Method performed by network node, method performed by node, network node, and node

Abstract

The disclosure relates to a 5G or 6G communication system for supporting a higher data transmission rate. The present disclosure provides a method performed by a network node in a wireless communication system, a method performed by a node in a wireless communication system, the network node and the node. A method performed by a first network node in a wireless communication system comprises: transmitting, to a first node, a first configuration message including at least one of configuration information of a scheduling window and energy saving configuration information; generating, based on a scheduling module trained by a generative network, scheduling information of at least one first node in at least one time unit; and scheduling the first node according to the scheduling information.

Description

METHOD PERFORMED BY NETWORK NODE, METHOD PERFORMED BY NODE, NETWORK NODE, AND NODE

[0001] The present disclosure generally relates to the field of wireless communication, and in particular to a method performed by a network node, a method performed by a node, the network node and the node.

[0002] 5G mobile communication technologies define broad frequency bands such that high transmission rates and new services are possible, and can be implemented not only in "Sub 6GHz" bands such as 3.5GHz, but also in "Above 6GHz" bands referred to as mmWave including 28GHz and 39GHz. In addition, it has been considered to implement 6G mobile communication technologies (referred to as Beyond 5G systems) in terahertz (THz) bands (for example, 95GHz to 3THz bands) in order to accomplish transmission rates fifty times faster than 5G mobile communication technologies and ultra-low latencies one-tenth of 5G mobile communication technologies.

[0003] At the beginning of the development of 5G mobile communication technologies, in order to support services and to satisfy performance requirements in connection with enhanced Mobile BroadBand (eMBB), Ultra Reliable Low Latency Communications (URLLC), and massive Machine-Type Communications (mMTC), there has been ongoing standardization regarding beamforming and massive MIMO for mitigating radio-wave path loss and increasing radio-wave transmission distances in mmWave, supporting numerologies (for example, operating multiple subcarrier spacings) for efficiently utilizing mmWave resources and dynamic operation of slot formats, initial access technologies for supporting multi-beam transmission and broadbands, definition and operation of BWP (BandWidth Part), new channel coding methods such as a LDPC (Low Density Parity Check) code for large amount of data transmission and a polar code for highly reliable transmission of control information, L2 pre-processing, and network slicing for providing a dedicated network specialized to a specific service.

[0004] Currently, there are ongoing discussions regarding improvement and performance enhancement of initial 5G mobile communication technologies in view of services to be supported by 5G mobile communication technologies, and there has been physical layer standardization regarding technologies such as V2X (Vehicle-to-everything) for aiding driving determination by autonomous vehicles based on information regarding positions and states of vehicles transmitted by the vehicles and for enhancing user convenience, NR-U (New Radio Unlicensed) aimed at system operations conforming to various regulation-related requirements in unlicensed bands, NR UE Power Saving, Non-Terrestrial Network (NTN) which is UE-satellite direct communication for providing coverage in an area in which communication with terrestrial networks is unavailable, and positioning.

[0005] Moreover, there has been ongoing standardization in air interface architecture / protocol regarding technologies such as Industrial Internet of Things (IIoT) for supporting new services through interworking and convergence with other industries, IAB (Integrated Access and Backhaul) for providing a node for network service area expansion by supporting a wireless backhaul link and an access link in an integrated manner, mobility enhancement including conditional handover and DAPS (Dual Active Protocol Stack) handover, and two-step random access for simplifying random access procedures (2-step RACH for NR). There also has been ongoing standardization in system architecture / service regarding a 5G baseline architecture (for example, service based architecture or service based interface) for combining Network Functions Virtualization (NFV) and Software-Defined Networking (SDN) technologies, and Mobile Edge Computing (MEC) for receiving services based on UE positions.

[0006] As 5G mobile communication systems are commercialized, connected devices that have been exponentially increasing will be connected to communication networks, and it is accordingly expected that enhanced functions and performances of 5G mobile communication systems and integrated operations of connected devices will be necessary. To this end, new research is scheduled in connection with eXtended Reality (XR) for efficiently supporting AR (Augmented Reality), VR (Virtual Reality), MR (Mixed Reality) and the like, 5G performance improvement and complexity reduction by utilizing Artificial Intelligence (AI) and Machine Learning (ML), AI service support, metaverse service support, and drone communication.

[0007] Furthermore, such development of 5G mobile communication systems will serve as a basis for developing not only new waveforms for providing coverage in terahertz bands of 6G mobile communication technologies, multi-antenna transmission technologies such as Full Dimensional MIMO (FD-MIMO), array antennas and large-scale antennas, metamaterial-based lenses and antennas for improving coverage of terahertz band signals, high-dimensional space multiplexing technology using OAM (Orbital Angular Momentum), and RIS (Reconfigurable Intelligent Surface), but also full-duplex technology for increasing frequency efficiency of 6G mobile communication technologies and improving system networks, AI-based communication technology for implementing system optimization by utilizing satellites and AI (Artificial Intelligence) from the design stage and internalizing end-to-end AI support functions, and next-generation distributed computing technology for implementing services at levels of complexity exceeding the limit of UE operation capability by utilizing ultra-high-performance communication and computing resources.

[0008] Embodiments of the present disclosure provide a method performed by a network node, a method performed by a node, the network node and the node. The embodiments of the present disclosure provide the following technical schemes.

[0009] An embodiment of the present disclosure provides a method performed by a first network node in a wireless communication system, comprising: transmitting, to a first node, a first configuration message including at least one of configuration information of a scheduling window and energy saving configuration information; generating, based on a scheduling module trained by a generative network, scheduling information of at least one first node in at least one time unit; and transmitting, to the first node, a first indication message for instructing the first node to perform downlink signal monitoring, wherein the scheduling information is obtained according to a matrix or an image generated by the scheduling module.

[0010] According to the present disclosure, scheduling efficiency in wireless communication systems is provided.

[0011] FIG. 1 is an exemplary system architecture of system architecture evolution (SAE).

[0012] FIG. 2 is a structural diagram of a conditional generative adversarial network.

[0013] FIG. 3 is a block diagram of an exemplary scheduling algorithm according to various embodiments of the present disclosure.

[0014] FIG. 4 is a block diagram of a generator according to an example embodiment of the present disclosure.

[0015] FIG. 5a is a block diagram of training and inferring according to an example embodiment of the present disclosure.

[0016] FIG. 5b is a block diagram of a self-optimizer according to an example embodiment of the present disclosure.

[0017] FIG. 6 is a block diagram of a device according to an example embodiment of the present disclosure.

[0018] Embodiments of the present disclosure provide a method performed by a network node, a method performed by a node, the network node and the node. The embodiments of the present disclosure provide the following technical schemes.

