Electronic equipment, methods, storage media, devices, and computer programs used in wireless communication systems

Abstract

The present disclosure relates to an electronic device, a method, and a storage medium used in a wireless communication system. Various embodiments using distributed machine learning in a wireless communication system have been described. In one embodiment, a user equipment transmits quantity information of training samples used in a current training period of a local model to a control entity, receives uplink resource information for uploading parameters of the local model, and the uplink resource indicated by the uplink resource information is allocated by the control entity based on the quantity information received from multiple user equipments so that the user equipment with a larger quantity indicated by the quantity information has a higher chance of being allocated sufficient uplink resources for uploading parameters of the local model, and uploads parameters of the local model to the control entity via the uplink resource indicated by the uplink resource information so that the control entity obtains a next global model.

Description

[Technical Field]

[0001] (Cross-reference of related applications) This application claims priority to the Chinese patent application filed on 5 July 2021, application number 202110756858.0, which is incorporated herein by reference as described in its entirety.

[0002] This disclosure relates to the field of wireless communication, more specifically to technologies employed when using distributed machine learning, and further to technologies used for data transmission. [Background technology]

[0003] As wireless networks and artificial intelligence develop, networks tend to become increasingly intelligent. Whether it's the rapidly developing 5G technology of today or the next generation of wireless network technologies of the future, the intelligence of wireless networks will be one of the important directions of their development. Specifically, federated learning (FL) has unique advantages in ensuring data privacy, security, and legal compliance, and can improve the effectiveness of machine learning models by collaboratively modeling across multiple devices. Therefore, it has become a very important distributed artificial intelligence framework or distributed machine learning framework, and its combination with wireless networks will be one of the main contents of future intelligent wireless network applications. Thus, how to design FL to effectively integrate with wireless network technology will have a significant impact on future artificial intelligence applications.

[0004] Due to the characteristics of FL applications, FL places higher demands on the QoS of wireless networks. For example, FL data traffic is latency-sensitive. Therefore, data traffic generated for FL needs to be transmitted using Ultra-Reliable Low-Latency Communications (URLLC) scenarios. Specifically, the transmission of local and global model parameters in FL requires URLLC support. However, providing URLLC data transmission services to a large number of user devices simultaneously places a significant burden on the wireless network and greatly increases communication costs. Therefore, to enable FL to operate successfully in wireless networks, appropriate supplements must be made to current wireless network standards so that the wireless network can meet the service requirements for model parameter transmission in federated learning. More preferably, FL should be able to operate in wireless networks at low cost and with high efficiency, avoiding excessive burden on the wireless network by FL applications and minimizing errors in trained global models.

[0005] Due to the finite resources in wireless networks, it may be insufficient for all user devices participating in distributed machine learning, such as FL, to upload local model parameters. The loss of local model parameters can increase the error in the aggregated global model, potentially making it difficult to achieve satisfactory distributed machine learning results.

[0006] Therefore, it is desirable to provide a method that can reduce the error of global models when applying distributed machine learning in wireless networks.

[0007] Meanwhile, 5G communication is attracting attention due to its superior data transmission performance, and research is becoming increasingly active. URLLc, one of the core application scenarios for 5G communication, has extremely stringent performance requirements for reliability and latency. Specifically, URLL is required to achieve 99.999% reliability and a latency of 1 ms. Furthermore, as wireless networks undergo generational changes and upgrades, many new applications will emerge, and these applications may have even stricter requirements for latency and reliability. For example, factory automation requires reliability higher than 79s (99.99999%) and latency of less than 1 ms. Next-generation wireless communication (B5G or 6G) is expected to have even higher and stricter requirements for latency and reliability, with reliability as high as 99s and latency of less than 0.1 ms.

[0008] In communications with high latency and reliability requirements, such as URLLC, the following problems arise. Firstly, while many currently developed and used technologies, such as short packet transmission, permissionless transmission, short time interval transmission, and time, frequency, and spatial diversity, contribute to improved reliability and reduced latency, current technologies often require the consumption of numerous resources to respond to abnormal situations, such as rapid changes in channel conditions, resulting in a lack of adaptability to wireless transmission environments and inflexibility in scheduling. Secondly, existing networks require the consumption of numerous resources for channel detection and estimation to achieve low-latency, high-reliability communications like URLLC, leading to significant overhead and making it difficult to provide more accurate channel information in the event of abnormal situations. Thirdly, real-time changes in wireless channels cause the channel sampling information referenced by base stations when scheduling to become outdated and distorted, completely or partially losing the scheduling value for the system and making it difficult to ensure timely and effective transmission of information.

[0009] Therefore, it is desirable to provide a method for improving downlink data transmission performance by enabling base stations to flexibly transmit downlink data with more appropriate resources according to the current channel state. [Overview of the project] [Means for solving the problem]

[0010] One aspect of this disclosure relates to electronic equipment used on the user equipment side in a wireless communication system. According to one embodiment, the electronic equipment may include a processing circuit system. The processing circuit system may transmit quantity information of training samples that the user equipment will use for the current training period of a local model to a control entity, receive uplink resource information for uploading parameters of the local model, and be allocated by the control entity based on quantity information from multiple user equipment such that the larger the quantity indicated by the quantity information for a user equipment, the greater the chance that sufficient uplink resources will be allocated to upload parameters of the local model, and upload parameters of the local model to the control entity via the uplink resources indicated by the uplink resource information so that the control entity obtains the next global model.

[0011] One aspect of the present disclosure relates to an electronic device used on the network device side in a wireless communication system. According to one embodiment, the electronic device includes a processing circuit system. The processing circuit system receives quantity information of training samples used by a user device during the current training period of the local model from the user device, transmits uplink resource information for uploading parameters of the local model, and the uplink resource indicated by the uplink resource information is allocated by the processing circuit system based on quantity information from a plurality of user devices such that the greater the quantity indicated by the quantity information of the user device, the greater the chance of allocating sufficient uplink resources for uploading the parameters of the local model, and is configured to receive parameters of the local model uploaded by the user device via the uplink resource indicated by the uplink resource information so as to obtain the next global model.

[0012] Another aspect of the present disclosure relates to a method used in a wireless communication system. In one embodiment, the method includes transmitting, by a user device, quantity information of training samples used during the current training period of a local model to a control entity, receiving uplink resource information for uploading parameters of the local model, where the uplink resource indicated by the uplink resource information is allocated by the control entity based on quantity information from a plurality of user devices such that the greater the quantity indicated by the quantity information of the user device, the greater the chance of allocating sufficient uplink resources for uploading the parameters of the local model, and uploading, by the control entity, the parameters of the local model to the control entity via the uplink resource indicated by the uplink resource information so as to obtain the next global model.

[0013] Another aspect of the present disclosure relates to a method used in a wireless communication system. In one embodiment, the method includes receiving, from a user equipment, quantity information of training samples that the user equipment uses during a current training period of a local model; transmitting uplink resource information for uploading parameters of the local model, wherein the uplink resource indicated by the uplink resource information is allocated based on quantity information from a plurality of user equipments such that a user equipment with a larger quantity indicated by the quantity information has an increased chance of being allocated sufficient uplink resources for uploading the parameters of the local model; and receiving, by the user equipment, parameters of the local model uploaded via the uplink resource indicated by the uplink resource information to obtain a next global model.

[0014] Yet another aspect of the present disclosure relates to a computer-readable storage medium storing one or more instructions. In some embodiments, when the one or more instructions are executed by one or more processors of an electronic device, the methods according to various embodiments of the present disclosure can be executed by the electronic device.

[0015] Yet another aspect of the present disclosure relates to various apparatuses including components or units that perform operations of the methods according to embodiments of the present disclosure.

[0016] The above summary is provided to summarize some exemplary embodiments in order to provide a basic understanding of various aspects of the subject matter described herein. Accordingly, the above features are merely examples and should not be construed as limiting the scope or gist of the subject matter described herein in any way. Other features, aspects, and advantages of the subject matter described herein will become apparent from the detailed description of the invention, which will be described in conjunction with the following drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] A better understanding of the contents of this disclosure can be obtained by considering the following specific description of embodiments in conjunction with the drawings. In all drawings, identical or similar parts are indicated by the same or similar reference numerals. All drawings, together with the following specific description, are included herein and constitute part of the specification, illustrating embodiments of this disclosure by example and interpreting the principles and merits of this disclosure. In the drawings,

[0018] [Figure 1] Figure 1 is a schematic diagram of a scenario using distributed machine learning in a wireless network according to an embodiment of the present disclosure. [Figure 2] Figure 2 is a schematic diagram of a system architecture using distributed machine learning in a wireless network according to an embodiment of this disclosure. [Figure 3] Figure 3 is a flowchart illustrating a method for applying distributed machine learning in a wireless network according to an embodiment of this disclosure. [Figure 4] Figure 4 illustrates the process of each iteration of distributed machine learning according to an embodiment of the present disclosure, and further shows the eMBB resources that the base station as a control entity may temporarily occupy. [Figure 5] Figure 5 shows the interaction process between a control entity and a user device according to an embodiment of the present disclosure. [Figure 6] Figure 6 shows the 5QI values ​​that can be used for transmitting model parameters of distributed machine learning according to the embodiments of this disclosure. [Figure 7] Figure 7 is a schematic diagram of a system architecture in which a base station transmits data to user equipment according to an embodiment of the present disclosure. [Figure 8] Figure 8 is a flowchart of the method used for data transmission according to the embodiment of this disclosure. [Figure 9] Figure 9 is a schematic diagram showing the training of a machine learning model according to an embodiment of this disclosure. [Figure 10] Figure 10 is a schematic diagram showing how to train a machine learning model in the first prediction mode according to an embodiment of the present disclosure. [Figure 11A]Figure 11A is a schematic diagram of an architecture for transmitting the predicted channel gain according to an embodiment of the present disclosure. [Figure 11B] Figure 11B is a sequence diagram showing the transmission of the channel gain predicted using the architecture in Figure 11A according to an embodiment of the present disclosure. [Figure 12A] Figure 12A is a flowchart of the communication process in which a base station transmits data to user equipment according to an embodiment of the present disclosure. [Figure 12B] Figure 12B is a flowchart of the communication process in which a user device receives data from a base station according to an embodiment of the present disclosure. [Figure 13] Figure 13 is a sequence diagram corresponding to the communication process shown in Figures 12A and 12B according to an embodiment of this disclosure. [Figure 14A] Figure 14A is a flowchart of another communication process in which a base station transmits data to user equipment according to an embodiment of the present disclosure. [Figure 14B] Figure 14B is a flowchart of another communication process in which user equipment receives data from a base station according to an embodiment of the present disclosure. [Figure 15] Figure 15 is a sequence diagram corresponding to the communication process shown in Figures 14A and 14B according to an embodiment of the present disclosure. [Figure 16] Figure 16 is a graph obtained by performing a simulation using the method shown in Figure 8 according to an embodiment of this disclosure. [Figure 17] Figure 17 is a block diagram showing an exemplary configuration of a personal computer, which is an information processing device that can be used in the embodiments of this disclosure. [Figure 18] Figure 18 is a block diagram showing a first example of an exemplary configuration of a gNB to which the technology of this disclosure can be applied. [Figure 19] Figure 19 is a block diagram showing a second example of an exemplary configuration of a gNB to which the technology of this disclosure can be applied. [Figure 20] Figure 20 is a block diagram showing an exemplary arrangement of a smartphone to which the technology of this disclosure can be applied. [Figure 21]Figure 21 is a block diagram showing an exemplary arrangement of a car navigation system to which the technology of this disclosure can be applied.

[0019] The embodiments described herein are subject to modification and alternative forms, the specific embodiments of which are illustrated in the accompanying drawings and described in detail herein. However, it should be understood that the accompanying drawings and their detailed descriptions do not limit the embodiments to any particular form of this disclosure, but rather include all modifications, equivalents, and alternative forms that fall within the essence and scope of the claims. [Modes for carrying out the invention]

[0020] The following describes typical applications in various embodiments of the apparatus and methods described herein. These examples are provided solely to add context and aid in understanding the embodiments described. It will therefore be apparent to those skilled in the art that the embodiments described below can be carried out without some or all of the specific details. In other cases, well-known process steps are not described in detail to avoid unnecessarily obscuring the embodiments described. Other applications are possible, and the technical applications of this disclosure are not limited to these examples.

