Adaptive modulation and coding method, network device, network element and storage medium

Abstract

The application discloses an adaptive modulation and coding method, a network device, a network element and a storage medium. The method comprises the following steps: when it is determined that base stations in a jurisdictional area satisfy a trigger condition of a smart AMC algorithm based on federated learning, the network device groups the base stations based on a similarity algorithm of base station feature information; a federated learning model network element corresponding to each group of base stations is determined; and the federated learning model network element is triggered to execute a life cycle process, and the life cycle process is used to obtain a smart AMC algorithm model, and all base stations in the same group share one smart AMC algorithm model. The embodiment can collect rich data resources of different devices, improve model accuracy, generalization and other performances; the multiple base stations in the federated learning model network element simultaneously perform model training, thereby saving the time for constructing the model; and the global model of the federated training suitable for the base station group is obtained by using a base station grouping rule to perform inference and prediction, thereby saving computing power and obtaining performance gain brought by the smart algorithm.

Description

Technical Field

[0001] This application relates to the field of mobile communication technology, specifically to an adaptive modulation and coding method, network equipment, network element, and storage medium. Background Technology

[0002] In 4G Long Term Evolution (LTE) and 5G New Radio (NR) wireless communication systems, Adaptive Modulation and Coding (AMC) technology is typically used to address the time-varying characteristics of the wireless channel and ensure the transmission quality of the wireless link. AMC is a technique that adaptively adjusts the modulation scheme and coding rate of the wireless link transmission according to changes in the wireless channel: when the channel quality is poor, a lower-order modulation and coding scheme (MCS) is selected, while when the channel quality is good, a higher-order MCS is selected, thereby maximizing the transmission efficiency of the wireless link while ensuring its reliability.

[0003] Traditional AMC schemes cannot accurately predict the channel quality at future uplink data transmission moments. When the channel environment fluctuates significantly, it can lead to problems such as high network bit error rate and decreased throughput performance, thus making it impossible to select a suitable MCS. Summary of the Invention

[0004] At least one embodiment of this application provides an adaptive modulation and coding method, network device, network element, and storage medium to solve the problem that the AMC scheme cannot select a suitable MCS.

[0005] To solve the above-mentioned technical problems, this application is implemented as follows:

[0006] In a first aspect, embodiments of this application provide an adaptive modulation and coding method applied to a network device, comprising:

[0007] When a network device determines that the base stations within its jurisdiction meet the triggering conditions of the Smart Adaptive Modulation and Coding (AMC) algorithm based on federated learning, it groups the base stations based on a similarity algorithm of base station feature information.

[0008] Determine the federated learning model network element corresponding to each group of base stations;

[0009] The federated learning model network element is triggered to execute the lifecycle process, which is used to obtain the intelligent AMC algorithm model. All base stations in the same group share one intelligent AMC algorithm model.

[0010] Optionally, the method further includes:

[0011] Receive performance information of the intelligent AMC algorithm model sent by the base station;

[0012] The performance of the intelligent AMC algorithm model is evaluated based on the performance information.

[0013] The similarity threshold in the similarity algorithm is adjusted based on the performance evaluation results.

[0014] Optionally, the determination that the base stations within the jurisdiction meet the triggering conditions of the federated learning-based intelligent adaptive modulation and coding (AMC) algorithm includes one of the following:

[0015] Receive first indication information sent by base stations within the jurisdiction, the first indication information being used to indicate that the base stations meet the triggering conditions of the intelligent AMC algorithm based on federated learning;

[0016] Receive network performance indicators and / or second indication information sent by base stations within the jurisdiction; determine, based on the network performance indicators and / or the second indication information, that the base station meets the triggering condition of the intelligent AMC algorithm based on federated learning; the second indication information is used to indicate that the base station meets the first condition.

[0017] Optionally, the triggering condition includes at least one of the following:

[0018] The network performance indicators of the base stations within the jurisdiction do not meet the user's Quality of Service (QoS) requirements;

[0019] The base stations within the jurisdiction meet the first condition;

[0020] The first condition includes at least one of the following:

[0021] The computing power of the base station is insufficient to support the training of artificial intelligence (AI) models;

[0022] Base station power consumption cannot support AI model training;

[0023] The prediction accuracy of the base station AI model training is below the first threshold.

[0024] The convergence time of the base station AI model training does not meet the user's QoS requirements;

[0025] Federated learning offers a performance gain over non-federated learning.

[0026] Optionally, the similarity algorithm based on base station feature information groups the base stations, including:

[0027] Based on the feature information of each base station, calculate the feature information similarity between different base stations;

[0028] The base stations are grouped according to the similarity of the feature information and a preset similarity threshold.

[0029] Optionally, the feature information includes at least one of the following:

[0030] Base station hardware information;

[0031] Base station configuration information;

[0032] Base station user information;

[0033] Base station wireless channel environment information;

[0034] Input characteristics of the intelligent algorithm of the base station;

[0035] Output characteristics of the base station's intelligent algorithm;

[0036] The model expression of the intelligent algorithm of the base station.

[0037] Optionally, determining the federated learning model network element corresponding to each group of base stations includes at least one of the following:

[0038] For a group of base stations, the local model training nodes of the group of base stations are determined based on the computing power resources of the base stations within the group and / or the similarity of feature information between the base stations.

[0039] For a group of base stations, a global model aggregation node is determined based on the computing power resources of the base stations in the group and / or the connection resources between the base stations in the group and the local model training node. The global model aggregation node is one of the base stations in the group or a network device.

[0040] Optionally, the local model training node satisfies a second condition, which includes at least one of the following:

[0041] The computing resources are sufficient to meet the needs of AI intelligent algorithms;

[0042] The characteristic information of the base station is the second threshold of all data information of the group of base stations.

[0043] Optionally, the global model aggregation node satisfies a third condition, which includes at least one of the following:

[0044] The computing resources are sufficient to meet the computing resource requirements when aggregating and / or updating model parameter information;

[0045] The connection resources can meet the latency requirements when aggregating and / or sending model parameter information;

[0046] To meet the synchronization requirements of sending different model parameter information to the base station.

[0047] Optionally, triggering the lifecycle process of the federated learning model network element includes:

[0048] A third instruction message is sent to the federated learning model network element, the third instruction message being used to instruct the federated learning model network element to execute a lifecycle process.

[0049] Optionally, the performance evaluation of the intelligent AMC algorithm model based on the performance information includes:

[0050] The AI ​​model performance metrics of the intelligent AMC algorithm model are evaluated based on the performance information.

[0051] The performance metrics of the AI ​​model include at least one of the following:

[0052] The accuracy of model training;

[0053] Data distribution coverage;

[0054] Convergence time of model training;

[0055] Generalization;

[0056] robustness;

[0057] Energy consumption.

[0058] Optionally, the performance evaluation of the intelligent AMC algorithm model based on the performance information includes:

[0059] The performance metrics of the intelligent AMC algorithm model are evaluated based on the performance information.

[0060] The performance metrics of the AMC algorithm include at least one of the following:

[0061] Throughput;

[0062] Communication reliability;

[0063] Delay;

[0064] Initial Block Error Rate (IBLER).

[0065] Secondly, embodiments of this application provide an adaptive modulation and coding method applied to network elements of a federated learning model, including:

[0066] The federated learning model network element executes the lifecycle process to obtain the intelligent AMC algorithm model;

[0067] The intelligent AMC algorithm model is sent to a group of base stations corresponding to the network elements of the federated learning model.

[0068] Optionally, the federated learning model network elements include: local model training nodes and global model aggregation nodes;

[0069] The execution lifecycle process, to obtain the intelligent AMC algorithm model, includes:

[0070] The local model training node downloads the initial model file from the global model aggregation node;

[0071] The local model training node performs local model training based on local data and the initial model file to obtain updated model information;

[0072] The local model training node sends the updated model information to the global model aggregation node;

[0073] The global model aggregation node integrates the updated model information sent by all local model training nodes to obtain the intelligent AMC algorithm model.

[0074] Optionally, the method further includes:

[0075] The third instruction information sent by the network device is received, which is used to instruct the federated learning model network element to execute the life cycle process.

[0076] Thirdly, embodiments of this application provide a network device, including a transceiver and a processor, wherein...

[0077] The processor is configured to group the base stations based on a similarity algorithm of base station feature information, provided that the base stations within its jurisdiction meet the triggering conditions of the intelligent adaptive modulation and coding (AMC) algorithm based on federated learning.

