Method and apparatus for a radio map construction in a wireless communication system

Abstract

The disclosure relates to a 5G or 6G communication system for supporting higher data rates. The disclosure provides a method performed by a base station in a wireless communication system. The method includes receiving, from a plurality of terminals, information on a data set including information on a location of a terminal and information on a channel quality corresponding to the location; determining a target terminal for a training among the plurality of terminals based on the information on the data set, wherein the training is associated with a prediction of a radio map; transmitting, to the target terminal, first information including a parameter related to a network model which is used for the prediction; receiving, from the target terminal, second information including an updated parameter corresponding to a result of the training; and updating the network model based on the second information.

Description

METHOD AND APPARATUS FOR A RADIO MAP CONSTRUCTION IN A WIRELESS COMMUNICATION SYSTEM

[0001] The disclosure relates to operations of a terminal and a base station in a wireless communication system. In particular, the disclosure relates to a method and an apparatus for a radio map construction.

[0002] Considering the development of wireless communication from generation to generation, the technologies have been developed mainly for services targeting humans, such as voice calls, multimedia services, and data services. Following the commercialization of 5th-generation (5G) communication systems, it is expected that the number of connected devices will exponentially grow. Increasingly, these will be connected to communication networks. Examples of connected things may include vehicles, robots, drones, home appliances, displays, smart sensors connected to various infrastructures, construction machines, and factory equipment. Mobile devices are expected to evolve in various form-factors, such as augmented reality glasses, virtual reality headsets, and hologram devices. In order to provide various services by connecting hundreds of billions of devices and things in the 6th-generation (6G) era, there have been ongoing efforts to develop improved 6G communication systems. For these reasons, 6G communication systems are referred to as beyond-5G systems.

[0003] 6G communication systems, which are expected to be commercialized around 2030, will have a peak data rate of tera (1,000 giga)-level bps and a radio latency less than 100μsec, and thus will be 50 times as fast as 5G communication systems and have the 1 / 10 radio latency thereof.

[0004] In order to accomplish such a high data rate and an ultra-low latency, it has been considered to implement 6G communication systems in a terahertz band (for example, 95GHz to 3THz bands). It is expected that, due to severer path loss and atmospheric absorption in the terahertz bands than those in millimeter wave bands introduced in 5G, technologies capable of securing the signal transmission distance (that is, coverage) will become more crucial. It is necessary to develop, as major technologies for securing the coverage, radio frequency (RF) elements, antennas, novel waveforms having a better coverage than orthogonal frequency division multiplexing (OFDM), beamforming and massive multiple input multiple output (MIMO), full dimensional MIMO (FD-MIMO), array antennas, and multiantenna transmission technologies such as large-scale antennas. In addition, there has been ongoing discussion on new technologies for improving the coverage of terahertz-band signals, such as metamaterial-based lenses and antennas, orbital angular momentum (OAM), and reconfigurable intelligent surface (RIS).

[0005] Moreover, in order to improve the spectral efficiency and the overall network performances, the following technologies have been developed for 6G communication systems: a full-duplex technology for enabling an uplink transmission and a downlink transmission to simultaneously use the same frequency resource at the same time; a network technology for utilizing satellites, high-altitude platform stations (HAPS), and the like in an integrated manner; an improved network structure for supporting mobile base stations and the like and enabling network operation optimization and automation and the like; a dynamic spectrum sharing technology via collision avoidance based on a prediction of spectrum usage; an use of artificial intelligence (AI) in wireless communication for improvement of overall network operation by utilizing AI from a designing phase for developing 6G and internalizing end-to-end AI support functions; and a next-generation distributed computing technology for overcoming the limit of user equipment (UE) computing ability through reachable super-high-performance communication and computing resources (such as mobile edge computing (MEC), clouds, and the like) over the network. In addition, through designing new protocols to be used in 6G communication systems, developing mechanisms for implementing a hardware-based security environment and safe use of data, and developing technologies for maintaining privacy, attempts to strengthen the connectivity between devices, optimize the network, promote softwarization of network entities, and increase the openness of wireless communications are continuing.

[0006] It is expected that research and development of 6G communication systems in hyper-connectivity, including person to machine (P2M) as well as machine to machine (M2M), will allow the next hyper-connected experience. Particularly, it is expected that services such as truly immersive extended reality (XR), high-fidelity mobile hologram, and digital replica could be provided through 6G communication systems. In addition, services such as remote surgery for security and reliability enhancement, industrial automation, and emergency response will be provided through the 6G communication system such that the technologies could be applied in various fields such as industry, medical care, automobiles, and home appliances.

[0007] Radio maps are key tools for wireless communication systems to satisfy 6G potential scenarios such as smart cities and smart factories. However, the radio maps constructed by the existing methods have high overhead.

[0008] The disclosure relates to operations of a terminal and a base station in a wireless communication system. In particular, the disclosure relates to a method and an apparatus for a radio map construction

[0009] Accordingly, an aspect of the disclosure is to provide a method and an apparatus for reducing the overhead of radio map construction with required precision and speed up the convergence of model.

[0010] In accordance with one aspect of the embodiments of the disclosure, a method performed by a base station in a wireless communication system is provided. The method includes receiving, from a plurality of terminals, information on a data set including information on a location of a terminal and information on a channel quality corresponding to the location; determining a target terminal for a training among the plurality of terminals based on the information on the data set, wherein the training is associated with a prediction of a radio map; transmitting, to the target terminal, first information including a parameter related to a network model which is used for the prediction; receiving, from the target terminal, second information including an updated parameter corresponding to a result of the training; and updating the network model based on the second information.

[0011] In accordance with one aspect of the embodiments of the disclosure, a method performed by a terminal in a wireless communication system is provided. The method includes transmitting, to a base station, information on a data set including information on a location of the terminal and information on a channel quality corresponding to the location; receiving, from the base station, first information including a parameter related to a network model associated with a prediction of a radio map, in case that the terminal is a target terminal which performs a training associated with the predication; performing the training based on the first information; and transmitting, to the base station, second information including an updated parameter corresponds to a result of the training, wherein the second information is used for updating the network model.

[0012] In accordance with one aspect of the embodiments of the disclosure, a base station in a wireless communication system is provided. The base station includes a transceiver; memory storing one or more programs; and one or more processors communicatively coupled to the transceiver and the memory, wherein the one or more programs include computer-executable instructions that, when executed by the one or more processors individually or collectively, cause the base station to: receive, from a plurality of terminals, information on a data set including information on a location of a terminal and information on a channel quality corresponding to the location, determine a target terminal for a training among the plurality of terminals based on the information on the data set, wherein the training is associated with a prediction of a radio map, transmit, to the target terminal, first information including a parameter related to a network model which is used for the prediction, receive, from the target terminal, second information including an updated parameter corresponding to a result of the training, and update the network model based on the second information.

[0013] In accordance with one aspect of the embodiments of the disclosure, a terminal in a wireless communication system is provided. The terminal includes a transceiver; memory storing one or more programs; and one or more processors communicatively coupled to the transceiver and the memory, wherein the one or more programs wherein the one or more programs include computer-executable instructions that, when executed by the one or more processors individually or collectively, cause the terminal to: transmit, to a base station, information on a data set including information on a location of the terminal and information on a channel quality corresponding to the location, receive, from the base station, first information including a parameter related to a network model associated with a prediction of a radio map, in case that the terminal is a target terminal which performs a training associated with the predication, perform the training based on the first information, and transmit, to the base station, second information including an updated parameter corresponds to a result of the training, wherein the second information is used for updating the network model.

[0014] In accordance with one aspect of the embodiments of the disclosure, there is provided a method executed by a wireless access network device in a communication system, including steps of:

[0015] receiving first information transmitted by at least one terminal device, the first information including position related information and / or wireless channel parameter related information of the terminal device;

[0016] determining at least one target terminal device from the at least one terminal device based on the first information transmitted by the at least one terminal device;

[0017] transmitting the first model parameter information to the at least one target terminal device, receiving second model parameter information transmitted by the at least one target terminal device, the second model parameter being obtained by training based on the first model parameter; and

[0018] fusing the second model parameter transmitted by the at least one target terminal device to obtain the updated first model parameter.

[0019] Optionally, the determining at least one target terminal device from the at least one terminal device based on the first information transmitted by the at least one terminal device includes:

[0020] determining characteristic information based on the first information transmitted by the at least one terminal device; and

[0021] determining at least one target terminal device from the at least one terminal device based on the characteristic information;

[0022] wherein the characteristic information includes at least one of the following:

[0023] the richness of a wireless channel parameter corresponding to each terminal device, the richness being related to a data volume and / or data fluctuation range; and

[0024] scarcity related information of the wireless channel parameter corresponding to each terminal device in at least one wireless channel parameter interval, the scarcity related information being related to the data volume in the at least one wireless channel parameter interval.

[0025] Optionally, the characteristic information further includes:

[0026] whether the wireless channel parameter corresponding to each terminal device includes a target wireless channel parameter, the target wireless channel parameter being related to at least one of the frequency of occurrence of each wireless channel parameter value, the variance of the wireless channel parameter value and the mean value of the wireless channel parameter value.

[0027] Optionally, the first information includes at least one of the following information and / or the richness of at least one of the following information, or the information corresponding to frequency bands and / or time units and / or different wireless access network devices, includes at least one of the following:

[0028] the data volume corresponding to the wireless channel parameter;

[0029] the difference between the maximum longitude and the minimum longitude among longitudes determined based on at least one piece of position information;

[0030] the difference between the maximum latitude and the minimum latitude among latitudes determined based on at least one piece of position information;

[0031] the difference between the maximum altitude and the minimum altitude among altitudes determined based on at least one piece of position information;

[0032] the difference between the maximum wireless channel parameter value and the minimum wireless channel parameter value among wireless channel parameter values corresponding to at least one piece of position information;

[0033] a first preset number of maximum distances among the respective distances determined based on each piece of position information;

[0034] a second preset number of minimum distances among the respective distances determined based on each piece of position information;

[0035] a position range determined based on each piece of position information;

[0036] a mean value of each piece of position information;

[0037] a mean value of each wireless channel parameter value;

[0038] a data volume of at least one wireless channel parameter interval;

[0039] a standard deviation of each piece of position information;

[0040] a standard deviation of each wireless channel parameter value; and

[0041] an average value of standard deviations of wireless channel parameters corresponding to different wireless access network devices.

[0042] Optionally, determining, based on the first information transmitted by the at least one terminal device, the richness of the wireless channel parameter corresponding to the terminal device includes at least one of the following ways:

[0043] for each terminal device, based on at least one of the data volume of the wireless channel parameter, the fluctuation range of the position and the fluctuation range of the wireless channel parameter corresponding to this terminal device, as well as the data volume of wireless channel parameters, the fluctuation range of positions and the fluctuation range of wireless channel parameters corresponding to all terminal devices, determining the richness of the wireless channel parameter corresponding to this terminal device; and

[0044] based on the first information and a corresponding threshold, determining the richness of the wireless channel parameter corresponding to this terminal device.

[0045] Optionally, determining, based on the first information transmitted by the at least one terminal device, that the wireless channel parameter corresponding to the terminal device includes a target wireless channel parameter includes:

[0046] if the variance of the wireless channel parameter value and / or the mean value of the wireless channel parameter value is greater than the first threshold and the frequency of occurrence of at least one wireless channel parameter value is less than a second threshold, determining that the wireless channel parameter corresponding to the terminal device includes a target wireless channel parameter.

[0047] Optionally, determining, based on the first information transmitted by the at least one terminal device, scarcity related information of the wireless channel parameter corresponding to the terminal device in at least one wireless channel parameter interval includes:

[0048] determining, based on the first information transmitted by the at least one terminal device, whether there is a target interval in at least one wireless channel parameter interval, the data volume corresponding to the wireless channel parameter of the at least one terminal device in the target interval satisfying a preset condition; and

[0049] if so, for each terminal device, the data volume in the target interval is used as scarcity related information for the wireless channel parameter corresponding to the terminal device in the at least one wireless channel parameter interval.

[0050] Optionally, the determining, based on the first information transmitted by the at least one terminal device, whether there is a target interval in at least one wireless channel parameter interval includes:

[0051] determining, from the first information transmitted by the at least one terminal device, a proportion of the data volume separately corresponding to the at least one wireless channel parameter interval in the total data volume;

[0052] for each wireless channel parameter interval in the at least one wireless channel parameter interval, determining the difference between the corresponding proportion and a predetermined proportion of this wireless channel parameter interval; and

[0053] determining the wireless channel parameter interval corresponding to the maximum difference as a target interval.

[0054] Optionally, determining at least one target terminal device from the at least one terminal device based on the characteristic information includes at least one of the following ways:

[0055] based on a channel state between the wireless access network device and the at least one terminal device and the characteristic information of each terminal device, determining a target terminal device from the at least one terminal device through a reinforcement learning model; and

[0056] determining a priority separately corresponding to the at least one terminal device based on the characteristic information of each terminal device, and determining a target terminal device from the at least one terminal device based on the priority.

[0057] Optionally, determining a priority separately corresponding to at least one terminal device based on the characteristic information of each terminal device includes:

[0058] based on a first weight separately corresponding to at least one piece of characteristic information of each terminal device, fusing at least one piece of characteristic information of each terminal device to obtain the priority separately corresponding to at least one terminal device.

[0059] Optionally, fusing the second model parameter transmitted by at least one target terminal device includes:

[0060] receiving a distribution characteristic of the wireless channel parameter;

[0061] grouping at least one target terminal device based on the distribution characteristic of the wireless channel parameter, and determining a second weight of at least one group of target terminal devices;

[0062] adjusting the second weight based on the similarity between the groups of target terminal devices and / or the model training precision of the second model parameter to obtain a third weight of at least one group of target terminal devices, and / or adjusting, based on the model training precision fed back by target terminal devices, the second weight of at least one target terminal device in at least one group of target terminal devices to obtain a fourth weight of at least one group of target terminal device; and

[0063] for the second model parameter information transmitted by each group of target terminal devices, fusing the second model parameter based on the third weight and / or the fourth weight.

[0064] Optionally, the wireless channel parameter includes at least one of the following:

[0065] a reference signal received power (RSRP) value;

[0066] a received signal strength (RSS) value;

[0067] a power spectral density (PSD) value;

[0068] a signal to interference plus noise ratio (SINR);

[0069] channel state information (CSI); and

[0070] reference signal receiving quality (RSRQ).

[0071] Optionally, the wireless channel parameter further includes at least one of the following:

[0072] frequency bands; and

[0073] time units.

[0074] Optionally, the transmitting first model parameter information to the at least one target terminal device includes:

[0075] determining a first element bit number reduction degree of the first model parameter, reducing the element bit number of the first model parameter based on the first element bit number reduction degree to obtain compressed first model parameter information, and transmitting the compressed first model parameter information to the at least one target terminal device.

[0076] Optionally, the receiving second model parameter information transmitted by the at least one target terminal device includes:

[0077] receiving the compressed second model parameter information transmitted by the at least one target terminal device, the compressed second model parameter information being obtained by compressing the second model parameter information based on the uplink data rate of the at least one target terminal device and / or the model training precision of the second model parameter, the uplink data rate being determined based on channel environment information between the at least one target terminal device and the wireless access network device.

[0078] Optionally, the compressed second model parameter information is obtained by compressing the second model parameter information based on the uplink data rate of the at least one target terminal device and / or the model training precision of the second model parameter includes at least one of the following situations:

[0079] the compressed second model parameter information is obtained by determining an element compression proportion of the second model parameter based on the uplink data rate and determining elements of the element compression proportion in the second model parameter;

[0080] the compressed second model parameter information is obtained by determining a second element bit number reduction degree of the second model parameter based on the uplink data rate and reducing the element bit number of the second model parameter based on the second element bit number reduction degree; and

[0081] the compressed second model parameter information is obtained by determining a model parameter cluster to be currently transmitted in the second model parameter based on the model training precision and determining a gradient to be currently transmitted in the model parameter cluster to be currently transmitted based on the model training precision.

[0082] In accordance with another aspect of embodiments of the disclosure, a method executed by a terminal device in a communication system is provided, including steps of:

[0083] transmitting first information to a wireless access network device, the first information including position related information and / or wireless channel parameter related information of the terminal device;

[0084] receiving first model parameter information transmitted by the wireless access network device;

[0085] in a case of being determined as a target terminal device by the wireless access network device based on the first information, training the first model parameter to obtain a second model parameter; and

[0086] transmitting the second model parameter information to the wireless access network device.

[0087] Optionally, being determined as a target terminal device by the wireless access network device based on the first information includes:

[0088] being determined as a target terminal device by the wireless access network device based on characteristic information, the characteristic information is determined based on the first information;

[0089] wherein the characteristic information includes at least one of the following:

[0090] the richness of a wireless channel parameter corresponding to the terminal device, the richness being related to a data volume and / or data fluctuation range; and

[0091] scarcity related information of the wireless channel parameter corresponding to the terminal device in at least one wireless channel parameter interval, the scarcity related information being related to the data volume in the at least one wireless channel parameter interval.

[0092] Optionally, the characteristic information further includes:

[0093] whether the wireless channel parameter corresponding to each terminal device comprises a target wireless channel parameter, the target wireless channel parameter being related to at least one of the frequency of occurrence of each wireless channel parameter value, the variance of the wireless channel parameter value and the mean value of the wireless channel parameter value.

[0094] Optionally, the first information includes at least one of the following information:

[0095] the data volume corresponding to the wireless channel parameter;

[0096] the difference between the maximum longitude and the minimum longitude among longitudes determined based on at least one piece of position information;

[0097] the difference between the maximum latitude and the minimum latitude among latitudes determined based on at least one piece of position information;

[0098] the difference between the maximum altitude and the minimum altitude among altitudes determined based on at least one piece of position information;

[0099] the difference between the maximum wireless channel parameter value and the minimum wireless channel parameter value among wireless channel parameter values corresponding to at least one piece of position information;

[0100] a first preset number of maximum distances among the respective distances determined based on each piece of position information;

[0101] a second preset number of minimum distances among the respective distances determined based on each piece of position information;

[0102] a position range determined based on each piece of position information;

[0103] a mean value of each piece of position information;

[0104] a mean value of each wireless channel parameter value;

[0105] a data volume of at least one wireless channel parameter interval;

[0106] a standard deviation of each piece of position information;

[0107] a standard deviation of each wireless channel parameter value; and

[0108] an average value of standard deviations of wireless channel parameters corresponding to different wireless access network devices.