[0021] First, with reference to Figure 1, a schematic diagram of a scenario using distributed machine learning in a wireless network 100 according to an embodiment of this disclosure will be described.

[0022] As shown in Figure 1, the wireless network 100 includes multiple user devices (UEs) 120-1 to 120-n and a base station (BS) 110. The wireless network 100 may be an existing network such as LTE (Long-Term Evolution), LTE-A, or NR (New Radio), or it may be a new wireless network that will emerge in the future. The base station 110 is connected to the core network and may be various 5G base stations such as Node B, eNode B, or HeNB, or it may be other access devices that provide wireless access services to user devices. Each of the user devices 120-1 to 120-n may be various terminal devices such as mobile phones, smartphones, tablet PCs, laptop PCs, notebook computers, digital televisions, or servers. By using the user devices, users can receive a wide range of network services. In the example in Figure 1, only three user devices 120-1, 120-2, and 120-3 are shown, but as a person skilled in the art will understand, the number of user devices communicating with a single base station may be more or less. For the sake of explanation, when it is not necessary to distinguish between different user devices, we will refer to the user device as user device 120 below.

[0023] The use of distributed machine learning in the wireless network 100 allows user devices 120-1 to 120-3 to jointly form a distributed machine learning group. Each user device 120 in the group may have one local model. The base station 110 may have one global model. User devices 120 can train their local models by collecting their local data and obtain the parameters of their local models. For example, user device 120-1 collects its local data to obtain data samples used for training.

number

number

number

[0024] These data samples can be input into an existing program (e.g., PyCharm) for training models on each user's device to fit the model parameters. Such models may be various machine learning models or artificial intelligence (AI) models, and may be composed of various existing neural networks, etc., according to the user's requirements. More broadly, any mathematical model that can be represented by parameters can be used as the local model and the global model. In the distributed machine learning employed in the embodiments of this disclosure, the specific configuration and achievable functions of the unified local model and global model are not particularly limited.

[0025] User equipment 120 may upload the local model trained in the current iteration to base station 110. Uploading a local model means uploading the parameters of the local model (hereinafter, model upload and model parameter upload have the same meaning), for example, uploading the parameters of each hierarchical node of the model. User equipment may use URLLC traffic resources to upload model parameters to meet latency and reliability needs. Based on the model parameters uploaded by each user device, base station 110 can aggregate to obtain a global model and download the parameters of the global model to each user device and transmit them, for example, via URLLC traffic resources. Furthermore, user equipment may also transmit other data traffic with the base station; for example, user equipment may transfer large files with the base station via enhanced mobile broadband (eMBB) traffic resources.

[0026] The distributed machine learning used in the wireless network 100 may be federated learning. As those skilled in the art can imagine, various machine learning methods may be implemented locally and centrally on the equipment, or they may be completed collaboratively by multiple nodes in a distributed manner. Therefore, all scenarios for performing distributed machine learning in a wireless network can apply the embodiments of this disclosure, and the embodiments of this disclosure do not particularly limit the specific type of distributed machine learning, as long as the user equipment can upload local model parameters to the base station and the base station can transmit global model parameters to the user equipment. For the sake of explanation, the technical proposal of this disclosure will be described below using the case where federated learning (FL) is distributed machine learning as an example. As those skilled in the art will understand, the embodiments of this disclosure are still applicable when other distributed machine learning methods are used.

[0027] When federated learning is applied in the wireless network 100, the interactions between each communication entity may be divided into four specific types, V1 to V4, as shown in Figure 2. It should be noted that Figure 2 includes user devices UE A, UE B, UE C and the base station, as well as a federation learning controller (FLCF) as a functional entity. The FLCF receives and aggregates local model parameters from each user device to obtain a global model, and transmits the global model parameters to each user device via the base station. Furthermore, the FLCF also receives state information and available radio resource information from the base station to determine how to allocate uplink resources for model uploading to different user devices. The location of the FLCF determines who the control entity is in federated learning. For example, the FLCF may be located within the base station, within a separate server that can communicate with the base station, or within a network element in the core network that can communicate with the base station. Accordingly, the control entities are the base station, the server, and the core network network element, respectively. In Figure 1, the FLCF is located within the base station, so the base station is the control entity for federation learning. If the FLCF is located in another device, the base station in Figure 1 should be replaced with the corresponding device.

[0028] Continuing to refer to Figure 2, UE A, UE B, and UE C each contain an FL application, and FLCF contains the FL application server. UE A, UE B, and UE C interface with the base station via the LTE-UU air interface. V1 between UE A, UE B, and UE C represents the exchange of user-side edge information. V2 between the FL application and the FL application server represents the request, delivery, upload, and download of models in the federated learning framework. If FLCF is not located at the base station, the information transmission of the processes included in V2 must be forwarded via the base station. V2 between UE A, UE B, and UE C and the base station represents the transmission of user device status information to the base station, such as the number of samples the user device uses to train the local model, the distance between the user device and the base station, etc. V3 between the base station and the FLCF indicates that the base station provides the FLCF with information necessary for uplink resource allocation, such as status information received in V2, and radio resources available during the local model upload phase (e.g., a finite given uplink resource dedicated to model upload and uplink resources that may be temporarily occupied). V3 may also indicate that the FLCF provides the base station with a QoS request for federation learning (e.g., 5QI values ​​specific to the transmission of parameters for federation learning) and sends an uplink resource allocation command to the base station for model upload, causing the base station to allocate appropriate uplink resources to each user device based on the said uplink resource allocation command to perform the model upload.

[0029] Due to the finite nature of wireless resources, dedicated wireless resources for federated learning cannot upload all local models, potentially leading to larger errors in the global model. The embodiments of this disclosure assume that, considering that local models on different user devices may have different levels of importance to the global model, prioritizing the uploading of more important local models can reduce errors in the global model. Based on this idea, V2 and V3 described above are introduced. V2 and V3 will be explained in detail below with reference to the flowchart of Method 300 shown in Figure 3. By implementing Method 300, uplink resources for uploading local model parameters can be rationally allocated to different user devices, thereby reducing errors in the global model.

[0030] In S310, the user device transmits information to the control entity regarding the quantity of training samples that the user device will use for the current training period of the local model.

[0031] In distributed machine learning frameworks such as federated learning, it is necessary to continuously update the parameters of the local model over multiple iterations to better match the actual situation of the current training samples and achieve higher accuracy. In each iteration, the user device trains its local model based on the collected training samples. During the training process, the user device determines the total quantity of training samples to be used for the training and feeds this quantity back to the control entity via messages (e.g., signaling).

[0032] The quantity of training samples reflects, to some extent, the importance of the local model on the user's device. The larger the number of training samples used by the user's device to train the local model, the more data information the model's parameters can reflect, resulting in higher confidence, more accurate predictions, and therefore, a more important model.

[0033] The control entity may be the network equipment where the FLCF in Figure 2 is located (e.g., base stations, servers, and core network elements), and is used to obtain a global model from the local model parameters at each user device in a distributed machine learning group (e.g., a federated learning group), and to send the parameters of the global model to the user devices. The control entity is also used to determine the allocation of uplink resources used for model uploading for each user device in the group.

[0034] In S320, user devices receive uplink resource information for uploading local model parameters. Here, the uplink resources indicated by the uplink resource information are allocated by the control entity based on quantity information from multiple user devices, such that user devices with larger quantities indicated by quantity information have a greater chance of being allocated sufficient uplink resources to upload local model parameters.

[0035] Each user device within the same federation learning group must provide the control entity with information on the quantity of training samples it will use to train its local model in the current iteration. The control entity can then determine the relative sizes of the training samples used by each user device. Generally, the more training samples a user device uses, the more accurate (or important) the trained local model will be, as it will contain more data information. Therefore, the control entity can ensure that more important local models can be uploaded by allocating sufficient uplink resources to user devices with larger sample quantities to upload their local model parameters. In particular, when wireless resources are limited, prioritizing the uploading of more important local models plays a crucial role in reducing errors in the global model. The specific uplink resource allocation method will be described later.

[0036] After the control entity allocates an uplink resource, information about the allocated uplink resource may be notified to the user device. This allows the user device to transmit model parameters over the allocated uplink resource.

[0037] In S330, the user device uploads the parameters of the local model to the control entity via the uplink resource indicated by the uplink resource information, so that the control entity obtains the next global model.

[0038] The user device sends the parameters of its local model to the control entity using its assigned uplink resource. The control entity can obtain a global model based on the parameters uploaded by different user devices, following existing techniques (e.g., aggregation techniques in federated learning). After obtaining the global model, the control entity may send the parameters of the global model to the user device. This allows the user device to update its local model and continue training the updated local model with the new training data in the next iteration.

[0039] The uplink resources indicated by the uplink resource information may be uplink resources for URLLC traffic. This ensures that parameter uploads have low latency and high reliability, contributing to a reduction in global model error. However, in wireless systems, wireless resources are often limited, so dedicated uplink resources for model uploads (e.g., URLLC resources) may be insufficient to allow all user devices to upload. In this case, it is conceivable to temporarily occupy some resources for model uploads. For example, uplink resources for non-latency sensitive traffic that all user devices in a federated learning group are using during the model upload phase, such as eMBB traffic, may be temporarily occupied. In this case, the occupied uplink resources are released by the original traffic and allocated to user devices lacking uplink resources to upload model parameters. This ensures that more model parameters can be uploaded to the control entity and aggregated, further reducing global model error. During the upload phase, by using a given finite number of URLLC traffic uplink resources (which are generally fewer than the number of user devices), and by using eMBB traffic uplink resources available for a small random allocation, it is possible to schedule uplink resources more appropriately and meet the needs of model uploads.

[0040] Depending on the location of the RLCF in Figure 2, S310 to S330 connect to different devices. For example, if the FLCF in Figure 2 is located inside the base station, the control entity is the base station. In this case, S310 and S330 connect to communication between the user equipment and the base station, and the uplink resource information in S320 is determined directly by the base station and transmitted to the user equipment. Alternatively, if the FLCF in Figure 2 may be located inside network equipment outside the base station (e.g., a server or core network element), the control entity is the server or core network element. In this case, the quantity information transmitted by the user equipment in S310 and the model parameters transmitted by the user equipment in S330 need to be transferred to the control entity via the base station, and the uplink resource information received by the user equipment in S320 needs to be first transmitted to the base station by the control entity and then transferred to the user equipment by the base station. The specific form of the uplink resource information transmitted by the control entity to the base station and the uplink resource information transferred by the base station to the user equipment may differ, but both refer to the same uplink resource. Since both refer to the same uplink resource, they can be considered substantially identical. Preferably, the FLCF resides within the base station, and the base station is the control entity. In this way, the delay in the exchange of model parameters can be reduced, and the system can be made more stable and reliable. Furthermore, preferably, the distributed machine learning used in the wireless network is federated learning.

[0041] According to the proposed technology, by transmitting training sample quantity information used to train local models to a control entity (e.g., a base station), the control entity can determine the importance of local models in different user devices based on the sample quantities reported by those devices. This increases the chances of more important local models being uploaded to the control entity using more training samples, allowing the global model aggregate to be based on more important local models and reducing errors in the global model.

[0042] According to embodiments of this disclosure, in addition to transmitting quantitative information to the control entity, user equipment may also transmit distance information between itself and the base station to the control entity. This distance information may reflect the physical distance between the user equipment and the base station, such as the straight-line distance of signal transmission. The reason for uploading distance information is to optimize the energy consumption of user equipment in a distributed machine learning group by allowing the control entity to prioritize user equipment closer to the base station when allocating uplink resources.

[0043] User devices require battery power to operate, and battery capacity is limited. To enable user devices to have a longer lifespan, a total power constraint may be set for all user devices in a federated learning group to ensure that the total energy consumed by individual user devices to transmit model parameters does not become excessive during the federated learning process. The total power constraint may be determined based on an average power constraint. The average power refers to the average value of the data transmission power of all user devices participating in federated learning, and may be determined by the base station based on the signal strength received from the user devices. The power constraint may be provided to the control entity so that it takes it into consideration when allocating uplink resources.

[0044] If a control entity knows not only the number of samples a user device uses to train its local model, but also the distance between the user device and the base station, the control entity can ensure that user devices with larger sample quantities and shorter distances have a greater chance of being allocated sufficient uplink resources to upload their local model parameters. A larger number of training samples used means that the corresponding local model is more important. A shorter distance from the base station means that the user device consumes less energy to upload the local model, which is advantageous in terms of energy consumption. In this way, not only can more important local models be ensured to be uploaded, but local models that are both more important and consume less energy to upload are given priority for uploading.