[0078] Determine the federated learning model network element corresponding to each group of base stations;

[0079] The federated learning model network element is triggered to execute the lifecycle process, which is used to obtain the intelligent AMC algorithm model. All base stations in the same group share one intelligent AMC algorithm model.

[0080] Optionally, the transceiver is used to: receive performance information of the intelligent AMC algorithm model sent by the base station;

[0081] The processor is further configured to: evaluate the performance of the intelligent AMC algorithm model based on the performance information; and adjust the similarity threshold in the similarity algorithm based on the performance evaluation results.

[0082] Optionally, the processor determines that the base stations within its jurisdiction meet the triggering conditions of the federated learning-based intelligent adaptive modulation and coding (AMC) algorithm, including one of the following:

[0083] Receive first indication information sent by base stations within the jurisdiction, the first indication information being used to indicate that the base stations meet the triggering conditions of the intelligent AMC algorithm based on federated learning;

[0084] Receive network performance indicators and / or second indication information sent by base stations within the jurisdiction; determine, based on the network performance indicators and / or the second indication information, that the base station meets the triggering condition of the intelligent AMC algorithm based on federated learning; the second indication information is used to indicate that the base station meets the first condition.

[0085] Optionally, the triggering condition includes at least one of the following:

[0086] The network performance indicators of base stations within the jurisdiction do not meet the QoS requirements of users;

[0087] The base stations within the jurisdiction meet the first condition;

[0088] The first condition includes at least one of the following:

[0089] The base station's computing power is insufficient to support AI model training;

[0090] Base station power consumption cannot support AI model training;

[0091] The prediction accuracy of the base station AI model training is below the first threshold.

[0092] The convergence time of the base station AI model training does not meet the user's QoS requirements;

[0093] Federated learning offers a performance gain over non-federated learning.

[0094] Optionally, the processor groups the base stations based on a similarity algorithm of base station feature information, including:

[0095] Based on the feature information of each base station, calculate the feature information similarity between different base stations;

[0096] The base stations are grouped according to the similarity of the feature information and a preset similarity threshold.

[0097] Optionally, the feature information includes at least one of the following:

[0098] Base station hardware information;

[0099] Base station configuration information;

[0100] Base station user information;

[0101] Base station wireless channel environment information;

[0102] Input characteristics of the intelligent algorithm of the base station;

[0103] Output characteristics of the base station's intelligent algorithm;

[0104] The model expression of the intelligent algorithm of the base station.

[0105] Optionally, the processor determines the federated learning model network element corresponding to each group of base stations, including at least one of the following:

[0106] For a group of base stations, the local model training nodes of the group of base stations are determined based on the computing power resources of the base stations within the group and / or the similarity of feature information between the base stations.

[0107] For a group of base stations, a global model aggregation node is determined based on the computing power resources of the base stations in the group and / or the connection resources between the base stations in the group and the local model training node. The global model aggregation node is one of the base stations in the group or a network device.

[0108] Optionally, the local model training node satisfies a second condition, which includes at least one of the following:

[0109] The computing resources are sufficient to meet the needs of AI intelligent algorithms;

[0110] The characteristic information of the base station is the second threshold of all data information of the group of base stations.

[0111] Optionally, the global model aggregation node satisfies a third condition, which includes at least one of the following:

[0112] The computing resources are sufficient to meet the computing resource requirements when aggregating and / or updating model parameter information;

[0113] The connection resources can meet the latency requirements when aggregating and / or sending model parameter information;

[0114] To meet the synchronization requirements of sending different model parameter information to the base station.

[0115] Optionally, the processor triggers the lifecycle process of the federated learning model element, including:

[0116] A third instruction message is sent to the federated learning model network element, the third instruction message being used to instruct the federated learning model network element to execute a lifecycle process.

[0117] Optionally, the processor performs a performance evaluation of the intelligent AMC algorithm model based on the performance information, including:

[0118] The AI ​​model performance metrics of the intelligent AMC algorithm model are evaluated based on the performance information.

[0119] The performance metrics of the AI ​​model include at least one of the following:

[0120] The accuracy of model training;

[0121] Data distribution coverage;

[0122] Convergence time of model training;

[0123] Generalization;

[0124] robustness;

[0125] Energy consumption.

[0126] Optionally, the processor performs a performance evaluation of the intelligent AMC algorithm model based on the performance information, including:

[0127] The performance metrics of the intelligent AMC algorithm model are evaluated based on the performance information.

[0128] The performance metrics of the AMC algorithm include at least one of the following:

[0129] Throughput;

[0130] Communication reliability;

[0131] Delay;

[0132] IBLER.

[0133] Fourthly, embodiments of this application provide a federated learning model network element, including a transceiver and a processor, wherein,

[0134] The processor is used to: execute lifecycle processes to obtain an intelligent AMC algorithm model;

[0135] The transceiver is used to send the intelligent AMC algorithm model to a group of base stations corresponding to the federated learning model network element.

[0136] Optionally, the federated learning model network elements include: local model training nodes and global model aggregation nodes;

[0137] The processor executes a lifecycle process to obtain an intelligent AMC algorithm model, including:

[0138] The local model training node downloads the initial model file from the global model aggregation node;

[0139] The local model training node performs local model training based on local data and the initial model file to obtain updated model information;

[0140] The local model training node sends the updated model information to the global model aggregation node;

[0141] The global model aggregation node integrates the updated model information sent by all local model training nodes to obtain the intelligent AMC algorithm model.

[0142] Optionally, the transceiver is further configured to: receive third indication information sent by the network device, the third indication information being used to instruct the federated learning model network element to execute a lifecycle process.

[0143] Fifthly, embodiments of this application provide an adaptive modulation and coding apparatus, comprising:

[0144] The grouping module is used to group the base stations based on a similarity algorithm of base station feature information, provided that the base stations within the jurisdiction meet the triggering conditions of the intelligent adaptive modulation and coding (AMC) algorithm based on federated learning.

[0145] The first determination module is used to determine the federated learning model network element corresponding to each group of base stations;

[0146] The first processing module is used to trigger the execution lifecycle process of the federated learning model network element. The lifecycle process is used to obtain the intelligent AMC algorithm model. All base stations in the same group share one intelligent AMC algorithm model.

[0147] Sixthly, embodiments of this application provide an adaptive modulation and coding apparatus, comprising:

[0148] The second processing module is used to execute the lifecycle process and obtain the intelligent AMC algorithm model.

[0149] The first sending module is used to send the intelligent AMC algorithm model to a group of base stations corresponding to the federated learning model network element.

[0150] In a seventh aspect, embodiments of this application provide a network device, including: a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the adaptive modulation and coding method described above.

[0151] Eighthly, embodiments of this application provide a federated learning model network element, characterized in that it includes: a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the adaptive modulation and coding method described above.

[0152] Ninthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the adaptive modulation and coding method described above.

[0153] In this embodiment, when a network device determines that a base station meets the triggering conditions of the intelligent AMC algorithm, it groups the base stations based on a similarity algorithm of base station feature information; it determines the federated learning model network element corresponding to each group of base stations and triggers the federated learning model network element to execute its lifecycle process, ultimately obtaining the intelligent AMC algorithm model. A group of base stations shares this intelligent AMC algorithm model. This embodiment can collect rich data resources from different devices, improving model accuracy, generalization, and other performance; the simultaneous model training of multiple base stations in the federated learning model network element greatly saves model construction time; by using base station grouping rules to find the base station group, a globally federated model suitable for that base station group can be obtained for inference and prediction, saving computing power and obtaining the performance gains brought by the intelligent algorithm. Attached Figure Description

[0154] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0155] Figure 1 This is one of the flowcharts illustrating the adaptive modulation and coding method according to an embodiment of this application;

[0156] Figure 2 This is a second schematic flowchart of the adaptive modulation and coding method according to an embodiment of this application;

[0157] Figure 3 This is the third flowchart illustrating the adaptive modulation and coding method according to an embodiment of this application;

[0158] Figure 4 This is one of the structural schematic diagrams of the adaptive modulation and coding apparatus according to an embodiment of this application;

[0159] Figure 5 This is a second schematic diagram of the adaptive modulation and coding apparatus according to an embodiment of this application;

[0160] Figure 6 This is one of the structural schematic diagrams of a network device according to an embodiment of this application;

[0161] Figure 7 This is one of the structural schematic diagrams of the network elements of the federated learning model in this application embodiment;

[0162] Figure 8 This is a second schematic diagram of the network device according to an embodiment of this application;

[0163] Figure 9 This is the second schematic diagram of the structure of the network element of the federated learning model in this application embodiment. Detailed Implementation

[0164] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0165] The terms “first,” “second,” etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The terms “and / or” in the specification and claims indicate at least one of the connected objects.