[0109] Optionally, transmitting the second model parameter information to the wireless access network device includes:

[0110] determining, based on channel environment information between the terminal device and the wireless access network device, an uplink data rate of the wireless access network device;

[0111] determining model training precision of the second model parameter;

[0112] compressing the second model parameter information based on the uplink data rate and / or model training precision; and

[0113] transmitting the compressed second model parameter information to the wireless access network device.

[0114] Optionally, compressing the second model parameter information based on the uplink data rate and / or model training precision includes at least one of the following ways:

[0115] determining an element compression proportion of the second model parameter based on the uplink data rate, and determining elements of the element compression proportion in the second model parameter, to obtain the compressed second model parameter information;

[0116] determining an element bit number reduction degree of the second model parameter based on the uplink data rate, and reducing the element bit number of the second model parameter based on the second element bit number reduction degree, to obtain the compressed second model parameter information; and

[0117] determining a model parameter cluster to be currently transmitted in the second model parameter based on the model training precision, and determining a gradient to be currently transmitted in the model parameter cluster to be currently transmitted based on the model training precision, to obtain the compressed second model parameter information.

[0118] Optionally, the richness of the wireless channel parameter corresponding to the terminal device is determined in at least one of the following ways:

[0119] the richness of the wireless channel parameter corresponding to the terminal device is determined based on at least one of the data volume of the wireless channel parameter, the fluctuation range of the position and the fluctuation range of the wireless channel parameter corresponding to the terminal device, as well as at least one of the data volume of wireless channel parameters, the fluctuation range of positions and the fluctuation range of wireless channel parameters corresponding to all terminal devices; and

[0120] the richness of the wireless channel parameter corresponding to this terminal device is determined based on the first information and the corresponding threshold.

[0121] Optionally, the wireless channel parameter corresponding to the terminal device including a target wireless channel parameter is determined based on a situation where the variance of the wireless channel parameter value and / or the mean value of the wireless channel parameter value is greater than a first threshold and the frequency of occurrence of at least one wireless channel parameter value is less than a second threshold.

[0122] Optionally, the scarcity related information of the wireless channel parameter corresponding to the terminal device in at least one wireless channel parameter interval is determined based on the data volume of the terminal device in a target interval in a case where there is a target interval in at least one wireless channel parameter interval based on first information sent by at least one terminal device, and the data volume corresponding to the wireless channel parameter of at least one terminal device in the target interval satisfies a preset condition.

[0123] Optionally, it is determined based on the wireless channel parameter interval corresponding to the maximum difference that there is a target interval in the at least one wireless channel parameter interval; the difference corresponding to each wireless channel parameter interval in the at least one wireless channel parameter interval is a difference between the proportion of the data volume corresponding to the wireless channel parameter interval in the total data volume and a predetermined proportion of this wireless channel parameter interval; and, the proportion of the data volume corresponding to the wireless channel parameter interval in the total data volume is determined based on the first information.

[0124] Optionally, the target terminal device is determined in at least one of the following ways:

[0125] the target terminal device is determined from at least one terminal device through a reinforcement learning model based on the channel state between the wireless access network device and the at least one terminal device and the characteristic information of each terminal device; and

[0126] the priority separately corresponding to the at least one terminal device is determined based on the characteristic information of each terminal device, and the target terminal device is determined from the at least one terminal device based on the priority.

[0127] Optionally, the priority separately corresponding to the at least one terminal device is obtained by fusing at least one piece of characteristic information corresponding to each terminal device based on the first weight separately corresponding to at least one piece of characteristic information of each terminal device.

[0128] Optionally, after the second model parameter information is transmitted to the wireless access network device, the second model parameter information is fused based on a third weight and / or a fourth weight, wherein the third weight is obtained by adjusting the second weight based on the similarity between each group of target terminal devices and / or the model training precision of the second model parameter; the fourth weight is obtained by adjusting the second weight of at least one target terminal device in at least one group of target terminal devices based on the model training precision fed back by the target terminal device; and, the second weight is the weight of at least one group of target terminal devices determined after grouping at least one target terminal device based on the distribution characteristic of the wireless channel parameter.

[0129] Optionally, receiving the first model parameter information transmitted by the wireless access network device includes:

[0130] receiving the compressed first model parameter information transmitted by the wireless access network device, the compressed first model parameter information is obtained by determining a first element bit number reduction degree of the first model parameter and reducing the element bit number of the first model element based on the first element bit number reduction degree.

[0131] Optionally, the wireless channel parameter includes at least one of the following:

[0132] a reference signal receiving power (RSRP) value;

[0133] a received signal strength (RSS) value;

[0134] a power spectral density (PSD) value;

[0135] a signal to interference plus noise ratio (SINR);

[0136] channel state information (CSI); and

[0137] reference signal receiving quality (RSRQ).

[0138] Optionally, the wireless channel parameter further includes at least one of the following:

[0139] frequency bands; and

[0140] time units.

[0141] In accordance with still another aspect of embodiments of the disclosure, a wireless access network device is provided, including:

[0142] a transceiver; and

[0143] a processor, which is coupled with the transceiver and configured to execute the method executed by a wireless access network device in a communication system according to the embodiments of the disclosure.

[0144] In accordance with yet another aspect of the embodiments of the disclosure, a terminal device is provided, including:

[0145] a transceiver; and

[0146] a processor, which is coupled with the transceiver and configured to execute the method executed by a terminal device in a communication system according to the embodiments of the disclosure.

[0147] In accordance with further another aspect of embodiments of the disclosure, a computer-readable storage medium is provided, the computer-readable storage medium having computer programs stored thereon that, when executed by a processor, implement the method executed by a wireless access network device or a terminal device in a communication system according to the embodiments of the disclosure.

[0148] In accordance with further another aspect of embodiments of the disclosure, a computer program product is provided, including computer programs that, when executed by a processor, implement the method executed by a wireless access network device or a terminal device in a communication system according to the embodiments of the disclosure.

[0149] In the communication method, the wireless access network device and the terminal device provided in embodiments of the disclosure, the wireless network device receives first information transmitted by at least one terminal device, the first information including position related information and / or wireless channel parameter related information of the terminal device; determines at least one target terminal device from the at least one terminal device based on the first information transmitted by the at least one terminal device; transmits first model parameter information to the at least one target terminal device, and receives second model parameter information transmitted by the at least one target terminal device, the second model parameter being trained based on the first model parameter; and, fuses the second model parameter transmitted by the at least one target terminal device to obtain the updated first model parameter.

[0150] According to an embodiment of the disclosure, the network model parameters used for prediction of a ratio map can be obtained by performing a federated learning in a suitable target terminal device.

[0151] Furthermore, according to an embodiment of the disclosure, the overhead generated during the federated learning can be reduced with required precision and the network model convergence can be speeded up.

[0152] In addition, according to an embodiment of the disclosure, the network model is locally trained by the terminal device, so that the privacy data of users can be protected to satisfy the confidentiality requirement.

[0153] To describe the technical schemes in embodiments of the disclosure more clearly, the drawings to be used in the description of embodiments of the disclosure will be briefly introduced below.

[0154] The above and other aspects, features, and advantages of certain embodiments of the disclosure will be more apparent from the following description taken in conjunction with the accompanying drawings, in which:

[0155] Figure 1 is a schematic diagram of a wireless network according to an embodiment of the disclosure;

[0156] Figure 2 is a schematic diagram of a base station according to an embodiment of the disclosure;

[0157] Figure 3 is a schematic diagram of a user equipment according to an embodiment of the disclosure;

[0158] Figure 4a is a flowchart of a method executed by a wireless access network device in a communication system according to an embodiment of the disclosure;

[0159] Figure 4b is a schematic diagram of a wireless network according to an embodiment of the disclosure;

[0160] Figure 4c is a schematic diagram of a radio map construction architecture according to an embodiment of the disclosure;

[0161] Figure 4d is a schematic diagram of a radio map visual representation according to an embodiment of the disclosure;

[0162] Figure 5 is a flowchart of a method executed by a terminal device in a communication system according to an embodiment of the disclosure;

[0163] Figure 6a is a schematic diagram of the relationship between generalized characteristics and grouping weights according to an embodiment of the disclosure;

[0164] Figure 6b is a schematic diagram of terminal device-level weight adjustment according to an embodiment of the disclosure;

[0165] Figure 6c is a schematic structure diagram of a model network structure according to an embodiment of the disclosure;

[0166] Figure 6d is a schematic diagram of parameter compression and quantization according to an embodiment of the disclosure;

[0167] Figure 6e is a schematic diagram of a distance relationship between RSRP and different wireless access network devices according to an embodiment of the disclosure;

[0168] Figure 6f is a schematic diagram of an angle relationship between RSRP and different wireless access network devices according to an embodiment of the disclosure;

[0169] Figure 6g is a schematic diagram of an AI model according to an embodiment of the disclosure;

[0170] Figure 7a is a schematic diagram of an overall radio map construction method according to an embodiment of the disclosure;

[0171] Figure 7b is a schematic diagram of UE selection with a high quality characteristic according to an embodiment of the disclosure;

[0172] Figure 7c is a schematic diagram of a radio map construction process according to an embodiment of the disclosure;

[0173] Figure 7d is a schematic diagram of another radio map construction process according to an embodiment of the disclosure;

[0174] Figure 8 is a schematic diagram of an architecture of a radio map construction device according to an embodiment of the disclosure;

[0175] Figure 9 is a schematic diagram of another radio map construction process according to an embodiment of the disclosure;

[0176] Figure 10 is a schematic diagram of some characteristics of an UE according to an embodiment of the disclosure; and

[0177] Figure 11 is a schematic structure diagram of an electronic device according to an embodiment of the disclosure.

[0178] Figure 1 through 11, discussed below, and the various embodiments used to describe the principles of the present disclosure in this patent document are by way of illustration only and should not be construed in any way to limit the scope of the disclosure. Those skilled in the art will understand that the principles of the present disclosure may be implemented in any suitably arranged system or device.

[0179] The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of various embodiments of the disclosure as defined by the claims and their equivalents. It includes various specific details to assist in that understanding but these are to be regarded as merely exemplary. Those of ordinary skill in the art will recognize that various changes and modifications of the various embodiments described herein can be made without departing from the scope and spirit of the disclosure. In addition, descriptions of well-known functions and constructions may be omitted for clarity and conciseness.

[0180] In order to make the objectives, technical schemes and advantages of the embodiments of the disclosure, a clear and complete description will be made with respect to the technical schemes of the embodiments of the disclosure, in conjunction with the accompanying drawings of the embodiments of the disclosure. Apparently, the described embodiments are a part of the embodiments of the disclosure, not all of the embodiments. Based on the described embodiments of the disclosure, all other embodiments obtained by common skilled in the art without creative labor belong to the protection scope of the disclosure.

[0181] The terms and words used in the following description and claims are not limited to the bibliographical meanings, but are merely used by the inventor to enable a clear and consistent understanding of the disclosure. Accordingly, it should be apparent to those skilled in the art that the following description of various embodiments of the disclosure is provided for illustration purpose only and not for the purpose of limiting the disclosure as defined by the appended claims and their equivalents.

[0182] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to "a component surface" includes reference to one or more of such surfaces.

[0183] Before undertaking the DETAILED DESCRIPTION below, it may be advantageous to set forth definitions of certain words and phrases used throughout this patent document. The term "couple" and its derivatives refer to any direct or indirect communication between two or more elements, whether those elements are in physical contact with one another. The terms "transmit," "receive," and "communicate," as well as derivatives thereof, encompass both direct and indirect communication. The terms "include" and "comprise," as well as derivatives thereof, mean inclusion without limitation. The term "or" is inclusive, meaning and / or. The phrase "associated with," as well as derivatives thereof, means to include, be included within, interconnect with, contain, be contained within, connect to or with, couple to or with, be communicable with, cooperate with, interleave, juxtapose, be proximate to, be bound to or with, have, have a property of, have a relationship to or with, or the like. The term "controller" means any device, system or part thereof that controls at least one operation. Such a controller may be implemented in hardware or a combination of hardware and software and / or firmware. The functionality associated with any particular controller may be centralized or distributed, whether locally or remotely. The phrase "at least one of," when used with a list of items, means that different combinations of one or more of the listed items may be used, and only one item in the list may be needed. For example, "at least one of: A, B, and C" includes any of the following combinations: A, B, C, A and B, A and C, B and C, and A and B and C. Likewise, the term "set" means one or more. Accordingly, a set of items can be a single item or a collection of two or more items.

[0184] Furthermore, the expressions "if" and "in case that" as used in the present specification or claims may, depending on the context, be interpreted to mean "when," "in response to," "based on," or "according to," and such expressions may be used interchangeably. In addition, other expressions having substantially the same meaning may also be used in place of these expressions, as long as the technical features of the present disclosure are not impaired. Furthermore, the term "configured" to indicate that predetermined information is set by a base station or a network may imply that the predetermined information is received via a predetermined message (for example, an RRC message).

[0185] Moreover, various functions described below can be implemented or supported by one or more computer programs, each of which is formed from computer readable program code and embodied in a computer readable medium. The terms "application" and "program" refer to one or more computer programs, software components, sets of instructions, procedures, functions, objects, classes, instances, related data, or a portion thereof adapted for implementation in a suitable computer readable program code. The phrase "computer readable program code" includes any type of computer code, including source code, object code, and executable code. The phrase "computer readable medium" includes any type of medium capable of being accessed by a computer, such as read only memory (ROM), random access memory (RAM), a hard disk drive, a compact disc (CD), a digital video disc (DVD), or any other type of memory. A "non-transitory" computer readable medium excludes wired, wireless, optical, or other communication links that transport transitory electrical or other signals. A non-transitory computer readable medium includes media where data can be permanently stored and media where data can be stored and later overwritten, such as a rewritable optical disc or an erasable memory device.

[0186] As described above, it should be noted that each block of the flowcharts and combinations of the flowcharts described in the disclosure may be performed by one or more computer programs including instructions. The entirety of the one or more computer programs may be stored in a single memory device, or the one or more computer programs may be stored in a plurality of memory devices in a distributed manner.

[0187] In addition, the functions or operations described in the disclosure may be processed by a single processor or a combination of processors. The single processor or the combination of processors may be a circuit that performs processing and may include an application processor (AP, e.g., a central processing unit (CPU)), a communication processor (CP, e.g., a modem), a graphics processing unit (GPU), a neural processing unit (NPU) (e.g., an artificial intelligence (AI) chip), a Wi-Fi chip, a Bluetooth® chip, a global positioning system (GPS) chip, a near-field communication (NFC) chip, a connectivity chip, a sensor controller, a touch controller, a fingerprint sensor controller, a display driver integrated circuit (IC), an audio codec (CODEC) chip, a universal serial bus (USB) controller, a camera controller, an image processing IC, a microprocessor unit (MPU), a system on chip (SoC), an IC, or a similar circuit.

[0188] Furthermore, it should be noted that various embodiments in the claims and descriptions of the disclosure may be implemented in the form of hardware, software, or a combination of hardware and software. Such software may be stored in a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium stores one or more computer programs (software modules), and the one or more computer programs include computer-executable instructions which, when executed individually or collectively by one or more processors of an electronic device, operate the electronic device to perform the method according to the disclosure.

[0189] The software may be stored in a transient or non-transitory storage device, for example, in the form of read-only memory (ROM) (regardless of whether it is erasable or rewritable), or random access memory (RAM), memory chips, devices, or integrated circuits (ICs). Also, the software may be stored in optically or magnetically readable media such as compact discs (CDs), digital versatile discs (DVDs), magnetic disks, or magnetic tapes. It should be understood that the storage devices and storage media are examples of non-transitory machine-readable storage media suitable for storing a program for implementing various embodiments of the disclosure.

[0190] Accordingly, various embodiments provide a program including code for implementing a device or method according to any one of the claims of the disclosure, and a non-transitory machine-readable storage medium storing such a program.

[0191] Definitions for other certain words and phrases are provided throughout this patent document. Those of ordinary skill in the art should understand that in many if not most instances, such definitions apply to prior as well as future uses of such defined words and phrases.

[0192] The figures included herein, and the various embodiments used to describe the principles of the disclosure are by way of illustration only and should not be construed in any way to limit the scope of the disclosure. Further, those skilled in the art will understand that the principles of the disclosure may be implemented in any suitably arranged wireless communication system.

[0193] Figures 1-3 below describe various embodiments of the disclosure implemented in wireless communications systems. The descriptions of Figures 1-3 are not meant to imply physical or architectural limitations to the manner in which different embodiments may be implemented. Different embodiments of the disclosure may be implemented in any suitably-arranged communications system.

[0194] Figure 1 illustrates an example wireless network according to embodiments of the disclosure. The embodiment of the wireless network shown in Figure 1 is for illustration only. Other embodiments of the wireless network 100 could be used without departing from the scope of the disclosure.

[0195] As shown in Figure 1, the wireless network includes a base station (next generation nodeB, gNB or gNodeB) 101, a gNB 102, and a gNB 103. The gNB 101 communicates with the gNB 102 and the gNB 103. The gNB 101 also communicates with at least one network 130, such as the internet, a proprietary internet protocol (IP) network, or other data network.

[0196] The gNB 102 provides wireless broadband access to the network 130 for a plurality of first user equipments (UEs) within a coverage area 120 of the gNB 102. The plurality of first UEs includes a UE 111, which may be located in a small business; a UE 112, which may be located in an enterprise (E); a UE 113, which may be located in a WiFi hotspot (HS); a UE 114, which may be located in a first residence (R1); a UE 115, which may be located in a second residence (R2); and a UE 116, which may be a mobile device (M), such as a cell phone, a wireless laptop, a wireless personal digital assistant (PDA), or the like. The gNB 103 provides wireless broadband access to the network 130 for a plurality of second UEs within a coverage area 125 of the gNB 103. The plurality of second UEs include the UE 115 and the UE 116, as well as subscriber stations (SS, for example, UEs) 117, 118 and 119. In some embodiments, one or more of the gNBs 101-103 may communicate with each other and with the UEs 111-116 using existing wireless communication techniques, and one or more of the UE 111-119 may communicate directly with each other (e.g., UEs 117-119) using wireless communication techniques.