[0019] An embodiment of the present disclosure provides a method performed by a first network node in a wireless communication system, comprising: transmitting, to a first node, a first configuration message including at least one of configuration information of a scheduling window and energy saving configuration information; generating, based on a scheduling module trained by a generative network, scheduling information of at least one first node in at least one time unit; and transmitting, to the first node, a first indication message for instructing the first node to perform downlink signal monitoring, wherein the scheduling information is obtained according to a matrix or an image generated by the scheduling module.

[0020] In accordance with the embodiment of the present disclosure, the scheduling module trained by the generative network comprises a classifier, at least one generator and a scheduler, the classifier is configured to generate different types of indication information about input data, the generator generates the scheduling information according to the different types of indication information generated by the classifier, and the scheduler schedules data transmission of the first node according to the scheduling information.

[0021] In accordance with the embodiment of the present disclosure, for one slot of one of the first nodes, the scheduling information further comprises at least one of bit number information, resource allocation information, and indication information of the type of transmission traffic.

[0022] In accordance with the embodiment of the present disclosure, the first configuration message including at least one of configuration information of a scheduling window and energy saving configuration information, and wherein the configuration information of the scheduling window comprises at least one of first starting offset information, first length information, first period information, and first valid time information.

[0023] In accordance with the embodiment of the present disclosure, the first indication message comprises at least one of sleep indication information and sleep time information.

[0024] In accordance with the embodiment of the present disclosure, the input data of the classifier comprises at least one of information related to the amount of buffered data, information related to the amount of arrived data, information related to the amount of transmitted data, information related to channel quality, information related to physical resources (e.g., resource blocks), information related to a user's location, and fingerprint information of a wireless environment, and output data of the classifier is the different types of indication information, and comprises at least one of a numerical value of a class, a vector of the class, a sequence of the class, and a token of the class.

[0025] In accordance with the embodiment of the present disclosure, the classifier can further comprise a predictor configured to predict information related to a next scheduling window, and an output of the predictor is used by the classifier to generate the output data.

[0026] In accordance with the embodiment of the present disclosure, the at least one generator is selected according to the output data of the classifier to generate the scheduling information.

[0027] In accordance with the embodiment of the present disclosure, the classifier can further generate first classification information and second classification information, and the first classification information is used by a first generator in the at least one generator to generate the scheduling information, and the second classification information is used by a second generator in the at least one generator to generate the scheduling information.

[0028] In accordance with the embodiment of the present disclosure, the at least one generator is trained according to a conditional generative adversarial network.

[0029] An embodiment of the present disclosure further provides a method performed by a first node in a wireless communication system, comprising: receiving, from a first network node, a first configuration message including at least one of configuration information of a scheduling window and energy saving configuration information, the scheduling being performed by generating, based on a scheduling module trained by a generative network, scheduling information of at least one first node in at least one time unit; and receiving, from the first network node, a first indication message for instructing the first node to perform downlink signal monitoring, wherein the scheduling information is obtained according to a matrix or an image generated by the scheduling module.

[0030] In accordance with the embodiment of the present disclosure, the scheduling module trained by the generative network comprises a classifier, at least one generator and a scheduler, the classifier is configured to generate different types of indication information about input data, the generator generates the scheduling information according to different types of indication information generated by the classifier, and the scheduler schedules data transmission of the first node according to the scheduling information.

[0031] In accordance with the embodiment of the present disclosure, for one slot of one of the first nodes, the scheduling information further comprises at least one of bit number information, resource allocation information, and indication information of the type of transmission traffic.

[0032] In accordance with the embodiment of the present disclosure, the first configuration message including at least one of configuration information of a scheduling window and energy saving configuration information, and wherein the configuration information of the scheduling window comprises at least one of first starting offset information, first length information, first period information, and first valid time information.

[0033] In accordance with the embodiment of the present disclosure, the first indication message comprises at least one of sleep indication information and sleep time information.

[0034] An embodiment of the present disclosure further provides a first network node in a wireless communication system, comprising: a transceiver configured to transmit and receive signals; and a controller coupled to the transceiver and configured to perform the method performed by the first network node in the wireless communication system.

[0035] An embodiment of the present disclosure further provides a first node in a wireless communication system, comprising: a transceiver configured to transmit and receive signals; and a controller coupled to the transceiver and configured to perform the method performed by the first node in the wireless communication system.

[0036] The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of various embodiments of the present 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. Accordingly, 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 present disclosure. In addition, descriptions of well-known functions and constructions may be omitted for clarity and conciseness.

[0037] 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 present disclosure. Accordingly, it should be apparent to those skilled in the art that the following description of various embodiments of the present disclosure is provided for illustration purpose only and not for the purpose of limiting the present disclosure as defined by the appended claims and their equivalents.

[0038] 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.

[0039] The term “include” or “may include” refers to the existence of a corresponding disclosed function, operation or component which can be used in various embodiments of the present disclosure and does not limit one or more additional functions, operations, or components. The terms such as “include” and / or “have” may be construed to denote a certain characteristic, number, step, operation, constituent element, component or a combination thereof, but may not be construed to exclude the existence of or a possibility of addition of one or more other characteristics, numbers, steps, operations, constituent elements, components or combinations thereof.

[0040] The term “or” used in various embodiments of the present disclosure includes any or all of combinations of listed words. For example, the expression “A or B” may include A, may include B, or may include both A and B.

[0041] Unless defined differently, all terms used herein, which include technical terminologies or scientific terminologies, have the same meaning as that understood by a person skilled in the art to which the present disclosure belongs. Such terms as those defined in a generally used dictionary are to be interpreted to have the meanings equal to the contextual meanings in the relevant field of art, and are not to be interpreted to have ideal or excessively formal meanings unless clearly defined in the present disclosure.

[0042] The accompanying drawings discussed below and various embodiments for describing the principles of the present disclosure in this patent document are only for illustration and should not be interpreted as limiting the scope of the disclosure in any way. Those skilled in the art will understand that the principles of the present disclosure can be implemented in any suitably arranged system or device.

[0043] In order to meet an increasing demand for wireless data communication services since a deployment of 4G communication system, efforts have been made to develop an improved 5G or pre-5G communication system. Therefore, the 5G or pre-5G communication system is also called “beyond 4G network” or “post LTE system”.

[0044] Wireless communication is one of the most successful innovations in modern history. Recently, a number of subscribers of wireless communication services has exceeded 5 billion, and it continues growing rapidly. With the increasing popularity of smart phones and other mobile data devices (such as tablet computers, notebook computers, netbooks, e-book readers and machine-type devices) in consumers and enterprises, a demand for wireless data traffics is growing rapidly. In order to meet rapid growth of mobile data traffics and support new applications and deployments, it is very important to improve efficiency and coverage of wireless interfaces.

[0045] In a mobile network, in order to satisfy the traffic requirements of different users, schedulers play a crucial role. Since conventional schedulers can only determine the scheduling of users in upcoming slots, a scheduling algorithm will be needed on the base station side in each slot, thereby affecting the performance of user equipment.