[0166] The technologies described in this document are not limited to NR systems and Long Time Evolution (LTE) / LTE-Advanced (LTE-A) systems, but can also be used in various wireless communication systems such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency-Division Multiple Access (SC-FDMA), and other systems. The terms "system" and "network" are often used interchangeably. CDMA systems can implement radio technologies such as CDMA2000 and Universal Terrestrial Radio Access (UTRA). UTRA includes Wideband Code Division Multiple Access (WCDMA) and other CDMA variants. TDMA systems can implement radio technologies such as the Global System for Mobile Communication (GSM). OFDMA systems can implement radio technologies such as Ultra Mobile Broadband (UMB), Evolution-UTRA (E-UTRA), IEEE 802.21 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20, and Flash-OFDM. UTRA and E-UTRA are part of the Universal Mobile Telecommunications System (UMTS). LTE and more advanced LTE (such as LTE-A) are newer versions of UMTS that use E-UTRA. UTRA, E-UTRA, UMTS, LTE, LTE-A, and GSM are described in documents from an organization called the 3rd Generation Partnership Project (3GPP). CDMA2000 and UMB are described in documents from an organization called 3rd Generation Partnership Project 2 (3GPP2).The techniques described herein can be used with the systems and radio technologies mentioned above, as well as with other systems and radio technologies. However, the following description describes NR systems for illustrative purposes, and NR terminology is used in most of the following description, although these techniques can also be applied to applications beyond NR systems.

[0167] The following description provides examples and is not intended to limit the scope, applicability, or configuration set forth in the claims. Changes may be made to the function and arrangement of the elements discussed without departing from the spirit and scope of this disclosure. Various procedures or components may be appropriately omitted, substituted, or added to the examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with reference to certain examples may be combined in other examples.

[0168] As described in the background section, existing technologies based on traditional AMC schemes cannot accurately predict the channel quality at future uplink data transmission times. When the channel environment fluctuates significantly, it can lead to problems such as high network bit error rate and decreased throughput performance, thus potentially preventing the selection of a suitable MCS. To address at least one of the above problems, embodiments of this application provide an adaptive modulation and coding method that can reduce or avoid the occurrence of the above situations, making the predicted channel quality at future uplink data transmission times more accurate, thereby helping the AMC module select a suitable MCS value.

[0169] Please refer to Figure 1 This application provides an adaptive modulation and coding method, applied to a network device, comprising:

[0170] Step 101: When the network device determines that the base stations within its jurisdiction meet the triggering conditions of the Smart Adaptive Modulation and Coding (AMC) algorithm based on federated learning, it groups the base stations based on the similarity algorithm of the base station feature information.

[0171] In this embodiment, the network device can be a base station or an upper-layer network device, such as: Operation, Administration and Maintenance (OAM) equipment, Near Real Time RAN Intelligent Controller (Near-RT RIC) defined by the Open Radio Access Network (O-RAN) architecture, Non-real time RAN Intelligent Controller (Non-RT RIC), etc.

[0172] Intelligent solutions based on AI / Machine Learning (ML) algorithms, such as intelligent AMC solutions, require model training on base stations, which has a much higher computational cost than traditional algorithms that are not based on AI / ML. However, intelligent algorithms are not always necessary. Network devices can determine whether the triggering conditions for intelligent AMC algorithms based on federated learning are met based on the network performance indicators of the base station. If the triggering conditions are met, the intelligent algorithm solution based on federated learning is triggered.

[0173] Step 102: Determine the federated learning model network element corresponding to each group of base stations;

[0174] The network device can use a similarity algorithm based on the feature information of base stations within its jurisdiction to group all or some of the base stations within the jurisdiction. Optionally, for example, it can group base stations that meet the triggering conditions. Each group of base stations shares a single intelligent AMC algorithm model, so the network device can determine federated learning model network elements for each group of base stations. The federated learning model network element includes: one or more local model training nodes and a global model aggregation node.

[0175] Step 103: Trigger the lifecycle process of the federated learning model network element. The lifecycle process is used to obtain the intelligent AMC algorithm model. All base stations in the same group share one intelligent AMC algorithm model.

[0176] In this embodiment, the network device triggers the federated learning model network element to perform model training, executes the lifecycle process, and finally obtains the intelligent AMC algorithm model. Specifically, the global model aggregation node in the federated learning model network element aggregates and processes the model information reported by each local model training node to obtain the intelligent AMC algorithm model. This intelligent AMC algorithm model is a shared intelligent AMC algorithm model among multiple base stations within a set of base stations. Adaptive modulation and coding can be implemented based on this intelligent AMC algorithm model.

[0177] In this embodiment, when a network device determines that a base station meets the triggering conditions of the intelligent AMC algorithm, it groups the base stations based on a similarity algorithm of base station feature information; it determines the federated learning model network element corresponding to each group of base stations and triggers the federated learning model network element to execute its lifecycle process, ultimately obtaining the intelligent AMC algorithm model. A group of base stations shares this intelligent AMC algorithm model. This embodiment can collect rich data resources from different devices, improving model accuracy, generalization, and other performance; the simultaneous model training of multiple base stations in the federated learning model network element greatly saves model construction time; by using base station grouping rules to find the base station group, a globally federated model suitable for that base station group can be obtained for inference and prediction, saving computing power and obtaining the performance gains brought by the intelligent algorithm.

[0178] As an optional embodiment, the determination that the base stations within the jurisdiction meet the triggering conditions of the federated learning-based intelligent adaptive modulation and coding (AMC) algorithm includes one of the following:

[0179] (1) Receive first indication information sent by a base station within the jurisdiction, wherein the first indication information is used to indicate that the base station meets the triggering conditions of the intelligent AMC algorithm based on federated learning;

[0180] (2) Receive network performance indicators and / or second indication information sent by base stations within the jurisdiction; determine the triggering conditions of the intelligent AMC algorithm based on federated learning based on the network performance indicators and / or the second indication information; the second indication information is used to indicate that the base station meets the first condition.

[0181] In this embodiment, the network device can determine whether the base station meets the triggering conditions of the federated learning-based intelligent AMC algorithm based on the indication information reported by the base station. That is, the base station itself can determine whether it meets the triggering conditions of the federated learning-based intelligent AMC algorithm based on network performance indicators, its own capabilities, etc., and report this to the network device. Alternatively, the base station may only report network performance indicators and / or second indication information to the network device. The second indication information can indicate its own capabilities and whether its capabilities meet the first condition. Then, the network device can determine whether the base station meets the triggering conditions of the federated learning-based intelligent AMC algorithm based on the base station's network performance indicators and / or the second indication information. Optionally, the network device can also monitor the base station's network performance indicators to determine whether the base station meets the triggering conditions of the federated learning-based intelligent AMC algorithm.

[0182] As an optional embodiment, the triggering condition includes at least one of the following:

[0183] (1) The network performance indicators of the base stations within the jurisdiction do not meet the user's QoS requirements; optionally, not meeting the user's QoS requirements may be that the network performance indicators are greater than or less than the threshold set for the network performance indicators.

[0184] Optionally, the network performance metrics include at least one of the following:

[0185] 1) IBLER;

[0186] 2) Throughput;

[0187] 3) Delay;

[0188] 4) Communication reliability (comprehensive evaluation of network performance indicators).

[0189] (2) The base stations within the jurisdiction meet the first condition; wherein the first condition includes at least one of the following:

[0190] a) The base station's computing power cannot support AI model training; for example, the base station's computing power does not support AI model training, or the base station's computing power-related parameter values ​​do not meet the relevant thresholds for AI model training.

[0191] b) Base station power consumption cannot support AI model training; for example: base station power consumption is limited and cannot support AI model training, or the base station power consumption parameter values ​​do not meet the relevant thresholds for AI model training.

[0192] c) The training and prediction accuracy of the base station AI model is lower than the first threshold;

[0193] d) The training convergence time of the base station AI model does not meet the user's QoS requirements; for example, when the training convergence time of the base station AI model is long, it leads to a decrease in network performance and fails to meet the user's QoS requirements.