[0197] Depending on the network type, the term "base station" or "BS" can refer to any component (or collection of components) configured to provide wireless access to a network, such as transmit point (TP), transmit-receive point (TRP), an enhanced (or "evolved") base station (eNodeB or eNB), a 5G base station (gNB), a macrocell, a femtocell, a wireless fidelity (WiFi) access point (AP), or other wirelessly enabled devices. Base stations may provide wireless access in accordance with one or more wireless communication protocols, e.g., 3GPP 5G New Radio (NR), Long Term Evolution (LTE), LTE Advanced (LTE-A), high speed packet access (HSPA), Wi-Fi 802.11a / b / g / n / ac, etc. For the sake of convenience, the various names for a base station-type apparatus and functionality are used interchangeably in this patent document to refer to network infrastructure components that provide wireless access to remote terminals. Also, depending on the network type, the term "user equipment" (UE) can refer to any component such as a mobile station (MS), subscriber station (SS), remote terminal, wireless terminal, receive point, or user device. For the sake of convenience, the various names for a user equipment-type device and functionality are used interchangeably in this patent document to refer to remote wireless equipment that wirelessly accesses a BS, whether the UE is a mobile device (such as a mobile telephone or smartphone) or is normally considered a stationary device (such as a desktop computer or vending machine).

[0198] Dotted lines show the approximate extents of the coverage areas 120 and 125, which are shown as approximately circular for the purposes of illustration and explanation only. It should be clearly understood that the coverage areas associated with gNBs, such as the coverage areas 120 and 125, may have other shapes, including irregular shapes, depending upon the configuration of the gNBs and variations in the radio environment associated with natural and man-made obstructions.

[0199] As described in more detail below, one or more of the UEs 111-119 include circuitry, programing, or a combination thereof. In certain embodiments, and one or more of the gNBs 101-103 includes circuitry, programing, or a combination thereof.

[0200] Although Figure 1 illustrates one example of a wireless network, various changes may be made to Figure 1. For example, the wireless network could include any number of gNBs and any number of UEs in any suitable arrangement. Also, the gNB 101 could communicate directly with any number of UEs and provide those UEs with wireless broadband access to the network 130. Similarly, each gNB 102-103 could communicate directly with the network 130 and provide UEs with direct wireless broadband access to the network 130. Further, the gNBs 101, 102, and / or 103 could provide access to other or additional external networks, such as external telephone networks or other types of data networks.

[0201] Figure 2 illustrates an example base station according to embodiments of the disclosure. The embodiment of the gNB 102 illustrated in Figure 2 is for illustration only, and the gNBs 101 and 103 of Figure 1 could have the same or similar configuration. However, gNBs come in a wide variety of configurations, and Figure 2 does not limit the scope of the disclosure to any particular implementation of a gNB.

[0202] As shown in FIG 2, the gNB 102 includes multiple antennas 200a-200n, multiple radio frequency (RF) transceivers 201a-201n, transmit (TX) processing circuitry 203, and receive (RX) processing circuitry 204. The gNB 102 also includes a controller / processor 205, a memory 206, and a backhaul or network interface (IF) 207.

[0203] The RF transceivers 201a-201n receive, from the antennas 200a-200n, incoming RF signals, such as signals transmitted by UEs in the network 100. The RF transceivers 201a-201n down-convert the incoming RF signals to generate intermediate frequency (IF) or baseband signals. The IF or baseband signals are sent to the RX processing circuitry 204, which generates processed baseband signals by filtering, decoding, and / or digitizing the baseband or IF signals. The RX processing circuitry 204 transmits the processed baseband signals to the controller / processor 205 for further processing.

[0204] The TX processing circuitry 203 receives analog or digital data (such as voice data, web data, electronic mail, or interactive video game data) from the controller / processor 205. The TX processing circuitry 203 encodes, multiplexes, and / or digitizes the outgoing baseband data to generate processed baseband or IF signals. The RF transceivers 201a-201n receive the outgoing processed baseband or IF signals from the TX processing circuitry 203 and up-converts the baseband or IF signals to RF signals that are transmitted via the antennas 201a-201n.

[0205] The controller / processor 205 can include one or more processors or other processing devices that control the overall operation of the gNB 102. For example, the controller / processor 205 could control the reception of forward channel signals and the transmission of reverse channel signals by the RF transceivers 201a-201n, the RX processing circuitry 204, and the TX processing circuitry 203 in accordance with well-known principles. The controller / processor 205 could support additional functions as well, such as more advanced wireless communication functions.

[0206] For instance, the controller / processor 205 could support beam forming or directional routing operations in which outgoing signals from multiple antennas 200a-200n are weighted differently to effectively steer the outgoing signals in a desired direction. Any of a wide variety of other functions could be supported in the gNB 102 by the controller / processor 205.

[0207] The controller / processor 205 is also capable of executing programs and other processes resident in the memory 206, such as an operating system (OS). The controller / processor 205 can move data into or out of the memory 206 as required by an executing process.

[0208] The controller / processor 205 is also coupled to the backhaul or network interface 207. The backhaul or network interface 207 allows the gNB 102 to communicate with other devices or systems over a backhaul connection or over a network. The interface 207 could support communications over any suitable wired or wireless connection(s). For example, when the gNB 102 is implemented as part of a cellular communication system (such as one supporting 5G, LTE, or LTE-A), the interface 207 could allow the gNB 102 to communicate with other gNBs over a wired or wireless backhaul connection. When the gNB 102 is implemented as an access point, the interface 207 could allow the gNB 102 to communicate over a wired or wireless local area network or over a wired or wireless connection to a larger network (such as the Internet). The interface 207 includes any suitable structure supporting communications over a wired or wireless connection, such as an Ethernet or RF transceiver.

[0209] The memory 206 is coupled to the controller / processor 205. Part of the memory 206 could include a random access memory (RAM), and another part of the memory 206 could include a Flash memory or other read only memory (ROM).

[0210] Although Figure 2 illustrates one example of gNB 102, various changes may be made to Figure 2. For example, the gNB 102 could include any number of each component shown in Figure 2. As a particular example, an access point could include a number of interfaces 207, and the controller / processor 205 could support routing functions to route data between different network addresses. As another particular example, while shown as including a single instance of TX processing circuitry 203 and a single instance of RX processing circuitry 204, the gNB 102 could include multiple instances of each (such as one per RF transceiver). Also, various components in Figure 2 could be combined, further subdivided, or omitted and additional components could be added according to particular needs.

[0211] Figure 3 illustrates an example user equipment according to embodiments of the disclosure. The embodiment of the UE 116 illustrated in Figure 3 is for illustration only, and the UEs 111-115 and 117-119 of Figure 1 could have the same or similar configuration. However, UEs come in a wide variety of configurations, and Figure 3 does not limit the scope of the disclosure to any particular implementation of a UE.

[0212] As shown in Figure 3, the UE 116 includes an antenna 301, a radio frequency (RF) transceiver 302, TX processing circuitry 303, a microphone 304, and receive (RX) processing circuitry 305. The UE 116 also includes a speaker 306, a controller or processor 307, an input / output (I / O) interface (IF) 308, an input device 309, a touchscreen display 310, and a memory 311. The memory 311 includes an OS 312 and one or more applications 313.

[0213] The RF transceiver 302 receives, from the antenna 301, an incoming RF signal transmitted by an gNB of the network 100. The RF transceiver 302 down-converts the incoming RF signal to generate an IF or baseband signal. The IF or baseband signal is sent to the RX processing circuitry 305, which generates a processed baseband signal by filtering, decoding, and / or digitizing the baseband or IF signal. The RX processing circuitry 305 transmits the processed baseband signal to the speaker 306 (such as for voice data) or to the processor 307 for further processing (such as for web browsing data).

[0214] The TX processing circuitry 303 receives analog or digital voice data from the microphone 304 or other outgoing baseband data (such as web data, e-mail, or interactive video game data) from the processor 307. The TX processing circuitry 303 encodes, multiplexes, and / or digitizes the outgoing baseband data to generate a processed baseband or IF signal. The RF transceiver 302 receives the outgoing processed baseband or IF signal from the TX processing circuitry 303 and up-converts the baseband or IF signal to an RF signal that is transmitted via the antenna 301.

[0215] The processor 307 can include one or more processors or other processing devices and execute the OS 312 stored in the memory 311 in order to control the overall operation of the UE 116. For example, the processor 307 could control the reception of forward channel signals and the transmission of reverse channel signals by the RF transceiver 302, the RX processing circuitry 305, and the TX processing circuitry 303 in accordance with well-known principles. In some embodiments, the processor 307 includes at least one microprocessor or microcontroller.

[0216] The processor 307 is also capable of executing other processes and programs resident in the memory 311, such as processes for CSI (Channel State Information) reporting on uplink channel. The processor 307 can move data into or out of the memory 311 as required by an executing process. In some embodiments, the processor 307 is configured to execute the applications 313 based on the OS 312 or in response to signals received from gNBs or an operator. The processor 307 is also coupled to the I / O interface 308, which provides the UE 116 with the ability to connect to other devices, such as laptop computers and handheld computers. The I / O interface 308 is the communication path between these accessories and the processor 307.

[0217] The processor 307 is also coupled to the touchscreen display 310. The user of the UE 116 can use the touchscreen display 310 to enter data into the UE 116. The touchscreen display 310 may be a liquid crystal display, light emitting diode display, or other display capable of rendering text and / or at least limited graphics, such as from web sites.

[0218] The memory 311 is coupled to the processor 307. Part of the memory 311 could include RAM, and another part of the memory 311 could include a Flash memory or other ROM.

[0219] Although Figure 3 illustrates one example of UE 116, various changes may be made to Figure 3. For example, various components in Figure 3 could be combined, further subdivided, or omitted and additional components could be added according to particular needs. As a particular example, the processor 307 could be divided into multiple processors, such as one or more central processing units (CPUs) and one or more graphics processing units (GPUs). Also, while Figure 3 illustrates the UE 116 configured as a mobile telephone or smartphone, UEs could be configured to operate as other types of mobile or stationary devices.

[0220] To make the objectives, technical schemes and advantages of the disclosure clearer, the implementations of the disclosure will be further described below in detail with reference to the drawings.

[0221] The embodiments of the disclosure provide an efficient and intelligent radio map construction method, which can protect user data privacy while satisfying the requirements of low overhead with required precision.

[0222] The efficient and intelligent radio map construction method provided by the disclosure adopts a distributed machine learning architecture based on federated learning. A terminal device trains a model based on a local data set and then uploads the model to a wireless access network device. The model is fused (or aggregated, or updated) in the wireless access network device, and the fused model is then re-issued to the terminal device. The radio map construction includes multiple rounds of model issuing, training, uploading, fusion and re-issuing. In each round, the wireless access network device needs to select a suitable terminal to train the model, thereby reducing overhead with required precision and accelerating the model convergence.

[0223] The technical schemes in the embodiments of the disclosure and the technical effects achieved by the technical schemes of the disclosure will be explained below by describing several exemplary implementations. It should be pointed out that the following implementations can be referred to, learned from or combined with each other, and the same terms, similar features and similar implementation steps in different implementations will not be repeated.

[0224] An embodiment of the disclosure provides a method executed by a wireless access network device in a communication system. As shown in Figure 4a, the method includes the following steps.

[0225] In step S101,the first information transmitted by at least one terminal device is received, and the first information includes position related information and / or wireless channel parameter related information of the terminal device.

[0226] At least one terminal device refers to the terminal device in the cell (or referred to as the serving cell) served by the wireless access network device. In the embodiment of the disclosure, the wireless access network device receives the first information reported by the terminal device in this cell. As an example, a typical wireless network is illustrated in Figure 4b, wherein a plurality of base stations cover one region, and a plurality of mobile UEs move in this region and can receive signals from at least one base station. Each UE acquires the first information (which can be used as data samples of the data set) and reports it to the wireless access network device, so as to obtain the input of radio map construction.

[0227] In the embodiment of the disclosure, the wireless access network device may include, but not limited to, a base station, a relay node, a combination of the both, or the like. For the convenience of description, the following description may be given by taking the wireless access network device being a base station as an example. It should be understood that the implementation of a relay node or the implementation of the combination of a base station and a relay node is similar to that on the base station side, and the similar processing process will not be repeated here.

[0228] In the embodiment of the disclosure, the position related information of the terminal device refers to the related information calculated or counted based on the position (e.g., longitude ( ), latitude ( ) and altitude ( ), etc.) of at least one time unit, instead of the position information itself, thereby protecting the privacy of users.

[0229] In the embodiment of the disclosure, the wireless channel parameter related information refers to the related information calculated or counted based on the wireless channel parameter corresponding to the position of at least one time unit, and may include the wireless channel parameter. Optionally, the wireless channel parameter includes at least one of the following: a reference signal receiving power (RSRP) value; a received signal strength (RSS) value; a power spectral density (PSD) value); a signal to interference plus noise ratio (SINR); channel state information (CSI); and, reference signal receiving quality (RSRQ). Optionally, the wireless channel parameter further includes at least one of the following: frequency bands; time units; or the like. Optionally, the wireless channel parameter may also be called a wireless environment parameter, or a wireless parameter for short.

[0230] As an example, the terminal device collects the position information and the wireless channel parameter of the corresponding position in real time, and stores them in the local data set in chronological order. In addition to the position information and the corresponding wireless channel parameter, the data collected by the terminal device also includes, but not limited to, frequency bands, time, or the like.

[0231] Optionally, the terminal device may also collect wireless channel parameters of a plurality of wireless access network devices, for example, including but not limited to, the base station of this cell, the base station of the neighboring cell or the like, and the subsequent processing process is similar and will not be repeated.

[0232] By taking the RSRP value as an example, each UE collects RSRP as a data sample of the local data set D. For thekthUE, the data set is represented as , where is the number of samples of , ( is thejthdata sample, represents the position and may be represented by 2D coordinates , and represents the RSRP of this position.

[0233] Optionally, a functionf( )may be obtained according to the corresponding RSRP value of each base station acquired at a particular positionp, so that a radio map is constructed, for example: , where represents the RSRP of the mthbase station. The RSRP of each base station is affected by the transmitting power, the path loss and the antenna gain. By taking the mthbase station as an example, , where is the transmitting power of the mthbase station; is the gain of the transmitting antenna, is the gain of the receiving antenna, and are determined by the antenna gain and direction; and, is the path loss from the transmitter to the receiver, including large-scale fading and small-scale fading. The large-scale fading mainly depends on the distance from the UE to the base station and the shadow fading, and the small-scale fading mainly depends on the multipath propagation and Doppler frequency shift.

[0234] In step S102,at least one target terminal device is determined from the at least one terminal device based on the first information transmitted by the at least one terminal device.

[0235] In the embodiment of the disclosure, a terminal device is selected for model training. The target terminal device refers to the terminal device selected by the wireless access network device. Specifically, the wireless access network device may select suitable terminal devices with reference to the first information transmitted by the at least one terminal device, and use high-quality data sets on these terminal devices to perform model training, thereby improving the model convergence speed.

[0236] Optionally, the wireless access network device notifies the selected target terminal device to perform model training.

[0237] In step S103, the first model parameter information is transmitted to the at least one target terminal device, and the second model parameter information transmitted by the at least one target terminal device is received, the second model parameter being trained based on the first model parameter.

[0238] The first model parameter information may be interpreted as the information related to global model parameters determined by the wireless access network device. The second model parameter information may be interpreted as the information related to local model parameters trained by the terminal device. Optionally, the model parameter information may refer to all or some parameters of the model, or variations with a specific model (e.g., before training or before fusion), or compressed model parameters, etc.

[0239] Optionally, the step of transmitting the first model parameter information to at least one target terminal device may also be performed before or at the same time as step S101 or S102. For example, if the wireless access network device transmits the first model parameter information to all terminal devices, at least one target terminal device may receive the first model parameter information.

[0240] In step S104, the second model parameter transmitted by the at least one target terminal device is fused(or updated) to obtain the updated first model parameter.

[0241] In the embodiment of the disclosure, the radio map construction method based on federated learning is adopted. A radio signal strength prediction network model is designed first according to the characteristics of the radio map, and the related parameters are obtained to initialize the neural network model. Optionally, as shown inFigure 4c, the wireless access network device may also communicate with a central server, and the central server will generate an initial global model. In each round (e.g., the th round, where ) of iteration, the wireless access network device downloads the global model, and selects, according to the first information transmitted by the terminal device in each round, a target terminal device (e.g., target terminal devices) participating in this round of federated learning, and the wireless access network device transmits the first model parameter information of the global model to the terminal device in advance or in real time. The selected target terminal device is locally trained based on the first model parameter issued by the wireless access network device, and the trained second model parameter information is uploaded to the wireless access network device. The wireless access network device fuses the second model parameter uploaded by the terminal device to update the first model parameter. Optionally, upon receiving the trained local model , the wireless access network device fuses the model and transmits the fused local model to the central server to fuse the global model. For example, the aggregation operation in the tthround of iteration may be represented as: . When the training reaches a training end condition (for example, the number of rounds reaches the set value, or a particular iteration time is reached, or the model converges, or the model precision tends to be stable, or the model reaches the predetermined effect, or the like), the model for predicting the radio map is trained; otherwise, the next round of terminal device selection, model parameter issuing, training, uploading and fusion is continued.

[0242] Optionally, the fusion mode includes, but not limited to, averaging, weighted averaging, weighted summation, etc.

[0243] That is, in the embodiment of the disclosure, in the process of training the model for predicting the radio map, the steps S101 to S104 may be performed multiple times (i.e., in the above multiple rounds), and the repeated processing process will not be repeated.

