Service transmission and reception point selection for user equipments

Abstract

Techniques are provided for determining service sending and receiving points of user equipment participating in an air joint learning process. The method comprises obtaining signal quality information of each user equipment in a group of user equipment. The method includes obtaining an indication from each user equipment. The method includes determining, based on the obtained signal quality information for each user equipment and the obtained indications from all user equipment, which user equipment is to be served by which transmitting and receiving point when performing the air joint learning process.

Description

Technical Field

[0001] Embodiments presented herein relate to methods, network nodes, computer programs, and computer program products for determining service transmission and reception points for a user equipment participating in an over-the-air federated learning process. Embodiments presented herein also relate to methods, user equipment, computer programs, and computer program products for assisting in determining service transmission and reception points for a user equipment when participating in an over-the-air federated learning process. Background Art

[0002] Growing concerns about data privacy have motivated consideration of collaborative machine learning (ML) systems that exploit discrete data, where training data is stored and processed locally by edge user devices (e.g., user equipment (UE)). Federated learning (FL) is a non-limiting example of a discrete learning topology, where multiple (potentially very large) agents, e.g., implemented in UEs, participate in training a shared global learning model by exchanging model updates with a centralized parameter server (PS) (e.g., implemented in a network node).

[0003] In general, FL can be viewed as an ML technique for training ML (or artificial intelligence; AI) models across multiple discrete agents, each of which uses local data samples to perform local model training. The technique requires multiple interactions of the model, but does not require the exchange of local data samples. In this regard, the ML or AI model can be viewed as a data-driven algorithm that applies AI / ML techniques to generate a set of outputs based on a set of inputs. Therefore, AI / ML model training can be viewed as a process of training an AI / ML model by learning the input / output relationship in a data-driven manner, and obtaining a trained AI / ML model for reasoning. The AI / ML model can be transmitted over the air interface; either the parameters of the model structure known at the receiving end, or a new model with parameters. In this regard, model download refers to the transmission of the model from the PS to the agent, and model upload refers to the transmission of the model from the agent to the PS.

[0004] In more detail, FL is an iterative process in which each global iteration (often referred to as an iteration round) is divided into three phases: In the first phase, the PS sends the current model parameter vector to all participating agents. In the second phase, each agent performs one or several steps of the stochastic gradient descent (SGD) process on its own training data based on the current model parameter vector and obtains a model update. In the third phase, the model updates from all agents are sent to the PS, which aggregates the received model updates and updates the parameter vector for the next iteration based on the model updates according to some aggregation rules. The first phase is then entered again, but with the updated parameter vector as the current model parameter vector.

[0005] In the following, it will be assumed that FL or a similar iterative learning process will be performed between a proxy placed in the UE and a parameter server placed on the network side of the wireless communication network. Therefore, some aspects of the wireless communication network will be disclosed next.

[0006] Denser wireless communication networks and systems operating on an increasing number of frequency bands are being developed. In the ongoing deployment of fifth generation (5G) telecommunication systems, further densification is required to meet the requirements, which means more network nodes are deployed. Denser networks and more frequencies make it more challenging to ensure that the serving cells and frequency carriers are allocated as optimally as possible for each served UE.

[0007] Traffic steering / load balancing is a technique for balancing the traffic load between overlapping cells in a network in order to utilize the capacity of different frequency layers. Figure 1 , Figure 1 Schematically, four cells 120a, 120b, 120c, 120d of the communication network are shown. Each cell 120a:120d may be served by a corresponding transmission and reception point. Alternatively, each transmission and reception point serves two or more cells 120a:120d. Alternatively, each cell 120a:120d may be served by two or more transmission and reception points. Figure 1 Also shown are UEs 300a, 300b, 300c, 300d, 300e served in cells 120a:120d. The double-headed arrows indicate potential opportunities for load balancing. The network node evaluates the traffic load in the cells 120a:120d. By executing a load balancing algorithm, the network node can identify whether it is necessary to move UEs between cells 120a:120d (i.e., move one or more of the UEs 300a:300e from one end of their double-headed arrows to the other end of their double-headed arrows) to balance the traffic load. If it is necessary to move the UE to balance the traffic load, the UE is selected and instructed to perform inter-frequency measurements. The network node then uses these measurements to direct the UE to the desired cell using, for example, a handover and / or a release with a redirection procedure.

[0008] In the case of training an ML model using an iterative learning process in which updates are sent over the air from an agent in the UE to a parameter server on the network side using analog modulation, the UE sends model updates on the same time-frequency resources, and adjusts its transmission to enable the parameter server to obtain the average model update. When UEs are distributed over a large geographical area, the efficiency of this iterative learning process depends in part on the number of UE groups that perform model aggregation in each update. For example, a group of UEs served in one cell may perform model updates in that cell, while another group of UEs participating in training the same model may perform model updates in another cell. The parameter server may create an aggregated model based on the updates from both groups of UEs, and broadcast the updates via the serving cell of each UE. This is particularly useful in Figure 2 It is shown in Figure 2 Schematically, a communication network with three transmission and reception points 110a, 110b, 110c is shown, each transmission and reception point serving its own cell 120a, 120b, 120c. The transmission and reception point 110a is serving UEs 300a, 300b in the cell 120a, the transmission and reception point 110b is serving UEs 300c, 300d, 300e in the cell 120b, and the transmission and reception point 110c is serving UEs 300f, 300g in the cell 120c. Thus, local updates for UEs 300a, 300b are sent via the transmission and reception point 110a, while local updates for UEs 300c, 300d, 300e are sent via the transmission and reception point 110b, and local updates for UEs 300f, 300g are sent via the transmission and reception point 110c.

[0009] In the scenario where multiple carrier frequencies are used in neighboring cells, UEs served in neighboring cells are therefore served on different carrier frequencies due to the load balancing configuration of the network. This will lead to inefficient resource utilization of the over-the-air joint learning process in which case one or more of the participating UEs are also in the coverage of another cell operating on a different carrier frequency.

[0010] Therefore, there is still a need for improved over-the-air federated learning. Summary of the invention

[0011] Embodiments herein aim to address the above issues by enabling resource-efficient traffic steering in the context of over-the-air federated learning.

[0012] According to a first aspect, a method for determining a serving sending and receiving point for a user equipment participating in an air-joint learning process is proposed. The method is performed by a network node. The method comprises: obtaining signal quality information for each user equipment in a set of user equipment. The signal quality information relates to the corresponding signal quality between the user equipment and each sending and receiving point pair in a set of sending and receiving points. The method comprises: obtaining an indication from each user equipment. The indication from a given user equipment relates to whether the given user equipment is to participate in an air-joint learning process with a parameter server via one of the sending and receiving points. The method comprises: determining which user equipment is to be served by which sending and receiving point when performing an air-joint learning process based on the obtained signal quality information for each user equipment and the obtained indications from all user equipment.

[0013] According to a second aspect, a network node for determining a serving sending and receiving point of a user equipment participating in an air joint learning process is proposed. The network node includes a processing circuit. The processing circuit is configured to enable the network node to obtain signal quality information for each user equipment in a group of user equipment. The signal quality information relates to the corresponding signal quality between the user equipment and each sending and receiving point pair in a group of sending and receiving points. The processing circuit is configured to enable the network node to obtain an indication from each user equipment. The indication from a given user equipment relates to whether the given user equipment is to participate in the air joint learning process with a parameter server via one of the sending and receiving points. The processing circuit is configured to enable the network node to determine which user equipment is to be served by which sending and receiving point when performing the air joint learning process based on the obtained signal quality information of each user equipment and the obtained indications from all user equipment.

[0014] According to a third aspect, a network node for determining a serving sending and receiving point of a user equipment participating in an air joint learning process is proposed. The network node includes an acquisition module, which is configured to obtain signal quality information of each user equipment in a group of user equipment. The signal quality information relates to the corresponding signal quality between the user equipment and each sending and receiving point pair in a group of sending and receiving points. The network node includes an acquisition module, which is configured to obtain an indication from each user equipment. The indication from a given user equipment relates to whether the given user equipment is to participate in the air joint learning process with a parameter server via one of the sending and receiving points. The network node includes a determination module, which is configured to determine which user equipment is to be served by which sending and receiving point when performing the air joint learning process based on the obtained signal quality information of each user equipment and the obtained indications from all user equipment.

[0015] According to a fourth aspect, a computer program for determining serving sending and receiving points of a user equipment participating in an over-the-air federated learning process is proposed, the computer program comprising computer program code which, when executed on a processing circuit of a network node, causes the network node to perform the method according to the first aspect.

[0016] According to a fifth aspect, a method for assisting in determining a serving transmission and reception point for a user equipment participating in an over-the-air joint learning process is proposed. The method is performed by a user equipment. The method comprises providing a network node with radio measurements made by the user equipment on a downlink reference signal sent by the transmission and reception point. The method comprises providing an indication to the network node that the user equipment will participate in an over-the-air joint learning process with a parameter server via one of the transmission and reception points. The method comprises obtaining from the network node an indication of which transmission and reception point will serve the user equipment when performing the over-the-air joint learning process.