[0045] Figure 4 shows the process of each iteration of distributed machine learning according to an embodiment of the present disclosure, in which, in addition to the uplink resource dedicated to model uploading, the resource that may be temporarily occupied is the eMBB uplink resource in the example of Figure 4.

[0046] As shown in Figure 4, the nth iteration and the (n+1)th iteration have the same iteration process, but the eMBB resources that may be temporarily occupied in each iteration are different. Each iteration consists of three stages. Stage 1 is the local model training stage for the user device, Stage 2 is the state information upload stage for the user device, and Stage 3 is the resource allocation, local model upload, and global model transmission stage for the control entity. In the example in Figure 4, the control entity is a base station, and the distributed machine learning is federated learning.

[0047] The federated learning group includes user device 1, user device 2, ..., user device U, with a total of U user devices. Each user device has a local model that is trained in stage 1 (e.g., A1-A2, B1-B2) according to the training samples that the user device has collected locally. In stage 2 (e.g., A2-A3, B2-B3), each user device i uses a sample quantity K to train its respective model. i The status information, including (n), is sent to the base station. The status information is the distance d between the user device i and the base station. i (n) may be further included. When the base station receives status information for all user devices in the group, the iteration process proceeds to stage 3 (e.g., A3 to B1, B3 to the point when the next iteration begins training the local model). At the point when stage 3 begins (e.g., A3 and B3), the base station may determine the eMBB traffic being conducted by it and all user devices in the group. The uplink resources of this eMBB traffic are the resources that the current iteration may occupy to upload local model parameters.

[0048] In Figure 4, at A3, two eMBB traffics appear, and the uplink resources of these two eMBB traffics are resources that may be temporarily occupied to transmit local model parameters. At B3, one eMBB traffic appears, and the uplink resources of this eMBB traffic are resources that may be temporarily occupied to transmit local model parameters. The uplink resources of temporarily occupied eMBB traffics may be allocated to any user device in the group that needs to upload a local model, regardless of which user device the eMBB traffic resources originate from.

[0049] FIG. 5 shows an interaction process 500 between a control entity and user equipment according to an embodiment of the present disclosure. In FIG. 5, the case where the control entity is the base station 505 is described as an example. As can be understood by those skilled in the art, when the control entity is other network equipment, a similar process may be performed.

[0050] As shown in FIG. 5, the time length from the start to the end of the nth iteration of distributed machine learning is T0, and the end time of the nth iteration is the time when the next (n + 1)th iteration starts. When the nth iteration of distributed machine learning starts, the user equipment i in the distributed machine learning group starts training a local machine learning model (abbreviated as local model) with the training samples collected locally. In S510, the user equipment i transmits the quantity K i (n) of the training samples used for training the local model and the distance d i (n) between the user equipment i and the base station 505 to the base station 505, for example, via a physical uplink shared channel (PUSCH). K i (n) and d i (n) may be transmitted simultaneously or separately.

[0051] In S520, the base station 505 determines the uplink resource X i (n) to be allocated to the user equipment i based on K i (n) and d i (n) received from all user equipment in the group. The specific resource allocation method will be described later. The uplink resource X i (n) may be a URLLC resource dedicated to model upload, or the base station uses K of all user equipment in the group i (n) and d iWhen (n) is received, the resources of non-delay-sensitive traffic between these user devices and the base station 505 may be, for example, eMBB resources. When the resources of non-delay-sensitive traffic need to be occupied to transmit the model parameters, those resources are fully released by the base station 505 and the corresponding user devices and used entirely to transmit the model parameters.

[0052] In S530, base station 505 has resource allocation result X i (n) is notified to user device i.

[0053] In S540, user device i is X i The parameter ω of the local model is the uplink resource indicated by (n). i Upload (n) to base station 505.

[0054] In S550, base station 505 aggregates the local models from user devices in the received group to obtain a global machine learning model (abbreviated as global model), and the parameters of the global model g i (n) is sent to user device i. As a result, user device i will g i Update the local model using (n).

[0055] Next, user device i starts the (n+1)th iteration to train the updated local model, and in S560, the quantity K of training samples used to train the local model is determined. i (n+1) and the distance d between user device i and base station 505. i (n+1) is transmitted to base station 505. When user device i is stationary, distance d i(n+1) may not be transmitted. Subsequently, as in S520-S550, base station 505 continues to allocate uplink resources based on the new status information and notifies user device i of the resource allocation result. User device i uploads local model parameters, and base station 505 transmits global model parameters. This completes a new iteration. The iteration process may be carried out continuously. This allows the machine learning model in each user device to match with more training samples and become more accurate.

[0056] The following describes the process by which a control entity allocates uplink resources based on state information, including quantity and distance information, transmitted by user devices, using federated learning as an example. The uplink resources used in federated learning include not only URLLC traffic resources dedicated to model uploading, but also eMBB resources that may be temporarily occupied.

[0057] When a control entity receives information from each user device in a federated learning group regarding the quantity of training samples used to train the local model in the current iteration, and the distance between the user device and the base station, the control entity can dynamically allocate uplink resources in real time so that the local model can be uploaded. Specifically, in one embodiment, the control entity may determine the importance of the local model that the user device is training, taking into account the sample quantity information that the user device uses to train the local model. In another embodiment, the control entity may consider the distance between the user device and the base station to satisfy the average power limit. In yet another embodiment, the control entity may determine resource finiteness, taking into account that the quantity of radio resources dedicated to federated learning is less than the quantity of user devices in the group. Based on the model importance, power limit, and / or radio resource limit, the control entity can determine the allocation of radio resources among user devices to achieve the requirement of minimizing errors in the global model.

[0058] To optimize energy consumption costs, wireless resource costs, and machine learning effects as simultaneously as possible, the inventors designed the following factors to consider when allocating uplink resources:

number

number

number

number

[0059] In equation (1), the factors to be considered in the uplink resource allocation can be divided into three parts. The first part is E(F(g(n+1))-F(g *)) represents the expected global model error after the nth iteration aggregate, corresponding to the machine learning effect or artificial intelligence (AI) training effect. Part 2

number

number

number

number

number

number

number

[0060] The above equation means searching for A(n) and L(n) such that equation (2) is minimized when restricted to equations (3) to (6). Here, A(n) is the uplink resource allocation for eMBB traffic occupied by user devices in the federated learning group.

number

[0061] The above optimization problem is a multi-objective optimization problem, and since finding the solution is complex, it is desirable to convert the above multi-objective optimization problem into a single-objective optimization problem. By minimizing the multi-objective linear weighting, the multi-objective optimization problem can be converted into a single-objective optimization problem. As shown in equations (7) to (9), the above optimization problem is converted as follows:

number

number

number

[0062] The difficulty in the above optimization problem lies in the first part of the objective function E(F(g(n+1))-F(g *The expression is as follows. The result of this expectation is generally not explicit, and its exact result depends on the specific AI model, making it difficult to determine by a precise formula. Fortunately, a non-patent document titled "A Joint Learning and Communications Framework for Federated Learning Over Wireless Networks," published in January 2021 in IEEE Transactions on Wireless Communications, Vol. 20, No. 1, pp. 269-283, by M. Chen, Z. Yang, W. Saad, C. Yin, HVPoor and S. Cui, provides an upper bound on this expectation, and this non-patent document is incorporated herein by reference. The aforementioned upper bound on this expectation is as follows:

number

number

number

[0063] Section 1

number

number

number

number

number

number

[0064] Equation (7) E(F(g(n+1))-F(g * Instead of using the first term above, we begin solving the optimization problem in equation (7). By setting the values ​​of η and μ, we can obtain A(n) and L(n) that minimize equation (7). If A(n) does not satisfy the constraint of equation (6) and / or L(n) does not satisfy the constraint of equation (3), we adjust the values ​​of η and μ to search again for A(n) and L(n) that minimize equation (7). If the obtained A(n) and L(n) both satisfy the constraints of equation (6) and equation (3), the obtained A(n) and L(n) correspond to the uplink resources that should be allocated to each user device. For example,

number

[0065] By adjusting η and μ, an optimal trade-off relationship between energy consumption, wireless resources, and AI model error can be achieved during the federated learning process, thereby dynamically adjusting energy consumption costs and wireless resource costs to obtain different levels of training effectiveness. Solving equation (7) above allows for the allocation of reasonable uplink resources for uploading models to each user device when wireless resources and energy consumption are limited. This ensures that uplink resources are preferentially allocated to more important local models and also to user devices closer to the base station.

[0066] After allocating an uplink resource to a user device, if the control entity is not a base station, the control device can transmit information to the user device via the base station to allocate the corresponding uplink resource. If the control entity is a base station, the base station allocates the corresponding uplink resource to the user device based on the allocation information obtained by solving equation (7). The user device transmits the parameters of the local model on the allocated uplink resource.

[0067] The transmission of local model parameters can satisfy specific quality of service (QoS) requirements. These QoS requirements may be specified by predetermined 5QI values. For example, the 5QI values ​​may have the form shown in Figure 6.

[0068] As shown in Figure 6, the 5QI value may be a value not currently in use, for example, 87. The QoS characteristics associated with the 5QI value may include at least one of the following: resource type, default priority level, packet delay budget (PDB), packet error rate (PER), default mean window, and default maximum data burst amount. In a scenario of model parameter transmission for distributed machine learning (e.g., federated learning), the resource type may be guaranteed bitrate (GBR), the default priority level may be 70, the packet delay budget may be 10ms, and the packet error rate may be 10 -6 The default average window may be 2000ms, and the default maximum data burst amount may not be defined. Defining such QoS requirements allows for good latency and reliability in the transmission of model parameters, thereby improving the accuracy of distributed machine learning.

[0069] These 5QI values ​​may be placed on user equipment, for example, by the user, core network equipment, or base station to ensure that user equipment adopts the corresponding QoS requests during model uploads.

[0070] Local model uploads must satisfy the QoS requirements defined by this 5QI value, and global model transmissions must also satisfy the QoS requirements defined by this 5QI value. For example, this 5QI value may be placed on a base station by core network equipment, administrators, or programs running on the base station so that the base station satisfies the corresponding QoS requirements when transmitting global model parameters.

[0071] The above describes the technologies used when utilizing distributed machine learning in wireless communication networks. Next, we will describe the technologies used when transmitting data in wireless communication networks.

[0072] First, with reference to Figure 7, a schematic diagram of the system architecture 700 in which a base station transmits data to user equipment according to an embodiment of this disclosure will be described.

[0073] As shown in Figure 7, the transmitting base station transmits downlink data to the receiving user equipment via a radio channel. The base station and user equipment in Figure 7 may be the same as those in Figure 1. For the sake of explanation, they are represented by the same symbols here.

[0074] To meet the latency requirements of a certain network service, data packets a[n] of that network service can arrive at the base station 110's transmit queue q[n] at a maximum of every Ta time, where Ta is the maximum delay each data packet can tolerate. The base station 110 may sample the channel state at every Th time. The channel sampling interval Th is much smaller than the allowable delay Ta of each data packet. When the base station 110 transmits a data packet from queue q[n], it transmits the data packet to the user equipment 120 via the radio channel with power corresponding to the channel quality, which is represented by a channel quality indicator (CQI) fed back by the user equipment 120.

[0075] In addition to receiving data packets from the base station 110, the user device 120 may predictively analyze future channel conditions based on historical channel conditions to obtain information that indicates the time slot with the best channel condition among several future time slots. By feeding back such information to the base station 110, the base station 110 can decide which time slot to transmit the data packets in. Furthermore, the user device 120 may also feed back the CQI measured in the time slot with the best predicted channel condition to the base station 110. This CQI can be used to determine the power allocation of the base station 110. In this way, the base station 110 can rationally allocate power while avoiding the use of time slots with poor channel conditions by transmitting data using power that matches the actual CQI of the time slot with a good channel condition, making the transmission of downlink data more rational and the use of resources more efficient. In addition, accurate transmission of downlink data can avoid resource overhead and waste caused by error correction later on.

[0076] The predictive analysis performed by the user device 120 plays a crucial role in network scheduling decisions, and this predictive analysis may be implemented based on a deep neural network (DNN) or the like. The predictive analysis, combined with feedback from the user device 120 to the CQI in the optimal time slot, contributes to improving downlink data transmission performance. A flowchart of the method 800 used for data transmission performed by the user device 120 will be specifically described below with reference to Figure 8.