[0046] Fig. 1 is an exemplary system architecture 200 according to various embodiments of the present disclosure. Other embodiments of the system architecture 200 can be used without departing from the scope of the present disclosure.

[0047] User equipment (UE) 201 is a terminal device for receiving data. A next generation radio access network (NG-RAN) 202 is a radio access network, which includes a base station (a gNB or an eNB connected to 5G core network 5GC, and the eNB connected to the 5GC is also called ng-gNB) that provides UE with interfaces to access the radio network. An access control and mobility management function entity (AMF) 203 is responsible for managing mobility context and security information of the UE. A user plane function entity (UPF) 204 mainly provides functions of user plane. A session management function entity SMF 205 is responsible for session management. A data network (DN) 206 includes, for example, services of operators, access of Internet and service of third parties.

[0048] The exemplary embodiments of the present disclosure will be further described below in conjunction with the drawings.

[0049] The text and the accompanying drawings are merely provided as examples to help in understanding the present disclosure. They should not be construed as limiting the scope of the present disclosure in any way. Although some embodiments and examples have been provided, based on the contents disclosed herein, it is apparent to those skilled in the art that changes can be made to the illustrated embodiments and examples, without departing from the scope of the present disclosure.

[0050] Before the introduction of specific contents, some assumptions and definitions of the present disclosure will be given below.

[0051] ■ The message names in the present disclosure are only examples, and other message names are also possible.

[0052] ■ "First", "second," and the like contained in the message names in the present disclosure are only used for distinguishing one message from another message, and do not indicate an execution or transmission order.

[0053] ■ The detailed description of steps irrelevant to the present disclosure will be omitted in the present disclosure.

[0054] ■ In the present disclosure, the steps in each process can be executed in combination or separately. The execution steps of each process are only examples, and other possible execution steps and / or orders are not excluded.

[0055] ■ In the present disclosure, a base station may be a 6G base station or a 5G base station (e.g., gNB, ng-eNB, etc.), or may be a 4G base station (e.g., eNB) or an RAN node, or may also be other types of access nodes.

[0056] ■ In the present disclosure, a user, a user equipment, a user terminal and a user terminal device are equivalent.

[0057] The nodes involved in the present disclosure include:

[0058] a first node: a user equipment, where the user equipment may be a mobile phone or a relay node; and

[0059] a second node: a base station, or a central unit (CU) of the base station, or a control panel part of the central unit of the base station, or a distributed unit of the base station.

[0060] The base station involved in the second node may be one of the following types (not excluding other types that can be used for user terminal access):

[0061] ■ a long term evolution (LTE) base station;

[0062] ■ a 5G base station;

[0063] ■ a 6G base station;

[0064] ■ an RAN node;

[0065] ■ a non-terrestrial network (NTN) base station;

[0066] ■ a high altitude platform station (HAPS) base station;

[0067] ■ a drone base station; and

[0068] ■ a WIFI access point.

[0069] The design of a scheduler involves the acquisition of various information, such as load information (e.g., the amount of data in a buffer, the amount of arrived traffic, etc.) and channel information (e.g., channel state, channel change, etc.). Some of these information can be directly acquired by the base station, while some thereof is reported by the user equipment. However, some information, such as the amount of future arrived traffic and a future channel state, are unknown to the base station. In order to overcome the uncertainty caused by unknown data, a conventional scheduling algorithm generally only determines the scheduling in a short time in the future, for example, the scheduling in a next slot. Such scheduling algorithm has a problem that the base station needs to execute the scheduling algorithm in each slot and the user needs to receive scheduling information (e.g., in 4G and 5G systems, the scheduling information contained in a DCI carried by a PDCCH) transmitted by the base station in each slot. As a result, it is disadvantageous for energy saving of the base station and the user equipment, and more signaling overhead is caused. This is a problem to be urgently solved.

[0070] The present invention proposes a new scheduling mechanism. By means of AI, the scheduling mechanism can generate the scheduling information of a plurality of users (the traffics of different users may be different or the same) in a future period of time (e.g., in the future period of time, the amount of data (e.g., the number of bits) to be transmitted by each user in each slot). Then, the base station allocates radio resources for each user according to the scheduling information, thereby satisfying the transmission of the number of bits indicated by the scheduling information as much as possible. A method designed by the present invention converts a scheduling problem into an image generation problem by means of the conditional generative adversarial network (C-GAN) in the field of AI. FIG. 2 shows a network structural diagram of the C-GAN. The C-GAN includes a generator and a discriminator. In training the C-GAN, training samples are real data (true data). If each piece of true data has a corresponding label or condition (y), the true data may be expressed as a conditional probability pg(x|y). If inputs of the generator are noises z and y, new data may be generated, that is, data pg(x|y) is generated. If input data of the discriminator will include the true data or the generated data and label, then the discriminator may determine, based on the label y, whether the input data is the true data or the generated data, so as to generate a predicted label for indicating a probability that the input data is the true data, i.e.,D(x|y). After the network is trained, it is expected to obtain a generator so that the discriminator cannot distinguish the generated data from the true data, and it is also expected to obtain a discriminator so that the discriminator can determine whether the input data is the true data or the generated data. The generator obtained by this method can generate corresponding images according to the condition. FIG. 3 shows a block diagram of a scheduling algorithm (the scheduling algorithm may perform downlink scheduling, uplink scheduling, or simultaneous uplink and downlink scheduling) design in the present invention. The scheduling algorithm includes three main parts:

[0071] ■ Classifier: this part mainly classifies the input data to generate different types of indication information (the indication information may be information of a scalar, or may be a vector, or may be a matrix, or may also be a token (the smallest unit of model processing). The data that can be obtained by the base station is fully classified by the classifier. The classifier may be in various forms as long as it can classify the input data (it may be a calculation formula, a machine learning network, or the like, e.g., LSTM or other networks).

[0072] ■ Generator: this part mainly generates scheduling data satisfying the condition according to input type information (e.g., type indication information), and outputs it to the scheduler for use. The generator may generate a variety of scheduling matrix data or image data. In one example, the generator may be obtained according to the C-GAN. In another example, the generator may be obtained according to other GANs. In still another example, the generator may be based on a diffusion model generative network. In yet another example, the generator may be a network such as a large model, or other types of AI models. The scheduling information generated by the generator is scheduling information of each user (or each type of traffic, or each type of traffic of each user) in a plurality of future slots. One example of the scheduling information is the number of bits transmitted in each slot, and another example thereof is resources (e.g., time resources, frequency resources, beam information, etc.) allocated in each slot, and still another example thereof is traffic information transmitted in each slot, or the like. In the present invention, the scheduling data generated by the generator may be used for scheduling the transmission of downlink data (data transmitted to the user by the base station) or the transmission of uplink data (data transmitted to the base station by the user).