[0244] The radio map constructed by the embodiment of the disclosure can reflect the information map of the relationship between the actual physical position and the wireless environment and can represent the radio communication environment, and each geographical position corresponds to an important wireless channel parameter value, for example, the received signal strength, the channel attenuation or the like. For example, as shown inFigure 4d, the radio map can provide the overall visual representation of the cellular network signal strength and the coverage region, so that the wireless access network device (e.g., gNB) has a comprehensive understanding of the radio environment, so as to better serve B5G, 6G and other communication networks that support massive connection, diversified ultra-wide band services and high-precision sensing. The radio map may also be referred to as a Radio Environment Map (REM). On the base station side, it can be used for positioning, radio resource management (e.g., switching management), transmission strategy optimization, power control, resource allocation, dynamic spectrum access, network planning and interference management (e.g., coordination and mitigation) or the like; while on the terminal side, it can assist the terminal to obtain the position, signal strength and other information. In addition, it can also be used for UAV path planning, base station address selection or the like.

[0245] Optionally, the radio map may be represented by using the function relationship between the geographical position and the RSRP, PSD and / or RSS. For example,{RSRP}=f(x, y), but it is not limited thereto.

[0246] In the method executed by a wireless access network device in a communication system provided in the embodiment of the disclosure, the first information transmitted by at least one terminal device is received, the first information including position related information and / or wireless channel parameter related information of the terminal device; at least one target terminal device is determined from the at least one terminal device based on the first information transmitted by the at least one terminal device; the first model parameter information is transmitted to the at least one target terminal device, and the second model parameter information transmitted by the at least one target terminal device is received, the second model parameter being trained based on the first model parameter; and, the second model parameter transmitted by the at least one target terminal device is fused to obtain the updated first model parameter. Thus, the model parameters based on more suitable terminal device federated learning can be obtained, and radio map is constructed with the required precision at low overhead.

[0247] In addition, compared with constructing a radio map by using buildings, streets, valleys and objects with different heights, in the method provided in the embodiment of the disclosure, the first information is reported by a mobile terminal device, and a terminal device is reasonably selected for model training, without being limited by the environmental radial characteristics in the real propagation environment. Thus, it can be applied to complex and heterogeneous propagation environments, and the requirements of the environmental and regional coverage for accurately constructing a radio map are satisfied.

[0248] Compared with constructing a radio map by an interpolation method, in the method provided in the embodiment of the disclosure, the complexity of radio map construction is effectively reduced through multiple rounds of AI model training and model parameter fusion.

[0249] Compared with constructing a radio map by using physical environmental map information or data collected by sensors, in the method provided in the embodiment of the disclosure, instead of using sensors or maps for data generation, a target terminal device is adaptively selected, and real-time training is performed by using the data collected by the terminal device, so that the construction cost for the radio map is reduced.

[0250] Moreover, in the method provided in the embodiment of the disclosure, it is unnecessary to obtain the original data such as real-time position information of the terminal device during the radio map construction, and the model is locally trained by the terminal device, so that the privacy data of users can be protected to satisfy the confidentiality requirement.

[0251] In the embodiment of the disclosure, in order to reduce the overall communication overhead, an optional implementation is provided for step S102. Target terminal devices are selected based on the distributed characteristic vector, so that efficient users with high quality characteristics are selected. The target terminal device that can use a high quality data training model can train model parameters and transmit them to the wireless access network device, thereby reducing the overall overhead. Specifically, this optional implementation may include: determining characteristic information based on the first information transmitted by the at least one terminal device; and, determining at least one target terminal device from the at least one terminal device based on the characteristic information.

[0252] The first information can represent some characteristics of the data set of the terminal device, including but not limited to, richness, specificity, scarcity, fluctuation, distribution characteristics, or the like. The characteristic information determined based on the first information corresponds to the characteristics of the data set of the terminal device, so that a target terminal device is selected. The suitable data set of the terminal device can be selected for model training, thereby improving the model convergence speed.

[0253] Optionally, compared with the samples used in the previous round of model training, each round of training may extract high quality characteristic information from the newly added samples in the data set.

[0254] In the embodiment of the disclosure, the characteristic information includes at least one of the following:

[0255] (1) The richness (or referred to as abundance) of the wireless channel parameter corresponding to each terminal device, the richness being related to the data volume and / or data fluctuation range.

[0256] That is, whether the terminal device is selected is related to the richness of the wireless channel parameter. For example, if the richness of the wireless channel parameter is higher, the possibility that the terminal device is selected is higher. The rich wireless channel parameter can improve the performance and generalization capability of the model. If the data set is richer (for example, the covered geographical positions are more, or the data volume in the data set is larger), the trained model is more accurate. In practical applications, those skilled in the art can set the determination criteria (e.g., data volume, position, data type, data fluctuation range, etc.) for the richness of the wireless channel parameter according to the actual situation. However, it will not be limited to the embodiment of the disclosure.

[0257] (2) Scarcity related information of the wireless channel parameter corresponding to each terminal device in at least one wireless channel parameter interval, the scarcity related information being related to the data volume in the at least one wireless channel parameter interval.

[0258] That is, whether the terminal device is selected is related to whether the wireless channel parameter has a wireless channel parameter interval with scarce data in the at least one wireless channel parameter interval as well as the specific situation in the wireless channel parameter interval with scarce data. For example, if the wireless channel parameter interval with scarce data is included, the probability that the terminal device is selected is higher. The wireless channel parameter with scarce data can improve the reliability of the model. The data contained in the data set is more scarce (for example, including the data collected at the position where other terminal devices rarely collect data), the trained model is more accurate. In practical applications, those skilled in the art can set the determination criteria (e.g., data volume, data proportion, etc.) for the scarcity related information according to the actual situation. However, it will not be limited to the embodiment of the disclosure.

[0259] Optionally, the characteristic information may further include:

[0260] (3) whether the wireless channel parameter corresponding to each terminal device includes a target wireless channel parameter, the target wireless channel parameter being related to at least one of the frequency of occurrence of each wireless channel parameter value, the variance of the wireless channel parameter value, and the mean value of the wireless channel parameter value.

[0261] That is, whether the terminal device is selected is related to whether the wireless channel parameter includes a special wireless channel parameter. For example, the probability that the terminal device including a special wireless channel parameter is selected is higher. The special wireless channel parameter can improve the reliability of the model. In practical applications, those skilled in the art can set the determination criteria for the special wireless channel parameter according to the actual situation. Optionally, the specificity of the data can be determined by at least one of the frequency of occurrence of each wireless channel parameter value, the variance of the wireless channel parameter value, and the mean value of the wireless channel parameter value, for example, special position, special wireless channel parameter value or the like. However, it will not be limited to the embodiment of the disclosure.

[0262] In the embodiment of the disclosure, the first information may include at least one of the following information and / or the richness of at least one of the following information (that is, the richness of the information may be determined and then directly reported by the terminal device), or the information corresponding to frequency bands and / or time units and / or different wireless access network devices includes at least one of the following.

[0263] (1) The data volume corresponding to the wireless channel parameter.

[0264] The local wireless channel parameter set of the terminal device may include wireless channel parameters of different time units. The data volume corresponding to the wireless channel parameter can be interpreted as the wireless channel parameters of time units, or can also be interpreted as the number of data samples, i.e., the number of samples.

[0265] (2) The difference between the maximum longitude and the minimum longitude among longitudes determined based on at least one piece of position information.

[0266] This information can be interpreted as the movement range of the terminal device in longitude.

[0267] (3) The difference between the maximum latitude and the minimum latitude among latitudes determined based on at least one piece of position information.

[0268] This information can be interpreted as the movement range of the terminal device in latitude.

[0269] (4) The difference between the maximum altitude and the minimum altitude among altitudes determined based on at least one piece of position information.

[0270] This information can be interpreted as the movement range of the terminal device in altitude.

[0271] (5) The difference between the maximum wireless channel parameter value and the minimum wireless channel parameter value among wireless channel parameter values corresponding to at least one piece of position information.

[0272] This information can be interpreted as the fluctuation range of the channel condition.

[0273] (6) A first preset number of maximum distances among the respective distances determined based on each piece of position information.

[0274] That is, the distance between every two pieces of position information can be determined based on the position information of each time unit of the terminal device, (a first preset number, ) maximum distances, e.g., top maximum distances, are selected. The value of the first preset number can be set according to the actual situation, and will not be limited to the embodiment of the disclosure.

[0275] (7) A second preset number of minimum distances among the respective distances determined based on each piece of position information.

[0276] That is, the distance between every two pieces of position information can be determined based on the position information of each time unit of the terminal device, (a second preset number, ) minimum distances, e.g., top minimum distances, are selected. The value of the second preset number can be set according to the actual situation, and will not be limited to the embodiment of the disclosure.

[0277] (8) A position range determined based on each piece of position information.

[0278] For example, the longitude , latitude and altitude in the position information of each time unit of the terminal device are counted, respectively, and the position range of the terminal device is obtained by subtracting the minimum from the maximum , subtracting the minimum from the maximum and subtracting the minimum from the maximum .

[0279] (9) A mean value of each piece of position information.

[0280] That is, the mean value of the position information in the local data set of the terminal device can be determined based on the position information of each time unit of the terminal device.

[0281] (10) A mean value of each wireless channel parameter value.

[0282] That is, the mean value of wireless channel parameters (e.g., RSRP, but not limited thereto) in the local data set of the terminal device can be determined based on the wireless channel parameter value of each time unit of the terminal device.

[0283] (11) The data volume of at least one wireless channel parameter interval.

[0284] The wireless channel parameter can be divided into a plurality of intervals, and the data volume in each interval can be counted.

[0285] (12) A standard deviation of each piece of position information.

[0286] That is, the standard deviation of the position information in the local data set of the terminal device can be determined based on the position information of each time unit of the terminal device. Optionally, the standard deviation can also be converted into the variance.

[0287] (13) A standard deviation of each wireless channel parameter value.

[0288] That is, the mean value of wireless channel parameters (e.g., RSRP, but not limited thereto) in the local data set of the terminal device can be determined based on the wireless channel parameter value of each time unit of the terminal device. Optionally, the standard deviation can also be converted into the variance.

[0289] (14) An average value of standard deviations of wireless channel parameters corresponding to different wireless access network devices.

[0290] That is, if the terminal device collects the wireless channel parameters of a plurality of wireless access network devices, the standard deviation of the wireless channel parameter corresponding to each wireless access network device can be counted, and the standard deviations corresponding to the plurality of wireless access network devices can be averaged. Optionally, the standard deviation can also be converted into the variance.

[0291] Some of the above information can also be interpreted as the characteristic information extracted from the first information. Optionally, the characteristic information may be executed by the terminal device and transmitted to the wireless access network device, or may be extracted by the wireless access network device based on the data reported by the terminal.

[0292] Optionally, an optional implementation is provided for determining the richness of the wireless channel parameter corresponding to the terminal device. Specifically, determining the richness of the wireless channel parameter corresponding to the terminal device based on the first information transmitted by the at least one terminal device may include at least one of the following ways.

[0293] (1) For each terminal device, based on at least one of the data volume of the wireless channel parameter, the fluctuation range of the position and the fluctuation range of the wireless channel parameter corresponding to this terminal device, as well as at least one of the data volume of wireless channel parameters, the fluctuation range of positions and the fluctuation range of wireless channel parameters corresponding to all terminal devices, the richness of the wireless channel parameter corresponding to this terminal device is determined.

[0294] That is, all terminal devices can be compared based on the richness of the wireless channel parameters corresponding to the terminal devices.

[0295] (2) The richness of the wireless channel parameter corresponding to this terminal device is determined based on the first information and a corresponding threshold.

[0296] That is, the richness of the wireless channel parameter corresponding to the terminal device can be determined by comparing at least one of the first information with the corresponding threshold.

[0297] In the embodiment of the disclosure, an optional implementation is provided for determining whether the wireless channel parameter corresponding to the terminal device includes a target wireless channel parameter. Specifically, the wireless access network device maintains a wireless channel parameter pool for each terminal device. The maintained information includes, but not limited to, the receiving time, the identity document (ID) of the terminal device, the wireless channel parameter value, or the like.

[0298] Optionally, the wireless access network device counts the received wireless channel parameter, including but not limited to the following statistical information.

[0299] 1. For all terminal devices, the frequency of occurrence of each wireless channel parameter value, the variance and mean value of the wireless channel parameter and the like are jointly counted.

[0300] 2. For different terminal devices, the frequency of occurrence of each wireless channel parameter value, the variance and mean value of the wireless channel parameter and the like are counted, respectively.

[0301] The wireless access network device may determine, according to the above statistical information, whether the terminal device reports special wireless channel parameters. For example, it is determined whether the terminal device collects wireless channel parameters with special positions.

[0302] Optionally, the determining, based on the first information transmitted by the at least one terminal device, whether the wireless channel parameter corresponding to the terminal device includes a target wireless channel parameter may include: if the variance of the wireless channel parameter value and / or the mean value of the wireless channel parameter value is greater than a first threshold and the frequency of occurrence of at least one wireless channel parameter value is less than a second threshold, determining that the wireless channel parameter corresponding to the terminal device includes a target wireless channel parameter.

[0303] For example, if a certain RSRP value of the terminal device appears for few times and the variance of the RSRP is larger, the wireless access network device can consider that this terminal device collects the RSRP with a special position.

[0304] Optionally, based on the inclusion of the target wireless channel parameter in the wireless channel parameter corresponding to the terminal device or the information (e.g.,indication information) related to the target wireless channel parameter included in the wireless channel parameter corresponding to the terminal device, it is determined whether this terminal device is selected.

[0305] In the embodiment of the disclosure, an optional implementation is provided for determining the scarcity related information of the wireless channel parameter corresponding to the terminal device in at least one wireless channel parameter interval (also referred to as segment or segmented interval). Specifically, the determining, based on the first information transmitted by the at least one terminal device, whether there is a target interval in the at least one wireless channel parameter interval may include: determining, based on the first information transmitted by the at least one terminal device, whether there is a target interval in the at least one wireless channel parameter interval, the data volume corresponding to the wireless channel parameter of at least one terminal device in the target interval satisfying a preset condition; and, if there is a target interval in the at least one wireless channel parameter interval, for each terminal device, using the data volume in the target interval as scarcity related information of the wireless channel parameter corresponding to the terminal device in the at least one wireless channel parameter interval.

[0306] Optionally, the determining, based on the first information transmitted by the at least one terminal device, whether there is a target interval in the at least one wireless channel parameter interval may include: determining, from the first information transmitted by the at least one terminal device, a proportion of the data volume separately corresponding to the at least one wireless channel parameter interval in the total data volume; for each wireless channel parameter interval in the at least one wireless channel parameter interval, determining the difference between the corresponding proportion and a predetermined proportion of this wireless channel parameter interval; and, determining the wireless channel parameter interval corresponding to the maximum difference as a target interval.

[0307] For example, the wireless access network device may obtain the predetermined distribution of this network in each wireless channel parameter interval in advance (e.g., the predetermined percentage occupied by each wireless channel parameter interval), compare the percentage of each wireless channel parameter interval of the terminal device in the total data volume, and determine the wireless channel parameter interval with the maximum percentage difference as a target interval, or take the wireless channel parameter interval with a percentage less than the predetermined percentage of the corresponding interval as a target interval.

[0308] In an example, by taking the wireless channel parameter being RSRP as an example, the wireless access network device obtains the predetermined percentage of each RSRP interval of this network in advance through a test set, and then compares the percentage of each RSRP interval corresponding to the RSRP data with the predetermined percentage in the test set to determine a scarce RSRP interval. As shown in Table 1, the percentage-test set corresponds to the predetermined percentage (as a reference value) of each RSRP interval. By using the total number and percentage of each RSRP interval reported by the terminal device and the gap between the percentage of each RSRP interval and the corresponding reference value, assuming that the gap of therthRSRP interval is , the RSRP interval with the maximum percentage gap is determined as a scarce interval. For example, by taking the proportion of RSRP of different intervals in the wireless data pool of the wireless network access network as a reference which is represented by , the base station calculates the proportion of the RSRP reported by the UE, which is represented by , . is the number of the terminal devices that report the first information to the served wireless access network device m during each iteration, represents the data volume of the terminal in the interval, and the scarce interval (sid) is the RSRP interval of the maximum value, that is, . In other words,

[0309]

[0310] ,where represents the scarce interval, and represents the number of RSRP intervals. The wireless access network device finds that the percentage of the RSRP in the interval -110 < RSRP ≤ -100 is less 15% than the predetermined percentage (25%). If the gap is maximal, this RSRP interval is a scarce interval (target interval).

[0311]

[0312] In the embodiment of the disclosure, the fluctuation of the wireless channel parameter can be evaluated by the standard deviations of the number of samples, RSRP, x coordinate value and y coordinate value.

[0313]

[0314] In the embodiment of the disclosure, an optional implementation is provided for the step of "determining, based on the first information transmitted by the at least one terminal device, whether there is a target interval in the at least one wireless channel parameter interval". Specifically, the target terminal device may be determined from the at least one terminal device through a reinforcement learning model based on the channel state between the wireless access network device and the at least one terminal device and the characteristic information of each terminal device.

[0315] As an example, the wireless access network device determines the richness of the data set, the special wireless channel parameter, the scarcity related information or the like based on the first information reported by the terminal, takes the channel state between it and the terminal device as an input of reinforcement learning, and uses the richness, the result containing the special wireless channel parameter and the scarcity related information as reward factors of reinforcement learning, respectively. It is also possible to use the real-time communication efficiency (e.g., channel power or the like, but not limited thereto) as a reward factor and output a terminal device (i.e., target terminal device) that requires model training through a reinforcement learning model.

[0316] In the embodiment of the disclosure, another optional implementation is provided for the step of "determining, based on the first information transmitted by the at least one terminal device, whether there is a target interval in the at least one wireless channel parameter interval". Specifically, it is possible to determine the priority separately corresponding to the at least one terminal device based on the characteristic information of each terminal device and determine a target terminal device from the at least one terminal device based on the priority.