[0077] In S810, the user device predicts information regarding the channel gains of multiple future time slots after the current time, based on the channel gains of multiple historical time slots prior to the current time. This information indicates the time slot with the highest channel gain among the multiple future time slots.

[0078] The user's device can obtain a predictive model for predictive analysis by training a machine learning model using a large amount of collected channel state information as training data. This predictive model may have the function of predicting the channel state of a certain number of future time slots based on the channel state of a certain number of past time slots. The implementation of the above function may employ existing methods for training machine learning models. For example, a dedicated development software package, such as the PyCharm development platform, can be used for creating and training neural networks. By training using the TensorFlow module construction model based on the PyCharm development platform, a model capable of implementing the corresponding function can be obtained.

[0079] After a large number of known training samples have been collected, it is necessary to label the training samples in order to obtain a large number of input and output data pairs for training the model. The channel states of N past time slots prior to a certain time t are labeled with their respective channel gains, and together they form the DNN input data x[t]=[h[tN-1],…,h[t-1],h[t]], where h[x] is the channel gain of the time slot corresponding to time x. Then, the channel states of M future time slots after this time t are labeled using at least one of the following four methods, and together they form the DNN output data y[t]. y[t] has M elements, the first element of which corresponds to the first time slot after time t, the second element of which corresponds to the second time slot after time t, and so on. N and M are the same or different positive integers greater than 1; for example, N may be equal to 8, 10, 16, etc., and M may be equal to 3, 5, 8, etc. Next, we input both DNN input data and DNN output data with numerous labeled correspondences into an existing DNN training program and fit each parameter in the DNN to obtain a predictive model that can predict the channel state of future time slots.

[0080] During the training process, different prediction modes can be obtained by labeling the DNN output data in different ways. Below, four different labeling methods are given to obtain four different prediction modes. These four prediction modes do not need to be used simultaneously. This means that a single communication system can use one or more of these prediction modes, or it can use all of them. The prediction modes used by different communication systems may be the same or different. If a single communication system can use multiple prediction modes, it may select which prediction mode to use for each prediction, or the administrator may configure which prediction mode to use. When a prediction is made using a particular prediction mode, the prediction result corresponding to that mode is obtained. These four prediction modes include the first prediction mode, the second prediction mode, the third prediction mode, and the fourth prediction mode. During the DNN training period, each of these four prediction modes has the DNN input data x[t] described above, and their output data will be described below.

[0081] In the first prediction mode, during the DNN training period, the time slot with the highest channel gain among the channel states of future time slots is labeled 1, and the other time slots are labeled 0, and these are jointly used as DNN output data. If multiple time slots have the same and highest channel gain, the oldest time slot among these multiple time slots is labeled 1, and the other time slots and the time slot with the lowest channel gain are all labeled 0, and these are jointly used as DNN output data. The DNN trained in this way receives the channel gains of N past time slots and outputs the time slot with the highest channel gain among M future time slots. For example, by using such a DNN, the user device 120 can obtain the output data y[t]=[0,1,0] from the channel gains of multiple historical time slots prior to the current time t, and determine that the channel state of the second time slot is the best among three future time slots.

[0082] In the second prediction mode, during the DNN training period, among the channel states of future time slots, time slots where the channel gain exceeds a preset threshold are labeled 1, and other time slots are labeled 0, and these are jointly used as DNN output data. The DNN thus trained receives the channel gains of N past time slots and outputs all time slots among M future time slots where the channel gain exceeds the threshold. For example, by using such a DNN, user equipment 120 can determine the output data y[t]=[0,1,1] from the channel gains of multiple historical time slots prior to the current time t, and determine that the channel gains in the future second and third time slots are above the set threshold and have acceptable channel quality, while the predicted channel gain for the future first time slot is poor and cannot be used for data transmission. Upon receiving such y[t] information, the base station can consider the second and third time slots as having the highest channel gains and choose to transmit data in them.

[0083] In the third prediction mode, during the DNN training period, the time slot with the highest channel gain among the channel states of future time slots is labeled as 1, and the other time slots are labeled as a percentage of their respective channel gains relative to the highest channel gain, and these are jointly used as the DNN output data. The DNN thus trained receives the channel gains of N past time slots and outputs the relative magnitude of the channel gains of each of the M future time slots. For example, using such a DNN, user device 120 can obtain the output data y[t]=[0.5,1,0.8] from the channel gains of multiple historical time slots prior to the current time t, and determine that among the three future time slots, the second time slot has the highest channel gain, the channel gain of the first time slot is 0.5 of the channel gain of the second time slot, and the channel gain of the third time slot is 0.8 of the channel gain of the second time slot. Upon receiving such y[t] information, the base station can determine that the channel conditions in the second time slot in the future are the best and choose to transmit data during that time.

[0084] In the fourth prediction mode, during the DNN training period, each future time slot is labeled with the channel gain value of that time slot and collectively becomes the output data of the DNN. The DNN thus trained receives the channel gains of N past time slots and outputs the channel gains of each of M future time slots. For example, by using such a DNN, the user device 120 can obtain the output data y[t]=[0.1,0.4,0.5] from the channel gains of multiple historical time slots prior to the current time t, and determine that h[t+1]=0.1, h[t+2]=0.4, h[t+3]=0.5.

[0085] While we have described the training and use of a predictive model using a DNN as an example of a machine learning model, as those skilled in the art can imagine, predictive models can also be trained using other machine learning models, such as recurrent neural networks and convolutional neural networks, and DNNs are not an limitation of this disclosure.

[0086] User equipment may use only one of the prediction modes described above to predict channel states, or it may use multiple of the prediction modes described above to predict channel states. If user equipment can use multiple prediction modes, it may use these multiple prediction modes simultaneously, or it may use a specific one for a certain period of time, depending on the user's or system's selection or settings. Regardless of which prediction mode the user equipment uses, the results obtained from the prediction must be fed back to the base station. This will allow the base station to easily perform resource scheduling and data transmission.

[0087] Figure 9 shows a schematic diagram of training a machine learning model. User device 120 collects channel state information by channel estimation during communication with base station 110 and uses it as sample data for training. For each of the four prediction modes described above, the sample data is labeled to obtain a large number of x[t] and y[t] pairs. These x[t] and y[t] pairs are then input into a machine learning model built on an existing development platform for training, thereby obtaining a predictive model capable of predicting the channel state. Regardless of the format of x[t] and y[t] adopted to train the model, a corresponding y[t] can be obtained based on x[t] during the process of using the model.

[0088] More specifically, Figure 10 shows a schematic diagram of training a machine learning model in the first prediction mode. The current time is represented as 0, the channel gains from time -5 to time 0 jointly constitute x[t], and future times 1 to 3 are labeled as y[t]=[0,1,0]. Each element in x[t] is input to each node at the input end of the model, and each element in y[t] is set to each node at the output end of the model. By inputting and performing calculations with a large number of x[t] and y[t] pairs into the model, it is possible to train a model that can predict the time slot with the highest channel gain in the future, based on the channel gains of historical time slots. As those skilled in the art will understand, for other prediction modes as well, a prediction model that achieves the function can be obtained by labeling the input and output data according to the function that the prediction mode achieves and training the model accordingly.

[0089] After the prediction model is trained, the user device 120 can input the channel gains of multiple historical time slots prior to the current time into the prediction model in real time to obtain a corresponding prediction result. Regardless of which prediction mode the user device 120 uses, the prediction result will relate to the channel gain of future time slots, and from the prediction result, it is possible to determine the time slot with the highest channel gain among future time slots.

[0090] Furthermore, the predictive operations performed by the user device using the predictive model may be initiated in response to a trigger command transmitted by the base station to start the prediction. For example, when data to be transmitted to the user device 120 arrives at the base station 110 or is generated by the base station 110, the base station 110 may send a trigger command to the user device 120. In response to receiving the trigger command, the user device 120 may start the predictive operations. The predictive operations of the user device 120 may also be configured as needed. For example, the user device 120 may be kept in a state where it can always perform channel prediction in cases such as when radio resources are insufficient, the radio channel is unstable, or there is extreme weather.

[0091] In the S820, the user device notifies the base station of the predicted information.

[0092] After the user device 120 obtains a prediction result using the prediction model, the user device 120 may immediately feed the prediction result back to the base station 110, for example, by transmitting the model's output result y[t] to the base station 110. Specifically, when the user device 120 performs a prediction using the first prediction mode, the user device 120 notifies the base station of the time slot with the highest channel gain among a plurality of future time slots. When the user device 120 performs a prediction using the second prediction mode, the user device 120 notifies the base station of all time slots among a plurality of future time slots in which the channel gain exceeds the first threshold. When the user device 120 performs a prediction using the third prediction mode, the user device 120 notifies the base station of the relative channel gain of each of the plurality of future time slots. When the user device 120 performs a prediction using the fourth prediction mode, the user device 120 notifies the base station of the channel gain of each of the plurality of future time slots. The prediction results may also be sent in a message responding to the trigger command described above (for example, feedback signaling).

[0093] Generally, since only one of the four prediction modes described above is used each time a prediction is made, the user device 120 may feed back only one type of prediction result to the base station 110. If multiple prediction modes are used simultaneously, the prediction results of all of these multiple prediction modes may be fed back to the base station. For example, from the viewpoint of computational complexity, it is preferable to use the first prediction mode. From the viewpoint of providing the base station with more channel state information so that the base station can better understand the channel status of each time slot, it is preferable to use the fourth prediction mode.

[0094] When user equipment 120 needs to feed back the channel gain of each future time slot based on the fourth prediction mode to base station 110, user equipment 120 may transmit the predicted channel gain of each future time slot to base station 110 via a compression model constructed by a deep neural network.

[0095] Figure 11A shows a schematic diagram of an architecture that transmits the predicted channel gain to the base station when the prediction model predicts the channel gain for each of several future time slots. Figure 11B is a sequence diagram corresponding to Figure 11A.

[0096] The prediction DNN 1110 and compression DNN 1120 may be located within the user equipment 120, and the decompression DNN 1130 may be located within the base station 110. The prediction DNN 1110 is used to predict the channel gain for a few future time slots after the current time from the channel gain for a few past time slots before the current time. The predicted channel gain is processed by the compression DNN 1120 to obtain intermediate data. The intermediate data is transmitted to the decompression DNN 1130 via the radio channel. The decompression DNN 1130 obtains decompressed data by processing the received intermediate data. This decompressed data is the restored predicted channel gain and is the same as the channel gain obtained by the prediction DNN 1110. The compression DNN 1120 and decompression DNN 1130 work together to perform compression and decompression processing of the predicted channel gain, thereby significantly reducing the amount of data transmitted via the radio channel compared to the amount of data of the predicted channel gain, reducing resource consumption for transmitting the prediction results, and improving system efficiency.

[0097] Compressed DNN 1120 and decompressed DNN 1130 may be trained together as a single machine learning model. For this model, one input layer, a predetermined number of hidden layers, and one output layer are constructed, with the number of nodes in each hidden layer being significantly smaller than the number of nodes in the input layer and the output layer. The same input and output data are also set for this model. When a machine learning development platform (e.g., PyCharm) converges the constructed model using the input and output data set in this way, it splits the model into two parts from one of the hidden layers (hereinafter referred to as the hidden layer), one of which constitutes compressed DNN 1120 and is deployed to the user equipment, and the other constitutes decompressed DNN 1130 and is deployed to the base station. In this way, compression and decompression processing can be achieved by constructing a machine learning model, enabling compressed transmission of channel gain with a completely new compression method, improving compression efficiency, and reducing radio resource consumption.

[0098] In Figure 11A, a compression module in user equipment and a decompression module in base station are explained using a DNN as an example. However, as those skilled in the art will understand, other machine learning modules may be used to construct the compression and decompression modules, provided that the amount of data transmitted over the radio channel is significantly less than the amount of data with the predicted channel gain.

[0099] Figure 11B shows the signal sequence diagram 1100 when a prediction model (e.g., Predictive DNN 1110) predicts the channel gain, and the user equipment transmits the channel gain to the base station.

[0100] In S1110, the base station transmits CSI-RS (Channel Status Information Reference Signal) signaling to the user equipment, instructing the user equipment to measure and provide feedback on channel quality.

