CN115119233A - Clustered wireless communication method and system - Google Patents

Clustered wireless communication method and system Download PDF

Info

Publication number
CN115119233A
CN115119233A CN202210656168.2A CN202210656168A CN115119233A CN 115119233 A CN115119233 A CN 115119233A CN 202210656168 A CN202210656168 A CN 202210656168A CN 115119233 A CN115119233 A CN 115119233A
Authority
CN
China
Prior art keywords
cluster
base station
edge
edge devices
aggregation
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.)
Pending
Application number
CN202210656168.2A
Other languages
Chinese (zh)
Inventor
刘翀赫
刘胜利
任金科
余官定
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.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
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 Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202210656168.2A priority Critical patent/CN115119233A/en
Publication of CN115119233A publication Critical patent/CN115119233A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/02—Traffic management, e.g. flow control or congestion control
    • H04W28/0226—Traffic management, e.g. flow control or congestion control based on location or mobility
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/02—Arrangements for optimising operational condition
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/16—Central resource management; Negotiation of resources or communication parameters, e.g. negotiating bandwidth or QoS [Quality of Service]
    • H04W28/18—Negotiating wireless communication parameters
    • H04W28/20—Negotiating bandwidth
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/16—Central resource management; Negotiation of resources or communication parameters, e.g. negotiating bandwidth or QoS [Quality of Service]
    • H04W28/18—Negotiating wireless communication parameters
    • H04W28/22—Negotiating communication rate
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Quality & Reliability (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

The invention discloses a clustered wireless communication method and a clustered wireless communication system, wherein the method comprises the following steps: step 1, a base station gathers all edge devices participating in communication into a plurality of clusters according to geographic positions, and the edge devices in each cluster use D2D links for communication; step 2, after each edge device carries out decentralized model parameter updating according to local data, the edge device exchanges and updates model parameters with the edge device communicated with the edge device through a D2D link, and parameter aggregation in a cluster is carried out by using a parameter averaging algorithm to obtain aggregation parameters; step 3, the base station selects one edge device from all edge devices in each cluster as a cluster head, and the cluster head uploads the aggregation parameters of the edge devices to the base station; and 4, the base station carries out global aggregation on the aggregation parameters uploaded by all the cluster heads and broadcasts the obtained global model to all the edge devices. The problem of communication flow congestion of the central node in a federated learning scene in which large-scale terminals participate together is solved, and the model convergence rate is improved.

