 201615 th International Conference Optical Communications and Networks (ICOCN), Handzhou, 2016, pp.1-3 ". The method comprises the steps of firstly calculating a candidate path set of services among different source and sink nodes in an off-line mode by adopting a front K shortest path algorithm, and then predicting the change time of future link flow through a Back Propagation Neural Network (BPNN). And in the routing stage, the duration overlap ratio, the path hop count and the spectrum utilization rate on the candidate paths are comprehensively considered for each candidate path, and the spectrum allocation is preferentially carried out on the candidate path with the minimum comprehensive weight. And directly blocking the service request when the spectrum blocks which meet the service request are not arranged on all the candidate paths. The method has the disadvantages that in the spectrum allocation stage, when the network resource can not provide the spectrum block meeting the service request, the service is directly blocked, and the adoption of the spectrum conversion technology to improve the flexibility of spectrum allocation is not considered, so that the service blocking rate is higher, and the network resource is wasted.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a routing and spectrum allocation RSA method adopting service prediction and spectrum conversion in an elastic optical network.
The specific thought of the invention is as follows: firstly, calculating 3 candidate paths for all source-destination node pairs of the network, then sequentially selecting service requests from all the service requests, and secondly, predicting the flow increasing time and duration caused by future service requests on each link in the candidate paths of the selected service requests by using a radial Basis Function network (RBFNN), namely predicting the flow change of the future link within the duration of the selected service requests. And then, comprehensively considering three factors of the duration contact ratio of the path, the path frequency spectrum occupation and the path hop count of each candidate path, calculating the comprehensive weight values of the candidate paths, and sequencing the candidate paths from small to large. And finally, sequentially judging whether a candidate path can process the selected service request according to the frequency spectrum constraint condition, and considering the frequency spectrum converter to establish a light path as much as possible and bear the selected service request when all the candidate paths do not meet the frequency spectrum constraint condition.
In order to achieve the purpose, the method comprises the following specific implementation steps:
(1) preprocessing network and service information:
(1a) inputting network topology, link spectrum state information and a service request set of an elastic optical network;
(1b) randomly selecting at least one node in the elastic optical network, and placing a frequency spectrum converter;
(1c) acquiring 3 candidate paths of each service request in a service set by using a K shortest path method;
(2) constructing a radial basis function neural network RBFNN:
building a three-layer radial basis function neural network RBFNN which sequentially comprises an input layer, a hidden layer and an output layer; setting the number of the neurons of the input layer to be 9, the number of the neurons of the hidden layer to be 10, and the number of the neurons of the output layer to be 1;
(3) selecting an unprocessed service request from the service request set in sequence;
(4) predicting change information of service request on link:
(4a) judging whether the number of the service requests borne on the link of each candidate route in the selected service request is less than 50, if so, marking the predicted flow increasing time caused by the future service request as 0 and then executing the step (4 e); otherwise, executing the step (4 b);
(4b) selecting 9 service requests from back to front according to the bearing time sequence for the service requests borne on the links of each candidate path in the selected service requests to form a prediction service request sample set, and then selecting 50 service requests from back to front according to the bearing time sequence to form a training service request sample set;
(4c) inputting arrival time information of all service requests in a training service request sample set into a Radial Basis Function Neural Network (RBFNN), and training the RBFNN structural parameters by adopting a supervised learning algorithm;
(4d) inputting the arrival time information of all service requests in the predicted service request sample set into a Radial Basis Function Neural Network (RBFNN), and predicting the flow increasing time caused by future service requests on a link of a candidate path of the selected service request according to the trained RBFNN structural parameters;
(4e) judging whether the flow increasing time caused by the predicted future service request is smaller than the leaving time of the selected service request, if so, executing the step (4f), otherwise, executing the step (5) after recording the duration time of the predicted future increased flow on the link as 0;
(4f) inputting duration information of all service requests in a training service request sample set into a Radial Basis Function Neural Network (RBFNN), and training the RBFNN structure parameters by adopting a supervised learning algorithm;
(4g) inputting the duration information of all service requests in the predicted service request sample set into an input layer of a Radial Basis Function Neural Network (RBFNN), predicting the duration of future increased flow on a link of a candidate path of the selected service request according to the trained RBFNN structural parameters, and executing the step (5);
(5) calculating the duration coincidence rate of each candidate path:
(5a) calculating the duration overlap ratio of each link of the candidate path of the selected service request;
(5b) calculating the duration coincidence rate of each link of the candidate path of the selected service request;
(5c) selecting the maximum link duration coincidence rate in the candidate paths as the duration coincidence rate of the candidate paths;
(6) calculating the comprehensive weight value of each candidate path:
(6a) calculating the spectrum occupancy rate of each candidate path of the selected service request;
(6b) calculating the hop count normalization value of each candidate path of the selected service request;
(6c) calculating the comprehensive weight value of each candidate path of the selected service request;
(6d) sorting all candidate paths of the selected service request according to the sequence of the comprehensive weight values from small to large;
(7) and judging whether the selected service request can be processed according to the spectrum constraint condition:
(7a) selecting a candidate path from all candidate paths of the selected service request in sequence according to the sequence of the comprehensive weight values;