[0317] Specifically, it is possible to fuse, based on a first weight separately corresponding to at least one piece of characteristic information of each terminal device, at least one piece of characteristic information of each terminal device to obtain the priority separately corresponding to the at least one terminal device.

[0318] In an example, the wireless access network device calculates the priority of each terminal device. The method of calculating the priority of thekthterminal device in thetthround is as following Math Figure 1:

[0319]

[0320] where is a richness factor having a range of (0-1), and is a scarcity factor having a range of (0-1); and optionally, .

[0321] represents the richness of the data set of the terminal device. For example, may be calculated in the following way: Math Figure 2

[0322]

[0323] where represents the data volume corresponding to the wireless channel parameter of the terminal device, and is the standard deviation. Specifically, represents the movement range of the terminal device in longitude, represents the movement range of the terminal device in latitude, represents the fluctuation range (standard deviation) of the channel condition of the wireless access network deviceucorresponding to the terminal devicek, and represents the average fluctuation range (average standard deviation) of channel conditions from the terminal devicekto different access network devices.Mrepresents the number of all terminal devices.

[0324] represents the scarcity related information of the wireless channel parameter of the terminal device. For example, may be represented by the data volume of this terminal device in the determined scarce RSRP interval (target interval). For example, may be calculated in the following way: Math Figure 3

[0325]

[0326] where represents the data volume of this terminal device in the determined scarce RSRP interval (target interval), andMrepresents the number of all terminal devices.

[0327] Further, terminal devices with the highest priority are selected as target terminal devices. The value of may be set by those skilled in the art according to the actual situation, or may be determined adaptively. However, it will not be limited to the embodiment of the disclosure.

[0328]

[0329] In the embodiment of the disclosure, an optional implementation is provided for step S104. Specifically, this optional implementation may include the following steps.

[0330] In step S1041, a distribution characteristic of the wireless channel parameter is received.

[0331] Optionally, the distribution characteristic of the wireless channel parameter includes, but not limited to, RSRP thermodynamic characteristic (e.g., density characteristic, fluctuation characteristic or the like, but not limited thereto), RSRP variance distribution or the like.

[0332] In step S1042, the at least target terminal devices is grouped based on the distribution characteristic of the wireless channel parameter, and a second weight of at least one group of target terminal devices is determined.

[0333] Optionally, the grouped terminal devices may include all target terminal devices, or may be terminal devices that have reported the second model parameter information among the target terminals (it is possible that the model parameters of some terminal devices during this training have not changed much and have not been reported), where .

[0334] Optionally, the distribution characteristic of the wireless channel parameter may be obtained based on the data volume (which may be interpreted as the density) of at least one wireless channel parameter interval, the mean value, variance and standard deviation of each wireless channel parameter value, the average value (which may be interpreted as the data fluctuation) of standard deviations of wireless channel parameters corresponding to different wireless access network devices.

[0335] Optionally, density and fluctuation characteristic sets (i.e., groups) are divided based on the RSRP density factor (RDF, radial distribution function) and the RSRP fluctuation factor (RFF, random Fourier feature). Specifically: Math Figure 4a

[0336]

[0337] where is the data volume corresponding to the (id)thRSRP interval corresponding to this terminal device. The RSRP interval number id is determined by comparing the average RSRP of samples in all data sets of this terminal device and the preset RSRP interval. If the average RSRP falls into a certain RSRP interval, the id is the interval number preset for this RSRP interval, and the data volume is counted through the preset RSRP interval, so that the data volume of the RSRP interval can be obtained through the segment number id. represents the data volume in the data sets of all terminal devices.

[0338] Or, may also be represented as: Math Figure 4b

[0339]

[0340] where represents the proportion of RSRP of different intervals in the wireless data pool of the wireless network access network as a reference, and represents the number of RSRP in the intervall: Math Figure 5a

[0341]

[0342] where represents the average fluctuation range (average standard deviation) of channel condition from the terminal device k to different access network devices, and represent the number of terminal devices selected in thetthround.

[0343] Or, may also be represented as: Math Figure 5b

[0344]

[0345]

[0346] The meaning of each parameter may refer to the description of Math Figure 2 and will not be repeated here.

[0347]

[0348] The grouping formula is: Math Figure 6a

[0349]

[0350] Further, the normalized group weight table may be predefined, and may be represented as: Math Figure 6b

[0351]

[0352] where represents the group weight corresponding to the group with an index of .

[0353]

[0354] The density and fluctuation characteristic sets may be divided according to Table 2 below. That is, the group IDs are mapped based on the fused density characteristics and fluctuation characteristics and the corresponding threshold of each group to obtain a group weight table:

[0355]

[0356] where , and corresponds to the threshold of theithgroup. In the predefined group weight table, the more obvious the generalized characteristic is, the larger the group weight is.

[0357] Optionally, if the generalized characteristic is more obvious, a larger group weight factor may be set, for example, , as shown inFigure6a. ， ，...， are weight factors of ， ，...， , respectively.

[0358] In step S1043-1, the second weight is adjusted based on the similarity among the groups of target terminal devices and / or the model training precision of the second model parameter to obtain a third weight of at least one group of target terminal devices.

[0359] In the embodiment of the disclosure, in order to overcome the blurring effect caused by grouping, additional fine group weight adjustment may be performed according to the correlation between groups and the difference in the number of terminal devices in each group.

[0360] Optionally, the similarity may be calculated by using the correlation coefficient of the average RSRP of each group of target terminal devices.

[0361] As an example, for each group, the correlation coefficient of this group with other groups is calculated by using the average RSRP of terminal devices in the group, to obtain a group most similar to this group.

[0362] It is assumed that, for , the weight factors of the groups and most similar to this group are and , respectively, and the similarity between and is .

[0363] Optionally, the weight of is adjusted, and the weight of is adjusted. That is, when the correlation coefficient is higher, other group weights will be affected more obviously.

[0364] In addition, when the difference in the number of terminal devices in each group is larger, a lower group weight adjustment coefficient may be set.

[0365] In an example, the example of fine group weight adjustment is as shown in Table 3 (taking and as an example).

[0366]

[0367] In step S1043-2,the second weight of at least one target terminal device in at least one group of target terminal devices is adjusted based on the model training precision fed back by the target terminal device to obtain a fourth weight of the at least one target terminal device.

[0368] Optionally, the model training precision may be represented by a verification error (or deviation). If the model verification error is smaller, the accuracy is higher. The model verification error may be verified by the test data set (also referred to as the verification data set) on the terminal device selected during each iteration. For example, the verification data set is a part (e.g., 20%) of the local data set, and the remaining 80% of the data set is used for model training, but not limited thereto.

[0369] For example, the verification error ( ) of thekthterminal device in thetthround may be calculated in the following way: Math Figure 7

[0370]

[0371] Or, is obtained by the method of calculating in the step S2024.

[0372] represents the data volume in the test data set of the terminal devicek; represents the number of wireless access network devices (e.g. , the serving cell and 3 neighboring cells, ) corresponding to the terminal devicek; represents the RSRP of thejthwireless access network device and theithsample predicted in thetthround based on the local model of the terminal devicek; and, represents the actual RSRP of thejthwireless access network device and theithsample. Optionally, at the end of thetthround of training, the terminal devicekreports the verification error to the wireless access network device. Thus, the wireless access network device can obtain the model training precision fed back by each target terminal device.

[0373] Optionally, the wireless access network device may calculate a confidence factor based on the model training precision fed back by the target terminal device, and then adjust the weight of the target terminal device based on the confidence factor to obtain the normalized dynamic adjustment weight , where is the group weight corresponding to the group number, and the confidence factor is calculated in the following way: Math Figure 8a

[0374]

[0375] In an example, the process of adjusting the weight of the target terminal device level according to the model training precision level is as shown in Figure 6b. On this basis, an example of UE weight adjustment is as shown in Table 4 (taking the fine group weight = 0.5 as an example):

[0376]

[0377] Further, the normalized sample density adjustment weight is obtained by using the data volume in the training data set of this user equipment: Math Figure 8b

[0378]

[0379] Further, the wireless access network device adjusts the model difference after thetthround of fusion: Math Figure 9

[0380]

[0381] where represents the number of reported gradients, represents thevthgradient of the model parameter reported by thekthterminal, and represents the gradient corresponding to in the model parameter issued by the (t-1)thwireless access network device. Or, , where . and are defined in the step S2044. Math Figure 10a

[0382]

[0383] When the normalized sample density adjustment weight is not taken into consideration, is calculated as follows: Math Figure 10b

[0384]

[0385] where represents the number of selected terminals in the cell of the mthbase station in the tthround.

[0386] The third weight of at least one target terminal device in at least one group of target terminal devices is adjusted based on the model training precision fed back by the target terminal device.

[0387] Or, the fourth weight may also be determined by clustering.

[0388] The RDF, RRF the fusion characteristic are calculated first: . For each UE in the tthround, the extracted characteristics are merged into a vector, which represents a sample point.

[0389] Then, the weight of each UE clustering is adjusted according to the average values of RDF and RRF:

[0390]

[0391] where represents the weight of the cluster m in the tthround; represents the initial weight of the cluster m; represents the average value of in the cluster m; and represents the average value of in the cluster m.

[0392] By taking a k-means clustering method as an example:

[0393] In step 1), K centre points are randomly selected to represent K clusters.

[0394] In step 2), the Euclidean distances between N sample points and K centre points are calculated.

[0395] In step 3), each sample point is classified to the cluster of the closest (smallest Euclidean distance) centre point, i.e., iteration 1.

[0396] In step 4), the mean of each cluster sample point is calculated to obtain K means, and the K means are used as new centres, i.e., iteration 2.

[0397] In step 5), the steps 2, 3 and 4 are repeated.

[0398] In step 6), after the convergence condition is satisfied, K centre points are obtained after convergence (the centre points remain unchanged).

[0399]

[0400] In step S1044,for the second model parameter information transmitted by each group of target terminal devices, the second model parameter is fused based on the third weight and / or the fourth weight.

[0401] Optionally, the model parameter is fused according to the fused model difference by FedAdam or other fusion methods to obtain a global model parameter at the end of this round of training. The process may be shown as follows: Math Figure 11a

[0402]

[0403] where: Math Figure 11b

[0404]

[0405] , Math Figure 11c

[0406]

[0407] where , and are hyper-parameters. The item may be very small and results in an unstable optimization behavior, and is used to maintain the stability of the numerical value.

[0408] In the method provided in the embodiment of the disclosure, the reported model parameter is fused by using grouping, feedback mechanism, group level weight adjustment, terminal device level weight adjustment and other means, and the local model is filtered by using these mechanisms, so that it is advantageous for global training, the training precision is improved, and the precision of radio map construction is thus improved.

[0409] Optionally, the wireless access network device fuses the model and then uploads it to the central server. The central service may fuse the global model with a plurality of fused model received from the wireless access network device by at least one of the above model parameter fusion method or by any method (e.g., average aggregation).

[0410] Optionally, the global model parameter may also be quantized, for example, quantized from 32 bits to 16 bits, thereby reducing the overhead of transmission from the wireless access network device to the terminal device.

[0411] In the embodiment of the disclosure, in the process of multiple rounds of model issuing, training, uploading and fusion during the radio map construction, the steps executed by the wireless access network device further include at least one of the following:

[0412] (1) The first model parameter is compressed, and the compressed first model parameter information is transmitted to the terminal device.

[0413] Optionally, a first element bit number reduction degree of the first model parameter is determined, the element bit number of the first model parameter is reduced based on the first element bit number reduction degree to obtain the compressed first model parameter information, and the compressed first model parameter information is transmitted to the at least one target terminal device.

[0414] The model parameter compression method may be referred to as a model quantization method.

[0415] Optionally, in the process of uploading model parameters or gradients, the uploaded model parameters or gradients are quantized, thereby reducing the bit number of elements and reducing the data traffic. According to the uplink data rate, under the condition of ensuring certain precision, different terminal devices reserve a larger number of bits when the wireless environment is good, and reserve a smaller number of bits when the wireless environment is poor.

[0416] Optionally, the wireless access network device may transmit the fused first model parameter information to all terminal devices in this cell before or after selecting the target terminal device, or may transmit the fused model parameter to the selected target terminal device after selecting the target terminal device.

[0417] That is, in the first round, the issuing mode may be broadcasting, multicasting, unicasting, anycasting, etc.

[0418] In an example, the format of the issued model parameter is as shown in Table 5:

[0419]

[0420] It should be understood that the terminal device may perform decompression and then perform corresponding training after receiving the model parameter transmitted by the wireless access network device.

[0421] (2) The second model parameter information transmitted by the at least one target terminal device is decompressed, and the decompressed second model parameter is fused.

[0422] The second model parameter information transmitted by the at least one target terminal device may be compressed, so the wireless access network device may perform decompression and then perform fusion.

[0423] (3) It is determined, based on the updated first model parameter, whether the radio map is constructed.

[0424] After each round of model parameter fusion or issuing, the wireless access network device may determine whether the radio map is constructed; if the radio map is constructed, the next round of process may be terminated; otherwise, the next round of terminal device selection, model parameter issuing, training, uploading, fusion and other processing is continued.

[0425] In the method executed by a wireless access network device provided in the embodiment of the disclosure, model training may be performed in combination with the richness, specificity, scarcity and other characteristics of the data set of the terminal side, thereby improving the model convergence speed. Moreover, after the terminal devices that report model parameters in each round are grouped, parameter fusion is performed based on the feedback, weight adjustment or the like, so that the convergence of model is accelerated.

[0426] In the embodiment of the disclosure,in step S103, the receiving second model parameter information transmitted by the at least one target terminal device may include: receiving compressed second model parameter information transmitted by the at least one target terminal device, the compressed second model parameter information being obtained by compressing the second model parameter information based on the uplink data rate of the at least one target terminal device and / or the model training precision of the second model parameter, the uplink data rate being determined based on channel environment information between the at least one target terminal device and the wireless access network device.

[0427] The specific implementation of this process may refer to the following description of the terminal device side and will not be repeated here.

[0428] In the embodiment of the disclosure, the compressed second model parameter information being obtained by compressing the second model parameter information based on the uplink data rate of the at least one target terminal device and / or the model training precision of the second model parameter includes at least one of the following situations:

[0429] the compressed second model parameter information is obtained by determining an element compression proportion of the second model parameter based on the uplink data rate and determining elements of the element compression proportion in the second model parameter;

[0430] the compressed second model parameter information is obtained by determining a second element bit number reduction degree of the second model parameter based on the uplink data rate and reducing the element bit number of the second model parameter based on the second element bit number reduction degree; and

[0431] the compressed second model parameter information is obtained by determining a model parameter cluster to be currently transmitted in the second model parameter based on the model training precision and determining a gradient to be currently transmitted in the model parameter cluster to be currently transmitted based on the model training precision.

[0432] The specific implementation of this process may also refer to the following description of the terminal device side and will not be repeated here.

[0433]

[0434] An embodiment of the disclosure provides a method executed by a terminal device in a communication system. As shown inFigure 5,the method includes the following steps.

[0435] In step S201, first information is transmitted to a wireless access network device, the first information including position related information and / or wireless channel parameter related information of the terminal device.

[0436] In the embodiment of the disclosure, the wireless access network device may include, but not limited to, a base station, a relay node, a combination of the both or the like in the cell where the terminal device is located.

[0437] In the embodiment of the disclosure, the wireless channel parameter includes at least one of the following: an RSRP value; an RSS value; a PSD value; an SINR; and, CSI. The wireless channel parameter further includes at least one of the following: frequency bands; time units; or the like.

[0438] Optionally, the terminal device acquires the position information and the wireless channel parameter of the corresponding position in real time, and stores them in the local data set in the chronological order.

[0439] Optionally, the terminal extracts (or counts) wireless channel parameter characteristics from the wireless channel parameter and then stores or reports them.

[0440] Optionally, the terminal device may also acquire the wireless channel parameters of a plurality of wireless access network devices, for example, including but not limited to, the base station of this cell, the base station of the neighboring cell or the like, and the subsequent processing process is similar and will not be repeated.

[0441] In step S202, the first model parameter information transmitted by the wireless access network device is received.

[0442] Optionally, this step may also be performed before or at the same time as the step S201.

[0443] The first model parameter information may be interpreted as the information related to global model parameters determined by the wireless access network device.

[0444] Optionally, the model parameter information may refer to all or some parameters of the model, or variations with a specific model (e.g., before training or before fusion), or compressed model parameters, etc.

[0445] In step S203, in a case of being determined as a target terminal device based on the first information by the wireless access network device, the first model parameter is trained to obtain a second model parameter.

[0446] The second model parameter information may be interpreted as the information related to local model parameters trained by the terminal device.

[0447] In the embodiment of the disclosure, the wireless access network device will select a terminal device model training. The target terminal device refers to the terminal device selected by the wireless access network device. Specifically, the wireless access network device may select suitable terminal devices with reference to the first information transmitted by the at least one terminal device, and use high-quality data sets on these terminal devices to perform model training, thereby improving the model convergence speed.

[0448] If the current terminal device is determined as a target terminal device by the wireless access network device, a notification from the wireless access network device may be received. Thus, the terminal device may train the model based on the local data set to obtain the second model parameter information.

[0449] In step S204, the second model parameter information is transmitted to the wireless access network device.

[0450] In the embodiment of the disclosure, a radio map construction method based on federated learning is adopted. The terminal device firstly receives the neural network model initialized by the wireless access network device. The wireless access network device selects a target terminal device participating in this round of federated learning according to the first information transmitted by the terminal device in each round. The selected target terminal device performs training locally based on the first model parameter issued by the wireless access network device, and then uploads the trained second model parameter information to the wireless access network device. The wireless access network device fuses the second model parameter uploaded by the terminal device to update the first model parameter, and transmits the fused model parameter to the terminal device. When the training reaches a training end condition (for example, the number of rounds reaches the set value, or the model converges, or the model precision tends to be stable, or the model reaches the predetermined effect, or the like), the model training for predicting the radio map is completed; otherwise, the next round of processing is continued.

[0451] That is, in the embodiment of the disclosure, in the process of training the model for predicting the radio map, the steps S201 to S204 may be performed for multiple times (i.e., in the above multiple rounds), and the repeated processing process will not be repeated.