Description

Clustered wireless communication method and system
Technical Field
The invention relates to the technical field of communication, in particular to a clustered wireless communication method and a clustered wireless communication system.
Background
In recent years, with the continuous development of artificial intelligence and big data technology, how to adopt a machine learning algorithm to perform effective information mining and application on massive data has gained a lot of attention and research. Typical machine learning algorithms typically require transmission of training data to a base station equipped with a base station and then centralized training of the machine learning model. However, as more and more edge devices such as mobile phones, smart wearable devices, optical sensors, and monitoring cameras access the internet, data for training models may be collected by various devices, and it is difficult to upload all collected training data to a base station through a wireless communication network due to security problems in data privacy and transmission. Therefore, how to securely access data on heterogeneous devices to effectively train a model has become an urgent research problem to be solved.
Research on the federal Learning Framework (FL) in distributed Learning theory has been the subject of rapid development in the scientific and technological community. The method provides a new model learning mode, and a high-quality global model is cooperatively trained by using a large amount of data distributed on mass equipment. The model training is dispersed to a plurality of terminals for carrying out, the terminals do not need to send original data to a base station, and the parameters or gradient information of the model are uploaded after the model training is carried out by utilizing local data. Thus, data privacy and communication efficiency can be maintained simultaneously on a highly distributed set of devices.
With the continuous development of mobile communication and the rapid growth of mobile intelligent equipment, the computing capacity of terminal equipment is greatly improved, and model training can be carried out. Therefore, by combining the federal learning and the wireless communication network, the computing resources of the terminal nodes can be fully utilized, more effective machine learning is carried out under the condition of ensuring the privacy of the user, however, due to the limited bandwidth resources, the wireless channel is unstable, and huge communication overhead is brought by the transmission of a large number of model parameters, so that larger communication delay is caused.
In a traditional federal learning scenario, terminal devices upload their local gradients or local models to the base station periodically, and the base station aggregates them to obtain a high-quality global model. The framework relies on the base station to collect model parameters or gradients. However, in consideration of practical applications, in a scenario where large-scale terminals participate in federal learning training together, due to a large amount of model information exchanged between the edge device and the base station, communication resources are a key factor that affects the convergence rate. The traditional federal learning architecture can cause traffic congestion problems at the central node, and the convergence rate can drop significantly when the network bandwidth available in the system is small. Therefore, how to solve the problem of communication traffic congestion of the central node in federal learning in a reasonable mode is a problem which needs to be researched urgently at present.
Disclosure of Invention
In view of the above, an object of the present invention is to provide a clustered wireless communication method and system, which can solve the problem of base station communication traffic congestion in large-scale federal learning, reduce communication delay in the training process, and improve convergence rate.
In order to achieve the above object, the embodiments provide the following technical solutions:
in a first aspect, an embodiment provides a clustered wireless communication method, where a system implementing the method includes a base station, and a plurality of edge devices in communication with the base station, and includes the following steps:
step 1, a base station gathers all edge devices participating in communication into a plurality of clusters according to geographic positions, and the edge devices in each cluster use D2D links for communication;
step 2, after each edge device updates the gradient descending model parameters according to the local data, the edge device exchanges and updates the model parameters with the edge device communicated with the edge device through a D2D link, and parameter aggregation in the cluster is carried out by using a parameter averaging algorithm to obtain aggregation parameters;
step 3, the base station selects one edge device from all edge devices in each cluster as a cluster head, and the cluster head uploads the aggregation parameters of the edge devices to the base station;
and 4, the base station carries out global aggregation on the aggregation parameters uploaded by all the cluster heads and broadcasts the obtained global model to all the edge devices.