[0017] According to a sixth aspect, a user equipment is proposed for assisting in determining a serving transmission and reception point of a user equipment to participate in an air-joint learning process. The user equipment comprises a processing circuit. The processing circuit is configured to cause the user equipment to provide a network node with radio measurements made by the user equipment on a downlink reference signal sent by the transmission and reception point. The processing circuit is configured to cause the user equipment to provide the network node with an indication that the user equipment will participate in an air-joint learning process with a parameter server via one of the transmission and reception points. The processing circuit is configured to cause the user equipment to obtain from the network node an indication of which transmission and reception point will serve the user equipment when performing the air-joint learning process.

[0018] According to a seventh aspect, a user equipment is proposed for assisting in determining a serving transmission and reception point of a user equipment to participate in an air-joint learning process. The user equipment comprises a providing module configured to provide a network node with radio measurements performed by the user equipment on a downlink reference signal transmitted by the transmission and reception point. The user equipment comprises a providing module configured to provide an indication to the network node that the user equipment will participate in an air-joint learning process with a parameter server via one of the transmission and reception points. The user equipment comprises an obtaining module configured to obtain from the network node an indication of which transmission and reception point will serve the user equipment when performing the air-joint learning process.

[0019] According to an eighth aspect, a computer program for assisting in determining service sending and receiving points of a user equipment participating in an over-the-air joint learning process is proposed, the computer program comprising computer program code which, when executed on a processing circuit of the user equipment, causes the user equipment to perform a method according to the fifth aspect.

[0020] According to a ninth aspect, a computer program product is provided, the computer program product comprising a computer program according to at least one of the fourth aspect and the eighth aspect and a computer-readable storage medium on which the computer program is stored. The computer-readable storage medium may be a non-transitory computer-readable storage medium.

[0021] Advantageously, these aspects provide resource efficient traffic steering of UEs participating in an over-the-air joint learning procedure.

[0022] Advantageously, these aspects enable finding the best transmit and receive points to serve each UE participating in the over-the-air joint learning process.

[0023] Advantageously, these aspects improve the resource efficiency of the over-the-air joint learning process, for example by reducing the number of groups into which the UEs are divided, or by selecting transmit and receive points with low traffic load (and therefore available network resources) to serve the UEs.

[0024] Advantageously, these aspects also result in energy savings and improved bit rates for UEs not participating in the over-the-air joint learning process due to more efficient overall usage of network resources.

[0025] Other objectives, features and advantages of the attached embodiments will become apparent from the following detailed disclosure, the attached dependent claims as well as the accompanying drawings.

[0026] Generally, unless otherwise expressly stated herein, all terms used in the claims are to be interpreted according to their ordinary meaning in the technical field. Unless otherwise expressly stated, all references to "a / an / the element, device, component, device, module, step, etc." should be openly interpreted as referring to at least one instance of an element, device, component, device, module, step, etc. Unless expressly stated otherwise, the steps of any method disclosed herein do not have to be strictly performed in the exact order disclosed. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The inventive concept will now be described by way of example with reference to the accompanying drawings, in which:

[0028] Figure 1 is a schematic diagram of a cell according to an embodiment;

[0029] Figure 2 is a schematic diagram of transmission and reception points serving a cell according to an embodiment;

[0030] Figure 3 and Figure 4 is a flow chart of a method according to an embodiment;

[0031] Figure 5is a schematic diagram of transmission and reception points serving a cell according to an embodiment;

[0032] Figure 6 is a signaling diagram of a method according to an embodiment;

[0033] Figure 7 is a schematic diagram of functional units of a network node according to an embodiment;

[0034] Figure 8 is a schematic diagram of functional modules of a network node according to an embodiment;

[0035] Fig. 9 is a schematic diagram of functional units of a user equipment according to an embodiment;

[0036] Fig.10 is a schematic diagram of functional modules of a user equipment according to an embodiment;

[0037] Fig.11 An example of a computer program product including computer readable means according to an embodiment is shown; and

[0038] Fig.12 is a schematic diagram illustrating a telecommunications network connected to a host computer via an intermediate network according to some embodiments; and

[0039] Fig.13 is a schematic diagram illustrating a host computer communicating with a terminal device over a partially wireless connection via a radio base station according to some embodiments. DETAILED DESCRIPTION

[0040] The inventive concept will now be described more fully below with reference to the accompanying drawings showing certain embodiments of the inventive concept. However, the inventive concept can be implemented in many different forms and should not be construed as being limited to the embodiments set forth herein; on the contrary, these embodiments are provided by way of example so that the disclosure will be comprehensive and complete and the scope of the inventive concept will be fully conveyed to those skilled in the art. Throughout this specification, similar reference numerals refer to similar elements. Any steps or features shown by dashed lines should be considered optional.

[0041] The statement that the first device obtains a certain data item, information, etc. should be interpreted as the data item or information being acquired, obtained, received, or otherwise made available to the first device. For example, the data item or information can be pushed from the second device to the first device or pulled from the second device by the first device. In addition, in order for the first device to obtain the data item or information, the first device can be configured to perform a series of operations, which may include interaction with the second device. Such operations or interactions may involve message exchanges, which include any of a request message for the data item or information, a response message including the data item or information, and a reply message for the data item or information. If the first device neither explicitly nor implicitly requests the data item or information, the request message can be omitted.

[0042] A statement that a first device provides a certain data item, information, etc. to a second device should be interpreted as the data item or information being sent by the first device to the second device or otherwise made available to the second device. For example, the data item or information may be pushed from the first device to the second device or pulled from the first device by the second device. In addition, in order for the first device to provide the data item or information to the second device, the first device and the second device may be configured to perform a series of operations to interact with each other. Such operations or interactions may involve a message exchange including any one of a request message for the data item or information, a response message including the data item or information, and a reply message for the data item or information. If the second device neither explicitly nor implicitly requests the data item or information, the request message may be omitted.

[0043] Furthermore, the statement that a data item, information, etc. is sent from a first device to a second device should be interpreted as the data item or information being sent directly from the first device to the second device or from the first device to the second device via at least one intermediate device.

[0044] It is assumed that the UE 300a:300K will be utilized during the iterative learning process, and the UE 300a:300K will report the computation results to the network node 200 as part of performing the iterative learning process. Therefore, the network node 200 may include, co-locate or integrate with a server entity. Each of the UEs 300a:300K includes, co-locates or integrates with a corresponding agent. In some aspects, the server entity and the agent are provided in a distributed computing architecture.

[0045] Consider a server entity and The invention relates to a system of multiple proxies (i.e. one proxy per UE). The server entity and the proxy communicate via the UE on a shared medium capable of broadcasting, i.e. sending information that can be received by several UEs simultaneously. A main example of such a medium is a wireless channel, where technologies standardized by the Third Generation Partnership Project (3GPP) such as Long Term Evolution (LTE) and New Radio (NR) are employed. Hereinafter, a wireless communication medium is assumed.

[0046] The following summarizes the process of one iteration of the iteratively trained learning model. First, the server entity broadcasts a parameter vector (which may be different from the parameter vector of the previously broadcast model, as described in detail below). It is assumed that this parameter vector is received by the UE. Then, each UE calculates a local model update based on the local training data and then sends its locally obtained update to the server entity. Finally, the server entity aggregates all received local updates to update the learning model.

[0047] Next, we will present an example of a nominal iterative learning process. Consider a Each transmission from the UE is assigned Resource Elements (REs). These REs can be time / frequency samples or spatial patterns.

[0048] The server entity is indexed by executing The global iteration updates its estimate of the learning model (maintained as the global model in step S0) ), is defined as the parameter vector . Assume that the parameter vector yes dimensional vector. In each iteration , perform the following steps:

[0049] Step S1: The server entity sends the current parameter vector of the learning model to the UE .

[0050] In this regard, the current parameter vector of the learning model is sent to the UE There can be different ways, such as unicast digital transmission, broadcast transmission or multicast transmission. Some properties of each of these types of transmission will be disclosed next.

[0051] For unicast digital transmissions, the server entity allocates orthogonal time, frequency and / or space resources to each UE. For each UE, the corresponding transmission rate and coding and modulation scheme are determined. The model or the difference update compared to the previously broadcast model is quantized and compressed using source code. For each UE, a modulation and coding scheme is applied, which is customized to adapt to the rate that the channel of the UE can support. Each UE then decodes the transmission containing the model (or its difference update).

[0052] For broadcast transmission, the server entity transmits in such a way that all UEs can decode the model. Due to the different properties of the radio propagation channel for each UE, for example resulting in different path losses for different UEs, this may require the selection of a very low transmission rate (using heavy error control coding). Another alternative is to use an error control code that can be decoded using bits that are only partially received. For example, if a channel code with a pseudo-random structure of its parity check matrix (e.g., a low-density parity check code; LDPC) is used, UEs with small path losses (and therefore high received signal-to-noise ratio SNR) can decode the broadcasted model after receiving a relatively small number of bits. In contrast, UEs with higher path losses will have to receive more parity bits before they can decode. UEs with relatively high SNR (compared to other UEs) can not only decode the model faster, but can also start their training calculations (e.g., stochastic gradients) earlier, thereby achieving a reduction in the clock frequency of the calculations, which in turn leads to lower power consumption.