[0194] e) Federated learning offers a performance gain over non-federated learning, as when federated learning offers a performance gain compared to non-federated learning schemes.

[0195] In this embodiment, the federated learning-based intelligent algorithm scheme trains the model on multiple base stations. However, not all situations require the intelligent scheme and federated learning model framework; therefore, it is necessary to design how to trigger the intelligent scheme and federated learning model framework. The federated learning-based intelligent algorithm is triggered when a base station meets certain conditions.

[0196] As an optional embodiment, the similarity algorithm based on base station feature information groups the base stations, including:

[0197] Based on the feature information of each base station, calculate the feature information similarity between different base stations;

[0198] The base stations are grouped according to the similarity of the feature information and a preset similarity threshold.

[0199] Optionally, the feature information includes at least one of the following:

[0200] 1) Base station hardware information; base station type, base station antenna information.

[0201] 2) Base station configuration information; base station time-frequency resource scheduling configuration parameters, and user multiple-input multiple-output (MIMO) related configuration parameters.

[0202] 3) Base station user information; user QoS requirements, user distribution, user mobility speed, terminal capabilities, and user service types.

[0203] 4) Base station wireless channel environment information; interference information, wireless channel information

[0204] 5) Input characteristics of the base station's intelligent algorithm; for example, the intelligent AMC algorithm includes, but is not limited to, the following aspects:

[0205] a: Configuration parameters related to the AMC algorithm: such as target IBLER, outer loop control acknowledgment (ACK) step size, outer loop control negative acknowledgment (NACK) step size, initial MCS, etc.

[0206] b: Time-frequency resource scheduling information related to the AMC algorithm: such as Channel Quality Indicator (CQI), MCS, Resource Block (RB) resource allocation, etc.

[0207] c: Intelligent algorithm models: such as Long Short Term Memory networks (LSTM), Convolutional Neural Networks (CNN) and model hyperparameters.

[0208] 6) Output characteristics of the base station's intelligent algorithm;

[0209] 7) Model expression of the intelligent algorithm of the base station.

[0210] In this embodiment, network devices can classify base stations based on their hardware and configuration information, user information, wireless channel environment, input / output characteristics of intelligent algorithms, and algorithm model expressions. This allows base stations within the same group to jointly train an intelligent algorithm model based on a federated learning framework. Specifically, the base station characteristic information related to base station grouping includes, but is not limited to, the aspects mentioned above.

[0211] Network devices group base stations based on their characteristic information. For example, if different base stations share the same intelligent algorithm-related features and models, they can be further grouped based on the similarity of their related feature data. Within the same group, federated learning can be used to jointly train the same algorithm model parameters.

[0212] Since the feature attributes of each base station are quite similar, and the data samples of each base station exist independently, a horizontal federated learning framework is adopted. Optionally, the cosine similarity algorithm can be used to calculate the similarity of feature information. For example, taking the time series of a certain feature of base station 1 as x and the time series of a certain feature of base station 2 as y, and substituting them into the following formula:

[0213]

[0214] The higher the similarity value calculated by the above formula, the more similar base station 1 and base station 2 are in a certain feature, and the more likely they are to be classified as the same group of base stations.

[0215] Optionally, base stations whose similarity values ​​meet a preset threshold can be grouped together. Optionally, the principles for grouping base stations include at least one of the following:

[0216] 1) The more similar the data information of the related features of different base stations, the higher the prediction accuracy after aggregation, but the performance is poor on unknown datasets, that is, the generalization is poor.

[0217] 2) The greater the difference in data information related to different base stations, the lower the prediction accuracy and the stronger the generalization.

[0218] Multiple thresholds can be set based on the prediction accuracy of the AI / ML model in the intelligent algorithm. When the prediction accuracy exceeds threshold 1, the number of base station classifications can be reduced; when the prediction accuracy is lower than threshold 2, the number of base station classifications can be increased.

[0219] As an optional embodiment, determining the federated learning model network element corresponding to each group of base stations includes at least one of the following:

[0220] (1) For a group of base stations, the local model training nodes of the group of base stations are determined based on the computing power resources of the base stations in the group and / or the similarity of feature information between the base stations.

[0221] Optionally, the local model training node satisfies a second condition, which includes at least one of the following:

[0222] The computing resources are sufficient to meet the needs of AI intelligent algorithms;

[0223] The characteristic information of the base station is the second threshold of all data information of the group of base stations.

[0224] In this embodiment, based on the results of base station grouping, a subset of base station nodes can be selected for local training of federated learning for each group of base stations. The selection principles may include: that the computing resources of the base station nodes within the group are sufficient to meet the computing power requirements of the AI ​​algorithm; and / or, that among the nodes with sufficient computing resources, at least a certain number of nodes are selected, and that their data information can represent a certain percentage of all data information within the group, based on the data similarity of the input features.

[0225] (2) For a group of base stations, the global model aggregation node of the group of base stations is determined based on the computing power resources of the base stations in the group and / or the connection resources between the base stations in the group and the local model training node. The global model aggregation node is one of the base stations in the group of base stations or a network device.

[0226] Optionally, the global model aggregation node satisfies a third condition, which includes at least one of the following:

[0227] The computing resources are sufficient to meet the computing resource requirements when aggregating and / or updating model parameter information;

[0228] The connection resources can meet the latency requirements when aggregating and / or sending model parameter information;

[0229] To meet the synchronization requirements of sending different model parameter information to the base station.

[0230] In this embodiment, for each group of base stations, a base station node or other intelligent network device can be selected as the aggregation node for the federated learning of that group to complete the aggregation and updating of model parameters. That is, the global model aggregation node can be a base station or an intelligent network device.

[0231] The principles for selecting aggregation nodes may include at least one of the following:

[0232] The computing resources are sufficient to meet the computing power requirements for aggregating and updating AI models.

[0233] The connectivity resources are sufficient to meet the latency requirements for aggregating and distributing AI model parameters to various local model training nodes.

[0234] Implementing intelligent algorithms for the same base station requires two AI / ML models, but the two models are located in different base station groups. Therefore, it is necessary to consider the synchronization of the parameters of different AI / ML models when they arrive at the base station at different aggregation nodes.

[0235] After selecting base station groups, model training nodes, and global model aggregation nodes, an initial template for the federated learning AI algorithm model is formulated.

[0236] Optionally, triggering the lifecycle process of the federated learning model network element includes:

[0237] A third instruction message is sent to the federated learning model network element, the third instruction message being used to instruct the federated learning model network element to execute a lifecycle process.

[0238] In this embodiment, the network device can instruct the federated learning model network elements to execute a lifecycle procedure. Optionally, the lifecycle procedure of the federated learning model includes, but is not limited to, the following steps:

[0239] 1) The local model training node downloads the initial model file from the global model aggregation node or other network devices;

[0240] 1) Local model training: Local model training nodes improve the model by learning from local data;

[0241] 3) Model generalization (autonomy): Local model training nodes generalize improvements to the model into a smaller update;

[0242] 4) Local model upload (autonomous): The local model training node encrypts the model update and sends it to the global model aggregation node;

[0243] 5) Model integration (joint): The global model aggregation node integrates the update information of multiple local models to obtain a shared model, namely the intelligent AMC algorithm model.

[0244] 6) Model distribution: The global model aggregation node distributes the shared model to all base stations within the group.

[0245] As an optional embodiment, the method further includes:

[0246] Receive performance information of the intelligent AMC algorithm model sent by the base station;

[0247] The performance of the intelligent AMC algorithm model is evaluated based on the performance information.

[0248] The similarity threshold in the similarity algorithm is adjusted based on the performance evaluation results.

[0249] In this embodiment, the network device can evaluate and optimize the obtained intelligent AMC algorithm model. The base station reports the performance information of the intelligent AMC algorithm model to the network device, and the network device evaluates the intelligent AMC algorithm model based on this information. Furthermore, the similarity threshold in the similarity algorithm can be adjusted according to the evaluation node, thereby optimizing the intelligent AMC algorithm model.

[0250] Optionally, the step of evaluating the performance of the intelligent AMC algorithm model based on the performance information includes: evaluating the AI ​​model performance indicators of the intelligent AMC algorithm model based on the performance information;

[0251] The performance metrics of the AI ​​model include at least one of the following:

[0252] The accuracy of model training;

[0253] Data distribution coverage;

[0254] Convergence time of model training;

[0255] Generalization;

[0256] robustness;

[0257] Energy consumption.