In step 1 of an embodiment, an apparatus connectivity graph is constructed for each cluster, where edge apparatuses are used as nodes, and a continuous edge is constructed between the nodes according to D2D links between the edge apparatuses to form the apparatus connectivity graph, and an adjacency matrix corresponding to the apparatus connectivity graph represents a D2D link connectivity situation between the nodes, where an element value in the adjacency matrix is 1, which represents that D2D links between two edge apparatuses are connected, and an element value is 0, which represents that D2D links between two edge apparatuses are not connected.
In step 2 of one embodiment, a parameter averaging algorithm is used for parameter aggregation within a cluster, comprising:
Figure BDA0003687889170000031
where i and j are the indices of the edge devices, ω i Which represents the parameters of the polymerization,
Figure BDA0003687889170000032
representing model parameters before update, n k Indicates the number of edge devices in cluster k, a j,i Indicating the D2D link connectivity between edge device i and edge device j:
Figure BDA0003687889170000033
in step 2 of one embodiment, D2D communications within all clusters share bandwidth resource B (1) The optimal bandwidth resource allocation when the communication time required for realizing intra-cluster polymerization is minimized is as follows:
Figure BDA0003687889170000041
wherein k is the index of the cluster, l is the iteration number, i, j represents the index of the edge device, M is the quantization bit number of the model parameter,
Figure BDA0003687889170000042
D2D link e for edge device i to edge device j transmission i,j Allocated bandwidth, P i k,l Is the transmit power of the terminal device i,
Figure BDA0003687889170000043
for D2D link e i,j Channel gain of, N 0 Is the variance of additive white Gaussian noise, E k Is the set of connected edges in the device connectivity graph corresponding to the cluster k.
In step 3 of an embodiment, selecting according to the connectivity of the nodes in the device connectivity graph corresponding to the cluster, that is, selecting the edge device corresponding to the node with the maximum connectivity in each cluster as a cluster head, includes:
Figure BDA0003687889170000044
wherein D (V) represents the degree of the equipment connected with the node V in the graph, namely the number of connected edges connected with the node V, V k And representing the set of nodes in the device connectivity graph corresponding to the cluster k.
In step 3 of an embodiment, selecting a cluster head according to an uplink channel state between an edge device and a base station and a transmit power of the edge device includes:
first, the data transmission rate R between edge device i and base station in cluster k k,i Is represented as follows:
Figure BDA0003687889170000045
then, the communication time T required for the edge device i in cluster k to transmit the model parameters to the base station i (2) Is represented as follows:
Figure BDA0003687889170000051
based on this, cluster head s corresponding to cluster k k Selection mode tableShown below:
Figure BDA0003687889170000052
wherein, B k Expressed as the bandwidth, P, allocated for communication of edge devices with a base station in cluster k k,i Is the transmission power of the edge device i, h k,i The gain of the uplink channel of the edge device i, M is the data quantity of the model parameter to be transmitted, n k Representing the total number of edge devices within cluster k.
In step 3 of one embodiment, the bandwidth resource allocated for communication between the cluster head and the base station is B (2) To minimize the communication time required for global aggregation, bandwidth resource B (2) The optimal allocation is performed as follows:
Figure BDA0003687889170000053
wherein k denotes an index of a cluster, B k Denotes cluster head s k Allocated bandwidth, P k Is a cluster head s k Transmit power of h k Is a cluster head s k Gain of uplink channel, N 0 Is the variance of additive white gaussian noise, and M is the number of quantization bits of the model parameter.
In a second aspect, embodiments provide a clustered wireless communication system, comprising a base station, a plurality of edge devices in communication with the base station,
the base station groups all edge devices participating in communication into a plurality of clusters according to geographic positions, and the edge devices in each cluster communicate by using a D2D link;
after each edge device updates gradient descending model parameters according to local data, the edge devices exchange updated model parameters with the edge devices communicated with the edge devices through D2D links, and parameter aggregation in a cluster is carried out by using a parameter averaging algorithm to obtain aggregation parameters;
the base station selects one edge device from all edge devices in each cluster as a cluster head, and the cluster head uploads the aggregation parameters of the edge devices to the base station;
and the base station carries out global aggregation on the aggregation parameters uploaded by all the cluster heads and broadcasts the obtained global model to all the edge devices.
Compared with the prior art, the invention has the beneficial effects that at least:
according to the geographical position of the edge device, the edge device is divided into a plurality of clusters by adopting a clustering algorithm, decentralized training is performed in each cluster in parallel, the wireless D2D network in the area is effectively utilized, and the problems of limited communication resources and base station flow congestion are solved. Only one cluster head in each cluster transmits model parameters with the central server, so that the data volume of the edge device and the base station is effectively reduced, the communication time is reduced, and the convergence rate of the model for achieving convergence is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a schematic structural diagram of a clustered wireless communication system provided by an embodiment;