[0053] With multicast transmission, UEs may be divided into different groups, where each group is assigned a respective uplink pilot resource. In this way, when beamforming transmission is used, multiple beams may be used, where each beam is adapted to serve a specific group of UEs. Multicast transmission may be viewed as a combination of unicast and broadcast techniques, and there may be situations where such grouping is preferred for performance reasons.

[0054] Step S2: Each UE performs Run stochastic gradient descent updates on steps to perform local optimization on the model;

[0055]

[0056] in, is the weight, and It is UE The objective function used at t(i) is the function that is used at t(i) (and it is based on its locally available training data).

[0057] Step S3: Each UE sends its model update to the server entity ;

[0058]

[0059] in, It is UE The model received from the server entity.

[0060] Step S4: The server entity adds the linear combination (weighted sum) of the updates received from the UE to its parameter vector The estimated value of is added to update the estimated value;

[0061]

[0062] in, is the weight.

[0063] Now suppose there is UE and therefore exists Model update. When the model is updated { } (where the time index has been removed for simplicity) from the UE via a wireless communication channel, there are particular benefits to using direct analog modulation. For analog modulation, UE can directly pass Resource Elements (RE) sent of Here, RE can be, for example: (i) one sample in time in a single-carrier system, or (ii) one subcarrier in an orthogonal frequency division multiplexing (OFDM) symbol in a multi-carrier system, or (iii) a specific spatial beam or a combination of a beam and a time / frequency resource.

[0064] One benefit of direct analog modulation is that the superposition property of wireless communication channels can be exploited to compute the aggregated update More specifically, the UE does not send a message to the server entity on a separate channel. , but use Each RE simultaneously sends model updates via linear analog modulation The server entity can then exploit the wave superposition property of the wireless communication channel, i.e. By ignoring noise and interference, the server entity will therefore receive the linear sum as expected. That is, the server entity ultimately updates only the aggregated model is of interest, rather than for each individual parameter vector Therefore, this technique can be referred to as iterative learning with over-the-air computation.

[0065] The over-the-air calculation assumes that appropriate power control is applied (so that the same power is received at the server entity) } all sent), and each sent Before transmission, phase rotation is performed appropriately to pre-compensate for the The phase rotation caused by the channel to the server entity.

[0066] One benefit of the over-the-air computation described above is the saving of radio resources. ), compared to standard FL, 50% of resources can be saved because the two UEs can send their model updates simultaneously in the same RE. UE, only the nominal required resources are needed part.

[0067] As disclosed above, there remains a need for improved over-the-air federated learning.

[0068] Embodiments disclosed herein relate to techniques for a network node 200 to determine a serving transmission and reception point 110a:110M for a UE 300a:300K participating in an over-the-air joint learning process. Embodiments disclosed herein also relate to a UE 300a:300K assisting a network node 200 in determining a serving transmission and reception point 110a for a UE 300a:300K when the UE 300a:300K participates in an over-the-air joint learning process.

[0069] At least some of these embodiments are based on which UE 300a:300K is to be served by which transmit and receive point 110a:110M when performing an over-the-air joint learning process based on a combination of radio environment information based on signal quality information of each UE 300a:300K on the one hand and an indication from the UE 300a:300K as to whether the UE 300a:300K will participate in the over-the-air joint learning process. This will, for example, improve efficiency by reducing the number of UE groups performing the over-the-air joint learning process.

[0070] Reference now Figure 3 , Figure 3 A method performed by the network node 200 for determining serving transmission and reception points 110a : 110M for a UE 300a : 300K participating in an over-the-air joint learning process according to an embodiment is shown.

[0071] S102: The network node 200 obtains signal quality information of each UE 300a:300K in a group of UEs 300a:300K. The signal quality information relates to the corresponding signal quality between each UE 300a:300K and a pair of transmission and reception points 110a:110M in a group of transmission and reception points 110a:110M.

[0072] S104: The network node 200 obtains an indication from each UE 300a:300K. The indication from a given UE 300a:300K relates to whether the given UE 300a:300K will participate in an over-the-air joint learning process with a parameter server via one of the sending and receiving points 110a:110M. In this regard, the network node 200 may obtain this information as a response to a previous request sent by the network node 200 to the UE 300a:300K. Thus, some of the UEs 300a:300K may respond to the network node 200 that these UEs 300a:300K will participate in an over-the-air joint learning process, while other of the UEs 300a:300K may respond to the network node 200 that these UEs 300a:300K will not participate in an over-the-air joint learning process (e.g., where the UE 300a:300K is not equipped with an agent).

[0073] S112: The network node 200 determines which UE 300a:300K is to be served by which transmission and reception point 110a:110M when performing the air joint learning process based on the obtained signal quality information of each UE 300a:300K and the obtained indications from all UEs 300a:300K. In some aspects, the determination involves selecting which UE 300a:300K is to be served by which transmission and reception point 110a:110M when performing the air joint learning process. Therefore, the terms determination and selection can be used interchangeably in this context.

[0074] Embodiments will now be disclosed relating to further details of the process performed by the network node 200 for determining the serving transmission and reception points 110a : 110M for the UE 300a : 300K participating in the over-the-air federated learning process.

[0075] In some examples, each of the transmission and reception points 110a: 110M is configured to provide service for its own cell. In some examples, the cells of at least two different transmission and reception points 110a: 110M have mutually different carrier frequencies. In some examples, two or more of the transmission and reception points 110a: 110M are co-located.

[0076] In some aspects, before starting the over-the-air joint learning process, during an initial phase, the UE 300a:300K may receive a configuration indicating instructions for performing inter-frequency carrier measurements from the network node 200. The measurement process will be disclosed in further detail below.

[0077] Next, how the network node 200 can obtain the signal quality information in step S102 will be disclosed.

[0078] In some aspects, the UE 300a:300K provides the network node 200 with signal quality measurements of the transmit and receive points 110a:110M that are candidates for participating in the over-the-air joint learning process. For example, each transmit and receive point 110a:110M serving the UE 300a:300K may configure inter-frequency measurements for carriers intended for use in the over-the-air joint learning process. Thus, in some embodiments, the signal quality information may be obtained from radio measurements performed by the UE 300a:300K on downlink reference signals sent by the transmit and receive point 110a:110M and / or from radio measurements performed by the transmit and receive point 110a:110M on uplink reference signals sent by the UE 300a:300K. In more detail, in order to perform the measurement, the UE 300a:300K may generally measure a reference signal transmitted by the transmission and reception point 110a:110M, for example, to evaluate the quality of the transmission beam transmitting the reference signal. Generally speaking, the reference signal transmitted by at least one transmission and reception point 110a:110M to the UE 300a:300K may include at least one of a channel state information reference signal (CSI-RS), a synchronization signal block (SSB), a primary synchronization signal (PSS), a secondary synchronization signal (SSS), and a cell-specific reference signal (CRS). In addition, the UE 300a:300K may evaluate the signal quality based on a reference signal received power (RSRP) or a reference signal received quality (RSRQ). More specifically, the UE 300a:300K may evaluate the beam quality via measurements of SSBs (e.g., corresponding to synchronization signal / physical broadcast channel (PBCH) blocks) in a 5G (e.g., NR) network or via measurements of CSI-RS resources in a 5G (e.g., NR) network or a 4G (e.g., LTE) network. The information destined for the network node 200 may be sent by the UE 300a:300K via its serving transmission and reception point 110a:110M and then from the transmission and reception point 110a:110M to the network node 200 via an XN interface.

[0079] In some aspects, the network node 200 predicts the signal quality of the UE 300a:300K. Thus, in some embodiments, the signal quality information is obtained from a signal quality prediction made by the network node 200 or from a signal quality prediction provided to the network node 200. In some non-limiting examples, this is achieved using a secondary carrier prediction (SCP), which provides a model for predicting other carrier or radio access technology (RAT) radio conditions based on measurements that are readily available or available on a source carrier. Using this method, the radio position of each UE 300a:300K can be given, and an estimate of the UE cell signal quality of another frequency can be estimated. The expected signal quality can include an RSRP estimate / measurement of a candidate transmit and receive point 110a:110M. Another method for finding a UE signal quality estimate on a candidate transmit and receive point 110a:110M operating on another carrier frequency is based on the geographic location of the UE 300a:300K. For example, a UE 300a:300K that is near a candidate transmission and reception point 110a:110M may be expected to have a higher predicted signal quality than a UE 300a:300K that is farther away from the same candidate transmission and reception point 110a:110M. The position of the UE 300a:300K may be estimated by any available LTE / NR positioning process, i.e., Assisted Global Navigation Satellite System (A-GNSS), Observed Time Difference of Arrival (OTODA), Uplink-Time Difference of Arrival (UTDOA), and Enhanced Cell Identity (E-CID) techniques.

[0080] Based on the measured or predicted signal quality of the UE 300a:300K, the network node 200 may request more information from the candidate transmission and reception points 110a:110M that is audible to the UE 300a:300K participating in the over-the-air joint learning process. An example of this information is provided next.

[0081] According to a first example, the information comprises service information.Therefore, in some embodiments, the network node 200 is configured to perform (optional) step S106.