[0452] The description of the radio map constructed in the embodiment of the disclosure may refer to the above description, and will not be repeated here.

[0453] In embodiments of the disclosure, being determined as a target terminal device based on the first information by the wireless access network device includes:

[0454] being determined as a target terminal device based on characteristic information by the wireless access network device, the characteristic information being determined based on the first information;

[0455] wherein the characteristic information includes at least one of the following:

[0456] the richness of the wireless channel parameter corresponding to the terminal device, the richness being related to a data volume and / or data fluctuation range; and

[0457] scarcity related information of the wireless channel parameter corresponding to the terminal device in at least one wireless channel parameter interval, the scarcity related information being related to the data volume in the at least one wireless channel parameter interval.

[0458] Optionally, the characteristic information may further include:

[0459] whether the wireless channel parameter corresponding to each terminal device includes a target wireless channel parameter, the target wireless channel parameter being related to at least one of the frequency of occurrence of each wireless channel parameter value, the variance of the wireless channel parameter value and the mean value of the wireless channel parameter value.

[0460] The detailed description of the characteristic information may refer to the above description, and will not be repeated here.

[0461] In the embodiment of the disclosure, the first information may include at least one of the following information and / or the richness of at least one of the following information, or the information corresponding to frequency bands and / or time units and / or different wireless access network devices, including at least one of the following:

[0462] the data volume corresponding to the wireless channel parameter;

[0463] the difference between the maximum longitude and the minimum longitude among longitudes determined based on at least one piece of position information;

[0464] the difference between the maximum latitude and the minimum latitude among latitudes determined based on at least one piece of position information;

[0465] the difference between the maximum altitude and the minimum altitude among altitudes determined based on at least one piece of position information;

[0466] the difference between the maximum wireless channel parameter value and the minimum wireless channel parameter value among wireless channel parameter values corresponding to at least one piece of position information;

[0467] a first preset number of maximum distances among the respective distances determined based on each piece of position information;

[0468] a second preset number of minimum distances among the respective distances determined based on each piece of position information;

[0469] a position range determined based on each piece of position information;

[0470] a mean value of each piece of position information;

[0471] a mean value of each wireless channel parameter value;

[0472] a data volume of at least one wireless channel parameter interval;

[0473] a standard deviation of each piece of position information;

[0474] a standard deviation of each wireless channel parameter value; and

[0475] an average value of standard deviations of wireless channel parameters corresponding to different wireless access network devices.

[0476] The detailed description of various first information may refer to the above description, and will not be repeated here.

[0477] In the embodiment of the disclosure, considering the radio map construction based on federated learning, the terminal device and the wireless access network device need to perform model transmission during the federated training process. In a federated learning scenario with many model parameters, limited communication bandwidth and many terminal devices, the data volume received by the wireless access network device is very large. In order to prevent from bringing a great communication pressure on the network and seriously affecting the overall training efficiency, the terminal device may dynamically compress the model parameter, so that the uplink communication overhead can be further reduced under the condition of satisfying the model precision.

[0478] As an example, it is assumed that the radio map is constructed to obtain a global model. This model may accurately predict the RSRP at any position in the corresponding region and has an optical transmission overhead. Thus, the minimum transmission overhead may be represented as , where the transmission overhead includes the overhead of wireless transmission between the wireless access network device and the terminal device, and T represents the iteration round. For each iteration, each wireless access network device will broadcast the global model to selected terminal devices, and each selected terminal device needs to upload the local model the wireless access network device for global model aggregation. is the number of base stations, and is the downlink transmission overhead of the tthiteration. is the uplink transmission overhead of terminal devices in the tthiteration. The minimum transmission overhead may also be represented as , where represents the data transmission size of the compressed global model, and represents the data transmission size of the compressed local model. In addition, is used for ensuring the model precision. represents the loss function of the global model on all selected terminal devices, represents the final global model, and , where represents the local loss function of the kthterminal device and may be evaluated by a normalized mean square error (NMSE). , where represents the amount of local samples of the data set of the kthterminal device, represents the real value of the mthbase station corresponding to the jthsample, and represents the predicted value of the mthbase station corresponding to the jthsample. Optionally, if is less than a precision threshold , the construction of the radio map may be terminated.

[0479] In the embodiment of the disclosure, the radio map construction based on federated learning requires multiple times of iterative training, and the transmission overhead increases with the increase of the number of iterations and the number of terminals participating in each iteration.

[0480] On this basis, in the embodiment of the disclosure, an optional implementation is provided for the step S204. Specifically, this optional implementation may include the following steps.

[0481] In step S2041, an uplink data rate of the wireless access network device is determined based on channel environment information between the terminal device and the wireless access network device.

[0482] Optionally, the channel environment information includes, but not limited to, channel state, channel bandwidth or the like.

[0483] In an optional implementation, it is assumed that the wireless access network device is configured with antennas and the signal transmitting end has single-antenna terminal devices. If all terminal devices use the same time-frequency resource, in the case of ideal channel state information (CSI), the uplink data rate of thekthterminal device may be represented as: Math Figure 12

[0484]

[0485] where represents the channel bandwidth; is a spatial channel matrix between the terminal device and the wireless access network device; represents thekthcolumn of elements of the matrixH; represents the conjugate transpose of ; is the average transmitting power of the terminal; and, ) is the additive white Gaussian noise (AWGN), where is a diagonal matrix with a size of .

[0486] In step S2042, model training precision of the second model parameter is determined.

[0487] In the embodiment of the disclosure, the model training precision of thekthterminal device in thetthround includes, but not limited to, the local training model verification error , the global model verification error in the (t-1)thround, the difference between the local training model verification error in thetthround and the global model verification error in the (t-1)thround or the like.

[0488] Optionally, may be determined based on the above formula (7).

[0489] may be determined based on the following formula: Math Figure 13a

[0490]

[0491] where represents the data volume in the test data set of the terminal devicek; represents the number of wireless access network devices corresponding to the terminal devicek; represents the RSRP of thejthwireless access network device and theithsample predicted based on the local model of the (t-1)thround; and, represents the actual RSRP of thejthwireless access network device and theithsample.

[0492] Optionally, the verification error function may be represented as: Math Figure 13b

[0493]

[0494] where represents the data volume in the test data set of the terminal devicekin the tthround, represents the number of wireless access network devices corresponding to the terminal device, and represents the actual RSRP.

[0495] The local model verification error and the global model verification error may be represented as follows, respectively:

[0496]

[0497]

[0498] where represents the predicted RSRP obtained using the local model based on the local test data set, and represents the predicted RSRP obtained using the global model based on the local test data set.

[0499] may be determined based on the following formula: Math Figure 14

[0500]

[0501] In step S2043, the second model parameter information is compressed based on the uplink data rate and / or model training precision.

[0502] Optionally, the federate learning compresses the model by reducing the data volume uploaded by the terminal device, for example, a model thinning method, a model quantization method, but not limited thereto. The terminal device may dynamically compress the model based on these compression methods by using the calculated uplink data rate and the local current model training precision, thereby reducing the uplink communication overhead.

[0503] In step S2044, the compressed second model parameter information is transmitted to the wireless access network device.

[0504] In the method provided in the embodiment of the disclosure, in view of the problem of high communication overhead for the model parameter, the model parameter is dynamically compressed according to the wireless channel environment, channel bandwidth, model training precision and other conditions, so that the communication overhead of uploading the model parameter is effectively reduced, and the efficiency of constructing the radio map is improved. For example, it can be applied to a wireless environment based on 5G mobile communication.

[0505] In the embodiment of the disclosure, an optional implementation is provided for the step S2043. Specifically, this optional implementation may include the following steps.

[0506] (1) An element compression proportion of the second model parameter is determined based on the uplink data rate, and elements of the element compression proportion in the second model parameter are determined to obtain the compressed second model parameter information.

[0507] The model parameter compression method may be referred to as a model sparsity method.

[0508] Optionally, in the process of uploading model parameters or gradients (variations), based on a Top-k sparse method, a depth gradient compression method, sparse distribution compression or other methods, the terminal device selects a certain proportion of elements in the model parameter or gradient, and uploads only these elements to the wireless access network device; and, fuses these elements on the wireless access network device side according to the actual positions of the elements, and then updates the model. Under the condition of ensuring certain precision, different terminal devices upload different proportions of elements according to the uplink data rate. When the wireless environment is good, the proportion in the model parameter or gradient is increased; and, when the wireless environment is poor, the proportion in the model parameter or gradient is decreased.

[0509] The dynamic model compression method combined with the channel environment information will be described below by taking uploading the model gradient and using the Top-k thinning method as an example.

[0510] It is assumed that is the energy consumed by the terminal devicekto sparse the gradient in thetthround: Math Figure 15

[0511]

[0512] where represents the total number of bits of communication of thekthterminal device intthround, represents the transmitting power, and represent the uplink data rate. By using the "Top-k" sparse method, the value and position of the non-zero gradient in the sparsified flattening tensor may be transmitted. It is assumed that A represents the data precision, for example, the single-precision floating point A = 32, and the double-precision floating point A = 64. The terminal device k may use A bits to represent the absolute value of each non-zero gradient, and the position of the gradient in the vector and the symbol of the gradient may be combined and represented by A precision, then: Math Figure 16

[0513]

[0514] Then, the compression proportion may be obtained: Math Figure 17

[0515]

[0516] The terminal determines the elements of in the second model parameter gradient and uploads the compressed model parameter gradient to the wireless access network device.

[0517] (2) An element bit number reduction degree of the second model parameter is determined based on the uplink data rate, and the element bit number of the second model parameter is reduced based on the second element bit number reduction degree, to obtain the compressed second model parameter information.

[0518] The model parameter compression method may be referred to as a model quantization method.

[0519] Optionally, in the process of uploading model parameters or gradients, the uploaded model parameter or gradient is quantized, thereby reducing the bit number of elements and reducing the data traffic. According to the uplink data rate, under the condition of ensuring certain precision, different terminal devices reserve a larger number of bits when the wireless environment is good, and reserve a smaller number of bits when the wireless environment is poor.

[0520] (3) Hybrid method

[0521] In order to further reduce the size of the uploaded data volume, the model sparsity method and the quantization method may be combined, that is, the elements in the sparse model are quantized. Optionally, the proportion of models or gradients and / or the number of bits for quantization and compression may be calculated by a mathematical method. According to the uplink data rate or wireless environment, under the condition of ensuring certain precision, different terminal devices increase the proportion in the model parameter or gradient and reserve a larger number of bits during quantization and compression when the wireless environment is good; and, decrease the proportion in the model parameter or gradient and reserve a smaller number of bits when the wireless environment is poor.

[0522] (4) A model parameter cluster to be currently transmitted is determined in the second model parameter based on the model training precision, and a gradient to be currently transmitted is determined in the model parameter cluster to be currently transmitted based on the model training precision, to obtain the compressed second model parameter information.

[0523] Since each selected terminal device needs to transmit an updated model parameter to the wireless access network device during each training iteration of the global model, considering that the deep neural network (DNN) model generally has millions of parameters, in order to reduce the communication overhead, an embodiment of the disclosure provides a model adaptive gradient compression method, wherein the gradient may be described as : Math Figure 18

[0524]

[0525] where represents the local model parameter of thekthUE obtained by thetthiteration; represents the global model parameter transmitted by the wireless access network in thetthiteration; ; ; represents the number of parameters in the model parameter; and represents the learning rate.

[0526] Optionally, the terminal may adaptively compress the model parameter based on the model training precision and .

[0527] Specifically, the terminal device may determine the number of model parameter clusters (or referred to as gradient clusters) according to the model training precision.

[0528] For thekthterminal device in thetthround of training, the terminal device determines the number of model parameter clusters uploaded currently according to the local model verification error in this round and the difference in the model verification error. If the local model verification error is less than a threshold and the difference in the verification error is less than a threshold , the numberNcof uploaded model parameter clusters will be decreased. And / or, if the local model precision remains unchanged in a very long period of time (in a certain range), the numberNcof uploaded model parameter clusters will be decreased.

[0529] Optionally, the terminal device determines to freeze layers, selects and quantizes the maximum number of parameter variations, and uploads the compression result to the serving wireless access network device of the terminal device. , and the quantization precision are determined based on the model verification error. If the model verification error is smaller, the parameter variations reported to the wireless access network device are less.

[0530] As an example, as shown in Figure 6c, the network structure has a total of 7 model parameter clusters (layers). When and , the parameters of five clusters from down to up may be frozen, and only the parameters of first two clusters are transmitted. For example, in Figure 6c, if the four bottom Mixer modules (clusters) transmitted in combination and the bottom one layer in the Mixer module transmitted separately are frozen and only the top two layers in the Mixer module transmitted separately are transmitted, the model parameter clusters to be currently transmitted are top two clusters.

[0531] Further, the terminal device may determine the number of gradients in the model cluster.

[0532] For thekthterminal device in thetthround of training, if , where represents the number of parameters of the model, represents theithmodel parameter of thekthterminal device after training in this round, represents theithmodel parameter received from the wireless access network device before training, represents the learning rate of the model, , and represents the reset threshold. It indicates that the local training model parameter is basically the same as the model parameter issued by the wireless access network device in the previous round, so the terminal may not report the second model parameter information in this round to the wireless access network device. Otherwise, the terminal selects, from the model parameter gradient vector, a predetermined number (e.g., ) of maximum gradients for reporting. Based on the local model verification error and the difference in the verification error, if the local training mode is more accurate and the difference from the local model is smaller, there are a smaller number of gradients to be reported to the wireless access network device, that is, the required is smaller.

[0533] Optionally, according to the model training precision of the local model, the selection of maximum gradients needs to satisfy the following condition: , where represents the selected gradient, wherein:

[0534]

[0535] Math Figure 19

[0536]

[0537] Optionally, the model parameter variation is defined as:

[0538]

[0539] where represents the local model parameter of thekthUE obtained in thetthiteration, and represents the global model parameter transmitted by the wireless access network device in thetthiteration. After layers are frozen, the parameters of the frozen layers are set as 0. If the local model parameter of the UE is updated as and the global model parameter is updated as , the model parameter variation is at this time. The selection of maximum parameter variations by the UE needs to satisfy the following condition:

[0540]

[0541]

[0542]

[0543]

[0544] where , , represents the number of model parameters, and and represent the thresholds of the local model verification error and the verification error difference, respectively. Based on the selected maximum parameter variation, is updated as , as shown in Figure 6d.

[0545] Further, the terminal device reports the quantified selected gradients, the positions and symbols of gradients, and the verification error to the wireless access network device.

[0546] Optionally, the gradients and the verification error are quantized as 16 bits (that is, the quantization precision is 0.000015). That is, the positions and symbols of gradients are combined and quantized by 16 bits (for example, the first bit represents the symbol, and the remaining bits represent the gradient position).

[0547] During quantization, when and , a higher quantization precision is used for the quantization of ; otherwise, a lower quantization precision is used for the quantization of . After quantization, is updated as , as shown in Figure 6d.

[0548] During quantization and compression, the terminal device may also perform adaptive quantization according to the channel condition. When the channel condition is good, a higher precision is used for quantization; otherwise, a lower precision is used for quantization. If it is assumed that the quantization precision of thekthterminal device in thetthround is , the quantization precision satisfies the following condition: Math Figure 20

[0549]

[0550] where represents the energy demand,Rrepresents the channel condition, and represents the UE power.

[0551] Optionally, the format for gradient reporting by the UE may be as shown in Table 6:

[0552]

[0553] Another method to determine the compression proportion based on the model precision, so that the compressed gradient can satisfy the following condition:

[0554]

[0555] where , and is the quantized and compressed gradient.

[0556] It should be understood that the wireless access network device may perform decompression and then perform corresponding fusion after receiving the second model parameter information transmitted by the terminal device.

[0557] In the embodiment of the disclosure, in multiple rounds of model issuing, training, uploading and fusion during the radio map construction, at least one optional implementation is provided for the step S203. Specifically, this step may include: enhancing the wireless channel parameter, and training the first model parameter based on the enhanced wireless channel parameter.

[0558] In the embodiment of the disclosure, the selected terminal device may perform data enhancement on the wireless channel parameter in the local data set, and then perform model training based on the enhanced data.

[0559] Optionally, the wireless channel parameter is enhanced based on the distance and / or angle from the wireless access network device, and the model is trained based on the enhanced wireless channel parameter.

[0560] The distance from the wireless access network device includes a two-dimensional distance and / or a three-dimensional distance.

[0561] As an example, the terminal device knows the position of the wireless access network device, and calculates the information of the two-dimensional spatial distance and horizontal position including the angle from the wireless access network device. If it is assumed that the altitude of the wireless access network device is 50, the information of the distance between the terminal device and the wireless access network device in the three-dimensional space is calculated. Logarithmic values are calculated for all distances at the same time, thereby increasing the dimension of training data. Then, the terminal device performs local model training based on the enhanced data.

[0562] Or, optionally, the terminal may also calculate the difference (e.g., the difference in RSRP) between every two groups of wireless channel parameters in the data set and the distance between two positions. If the difference in wireless channel parameter is smaller and the distance is less than a threshold (e.g., 3 meters), interpolation is performed in the two groups of wireless channel parameters, for example, interpolating a group of data every half a meter; and, if the difference in wireless channel parameter is larger but the distance is very small, it indicates that this position is blocked. At this time, the data may be enhanced by oversampling. Then, the terminal device performs local model training based on the enhanced data.

[0563] In the embodiment of the disclosure, the input characteristic of the AI model may also be determined based on the distance and / or angle from the wireless access network device. By taking the wireless channel parameter RSRP as an example, considering the factors affecting the RSRP, the two characteristics can better reflect the geometrical relationship between the position and the RSRP, thereby greatly improving the performance of the AI model.

[0564] For example, referring to Figure 6e, the distance relationship between the RSRP and different wireless access network devices is shown. It can be seen that, even if the distance is increased, the fluctuation of the RSRP demonstrates the common range and upper limit for most samples, reflecting the path loss characteristics of wireless access network devices. However, different single fluctuation ranges are demonstrated at different distances, reflecting the diversity of the environments of different wireless access network devices.