[0073] ■ Scheduler: this part converts the scheduling data of the generator into the scheduling information of each user in each slot, for example, the number of bits transmitted in each slot, the resources (e.g., time resources, frequency resources, beam information, etc.) allocated in each slot, the traffic information transmitted in each slot, or the like, and allocates the resources for each user according to a channel condition of the user, to ensure that the data transmission indicated by the "scheduling information of each user in each slot" determined by the generator can be completed in each slot as far as possible. In one example, when the generated scheduling information is the number of bits transmitted in each slot, the scheduler will allocate the resources for each user as far as possible to ensure that "the number of bits transmitted in each slot" can be transmitted to the user (in downlink), or schedule sufficient resources so that the user can transmit "the number of bits transmitted in each slot" to the base station (in uplink).

[0074] The design objective of the present invention is to generate a matrix containing the scheduling information. If each row of the matrix represents scheduling related information of one user or traffic and each column thereof represents the scheduling related information of one user or traffic in one slot, each element of the matrix represents the scheduling information of one user or traffic in one slot. The scheduling information may include at least one of:

[0075] ■ bit number information: the number of bits is the number of bits generated by the scheduling algorithm, and the scheduler will satisfy the transmission of the number of bits indicated by the information as far as possible; and, if the transmission of the number of bits indicated by the information cannot be satisfied, the scheduler will satisfy it according to its maximum capacity;

[0076] ■ resource allocation information: the information may be indication information of time resources (e.g., the number of symbols, sub-frame number, frame number, etc.), indication information of frequency resources (e.g., a frequency point, a frequency band range, a subcarrier number, the number of subcarriers, etc.), indication information of time-frequency resources (e.g., a PRB number, a RB number, etc.), indication information of beams (e.g., identifier information of beams, transmission configuration information, etc.), starting position information of resources, ending position information of resources, range information of resources, or the like; and

[0077] ■ indication information of the type of transmission traffic, e.g., a DRB ID, an SRB ID, a QoS flow ID, a logic channel ID, a PDU session ID, or the like.

[0078] The following description will be given by taking bit number information as an example, and the following method is also applied to other types of scheduling information. In addition, the following method will be described by taking downlink scheduling as an example, and the following method can also be used for uplink scheduling.

[0079] In the present invention, in order to set forth the design of a specific algorithm, the concept of time window is defined. The time window includes a certain number of time units (e.g., slots). In one example, the time window may only include one slot. In another example, the time window may include a fixed number of slots. In another example, the time window may include a variable number of slots, that is, different time windows may have different numbers of slots. During AI model training, historical data may be divided into data in different time windows. During AI model inference, outputs may be defined as information in one time window, or information in a plurality of time windows.

[0080] In order to explain the scheduling algorithm designed by the present invention in detail, the description can be given from the following several aspects:

[0081] First aspect: theoretical modeling, i.e., how to convert a scheduling problem into an image generation problem

[0082] It is considered that a base station (BS) schedules N users running different traffics in a cell, and the smallest time unit of scheduling is a slot. Generally, the scheduling scheme makes a decision (i.e., a scheduling decision) on a window ω with L slots. It may be represented by a matrix, e.g., T(ω)=[ti,j(ω)]N×L, where ti,j(ω) is the number of bits transmitted to a user i in a slot j. The following aspects are taken into consideration during making the decision (in the present invention, unless otherwise explicitly stated, the following variables are variables corresponding to the scheduling window ω, and ω in each variable symbol is omitted for the sake of simplicity):

[0083] ■ Initial traffic load: it reflects the number of bits preserved in the buffer during making the scheduling decision, and it may be represented by a vector B=[bi]N×1, where biis the number of bits in a buffer of the user i (for downlink, the buffer is a buffer of the user i at the base station, while for uplink, the buffer is a buffer of the user i).

[0084] ■ Arrived traffic load: it indicates the newly arrived load after making the scheduling decision (e.g., load arrived in subsequent L slots). It is closely related to a traffic arrival model of traffics running in the user equipment, and it may be represented by a matrix A=[ai,j]N×L, where ai,jis the number of arrived bits of the user i in the slot j.

[0085] ■ Channel state: it reflects the transmission capability of each user in each slot. It may be represented by a matrix C=[ci,j]N×L, where ci,jis the channel state (e.g., channel quality indicator (CQI)) of the user i in the slot j.

[0086] ■ QoS requirement: for a service running in the user equipment, quality-of-service performance achieved by the scheduling scheme may be described by K different parameters, such as a throughput, a delay and a packet loss rate, to satisfy the user experience. Each QoS parameter of the user i has a certain limit Qi,k(j = 1,...,K). For example, Qi,kdefines a lower bound of the throughput, or Qi,krepresents an upper bound of the delay or the packet loss rate. Therefore, the quality-of-service requirements of N users may be represented by a matrix Qreq=[Qi,j]N×K.

[0087] In the present invention, the theoretical modeling is performed from two different perspectives:

[0088] ■ Perspective 1: modeling for one time window

[0089] A scheduling scheme that ensures Qreqmay be generated based on the above aspects by a function of formula G(*), for example:

[0090]

[0091] In practice, the transmission of a transmission network and the generation of application layer data will cause A to be random. Similarly, due to a change of the channel state caused by a radio environment and user movement, C is also random. B is the only determined parameter during making the scheduling decision. Therefore, T may be regarded as a random variable with a probability of p(T|B). In addition, the QoS performance achieved by the scheduling scheme may be represented as a function of T. For example, Q=F(T), where Q=[qi,k(T)]N×K) and qi,k(T) is a value of the kth QoS parameter of the user i. Thus, the mean of each QoS parameter of each user may be calculated by the following formula:

[0092]

[0093] where . For the given formula (2), the mean of the QoS parameters which can be separately realized by the traditional scheduling scheme and the newly designed scheduling scheme may be represented as and , respectively (in the present invention, the variables with (tilde) and (hat) represent variables of the traditional scheme and the newly designed scheme, respectively). In order to describe the difference between and , the Euclidean distance may be defined as:

[0094]

[0095] where and may be obtained for the traditional scheduling scheme and the newly designed scheduling scheme by the formula (2), respectively. If it is assumed that the traditional scheduling scheme can ensure the QoS requirement (defined as Qreq) of each user, the new scheduling scheme also needs to achieve a similar QoS performance. Thus, the optimization problem of the new scheduling scheme is given below. For example, G(*) is formulated under the constraint of the quality-of-service requirement of each user:

[0096]

[0097] It is difficult to solve the problem (4) due to the following difficulties: 1) during the calculation of (3), it is difficult to deduce closed forms of F(T) and p(T|B) in the formula (2); and 2) it is difficult to determine a mathematical relationship between the scheduling scheme and the generated T. Therefore, the problem (4) needs to be converted.