[0082] S106: The network node 200 obtains service information. The service information relates to any one of the following items: the number of UEs 300a:300K currently served by each sending and receiving point 110a:110M, or the prediction of the number of UEs 300a:300K to be served by each sending and receiving point 110a:110M, the service load of each sending and receiving point 110a:110M, and the available resources at each sending and receiving point 110a:110M for participating in the air joint learning process. Then, which UE 300a:300K is to be served by which sending and receiving point 110a:110M can also be determined based on the service information. Therefore, the service information can relate to the number of UEs 300a:300K served or the predicted number of UEs 300a:300K served. The service information can also relate to the load of these UEs 300a:300K (predicted or actually served). The load may relate to resources (eg, physical resource blocks (PRBs) or subframes). The transmission and reception points 110a: 110M may also provide an estimate of the amount of time-frequency resources that may be used for the over-the-air joint learning process.

[0083] According to a second example, the information includes hardware information. The hardware information may relate to how accurately the transmission and reception point 110a:110M can estimate the channel to each UE 300a:300K, or to other impairments (e.g., phase noise). Therefore, in some embodiments, the network node 200 is configured to perform (optional) step S108.

[0084] S108: The network node 200 obtains accuracy information. The accuracy information relates to the accuracy level of the channel estimation performed by each of the transmission and reception points 110a:110M for the UE 300a:300K. Then, it can also be determined based on the accuracy information which UE 300a:300K is to be served by which transmission and reception point 110a:110M.

[0085] According to a third example, the information comprises the link quality to the UE 300a:300K.

[0086] The UE 300a:300K may also indicate to the network node 200 the importance of its data set. Thus, in some embodiments, each user equipment 300a:300K to participate in the over-the-air joint learning process is configured to train on a local data set as part of the over-the-air joint learning process. The network node 200 may take this information into account when determining which user equipment 300a:300K is to be served by which transmission and reception point 110a:110M. The importance or description may, for example, include statistical information of its response variable, the number of data samples, etc. Thus, in some embodiments, the network node 200 is configured to perform (optional) step S110.

[0087] S110: The network node 200 obtains data set importance information from the UE 300a:300K. The data set importance information relates to the uniqueness level of each local data set. The data set importance information may relate to information directly indicating the importance of the data set, or information that can be used by the network node 200 to infer the importance of the data set. Then, it may also be determined based on the data set importance information which UE 300a:300K is to be served by which transmission and reception point 110a:110M.

[0088] As an example, in a heterogeneous data scenario, a given UE 300a:300K may have access to very unique data such that the given UE 300a:300K should not be ignored in the learning process. As a non-limiting illustrative example, consider a federated learning process involving handwritten digit recognition, where there are 100 participating UEs 300a:300K. If 99 UEs 300a:300K access samples of digits "0" to "8", but only 1 UE 300a:300K accesses digit "9", it is important that the network node 200 receives the results provided by this UE 300a:300K during the federated learning process.

[0089] Next, aspects of how the network node 200 in step S112 can determine which user equipment 300a:300K is to be served by which transmission and reception point 110a:110M will be disclosed.

[0090] One criterion is to minimize the number of handovers. That is, a transmission and reception point 110a:110M (cell) to which most UEs 300a:300K are already connected is selected. Therefore, in some embodiments, before determining which UE 300a:300K is to be served by which transmission and reception point 110a:110M, the UE 300a:300K is served by an initial transmission and reception point 110a:110M in the group of transmission and reception points 110a:110M. Then, which UE 300a:300K is to be served by which transmission and reception point 110a:110M can be determined based on a handover criterion for reducing the number of handovers from the initial transmission and reception point 110a:110M.

[0091] One criterion is to select the transmission and reception points 110a:110M so that the signal quality of all UEs 300a:300K is above a certain threshold for each transmission and reception point 110a:110M. Here, signal quality may refer to a signal-to-noise ratio (SNR) or a signal-to-interference-plus-noise ratio (SINR). Signal quality may also refer to a statistical characterization of a link blocking probability parameterized, for example, in terms of a blocking probability or a Markov chain (with the parameters of the Gilbert-Elliott model as one possible example).

[0092] One criterion is to select the transmission and reception points 110a:110M taking into account the time variability of the wireless channel, quantified, for example, in terms of the channel coherence time (Doppler frequency), as this may ultimately determine how the UE 300a:300K adjusts its transmission to adapt to the over-the-air joint learning process. A transmission and reception point 110a:110M whose wireless channel to the UE 300a:300K varies more slowly may be more adaptable to the over-the-air joint learning process and may therefore be preferred in the selection.

[0093] One criterion is to select a transmit and receive point 110a:110M with suitable processing capabilities (e.g., sufficient antennas with suitable digital processing) to support the over-the-air joint learning process transmission of the UE 300a:300K. For example, a transmit and receive point 110a:110M with a small number or a single antenna may benefit more from the over-the-air joint learning process transmission than a multi-antenna (MIMO) transmit and receive point 110a:110M because the MIMO transmit and receive point 110a:110M may provide conventional spatial multiplexing of the UE 300a:300K in the same resource block. Therefore, in some embodiments, each of the transmit and receive points 110a:110M has processing capabilities for participating in the over-the-air joint learning process, and which UE 300a:300K is to be served by which transmit and receive point 110a:110M is determined based on the processing capabilities of each transmit and receive point 110a:110M.

[0094] One criterion is to select the transmission and reception point 110a : 110M with the most resources available for the over-the-air joint learning process (eg, the transmission and reception point 110a : 110M with the least number of served UEs 300a : 300K).

[0095] One criterion is to select the transmission and reception points 110a:110M that minimize the number of groups of UEs 300a:300K required. Thus, in some embodiments, which UE 300a:300K is to be served by which transmission and reception point 110a:110M is determined based on a selection criterion for reducing the total number of transmission and reception points 110a:110M required to serve the UEs 300a:300K.

[0096] One criterion is to select the UE 300a:300K with the most unique data sets with the transmit and receive point 110a:110M with the best path gain. An example of this criterion is the heterogeneous data scenario disclosed above. That is, as described above, each UE 300a:300K to participate in the air joint learning process can be configured to train on a local data set as part of the air joint learning process, wherein each local data set has a corresponding uniqueness level, and wherein the signal quality reflects the path gain between the UE 300a:300K and the transmit and receive point 110a:110M. Then, the UE 300a:300K with the local data set with the highest uniqueness level can be determined to be served by the transmit and receive point 110a:110M with the highest path gain. Additionally or alternatively, the UE 300a:300K with the highest requirement for energy efficiency can be determined to be served by the transmit and receive point 110a:110M with the highest path gain.

[0097] As disclosed above, before determining which UE 300a:300K is to be served by which transmission and reception point 110a:110M, the UE 300a:300K may be served by the initial transmission and reception point 110a:110M in the group of transmission and reception points 110a:110M. Therefore, once the decision is made (i.e., when it has been determined which UE 300a:300K is to be served by which transmission and reception point 110a:110M), the corresponding configuration may be sent to the UE 300a:300K. This may include instructions for some of the UEs 300a:300K to initiate a handover to a target cell. Therefore, in some embodiments, the network node 200 is configured to perform (optional) step S114.

[0098] S114: The network node 200 initiates a first handover for each UE 300a:300K determined to be served by another transmission and reception point 110a:110M than its initial transmission and reception point 110a.

[0099] Once the over-the-air joint learning process is completed, some of the UEs 300a:300K may optionally be instructed to return to their previous serving cell. This configuration may be sent a priori (eg, during an initial phase). Thus, in some embodiments, the network node 200 is configured to perform (optional) step S116.

[0100] S116: When the over-the-air joint learning process has been terminated, the network node 200 initiates a second handover back to the initial transmission and reception point 110a: 110M for each UE 300a: 300K for which the first handover was performed.

[0101] Reference now Figure 4 , Figure 4 A method for a UE 300a : 300K to assist in determining a serving transmission and reception point 110a for the UE 300a : 300K when participating in an over-the-air joint learning process according to an embodiment is shown.

[0102] S202: The UE 300a:300K provides the network node 200 with radio measurements performed by the UE 300a:300K on the downlink reference signal transmitted by the transmission and reception point 110a:110M.

[0103] S204: The UE 300a:300K provides an indication to the network node 200 that the UE 300a:300K will participate in an over-the-air joint learning process with the parameter server via one of the transmission and reception points 110a:110M.

[0104] S208: The UE 300a:300K obtains an indication from the network node 200 as to which transmission and reception point 110a:110M will provide service to the UE 300a:300K when performing the over-the-air joint learning procedure.

[0105] Embodiments will now be disclosed relating to further details of operations performed by the UE 300a: 300K to assist in determining the serving transmission and reception points 110a.

[0106] The UE 300a:300K is configured to perform training on a local dataset as part of an over-the-air federated learning process. As disclosed above, the UE 300a:300K may also indicate its dataset importance to the network node 200. Therefore, in some embodiments, the UE 300a:300K is configured to perform (optional) step S206.

[0107] S206: UE 300a: 300K provides data set importance information to the network node 200. The data set importance information relates to the uniqueness level of the local data set.