fig. 2 is a flow chart of a clustered wireless communication method provided by an embodiment;
FIG. 3 is a schematic diagram of wireless communication interaction provided by an embodiment;
fig. 4 is a schematic diagram of wireless communication interaction provided by another embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is a schematic structural diagram of a clustered wireless communication system according to an embodiment. Embodiments provide a wireless communication system including a base station equipped with a central server, a plurality of edge devices communicating with the base station as terminals. Wherein, the edge devices respectively maintain local data for model training, and the local data and the base station complete the model training together. The clustered wireless communication system can realize the federal learning of large-scale edge devices participating together. Specifically, after clustering edge devices according to geographic positions, each cluster independently conducts decentralized training and model parameter aggregation based on a D2D network, one cluster head is selected from each cluster according to a cluster head selection strategy, and aggregation parameters of the cluster heads are uploaded to a base station to conduct global model aggregation. The problem of central server traffic congestion is solved. The data volume exchanged between the terminal equipment and the base station is effectively reduced, the communication time delay in the training process is reduced, and the model convergence rate is improved.
Fig. 2 is a flow chart of a clustered wireless communication method provided by an embodiment. As shown in fig. 2, the clustered wireless communication method provided by the embodiment applies to the system shown in fig. 1, and includes the following steps:
in step 201, the base station groups all edge devices participating in communication into a plurality of clusters according to the geographical positions, and the edge devices in each cluster communicate by using a D2D link.
In the embodiment, the base station adopts a clustering algorithm to cluster all the edge devices into K clusters according to the address positions, the edge devices in the same cluster are subjected to decentralized machine learning training, and the communication between the edge devices is based on a wireless D2D network.
Due to the influence of path loss, noise and the like, D2D links between some devices are blocked, so that a device connectivity graph is constructed for each cluster to represent the D2D link communication condition of edge devices in the cluster, and the connectivity structure between the devices in the cluster is unchanged in the whole training process. Specifically, edge devices are taken as nodes, and connecting edges are constructed between the nodes according to D2D links between the edge devices to form a device connectivity graph G k (V k ,E k ) In which V is k A set of nodes is represented that is,
Figure BDA0003687889170000081
representing a set of connected edges. The matrix corresponding to the device connectivity graph represents the link connectivity of the D2D between the nodes, where an element value in the matrix is 1, which indicates that the D2D link connectivity between two edge devices is achieved, and an element value is 0, which indicates that the D2D link connectivity between two edge devices is not achieved.
Step 202, after each edge device updates the gradient-descending model parameters according to the local data, the edge device exchanges the updated model parameters with the edge device communicated with the edge device through the D2D link, and performs parameter aggregation in the cluster by using a parameter averaging algorithm to obtain aggregation parameters.
In an embodiment, the edge device i in the cluster k updates the model parameters using a stochastic gradient descent algorithm, specifically according to the local data set D i Middle random sampling data sample xi t,l Carrying out gradient descent updating on the model parameters, and obtaining the model parameters through the l iteration of the t round of training
Figure BDA0003687889170000082
Expressed as:
Figure BDA0003687889170000083
wherein the content of the first and second substances,
Figure BDA0003687889170000084
representing the model parameter, η, obtained in the first-1 iteration t,l For the learning rate of the ith iteration of the tth round of training,. v.is the gradient operator, and L is the loss function.
In the embodiment, the edge device i in the cluster k exchanges model parameters with the edge device communicated with the edge device i based on the D2D network communication to perform aggregation of the model parameters in the cluster, the aggregation parameter of the edge device i after the model parameters are averaged is used as the starting point of the (l + 1) th iteration, and the aggregation parameter is
Figure BDA0003687889170000085
Expressed as:
Figure BDA0003687889170000086
wherein n is k Indicates the number of edge devices in cluster k, a j,i Indicating the D2D link connectivity between edge device i and edge device j:
Figure BDA0003687889170000087
in an embodiment, the available bandwidth resource of the wireless communication system is B. For the intra-cluster decentralized training phase, all intra-cluster D2D communications share bandwidth resource B (1) . The bandwidth resource used by the cluster head and the base station for communication is B (2) 。