[0258] Optionally, the step of evaluating the performance of the intelligent AMC algorithm model based on the performance information includes: evaluating the AMC algorithm performance indicators of the intelligent AMC algorithm model based on the performance information;

[0259] The performance metrics of the AMC algorithm include at least one of the following:

[0260] Throughput;

[0261] Communication reliability;

[0262] Delay;

[0263] IBLER.

[0264] Optionally, network devices can dynamically adjust the similarity threshold based on the performance metrics of the training model. For example, if the training model is overfitting, it indicates that the dataset is not rich enough, and the similarity threshold can be lowered to allow more base stations to form federated learning groups. If the training model is underfitting, it indicates that the model complexity is insufficient to express the information of the dataset, and the similarity threshold can be increased to reduce the number of base stations and the diversity of the data.

[0265] The implementation method of the adaptive modulation and coding is illustrated below through specific embodiments.

[0266] like Figure 2 As shown, it includes:

[0267] Step 21: The intelligent network device triggers the intelligent AMC algorithm scheme based on federated learning.

[0268] Specifically, step 211: monitoring the performance indicators related to the intelligent AMC algorithm.

[0269] When the network performance indicators of base station nodes, such as IBERL, throughput, and latency, fail to meet the threshold, the intelligent AMC algorithm scheme can be triggered.

[0270] Optionally, network devices can send subscription requests to the base station to subscribe to the base station's network performance metrics. For example, the base station might trigger a report when its network performance metrics are below or above a threshold.

[0271] Step 212: Trigger the intelligent AMC algorithm scheme based on federated learning.

[0272] Optionally, for base stations whose network performance indicators do not meet the threshold, if their computing power is insufficient to support AI model training, or their energy-saving conditions are insufficient to support AI model training, or their training accuracy cannot reach the threshold for model deployment; or if the convergence time of the AI ​​model training of this base station does not meet the user's QoS requirements, then the intelligent AMC algorithm scheme based on federated learning can be triggered.

[0273] The base station or upper-layer network device, acting as an intelligent network device, initiates a federated learning request to jointly train an intelligent AMC model with other base stations and receives model parameters that have reached the required accuracy for model deployment, thus triggering the intelligent AMC algorithm mechanism based on federated learning.

[0274] Step 22: Collaborative orchestration and management of network elements in the federated learning model.

[0275] Step 221: Implement base station grouping and cluster base stations based on similarity algorithms.

[0276] This embodiment requires further aggregation and classification of base stations based on the input and output feature dimensions and model expressions of the intelligent algorithm, so that the intelligent AMC algorithm model can be shared among base stations in the same group.

[0277] In the intelligent AMC algorithm designed in this embodiment, the models that need to be trained are the uplink interference time series prediction model and the channel matrix time series prediction model, using algorithms such as LSTM.

[0278] The input of the uplink interference time series prediction model 1 is the uplink interference prediction value RIP, and the output is the uplink interference prediction value RIP(n+Δ) at the future time Δ. Its input-output logic relationship is universal among different base stations. Base stations can be grouped based on similarity algorithms, and multiple base station devices can cooperate to train a time series prediction model.

[0279] The input to the channel matrix time series prediction model 2 is the channel matrix H(k,r,p), and the output is the channel matrix H at the future time Δ. n+Δ (k,r,p) has a universal input-output logic relationship across different base stations. It can be used to group base stations based on similarity algorithms, and multiple base station devices can cooperate to train a time series prediction model.

[0280] For the uplink interference time series prediction model 1, the time series of the uplink interference prediction value RIP has different time fluctuation period, trend and amplitude in different base stations. By using a similarity algorithm, a group of base stations with a similarity of uplink interference prediction value RIP sequence higher than the threshold 1 over a period of time can be grouped together to train a smart AMC algorithm model.

[0281] The input channel matrix H of the channel matrix time series prediction model Model 2 n+Δ The time series (k,r,p) has different time fluctuation periods, trends, and amplitudes at different base stations. By using a similarity algorithm, base stations with a similarity of uplink interference prediction values ​​(such as received interference power (RIP)) over a historical period of time can be grouped together and trained on a single model.

[0282] The process of base station grouping in the federated learning framework for uplink interference time series prediction model 1 and channel matrix time series prediction model 2 is independent of each other, meaning that different partitioning methods can be used.

[0283] Optionally, the grouping principle for base stations is as follows:

[0284] 1) The more similar the data information related to different base stations, the higher the prediction accuracy after aggregation, but the poorer the generalization.

[0285] 2) The greater the difference in data information related to different base stations, the lower the prediction accuracy and the stronger the generalization.

[0286] Step 222: Select local model training nodes based on the data similarity between base stations within the group and the computing power resources of each base station.

[0287] Based on the base station grouping results, for each group, a subset of base station nodes are selected as local model training nodes for federated learning local training. The selection criteria will not be elaborated here.

[0288] Step 223: Select a global model aggregation node based on the connection resources and computing power resources between the node and the local training node. The selection principles will not be elaborated here.

[0289] Step 23: Trigger the lifecycle process of the federated learning model.

[0290] Step 24: The intelligent network device subscribes to the performance information of the intelligent AMC algorithm model from the base station, and evaluates the performance indicators of the AI ​​model and the intelligent AMC algorithm based on the performance information of the intelligent AMC algorithm model.

[0291] The intelligent AMC scheme in this embodiment obtains the uplink channel noise ratio UL_SINR at the time of future Physical Uplink Shared Channel (PUSCH) data service transmission by predicting uplink interference and channel matrix using the following formula (3), thereby improving the accuracy of base station selection of MCS and meeting the requirements of high reliability and low latency of wireless network.

[0292]

[0293]

[0294]

[0295] Among them, S n+Δ(k,r) represents the useful signal strength received by the base station for each subcarrier; S(n+Δ) is the average useful signal strength per RB / resource block group (RBG) for scheduled user i in a certain frequency domain. UL_SINR(n+Δ) is the uplink channel noise ratio at time (n+Δ). Specifically, the channel matrix prediction value at time tti(n+Δ) is obtained using channel prediction. The useful signal strength S received by the base station for each subcarrier is further calculated using formula (1). n+Δ (k,r). X n+Δ (k,p) is the reference signal transmitted by the user at time tti(n+Δ), which is known information of the base station based on the base station's configuration of the transmitter and uplink power control function. K is the total number of subcarriers of the scheduled user i in a certain frequency domain, R is the number of receiving antennas of the base station, P is the total number of antennas of the transmitter, and N_sc is the number of subcarriers contained in each RB / RBG. The average useful signal strength S(n+Δ) per RB / RBG for the scheduled user i in a certain frequency domain can be calculated by formula (2). Finally, UL_SINR(n+Δ) is obtained by formula (3).

[0296] The uplink interference prediction input in formula (3) is characterized by the uplink interference value RIP received by each RB / RBG within the working frequency band of the serving cell, including historical data and the interference value in the current statistical period, i.e., the time series of the uplink interference value of each RB / RBG. The output is the uplink interference prediction value RIP(n+Δ) at the future time Δ.

[0297] The channel prediction input in formula (3) consists of channel data from the requesting user, which is a K*R*P dimensional matrix H(k,r,p), where K is the number of working frequency domain subcarriers of the user, R is the number of receiving antennas, and P is the number of transmitting antennas. It includes historical data and channel data for the current statistical period, i.e., the R*P dimensional matrix time series of each subcarrier. The output is the channel matrix H at a future time Δ. n+Δ (k,r,p).

[0298] Uplink interference prediction and channel prediction both fall under the category of time series prediction, and can be implemented using LSTM models, or RNN, STL+GRU models, etc. The output is the uplink channel data at the next k time steps.

[0299] Based on the uplink interference prediction and channel data at the future time Δ, the uplink channel noise ratio UL_SINR at the future time Δ is obtained.

[0300] In the intelligent AMC algorithm, the models that need to be trained are uplink interference time series prediction models and channel matrix time series prediction models, such as LSTM models. The input and output feature dimensions of the models, as well as the input and output logical relationships, are universal across different base stations. Therefore, multiple base station devices can be used in a federated learning framework to collaboratively train a time series prediction model.