[0108] As disclosed above, the UE 300a:300K is served by the initial sending and receiving point 110a before obtaining an indication of which sending and receiving point 110a:110M will serve the UE 300a:300K. As disclosed above, some UEs 300a:300K may be instructed to initiate a handover to a target cell. Therefore, in some embodiments, the UE 300a:300K is configured to perform (optional) step S210.

[0109] S210: When performing the over-the-air joint learning process, the UE 300a:300K performs a first handover for the UE 300a:300K to be served by another transmission and reception point 110a:110M other than its initial transmission and reception point 110a.

[0110] As disclosed above, once the over-the-air joint learning process is completed, some of the UEs 300a:300K may optionally be instructed to return to their previous serving cell.Therefore, in some embodiments, the UE 300a:300K is configured to perform (optional) step S212.

[0111] S212: When the over-the-air joint learning process has been terminated, the UE 300a: 300K performs a second handover back to the initial transmission and reception point 110a.

[0112] Next reference Figure 5 , Figure 5 A communication network with three transmission and reception points 110a, 110b, 110c is schematically shown, each transmission and reception point serving its own cell 120a, 120b, 120c. Initially, the transmission and reception point 110a is serving UEs 300a, 300b in cell 120a, the transmission and reception point 110b is serving UEs 300c, 300d, 300e in cell 120b, and the transmission and reception point 110c is serving UEs 300f, 300g in cell 120c. From a purely traffic information point of view, this may be the best allocation of which transmission and reception point is to serve which UE. However, when it is also considered that UEs 300a:300g will participate in the over-the-air joint learning process, it may be better to have the transmission and reception point 110b serve all UEs 300a:300g. Thus, a handover to the transmission and reception point 110b may be initiated for the UE 300a, 300b, 300f, 300g.

[0113] Next, we will refer to Figure 6 A specific embodiment for determining the serving sending and receiving point 110a of the UE 300a participating in the over-the-air joint learning process is disclosed with reference to a signaling diagram of FIG. Here, it is assumed that a proxy is co-located with the UE 300a and the parameter server is co-located with the network node 200. Therefore, reference will be made to the UE 300a and the network node 200, rather than the proxy and the parameter server.

[0114] S301: An Over-the-Air Associated Learning (OtAFL) procedure is initiated between the UE 300a and the network node 200. Thereby, the UE 300a also indicates that it will participate in the Over-the-Air Associated Learning procedure with the parameter server. The UE 300a is initially served by an initial transmission and reception point.

[0115] S302: UE 300a provides data set importance information to the network node 200.

[0116] S303: The network node 200 provides configuration related to instructions for transmitting downlink reference signals to a group of transmission and reception points 110a:110M. The group of transmission and reception points 110a:110M defines candidate transmission and reception points 110a:110M for serving the UE 300a.

[0117] S304: The UE 300a receives, from the network node 200, configuration related to instructions for performing radio measurements on downlink reference signals transmitted from a set of transmission and reception points 110a: 110M.

[0118] S305: The UE 300a obtains a radio measurement value by measuring the downlink reference signal, and provides the radio measurement value to the network node 200 as signal quality information.

[0119] S306: The network node 200 obtains service information from the group of sending and receiving points 110a: 110M.

[0120] S307: The network node 200 determines which transmission and reception point (TRP) 110a:110M will serve the UE 300a when performing the air-joint learning process based on the obtained signal quality information and the obtained service information (and the fact that the UE 300a will participate in the air-joint learning process with the parameter server).

[0121] S308a, S308b: The network node 200 initiates handover of the UE 300a from the initial transmission and reception point to the transmission and reception point 110a selected to serve the UE 300a during the over-the-air joint learning process.

[0122] S309: Execute the air joint learning process.

[0123] S310a, S310b: The network node 200 initiates handover of the UE 300a from the selected initial transmission and reception point back to the initial transmission and reception point.

[0124] Next, non-limiting examples of scenarios in which the embodiments disclosed herein can be applied will be disclosed.

[0125] According to a first example, the computation task involves predicting the best secondary carrier frequency to be used by the UE 300a:300K. The data obtained locally by the UE 300k may then represent measurements of the serving carrier of the UE 300k. In this regard, the best secondary carrier frequency for the UE 300a:300K may be predicted based on measurement reports on its serving carrier. Thus, the reported secondary carrier frequency defines the result of the computation. To implement this technique, the UE 300a:300K may be trained by the network node 200, wherein each UE 300k takes as input the measurement reports on the serving carrier (and possibly other available reports, such as time advance, etc.) according to a machine learning model, and takes as output a prediction of whether the UE 300a:300K has coverage on the secondary carrier frequency.

[0126] According to a second example, the computational task involves compressing channel state information using an autoencoder, wherein the network node 200 implements a decoder of the autoencoder and each of the UEs 300a:300K implements a corresponding encoder of the autoencoder according to a machine learning model. The autoencoder can be viewed as a type of neural network for learning efficient data representations.

[0127] According to a third example, the computational task involves signal quality degradation prediction. The signal quality degradation prediction is based on the measurement of the radio link used by the UE 300a: 300K. In this respect, based on the received data, according to the computational results, in the report, the network node 200 can learn, for example, which signal quality measurement sequence (e.g., reference signal received power; RSRP) causes the signal quality to drop significantly. After training the machine learning model, for example, by using an iterative learning process, the network node 200 can provide the model to the UE 300a: 300K. The model can be provided to the UE 300a: 300K or other UE 300a: 300K that has participated in the training. Then, the UE 300a: 300K can apply the model to predict future signal quality values. Then, the signal quality prediction can be used in the context of any of the following items: initiating inter-frequency handover, setting handover and / or reselection parameters, changing the device scheduler priority so as to schedule the UE 300a: 300K when the expected signal quality is good. The data used to train such a model is located at the UE side, so future signal quality predictions can be efficiently learned using the iterative learning process disclosed in this article.

[0128] According to a fourth example, an iterative learning process involves mapping one or more geographic locations to a radio signal quality value or a bit rate value. This can be used by the network node 200 to understand potential network coverage holes in a cell and how the bit rate varies within the cell. However, the geographic location reported by the device may violate the privacy issues of the UE 300a:300K, so an iterative learning process can be used because it only includes the signaling of the parameter vector. As an illustrative example, assume that the UE 300a:300K has a data set mapped with the geographic location and the signal to interference plus noise ratio (SINR) according to the corresponding area (where the SINR takes a normalized value from 0 to 1). Then, the UE 300a:300K performs an iterative learning process with the network node 200 to include its data set information in the global machine learning model to be used by the network.

[0129] According to a fifth example, an iterative learning process involves using a network of cameras to monitor a particular area (public or private). Each camera is associated with its own UE 300a:300K. The network is configured to learn a global machine learning model that can be used to detect whether suspicious activities are ongoing in any video image captured by the camera. Suspicious activities may involve identifying unauthorized persons in a private area, identifying vehicles in the area, or monitoring the area for any criminal activity. However, for privacy reasons, these cameras are not allowed to share their raw image data. The global machine learning model can then be trained using an iterative learning process, in which the cameras only share model (gradient) updates based on the parameter vector, rather than their observations. In this case, the parameter vector can represent a prediction of the content in the image (e.g., parameters of a face recognition algorithm).

[0130] According to a sixth example, the iterative learning process involves a natural language processing application. The UE 300a:300K can learn a language model representing a machine learning model, for example for text prediction when typing on the UE 300a:300K (e.g., a mobile device or smart phone). By using an iterative learning process, only updates to the model need to be shared, but the original text data will not be shared.

[0131] Figure 7 The components of the network node 200 according to the embodiment are schematically shown in the form of a plurality of functional units. Fig.11 The processing circuit 210 may be provided by any combination of one or more of a suitable central processing unit (CPU), a multiprocessor, a microcontroller, a digital signal processor (DSP), etc., which may be implemented in accordance with software instructions stored in a storage medium (e.g., in the form of a storage medium 230). The processing circuit 210 may also be provided as at least one application specific integrated circuit (ASIC) or a field programmable gate array FPGA.

[0132] Specifically, the processing circuit 210 is configured to cause the network node 200 to perform a set of operations or steps as disclosed above. For example, the storage medium 230 may store the set of operations, and the processing circuit 210 may be configured to retrieve the set of operations from the storage medium 230 to cause the network node 200 to perform the set of operations. The set of operations may be provided as an executable instruction set. Therefore, the processing circuit 210 is thereby arranged to perform the method as disclosed herein.

[0133] The storage medium 230 may also include a persistent storage device, which may be, for example, any single memory or any combination of magnetic memory, optical memory, solid-state memory, or even remotely mounted memory.

[0134] The network node 200 may also include a communication interface 220 for communicating directly or indirectly with other entities, functions, nodes and devices (e.g., UE 300a:300K and transmission and reception points 110a:110M). Thus, the communication interface 220 may include one or more transmitters and receivers including analog components and digital components.

[0135] The processing circuit 210 controls the overall operation of the network node 200, for example, by sending data and control signals to the communication interface 220 and the storage medium 230, by receiving data and reports from the communication interface 220, and by obtaining data and instructions from the storage medium 230. Other components of the network node 200 and related functions are omitted so as not to obscure the concepts presented herein.