[0101] In S1120, the user device predicts the channel gains of multiple future time slots using a fourth prediction mode based on the channel gains of multiple historical time slots, and obtains the channel gain for each of the multiple future time slots.

[0102] In S1130, the compressed DNN in the user's device processes the predicted multiple channel gains and obtains intermediate data from the intermediate node of the hidden layer, which is less data than the multiple channel gains input to the compressed DNN.

[0103] In S1140, the intermediate data output by the intermediate node is quantized and fed back.

[0104] In S1150, the expanded DNN at the base station processes the received intermediate data to recover the channel gain for several future time slots predicted by the user equipment. Based on these channel gains, the base station can determine the optimal future transmit time slot.

[0105] In S1160, the base station selects to transmit downlink data to the user equipment during the optimal transmission time slot in the future.

[0106] Returning to Figure 8, in S830, the user device notifies the base station of the channel quality instruction for the time slot having the highest channel gain indicated by the predicted information, and instructs the base station to transmit data in that time slot based on the channel quality instruction.

[0107] The base station 110 typically periodically transmits signaling to the user equipment 120 for channel quality measurement. For example, the base station 110 may transmit CSI-RS signaling to the user equipment 120 in each time slot. In response to receiving the CSI-RS signaling, the user equipment 120 may feed back the current CQI to the base station 110.

[0108] However, if user equipment 120 determines that the available uplink resources fall below a second threshold (which indicates a shortage of uplink resources), in order to use the limited uplink resources more rationally, user equipment 120 may not feed back the CQI each time it receives CSI-RS signaling, but wait for the optimal time slot indicated by predictive information, measure the channel only during that time slot, and feed back the actual CQI of that optimal time slot to base station 110. As a person skilled in the art will understand, if base station determines, for example, from recorded resource usage, information reported by user equipment, that the available uplink resources fall below a certain threshold, it may send a message to user equipment instructing user equipment to feed back the CQI only during the optimal time slot.

[0109] Furthermore, if user equipment 120 determines that the uplink feedback channel quality falls below a third threshold (which indicates poor uplink feedback channel quality), in order to avoid the problem of wasted transmission power and resources due to the failure to feed back CQI via a poor-quality uplink feedback channel, user equipment 120 may wait for an optimal time slot indicated by predictive information, rather than responding to all CSI-RS signaling, and only during that time slot, feed back the actual CQI for that optimal time slot to base station 110. As those skilled in the art will understand, if base station determines, for example, from previous data decoding operations, that the uplink feedback channel quality falls below a certain threshold, it may send a message to user equipment instructing user equipment to feed back CQI only during the optimal time slot.

[0110] The two situations described above may exist simultaneously in one embodiment or separately in different embodiments. The thresholds described above may be set as appropriate according to actual needs, provided that the objective of rational resource utilization and / or resource conservation is achieved.

[0111] Furthermore, to enable base stations to transmit downlink data with more appropriate power and in the optimal time slot, user equipment may actively feed back the actual CQI for the optimal time slot, regardless of whether the base station transmits CSI-RS signaling. Upon receiving the actual CQI for the optimal time slot, the base station may, as in related technologies, make an appropriate power allocation based on the CQI and transmit downlink data with appropriate power and in the optimal time slot.

[0112] User equipment may use one of the following two methods to feed back CQI to the base station. The first method is shown in Figures 12A, 12B, and 13, in which user equipment responds only to CSI-RS signaling for the optimal time slot and feeds back the CQI for that optimal time slot to the base station. The second method is shown in Figures 14A, 14B, and 15, in which user equipment, upon receiving CSI-RS signaling for a given time slot, feeds back the CQI for that time slot to the base station. Next, Figures 12 to 15 will be described.

[0113] Corresponding to the first method, Figure 12A shows a flowchart of data transmission method 1200-T performed by the base station, and Figure 12B shows a flowchart of data transmission method 1200-R performed by the user equipment.

[0114] As shown in Figure 12A, Method 1200-T begins. In S1210-T, a data packet to be sent to the user device arrives, and the base station sends a trigger command to the user device to initiate prediction. In S1220-T, the base station receives feedback from the user device containing prediction information. This feedback indicates the optimal time slot with the highest channel gain among several future time slots. In S1230-T, the base station waits for the optimal time slot with the highest channel gain. In S1240-T, the base station sends a data packet to the user device based on the CQI of the optimal time slot. Method 1200-T then ends.

[0115] As shown in Figure 12B, method 1200-R begins. In S1205-R, the user equipment continues to receive CSI-RS transmitted from the base station and obtains the channel gain for each time slot. This step is always performed and does not have a specific sequence relationship with the other steps. In S1210-R, the user equipment receives a trigger command from the base station to start prediction. In S1220-R, the user equipment predicts information about the channel gains of several future time slots after the current time, based on the channel gains of several past time slots before the current time. This information indicates the time slot with the highest channel gain. In this example, the first prediction mode is employed, and the prediction result is the optimal time slot with the highest channel gain. In S1230-R, the user equipment feeds back the predicted optimal time slot to the base station. In S1240-R, the user equipment waits for the predicted optimal time slot and feeds back the CQI for that time slot to the base station. In S1250-R, the user equipment receives data packets transmitted by the base station in the optimal time slot. Then, method 1200-R is terminated.

[0116] Figure 13 shows a sequence diagram corresponding to the method shown in Figures 12A and 12B.

[0117] In S1310, the base station transmits CSI-RS signaling to the user equipment in each time slot. Although this step is described only in S1310, it is always performed periodically, regardless of the execution order of the other steps.

[0118] In S1320, the user device performs channel estimation based on the received data and obtains channel state information, such as the channel gain for each time slot. Although this step is described only in S1320, this step is always performed, regardless of the execution order of the other steps.

[0119] In S1330, a new data packet sent to the user device reaches the base station.

[0120] In S1340, the base station sends a trigger command to the user equipment to initiate prediction.

[0121] In S1350, the user equipment feeds back the predicted optimal time slot to the base station.

[0122] In S1360, the user device waits for the arrival of the optimal time slot.

[0123] In S1370, the user device measures the CQI of the optimal time slot and feeds that CQI back to the base station.

[0124] In S1380, the base station transmits data packets to the user equipment in the optimal time slot based on the feedbacked CQI.

[0125] Corresponding to the second method, Figure 14A shows a flowchart of data transmission method 1400-T performed by the base station, and Figure 14B shows a flowchart of data transmission method 1400-R performed by the user equipment.

[0126] Data transmission method 1400-T is almost the same as data transmission method 1200-T, and steps S1410-T to S1440-T are almost identical to steps S1210-T to S1240-T, so no further explanation is provided here.

[0127] As shown in Figure 14B, method 1400-R begins. In S1405-R, the user equipment continues to receive CSI-RS transmitted from the base station and obtains the channel gain for each time slot. In S1409-R, the user equipment continues to feed back the CQI for each time slot, measured in response to the CSI-RS, to the base station. Steps S1405-R and S1409-R do not have a specific sequence relationship with the other steps and are always performed. Steps S1410-R to S1430-R are substantially the same as steps S1210-R to S1230-R. In S1440-R, the user equipment receives data packets transmitted by the base station in the optimal time slot. Method 1400-R then ends.

[0128] Figure 15 shows a sequence diagram corresponding to the method shown in Figures 14A and 14B.

[0129] S1510 is almost identical to S1310. In S1515, the user device feeds back the CQI to the base station in each time slot. The operations of S1510 and S1515 are performed only once and are shown to be performed from the beginning in Figure 15, but S1510 and S1515 are always performed in each time slot and are not particularly related to the execution order of the other steps. S1520 to S1560 are almost identical to S1320 to S1360. While the user device waits for the optimal time slot to arrive, the base station also waits for the optimal time slot to arrive and transmits a data packet in that time slot. In S1570, the base station transmits a data packet to the user device in the optimal time slot based on the CQI of the optimal time slot.

[0130] According to the proposed technology shown in Figure 8, by predicting the channel gain of future time slots based on the channel gain of historical time slots, the base station can determine the future time slot with the highest channel gain. This allows downlink data to be transmitted with power matching the channel quality indication of the time slot in which the channel quality is relatively good. This enables flexible determination of more appropriate transmission resources according to the current channel state, mitigating the problem of wasting power and resources by transmitting data in time slots with poor channel quality, and improving the transmission performance of downlink data.

[0131] The method described above can predict the time slots in which a base station will transmit downlink data. In other scenarios, the arrival delay of past data packets and the size of the data packets may be used to predict the incoming data traffic.

[0132] Figure 16 shows simulation results obtained using the data transmission method according to the embodiment of this disclosure. In the simulation process, the channel is a Rayleigh fading channel and the channel gain is exponentially distributed.

number

number

number

[0133] In the simulation diagram, Ta=1 indicates that no prediction is made, and the data packet is immediately transmitted by the base station in the next time slot after it arrives. Ta=2 indicates that the optimal time slot is predicted for the next two time slots, and Ta=3 indicates that the optimal time slot is predicted for the next three time slots; the rest can be inferred from these. From the simulation results, it can be seen that when data packets are transmitted immediately without any prediction, the average power consumption of the base station is maximized, and when prediction is adopted, the average power consumption of the base station decreases sharply. As the number of predictable time slots increases (the number of predictable time slots must not exceed the maximum delay that the data packet can tolerate), the average power consumption of the base station gradually decreases. These simulation results show that power consumption can be significantly reduced when the base station schedules and allocates power based on prediction results for user equipment.

[0134] Each exemplary electronic device and method according to the embodiments of this disclosure has been described above. It should be understood that the operation or function of these electronic devices may be combined with each other to achieve more or fewer operations or functions than those described. The operation steps of each method may also be combined with each other in any suitable order to achieve more or fewer operations than those described.

[0135] It should be understood that the machine-executable instructions in the machine-readable storage medium or program product according to the embodiments of this disclosure may be configured to perform operations corresponding to the embodiments of the above-described equipment and methods. When referring to the embodiments of the above-described equipment and methods, the embodiments of the machine-readable storage medium or program product will be obvious to those skilled in the art and will not be described further. Machine-readable storage medium or program product that contains or includes the above-described machine-executable instructions are also within the scope of this disclosure. Such storage media include, but are not limited to, floppy disks, optical disks, magneto-optical disks, memory cards, memory sticks, and the like.

[0136] Furthermore, it should be understood that the above series of processes and devices may be implemented by software and / or firmware. When implemented by software and / or firmware, the programs constituting this software are installed from a storage medium or network to a computer having a dedicated hardware configuration, for example, the general-purpose personal computer 1300 shown in Figure 17, and this computer can perform various functions when various programs are installed. Figure 17 is a block diagram showing an exemplary configuration of a personal computer, which is an information processing device that can be adopted in embodiments of this disclosure. In one example, the personal computer can correspond to the exemplary terminal device described in this disclosure.

[0137] In Figure 17, the central processing unit (CPU) 1301 executes various processes based on programs stored in read-only memory (ROM) 1302 or programs loaded from storage 1308 into random access memory (RAM) 1303. RAM 1303 also stores data necessary for the CPU 1301 to execute various processes as needed.

[0138] The CPU 1301, ROM 1302, and RAM 1303 are connected to each other via bus 1304. The input / output interface 1305 is also connected to bus 1304.

[0139] The input unit 1306, which includes a keyboard and mouse, the output unit 1307, which includes a display such as a cathode ray tube (CRT) or liquid crystal display (LCD) and speakers, the storage unit 1308, which includes a hard disk, and the communication unit 1309, which includes a network interface card such as a LAN card or modem, are connected to the input / output interface 1305. The communication unit 1309 performs communication processing via a network, such as the Internet.

[0140] If necessary, drive 1310 is also connected to input / output interface 1305. Removable media 1311, such as magnetic disks, optical disks, magneto-optical disks, or semiconductor memory, are installed on drive 1310 as necessary, and computer programs read from them are installed on storage 1308 as necessary.

[0141] When the above series of processes are implemented using software, the programs that make up the software are installed from a network, such as the Internet, or from a storage medium, such as removable media 1311.

[0142] Those skilled in the art should understand that such a storage medium is not limited to the removable media 1311 shown in Figure 17, which stores the program and is distributed separately from the device to provide the program to the user. Examples of removable media 1311 include magnetic disks (including floppy disks®), optical disks (including optical disk read-only memory (CD-ROM) and digital versatile disks (DVD)), magneto-optical disks (including MiniDisc (MD)®), and semiconductor memory. Alternatively, the storage medium may be a hard disk included in ROM 1302 or storage 1308, which stores the program and is distributed to the user together with the device containing them.