[0301] This application provides application scenarios for intelligent AMC algorithm schemes that can utilize federated learning frameworks. The input and output feature dimensions and model expressions of the intelligent AMC algorithm are universal across different base station devices, making it suitable for use with network learning frameworks. This application design includes triggering conditions for the federated learning-based intelligent AMC scheme to save computing power when there is sufficient demand. It also designs a method to classify base stations with universal intelligent AMC algorithm models into one category based on the similarity of data distribution, and to apply federated learning among base stations of the same category. Based on data differences and remaining computing power, base station nodes are selected to train local model parameters. Base stations or other network devices with low transmission costs and sufficient computing power are selected as nodes for global model aggregation to establish a shared machine learning model.

[0302] In this application, the network device can dynamically adjust the similarity threshold based on the performance metrics of the training model. When the training model is overfitting, it indicates that the dataset is not rich enough, and the similarity threshold can be lowered to allow more base stations to form federated learning groups. When the training model is underfitting, it indicates that the model complexity is insufficient to express the information of the dataset, and the similarity threshold can be increased to reduce the number of base stations and the diversity of data.

[0303] The adaptive modulation and coding scheme (AMCS) of this application is based on uplink interference and channel matrix prediction values ​​to obtain the channel quality at the future uplink data service transmission time, which can improve the accuracy of base station selection of MCS. The method first uses uplink interference data and channel matrix data to predict the uplink interference and channel matrix values ​​at the future uplink data transmission time. The base station then processes the predicted values ​​appropriately, converting them into uplink interference and channel matrix values ​​per RB / RBG in a certain frequency domain for scheduled user i. These values ​​are then provided to the base station's AMC module to calculate the signal-to-interference-plus-noise ratio (SINR) of the channel quality, thereby helping the AMC module select a suitable MCS value.

[0304] The embodiments of this application can collect data from different devices to enrich data resources and improve model accuracy, generalization and other performance; multiple base stations can train the model at the same time, which greatly saves the time of building the model; for base station nodes with insufficient computing power, inference and prediction can be performed through a global model trained in a federated manner, saving computing power.

[0305] like Figure 3 As shown in the embodiments of this application, an adaptive modulation and coding method is also provided, applied to network elements of a federated learning model, including:

[0306] Step 301: The federated learning model network element executes the lifecycle process to obtain the intelligent AMC algorithm model;

[0307] Step 302: Send the intelligent AMC algorithm model to a group of base stations corresponding to the federated learning model network element.

[0308] In this embodiment, after grouping base stations, the network device can determine the corresponding federated learning model network element for each group of base stations and trigger the execution lifecycle process of the federated learning model network element. The execution lifecycle process of the federated learning model network element can obtain an intelligent AMC algorithm model. For the same base station group, all base stations within the group share a single intelligent AMC algorithm model. The federated learning model network element can distribute the intelligent AMC algorithm model to the private base stations within the group.

[0309] Optionally, the federated learning model network elements include: local model training nodes and global model aggregation nodes;

[0310] The execution lifecycle process, to obtain the intelligent AMC algorithm model, includes:

[0311] The local model training node downloads the initial model file from the global model aggregation node;

[0312] The local model training node performs local model training based on local data and the initial model file to obtain updated model information;

[0313] The local model training node sends the updated model information to the global model aggregation node;

[0314] The global model aggregation node integrates the updated model information sent by all local model training nodes to obtain the intelligent AMC algorithm model.

[0315] Optionally, the method further includes:

[0316] The third instruction information sent by the network device is received, which is used to instruct the federated learning model network element to execute the life cycle process.

[0317] In this embodiment, the network device can instruct the federated learning model network element to execute the lifecycle process through third instruction information.

[0318] In the embodiments of this application, the simultaneous model training by multiple base stations greatly saves the time for building the model; for base station nodes with insufficient computing power, inference and prediction can be performed through a globally federated model, saving computing power.

[0319] The various methods of the embodiments of this application have been described above. Apparatus for implementing the above methods will now be provided.

[0320] Please refer to Figure 4 This application also provides an adaptive modulation and coding apparatus 400, applied to a network device, comprising:

[0321] The grouping module 410 is used to group the base stations based on a similarity algorithm of base station feature information when it is determined that the base stations within the jurisdiction meet the triggering conditions of the intelligent adaptive modulation and coding (AMC) algorithm based on federated learning.

[0322] The first determining module 420 is used to determine the federated learning model network element corresponding to each group of base stations;

[0323] The first processing module 430 is used to trigger the execution lifecycle process of the federated learning model network element. The lifecycle process is used to obtain the intelligent AMC algorithm model. All base stations in the same group share one intelligent AMC algorithm model.

[0324] Optionally, the device further includes:

[0325] The first receiving module is used to receive the performance information of the intelligent AMC algorithm model sent by the base station;

[0326] The evaluation module is used to evaluate the performance of the intelligent AMC algorithm model based on the performance information.

[0327] An optimization module is used to adjust the similarity threshold in the similarity algorithm based on the performance evaluation results.

[0328] Optionally, the apparatus further includes: a second determining module, configured to perform one of the following:

[0329] Receive first indication information sent by base stations within the jurisdiction, the first indication information being used to indicate that the base stations meet the triggering conditions of the intelligent AMC algorithm based on federated learning;

[0330] Receive network performance indicators and / or second indication information sent by base stations within the jurisdiction; determine, based on the network performance indicators and / or the second indication information, that the base station meets the triggering condition of the intelligent AMC algorithm based on federated learning; the second indication information is used to indicate that the base station meets the first condition.

[0331] Optionally, the triggering condition includes at least one of the following:

[0332] The network performance indicators of base stations within the jurisdiction do not meet the QoS requirements of users;

[0333] The base stations within the jurisdiction meet the first condition;

[0334] The first condition includes at least one of the following:

[0335] The base station's computing power is insufficient to support AI model training;

[0336] Base station power consumption cannot support AI model training;

[0337] The prediction accuracy of the base station AI model training is below the first threshold.

[0338] The convergence time of the base station AI model training does not meet the user's QoS requirements;

[0339] Federated learning offers a performance gain over non-federated learning.

[0340] Optionally, the grouping module is specifically used for:

[0341] Based on the feature information of each base station, calculate the feature information similarity between different base stations;

[0342] The base stations are grouped according to the similarity of the feature information and a preset similarity threshold.

[0343] Optionally, the feature information includes at least one of the following:

[0344] Base station hardware information;

[0345] Base station configuration information;

[0346] Base station user information;

[0347] Base station wireless channel environment information;

[0348] Input characteristics of the intelligent algorithm of the base station;

[0349] Output characteristics of the base station's intelligent algorithm;

[0350] The model expression of the intelligent algorithm of the base station.

[0351] Optionally, the first determining module is specifically configured to perform at least one of the following:

[0352] For a group of base stations, the local model training nodes of the group of base stations are determined based on the computing power resources of the base stations within the group and / or the similarity of feature information between the base stations.

[0353] For a group of base stations, a global model aggregation node is determined based on the computing power resources of the base stations in the group and / or the connection resources between the base stations in the group and the local model training node. The global model aggregation node is one of the base stations in the group or a network device.

[0354] Optionally, the local model training node satisfies a second condition, which includes at least one of the following:

[0355] The computing resources are sufficient to meet the needs of AI intelligent algorithms;

[0356] The characteristic information of the base station is the second threshold of all data information of the group of base stations.

[0357] Optionally, the global model aggregation node satisfies a third condition, which includes at least one of the following:

[0358] The computing resources are sufficient to meet the computing resource requirements when aggregating and / or updating model parameter information;

[0359] The connection resources can meet the latency requirements when aggregating and / or sending model parameter information;

[0360] To meet the synchronization requirements of sending different model parameter information to the base station.

[0361] Optionally, the first processing module is specifically used for:

[0362] A third instruction message is sent to the federated learning model network element, the third instruction message being used to instruct the federated learning model network element to execute a lifecycle process.

[0363] Optionally, the evaluation module is specifically used for:

[0364] The AI ​​model performance metrics of the intelligent AMC algorithm model are evaluated based on the performance information.

[0365] The performance metrics of the AI ​​model include at least one of the following:

[0366] The accuracy of model training;

[0367] Data distribution coverage;

[0368] Convergence time of model training;

[0369] Generalization;

[0370] robustness;

[0371] Energy consumption.

[0372] Optionally, the evaluation module is specifically used for:

[0373] The performance metrics of the intelligent AMC algorithm model are evaluated based on the performance information.

[0374] The performance metrics of the AMC algorithm include at least one of the following:

[0375] Throughput;

[0376] Communication reliability;

[0377] Delay;

[0378] IBLER.