[0098] In the formula (2), the average value of each QoS parameter is determined by p(T|B). When the traffic load in the buffer is higher, B may be regarded as a main factor for determining T, because the scheduling algorithm may be unable to empty the buffer. However, when the load in the buffer is lower, B is insufficient to determine T, because the scheduler may transmit data that arrives during the scheduling window by allocating resources. Therefore, by introducing A, p(T|B) may be further calculated as:

[0099]

[0100] where p(A|B) = p(A) is due to the independence of B and A, and p(T|B,A) is related to the channel state C during one scheduling window. PD=p( | , ) and PG=p( | , ) are given for the traditional scheme and the new scheme, respectively, and the formula (3) may be further deformed as follows according to the formula (5):

[0101]

[0102] Since the probability of A is only related to the traffic arrival model and not related to the used scheduling algorithm, p( )=p( )=p(A). Thus, the optimization of the target formula (4) is determined by calculating qi,k( )PD-qi,k( )PGin the formula (6). Ideally, if the newly designed scheme can obtain PGsame as PD, the formula (6) will be minimized as 0. In other words, the minimization of D[ , ] may be converted into the design of a new scheduling scheme, so that PGand PDcorresponding to this scheme are approximate. In theory, the similarity between two probability distributions may be calculated by a Jensen-Shannon divergence (JSD), that is:

[0103]

[0104] where DKL(p|q) is a formula of calculating a Kullback-Leibler (KL) divergence of the two probability distributions (e.g., p(x) and p(y)):

[0105]

[0106] Thus, the problem (4) is converted into the design of G to minimize the JSD in the formula (7), and may be expressed as:

[0107]

[0108] An interesting finding is that the problem (9) has an optimization goal similar to a popular artificial intelligence (AI) model, i.e., a conditional generative adversarial network (C-GAN), which is evolved from the GAN. The C-GAN includes two models, i.e., a generator and a discriminator. Training samples are real data, and each sample has a special label. That is, the generator is responsible for generating data according to inputs of a noise variable and a label and then directly providing the generated data to the discriminator. The discriminator distinguishes, according to the input label, whether the input data is real data or the generated data. The goal of the C-GAN is to obtain a generator capable of generating data so that the discriminator cannot distinguish the generated data from the real data, and the C-GAN can also obtain a discriminator which can distinguish the real data from the generated data. In other words, the discriminator and the generator play a minimization / maximization game involving two players, the optimization goal of which is:

[0109]

[0110] where V(G,D) may be defined as:

[0111]

[0112] where pd(x|y) and pg(z|y) represent probabilities of the real data and the generated data under the condition y, respectively. D(*)∈[0,1] is an output of the discriminator, and represents the probability that the input data is from the real data rather than the generated data. For the problem (10):

[0113]

[0114] Thus, the problem (10) may be converted into obtaining a generator, so that it can optimize the following target:

[0115]

[0116] If pd(x│y) and pg(z|y) in the formula (13) are substituted with PD=p( | , ) and PG=p( | , ), the problem (13) is equivalent to the problem (9), that is:

[0117]

[0118] Thus, the scheduler design in the problem (4) is converted into a problem which can be solved by the C-GAN.

[0119] ■ Perspective 2: modeling for a plurality of time windows

[0120] In another implementation, the QoS performance of the user is counted for a period of time, and this period of time may include a plurality of windows, e.g., W windows. The scheduling matrix of the scheduling of the user by the base station may be represented as: . The QoS performance that can be achieved may be represented as , where and is a value of the kthQoS parameter of the user i. The obtained scheduling decision may be represented as , where B1represents an initial buffer size at the beginning of W windows, and and represent a traffic arrival matrix and a channel state matrix in the W windows, respectively. Due to the generation and transmission of the application layer data, is a random variable. And, due to a change of the channel state, is also random. For each window, Aωand Cωare unknown. Thus, B1is the only known parameter at the beginning of scheduling, and is a random variable with a probability of . Further, the mean of each QoS parameter of each user may be calculated by the following formula:

[0121]

[0122] where . Similar to the formula (3), in order to describe the difference between and , an Euclidean distance may be defined as:

[0123]

[0124] Thus, the problem to be solved by the scheduling algorithm is:

[0125]

[0126] In order to solve the problem (17), a problem conversion is needed. The above may be calculated by the following method:

[0127]

[0128] If an initial buffer load of each scheduling window is Bω(ω=1,…,W), the following may be obtained:

[0129]

[0130] where I is a constant vector [1,1,…,1]T, and "T" represents a transposition operation. Based on the above formulae, the following may be obtained:

[0131]

[0132] similarly,

[0133]

[0134] by that analogy, may be calculated by the following formula:

[0135]

[0136]

[0137] Thus, the problem (17) is converted into:

[0138]

[0139] When generated by the new scheduling mechanism is similar to and , the problem (20) can be solved. Ideally, the transmission matrix of each window generated by the new scheduling mechanism needs to have the same distribution as the traditional transmission matrix, that is, , and . If and are substituted into the formula (7), according to the above analysis "for one time window" (e.g., (7) to (14)), the best solution of the problem (17) is shown by the following formula:

[0140]

[0141] Thus, the scheduler shown by the formula (21) may also be designed through the C-GAN.

[0142] According to the above modeling from two different perspectives, the design of a new scheduler may be converted into the problem to be solved by the C-GAN. In the field of artificial intelligence, the C-GAN has become very popular in image generation due to its excellent performance in generating any desired image. Inspired by this, the scheduling decision matrix T may be further visualized as an image (i.e., a scheduling image) by the following steps.

[0143] In step 1 (normalization), through the formula (22), the scheduling decision matrix T may be normalized as:

[0144]

[0145] where tmaxis the number of bits which can be transmitted by each user in one slot, and is determined by the system bandwidth and the highest modulation and coding scheme (MCS).

[0146] In step 2 (visualization): τ is multiplied by 255, so that it becomes a gray image. The gray level of each pixel point may represent the number of bits transmitted in each slot. Specifically, a white pixel point indicates that the number of bits transmitted in one slot is tmax, while a black pixel point indicates that no bit is transmitted.

[0147] Finally, the design of the new scheduling algorithm is converted into obtaining a generator through the C-GAN, so that it can generate a scheduling image corresponding to T under the conditions B and A. In one example, this scheduler is called a buffer-arrival-transmission scheduler (B.A.T-scheduler).

[0148] Second aspect: design of classifier

[0149] FIG. 4 shows a block diagram of a classifier. Input parameters of the classifier may be at least one of the following parameters.