[0565] For another example, referring to Figure 6f, the angle relationship between the RSRP and different wireless access network devices is shown. With the change of the angle, a trend similar to distance can also be seen.

[0566] Optionally, in order to improve the performance of the model to the greatest extent, the AI model provided in the embodiment of the disclosure considers global characteristics and single characteristics equally. In the embodiment of the disclosure, the richness of the wireless channel parameter corresponding to the terminal device is determined in at least one of the following ways:

[0567] the richness of the wireless channel parameter corresponding to the terminal device is determined based on at least one of the data volume of the wireless channel parameter, the fluctuation range of the position and the fluctuation range of the wireless channel parameter corresponding to the terminal device, as well as at least one of the data volume of wireless channel parameters, the fluctuation range of positions and the fluctuation range of wireless channel parameters corresponding to all terminal devices; and

[0568] the richness of the wireless channel parameter corresponding to this terminal device is determined based on the first information and a corresponding threshold.

[0569] In the embodiment of the disclosure, the wireless channel parameter corresponding to the terminal device including a target wireless channel parameter is determined based on a situation where the variance of the wireless channel parameter value and / or the mean value of the wireless channel parameter value is greater than a first threshold and the frequency of occurrence of at least one wireless channel parameter value is less than a second threshold.

[0570] In the embodiment of the disclosure, the wireless channel parameter corresponding to the terminal device is determined based on the data volume of the terminal device in a target interval in a case where the scarcity rated information in at least one wireless determines that a target interval exists in at least one wireless channel parameter interval based on the first information transmitted by the at least one terminal device, the data volume corresponding to the wireless channel parameter of at least one terminal device in the target interval satisfies a preset condition.

[0571] In the embodiment of the disclosure, it is determined based on the wireless channel parameter interval corresponding to the maximum difference that there is a target interval in the at least one wireless channel parameter interval; the difference corresponding to each wireless channel parameter interval in the at least one wireless channel parameter interval is a difference between the proportion of the data volume corresponding to the wireless channel parameter interval in the total data volume and a predetermined proportion of this wireless channel parameter interval; and, the proportion of the data volume corresponding to the wireless channel parameter interval in the total data volume is determined based on the first information.

[0572] In the embodiment of the disclosure, the target terminal device is determined in at least one of the following ways:

[0573] the target terminal device is determined from the at least one terminal device through a reinforcement learning model based on the channel state between the wireless access network device and the at least one terminal device and the characteristic information of each terminal device; and

[0574] the priority separately corresponding to the at least one terminal device is determined based on the characteristic information of each terminal device, and the target terminal device is determined from the at least one terminal device based on the priority.

[0575] In the embodiment of the disclosure, the priority separately corresponding to the at least one terminal device is determined by fusing at least one piece of characteristic information corresponding to each terminal device based on the first weight separately corresponding to at least one piece of characteristic information of each terminal device.

[0576] In the embodiment of the disclosure, after the second model parameter information is transmitted to the wireless access network device, the second model parameter information is fused based on a third weight and / or a fourth weight, wherein the third weight is obtained by adjusting the second weight based on the similarity among the groups of target terminal devices and / or the model training precision of the second model parameter; the fourth weight is obtained by adjusting the second weight of at least one target terminal device in at least one group of target terminal devices based on the model training precision fed back by the target terminal device; and, the second weight is the weight of at least one group of target terminal devices determined after grouping the at least one target terminal device based on the distribution characteristic of the wireless channel parameter.

[0577] In the embodiment of the disclosure, the receiving first model parameter information transmitted by the wireless access network device includes:

[0578] receiving compressed first model parameter information transmitted by the wireless access network device, the compressed first model parameter information being obtained by determining a first element bit number reduction degree of the first model parameter and reducing the element bit number of the first model element based on the first element bit number reduction degree.

[0579] The implementation principles and steps of the method executed by a terminal device provided in the embodiment of the disclosure correspond to those of the method executed by a wireless access network device, and have the corresponding technical effects. The details of the terminal device side may specifically refer to the above description of the corresponding method on the wireless access network device side and will not be repeated here.

[0580] In the radio map construction method provided in the embodiment of the disclosure, in order to improve the accuracy of the training model, the RSRP fluctuation and scarcity vectors are used, and the selection priority is formulated to select a suitable UE to participate in local model training. Secondly, an adaptive method of selecting gradient clusters (e.g., the above model parameter clusters, the number of clusters being determined by the model training precision) and compressing gradients are used, and the quantized gradient and the model training precision are reported, so that the overhead of uploading or downloading model parameters between the UE and the gNB is reduced. In addition, in order to accelerate the model convergence, a model parameter fusion method based on the data distribution characteristic and confidence is used, UEs are grouped according to the generalized characteristic, and group-level and UE-level accurate weight adjustment is correspondingly performed to obtain accurate adjustment weights, thereby ensuring the accuracy of training. Compared with the existing radio map construction method, through simulation based on the open data set, on the basis of protecting user privacy, the training overhead is significantly reduced, the training grain is significantly improved, and the precision loss is significantly reduced.

[0581] Optionally, the radio map construction method is specifically a radio map construction method based on a distribution characteristic vector (DCV). The DCV is used to extract the characteristics of the model and the data set, including position characteristics and corresponding wireless environment characteristics. In order to improve the precision of the radio map, according to the richness, scarcity and other characteristics of data defined in the DCV, UEs with a high data quality (the high data quality may mean a large data volume and a wide geographical position coverage, including special positions that other UEs do not have, etc.) may be selected for model training on the UE side, so that the accuracy of the radio map can be improved. Moreover, in order to reduce the overhead of model parameter interaction, the gradient to be interacted can be selected according to the model deviation estimation (e.g., the above model training precision or verification error) defined in the DCV. The model fusion method based on a model confidence factor and a data representative factor (used for weight adjustment) further improves the precision of the radio map. For example, the above can be construed as the confidence factor, and the above can be construed as the data representative factor.

[0582]

[0583] Based on at least one of the above embodiments, in an embodiment of the disclosure, as shown in Figure 7a, an overall radio map construction method is provided. Specifically, for the UE side, data collection, model, model parameter compression, model parameter uploading and other operations are included; for the gNB side, UE selection, model local fusion, UE selection result notification, uploading the model parameter to the server and other operations are included; and, for the central server, global model broadcasting, global model parameter fusion and other operations are included, throughout the whole radio map construction process based on learning.

[0584] The situation of UE selection based on a high quality characteristic is shown in Figure 7b. It can be seen that, compared with the reference, UEs with a large RSRP fluctuation (rich data) and UEs with data in scarce regions can be selected as UEs with high data quality by the radio access network device.

[0585] Based on at least one of the above embodiments, in an embodiment of the disclosure, as shown in Figure 7c, a radio map construction process is provided. Specifically, this process mainly includes the following steps.

[0586] The wireless access network device issues an initial global model to terminal devices. For each roundt, the terminal devices collect wireless channel parameters in real time to obtain data sets, extract data characteristics of the wireless channel parameters and report them to the wireless access network device. The wireless access network device selects a target terminal device from the terminal devices according to these reported data, and notifies the selection result to the corresponding terminal devices. The selected terminal device performs data enhancement on the local data set, and performs model training to obtain a second model parameter . Then, the second model parameter is subjected to gradient compression to obtain the compressed model gradient , and then transmitted to the wireless access network device. The wireless access network device updates the first (global) model parameter , and determines whether the radio map has been constructed; if the radio map has not been constructed, the next round of terminal device selection, model parameter issuing, training, uploading and fusion is continued; and, if the radio map has been constructed, the trained radio map model is obtained.

[0587] For the data enhancement step, the data is enhanced according to the RSRP difference and the distance between two adjacent samples. If the RSRP difference between two samples is very small but the distance is very large, the samples are interpolated. If the RSRP difference between two samples is very large but the distance is very small, the model learns special situations by oversampling.

[0588] Optionally, by taking four groups of RSRP measured from the serving cell and three neighboring cells as an example, the model may use Mixer, which has a flexible trunk-branch network (e.g., TBNet) structures in which the trunk part and the branch part are separately used for learning global and local characteristics, and has low overhead and strong data fitting capability. As an example, the model may include a seven-layer Mixer. Referring to Figure 6g, in the bottom layer of the model, data characteristics are extracted by a unified four-layer Mixer; while in the top layer of the model, the PRRP values measured from different gNBs are independently predicted by using four groups of three-layer Mixer. This mixer framework may be flexibly deployed according to the network configuration and applied in various network scenarios.

[0589] Based on at least one of the above embodiments of the disclosure, in an embodiment of the disclosure, as shown in Figure 7d, another radio map construction process (based on a distribution characteristic vector (DCV)) is provided. Specifically, this process mainly includes the following steps.

[0590] Terminal devices collect wireless channel parameters in real time to obtain data sets, extract distribution characteristic vectors based on the data sets to obtain distribution characteristic vectors 1 (e.g., the following DCV1), and report them to the wireless access network device. The wireless access network device selects a target terminal device from the terminal devices according to these reported data and notifies the selection result to the corresponding terminal devices. The selected terminal device performs model training on the local data set, then performs gradient compression on the second model parameter to obtain a distribution characteristic vector 2 (e.g., the following DCV2), and transmits it to the wireless access network device. The wireless access network device fuses the model parameter and determines whether the radio map has been constructed. If the radio map has not been constructed, the next round of terminal device selection, model parameter issuing, training, uploading and fusion is continued; and, if the radio map has been constructed, the trained radio map model is obtained.

[0591]

[0592] In an embodiment of the disclosure, as shown in Figure 8, an architecture of a radio map construction device is provided. Specifically, the radio map construction device includes the linkage of a terminal device and a wireless access network device. A data acquisition entity, a local data set, a data counting (extraction) entity, an AI model, and a model compression entity may be deployed in the terminal device. An AI model initial parameter entity, a data pool maintenance entity, a terminal device selection entity, a model parameter fusion entity, and a radio map construction end determination entity are deployed in the wireless access network device.

[0593] The data acquisition entity acquires real-time data (position and corresponding wireless channel parameter value) and stores the data in the local data set. The data counting (extraction) entity counts (extracts) the corresponding wireless channel parameter in the local data set. When the terminal device is selected by the wireless access network device, this terminal device trains the model parameter according to the local data set. Then, the model compression entity dynamically compresses the model parameter according to the channel environment, bandwidth, and other conditions, and uploads the compressed model parameter to the model fusion entity of the wireless access network device.

[0594] The AI model initial parameter entity in the wireless access network device issues the initial parameter of the model to the terminal device. The data pool maintenance entity maintains the wireless channel parameter reported by the terminal device. The terminal device selection entity selects a suitable terminal device for model training. The model fusion entity fuses the model parameters reported by different terminal devices and broadcasts the fused model to the terminal device. The radio map construction end entity determines whether the radio map has been constructed.

[0595] In an embodiment of the disclosure, as shown in Figure 9, another radio map construction process is provided. Specifically, this process mainly includes the following steps.

[0596] In the initialization process:

[0597] A BS collects UEs' capabilities, including computing capability, storage capacity or the like.

[0598] The BS determines, according to the UEs' capabilities, UEs' qualification to participate in radio map construction.

[0599] The BS notifies a radio map construction task to the qualified UEs.

[0600] UEs agree to participate in model training and testing and report data characteristics.

[0601] The BS selects a UE to participate in model training and testing according to the UE's data characteristics.

[0602] The BS transmits the compressed initial global model parameter to the selected US.

[0603] In each iteration process of model training:

[0604] The UE decompresses the parameter, and trains and tests the model locally.

[0605] The UE calculate a verification error of the local model, obtains the channel condition of compressing the local model parameter, and compresses the parameter according to the verification error and the channel condition (an interface for transmitting the compressed parameter may be newly defined).

[0606] The UE uploads the compressed local model parameter and the verification error.

[0607] The BS fuses the parameter uploaded by the selected UE and updates the model, and determines whether the model converges. The BS groups UEs according to the RSRP thermodynamic characteristic and RSRP distribution characteristic, and performs additional fine group weight adjustment according to the correlation between groups and the difference in the number of terminal devices in each group.

[0608] The BS transmits the compressed global model parameter to the selected UE.

[0609] After multiple iterations, after fusing the parameter uploaded by the selected UE and updating the model, the BS ends the radio map construction if the model converges.

[0610] Based on at least one of the above embodiments, several examples of the radio map construction map will be given below.

[0611] In an example, the radio map construction mainly includes the following steps.

[0612] In step 3.0, the wireless access network device issues an initial neural network model parameter.

[0613] In step 3.1, the wireless access network device selects a suitable terminal device for model training. The step 3.1 may include the following steps.

[0614] In step 3.1-1, the terminal device acquires position information and the corresponding wireless channel parameter value in real time. The specific data content may refer to the above description and will not be repeated here.

[0615] In step 3.1-2, the terminal device counts the wireless channel parameter characteristic in the local data set and reports it to the wireless access network device (optionally, to ensure data security, the UE may only report the wireless channel parameter characteristic result).

[0616] In step 3.1-3, the wireless access network device maintains the wireless channel parameter pool reported by the terminal device.

[0617] In step 3.1-4, the wireless network access device determines whether the terminal device performs model training.

[0618] For example, the wireless access network device determines, based on at least one of the above characteristics (e.g., richness, specificity, scarcity, etc.) of the wireless channel parameter and the channel state between the wireless access network device and the terminal device (optional), whether the terminal device performs model training. The determination method includes, but not limited to, a reinforcement learning based method, or the like. The wireless access network device notifies the selected terminal device to perform model training.

[0619] In step 3.2, the terminal device performs data enhancement on the data in the local data set, performs model training based on the enhanced data, and evaluates the model training precision based on the test set.

[0620] In step 3.3, the terminal device compresses the model parameter and uploads the compressed model parameter to the wireless access network device. The step 3.3 may include the following steps.

[0621] In step 3.3-1, the terminal device acquires the channel environment information such as the channel state between the terminal device and the wireless access network device and the channel bandwidth, as well as the local model training precision in this round.

[0622] In step 3.3-2, the terminal device dynamically compresses the model based on the channel environment information, the local model training precision in this round, and the like.

[0623] In step 3.4, the wireless access network device fuses the parameter, then compresses the fused model parameter and issues it to the terminal device.

[0624] In step 3.5, when the model precision tends to be stable, the radio map has been constructed; otherwise, the next round of terminal device selection, model parameter issuing, training, uploading, fusion, and the like is continued.

[0625] In another example, by taking the construction of a single-wireless access network device RSRP radio map as an example, a radio map construction process mainly includes the following steps.

[0626] In step 4.1, the wireless access network device issues an initial neural network model parameter.

[0627] In step 4.2, the terminal device acquires position information and corresponding RSRP in real time. The acquired data is stored in the local data set in the chronological order, as shown in Table 7:

[0628]

[0629] In step 4.3, the terminal device counts the characteristics of the position data and corresponding RSRP in the local data set, and reports them to the wireless access network device.

[0630] For example, the terminal device may count the following data:

[0631] 1) the data volume in the local data set is greater than a threshold threshold1;

[0632] 2) maximum distances in the local data set are greater than a threshold threshold2;

[0633] 3) the variance of the position information is greater than a threshold threshold3; and

[0634] 4) the variance of the RSRP is greater than a threshold threshold4.

[0635] In step 4.4, the wireless access network device maintains the RSRP data pool reported by the terminal device in this cell.

[0636] For example, the wireless access network device may maintain the information shown in Table 8 below:

[0637]

[0638] In step 4.5, the wireless network access device determines whether the terminal device performs model training.

[0639] For example, the wireless access network device determines the richness of the data set based on the acquired or counted data reported by the terminal, counts the frequency of occurrence of each RSRP value, the variance and mean value of RSRP or the like of all terminal devices, and counts the frequency of occurrence of each RSRP value, the variance and mean value of RSRP or the like of different terminal devices, so as to determine whether the terminal device acquires the RSRP with a special position. The channel state between the wireless access network device and the terminal device is used as an input of reinforcement learning, the results of the determination of the richness and specificity are used as a reward factor 1 and a reward factor 2 of reinforcement learning, and the real-time communication efficiency is used as a reward factor 3. The terminal that needs to perform model training is output through a reinforcement learning model. The wireless access network device notifies the selected terminal device to perform model training.

[0640] In step 4.6, the selected terminal enhances the data in the local data set, and then trains the local model based on the enhanced data. The data enhancement mode may refer to the above description and will not be repeated here.

[0641] In step 4.7, the terminal device acquires the channel environment information such as the channel state between the terminal device and the wireless access network device and the channel bandwidth, and calculates the uplink data rate. For example, the calculation is performed by the above formula (12).

[0642] In step 4.8, the terminal device dynamically compresses the model in combination with the channel environment information. For example, model compression is performed by the Top-k thinning method shown by the above formulae (15) to (17).

[0643] In step 4.9, the wireless access network device fuses the model parameter, quantizes and compresses the fused model parameter, and issues it to the terminal device.

[0644] In step 4.10, when the model precision tends to be stable, the radio map has been constructed; otherwise, the next round of terminal device selection, model parameter issuing, training, uploading, and fusion is continued.

[0645] The efficient and intelligent radio map construction method provided in this example can be used to construct a single-wireless access network device RSRP radio map.

[0646] In another example, by taking the construction of a multi-wireless access network device RSRP radio map as an example, a radio map construction process mainly includes the following steps (taking two wireless access network devices as an example).

[0647] In step 5.1, the wireless access network device issues an initial neural network model parameter to the terminal device in this cell.

[0648] In step 5.2, the terminal device acquires position information and measured RSRP of a plurality of corresponding base stations in real time. The acquired data is stored in the local data set in the chronological order, as shown in Table 9:

[0649]

[0650] In step 5.3, the terminal device counts the characteristics of the position data and corresponding RSRP in the local data set, and reports them to the wireless access network devices in the serving cell.