B＝B (1) +B (2)
Each terminal device transmits the model parameters to the edge devices communicated with the terminal device in the device communication graph, and because the geographic distance between clusters is long, the communication interference between the edge devices of different clusters can be regarded as noise.
For the cluster k, a Frequency Division Multiple Access (FDMA) mode is adopted for B (1) Dividing the channel into a plurality of sub-channels, wherein each sub-channel is allocated to one directed edge (i → j) of the device connectivity graph and is marked as e i,j . Consider the D2D link e that edge i transmits to edge j when the l-th D2D communication inside cluster k is taken into account i,j Allocated bandwidth of
Figure BDA0003687889170000091
The relationship is as follows:
Figure BDA0003687889170000092
transmission rate of communication when edge device i transmits model parameters to edge device j
Figure BDA0003687889170000093
Is represented as follows:
Figure BDA0003687889170000094
wherein, P i k,l Is the transmit power of the terminal device i,
Figure BDA0003687889170000095
for D2D link e i,j Channel gain of, N 0 Is the variance of additive white gaussian noise.
Consider the maximum time required for edge device i to communicate with edge device j the l < th > D2D assuming that the amount of model parameters that need to be transmitted each time is M bits
Figure BDA0003687889170000096
Expressed as:
Figure BDA0003687889170000097
accumulating the time required for the cluster k to perform the L-round in-cluster decentralized training
Figure BDA0003687889170000098
Is represented as follows:
Figure BDA0003687889170000101
the communication time T required by all K clusters to finish L rounds of intra-cluster decentralized training (1) Is represented as follows:
Figure BDA0003687889170000102
in order to minimize the communication time T required for intra-cluster polymerization (1) Resource B with bandwidth (1) The optimal allocation is performed as follows:
Figure BDA0003687889170000103
step 203, the base station selects one edge device from all edge devices in each cluster as a cluster head, and the cluster head uploads the aggregation parameters to the base station.
In an embodiment, after L times of model parameter updating and intra-cluster model parameter aggregation are performed in K clusters using step 2, according to a cluster head selection policy, one edge device is selected as a cluster head in each cluster, and a set of cluster head devices is represented by S ═ S { (S) } S 1 ,s 2 ,...,s K }。
In one possible implementation manner, the adopted cluster head selection strategy is to select according to the connectivity of the nodes of the device connectivity graph in the cluster. The connectivity D (v) of the node v in the device connectivity graph refers to the number of edges connected with the node v. The larger the D (v), the larger the number of other nodes connected with the node, that is, the larger the connectivity of the node represented by the edge device, the larger the model parameter can be exchanged with more devices when the intra-cluster parameter aggregation is performed. And after parameter averaging, more accurate aggregation parameters representing the model after local training of the devices in the cluster can be obtained. Therefore, the edge device with the largest degree of connectivity in each cluster is selected as a cluster head, and the aggregation parameters are uploaded to the base station. Cluster head s in cluster k k The selection is represented as follows:
Figure BDA0003687889170000111
in another possible real-time mode, the cluster head selection strategy is selected according to the uplink channel state between the edge device and the base station and the transmission power of the edge device. Specifically, considering the communication delay of the system, the bandwidth resource allocated by the communication between the device and the server is B (2) Data transmission rate R between edge device i and base station in cluster k k,i Is represented as follows:
Figure BDA0003687889170000112
wherein, B k Expressed as the bandwidth, P, allocated for communication of edge devices with a base station in cluster k k,i Is the transmission power of the edge device i, h k,i For the uplink channel gain of the edge device i, the communication time T between the edge device i and the base station in the cluster k i (2) Is represented as follows:
Figure BDA0003687889170000113
wherein, M is the data amount of the model parameter to be transmitted.
Communication time T of edge device i i (2) The shorter the uplink channel communication quality is, the better the uplink channel communication quality is, the training time required by the whole federal learning to reach convergence can be reduced, and the whole convergence rate of the model is improved. T is i (2) Depending on the transmit power of the device at that time and its uplink channel gain. Therefore, the cluster head selection strategy selects the edge device with the best channel state in each cluster as a cluster head, and sends the aggregation parameters to the base station. Cluster head s in cluster k k The selection is represented as follows:
Figure BDA0003687889170000114
in the embodiment, the K cluster heads upload the cluster aggregation parameters to the base station, and the bandwidth resource allocated by the cluster heads and the base station in communication is B (2) 。