[0150] ■ Information related to the amount of buffered data: the information is used to reflect the amount of buffered data of the user on the base station side, or the amount of buffered data on each user side. One possible implementation of the information is at least one of:

[0151] ■ Mode 1: the information is a scalar. The scalar represents the amount of buffered data of all users (or traffics) at a certain moment (t), and the scalar is information related to the amount of buffered data of all users. In one example, the information is obtained by summating the amount of buffered data of all users or obtained according to a particular function (for example, adding the amount of buffered data of each user in proportion, e.g., calculating a ratio of the sum of the amount of buffered data of all users to the maximum number of bits that can be buffered by users in one slot, which is a normalization operation of the amount of buffered data of the user);

[0152] ■ Mode 2: the information is a vector. The vector represents the amount of buffered data of different users (or traffics) at a certain moment (t), and each element of the vector is information related to the amount of buffered data of one user. In one example, the information is obtained for the amount of buffered data of one user according to a particular function (for example, calculating a ratio of the amount of buffered data of the user to a constant or variable, e.g., calculating a ratio of the amount of buffered data of the user to the maximum number of bits that can be buffered by the user in one slot, which is a normalization operation of the amount of buffered data of the user); and

[0153] ■ Mode 3: the information is a matrix. The matrix represents the amount of buffered data of different users (or traffics) at different moments. Each element in the matrix represents the amount of buffered data of different users at a certain moment (t), and each element is information related to the amount of buffered data of one user. In one example, the information is obtained for the amount of buffered data of one user according to a particular function (for example, calculating a ratio of the amount of buffered data of the user to a constant or variable, e.g., calculating a ratio of the amount of buffered data of the user to the maximum number of bits that can be buffered by the user in one slot, which is a normalization operation of the amount of buffered data of the user).

[0154] ■ Information related to the amount of arrived data: the information is used to reflect the data amount of downlink data received by the user on the base station side or the data amount of uplink data on the user side. Possible implementations of the information are as follows.

[0155] ■ Mode 1: the information is a scalar. The scalar represents the amount of arrived data of all users (or traffics) at a certain moment (t), and the scalar is information related to the amount of arrived data of all users. In one example, the information is obtained by summating the amount of arrived data of all users or obtained according to a particular function (for example, calculating a ratio of the amount of arrived data of each user to the maximum number of bits that can be arrived by the user in one slot, which is a normalization operation of the amount of arrived data of the user);

[0156] ■ Mode 2: the information is a vector. The vector represents the amount of arrived data of different users (or traffics) at a certain moment (t), and each element of the vector is information related to the amount of arrived data of one user. In one example, the information is obtained for the amount of arrived data of one user according to a particular function (for example, calculating a ratio of the amount of arrived data of the user to a constant or variable, e.g., calculating a ratio of the amount of arrived data of the user to the maximum number of bits that can be arrived by the user in one slot, which is a normalization operation of the amount of arrived data of the user); and

[0157] ■ Mode 3: the information is a matrix. The matrix represents the amount of arrived data of different users (or traffics) at different moments. Each element in the matrix represents the amount of arrived data of different users at a certain moment (t), and each element is information related to the amount of arrived data of one user. In one example, the information is obtained for the amount of arrived data of one user according to a particular function (for example, calculating a ratio of the amount of arrived data of the user to a constant or variable, e.g., calculating a ratio of the amount of arrived data of the user to the maximum number of bits that can be arrived by the user in one slot, which is a normalization operation of the amount of arrived data of the user).

[0158] ■ Information related to the amount of transmitted data: the information is used to reflect the amount of downlink data transmitted to the user by the base station or the amount of uplink data transmitted by the user scheduled by the base station. Possible implementations of the information are as follows.

[0159] ■ ■ Mode 1: the information is a scalar. The scalar represents the amount of transmitted data of all users (or traffics) at a certain moment (t), and the scalar is information related to the amount of transmitted data of all users. In one example, the information is obtained by summating the amount of transmitted data of all users or obtained according to a particular function (for example, adding the amount of arrived data of each user in proportion, e.g., calculating a ratio of the sum of the amount of transmitted data of all users to the maximum number of bits that can be transmitted by users in one slot, which is a normalization operation of the amount of transmitted data of the user);

[0160] ■ ■ Mode 2: the information is a vector. The vector represents the amount of transmitted data of different users (or traffics) at a certain moment (t), and each element of the vector is information related to the amount of transmitted data of one user. In one example, the information is obtained for the amount of transmitted data of one user according to a particular function (for example, calculating a ratio of the amount of transmitted data of the user to a constant or variable, e.g., calculating a ratio of the amount of transmitted data of the user to the maximum number of bits that can be transmitted by the user in one slot, which is a normalization operation of the amount of transmitted data of the user); and

[0161] ■ ■ Mode 3: the information is a matrix. The matrix represents the amount of transmitted data of different users (or traffics) at different moments. Each element in the matrix represents the amount of transmitted data of different users at a certain moment (t), and each element is information related to the amount of transmitted data of one user. In one example, the information is obtained for the amount of transmitted data of one user according to a particular function (for example, calculating a ratio of the amount of transmitted data of the user to a constant or variable, e.g., calculating a ratio of the amount of transmitted data of the user to the maximum number of bits that can be transmitted by the user in one slot, which is a normalization operation of the amount of transmitted data of the user).

[0162] ■ Information related to channel quality: the information is used to reflect uplink and downlink channel quality (e.g., CQI) of the user. Possible implementations of the information are as follows.

[0163] ■ ■ Mode 1: the information is a scalar. The scalar represents the channel quality of all users (or traffics) at a certain moment (t), and the scalar is information related to the channel quality of all users. In one example, the information is obtained by summating the channel quality information of all users or obtained according to a particular function (for example, adding the channel quality of each user in proportion, e.g., calculating a ratio of the channel quality of all users to the best channel quality that can be achieved by the user in one slot, which is a normalization operation for the channel quality of the user, e.g., the average value, maximum value, minimum value, median value, or the like of the channel quality of all users);

[0164] ■ ■ Mode 2: the information is a vector. The vector represents the channel quality of different users (or traffics) at a certain moment (t), and each element of the vector is information related to the channel quality of one user. In one example, the information is obtained for the channel quality of one user according to a particular function (for example, calculating a ratio of the channel quality of the user to a constant or variable, e.g., calculating a ratio of the channel quality of the user to the best channel quality that can be achieved by the user in one slot, which is a normalization operation for the channel quality of the user, e.g., the average value, maximum value, minimum value, median value, or the like of the channel quality of the user); and

[0165] ■ ■ Mode 3: the information is a matrix. The matrix represents the channel quality of different users (or traffics) at different moments. Each element in the matrix represents the channel quality of different users at a certain moment (t), and each element is information related to the channel quality of one user. In one example, the information is obtained for the channel quality of one user according to a particular function (for example, calculating a ratio of the channel quality of the user to a constant or variable, e.g., calculating a ratio of the channel quality of the user to the best channel quality that can be achieved by the user in one slot, which is a normalization operation for the channel quality of the user, e.g., the average value, maximum value, minimum value, median value, or the like of the channel quality of the user).

[0166] ■ Information related to physical resources (e.g., resource blocks): the information is used to reflect the user's usage of resources. In one example, the information is the data amount of physical resources used by the user on the base station side, e.g., the number of RBs or REs or PRBs, or the number of sub-frames, or the number of frames, or the like.