[0143] The technology of this disclosure can be applied to a variety of products. For example, the base station referred to in this disclosure may be implemented as an evolved node B (gNB) of any type, such as macro gNBs and small gNBs. A small gNB may be a gNB that covers cells smaller than macrocells, such as a pico gNB, micro gNB, or femto gNB. Alternatively, the base station may be implemented as any other type of base station, such as a Node B and a Base Transceiver Station (BTS). The base station may include an entity (also called base station equipment) configured to control the radio communication and one or more remote radio heads (RRHs) located separately from the entity. Furthermore, various types of terminals described later can operate as base stations by performing base station functions temporarily or semi-permanently.

[0144] For example, the terminal equipment referred to in this disclosure may, in some examples, also be called user equipment and may be implemented as a mobile terminal (e.g., a smartphone, tablet personal computer (PC), notebook PC, portable game console, portable / dongle mobile router, and digital imaging device) or an in-vehicle terminal (e.g., a car navigation system). User equipment may be implemented as a terminal that performs machine-to-machine (M2M) communication (also called a machine-type communication (MTC) terminal). User equipment may also be a wireless communication module (e.g., an integrated circuit module including a single chip) mounted on each of the above terminals.

[0145] Examples of applications of this disclosure will be explained below with reference to Figures 18 to 21.

[0146] [Application examples for base stations] It should be understood that the term "base station" in this disclosure has the full scope of its ordinary meaning and includes radio communication stations that are at least part of a radio communication system or radio system in order to perform communications. Examples of base stations include, for example, one or both of a base station transceiver (BTS) and / or a base station controller (BSC) in a GSM® system, one or both of a radio network controller (RNC) and / or a Node B in a WCDMA® system, an eNB in ​​LTE and LTE-Advanced systems, or a corresponding network node in a future communication system (e.g., a gNB, eLTE eNB, etc., which may appear in a 5G communication system). Some functions of base stations in this disclosure may be implemented as entities with control functions over communications in D2D, M2M and V2V communication scenarios, or as entities with spectrum tuning functions in cognitive radio communication scenarios.

[0147] (First application example) Figure 18 is a block diagram showing a first example of an exemplary configuration of a gNB to which the technology of the present disclosure can be applied. The gNB 1400 includes a plurality of antennas 1410 and base station equipment 1420. The base station equipment 1420 and each antenna 1410 can be connected to each other via RF cables. In one embodiment, the gNB 1400 (or base station equipment 1420) herein can correspond to the electronic equipment 300A, 1300A and / or 1500B.

[0148] Each of the antennas 1410 includes one or more antenna elements (for example, multiple antenna elements included in a multi-input multiple-output (MIMO) antenna) and is used by the base station equipment 1420 to transmit and receive radio signals. As shown in Figure 18, the gNB 1400 may include multiple antennas 1410. For example, the multiple antennas 1410 may be compatible with multiple frequency bands used by the gNB 1400.

[0149] The base station equipment 1420 includes a controller 1421, memory 1422, network interface 1423, and wireless communication interface 1425.

[0150] The controller 1421 may be, for example, a CPU or DSP, and may operate various functions of the upper layer of the base station equipment 1420. For example, the controller 1421 may generate data packets based on data in signals processed by the wireless communication interface 1425 and transmit the generated packets via the network interface 1423. The controller 1421 may bundle data from multiple baseband processors to generate bundle packets and transmit the generated bundle packets. The controller 1421 may have logical functions to perform the following controls, such as radio resource control, radio bearer control, mobility management, acceptance control, and scheduling. This control may be performed by connecting nearby gNBs or core network nodes. The memory 1422 includes RAM and ROM and stores programs executed by the controller 1421 and various types of control data (e.g., terminal lists, transmission power data, and scheduling data).

[0151] Network interface 1423 is a communication interface for connecting base station equipment 1420 to the core network 1424. Controller 1421 can communicate with core network nodes or other gNBs via network interface 1423. In this case, gNB 1400 and the core network nodes or other gNBs can be interconnected by logic interfaces (e.g., S1 interface and X2 interface). Network interface 1423 can also be a wired communication interface or a wireless communication interface used for a wireless backhaul line. If network interface 1423 is a wireless communication interface, it can be used for wireless communication using a higher frequency band compared to the frequency band used by wireless communication interface 1425.

[0152] The wireless communication interface 1425 supports any cellular communication scheme (e.g., Long Term Evolution (LTE) and LTE-Advanced) and provides wireless connectivity to terminals in cells located on the gNB 1400 via the antenna 1410. The wireless communication interface 1425 typically includes, for example, a baseband (BB) processor 1426 and RF circuitry 1427. The BB processor 1426 can perform, for example, encoding / decoding, modulation / demodulation, and multiplexing / demultiplexing, as well as various types of signal processing at different layers (e.g., L1, Media Access Control (MAC), Radio Link Control (RLC), Packet Data Aggregation Protocol (PDCP)). Instead of the controller 1421, the BB processor 1426 may have some or all of the above-described logical functions. The BB processor 1426 may be a memory in which a communication control program is stored, or it may be a module containing a processor and associated circuitry configured to execute the program. Program updates can change the functionality of the BB processor 1426. This module may be a card or bread inserted into a slot of base station equipment 1420. Alternatively, this module may be a chip mounted on a card or bread. Simultaneously, the RF circuit 1427, including, for example, a mixer, filter, and amplifier, can transmit and receive radio signals via the antenna 1410. Figure 18 shows an example in which one RF circuit 1427 is connected to one antenna 1410, but the disclosure is not limited to this illustration, and one RF circuit 1427 may be connected to multiple antennas 1410 simultaneously.

[0153] As shown in Figure 18, the wireless communication interface 1425 may include multiple BB processors 1426. For example, the multiple BB processors 1426 can be compatible with multiple frequency bands used by the gNB1400. As shown in Figure 18, the wireless communication interface 1425 may include multiple RF circuits 1427. For example, the multiple RF circuits 1427 can be compatible with multiple antenna elements. Although Figure 18 shows an example in which the wireless communication interface 1425 includes multiple BB processors 1426 and multiple RF circuits 1427, the wireless communication interface 1425 may include a single BB processor 1426 or a single RF circuit 1427.

[0154] (Second application example) Figure 19 is a block diagram showing a second example of an exemplary configuration of a gNB to which the technology of this disclosure can be applied. The gNB 1530 includes a plurality of antennas 1540, a base station device 1550, and an RRH 1560. The RRH 1560 and each antenna 1540 can be interconnected via an RF cable. The base station device 1550 and the RRH 1560 can be interconnected via a high-speed line such as an optical fiber cable. In one implementation, the gNB 1530 (or base station device 1550) herein can correspond to the electronic equipment 300A, 1300A, and / or 1500B.

[0155] Each of the antennas 1540 includes one or more antenna elements (for example, multiple antenna elements included in a MIMO antenna) and is used to transmit and receive radio signals of the RRH1560. As shown in Figure 19, the gNB1530 may include multiple antennas 1540. For example, the multiple antennas 1540 can be compatible with multiple frequency bands used by the gNB1530.

[0156] The base station equipment 1550 includes a controller 1551, a memory 1552, a network interface 1553, a wireless communication interface 1555, and a connection interface 1557. The controller 1551, memory 1552, and network interface 1553 are the same as the controller 1421, memory 1422, and network interface 1423 described with reference to Figure 18.

[0157] The wireless communication interface 1555 supports any cellular communication method (e.g., LTE and LTE-Advanced) and provides wireless communication to a terminal located in the sector corresponding to the RRH1560 via the RRH1560 and antenna 1540. The wireless communication interface 1555 may typically include, for example, a BB processor 1556. The BB processor 1556 is the same as the BB processor 1426 described with reference to Figure 18, except that the BB processor 1556 is connected to the RF circuit 1564 of the RRH1560 via a connection interface 1557. As shown in Figure 19, the wireless communication interface 1555 may include multiple BB processors 1556. For example, multiple BB processors 1556 can be compatible with multiple frequency bands used by the gNB1530. Although Figure 19 shows an example in which the wireless communication interface 1555 includes multiple BB processors 1556, the wireless communication interface 1555 may include a single BB processor 1556.

[0158] The connection interface 1557 is an interface for connecting the base station equipment 1550 (wireless communication interface 1555) to the RRH1560. The connection interface 1557 may also be a communication module for communication in the high-speed line described above that connects the base station equipment 1550 (wireless communication interface 1555) to the RRH1560.

[0159] The RRH1560 includes a connection interface 1561 and a wireless communication interface 1563.

[0160] The connection interface 1561 is an interface for connecting the RRH1560 (wireless communication interface 1563) to the base station equipment 1550. The connection interface 1561 may also be a communication module for communication on the high-speed line described above.

[0161] The wireless communication interface 1563 transmits and receives radio signals via the antenna 1540. The wireless communication interface 1563 may typically include, for example, an RF circuit 1564. The RF circuit 1564 may include, for example, a mixer, a filter, and an amplifier, and transmit and receive radio signals via the antenna 1540. Figure 19 shows an example in which one RF circuit 1564 is connected to one antenna 1540, but the disclosure is not limited to this illustration, and one RF circuit 1564 may be connected to multiple antennas 1540 simultaneously.

[0162] As shown in Figure 19, the wireless communication interface 1563 may include multiple RF circuits 1564. For example, multiple RF circuits 1564 can support multiple antenna elements. Although Figure 19 shows an example in which the wireless communication interface 1563 includes multiple RF circuits 1564, the wireless communication interface 1563 may also include a single RF circuit 1564.

[0163] [Examples of applications related to user devices] (First application example) Figure 20 is a block diagram showing an exemplary configuration of a smartphone 1600 to which the technology of the present disclosure can be applied. The smartphone 1600 includes a processor 1601, memory 1602, storage device 1603, external connection interface 1604, imaging device 1606, sensor 1607, microphone 1608, input device 1609, display device 1610, speaker 1611, wireless communication interface 1612, one or more antenna switches 1615, one or more antennas 1616, bus 1617, battery 1618, and auxiliary controller 1619. In one embodiment, the smartphone 1600 (or processor 1601) herein can correspond to the terminal devices 300B and / or 1500A.

[0164] The processor 1601 is, for example, a CPU or a system-on-a-chip (SoC) and can control the application layer and other layer functions of the smartphone 1600. The memory 1602 includes RAM and ROM and stores data and programs executed by the processor 1601. The storage device 1603 may include, for example, semiconductor memory and a storage medium such as a hard disk. The external connection interface 1604 is an interface for connecting external devices (e.g., memory cards and Universal Serial Bus (USB) devices) to the smartphone 1600.

[0165] The imaging device 1606 includes an image sensor (e.g., a charge-coupled device (CCD) and a complementary metal-oxide-semiconductor (CMOS)) and generates a captured image. Sensor 1607 may include a set of sensors such as a measuring sensor, a gyroscope, a geomagnetic sensor, and an accelerometer. Microphone 1608 converts sound input to the smartphone 1600 into an audio signal. Input device 1609 includes, for example, a touch sensor, keypad, keyboard, button, or switch positioned to detect touches on the screen of the display device 1610 and receives operations or information input from the user. Display device 1610 includes a screen (e.g., a liquid crystal display (LCD) and an organic light-emitting diode (OLED) display) and displays the output image from the smartphone 1600. Speaker 1611 converts the audio signal output from the smartphone 1600 into sound.

[0166] The wireless communication interface 1612 supports any cellular communication method (e.g., LTE and LTE-Advanced) and performs wireless communication. The wireless communication interface 1612 may typically include, for example, a broadband processor 1613 and an RF circuit 1614. The broadband processor 1613 can perform various types of signal processing for wireless communication, such as encoding / decoding, modulation / demodulation, and multiplexing / demultiplexing. At the same time, the RF circuit 1614, for example, includes a mixer, filter, and amplifier, and can transmit and receive wireless signals via the antenna 1616. The wireless communication interface 1612 may be a single chip module on which the broadband processor 1613 and the RF circuit 1614 are integrated. As shown in Figure 20, the wireless communication interface 1612 may include multiple broadband processors 1613 and multiple RF circuits 1614. Figure 20 shows an example in which the wireless communication interface 1612 includes multiple BB processors 1613 and multiple RF circuits 1614, but the wireless communication interface 1612 may also include a single BB processor 1613 or a single RF circuit 1614.