Using Orthogonal Frequency Division Multiple Access (OFDMA) to convert B into B (2) Divided into K sub-channels. Cluster head s k Allocated bandwidth of B k Then, the relationship is as follows:
Figure BDA0003687889170000121
cluster head s k Data transmission rate R with base station k Is represented as follows:
Figure BDA0003687889170000122
wherein, P k To cluster heads s k Transmit power of h k Is a cluster head s k Gain of uplink channel, N 0 Is the variance of additive white gaussian noise.
The communication time T required for one global aggregation (2) Is represented as follows:
Figure BDA0003687889170000123
to minimize the communication time T required for global aggregation (2) Resource B with bandwidth (2) The optimal allocation is performed as follows:
Figure BDA0003687889170000124
and 204, the base station performs global aggregation on the aggregation parameters uploaded by all the cluster heads, and broadcasts the obtained global model to all the edge devices.
In the embodiment, K cluster heads respectively train the aggregation parameters
Figure BDA0003687889170000125
Uploading the K aggregation parameters to a base station, and averaging the K aggregation parameters by the base station to obtain a global model omega t+1 Expressed as:
Figure BDA0003687889170000131
the base station broadcasts the global model omega to all N devices t+1 。
205, repeating the above steps 201-204, and performing T-round training until the global model converges.
According to the clustering wireless communication method, the edge equipment is divided into different clusters, so that the problems of limited communication resources and central server flow congestion are solved; independent decentralized training is carried out in each cluster, and the wireless D2D network in the region is effectively utilized; and a cluster head equipment selection strategy is adopted to select cluster heads to upload aggregation parameters, so that the convergence rate is improved.
Fig. 3 is a schematic diagram of wireless communication interaction provided by an embodiment. As shown in fig. 3, the method comprises the following steps:
301. and performing a plurality of decentralized training times on all the clusters in parallel, wherein the decentralized training comprises local training by using local data and intra-cluster parameter aggregation by using wireless D2D communication.
302. And all the edge devices upload the connectivity information of the edge devices to the base station.
303. And the scheduler on the base station selects the edge equipment with the maximum intra-cluster connectivity as a cluster head according to the collected connectivity information.
304. And allocating bandwidth for the uplink, and uploading the aggregation parameters to the base station by the cluster head of each cluster.
305. And the base station aggregates the global model through parameter averaging.
306. The base station transmits the global model to all terminal devices in a broadcast manner.
Fig. 4 is a schematic diagram of wireless communication interaction provided by another embodiment. As shown in fig. 4, the method comprises the following steps:
401. and performing a plurality of decentralized training times on all the clusters in parallel, wherein the decentralized training comprises local training by using local data and intra-cluster parameter aggregation by using wireless D2D communication.
402. And all the edge equipment terminals upload the uplink channel state information of the edge equipment terminals to the base station.
403. And the scheduler on the base station selects the edge equipment with the best channel state in each cluster as a cluster head according to the collected uplink channel state information.
404. And allocating bandwidth for the uplink, and uploading the aggregation parameters to the base station by the cluster head of each cluster.
405. And the base station aggregates the global model through parameter averaging.
406. The base station transmits the global model to all terminal devices in a broadcast manner.
In the embodiment, the communication mode between the edge device and the base station is a wireless communication mode, and may be an existing mobile communication network, that is, an LTE (Long-term Evolution) or 5G network, or a WiFi network.
In an embodiment, the base station comprises a central server with processor capability far exceeding the computing capability of the edge device and the capability of independent model training. The edge device can be a mobile terminal which can support model training, such as a modern smart phone, a tablet computer, a notebook computer, an automatic driving automobile and the like, is provided with a wireless communication system, and can be accessed to mainstream wireless communication networks such as a mobile communication network and WiFi.
In the wireless communication method provided by the embodiment, in a federal learning scene, the device is divided into a plurality of clusters according to the geographical position through a clustering algorithm, and decentralized training is performed in each cluster in parallel, so that the problem of central node communication flow congestion in the federal learning scene in which large-scale terminals participate together is solved, the data volume exchanged between the terminal device and the base station is effectively reduced, the communication delay in the training process is reduced, and the model convergence rate is improved.
The above-mentioned embodiments are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions, equivalents, etc. made within the scope of the principles of the present invention should be included in the scope of the present invention.