[0167] ■ Information related to a user’s location: the information indicates location information of the user, e.g., cell identifier information, or GPS location information, or the like.

[0168] ■ Fingerprint of the wireless environment: the information is mainly used to indicate the channel condition of the wireless environment based on the location information under the current base station.

[0169] ■ Indication information of traffic type: the information indicates the type of traffic, such as video, voice, game, or the like.

[0170] ■ QoS parameter information of traffic: the information indicates QoS parameters of the traffic, such as latency requirement, rate requirement, latency jitter requirement, or the like.

[0171] Outputs of the classifier may be at least one of the following information.

[0172] ■ Numerical value of class: the information indicates a class of the input parameter, e.g., 0, 1, 2, 3, …. The numerical value may be obtained by a supervised or unsupervised method, or by calculation.

[0173] ■ Vector of class: the information identifies each class [C1, C2, C3] in a class sequence to which the input parameter belongs, where C1 represents the class of the input data in the first category, C2 represents a first sub-class of the input data in the class identified by C1, and C3 represents a second sub-class of the input data in the classes identified by C1 and C2.

[0174] ■ Sequence of class: the sequence may be a token sequence of words, or a token sequence composed of communication signaling primitives and the description of current communication state between the base station and the user, and is cascaded by mapping or aligning with historical data to be generated and then obtained by pre-training. The sequence may be similar to the input class to a certain extent.

[0175] An implementation of the classifier may be at least of:

[0176] ■ a mathematical method, for example, being obtained by the result of calculation, where the calculation is obtained according to a calculation formula;

[0177] ■ a supervised neural network (e.g., RNN, GRU, ARMIA, LSTM, TSMixer, PatchMixer, Transformer, etc.);

[0178] ■ an unsupervised network (e.g., a clustering network, which is applied in the scheduler to cluster according to characteristics of the communication data, such as traffic density);

[0179] ■ a multi-modal large model network (similar to the concept of the existing chatgpt large model, the concept needs to be further improved when it is applied to the communication field, for example: a tokenizer pre-trained based on the communication primitives or state description corpus is used to generate a token word list, for example, which may be a digit sequence of 1, 23, 45…, or an alphabetic word sequence of a, ab, cde, and the token word list (digit sequence 11, 23, 56) of a patch block is trained based on data matrix patch segmentation using a variational self-encoder; the two token sequence corpuses are cascaded for training a multi-modal large model to obtain a pre-trained large model network); and

[0180] ■ combinations of the above methods.

[0181] An implementation of the classifier includes at least of:

[0182] Classifier embodiment 1: the amount of arrived data is predicted by using a time sequence regression network, the range of numerical values of the transmitted data is calculated according to a formula and then classified, and the required data is finally generated by the generator according to the class.

[0183] ■ Input parameters: 1) the information related to the amount of buffered data (e.g., the information according to the above mode 1 of the "information related to the amount of buffered data"), which is buffered information of the current user in one example; and 2) the information related to the amount of arrived data in each time window in a past period of time (e.g., the information according to the above mode 1 of the "information related to the amount of arrived data").

[0184] ■ Output parameter: a classification value of the amount of data to be transmitted in a next time window.

[0185] ■ Formula-based classification: in order to obtain different classes according to the predicted amount of arrived data, the following variables are defined:

[0186] ■ τ: the normalized number of transmitted bits, in one example, a ratio of the sum of the amount of transmitted data of all users in the scheduling window to the maximum amount of data that can be transmitted by one user in one slot, which is calculated by the following formula:

[0187]

[0188] ■ α: the normalized number of arrived bits, in one example, a ratio of the sum of the amount of data of all users (or traffics) received by the base station in the scheduling window to the maximum amount of arrived data, which is calculated by the following formula:

[0189]

[0190] where amaxis the maximum number of bits that can arrived of all users in one slot.

[0191] ■ β: the normalized number of buffered bits at the beginning of the scheduling window, in one example, a ratio of the amount of buffered data of all users at the beginning of the scheduling window to the maximum buffer size, which is calculated by the following formula:

[0192]

[0193] where bmaxis the maximum buffer size of each user.

[0194] For each scheduling window, the total number of transmitted bits should not be greater than the sum of the number of bits in an initial buffer and the number of arrived bits, that is:

[0195]

[0196] According to the above defined τ, α and β, the following may be further obtained by calculation:

[0197]

[0198] The traffic load (including the buffer load β and the arrived load α) may be redefined as μ, for example:

[0199]

[0200] where β=0 and α=0 indicate that there is no data traffic, so the value of μ is defined as 6, rather than being infinite. In addition, considering that the maximum value of τ is 1, it may be limited as follows.

[0201] is a constant. The μ defined by the formula (25) may be used for the first classification, and the types such as class 1, class 2, …, class K may be determined according the value range of μ. As an example classification method, 5 classes are classified:

[0202] ■ class 1 (μ < 2): heavy load, where all samples are bounded by τ=1;

[0203] ■ class 2 (2 ≤ μ < 3): medium load, where almost all samples are bounded by τ<0.6;

[0204] ■ class 3 (3 ≤ μ < 4): low load, where the samples are strictly bounded by δ×10-μ:

[0205]

[0206] ■ class 4 (4 ≤ μ < 6): ultra-low load, where the samples are strictly bounded by δ×10-μ; and

[0207] ■ class 5 (μ = 6): no traffic load.

[0208] Further, the class 1 indicates the load in the buffer of the user equipment is heavy. For a well-designed traditional scheme, the number of heavy-load samples is much less than that of other classes. If the training of the C-GAN is directly carried out among these five classes, the imbalance in the number of samples will have a great impact on the convergence of the C-GAN. In order to solve this problem, samples of the class 1 may be trained separately, and may be further classified. Specifically, in order to ensure the QoS, in the traditional scheme, the data of users with high buffer load will be transmitted out as much as possible during scheduling. Inspired by this, based on the buffer load of each user, the samples of the class 1 are further classified (called second classification). The samples with the highest buffer load corresponding to the same user are classified into one class, and the class 1 is further classified into class 1-1, class 1-2, …, class 1-N according to the number of users. Each class corresponds to one user equipment. For example, samples of the class 1-2 are samples with the highest buffer load of the user equipment 2.

[0209] ■ Structure of the classifier: the classifier includes a predictor. The predictor may perform classification by a long short-term memory (LSTM) neural network (or similar AI networks that can be applied to time sequence prediction, such as RNN, GRU, ARMIA, TSMixer, PatchMixer, Transformer or other networks) and according to the description in the above "Formula-based classification". The classifier trains a network that predicts the arrived data. One implementation includes the following steps.

[0210] ■ ■ Step 1: the historical scheduling data generated by the base station is acquired first (referring to the parameter description in the design of classifier).

[0211] ■ ■ Step 2: the base station data obtained in each slot is processed, and α corresponding to each time window may be obtained for each time window.