[0136] Figure 8 The components of the network node 200 according to the embodiment are schematically shown in the form of a plurality of functional modules. Figure 8 The network node 200 includes multiple functional modules; an obtaining module 210a configured to execute step S102, an obtaining module 210b configured to execute step S104, and a determining module 210f configured to execute step S112. Figure 8 The network node 200 may further include a plurality of optional functional modules, such as an obtaining module 210c configured to execute step S106, an obtaining module 210d configured to execute step S108, an obtaining module 210e configured to execute step S110, an initiating module 210g configured to execute step S114, and an initiating module 210h configured to execute step S116.

[0137] Typically, each functional module 210a:210h may be implemented in hardware or software. Preferably, one or more or all functional modules 210a:210h may be implemented by a processing circuit 210 that may cooperate with a communication interface 220 and / or a storage medium 230. Therefore, the processing circuit 210 may be arranged to obtain instructions provided by the functional modules 210a:210h from the storage medium 230 and execute these instructions, thereby performing any steps of the network node 200 as disclosed herein.

[0138] The network node 200 may be provided as a standalone device or as part of at least one further device. For example, the network node 200 may be arranged in a node of a radio access network or a node of a core network. Alternatively, the functionality of the network node 200 may be distributed between at least two devices or nodes. These at least two nodes or at least two devices may be part of the same network portion (such as a radio access network or a core network), or may be distributed between at least two such network portions. In general, instructions that need to be executed in real time may be executed in a device or node that is closer to the cell than instructions that do not need to be executed in real time. Thus, a first portion of the instructions executed by the network node 200 may be executed in a first device, and a second portion of the instructions executed by the network node 200 may be executed in a second device; the embodiments disclosed herein are not limited to any specific number of devices on which the instructions executed by the network node 200 may be executed. Therefore, the method according to the embodiments disclosed herein is suitable for execution by a network node 200 residing in a cloud computing environment. Therefore, although in Figure 7 A single processing circuit 210 is shown in FIG. 1 , but the processing circuit 210 may be distributed among multiple devices or nodes. The same applies to Figure 8 Functional modules 210a:210h and Fig.11 Computer program 1120a.

[0139] Fig. 9 The components of the UE 300a:300K according to the embodiment are schematically shown in the form of a plurality of functional units. Fig.11 The processing circuit 310 may be provided by any combination of one or more of a suitable central processing unit (CPU), a multiprocessor, a microcontroller, a digital signal processor (DSP), etc., which may be implemented in accordance with software instructions stored in a storage medium (e.g., in the form of a storage medium 330). The processing circuit 310 may also be provided as at least one application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

[0140] Specifically, the processing circuit 310 is configured to cause the UE 300a:300K to perform a set of operations or steps as disclosed above. For example, the storage medium 330 may store the set of operations, and the processing circuit 310 may be configured to obtain the set of operations from the storage medium 330 to cause the UE 300a:300K to perform the set of operations. The set of operations may be provided as an executable instruction set. Therefore, the processing circuit 310 is thereby arranged to perform the method as disclosed herein.

[0141] The storage medium 330 may also include a persistent storage device, which may be, for example, any single memory or any combination of magnetic memory, optical memory, solid-state memory, or even remotely mounted memory.

[0142] The UE 300a:300K may also include a communication interface 320 for communicating directly or indirectly with other entities, functions, nodes and devices (e.g., the network node 200 and the transmission and reception points 110a:110M). Thus, the communication interface 320 may include one or more transmitters and receivers including analog components and digital components.

[0143] The processing circuit 310 controls the overall operation of the UE 300a: 300K, for example, by sending data and control signals to the communication interface 320 and the storage medium 330, by receiving data and reports from the communication interface 320, and by retrieving data and instructions from the storage medium 330. Other components of the UE 300a: 300K and related functions are omitted so as not to obscure the concepts presented herein.

[0144] Fig.10 The components of the UE 300a:300K according to the embodiment are schematically shown in the form of multiple functional modules. Fig.10 The UE 300a:300K includes multiple functional modules: a providing module 310a configured to execute step S202, a providing module 310b configured to execute step S204, and an obtaining module 310d configured to execute step S208. Fig.10 The UE 300a:300K may further include a plurality of optional functional modules, such as a providing module 310c configured to execute step S206, a switching module 310e configured to execute step S210, and a switching module 310f configured to execute step S212.

[0145] In general, each functional module 310a:310f may be implemented in hardware or software. Preferably, one or more or all functional modules 310a:310f may be implemented by a processing circuit 310 that may cooperate with a communication interface 320 and / or a storage medium 330. Therefore, the processing circuit 310 may be arranged to obtain instructions provided by the functional module 310a:310f from the storage medium 330 and execute these instructions, thereby performing any steps of the UE 300a:300K as disclosed herein.

[0146] Fig.11An example of a computer program product 1110a, 1110b comprising a computer readable device 1130 is shown. On the computer readable device 1130, a computer program 1120a may be stored, which may cause the processing circuit 210 and entities and devices operatively coupled to the processing circuit 210 (e.g., communication interface 220 and storage medium 230) to perform methods according to the embodiments described herein. Therefore, the computer program 1120a and / or the computer program product 1110a may provide means for performing any steps of the network node 200 disclosed herein. On the computer readable device 1130, a computer program 1120b may be stored, which may cause the processing circuit 310 and entities and devices operatively coupled to the processing circuit 310 (e.g., communication interface 320 and storage medium 330) to perform methods according to the embodiments described herein. Therefore, the computer program 1120b and / or computer program product 1110b may provide means for performing any steps of the UE 300a: 300K disclosed herein.

[0147] exist Fig.11 In the example of , the computer program products 1110a, 1110b are shown as an optical disc, such as a CD (Compact Disc) or a DVD (Digital Versatile Disc) or a Blu-ray Disc. The computer program products 1110a, 1110b may also be embodied as a memory, such as a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), or an electrically erasable programmable read-only memory (EEPROM), and more specifically as a non-volatile storage medium of a device in an external memory, such as a USB (Universal Serial Bus) memory or a flash memory, such as a compact flash. Thus, although the computer programs 1120a, 1120b are schematically shown here as tracks on the depicted optical disc, the computer programs 1120a, 1120b may be stored in any manner suitable for the computer program products 1110a, 1110b.

[0148] Fig.124 is a schematic diagram showing a telecommunication network connected to a host computer 430 via an intermediate network 420 according to some embodiments. According to an embodiment, the communication system includes a telecommunication network 410 (e.g., a cellular network of the 3GPP type), which includes an access network 411 and a core network 414. The access network 411 includes a plurality of radio access network nodes 412a, 412b, 412c (e.g., NB, eNB, gNB (each corresponding to the network node 200) or other types of wireless access points), each of which defines a corresponding coverage area or cell 413a, 413b, 413c. Each radio access network node 412a, 412b, 412c can be connected to the core network 414 via a wired or wireless connection 415. A first UE 491 located in the coverage area 413c is configured to be wirelessly connected to or paged by the corresponding network node 412c. A second UE 492 in the coverage area 413a can be wirelessly connected to the corresponding network node 412a. Although multiple UEs 491, 492 are shown in this example, the disclosed embodiments are equally applicable to the case where only one UE is in the coverage area or only one UE is connected to the corresponding base station 412. UE 491, 492 corresponds to UE 300a:300K.

[0149] The telecommunications network 410 itself is connected to a host computer 430, which may be embodied in the hardware and / or software of a standalone server, a cloud-implemented server, a distributed server, or as a processing resource in a server cluster. The host computer 430 may be owned or under the control of a service provider, or may be operated by or on behalf of a service provider. The connections 421 and 422 between the telecommunications network 410 and the host computer 430 may extend directly from the core network 414 to the host computer 430, or may be made through an optional intermediate network 420. The intermediate network 420 may be one or a combination of more than one of a public, private, or carrier network; the intermediate network 420 (if present) may be a backbone network or the Internet; specifically, the intermediate network 420 may include two or more sub-networks (not shown).

[0150] Fig.12The communication system in the embodiment as a whole implements a connection between the connected UE 491, 492 and the host computer 430. The connection can be described as an over-the-top (OTT) connection 450. The host computer 430 and the connected UE 491, 492 are configured to transmit data and / or signaling via the OTT connection 450 using the access network 411, the core network 414, any intermediate network 420 and possible other intermediate infrastructure (not shown). The OTT connection 450 can be transparent in the sense that the participating communication devices through which the OTT connection 450 passes do not know the routing of the uplink and downlink communications. For example, the network node 412 may not be informed or may not need to be informed of the past routing of the incoming downlink communication with data originating from the host computer 430 to be forwarded (e.g., handed over) to the connected UE 491. Similarly, the network node 412 does not need to be aware of the future routing of the outgoing uplink communication originating from the UE 491 to the host computer 430.