Claims (9)

1. A clustered wireless communication method, a system implementing the method comprising a base station, a plurality of edge devices in communication with the base station, comprising the steps of:
step 1, a base station gathers all edge devices participating in communication into a plurality of clusters according to geographic positions, and the edge devices in each cluster use D2D links for communication;
step 2, after each edge device updates the gradient descending model parameters according to the local data, the edge device exchanges and updates the model parameters with the edge device communicated with the edge device through a D2D link, and parameter aggregation in the cluster is carried out by using a parameter averaging algorithm to obtain aggregation parameters;
step 3, the base station selects one edge device from all edge devices in each cluster as a cluster head, and the cluster head uploads the aggregation parameters of the edge devices to the base station;
and 4, the base station carries out global aggregation on the aggregation parameters uploaded by all the cluster heads and broadcasts the obtained global model to all the edge devices.
2. The clustered wireless communication method according to claim 1, wherein in step 1, a device connectivity graph is constructed for each cluster, wherein, with edge devices as nodes, connectivity edges are constructed between the nodes according to D2D links between the edge devices to form the device connectivity graph, and an adjacency matrix corresponding to the device connectivity graph characterizes the D2D link connectivity between the nodes, wherein an element value in the adjacency matrix is 1, which indicates that D2D links between two edge devices are connected, and an element value is 0, which indicates that D2D links between two edge devices are not connected.
3. The clustered wireless communication method of claim 1 wherein the step 2 of performing intra-cluster parameter aggregation using a parameter averaging algorithm comprises:
Figure FDA0003687889160000011
where i and j are the indices of the edge devices, ω i Which represents the parameters of the polymerization,
Figure FDA0003687889160000012
representing model parameters before update, n k Indicates the number of edge devices in cluster k, a j,i Indicating the D2D link connectivity between edge device i and edge device j:
Figure FDA0003687889160000021
4. the method of claim 1, wherein in step 2, D2D communication in all clusters shares bandwidth resource B (B:) 1 ) The optimal bandwidth resource allocation when the communication time required for realizing intra-cluster polymerization is minimized is as follows:
Figure FDA0003687889160000022
wherein k is the index of the cluster, l is the iteration number, i, j represents the index of the edge device, M is the quantization bit number of the model parameter,
Figure FDA0003687889160000023
D2D link e for edge device i to edge device j transmission i,j Allocated bandwidth, P i k,l Is the transmit power of the terminal device i,
Figure FDA0003687889160000024
for D2D link e i,j Channel gain of, N 0 Is the variance of additive white Gaussian noise, E k Is the set of connected edges in the device connectivity graph corresponding to the cluster k.
5. The clustered wireless communication method according to claim 2, wherein in the step 3, selecting according to connectivity of nodes in the device connectivity graph corresponding to the cluster, that is, selecting an edge device corresponding to a node with the largest connectivity in each cluster as a cluster head, includes:
Figure FDA0003687889160000025
wherein D (V) represents the degree of the equipment connected with the node V in the graph, namely the number of connected edges connected with the node V, V k Represents a cluster k pairThe corresponding device communicates with the set of nodes in the graph.
6. The clustered wireless communication method as claimed in claim 1, wherein the step 3 of selecting the cluster head according to the uplink channel status between the edge device and the base station and the transmission power of the edge device comprises:
first, the data transmission rate R between edge device i and base station in cluster k k,i Is represented as follows:
Figure FDA0003687889160000031
then, the communication time T required for the edge device i in cluster k to transmit the model parameters to the base station i (2) Is represented as follows:
Figure FDA0003687889160000032
based on this, the cluster head s corresponding to the cluster k k The selection is represented as follows:
Figure FDA0003687889160000033
wherein, B k Expressed as the bandwidth, P, allocated for communication of edge devices with a base station in cluster k k,i Is the transmission power of the edge device i, h k,i The gain of the uplink channel of the edge device i, M is the data quantity of the model parameter to be transmitted, n k Representing the total number of edge devices within cluster k.
7. The method of claim 1 wherein in step 3, the bandwidth resource allocated for communication between the cluster head and the base station is B (2) To minimize the communication time required for global aggregation, bandwidth resource B (2) The optimal allocation is performed as follows:
Figure FDA0003687889160000041
where k denotes the index of the cluster, B k Denotes cluster head s k Allocated bandwidth, P k Is a cluster head s k Transmit power of h k Is a cluster head s k Gain of uplink channel, N 0 Is the variance of additive white gaussian noise, and M is the number of quantization bits of the model parameter.
8. The clustered wireless communication method of claim 1, wherein in step 1, a clustering algorithm is used to cluster all edge devices into a plurality of clusters according to geographical locations;
in step 2, all edge devices in each cluster adopt a random gradient descent algorithm to update model parameters.
9. A clustered wireless communication system comprising a base station, a plurality of edge devices in communication with the base station,
the base station groups all the edge devices participating in communication into a plurality of clusters according to the geographical positions, and the edge devices in each cluster communicate by using a D2D link;
after each edge device updates gradient descending model parameters according to local data, the edge devices exchange updated model parameters with the edge devices communicated with the edge devices through D2D links, and parameter aggregation in a cluster is carried out by using a parameter averaging algorithm to obtain aggregation parameters;
the base station selects one edge device from all edge devices in each cluster as a cluster head, and the cluster head uploads the aggregation parameters of the edge devices to the base station;
and the base station carries out global aggregation on the aggregation parameters uploaded by all the cluster heads and broadcasts the obtained global model to all the edge devices.
CN202210656168.2A 2022-06-10 2022-06-10 Clustered wireless communication method and system Pending CN115119233A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210656168.2A CN115119233A (en) 2022-06-10 2022-06-10 Clustered wireless communication method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210656168.2A CN115119233A (en) 2022-06-10 2022-06-10 Clustered wireless communication method and system