[0167] In addition to the cellular communication method, the wireless communication interface 1612 can support other types of wireless communication methods, such as short-range wireless communication, proximity communication, and wireless local network (LAN) methods. In this case, the wireless communication interface 1612 may include a BB processor 1613 and an RF circuit 1614 for each wireless communication method.

[0168] Each of the antenna switches 1615 switches the destination of the antenna 1616 among multiple circuits included in the wireless communication interface 1612 (for example, circuits used for different wireless communication methods).

[0169] Each of the antennas 1616 includes one or more antenna elements (for example, multiple antenna elements included in a MIMO antenna) and is used to transmit and receive radio signals of the wireless communication interface 1612. As shown in Figure 20, the smartphone 1600 may include multiple antennas 1616. Although Figure 20 shows an example in which the smartphone 1600 includes multiple antennas 1616, the smartphone 1600 may also include a single antenna 1616.

[0170] The smartphone 1600 may also include antennas 1616 for each wireless communication method. In this case, the antenna switch 1615 may be omitted from the arrangement of the smartphone 1600.

[0171] Bus 1617 connects the processor 1601, memory 1602, storage device 1603, external connection interface 1604, imaging device 1606, sensor 1607, microphone 1608, input device 1609, display device 1610, speaker 1611, wireless communication interface 1612, and auxiliary controller 1619 to each other. Battery 1618 provides power to each block of the smartphone 1600 shown in Figure 20 via power lines. Power lines are partially indicated by dotted lines in the drawing. The auxiliary controller 1619 operates the minimum necessary functions of the smartphone 1600, for example, in sleep mode.

[0172] (Second application example) Figure 21 is a block diagram showing an exemplary arrangement of a car navigation device 1720 to which the technology of the present disclosure can be applied. The car navigation device 1720 includes a processor 1721, memory 1722, a global positioning system (GPS) module 1724, a sensor 1725, a data interface 1726, a content player 1727, a storage medium interface 1728, an input device 1729, a display device 1730, a speaker 1731, a wireless communication interface 1733, one or more antenna switches 1736, one or more antennas 1737, and a battery 1738. In one implementation, the car navigation device 1720 (or processor 1721) herein can correspond to the terminal devices 300B and / or 1500A.

[0173] The processor 1721 is, for example, a CPU or SoC, and can control the navigation and other functions of the car navigation device 1720. The memory 1722 includes RAM and ROM and stores data and programs executed by the processor 1721.

[0174] The GPS module 1724 measures the position (e.g., latitude, longitude, altitude) of the car navigation device 1720 using GPS signals received from GPS satellites. The sensor 1725 may include a set of sensors, such as a gyro sensor, a geomagnetic sensor, and a barometric pressure sensor. The data interface 1726 connects to, for example, an in-vehicle network 1741 via a terminal (not shown) to acquire data generated by the vehicle (e.g., vehicle speed data).

[0175] The content player 1727 plays content stored on a storage medium (e.g., CDs and DVDs). This storage medium is inserted into the storage medium interface 1728. The input device 1729 includes, for example, a touch sensor, button, or switch positioned to detect touches on the screen of the display device 1730, and receives operations or information input from the user. The display device 1730 includes, for example, an LCD or OLED display screen, and displays images of the navigation function or the played content. The speaker 1731 outputs sounds of the navigation function or the played content.

[0176] The wireless communication interface 1733 supports any cellular communication method (e.g., LTE and LTE-Advanced) and can perform wireless communication. The wireless communication interface 1733 may typically include, for example, a broadband processor 1734 and an RF circuit 1735. The broadband processor 1734 can perform various types of signal processing for wireless communication, such as encoding / decoding, modulation / demodulation, and multiplexing / demultiplexing. At the same time, the RF circuit 1735, which includes, for example, a mixer, filter, and amplifier, can transmit and receive wireless signals via the antenna 1737. The wireless communication interface 1733 may be a single chip module on which the broadband processor 1734 and the RF circuit 1735 are integrated. As shown in Figure 21, the wireless communication interface 1733 may include multiple broadband processors 1734 and multiple RF circuits 1735. Figure 21 shows an example in which the wireless communication interface 1733 includes multiple BB processors 1734 and multiple RF circuits 1735, but the wireless communication interface 1733 may also include a single BB processor 1734 or a single RF circuit 1735.

[0177] In addition to the cellular communication method, the wireless communication interface 1733 can support other types of wireless communication methods, such as short-range wireless communication, proximity communication, and wireless LAN. In this case, the wireless communication interface 1733 may include a BB processor 1734 and an RF circuit 1735 for each wireless communication method.

[0178] Each of the antenna switches 1736 switches the destination of the antenna 1737 between multiple circuits in the wireless communication interface 1733 (for example, circuits used for different wireless communication methods).

[0179] Each of the antennas 1737 includes one or more antenna elements (for example, multiple antenna elements included in a MIMO antenna) and is used to transmit and receive radio signals of the wireless communication interface 1733. As shown in Figure 21, the car navigation device 1720 may include multiple antennas 1737. Although Figure 21 shows an example in which the car navigation device 1720 includes multiple antennas 1737, the car navigation device 1720 may include a single antenna 1737.

[0180] The car navigation device 1720 may also include antennas 1737 for each wireless communication method. In this case, the antenna switch 1736 may be omitted from the arrangement of the car navigation device 1720.

[0181] Battery 1738 supplies power to each block of the car navigation device 1720 shown in Figure 21 via power supply lines. The power supply lines are partially indicated by dotted lines in the drawing. Battery 1738 stores the power supplied from the vehicle.

[0182] The technology described herein may be implemented as an in-vehicle system (or vehicle) 1740 which includes one or more blocks of a car navigation device 1720, an in-vehicle network 1741, and a vehicle module 1742. The vehicle module 1742 generates vehicle data (e.g., vehicle speed, engine speed, fault information) and outputs the generated data to the in-vehicle network 1741.

[0183] While exemplary embodiments of the present disclosure have been described above with reference to the drawings, the present disclosure is, of course, not limited to these examples. Those skilled in the art will understand that various changes and modifications can be made within the scope of the appended claims, and that such changes and modifications will fall within the scope of the art of the present disclosure.

[0184] For example, the multiple functions included in one unit in the above embodiments can be implemented by separate devices. Alternatively, the multiple functions implemented by multiple units in the above embodiments can each be implemented by separate devices. Furthermore, one of the above functions can be implemented by multiple units. Of course, such arrangements are within the scope of the art of this disclosure.

[0185] In this specification, the steps described in the flowchart include not only processes that are executed chronologically in the order described, but also processes that are not necessarily executed chronologically but in parallel or individually. Furthermore, of course, the order of the steps that are executed chronologically may also be changed as appropriate.

[0186] While the present disclosure and its merits have been described in detail, it should be understood that various modifications, substitutions, and transformations are possible, provided they do not fall outside the spirit and scope of the present disclosure, which is limited to the claims attached. Furthermore, the terms “including,” “incorporating,” or any other variant thereof in the embodiments of the present disclosure, by their non-exclusive nature, mean that a process, method, article, or apparatus containing a set of elements includes not only those elements but also other elements not explicitly stated, or elements specific to such a process, method, article, or apparatus. Unless further restrictions are imposed, an element limited by “including one…” does not exclude other identical elements from being included in a process, method, article, or apparatus containing such elements.