[0151] Fig.13 is a schematic diagram showing a host computer communicating with a UE via a radio access network node over a partially wireless connection according to some embodiments. Fig.13 5. The example implementation of the UE, the radio access network node and the host computer according to the embodiment discussed in the previous paragraph is described. In the communication system 500, the host computer 510 includes hardware 515, which includes a communication interface 516, which is configured to establish and maintain a wired or wireless connection with the interface of different communication devices of the communication system 500. The host computer 510 also includes a processing circuit 518, which may have storage and / or processing capabilities. In particular, the processing circuit 518 may include one or more programmable processors, application-specific integrated circuits, field programmable gate arrays, or a combination of such devices (not shown) suitable for executing instructions. The host computer 510 also includes software 511, which is stored in or accessible by the host computer 510 and can be executed by the processing circuit 518. The software 511 includes a host application 512. The host application 512 is operable to provide services to a remote user (e.g., UE 530), and the UE 530 is connected via an OTT connection 550 terminated at the UE 530 and the host computer 510. UE 530 corresponds to UE 300a: 300K. In providing services to remote users, host application 512 may provide user data sent using OTT connection 550.

[0152] The communication system 500 also includes a radio access network node 520, which is arranged in the telecommunication system and includes hardware 525 that enables it to communicate with the host computer 510 and the UE 530. The radio access network node 520 corresponds to the network node 200. The hardware 525 may include: a communication interface 526 for establishing and maintaining a wired or wireless connection between interfaces of different communication devices of the communication system 500; and a radio interface 527 for establishing and maintaining communication with a network device located in a coverage area (in a region) served by the radio access network node 520. Fig.13 The communication interface 526 may be configured to facilitate a connection 560 with the host computer 510. The connection 560 may be direct, or it may be through a core network (eg, a wireless network) of the telecommunications system. Fig.13 520) and / or through one or more intermediate networks outside the telecommunications system. In the illustrated embodiment, the hardware 525 of the radio access network node 520 also includes a processing circuit 528, which may include one or more programmable processors, application specific integrated circuits, field programmable gate arrays, or a combination thereof (not shown) adapted to execute instructions. The radio access network node 520 also has software 521 stored internally or accessible via an external connection.

[0153] The communication system 500 also includes the UE 530 already mentioned. Its hardware 535 may include a radio interface 537, which is configured to establish and maintain a wireless connection 570 with a radio access network node that provides services for the coverage area currently located by the UE 530. The hardware 535 of the UE 530 also includes a processing circuit 538, which may include one or more programmable processors, application-specific integrated circuits, field programmable gate arrays, or a combination of such devices (not shown) suitable for executing instructions. The UE 530 also includes software 531, which is stored in or accessible by the UE 530 and can be executed by the processing circuit 538. The software 531 includes a client application 532. The client application 532 may be operable to provide services to human or non-human users via the UE 530 under the support of the host computer 510. In the host computer 510, the executing host application 512 can communicate with the executing client application 532 via the OTT connection 550, which is terminated at the UE 530 and the host computer 510. When providing services to users, the client application 532 can receive request data from the host application 512 and provide user data in response to the request data. The OTT connection 550 can transmit both the request data and the user data. The client application 532 can interact with the user to generate the user data it provides.

[0154] Notice, Fig.13 The host computer 510, the radio access network node 520 and the UE 530 shown can be respectively connected to Fig.12 The host computer 430, one of the network nodes 412a, 412b, 412c, and one of the UEs 491, 492 in the example embodiment are similar or identical. That is, the internal workings of these entities may be similar to or identical to those of FIG. Fig.13 As shown, and independently, the surrounding network topology can be Fig.12 network topology.

[0155] exist Fig.13 In the embodiment, an OTT connection 550 has been abstractly drawn to illustrate the communication between the host computer 510 and the UE 530 via the network node 520, but without explicitly mentioning any intermediate devices and the exact routing messages via these devices. The network infrastructure can determine the route, which can be configured to be hidden from the UE 530 or hidden from the service provider operating the host computer 510, or hidden from both. While the OTT connection 550 is active, the network infrastructure can also make decisions to dynamically change the route (e.g., based on load balancing considerations or reconfiguration of the network).

[0156] The wireless connection 570 between the UE 530 and the radio access network node 520 is consistent with the teachings of the embodiments described throughout this disclosure. One or more of the various embodiments improves the performance of an OTT service provided to the UE 530 using the OTT connection 550, in which the wireless connection 570 forms the final part. More specifically, the teachings of these embodiments can reduce interference due to the improved classification capabilities of airborne UEs that can cause significant interference.

[0157] A measurement process may be provided for monitoring data rate, latency, and other factors that are the object of improvement of one or more embodiments. There may also be an optional network function for reconfiguring the OTT connection 550 between the host computer 510 and the UE 530 in response to changes in the measurement results. The measurement process and / or network function for reconfiguring the OTT connection 550 may be implemented in software 511 and hardware 515 of the host computer 510 or in software 531 and hardware 535 of the UE 530 or in both. In an embodiment, a sensor (not shown) may be deployed in or associated with a communication device through which the OTT connection 550 passes; the sensor may participate in the measurement process by providing the value of the monitoring quantity exemplified above or providing the value of other physical quantities that the software 511, 531 can use to calculate or estimate the monitoring quantity. The reconfiguration of the OTT connection 550 may include message format, retransmission settings, preferred routing, etc.; the reconfiguration does not need to affect the network node 520, and the reconfiguration may be unknown or imperceptible to the radio access network node 520. Such processes and functions may be known and practiced in the art. In certain embodiments, the measurements may involve proprietary UE signaling that facilitates host computer 510 measurements of throughput, propagation time, latency, etc. The measurements may be implemented as follows: Software 511 and 531 enables sending messages (specifically, empty or "fake" messages) using OTT connection 550 while it monitors propagation time, errors, etc.

[0158] The inventive concept has mainly been described above with reference to a few embodiments. However, it is readily appreciated by a person skilled in the art that other embodiments than the ones disclosed above are equally possible within the scope of the inventive concept as defined by the appended patent claims.

Claims

1. A method for determining a serving sending and receiving point (110a:110M) of a user equipment (300a:300K) participating in an over-the-air joint learning process, the method being performed by a network node (200), the method comprising: Obtaining (S102) signal quality information of each user equipment (300a:300K) in a group of user equipment (300a:300K), wherein the signal quality information relates to the corresponding signal quality between each user equipment (300a:300K) and a pair of transmission and reception points (110a:110M) in a group of transmission and reception points (110a:110M); obtaining (S104) an indication from each user equipment (300a:300K), wherein the indication from a given user equipment (300a:300K) relates to whether the given user equipment (300a:300K) is to participate in the over-the-air federated learning process with a parameter server via one of the transmission and reception points (110a:110M); and Based on the obtained signal quality information of each user equipment (300a:300K) and the obtained indications from all the user equipment (300a:300K), it is determined (S112) which user equipment (300a:300K) is to be served by which transmission and reception point (110a:110M) when performing the over-the-air joint learning process.

2. The method according to claim 1, wherein: Each of the transmission and reception points (110a: 110M) is configured to serve its own cell.

3. The method according to claim 2, wherein: The cells of at least two different transmission and reception points (110a:110M) have mutually different carrier frequencies.

4. The method according to claim 1, wherein: The signal quality information is obtained from radio measurements performed by the user equipment (300a:300K) on a downlink reference signal sent by the transmission and reception point (110a:110M) and / or from radio measurements performed by the transmission and reception point (110a:110M) on an uplink reference signal sent by the user equipment (300a:300K).

5. The method according to claim 1, wherein: The signal quality information is obtained from a signal quality prediction made by the network node (200) or from a signal quality prediction provided to the network node (200).

6. The method according to claim 1, wherein: The method further comprises: Obtaining (S106) traffic information, wherein the traffic information relates to any one of the following items: the number of user equipments (300a:300K) currently served by each transmission and reception point (110a:110M), or a prediction of the number of user equipments (300a:300K) to be served by each transmission and reception point (110a:110M), the traffic load of each transmission and reception point (110a:110M), and available resources at each transmission and reception point (110a:110M) for participating in the over-the-air joint learning process, and Which user equipment (300a: 300K) is to be served by which sending and receiving point (110a: 110M) is also determined based on the service information.

7. The method according to claim 1, wherein: The method further comprises: obtaining (S108) accuracy information, wherein the accuracy information relates to an accuracy level of a channel estimation performed by each of the transmitting and receiving points (110a:110M) for the user equipment (300a:300K), and Which user equipment (300a: 300K) is to be served by which transmission and reception point (110a: 110M) is also determined based on the accuracy information.

8. The method according to claim 1, wherein: Each user equipment (300a:300K) to participate in the over-the-air federated learning process is configured to train on a local dataset as part of the over-the-air federated learning process, wherein the method further comprises: obtaining (S110) data set importance information from the user device (300a:300K), wherein the data set importance information relates to a uniqueness level of each local data set, and Which user equipment (300a: 300K) is to be served by which transmission and reception point (110a: 110M) is also determined based on the data set importance information.

9. The method according to claim 1, wherein: Prior to determining which user equipment (300a:300K) is to be served by which transmission and reception point (110a:110M), the user equipment (300a:300K) is served by an initial transmission and reception point (110a:110M) in the set of transmission and reception points (110a:110M), and Which user equipment (300a:300K) is to be served by which transmission and reception point (110a:110M) is determined according to a switching criterion for reducing the number of switchings from the initial transmission and reception point (110a:110M).

10. The method according to claim 1, wherein: Each of the transmitting and receiving points (110a:110M) has processing capabilities for participating in the over-the-air joint learning process, and Which user equipment (300a:300K) is to be served by which transmission and reception point (110a:110M) is determined based on the processing capability of each transmission and reception point (110a:110M).