(7b) judging whether a free spectrum block meeting the spectrum constraint condition exists on the selected candidate path, if so, executing the step (7c), otherwise, executing the step (7 d);
(7c) after allocating the first idle spectrum block meeting the spectrum constraint condition to the selected service request, executing step 9;
(7d) judging whether the comprehensive weight value of the selected candidate path is maximum or not, if so, executing the step (8), otherwise, executing the step (7 a);
(8) considering the spectrum converter, the selected service request is processed:
(8a) judging whether a candidate path passes through a node for placing a spectrum converter in all candidate paths, if so, executing the step (8b), otherwise, executing the step (9) after blocking the selected service request;
(8b) judging whether idle spectrum blocks meeting the frequency spectrum adjacency condition exist on links before and after the frequency spectrum conversion node in a candidate path passing through the node where the frequency spectrum converter is placed, if so, executing the step (8c), otherwise, executing the step (9) after the selected service request is blocked;
(8c) allocating the first free spectrum block meeting the spectrum adjacency condition to the selected service request, and executing the step (9);
(9) updating the state of the network resource;
(10) judging whether the service request set is empty, if so, executing the step (11), otherwise, executing the step (3);
(11) and processing all service requests.
Compared with the prior art, the invention has the following advantages:
firstly, the invention uses the radial basis function neural network RBFNN to predict the change information of the traffic on the link, firstly judges whether the selected service request can be processed according to the frequency spectrum constraint condition, when all the candidate paths do not meet the frequency spectrum constraint condition, the frequency spectrum converter is considered to process the service request, and the invention overcomes the problem that the network resource can not be fully utilized because two factors considered in the prior art during the route selection are based on the current network state information and the influence of the link traffic change caused by the future service request on the current service route selection is ignored.
Secondly, because the invention uses the radial basis function neural network RBFNN to predict the change information of the service volume on the link, firstly, according to the frequency spectrum constraint condition, the invention judges whether the selected service request can be processed, when all candidate paths do not meet the frequency spectrum constraint condition, the frequency spectrum converter is considered to process the service request, thereby overcoming the defect that the service is directly blocked when the candidate paths can not provide continuous idle frequency slot blocks meeting the service request in the current state in the frequency spectrum allocation stage in the prior art.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a simulation of the present invention.
The specific implementation mode is as follows:
the present invention is described in further detail below with reference to the attached drawing figures.
The specific implementation steps of the present invention are further described with reference to fig. 1.
Step 1, preprocessing network and service information.
The method comprises the steps of inputting a set of network topology, link spectrum state information and service requests of the elastic optical network.
And randomly selecting at least one node in the elastic optical network and placing a spectrum converter.
And 3 candidate paths of each service request in the service set are obtained by using a K shortest path method.
And 2, constructing a radial basis function neural network RBFNN.
Building a three-layer radial basis function neural network RBFNN which sequentially comprises an input layer, a hidden layer and an output layer; the number of neurons of the input layer is set to 9, the number of neurons of the hidden layer is set to 10, and the number of neurons of the output layer is set to 1.
And 3, selecting an unprocessed service request from the service request set in sequence.
And 4, predicting the change information of the service request on the link.
(4.1) judging whether the number of the service requests borne on the link of each candidate path in the selected service request is less than 50, if so, recording the predicted flow increasing time caused by the future service request as 0, and then executing 4.5; otherwise, 4.2 is executed.
(4.2) selecting 9 service requests from back to front according to the bearing time sequence for the service requests borne on the link of each candidate path in the selected service requests to form a predicted service request sample set, and selecting 50 service requests from back to front according to the bearing time sequence to form a training service request sample set;
and (4.3) inputting the arrival time information of all service requests in the training service request sample set into a Radial Basis Function Neural Network (RBFNN), and training the RBFNN structure parameters by adopting a supervised learning algorithm.
The specific steps of the supervised learning algorithm are as follows:
step 1, randomly selecting arrival time information of service requests in a training service request sample set, and initializing a central vector, an expanded width vector and a weight from a hidden layer to an output layer of a radial basis function.
Step 2, inputting training sample data at the input layer side of the RBFNN, and calculating the output value of each neuron in the hidden layer of the RBFNN according to the following formula:
Figure BDA0002019321990000051
in the formula, hjRepresenting the output value of the j-th neuron in the hidden layer of the RBFNN, j being 1,2,3, …, p, p representing the total number of neurons in the hidden layer, exp (·) representing exponential operation with e as the base, | | | · | | | representing European norm operation, X representing the input vector in the RBFNN, CjRepresents the center vector, D, of the jth neuron in the hidden layer of the RBFNNjRepresents the extended width vector of the jth neuron in the hidden layer of the RBFNN, which influences the range of action of the neuron on the input vector, DjThe larger the hidden layer has, the larger the range of influence on the input vector, and the smoothing between neuronsThe better the degree.
And 3, carrying out linear change on the output of the hidden layer of the radial basis function neural network RBFNN according to the following formula to obtain the value of an output neuron:
Figure BDA0002019321990000061
in the formula, yiRepresents the output value of the ith neuron in the output layer of the RBFNN, sigma represents the summation operation, wijAnd representing the weight value of the jth neuron in the hidden layer to the ith neuron in the output layer of the RBFNN.