[0379] It should be noted that the device in this embodiment corresponds to the method applied to network devices described above. The implementation methods in the above embodiments are also applicable to the embodiments of this device and can achieve the same technical effect. The parts and beneficial effects that are the same as those in the method embodiments will not be described in detail here.

[0380] Please refer to Figure 5 This application also provides an adaptive modulation and coding apparatus 500, comprising:

[0381] The second processing module 510 is used to execute the lifecycle process and obtain the intelligent AMC algorithm model.

[0382] The first sending module 520 is used to send the intelligent AMC algorithm model to a group of base stations corresponding to the federated learning model network element.

[0383] Optionally, the federated learning model network elements include: local model training nodes and global model aggregation nodes;

[0384] The second processing module 510 is specifically used for:

[0385] The local model training node downloads the initial model file from the global model aggregation node;

[0386] The local model training node performs local model training based on local data and the initial model file to obtain updated model information;

[0387] The local model training node sends the updated model information to the global model aggregation node;

[0388] The global model aggregation node integrates the updated model information sent by all local model training nodes to obtain the intelligent AMC algorithm model.

[0389] Optionally, the device further includes:

[0390] The second receiving module receives third instruction information sent by the network device, the third instruction information being used to instruct the federated learning model network element to execute the lifecycle process.

[0391] It should be noted that the device in this embodiment corresponds to the method applied to network elements of the federated learning model described above. The implementation methods in the above embodiments are also applicable to the embodiments of this device and can achieve the same technical effect. The parts and beneficial effects that are the same as those in the method embodiments will not be described in detail here.

[0392] Please refer to Figure 6 This application also provides a network device 600, including: a transceiver 601 and a processor 602;

[0393] The processor 602 is used to group the base stations based on a similarity algorithm of base station feature information when it is determined that the base stations within the jurisdiction meet the triggering conditions of the intelligent adaptive modulation and coding (AMC) algorithm based on federated learning.

[0394] Determine the federated learning model network element corresponding to each group of base stations;

[0395] The federated learning model network element is triggered to execute the lifecycle process, which is used to obtain the intelligent AMC algorithm model. All base stations in the same group share one intelligent AMC algorithm model.

[0396] Optionally, the transceiver is used to: receive performance information of the intelligent AMC algorithm model sent by the base station;

[0397] The processor is further configured to: evaluate the performance of the intelligent AMC algorithm model based on the performance information; and adjust the similarity threshold in the similarity algorithm based on the performance evaluation results.

[0398] Optionally, the processor determines that the base stations within its jurisdiction meet the triggering conditions of the federated learning-based intelligent adaptive modulation and coding (AMC) algorithm, including one of the following:

[0399] Receive first indication information sent by base stations within the jurisdiction, the first indication information being used to indicate that the base stations meet the triggering conditions of the intelligent AMC algorithm based on federated learning;

[0400] Receive network performance indicators and / or second indication information sent by base stations within the jurisdiction; determine, based on the network performance indicators and / or the second indication information, that the base station meets the triggering condition of the intelligent AMC algorithm based on federated learning; the second indication information is used to indicate that the base station meets the first condition.

[0401] Optionally, the triggering condition includes at least one of the following:

[0402] The network performance indicators of base stations within the jurisdiction do not meet the QoS requirements of users;

[0403] The base stations within the jurisdiction meet the first condition;

[0404] The first condition includes at least one of the following:

[0405] The base station's computing power is insufficient to support AI model training;

[0406] Base station power consumption cannot support AI model training;

[0407] The prediction accuracy of the base station AI model training is below the first threshold.

[0408] The convergence time of the base station AI model training does not meet the user's QoS requirements;

[0409] Federated learning offers a performance gain over non-federated learning.

[0410] Optionally, the processor groups the base stations based on a similarity algorithm of base station feature information, including:

[0411] Based on the feature information of each base station, calculate the feature information similarity between different base stations;

[0412] The base stations are grouped according to the similarity of the feature information and a preset similarity threshold.

[0413] Optionally, the feature information includes at least one of the following:

[0414] Base station hardware information;

[0415] Base station configuration information;

[0416] Base station user information;

[0417] Base station wireless channel environment information;

[0418] Input characteristics of the intelligent algorithm of the base station;

[0419] Output characteristics of the base station's intelligent algorithm;

[0420] The model expression of the intelligent algorithm of the base station.

[0421] Optionally, the processor determines the federated learning model network element corresponding to each group of base stations, including at least one of the following:

[0422] For a group of base stations, the local model training nodes of the group of base stations are determined based on the computing power resources of the base stations within the group and / or the similarity of feature information between the base stations.

[0423] For a group of base stations, a global model aggregation node is determined based on the computing power resources of the base stations in the group and / or the connection resources between the base stations in the group and the local model training node. The global model aggregation node is one of the base stations in the group or a network device.

[0424] Optionally, the local model training node satisfies a second condition, which includes at least one of the following:

[0425] The computing resources are sufficient to meet the needs of AI intelligent algorithms;

[0426] The characteristic information of the base station is the second threshold of all data information of the group of base stations.

[0427] Optionally, the global model aggregation node satisfies a third condition, which includes at least one of the following:

[0428] The computing resources are sufficient to meet the computing resource requirements when aggregating and / or updating model parameter information;

[0429] The connection resources can meet the latency requirements when aggregating and / or sending model parameter information;

[0430] To meet the synchronization requirements of sending different model parameter information to the base station.

[0431] Optionally, the processor triggers the lifecycle process of the federated learning model element, including:

[0432] A third instruction message is sent to the federated learning model network element, the third instruction message being used to instruct the federated learning model network element to execute a lifecycle process.

[0433] Optionally, the processor performs a performance evaluation of the intelligent AMC algorithm model based on the performance information, including:

[0434] The AI ​​model performance metrics of the intelligent AMC algorithm model are evaluated based on the performance information.

[0435] The performance metrics of the AI ​​model include at least one of the following:

[0436] The accuracy of model training;

[0437] Data distribution coverage;

[0438] Convergence time of model training;

[0439] Generalization;

[0440] robustness;

[0441] Energy consumption.

[0442] Optionally, the processor performs a performance evaluation of the intelligent AMC algorithm model based on the performance information, including:

[0443] The performance metrics of the intelligent AMC algorithm model are evaluated based on the performance information.

[0444] The performance metrics of the AMC algorithm include at least one of the following:

[0445] Throughput;

[0446] Communication reliability;

[0447] Delay;

[0448] IBLER.

[0449] It should be noted that the device in this embodiment corresponds to the method applied to network devices described above. The implementation methods in each of the above embodiments are also applicable to the embodiments of this device and can achieve the same technical effect. The parts that are the same as those in the method embodiments and their beneficial effects will not be described in detail here.

[0450] Please refer to Figure 7 This application embodiment also provides a federated learning model network element 700, including: a transceiver 701 and a processor 702;

[0451] The processor 702 is used to: execute a lifecycle process to obtain an intelligent AMC algorithm model;

[0452] The transceiver 701 is used to send the intelligent AMC algorithm model to a group of base stations corresponding to the federated learning model network element.

[0453] Optionally, the federated learning model network elements include: local model training nodes and global model aggregation nodes;

[0454] The processor executes a lifecycle process to obtain an intelligent AMC algorithm model, including:

[0455] The local model training node downloads the initial model file from the global model aggregation node;

[0456] The local model training node performs local model training based on local data and the initial model file to obtain updated model information;

[0457] The local model training node sends the updated model information to the global model aggregation node;

[0458] The global model aggregation node integrates the updated model information sent by all local model training nodes to obtain the intelligent AMC algorithm model.

[0459] Optionally, the transceiver is further configured to: receive third indication information sent by the network device, the third indication information being used to instruct the federated learning model network element to execute a lifecycle process.

[0460] It should be noted that the device in this embodiment corresponds to the method described above for application to network elements of a federated learning model. The implementation methods in the above embodiments are all applicable to the embodiments of this device and can achieve the same technical effect. The parts and beneficial effects that are the same as those in the method embodiments will not be described in detail here.

[0461] Please refer to Figure 8 This application embodiment also provides a network device 800, including a transceiver 810, a processor 800, a memory 820, and a program or instructions stored in the memory 820 and executable on the processor 800; when the processor 800 executes the program or instructions, it implements the steps of the adaptive modulation and coding method executed by the network device described above.