[0187] From the description herein, it can be understood that embodiments of this disclosure may be configured as follows: 1. An electronic device used on the user equipment side in a wireless communication system, which includes a processing circuit system, the processing circuit system is The control entity sends information to the user device about the quantity of training samples to be used for the current training period of the local model. The system receives uplink resource information for uploading local model parameters, and the uplink resources indicated by the uplink resource information are allocated by the control entity based on quantity information from multiple user devices, such that the larger the quantity indicated by the quantity information for a user device, the greater the chance that sufficient uplink resources will be allocated for uploading local model parameters. An electronic device configured to upload parameters of a local model to a control entity via an uplink resource indicated by the uplink resource information, so that the control entity obtains the next global model. 2. The processing circuit system further includes: It is configured to transmit distance information indicating the distance between the user device and the base station to the control entity. The electronic device described in Clause 1, wherein the uplink resources indicated by the uplink resource information are allocated by the control entity based on the quantity information and distance information from a plurality of user devices, such that the larger the quantity indicated by the quantity information and the smaller the distance indicated by the distance information, the greater the chance that sufficient uplink resources will be allocated to upload local model parameters for the user device. 3. The uplink resources indicated by the uplink resource information include the electronic devices described in Clause 1, including the uplink resources for URLLC traffic. 4. The uplink resources indicated by the uplink resource information include the uplink resources of non-delay-sensitive traffic conducted between the base station and the user devices, as described in Clause 1, when the control entity receives quantity information from the user devices. 5. The electronic device described in Clause 4, wherein the non-delay-sensitive traffic is eMBB traffic. 6. The electronic device described in Clause 1, wherein the control entity and the plurality of user devices jointly achieve federated learning, and the control entity is a base station. 7. Uploading parameters for a local model is performed by electronic equipment as described in Clause 1, which satisfies the specified QoS requirements specified by the given 5QI values. 8. The processing circuit system further includes: It is configured to receive the parameters of the next global model from the control entity and update the parameters of the local model, The transmission of the parameters of the following global model is performed by the electronic equipment described in Clause 1, which satisfies a predetermined QoS requirement specified by a predetermined 5QI value. 9. The predetermined 5QI values ​​are, The resource type is guaranteed bitrate GBR. The default priority level is 70. The packet delay budget is 10ms. Packet error rate is 10 -6 Being Electronic equipment as defined in Clause 7 or 8, which defines at least one of the following: the default average window is 2000ms. 10. The electronic device described in Clause 9, wherein the 5QI value is 87. 11. Electronic equipment used on the network equipment side in a wireless communication system, which includes a processing circuit system, the processing circuit system is The user device receives information from the user device regarding the quantity of training samples that the user device will use for the current training period of the local model. Uplink resource information for uploading local model parameters is transmitted, and the uplink resources indicated by the uplink resource information are allocated by the processing circuit system based on quantity information from multiple user devices, such that the larger the quantity indicated by the quantity information for a user device, the greater the chance that sufficient uplink resources will be allocated for uploading local model parameters. An electronic device configured to receive parameters of a local model uploaded by a user device via an uplink resource indicated by the uplink resource information, in order to obtain the next global model. 12. The processing circuit system further includes: It is configured to receive distance information from the user device indicating the distance between the user device and the base station. The electronic device according to Clause 11, wherein the uplink resources indicated by the uplink resource information are allocated by the processing circuit system based on the quantity information and distance information from the plurality of user devices, such that the larger the quantity indicated by the quantity information and the smaller the distance indicated by the distance information, the greater the chance that sufficient uplink resources will be allocated to upload the parameters of the local model. 13. The uplink resources indicated by the uplink resource information include the electronic devices described in Clause 11, including uplink resources for URLLC traffic. 14. The electronic device described in Clause 11, wherein the uplink resources indicated by the uplink resource information include uplink resources of the base station and the plurality of user devices when the processing circuit system receives quantity information from the plurality of user devices, and the plurality of user devices are performing non-delay sensitive traffic. 15. The electronic device described in Clause 14, wherein the non-delay-sensitive traffic is eMBB traffic. 16. The electronic device described in Clause 11, wherein the electronic device and the plurality of user devices jointly achieve federated learning, and the electronic device is a base station. 17. Uploading parameters for a local model is performed by electronic equipment as described in Clause 11, which satisfies the specified QoS requirements specified by the given 5QI values. 18. The processing circuit system further includes: The system is configured to send the parameters of the next global model to the user device, causing the user device to update the parameters of the local model. The transmission of the parameters of the following global model is performed by the electronic equipment described in Clause 11, which satisfies a predetermined QoS requirement specified by a predetermined 5QI value. 19. The predetermined 5QI values ​​are, The resource type is guaranteed bitrate GBR. The default priority level is 70. The packet delay budget is 10ms. Packet error rate is 10 -6 Being Electronic equipment as defined in Clause 17 or 18, which defines at least one of the following: a default average window of 2000ms. 20. The electronic device described in Clause 19, wherein the 5QI value is 87. 21. A method used in a wireless communication system, The control entity transmits information about the quantity of training samples that the user device will use for the current training period of the local model, The process involves receiving uplink resource information for uploading local model parameters, where the uplink resources indicated by the uplink resource information are allocated by the control entity based on quantity information from multiple user devices, such that user devices with larger quantities indicated by quantity information have a greater chance of being allocated sufficient uplink resources to upload local model parameters. A method comprising uploading parameters of a local model to a control entity via an uplink resource indicated by the uplink resource information, so that the control entity obtains the following global model. 22. A method used in a wireless communication system, Receiving information from the user device regarding the quantity of training samples that the user device will use for the current training period of the local model, The transmission of uplink resource information for uploading local model parameters, wherein the uplink resources indicated by the uplink resource information are allocated based on quantity information from multiple user devices, such that user devices with larger quantities indicated by quantity information have a greater chance of being allocated sufficient uplink resources to upload local model parameters. A method comprising receiving parameters of a local model uploaded by a user device via an uplink resource indicated by the uplink resource information, in order to obtain the next global model. 23. A computer-readable storage medium storing one or more instructions that, when executed by one or more processors of an electronic device, cause the electronic device to perform the method described in Clause 21 or 22. 24. An apparatus used in a wireless communication system, including a component for performing an operation as described in Clause 21 or 22. 25. Electronic equipment used on the user equipment side in a wireless communication system, which includes a processing circuit system, Predict information regarding channel gains of a plurality of future time slots after the current time based on channel gains of a plurality of history time slots before the current time, the information indicating a time slot having the highest channel gain among the plurality of future time slots, Notify the base station of the information, Notify the base station of a channel quality indication in a time slot having the highest channel gain indicated by the information, and configure the base station to transmit data based on the channel quality indication in the time slot. An electronic device. 26. The processing circuit system further Is configured to receive a trigger command for starting a prediction transmitted from a base station, The foregoing prediction of information regarding channel gains of a plurality of future time slots after the current time The electronic device according to clause 25, further comprising predicting information regarding channel gains of a plurality of future time slots after the current time in response to receiving a trigger command. 27. The foregoing receiving a trigger command for starting a prediction transmitted from a base station The electronic device according to clause 26, including receiving a trigger command for starting a prediction transmitted by the base station in response to the appearance of data to be transmitted to the user equipment. 28. The information is Among the plurality of future time slots, a time slot having the highest channel gain, and Among the plurality of future time slots, all time slots whose channel gains exceed a first threshold, and The relative channel gain of each of the plurality of future time slots, and The electronic device according to clause 25, including at least one of each channel gain of the plurality of future time slots. 29. The electronic device according to Clause 28, wherein the information is the channel gain of each of the plurality of future time slots, and transmits the channel gain of each of the plurality of future time slots to a base station via a compression model constructed by a deep neural network. 30. Notifying the base station of the channel quality instruction in the time slot having the highest channel gain as indicated by the aforementioned information is: The electronic device described in Clause 25, which, when it is determined that available uplink resources fall below a second threshold, or when it is determined that uplink feedback channel quality falls below a third threshold, notifies the base station only of a channel quality instruction for that time slot, and does not notify channel quality instructions for other time slots among the plurality of future time slots. 31. Electronic equipment used on the base station side in a wireless communication system, which includes a processing circuit system, the processing circuit system is The user device receives information regarding channel gains in multiple future time slots after the current time, the information being predicted by the user device based on the channel gains of multiple historical time slots prior to the current time, and the information indicating the time slot with the highest channel gain among the multiple future time slots. The user device receives a channel quality instruction for the time slot having the highest channel gain indicated by the information, An electronic device configured to transmit data based on the channel quality instruction in a time slot having the highest channel gain indicated by the aforementioned information. 32. The processing circuit system further includes: The electronic device described in Clause 31, which is configured to send a trigger command to a user device to initiate a prediction. 33. Sending the trigger command to initiate the prediction, as mentioned above, to the user's device, The electronic device described in Clause 32, which includes sending a trigger command to the user device to initiate prediction in response to the appearance of data to be sent to the user device. 34. The above information is Of the aforementioned multiple future time slots, the time slot having the highest channel gain, Of the aforementioned multiple future time slots, all time slots in which the channel gain exceeds the first threshold, The relative channel gain of each of the aforementioned plurality of future time slots, The electronic device according to Clause 31, comprising at least one of the channel gains of each of the plurality of future time slots. 35. If the information is the channel gain for each of the plurality of future time slots, the electronic device described in Clause 34, wherein the channel gain for each time slot is received from the user device via a compression model constructed by a deep neural network. 36. Receiving a channel quality instruction from the user equipment in the time slot having the highest channel gain as indicated by the aforementioned information is: The electronic device described in Clause 31, which, when it is determined that available uplink resources fall below a second threshold, or when it is determined that uplink feedback channel quality falls below a third threshold, receives only a channel quality instruction from the user device for that time slot, and does not receive channel quality instructions for other time slots among the plurality of future time slots. 37. A method used in a wireless communication system, Predicting information regarding the channel gains of multiple future time slots after the current time, based on the channel gains of multiple historical time slots prior to the current time, wherein the information indicates the time slot having the highest channel gain among the multiple future time slots. To notify the base station of the aforementioned information, A method comprising notifying a base station of a channel quality instruction for a time slot having the highest channel gain indicated by the aforementioned information, and causing the base station to transmit data in that time slot based on the channel quality instruction. 38. A method used in a wireless communication system, Receiving information from a user device regarding channel gains in multiple future time slots after the current time, wherein the information is predicted by the user device based on the channel gains of multiple historical time slots prior to the current time, and the information indicates the time slot with the highest channel gain among the multiple future time slots. Receiving a channel quality instruction from the user device in the time slot having the highest channel gain indicated by the information, A method comprising transmitting data based on the channel quality instruction in a time slot having the highest channel gain indicated by the aforementioned information. 39. A computer-readable storage medium storing one or more instructions that, when executed by one or more processors of an electronic device, cause the electronic device to perform the actions described in Clause 37 or 38. 40. Apparatus used in a wireless communication system, including a component for performing an operation as described in Clause 37 or 38.

Claims

1. An electronic device used on the user equipment side in a wireless communication system, which includes a processing circuit system, the processing circuit system is The control entity sends information to the user device about the quantity of training samples to be used for the current training period of the local model. The system receives uplink resource information for uploading local model parameters, and the uplink resources indicated by the uplink resource information are allocated by the control entity based on quantity information from multiple user devices, such that the larger the quantity indicated by the quantity information for a user device, the greater the chance that sufficient uplink resources will be allocated for uploading local model parameters. An electronic device configured to upload parameters of a local model to a control entity via an uplink resource indicated by the uplink resource information, so that the control entity obtains the next global model.

2. The processing circuit system further includes, It is configured to transmit distance information indicating the distance between the user device and the base station to the control entity. The electronic device according to claim 1, wherein the uplink resources indicated by the uplink resource information are allocated by a control entity based on the quantity information and distance information from a plurality of user devices, such that the larger the quantity indicated by the quantity information and the smaller the distance indicated by the distance information, the greater the chance that sufficient uplink resources will be allocated to upload local model parameters for the user device.

3. The uplink resources indicated by the aforementioned uplink resource information include and / or uplink resources for URLC traffic. The uplink resources indicated by the uplink resource information include, when the control entity receives quantity information from the plurality of user devices, the uplink resources of the base station and the plurality of user devices that are conducting non-delay-sensitive traffic, and / or The control entity and the multiple user devices jointly achieve federated learning, and the control entity is a base station and / or The electronic device according to claim 1, wherein uploading parameters of a local model satisfies a predetermined QoS requirement specified by a predetermined 5QI value.

4. The processing circuit system further includes, It is configured to receive the parameters of the next global model from the control entity and update the parameters of the local model, The electronic device according to claim 1, wherein the transmission of the parameters of the following global model satisfies a predetermined QoS requirement specified by a predetermined 5QI value.

5. The predetermined 5QI value is, The resource type is guaranteed bitrate GBR. The default priority level is 70. The packet delay budget is 10 ms. Packet error rate is 10 －６ Being The electronic device according to claim 3 or 4, defining at least one of the following: the default average window is 2000 ms.

6. The electronic device according to claim 5, wherein the 5QI value is 87.

7. The electronic device according to claim 1, wherein the uplink resource indicated by the uplink resource information includes uplink resources of non-delay-sensitive traffic being conducted by the base station and the plurality of user devices when the control entity receives quantity information from the plurality of user devices, and the non-delay-sensitive traffic is eMBB traffic.

8. An electronic device used on the network equipment side in a wireless communication system, which includes a processing circuit system, the processing circuit system is The user device receives information from the user device regarding the quantity of training samples that the user device will use for the current training period of the local model. Uplink resource information for uploading local model parameters is transmitted, and the uplink resources indicated by the uplink resource information are allocated by the processing circuit system based on quantity information from multiple user devices, such that the larger the quantity indicated by the quantity information for a user device, the greater the chance that sufficient uplink resources will be allocated for uploading local model parameters. An electronic device configured to receive parameters of a local model uploaded by a user device via an uplink resource indicated by the uplink resource information, in order to obtain the next global model.

9. The processing circuit system further includes, It is configured to receive distance information from the user device indicating the distance between the user device and the base station. The electronic device according to claim 8, wherein the uplink resources indicated by the uplink resource information are allocated by the processing circuit system based on the quantity information and distance information from the plurality of user devices, such that the larger the quantity indicated by the quantity information and the smaller the distance indicated by the distance information, the greater the chance that sufficient uplink resources will be allocated to upload the parameters of the local model for the user device.

10. The uplink resources indicated by the aforementioned uplink resource information include and / or uplink resources for URLC traffic. The uplink resources indicated by the uplink resource information include, when the processing circuit system receives quantity information from the plurality of user devices, the uplink resources of the base station and the plurality of user devices that are performing non-delay sensitive traffic, and / or The aforementioned electronic device and the multiple user devices jointly realize federated learning, and the aforementioned electronic device is a base station and / or, The electronic device according to claim 8, wherein uploading parameters of a local model satisfies a predetermined QoS requirement specified by a predetermined 5QI value.

11. The processing circuit system further includes, The system is configured to send the parameters of the next global model to the user device, causing the user device to update the parameters of the local model. The electronic device according to claim 8, wherein the transmission of the parameters of the next global model satisfies a predetermined QoS requirement specified by a predetermined 5QI value.

12. The predetermined 5QI value is, The resource type is guaranteed bitrate GBR. The default priority level is 70. The packet delay budget is 10 ms. Packet error rate is 10 －６ Being The electronic device according to claim 10 or 11, defining at least one of the following: the default average window is 2000 ms.

13. The electronic device according to claim 12, wherein the 5QI value is 87.

14. The electronic device according to claim 8, wherein the uplink resources indicated by the uplink resource information include uplink resources of non-delay sensitive traffic being conducted by the base station and the plurality of user devices when the processing circuit system receives quantity information from the plurality of user devices, and the non-delay sensitive traffic is eMBB traffic.

15. A method used in wireless communication systems, The control entity transmits information about the quantity of training samples that the user device will use for the current training period of the local model, The process involves receiving uplink resource information for uploading local model parameters, where the uplink resources indicated by the uplink resource information are allocated by the control entity based on quantity information from multiple user devices, such that user devices with larger quantities indicated by quantity information have a greater chance of being allocated sufficient uplink resources to upload local model parameters. A method comprising uploading parameters of a local model to a control entity via an uplink resource indicated by the uplink resource information, so that the control entity obtains the following global model.

16. A method used in wireless communication systems, Receiving information from the user device regarding the quantity of training samples that the user device will use for the current training period of the local model, The transmission of uplink resource information for uploading local model parameters, wherein the uplink resources indicated by the uplink resource information are allocated based on quantity information from multiple user devices, such that user devices with larger quantities indicated by quantity information have a greater chance of being allocated sufficient uplink resources to upload local model parameters. A method comprising receiving parameters of a local model uploaded by a user device via an uplink resource indicated by the uplink resource information, in order to obtain the next global model.

17. A computer-readable storage medium storing one or more instructions that, when executed by one or more processors of an electronic device, cause the electronic device to perform the method according to claim 15 or 16.

18. An apparatus used in a wireless communication system, comprising a component for performing an operation according to claim 15 or 16.

19. A computer program comprising one or more instructions that, when executed by one or more processors of an electronic device, cause the electronic device to perform the method according to claim 15 or 16.

Images