And 4, calculating an objective function value of the RBFNN according to the following formula:
Figure BDA0002019321990000062
wherein E represents the objective function value of the RBFNN, n represents the number of training samples, q represents the number of neurons of the output layer of the RBFNN, and y represents the number of training samplesikRepresents the net output value, o, of the k-th output neuron of the RBFNN at the i-th input sampleikRepresents the expected output value of the k output neuron of the radial basis function neural network RBFNN at the i input sample.
And 5, judging whether the objective function value is greater than the iteration termination precision, if so, executing the 6 th step of the step, and otherwise, finishing the training.
Step 6, respectively updating the central vector C of the radial basis function neural network RBFNN according to the following three formulasj(t) extended Width vector Dj(t) and weight W from hidden layer to output layerj(t) returning to the 4th step of the step:
Figure BDA0002019321990000063
Figure BDA0002019321990000064
Figure BDA0002019321990000071
in the formula, Cj(t) denotes the centre vector of the t-th update of the jth neuron in the hidden layer of the RBFNN, Cj(t-1) denotes the center vector of the t-1 th update of the jth neuron in the hidden layer of the RBFNN, η1A learning factor representing a radial basis function neural network RBFNN center vector,
Figure BDA0002019321990000072
denotes a partial derivative operation, Dj(t) represents the t-th updated expanded wideband vector of the jth neuron in the hidden layer of the RBFNN, Dj(t-1) represents the expanded width vector of the t-1 th update of the jth neuron in the hidden layer of the radial basis function neural network RBFNN, η2Learning factor, W, representing the RBFNN extended wideband vector of a radial basis function neural networkj(t) represents the output weight vector updated t times by the jth neuron in the hidden layer of the RBFNN, Wj(t-1) represents the output weight vector of t-1 updating of the jth neuron in the hidden layer of the RBFNN, η3And the learning factor represents an output weight vector in the RBFNN hidden layer of the radial basis function neural network.
And (4.4) inputting the arrival time information of all service requests in the predicted service request sample set into a Radial Basis Function Neural Network (RBFNN), and predicting the flow increasing time caused by future service requests on links of candidate paths in the selected service requests according to the trained RBFNN network structure parameters.
(4.5) judging whether the flow increasing time caused by the predicted future service request is less than the leaving time of the selected service request, if so, executing 4.6, otherwise, executing step 5 after recording the duration time of the predicted future increased flow as 0.
And (4.6) inputting the duration information of all the service requests in the training service request sample set into the RBFNN, and training the RBFNN structure parameters by adopting a supervised learning algorithm.
The specific steps of the supervised learning algorithm are as follows:
step 1, randomly selecting data in a duration time sample set, and initializing a central vector, an expanded width vector and a weight from a hidden layer to an output layer of a radial basis function.
Step 2, inputting training sample data at the input layer side of the RBFNN, and calculating the output value of each neuron in the hidden layer of the RBFNN according to the following formula:
Figure BDA0002019321990000073
in the formula, hjRepresenting the output value of the j-th neuron in the hidden layer of the RBFNN, j being 1,2,3, …, p, p representing the total number of neurons in the hidden layer, exp (·) representing exponential operation with e as the base, | | | · | | | representing European norm operation, X representing the input vector in the RBFNN, CjRepresents the center vector, D, of the jth neuron in the hidden layer of the RBFNNjRepresents the extended width vector of the jth neuron in the hidden layer of the RBFNN, which influences the range of action of the neuron on the input vector, DjThe larger the hidden layer has the greater its scope of influence on the input vector, and the better the smoothness between neurons.
And 3, carrying out linear change on the output of the hidden layer of the radial basis function neural network RBFNN according to the following formula to obtain the value of an output neuron:
Figure BDA0002019321990000081
in the formula, yiRepresents the output value of the ith neuron in the output layer of the RBFNN, sigma represents the summation operation, wijAnd representing the weight value of the jth neuron in the hidden layer to the ith neuron in the output layer of the RBFNN.
And 4, calculating an objective function value of the RBFNN according to the following formula:
Figure BDA0002019321990000082
wherein E represents the objective function value of the RBFNN, n represents the number of training samples, q represents the number of neurons in the output layer of the RBFNN, and y represents the number of the RBFNNikRepresents the net output value, o, of the k-th output neuron of the RBFNN at the i-th input sampleikRepresents the expected output value of the k output neuron of the radial basis function neural network RBFNN at the i input sample.
And 5, judging whether the objective function value is greater than the iteration termination precision, if so, executing the 6 th step of the step, and otherwise, finishing the training.
Step 6, respectively updating the central vector C of the radial basis function neural network RBFNN according to the following three formulasj(t) extended Width vector Dj(t) and weight W from hidden layer to output layerj(t) returning to the 4th step of the step:
Figure BDA0002019321990000083
Figure BDA0002019321990000084
Figure BDA0002019321990000091
in the formula, Cj(t) denotes the centre vector of the t-th update of the jth neuron in the hidden layer of the RBFNN, Cj(t-1) denotes the center vector of the t-1 th update of the jth neuron in the hidden layer of the RBFNN, η1A learning factor representing a radial basis function neural network RBFNN center vector,
Figure BDA0002019321990000092
denotes a partial derivative operation, Dj(t) represents the t-th updated expanded wideband vector of the jth neuron in the hidden layer of the RBFNN, Dj(t-1) represents the expanded width vector of the t-1 th update of the jth neuron in the hidden layer of the radial basis function neural network RBFNN, η2Learning factor, W, representing the RBFNN extended wideband vector of a radial basis function neural networkj(t) represents the output weight vector updated t times by the jth neuron in the hidden layer of the RBFNN, Wj(t-1) represents the output weight vector of t-1 updating of the jth neuron in the hidden layer of the RBFNN, η3And the learning factor represents an output weight vector in the RBFNN hidden layer of the radial basis function neural network.