11. The method according to claim 1, wherein: Which user equipment (300a:300K) is to be served by which transmission and reception point (110a:110M) is determined according to selection criteria for reducing the total number of transmission and reception points (110a:110M) required to serve the user equipment (300a:300K).

12. The method according to claim 1, wherein: Each user device (300a:300K) to participate in the over-the-air federated learning process is configured to train a local data set as part of the over-the-air federated learning process, wherein each local data set has a corresponding uniqueness level, wherein the signal quality reflects a path gain between the user device (300a:300K) and the transmission and reception point (110a:110M), and Among them, the user equipment (300a:300K) with the local data set having the highest uniqueness level is determined to be served by the transmission and reception point (110a:110M) with the highest path gain.

13. The method according to claim 1, wherein: Each user device (300a:300K) to participate in the over-the-air federated learning process is configured to train a local data set as part of the over-the-air federated learning process, wherein each local data set has a corresponding uniqueness level, wherein the signal quality reflects a path gain between the user device (300a:300K) and the transmission and reception point (110a:110M), and Among them, the user equipment (300a:300K) with the highest requirement for energy efficiency is determined to be served by the transmission and reception point (110a:110M) with the highest path gain.

14. The method according to claim 1, wherein: Prior to determining which user equipment (300a:300K) is to be served by which transmission and reception point (110a:110M), the user equipment (300a:300K) is served by an initial transmission and reception point (110a:110M) in the set of transmission and reception points (110a:110M), and wherein the method further comprises: A first handover is initiated (S114) for each user equipment (300a:300K) determined to be served by another transmission and reception point (110a:110M) other than its initial transmission and reception point (110a).

15. The method according to claim 14, wherein: The method further comprises: When the over-the-air joint learning process has been terminated, a second handover back to the initial transmission and reception point (110a:110M) is initiated (S116) for each user equipment (300a:300K) for which the first handover was performed.

16. A method for assisting in determining a serving transmission and reception point (110a) of a user equipment (300a:300K) participating in an over-the-air joint learning process, the method being performed by the user equipment (300a:300K), the method comprising: providing (S202) to the network node (200) radio measurements performed by the user equipment (300a:300K) on downlink reference signals transmitted by the transmission and reception point (110a:110M); providing (S204) to the network node (200) an indication that the user equipment (300a:300K) will participate in the over-the-air joint learning process with a parameter server via one of the transmission and reception points (110a:110M); and An indication is obtained (S208) from the network node (200) as to which transmission and reception point (110a:110M) will serve the user equipment (300a:300K) when performing the over-the-air joint learning process.

17. The method according to claim 16, wherein: The user equipment (300a:300K) is configured to train on a local dataset as part of the over-the-air federated learning process, and wherein the method further comprises: Providing (S206) data set importance information to the network node (200), wherein the data set importance information relates to a uniqueness level of the local data set.

18. The method according to claim 16, wherein: Prior to obtaining an indication of which transmission and reception point (110a:110M) will serve the user equipment (300a:300K), the user equipment (300a:300K) is served by an initial transmission and reception point (110a), and wherein the method further comprises: While performing the over-the-air joint learning process, a first handover is performed (S210) for a user equipment (300a:300K) to be served by another transmission and reception point (110a:110M) other than its initial transmission and reception point (110a).

19. The method according to claim 18, wherein: The method further comprises: When the over-the-air joint learning process has been terminated, a second switch back to the initial transmission and reception point (110a) is performed (S212).

20. A network node (200) for determining a serving sending and receiving point (110a:110M) of a user equipment (300a:300K) participating in an over-the-air joint learning process, the network node (200) comprising a processing circuit (210) configured to cause the network node (200) to: Signal quality information of each user equipment (300a:300K) in a group of user equipment (300a:300K) is obtained, wherein: The signal quality information relates to the respective signal quality between each user equipment (300a:300K) and a pair of transmission and reception points (110a:110M) in a set of transmission and reception points (110a:110M); obtaining an indication from each user equipment (300a:300K), wherein the indication from a given user equipment (300a:300K) relates to whether the given user equipment (300a:300K) is to participate in the over-the-air federated learning process with a parameter server via one of the transmission and reception points (110a:110M); and Based on the obtained signal quality information of each user equipment (300a:300K) and the obtained indications from all the user equipment (300a:300K), it is determined which user equipment (300a:300K) is to be served by which transmission and reception point (110a:110M) when performing the over-the-air joint learning process.

21. A network node (200) for determining a service sending and receiving point (110a:110M) of a user equipment (300a:300K) participating in an over-the-air joint learning process, the network node (200) comprising: An obtaining module (210a) configured to obtain signal quality information of each user equipment (300a:300K) in a group of user equipment (300a:300K), wherein the signal quality information relates to the corresponding signal quality between each user equipment (300a:300K) and a pair of transmission and reception points (110a:110M) in a group of transmission and reception points (110a:110M); an obtaining module (210b) configured to obtain an indication from each user equipment (300a:300K), wherein the indication from a given user equipment (300a:300K) relates to whether the given user equipment (300a:300K) is to participate in the over-the-air joint learning process with a parameter server via one of the transmission and reception points (110a:110M); and The determination module (210f) is configured to determine which user equipment (300a:300K) is to be served by which transmission and reception point (110a:110M) when performing the over-the-air joint learning process based on the obtained signal quality information of each user equipment (300a:300K) and the obtained instructions from all the user equipment (300a:300K).

22. The network node (200) according to claim 20 or 21, further configured to perform the method according to any one of claims 2 to 15.

23. A user equipment (300a:300K) for assisting in determining a serving transmission and reception point (110a) of a user equipment (300a:300K) to participate in an over-the-air joint learning process, the user equipment (300a:300K) comprising a processing circuit (310) configured to cause the user equipment (300a:300K): providing to the network node (200) radio measurements performed by the user equipment (300a:300K) on downlink reference signals transmitted by the transmission and reception points (110a:110M); providing an indication to the network node (200) that the user equipment (300a:300K) will participate in the over-the-air joint learning process with a parameter server via one of the transmission and reception points (110a:110M); and Obtaining an indication from the network node (200) as to which transmission and reception point (110a:110M) will serve the user equipment (300a:300K) when performing the over-the-air joint learning process.

24. A user equipment (300a:300K) for assisting in determining a serving sending and receiving point (110a) of a user equipment (300a:300K) to participate in an over-the-air joint learning process, the user equipment (300a:300K) comprising: A providing module (310a) configured to provide to the network node (200) radio measurements performed by the user equipment (300a:300K) on downlink reference signals transmitted by the transmission and reception points (110a:110M); a providing module (310b) configured to provide an indication to the network node (200) that the user equipment (300a:300K) will participate in the over-the-air joint learning process with a parameter server via one of the transmission and reception points (110a:110M); as well as The obtaining module (310d) is configured to obtain an indication from the network node (200) as to which transmission and reception point (110a:110M) will provide service for the user equipment (300a:300K) when performing the over-the-air joint learning process.

25. The user equipment (300a:300K) according to claim 23 or 24, further configured to perform the method according to any one of claims 17 to 19.

26. A computer program (1120a) for determining serving sending and receiving points (110a:110M) of a user equipment (300a:300K) participating in an over-the-air federated learning process, the computer program comprising computer code which, when executed on a processing circuit (210) of a network node (200), causes the network node (200) to: Obtaining (S102) signal quality information of each user equipment (300a:300K) in a group of user equipment (300a:300K), wherein: The signal quality information relates to the respective signal quality between each user equipment (300a:300K) and a pair of transmission and reception points (110a:110M) in a set of transmission and reception points (110a:110M); obtaining (S104) an indication from each user equipment (300a:300K), wherein the indication from a given user equipment (300a:300K) relates to whether the given user equipment (300a:300K) is to participate in the over-the-air federated learning process with a parameter server via one of the transmission and reception points (110a:110M); as well as Based on the obtained signal quality information of each user equipment (300a:300K) and the obtained indications from all the user equipment (300a:300K), it is determined (S112) which user equipment (300a:300K) is to be served by which transmission and reception point (110a:110M) when performing the over-the-air joint learning process.

27. A computer program (1120b) for assisting in determining serving sending and receiving points (110a) of a user equipment (300a:300K) participating in an over-the-air federated learning process, the computer program comprising computer code which, when executed on a processing circuit (310) of the user equipment (300a:300K), causes the user equipment (300a:300K) to: providing (S202) to the network node (200) radio measurements performed by the user equipment (300a:300K) on downlink reference signals transmitted by the transmission and reception point (110a:110M); providing (S204) to the network node (200) an indication that the user equipment (300a:300K) will participate in the over-the-air joint learning process with a parameter server via one of the transmission and reception points (110a:110M); and An indication is obtained (S208) from the network node (200) as to which transmission and reception point (110a:110M) will serve the user equipment (300a:300K) when performing the over-the-air joint learning process.

28. A computer program product (1110a, 1110b) comprising a computer program (1120a, 1120b) according to at least one of claims 26 and 27 and a computer readable storage medium (1130) on which the computer program is stored.

Images