[0651] For example, the terminal device may count the following data:

[0652] 1) the RSRP data volume related to the serving cell in the local data set is greater than a threshold threshold5;

[0653] 2) maximum distances related to the serving cell in the local data set is greater than a threshold threshold6;

[0654] 3) the variance of the position information related to the serving cell is greater than a threshold threshold7; and

[0655] 4) the variance of the RSRP value related to the serving cell is greater than a threshold threshold8.

[0656] In step 5.4, the wireless access network device maintains the RSRP data pool reported by the terminal device in this cell (which may be the same as the step 4.4).

[0657] In step 5.5, the wireless network access device determines whether the terminal device in this cell performs model training (which may be the same as the step 4.5).

[0658] In step 5.6, the selected terminal device performs model training.

[0659] In step 5.7, the terminal device acquires the channel environment information such as the channel state between the terminal device and the wireless access network device in the serving cell and the channel bandwidth, and calculates the uplink data rate.

[0660] In step 5.8, the terminal device dynamically compresses and uploads the model in combination with the channel environment information.

[0661] In step 5.9, the wireless access network device fuses, quantizes and compresses the model parameter, and issues it to the terminal device.

[0662] In step 5.10, when the model precision tends to be stable, the radio map has been constructed; otherwise, the next round of terminal device selection, model parameter issuing, training, uploading and fusion is continued.

[0663] The efficient and intelligent radio map construction method provided in this example can be used to construct a multi-wireless access network device RSRP radio map.

[0664] In still another example, also taking the construction of a multi-wireless access network device RSRP radio map as an example, another radio map construction process mainly includes the following steps.

[0665] In step 6.1, the wireless access network device issues an initial neural network model parameter.

[0666] In step 6.2, the terminal device acquires position information and corresponding RSRP in real time (which may be the same as the step 5.2).

[0667] In step 6.3, the terminal device counts the position information and corresponding RSRP characteristic of each surrounding wireless access network device in the local data set, including:

[0668] 1) Amount_of_samples: indicating the number of data samples;

[0669] 2) Fluctuations_of_x and Fluctuations_of_y: indicating the movement range of the object, where x and y are the horizontal coordinate and vertical coordinate of the position of the terminal, Fluctuations_of_x is the value obtained by subtracting the minimum x from the maximum x, and Fluctuations_of_y is the value obtained by subtracting the minimum y from the maximum y;

[0670] 3) Fluctuations_of_RSRP: indicating the fluctuation range of the channel condition, which may be the value of obtained by subtracting the minimum RSRP from the maximum RSRP; and

[0671] 4) the number of samples in each RSRP range.

[0672] The terminal reports this vector to the wireless access network device. This vector may be used to evaluate the quality of data samples. The vector may be as shown in Table 10:

[0673]

[0674] Based on this vector, some characteristics of the UE may be speculated. As an example, referring to Figure 10:

[0675] For example, for a UE with large fluctuations ofxandyand small fluctuations of RSRP, it is possible that the UE is moving dramatically, but the distance from the base station basically remains unchanged.

[0676] For another example, for a UE with small fluctuations ofxandyand large fluctuations of RSRP, it indicates that the UE moves in a small range, but the channel changes dramatically, and there may be an obstacle.

[0677] Optionally, the terminal device further constructs a distribution characteristic vector 1 (extracted from the data set) and reports it to the wireless access network. For example, the distribution characteristic vector 1 may be as show in Table 11:

[0678]

[0679] Optionally, during each model training iteration, Table 10 or Table 11 will be reported to the wireless access network device for terminal device selection, so that these terminal devices are trained by the local model.

[0680] In step 6.4, the wireless access network device maintains the RSRP data pool (or SINR data pool) reported by the terminal device in this cell.

[0681] As an example, the RSRP data pool may be as shown in Table 1, and the scarce RSRP interval (target interval) may be determined from the RSRP data pool.

[0682] Optionally, the terminal device reports the DCV1 to the wireless access network device, and the DCV1 provides a reference for RSRP or SINR percentage.

[0683] In step 6.5, the wireless network access device determines whether the terminal device performs model training.

[0684] Optionally, the wireless access network device calculates the priority of each terminal device in the way shown in Formula (1), and selects terminal devices with the highest priority.

[0685] Optionally, a terminal device with high data quality is selected for model training. The data quality of the terminal device may be evaluated by the DCV1. By taking RSRP as an example, the wireless access network device calculated the data richness priority and scarcity priority of each terminal device, and selects a terminal device with the highest priority for training.

[0686] In step 6.6, the selected terminal device enhances the data in the local data set, and then trains the local model based on the enhanced data. The data enhancement mode may refer to the above description and will not be repeated here.

[0687] In step 6.7, the terminal device acquires the local model training precision (verification error). Optionally, the calculation method may refer to the above Formula (7), Formula (14), etc.

[0688] Optionally, the terminal device further constructs a distribution characteristic vector 2 (extracted from the model parameter) and reports it to the wireless access network. For example, the distribution characteristic vector 2 may be as shown in Table 12:

[0689]

[0690] In step 6.8, the terminal device adaptively compresses the model parameter based on the local model training precision.

[0691] Optionally, the model deviation in the DCV2 is used for model compression.

[0692] Optionally, the step 6.8 may include the following steps.

[0693] In step 6.8-1, the terminal device determines a model parameter (gradient) cluster to be transmitted according to the trained model training precision. For example, the network structure has a total of seven clusters. When and , only the model parameters of first two clusters are transmitted.

[0694] In step 6.8-2, the terminal device determines a gradient to be transmitted in the model parameter gradient cluster. For example, maximum gradients are selected according to the verification error of the local model.

[0695] In step 6.8-3, the terminal device quantizes and reports selected gradients, the positions and symbols of gradients, and the verification error to the wireless access network device.

[0696] In step 6.9, the wireless access network device fuses the reported model parameter, compresses the fused model parameter, and issues it to the terminal device.

[0697] Optionally, the model parameter may be fused based on the DCV2 reported by the terminal device participating in model training.

[0698] Optionally, the step 6.9 may include the following steps.

[0699] In step 6.9-1, terminal devices are grouped based on the RSRP thermodynamic characteristic and the RSRP variance distribution to obtain density and fluctuation characteristic sets and the weight corresponding to each set. For example, the reference can be made to the contents show in the above Table 2 and Formulae (4) to (6-2).

[0700] In step 6.9-2, the group weight factor of each group and the terminal device-level weight in each group are adjusted based on the similarity between groups and the number of user equipment in each user equipment group. The reported model parameter is fused based on the adjusted weight. For example, the adjustment mode and the fusion mode may refer to the above description and will not be repeated here.

[0701] In step 6.10, when the model precision tends to be stable, the radio map has been constructed; otherwise, the next round of terminal device selection, model parameter issuing, training, uploading, and fusion is continued.

[0702] The efficient and intelligent radio map construction method provided in this example can be used to construct a multi-wireless access network device RSRP radio map.

[0703] In the efficient and intelligent radio map construction method provided in the embodiments of the disclosure, a distributed machine learning architecture based on federated learning is used, and the radio map construction includes multiple rounds of model issuing, training, uploading, and fusion. In each round, the wireless access network device selects a suitable terminal device (e.g., with a higher priority) for model training by extracting radio signal characteristics and data characteristics and deducing the characteristics (e.g., the diversity of channel variation, the scarcity of signal distribution, etc.) of the data set on the terminal device, thereby improving the model precision, accelerating the model convergence and reducing the number of communication rounds. In combination with the channel condition between the terminal side and the wireless access network device, model quantization and compression are selectively performed (for example, considering the model training precision, some parameters are transmitted; for example, during training the UE, only the model parameters with higher precision are uploaded, but not all the parameter are uploaded), thereby reducing the communication cost on the terminal side. The terminal device dynamically compresses and uploads the model parameter, thereby reducing the communication overhead. The wireless access network device uses an adaptive model parameter fusion method based on the model confidence and the data representative factor, so that the fusion precision of the model is improved and the training can converge quickly. The radio map construction method provided in the embodiments of the disclosure can satisfy the requirements of low communication overhead, high precision and low model complexity.

[0704] Through lots of experiments, the inventor of the disclosure has found that the radio map construction provided in the embodiments of the disclosure reduces 96% of the transmission overhead at a very small precision loss (<4%), and the model has low complexity and can be easily deployed in actual networks.

[0705] An embodiment of the disclosure provides an electronic device, including: a transceiver, which is configured to transmit and receive signals; and, a processor, which is coupled to the transceiver and configured to implement the steps in the above method embodiments. Optionally, the electronic device may be a wireless access network device, and the processor is configured to implement the steps in the embodiments of the method executed by a wireless access network device. The detailed functional descriptions and the achieved beneficial effects can refer to the above description of the embodiments of the method executed by a wireless access network device and will not be repeated here. Optionally, the electronic device may be a terminal device, and the processor is configured to implement the steps in the embodiments of the method executed by a terminal device. The detailed functional descriptions and the achieved beneficial effects can refer to the above description of the embodiments of the method executed by a terminal device and will not be repeated here. In practical applications, the wireless access network device or the terminal device can be interpreted as different network nodes.

[0706] An embodiment of the disclosure further provides an electronic device, including at least one transceiver and at least one processor coupled to the at least one transceiver. The at least one processor is configured to implement the method provided in any one of optional embodiments of the disclosure.

[0707]

[0708] Figure 11 shows a schematic structure diagram of an electronic device 4000 to which the solution of the embodiment of the disclosure is applied. As shown in Figure 11, the electronic device 4000 shown in Figure 11 may include a processor 4001 and a memory 4003. The processor 4001 is connected to the memory 4003, for example, through a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004. It should be noted that, in practical applications, the transceiver 4004 is not limited to one, and the structure of the electronic device 4000 does not constitute any limitations to the embodiments of the disclosure.

[0709] The processor 4001 may be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), or a field programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logical blocks, modules and circuits described in connection with the disclosure. The processor 4001 may also be a combination for realizing computing functions, for example, a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0710] The bus 4002 may include a path to transfer information between the components described above. The bus 4002 may be a peripheral component interconnect (PCI) bus, or an extended industry standard architecture (EISA) bus, etc. The bus 4002 may be an address bus, a data bus, a control bus, etc. For ease of presentation, the bus is represented by only one thick line in Figure 11. However, it does not mean that there is only one bus or one type of buses.

[0711] The memory 4003 may be, but not limited to, read only memories (ROMs) or other types of static storage devices that can store static information and instructions, random access memories (RAMs) or other types of dynamic storage devices that can store information and instructions, may be electrically erasable programmable read only memories (EEPROMs), compact disc read only memories (CD-ROMs) or other optical disk storages, optical disc storages (including compact discs, laser discs, discs, digital versatile discs, blue-ray discs, etc.), magnetic storage media or other magnetic storage devices, or any other media that can carry or store desired program codes in the form of instructions or data structures and that can be accessed by computers.

[0712] The memory 4003 is used to store computer program for executing the solutions of the disclosure, and is controlled by the processor 4001. The processor 4001 is used to execute the computer program stored in the memory 4003 to implement the solution provided in any method embodiment described above.

[0713] Embodiments of the disclosure provide a computer-readable storage medium having a computer program stored on the computer-readable storage medium, the computer program, when executed by a processor, implements the steps and corresponding contents of the foregoing method embodiments.

[0714] Embodiments of the disclosure also provide a computer program product including a computer program, the computer program when executed by a processor realizing the steps and corresponding contents of the preceding method embodiments.

[0715] The terms "first", "second", "third", "fourth", "1", "2", etc. (if present) in the specification and claims of this application and the accompanying drawings above are used to distinguish similar objects and need not be used to describe a particular order or sequence. It should be understood that the data so used is interchangeable where appropriate so that embodiments of the disclosure described herein can be implemented in an order other than that illustrated or described in the text.

[0716] It should be understood that while the flow diagrams of embodiments of the disclosure indicate the individual operational steps by arrows, the order in which these steps are performed is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of embodiments of the disclosure, the implementation steps in the respective flowcharts may be performed in other orders as desired. In addition, some, or all of the steps in each flowchart may include multiple sub-steps or multiple phases based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same moment, and each of these sub-steps or stages can also be executed at different moments separately. The order of execution of these sub-steps or stages can be flexibly configured according to requirements in different scenarios of execution time, and the embodiments of the disclosure are not limited thereto.

[0717] The above-mentioned description and the drawings are provided merely as examples to help readers to understand the disclosure, and they should not be interpreted or aim to limit the scope of the disclosure in any way. Although some embodiments are provided, it is apparent for those skilled in the art to adopt other similar implementation means based on the technical idea of the disclosure without departing from the technical concept of the solution of the disclosure.

Claims

1.A method performed by a base station in a wireless communication system, the method comprising:receiving, from a plurality of terminals, information on a data set including information on a location of a terminal and information on a channel quality corresponding to the location;determining a target terminal for a training among the plurality of terminals based on the information on the data set, wherein the training is associated with a prediction of a radio map;transmitting, to the target terminal, first information including a parameter related to a network model which is used for the prediction;receiving, from the target terminal, second information including an updated parameter corresponding to a result of the training; andupdating the network model based on the second information.2.The method of claim 1, wherein the determining the target terminal further includes:generating characteristic information for the terminal based on the information on the data set; anddetermining whether to select the terminal as the target terminal based on the characteristic information,wherein the characteristic information includes richness information associated with a data volume and data fluctuation range of the data set, scarcity information associated with to the data volume in a specific interval of a channel quality parameter, and specificity information associated with whether data that the base station prefers is included in the data set.3.The method of claim 1,wherein the second information is compressed based on an uplink data transmission rate associated with a channel quality of the target terminal and a verification error associated with the result of the training.4.The method of claim 1, wherein the updating the network model further includes:identifying a distribution characteristic of the updated parameter, wherein the distribution characteristic includes density and fluctuation characteristic associated with a channel quality of the target terminal,generating a terminal group including the target terminal and other terminals which are determined as the target terminal among the plurality of the terminals,configuring a weight for the terminal group based on the distribution characteristic;adjusting the weight based on a verification error associated with the result of the training, wherein information on the verification error is received from the target terminal; andfusing the updated parameter based on the adjusted weight,wherein the network model is updated based on the fused updated parameter.5.A method performed by a terminal in a wireless communication system, the method comprising:transmitting, to a base station, information on a data set including information on a location of the terminal and information on a channel quality corresponding to the location;receiving, from the base station, first information including a parameter related to a network model associated with a prediction of a radio map, in case that the terminal is a target terminal which performs a training associated with the predication;performing the training based on the first information; andtransmitting, to the base station, second information including an updated parameter corresponds to a result of the training,wherein the second information is used for updating the network model.6.The method of claim 5, further comprising:obtaining information on initial model of the network model;performing a data enhancement on the data set; andgenerating the updated parameter, based on the training and the enhanced data set by the data enhancement,wherein the data enhancement includes an interpolation between two samples included in the data set or an oversampling, andwherein whether the terminal performs the interpolation or the oversampling is determined based on a reference signals received power (RSRP) difference between the two samples and a distance between the two samples.7.The method of claim 5,wherein the second information is compressed based on an uplink data transmission rate associated with a channel quality of the target terminal and a verification error associated with the result of the training.8.A base station in a wireless communication system, the base station comprising:a transceiver;memory storing one or more programs; andone or more processors communicatively coupled to the transceiver and the memory,wherein the one or more programs include computer-executable instructions that, when executed by the one or more processors individually or collectively, cause the base station to:receive, from a plurality of terminals, information on a data set including information on a location of a terminal and information on a channel quality corresponding to the location,determine a target terminal for a training among the plurality of terminals based on the information on the data set, wherein the training is associated with a prediction of a radio map,transmit, to the target terminal, first information including a parameter related to a network model which is used for the prediction,receive, from the target terminal, second information including an updated parameter corresponding to a result of the training, andupdate the network model based on the second information.9.The base station of claim 8, wherein a determining the target terminal for the training further includes:generating characteristic information for the terminal based on the information on the data set; anddetermining whether to select the terminal as the target terminal based on the characteristic information,wherein the characteristic information includes richness information associated with a data volume and data fluctuation range of the data set, scarcity information associated with to the data volume in a specific interval of a channel quality parameter, and specificity information associated with whether data that the base station prefers is included in the data set.10.The base station of claim 8,wherein the second information is compressed based on an uplink data transmission rate associated with a channel quality of the target terminal and a verification error associated with the result of the training.11.The base station of claim 8, wherein an updating the network model based on the second information further includes:identifying a distribution characteristic of the updated parameter, wherein the distribution characteristic includes density and fluctuation characteristic associated with a channel quality of the target terminal,generating a terminal group including the target terminal and other terminals which are determined as the target terminal among the plurality of the terminals,configuring a weight for the terminal group based on the distribution characteristic;adjusting the weight based on a verification error associated with the result of the training, wherein information on the verification error is received from the target terminal; andfusing the updated parameter based on the adjusted weight,wherein the network model is updated based on the fused updated parameter.12.A terminal in a wireless communication system, the terminal comprising:a transceiver;memory storing one or more programs; andone or more processors communicatively coupled to the transceiver and the memory,wherein the one or more programs wherein the one or more programs include computer-executable instructions that, when executed by the one or more processors individually or collectively, cause the terminal to:transmit, to a base station, information on a data set including information on a location of the terminal and information on a channel quality corresponding to the location,receive, from the base station, first information including a parameter related to a network model associated with a prediction of a radio map, in case that the terminal is a target terminal which performs a training associated with the predication,perform the training based on the first information, andtransmit, to the base station, second information including an updated parameter corresponds to a result of the training,wherein the second information is used for updating the network model.13.The terminal of claim 12, wherein the computer-executable instructions that, when executed by the one or more processors individually or collectively, further cause the terminal to:perform a data enhancement on the data set, andgenerate the updated parameter, based on the training and the enhanced data set by the data enhancement,wherein the data enhancement includes an interpolation between two samples included in the data set or an oversampling, andwherein whether the terminal performs the interpolation or the oversampling is determined based on a reference signals received power (RSRP) difference between the two samples and a distance between the two samples.14.The terminal of claim 12,wherein the second information is compressed based on an uplink data transmission rate associated with a channel quality of the target terminal and a verification error associated with the result of the training.

Images