And 4.7, inputting the duration information of all service requests in the predicted service request sample set into an input layer of the RBFNN, predicting the duration of future increased flow on a link of a candidate path of the selected service request according to the trained RBFNN structural parameters, and executing the step 5.
And 5, calculating the duration coincidence rate of each candidate path.
The duration overlap ratio of each link of the candidate path of the selected service request is calculated.
The duration coincidence refers to the duration of the service carried on each link of the candidate path of the selected service, the duration of the selected service and the duration of the predicted future service, and the duration coincidence degree refers to the degree of duration coincidence, which reflects the resource competition degree of other services on the link and the selected service within the duration range of the selected service, and the larger the duration coincidence degree of the link is, the more intense the competition degree on the link is.
The duration overlap ratio calculation formula is as follows:
Figure BDA0002019321990000093
in the formula (I), the compound is shown in the specification,
Figure BDA0002019321990000101
link/in a candidate path representing a selected service request imnLink duration overlap of the up-loaded service request or predicted future service request j with the selected service request i,
Figure BDA0002019321990000108
indicating the departure time of the service request j,
Figure BDA0002019321990000102
indicating the arrival time of the selected service request i,
Figure BDA0002019321990000103
indicating the arrival time of the service request j,
Figure BDA0002019321990000104
indicating the departure time of the selected service request i,
Figure BDA0002019321990000109
indicating the duration of the service request j,
Figure BDA0002019321990000105
indicating the duration of the selected service request i.
The time duration coincidence rate of the selected service request on each link of the candidate path is calculated.
The normalized calculation formula is as follows:
Figure BDA0002019321990000106
in the formula, RT(lmn) Link/in candidate path representing selected service requestmnIs represented as belonging to the symbol, omegamnLink/in candidate path representing selected service requestmnService request set of, NiChain of candidate paths representing selected service requestsRoad lmnThe total number of service requests that coincide with the duration of the selected service request.
And selecting the maximum link duration coincidence rate in the candidate paths as the duration coincidence rate of the candidate paths.
And 6, calculating the comprehensive weight value of each candidate path.
The spectrum occupancy of each candidate path of the selected service request is calculated.
The spectrum occupancy rate formula is as follows:
RU(Pk)＝fU(Pk)/Nfs.
in the formula, RU(Pk) Candidate path P representing selected service requestkSpectrum occupancy of fU(Pk) Indicating a selected service request candidate path PkTotal number of upper occupied frequency slots, NfsRepresenting the total number of frequency slots on the link in each candidate path.
And calculating the hop count normalization value of each candidate path of the selected service request.
The path hop count normalization formula is as follows:
Figure BDA0002019321990000107
in the formula, RL(Pk) Candidate path P representing selected service requestkNormalized value of the number of path hops, LpkIndicating a selected service request candidate path PkL represents the maximum of the number of hops of all candidate paths requested by the selected service.
And 6.3, calculating the comprehensive weight value of each candidate path of the selected service request by using a comprehensive weight value calculation formula.
The comprehensive weight value calculation formula is as follows:
Weight(Pk)＝αRT(Pk)+βRU(Pk)+γRL(Pk)
in the formula, Weight (P)k) Representing candidate paths Pkα denotes a weight factor for the duration coincidence, RT(Pk) Indicating a candidate path Pkβ denotes a weighting factor for the spectrum occupancy, γ denotes a weighting factor for the number of path hops, and α + β + γ is equal to 1.
And sequencing all candidate paths of the selected service request according to the sequence of the comprehensive weight values from small to large.
And 7, judging whether the selected service request can be processed or not according to the spectrum constraint condition.
The spectrum constraint condition means that the spectrum has continuity and adjacency, wherein the continuity means that the same idle spectrum resources are allocated on each link along a path for an arriving service request; the adjacency is that the spectrum resources allocated for each service request are continuous, and the spectrum resources must be equal to the size of the spectrum resources of the service request.
And (7.1) sequentially selecting a candidate path from all the candidate paths of the selected service request according to the sequence of the comprehensive weight values.
And (7.3) judging whether the selected candidate path has a free spectrum block meeting the spectrum constraint condition, if so, executing 7.3, and otherwise, executing 7.4.
(7.3) after allocating the first free spectrum block satisfying the spectrum constraint condition to the selected service request, step 9 is executed.
(7.4) judging whether the comprehensive weight value of the selected candidate path is the maximum or not, if so, executing the step 8, otherwise, executing the step 7.1.
Step 8, considering the spectrum converter, the selected service request is processed.
(8.1) judging whether a candidate path passes through the node for placing the spectrum converter in all the candidate paths, if so, executing a step 8.2, otherwise, executing a step 9 after the selected service request is blocked.
(8.2) in the candidate path passing through the node for placing the spectrum converter, judging whether idle spectrum blocks meeting the spectrum adjacency condition exist on the links before and after passing through the spectrum conversion node, if so, executing a step 8.3, otherwise, executing a step 9 after the selected service request is blocked.
(8.3) allocating the first free spectrum block satisfying the spectrum adjacency condition to the selected service request, and executing step 9.
And 9, updating the network resource state.
And step 10, judging whether the service request set is empty, if so, executing step 11, otherwise, executing step 3.
And step 11, processing all service requests.
The effect of the method is verified through simulation of the invention.
1. Simulation conditions are as follows:
the method is realized by C + + programming by using software Visual Studio 2013 on a Windows system, simulation parameters are set as follows, the optical fiber bandwidth capacity on each link in an elastic optical network is set to be 4THZ, the bandwidth of each frequency slot is 12.5GHz, namely each link comprises 320 frequency slots, the bandwidth of each service request obeys the uniform distribution of [1,6] frequency slots, the protection bandwidth is 1 frequency slot, the total number of the service requests is 100000, a Poisson process with the reaching rate of lambda and a self-similarity process are respectively obtained in the reaching process, a multi-fractal wavelet model is adopted in the self-similarity process, the duration obeys the negative exponential distribution of the parameter mu, 3 nodes are randomly selected to place a frequency spectrum converter, and during path weight calculation, α, β and gamma respectively take 0.5, 0.3 and 0.2.
2. Simulation content and result analysis thereof:
the simulation experiment of the present invention is to utilize the method of the present invention and two prior art methods, in the NSFNET network topology, aiming at the poisson service source and the self-similar service source, under different service intensities, the service blocking rate and the spectrum resource utilization rate are evaluated and calculated, wherein the NSFNET network topology has 14 nodes and 21 unidirectional links, as shown in fig. 2 (a). The prior art is the RSA method based on K shortest path and first match and the RSA method based on Prediction, which are proposed in the paper "An effective Routing and spectrum assignment Algorithm Using Prediction for Elastic Optical Networks" (2016International Conference on Information System & Intelligent Association, Hong Kong, 2016, pp.89-93.) published by Wenbin Jia et al. The RSA method based on the K shortest path and the first matching only considers the length of the path in the route selection stage and adopts the first matching strategy in the spectrum allocation stage; in the prediction-based RSA method, not only the length of a path and the influence of the current network state are considered during route selection, but also the influence of prediction factors is considered, and a first-time matching strategy is still adopted in a spectrum allocation stage.
In the simulation fig. 2(b), the simulation fig. 2(c), the simulation fig. 2(d) and the simulation fig. 2(e), the leftmost forward-diagonal rectangular box represents the result of the RSA method based on the K shortest path and the first matching, the square rectangular box represents the result of the RSA method based on prediction, and the rightmost backward-diagonal rectangular box represents the method of the present invention.
Fig. 2(b) is a simulation of a change of a blocking rate of a service in a network with a service intensity in a poisson service source model. It can be seen from the figure that the blocking rate of the method of the present invention is the lowest, followed by the predicted RSA method, and the highest blocking rate obtained by the RSA method based on the K shortest path and the first match. Specifically, the blocking rate of the present invention is 12.30% lower on average than that of the K-shortest-circuit and first-match-based RSA method, and 8.29% lower on average than that of the predicted RSA method. The RSA method based on the K shortest path and the first matching does not consider the prediction factor to reserve resources in advance, and the RSA method based on the prediction and the method not only consider the current loaded service, but also consider the arrival of the service on a future link, so the performances of the two methods are superior to the RSA method based on the K shortest path and the first matching. In addition, the invention also introduces spectrum transformation, and tries to perform spectrum transformation on the candidate path for the service with unsuccessful primary resource allocation so as to further improve the network performance, therefore, the invention is superior to the prediction-based RSA method in terms of the blocking rate performance.
Simulation fig. 2(c) shows the change of the spectrum utilization rate with the service intensity in the poisson service source model. Similarly, the method of the invention considers the time information of the carried service and the future service to calculate the network resource, and simultaneously utilizes the spectrum transformation technology to enable more service requests to be carried, thereby effectively improving the utilization rate of the network resource. Therefore, the spectrum utilization rate obtained by the method is the highest, which is 6.35% higher than the spectrum utilization rate of the RSA method based on the K shortest path and the first matching on average and 4.62% higher than the spectrum utilization rate of the RSA method based on prediction on average.
Fig. 2(d) is a simulation of the change of the blocking rate of the service in the network with the service intensity under the self-similar service source model. Due to the burstiness of self-similar service arrival, the blocking rate is higher than that of a poisson service source. In addition, the method of the invention and the method based on the predicted RSA fully excavate the historical bearing information of the link, predict the coming of the future service request, reserve resources for the sudden service request in advance and slow down the competition degree of the self-similar service to the frequency spectrum resources. Meanwhile, the invention introduces a spectrum transformation strategy to further reduce the blocking performance, but the improvement of the blocking performance is not very obvious. Under a self-similar service model, the blocking rate obtained by the RSA method based on the K shortest circuit and the first matching is the highest, the blocking rate obtained by the method is the lowest, the average of the blocking rates is 3.66% lower than that of the RSA method based on the K shortest circuit and the first matching, and the blocking rate obtained by the RSA method based on the prediction is 1.58% lower than that of the RSA method based on the K shortest circuit and the first matching.
Simulation fig. 2(e) shows the variation of spectrum utilization with the service intensity under the self-similar service source model. Simulation results show that the spectrum utilization rate obtained by the RSA method based on the K shortest path and the first matching is the lowest, and the spectrum utilization rate obtained by the method is the highest based on the frequency spectrum utilization rate obtained by the predicted RSA method. In the aspect of spectrum utilization rate performance, the average result obtained by the method is 3.74% higher than that obtained by the RSA method based on the K shortest circuit and the first matching, and the average result obtained by the method is 1.89% higher than that obtained by the RSA method based on prediction.
Comparing the simulation fig. 2(b), the simulation fig. 2(c), the simulation fig. 2(d) and the simulation fig. 2(e), it can be known that the spectrum conversion strategy has an influence on both the blocking frequency and the spectrum utilization rate of the poisson service source and the self-similar service source, but has a larger influence on the blocking probability and the spectrum utilization rate of the poisson source.