Publications (1)

Publication Number Publication Date
CN115119233A true CN115119233A (en) 2022-09-27

Family

ID=83325808

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210656168.2A Pending CN115119233A (en) 2022-06-10 2022-06-10 Clustered wireless communication method and system

Country Status (1)

Country Link
CN (1) CN115119233A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115580891A (en) * 2022-12-09 2023-01-06 北京邮电大学 Flow prediction model training method, prediction method and device based on federal learning
CN115828638A (en) * 2023-01-09 2023-03-21 西安深信科创信息技术有限公司 Automatic driving test scene script generation method and device and electronic equipment

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115580891A (en) * 2022-12-09 2023-01-06 北京邮电大学 Flow prediction model training method, prediction method and device based on federal learning
CN115828638A (en) * 2023-01-09 2023-03-21 西安深信科创信息技术有限公司 Automatic driving test scene script generation method and device and electronic equipment

Similar Documents

Publication Publication Date Title
CN108718463B (en) Resource allocation method based on multi-time scale collaborative optimization under H-CRAN
CN112488322B (en) Federal learning model training method based on data feature perception aggregation
CN111447083B (en) Federal learning framework under dynamic bandwidth and unreliable network and compression algorithm thereof
CN112020103B (en) Content cache deployment method in mobile edge cloud
CN115119233A (en) Clustered wireless communication method and system
WO2019096005A1 (en) Networking method and apparatus, network access method, and user equipment
EP3854125B1 (en) Fronthaul configuration based on facilitating cell allocation to baseband units and hubs of a communications network
Klymash et al. State of the art and further development of information and communication systems
CN109699033B (en) LoRa power Internet of things base station deployment method and device for cost and load balancing
CN110213776B (en) WiFi unloading method based on Q learning and multi-attribute decision
TWI698111B (en) COMMUNICATION SYSTEM OF QoE-ORIENTED CROSS-LAYER BEAM ALLOCATION AND ADMISSION CONTROL FOR FUNCTIONAL SPLITTED WIRELESS FRONTHAUL COMMUNICATIONS
CN115802389B (en) Federal learning method trained by digital twin auxiliary model
CN109905864A (en) A kind of cross-layer Resource Allocation Formula towards electric power Internet of Things
CN115392481A (en) Federal learning efficient communication method based on real-time response time balancing
Li et al. Cognitive carrier resource optimization for internet-of-vehicles in 5g-enhanced smart cities
CN115865930A (en) MEC dynamic adjustment method based on 5G power Internet of things
CN105530203B (en) The connection control method and system of D2D communication links
Ganjalizadeh et al. Interplay between distributed AI workflow and URLLC
US20240031427A1 (en) Cloud-network integration oriented multi-access edge computing architecture
Chilukuri et al. Achieving optimal cache utility in constrained wireless networks through federated learning
CN113055860A (en) D2D many-to-many resource allocation method in cellular network
Dinh et al. Enabling large-scale federated learning over wireless edge networks
CN116133082A (en) Multi-hop clustering method for improving topology duration of aviation ad hoc network
CN114390489B (en) End-to-end network slice servitization deployment method
CN114423070A (en) D2D-based heterogeneous wireless network power distribution method and 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