[0462] The transceiver 810 is used to receive and send data under the control of the processor 800.

[0463] Among them, Figure 8 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits together, represented by one or more processors (processor 800) and memory (memory 820). The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 810 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. The processor 800 is responsible for managing the bus architecture and general processing, and the memory 820 can store data used by the processor 800 during operation.

[0464] Please refer to Figure 9 This application embodiment also provides a federated learning model network element 900, including a transceiver 910, a processor 900, a memory 920, and a program or instructions stored in the memory 920 and executable on the processor 900; when the processor 900 executes the program or instructions, it implements the steps of the adaptive modulation and coding method executed by the federated learning model network element described above.

[0465] The transceiver 910 is used to receive and send data under the control of the processor 900.

[0466] Among them, Figure 9In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits together, represented by one or more processors (processor 900) and memory (memory 920). The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 910 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. The processor 900 is responsible for managing the bus architecture and general processing, and the memory 920 can store data used by the processor 900 during operation.

[0467] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described adaptive modulation and coding method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0468] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0469] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0470] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An adaptive modulation and coding method applied to a network device, characterized in that, The method comprises the following steps: The network device groups the base stations in the jurisdictional area based on a similarity algorithm of base station feature information, in a case where it is determined that the base stations in the jurisdictional area meet a trigger condition of a smart adaptive modulation and coding (AMC) algorithm based on federated learning; A federated learning model network element corresponding to each group of base stations is determined; The network device triggers the federated learning model network element to execute a life cycle process, which is used to obtain a smart AMC algorithm model, and all base stations in the same group share one smart AMC algorithm model.

2. The method of claim 1, wherein, The method further comprises the following steps: Receiving performance information of the smart AMC algorithm model sent by the base stations; Performing performance evaluation on the smart AMC algorithm model according to the performance information; Adjusting a similarity threshold in the similarity algorithm according to the performance evaluation result.

3. The method of claim 1, wherein, The determination that the base stations in the jurisdictional area meet the trigger condition of the smart AMC algorithm based on federated learning comprises one of the following: Receiving first indication information sent by the base stations in the jurisdictional area, the first indication information being used to indicate that the base stations meet the trigger condition of the smart AMC algorithm based on federated learning; Receiving network performance indicators and / or second indication information sent by the base stations in the jurisdictional area; and determining that the base stations meet the trigger condition of the smart AMC algorithm based on federated learning according to the network performance indicators and / or the second indication information; The second indication information is used to indicate that the base stations meet a first condition.

4. The method according to claim 1 or 3, characterized in that, The trigger condition comprises at least one of the following: The network performance indicators of the base stations in the jurisdictional area do not meet user QoS requirements; The base stations in the jurisdictional area meet the first condition; The first condition comprises at least one of the following: The base station computing power condition cannot support artificial intelligence (AI) model training; The base station power consumption cannot support AI model training; The base station AI model training prediction accuracy is lower than a first threshold value; The base station AI model training convergence time does not meet user quality of service (QoS) requirements; The use of federated learning has performance gain compared with non-federated learning.

5. The method of claim 1, wherein, The grouping of the base stations based on the similarity algorithm of base station feature information comprises the following steps: Calculating feature information similarity between different base stations according to the feature information of each base station; Grouping the base stations according to the feature information similarity and a preset similarity threshold value.

6. The method of claim 5, wherein, The feature information comprises at least one of the following: Hardware information of the base station; Configuration information of the base station; User information of the base station; Wireless channel environment information of the base station; Input features of the smart algorithm of the base station; Output features of the smart algorithm of the base station; Model expression of the smart algorithm of the base station.

7. The method of claim 1, wherein, The determination of the federated learning model network element corresponding to each group of base stations comprises at least one of the following: For a group of base stations, a local model training node of the group of base stations is determined according to the computing power resources of the base stations in the group and / or the feature information similarity between the base stations; For a group of base stations, a global model aggregation node of the group of base stations is determined according to the computing power resources of the base stations in the group and / or the connection resources between the base stations in the group and the local model training node, the global model aggregation node being one of the base stations in the group or a network device.

8. The method of claim 7, wherein, The local model training node satisfies a second condition, and the second condition includes at least one of the following: The computing resource can meet the demand of the AI intelligent algorithm; The characteristic information of the base station is a second threshold of all data information of the group of base stations.

9. The method of claim 7, wherein, The global model aggregation node satisfies a third condition, and the third condition includes at least one of the following: The computing resource can meet the demand of the computing resource when aggregating and / or updating the model parameter information; The connection resource can meet the demand of the time delay when aggregating and / or sending the model parameter information; Satisfy the synchronization requirement of different model parameter information sent to the base station.

10. The method of claim 1, wherein, The trigger includes: Sending third indication information to the federated learning model network element, the third indication information being used to instruct the federated learning model network element to execute the life cycle process.

11. The method of claim 2, wherein, The performance evaluation includes: According to the performance information, evaluating the AI model performance index of the intelligent AMC algorithm model; The AI model performance index includes at least one of the following: Accuracy of model training; Data distribution coverage; Convergence time of model training; Generalization; Robustness; Energy consumption.

12. The method of claim 2, wherein, The performance evaluation includes: According to the performance information, evaluating the AMC algorithm performance index of the intelligent AMC algorithm model; The AMC algorithm performance index includes at least one of the following: Throughput; Communication reliability; Latency; Initial transmission block error rate IBLER.

13. An adaptive modulation and coding method applied to a federated learning model network element, characterized in that, The federated learning model network element executes the life cycle process to obtain the intelligent AMC algorithm model, and sends the intelligent AMC algorithm model to a group of base stations corresponding to the federated learning model network element. The federated learning model network element includes a local model training node and a global model aggregation node. The execution of the life cycle process to obtain the intelligent AMC algorithm model includes:

14. The method of claim 13, wherein, The local model training node downloads an initial model file from the global model aggregation node; The local model training node performs local model training according to local data and the initial model file to obtain updated model information; The local model training node sends the updated model information to the global model aggregation node; The global model aggregation node integrates the updated model information sent by all local model training nodes to obtain the intelligent AMC algorithm model. The method further includes: Receiving third indication information sent by a network device, the third indication information being used to instruct the federated learning model network element to execute the life cycle process.

15. The method of claim 13, wherein, The transceiver and the processor are included, and the processor is configured to: Group the base stations based on a similarity algorithm of base station characteristic information when it is determined that the base stations in the jurisdictional area meet the trigger condition of the federated learning-based intelligent AMC algorithm; 16. A network device, comprising: Determine the federated learning model network element corresponding to each group of base stations; ​ ​ trigger the federated learning model network element to execute a life cycle process, the life cycle process being used to obtain an intelligent AMC algorithm model, and all base stations in the same group sharing one intelligent AMC algorithm model.

17. A federated learning model network element, comprising: comprising a transceiver and a processor, wherein the processor is configured to execute a life cycle process to obtain an intelligent AMC algorithm model; the transceiver is configured to send the intelligent AMC algorithm model to a group of base stations corresponding to the federated learning model network element.

18. An adaptive modulation and coding apparatus, characterized by comprising: comprising: a grouping module configured to group base stations in a jurisdictional area based on a similarity algorithm of base station feature information, if the base stations meet a triggering condition of a federated learning based intelligent adaptive modulation and coding (AMC) algorithm; a first determination module configured to determine a federated learning model network element corresponding to each group of base stations; a first processing module configured to trigger the federated learning model network element to execute a life cycle process, the life cycle process being used to obtain an intelligent AMC algorithm model, and all base stations in the same group sharing one intelligent AMC algorithm model.

19. An adaptive modulation and coding apparatus, characterized by comprising: comprising: a second processing module configured to execute a life cycle process to obtain an intelligent AMC algorithm model; a first sending module configured to send the intelligent AMC algorithm model to a group of base stations corresponding to the federated learning model network element.

20. A network device, comprising: comprising: a processor, a memory, and a program stored in the memory and executable on the processor, the program, when executed by the processor, implements the steps of the method of any one of claims 1 to 12.

21. A federated learning model network element, comprising: comprising: a processor, a memory, and a program stored in the memory and executable on the processor, the program, when executed by the processor, implements the steps of the method of any one of claims 13 to 15.

22. A computer-readable storage medium, characterized in that, The computer program is stored in the computer readable storage medium, and the computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 12, or implements the steps of the method of any one of claims 13 to 15.

Images