Claims (5)

1. An RSA method using service prediction and spectrum conversion in elastic optical network is characterized in that: predicting the change information of the service request on the link by using a radial basis function neural network RBFNN, firstly judging whether the selected service request can be processed according to a frequency spectrum constraint condition, and when all candidate paths do not meet the frequency spectrum constraint condition, considering a frequency spectrum converter to process the service request, wherein the method comprises the following specific steps:
(1) preprocessing network and service information:
(1a) inputting network topology, link spectrum state information and a service request set of an elastic optical network;
(1b) randomly selecting at least one node in the elastic optical network, and placing a frequency spectrum converter;
(1c) acquiring 3 candidate paths of each service request in a service set by using a K shortest path method;
(2) constructing a radial basis function neural network RBFNN:
building a three-layer radial basis function neural network RBFNN which sequentially comprises an input layer, a hidden layer and an output layer; setting the number of the neurons of the input layer to be 9, the number of the neurons of the hidden layer to be 10, and the number of the neurons of the output layer to be 1;
(3) selecting an unprocessed service request from the service request set in sequence;
(4) predicting change information of service request on link:
(4a) judging whether the number of the service requests borne on the link of each candidate route in the selected service request is less than 50, if so, marking the predicted flow increasing time caused by the future service request as 0 and then executing the step (4 e); otherwise, executing the step (4 b);
(4b) selecting 9 service requests from back to front according to the bearing time sequence for the service requests borne on the links of each candidate path in the selected service requests to form a prediction service request sample set, and then selecting 50 service requests from back to front according to the bearing time sequence to form a training service request sample set;
(4c) inputting arrival time information of all service requests in a training service request sample set into a Radial Basis Function Neural Network (RBFNN), and training the RBFNN structural parameters by adopting a supervised learning algorithm;
(4d) inputting the arrival time information of all service requests in the predicted service request sample set into a Radial Basis Function Neural Network (RBFNN), and predicting the flow increasing time caused by future service requests on a link of a candidate path of the selected service request according to the trained RBFNN structural parameters;
(4e) judging whether the flow increasing time caused by the predicted future service request is smaller than the leaving time of the selected service request, if so, executing the step (4f), otherwise, executing the step (5) after recording the duration time of the predicted future increased flow on the link as 0;
(4f) inputting duration information of all service requests in a training service request sample set into a Radial Basis Function Neural Network (RBFNN), and training the RBFNN structure parameters by adopting a supervised learning algorithm;
(4g) inputting the duration information of all service requests in the predicted service request sample set into a Radial Basis Function Neural Network (RBFNN), predicting the duration of future increased flow on a link of a candidate path of the selected service request according to the trained RBFNN structural parameters, and executing the step (5);
(5) calculating the duration coincidence rate of each candidate path:
(5a) calculating the duration overlap ratio on each link of the candidate path of the selected service request;
(5b) calculating the duration coincidence rate on each link of the candidate path of the selected service request;
(5c) selecting the maximum link duration coincidence rate in the candidate paths as the duration coincidence rate of the candidate paths;
(6) calculating the comprehensive weight value of each candidate path:
(6a) calculating the spectrum occupancy rate of each candidate path of the selected service request;
(6b) calculating the hop count normalization value of each candidate path of the selected service request;
(6c) calculating the comprehensive weight value of each candidate path of the selected service request;
(6d) sorting all candidate paths of the selected service request according to the sequence of the comprehensive weight values from small to large;
(7) and judging whether the selected service request can be processed according to the spectrum constraint condition:
(7a) selecting a candidate path from all candidate paths of the selected service request in sequence according to the sequence of the comprehensive weight values;
(7b) judging whether a free spectrum block meeting the spectrum constraint condition exists on the selected candidate path, if so, executing the step (7c), otherwise, executing the step (7 d);
(7c) after allocating the first idle spectrum block meeting the spectrum constraint condition to the selected service request, executing step 9;
(7d) judging whether the comprehensive weight value of the selected candidate path is maximum or not, if so, executing the step (8), otherwise, executing the step (7 a);
(8) considering the spectrum converter, the selected service request is processed:
(8a) judging whether a candidate path passes through a node for placing a spectrum converter in all candidate paths, if so, executing the step (8b), otherwise, executing the step (9) after blocking the selected service request;
(8b) judging whether idle spectrum blocks meeting the frequency spectrum adjacency condition exist on links before and after the frequency spectrum conversion node in a candidate path passing through the node where the frequency spectrum converter is placed, if so, executing the step (8c), otherwise, executing the step (9) after the selected service request is blocked;
(8c) allocating the first free spectrum block meeting the spectrum adjacency condition to the selected service request, and executing the step (9);
(9) updating the state of the network resource;
(10) judging whether the service request set is empty, if so, executing the step (11), otherwise, executing the step (3);
(11) and processing all service requests.
2. The RSA method with service prediction and spectrum transformation in elastic optical network as claimed in claim 1, wherein the relevant information of service request is trained by the supervised learning algorithm in steps (4c) and (4f) to predict the change information of service request on link, and reserve resources for future service request in advance, so as to effectively select more suitable optical path for the currently arriving service request; the specific steps of the supervised learning algorithm are as follows:
the method comprises the following steps that firstly, data in a training sample set are randomly selected, and a central vector, an expanded width vector and a weight from a hidden layer to an output layer of a radial basis function are initialized, wherein the training sample set in the step (4c) is composed of arrival time information of all service requests in a training service request sample set, and the training sample set in the step (4f) is composed of duration information of all service requests in the training service request sample set;
secondly, inputting training sample data at the input layer side of the RBFNN, and calculating the output value of each neuron in the RBFNN hidden layer of the RBFNN according to the following formula:
Figure FDA0002421793840000031
in the formula, hjRepresenting the output value of the j-th neuron in the hidden layer of the RBFNN, j being 1,2,3, …, p, p representing the total number of neurons in the hidden layer, exp (·) representing exponential operation with e as the base, | | | · | | | representing European norm operation, X representing the input vector in the RBFNN, CjRepresents the center vector, D, of the jth neuron in the hidden layer of the RBFNNjRepresents the extended width vector of the jth neuron in the hidden layer of the RBFNN, which influences the range of action of the neuron on the input vector, DjThe larger the hidden layer has, the larger the influence range of the hidden layer on the input vector is, and the better the smoothness among the neurons is;
thirdly, carrying out linear change on the output of the hidden layer of the radial basis function neural network RBFNN according to the following formula to obtain the value of an output neuron:
Figure FDA0002421793840000041
in the formula, yiRepresents the output value of the ith neuron in the output layer of the RBFNN, sigma represents the summation operation, wijRepresenting the weight from the jth neuron in the hidden layer of the radial basis function neural network RBFNN to the ith neuron in the output layer;
fourthly, calculating an objective function value of the RBFNN according to the following formula:
Figure FDA0002421793840000042
wherein E represents the objective function value of the RBFNN, n represents the number of training samples, q represents the number of neurons in the output layer of the RBFNN, and y represents the number of the RBFNNikRepresents the net output value, o, of the k-th output neuron of the RBFNN at the i-th input sampleikRepresenting an expected output value of a k output neuron of the RBFNN at an ith input sample;
fifthly, judging whether the objective function value is greater than the iteration termination precision, if so, executing the sixth step, otherwise, ending the training, wherein the iteration termination precision in the step (4c) is 0.01, and the iteration termination precision in the step (4f) is 0.1;
sixthly, respectively updating the central vector C of the radial basis function neural network RBFNN according to the following three formulasj(t) extended Width vector Dj(t) and weight W from hidden layer to output layerj(t) returning to the fourth step:
Figure FDA0002421793840000043
Figure FDA0002421793840000051
Figure FDA0002421793840000052
in the formula, Cj(t) denotes the centre vector of the t-th update of the jth neuron in the hidden layer of the RBFNN, Cj(t-1) denotes the center vector of the t-1 th update of the jth neuron in the hidden layer of the RBFNN, η1A learning factor representing a radial basis function neural network RBFNN center vector,
Figure FDA0002421793840000053
denotes a partial derivative operation, Dj(t) represents the t-th updated expanded wideband vector of the jth neuron in the hidden layer of the RBFNN, Dj(t-1) represents the expanded width vector of the t-1 th update of the jth neuron in the hidden layer of the radial basis function neural network RBFNN, η2Learning factor, W, representing the RBFNN extended wideband vector of a radial basis function neural networkj(t) represents the output weight vector updated t times by the jth neuron in the hidden layer of the RBFNN, Wj(t-1) represents the output weight vector of t-1 updating of the jth neuron in the hidden layer of the RBFNN, η3And the learning factor represents an output weight vector in the RBFNN hidden layer of the radial basis function neural network.
3. The RSA method with service prediction and spectrum transformation in a resilient optical network as claimed in claim 1, wherein the duration coincidence in step (5a) is the coincidence of the duration of the service request carried on each link of the candidate path of the selected service, the duration of the selected service request and the predicted future increased traffic duration, and the duration coincidence is the degree of duration coincidence, which reflects the resource competition degree of other service requests on the link with the selected service request in the duration range of the selected service request, and the larger the duration coincidence of the link is, the more intense the competition degree on the link is.
4. The RSA method with traffic prediction and spectrum transformation in a resilient optical network as claimed in claim 1, wherein the duration overlap ratio calculation formula in step (5a) is as follows:
Figure FDA0002421793840000061
in the formula (I), the compound is shown in the specification,
Figure FDA0002421793840000062
link/in candidate path representing selected service requestmnLink duration overlap of the up-loaded service request or predicted future service request j with the selected service request i,
Figure FDA0002421793840000063
indicating the departure time of the service request j,
Figure FDA0002421793840000064
indicating the arrival time of the selected service request i,
Figure FDA0002421793840000065
indicating the arrival time of the service request j,
Figure FDA0002421793840000066
indicating the departure time of the selected service request i,
Figure FDA0002421793840000067
indicating the duration of the service request j,
Figure FDA0002421793840000068
indicating the duration of the selected service request i.
5. An RSA method with service prediction and spectrum transformation in a resilient optical network as claimed in claim 1, wherein the formula for calculating the comprehensive weight of the candidate paths containing spectrum transformer in the network in step (6c) is as follows:
Weight(Pk)＝αRT(Pk)+βRU(Pk)+γRL(Pk)
in the formula, Weight (P)k) Representing candidate paths Pkα denotes a weight factor for the duration coincidence, RT(Pk) Representing candidate paths Pkβ represents a weighting factor for the spectral occupancy, RU(Pk) Representing candidate paths PkY represents a weighting factor for the number of path hops, α + β + y is 1, RL(Pk) Representing candidate paths PkNormalized value of the number of path hops.
CN201910273876.6A 2019-04-07 2019-04-07 RSA method adopting service prediction and spectrum conversion in elastic optical network Active CN110011922B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910273876.6A CN110011922B (en) 2019-04-07 2019-04-07 RSA method adopting service prediction and spectrum conversion in elastic optical network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910273876.6A CN110011922B (en) 2019-04-07 2019-04-07 RSA method adopting service prediction and spectrum conversion in elastic optical network

Publications (2)

Publication Number Publication Date
CN110011922A CN110011922A (en) 2019-07-12
CN110011922B true CN110011922B (en) 2020-06-05

Family

ID=67170136

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910273876.6A Active CN110011922B (en) 2019-04-07 2019-04-07 RSA method adopting service prediction and spectrum conversion in elastic optical network

Country Status (1)

Country Link
CN (1) CN110011922B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111385680B (en) * 2020-02-20 2022-02-11 重庆邮电大学 Spectrum allocation method based on mixed spectrum conversion resource pool in elastic optical network
CN111654442B (en) * 2020-04-20 2021-03-19 北京邮电大学 Dynamic service grooming processing method and device in elastic optical network
CN111935008B (en) * 2020-07-29 2022-02-18 中国南方电网有限责任公司 Optical network routing method and system based on physical layer damage constraint of machine learning
CN114205284A (en) * 2021-11-01 2022-03-18 国网宁夏电力有限公司电力科学研究院 Routing resource allocation method and device
CN114520939B (en) * 2022-03-14 2023-02-10 西安电子科技大学 Optical network resource allocation method based on index perceptual evolution
CN114584871B (en) * 2022-04-28 2022-08-05 华南师范大学 Spectrum allocation method, device, storage medium and equipment of elastic optical network

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103595495A (en) * 2013-10-27 2014-02-19 西安电子科技大学 Routing and spectrum resource allocation method for static service flow in elastic optical network
CN104320270A (en) * 2014-10-08 2015-01-28 武汉邮电科学研究院 Elastic optical network control system based on SDN architecture and method thereof
CN108337043A (en) * 2017-12-26 2018-07-27 广东电网有限责任公司电力调度控制中心 Fault recovery method with area fault tolerance in multilayer SDN optical-fiber networks

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103051547B (en) * 2012-12-24 2015-08-05 北京邮电大学 The route of resource perception and frequency spectrum resource allocation method and system in elastic optical network
CN106792285B (en) * 2017-01-03 2020-01-17 重庆邮电大学 Double-weight maximal-cluster scheduling method of elastic optical multicast switching node structure based on shared spectrum converter
CN108809839B (en) * 2018-07-17 2020-12-01 湖南理工学院 Wireless Mesh backbone network flow control method and device

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103595495A (en) * 2013-10-27 2014-02-19 西安电子科技大学 Routing and spectrum resource allocation method for static service flow in elastic optical network
CN104320270A (en) * 2014-10-08 2015-01-28 武汉邮电科学研究院 Elastic optical network control system based on SDN architecture and method thereof
CN108337043A (en) * 2017-12-26 2018-07-27 广东电网有限责任公司电力调度控制中心 Fault recovery method with area fault tolerance in multilayer SDN optical-fiber networks

Also Published As

Publication number Publication date
CN110011922A (en) 2019-07-12

Similar Documents

Publication Publication Date Title
CN110011922B (en) RSA method adopting service prediction and spectrum conversion in elastic optical network
Alvizu et al. Matheuristic with machine-learning-based prediction for software-defined mobile metro-core networks
CN101895422B (en) Method for optimizing resources of static-dynamic mixed service in three-layer network
CN111445111A (en) Electric power Internet of things task allocation method based on edge cooperation
CN101127689B (en) Method for minimizing priority seizing cost
CN110519798A (en) A kind of fiber radio integration Internet resources distribution mechanism and income calculation method
CN113220364A (en) Task unloading method based on vehicle networking mobile edge computing system model
JP2019514309A (en) System and method for communication network service connectivity
CN111885493B (en) Micro-cloud deployment method based on improved cuckoo search algorithm
CN114629769B (en) Traffic map generation method of self-organizing network
CN114666805B (en) Optical network planning method and system suitable for multi-granularity service
CN116208567A (en) Method and system for flow scheduling of SDN network resources of cross-domain data center
Yu et al. A deep learning based RSA strategy for elastic optical networks
CN112187535A (en) Server deployment method and device in fog computing environment
CN103618674B (en) A united packet scheduling and channel allocation routing method based on an adaptive service model
Ghanshala et al. Self-organizing sustainable spectrum management methodology in cognitive radio vehicular adhoc network (cravenet) environment: A reinforcement learning approach
CN107689884A (en) A kind of method of half network plan simulation in kind of power communication
Nguyen et al. Efficient virtual network embedding with node ranking and intelligent link mapping
Sen Intelligent admission and placement of O-RAN slices using deep reinforcement learning
CN100440867C (en) Method of real time soft seizing wave length route photo network
Senouci et al. Dynamic channel assignment in cellular networks: a reinforcement learning solution
Varsha et al. Hybrid TABU-GA search for energy efficient routing in WSN
CN114615705B (en) Single-user resource allocation strategy method based on 5G network
CN115134928B (en) Wireless Mesh network congestion control method with optimized frequency band route
Hsu et al. Performance of adaptive routing strategies in wavelength-routed networks

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