[0212] ■ ■ Step 3: based on the data of each time window obtained in the step 2, a data set based on the historical data of a plurality of consecutive time windows in a past period of time (the time length may be arbitrarily set) is constructed. The following description will be given by taking the period of time being 100 time windows as an example. The data set includes multiple groups of data, and each group of data contains the information of 100 time windows, e.g., αiof 100 time windows, followed by a next αiof the 100 time windows. For example, this group of data includes α1,α2,…,α100and α101. This group of data may be used to train a group of sample data of the model. Based on the method, a training set containing multiple groups of data may be formed.

[0213] ■ ■ Step 4: a neural network is constructed and trained using the above constructed training set. Every time 100 αi(e.g., αi+1,αi+2,…,αi+100) are input, the network will generate a predicted . One or more predicted values are compared with the corresponding real value αi+101to obtain an offset (loss), and network parameters is modified through back propagation, so that the trained network gradually converges (the loss value gradually decreases).

[0214] An example structure of the LSTM network is as follows. The inputs are α1,α2,…,αMof M scheduling windows. The LSTM structure includes 5 LSTM layers. Each layer consists of a plurality of LSTM units, and the number of LSTM units is the same as the number of inputs. There is a fully connected layer after the LSTM layers, and there are 50 hidden layers.

[0215] ■ Classifier inferring: according to the above steps, the trained LSTM may be used as a predictor. The predictor can predict the load arrived in the future, e.g., α of a next scheduling window. Specifically, α1,α2,…,αMof the arrived data of the LSTM network in a period of time before the current moment (the time interval may be flexibly set, but needs to be consistent with that in training, e.g., 100 time windows) are input as inputs of the network to predict the arrived data . In combination with β of the next window, μ is calculated according to the above formula (25), and the first classification (e.g., one of the classes 1 to 5) of the next window is determined according to a size of μ. If the next window belongs to the class 1, the second classification (e.g., class 1-1, class 1-2, …, class 1-N) is performed according to the buffered load of each user at the beginning of the next window. Thus, a final class of the next window may be determined.

[0216] A main reason for classification by the above method is to predict what class of data is transmitted in the next window. A main condition that affects the numerical value or class of transmission at the next moment is derived from (the arrived load and the buffered load). The initial buffered load may be directly obtained. For the arrived load, due to different types of traffics used by users, the arrival of the service traffic may be predicted from data of the time sequence at the past moment.

[0217] The classification method based on the formulae may also be based on other formulae. For example, according to the number of UE accesses, the traffic types are classified first and the formulae are then classified. Or, the formulae are generated by using more information as the input condition. For example, the data such as the location and the channel condition (CQI) may also be used as the input condition. That is, all factors affecting the number of bits transmitted at the next moment can be used as the condition.

[0218] Classifier embodiment 2: a class of the amount of data transmitted at the next moment is directly predicted by using a time sequence regression network, and the data required by the network is finally generated according to the class.

[0219] ■ Input parameters: 1) the information related to the amount of buffered data (e.g., the information according to the above mode 1 of the "information related to the amount of buffered data"), which is buffered information of the current user in one example; and 2) the information related to the amount of arrived data in each time window in a past period of time (e.g., the information according to the above mode 1 of the "information related to the amount of arrived data").

[0220] ■ Output parameter: a classification value of data to be transmitted at the next moment.

[0221] ■ Structure of the classifier: an LSTM neural network (or similar AI networks that can be applied to time sequence prediction, such as RNN, GRU, ARMIA, TSMixer, PatchMixer, Transformer or other networks) + a fully connected classification network.

[0222] The existing network is modified by using the existing time sequence regression prediction model such as LSTM / RNN / TSMixer, a fully connected layer is added to extract characteristics of the input time sequence, and the characteristics are classified, so that the class of the data at a future moment is predicted.

[0223] Thus, a cross entropy Loss can be calculated by using classification labels of the data, so that the class at the future moment is predicted through the data of the time sequence in a past period of time. The classification labels may be constrained by the formulae in the above classifier embodiment 1 or other external measured parameters. For example, the classification network may be trained according to the labels of the input (αi, βi, i) and xi+1of different time sequences, so that it can predict the label of the data at a next moment. The classification labels may also label the data classification by unsupervised methods such as principal component analysis (PCA) and clustering, or by using some paired primitives or semantic sequences. In one example, the training of the classifier may include the following steps.

[0224] ■ ■ Step 1: the historical scheduling data generated by the base station is acquired first (referring to the parameter description in the design of classifier). A window matrix or window image is formed as a data element according to the number of user equipments and a plurality of (settable) slots, and the total amount of arrived data, the total amount of initial buffered data and the total amount of transmitted data included in the matrix are calculated as inputs of the next step.

[0225] ■ ■ Step 2: a data set based on the historical data in consecutive time window intervals is constructed (the time interval may be flexibly set). The input data constructed by the data set contains a set of the amount of arrived data at each moment in one time window. The data set includes two classes of elements (x1, x2, y), where x1 and x2 are used as inputs of the training data, and y is used as a value of regression prediction of the training data. For example, for x1, according to the amount of arrived data A(t-100) to A(t-1) at each moment, 100 numbers form an array, and the array is used as an input x1. For x2, according to the amount of buffered data B(t-100) to B(t-1) at each moment, 100 numbers form an array, and the array is used as an input x2. The y is a classification value (which may label the historical data according to the formulae or in other ways).

[0226] ■ ■ Step 3: the LSTM network is modified as a classification neural network and trained by using the historical data, and the network parameters are modified through back propagation by using the cross entropy (loss) generated by comparing the predicted value of class Y and the historical value of class Y, so that the trained network gradually converges (the loss value gradually decreases).

[0227] ■ Inferring of the classifier: the data amount A of the arrived data in a period of time (the time interval may be flexibly set, but needs to be consistent with that in training) before the current moment as and the value of the buffered data β at the current moment are input into the network trained in the above steps, and the class of the transmitted data (the total amount of the transmitted data in a window arrived at the next moment) is directly predicted.

[0228] Classifier embodiment 3: the amount of transmitted data is predicted by using a time sequence regression network, the range of the transmission efficiency of the transmitted data is calculated according to a formula and then classified, and the data required by the network is generated according to the class.

[0229] ■ Input parameter: the amount of transmitted data at each moment in a pervious period of time.

[0230] ■ Output parameter: the data amount of data to be transmitted at the next moment.

[0231] ■ Structure of the classifier: an LSTM neural network (or similar AI networks that can be applied to time sequence prediction, such as RNN, GRU, ARMIA, TSMixer, PatchMixer, Transformer or other networks) + a formula classification module (a spectrum efficiency formula). In one example, the training of the classifier may include the following steps.

[0232] ■ ■ Step 1: the historical scheduling data generated by the base station is acquired first (referring to the parameter description in the design of classifier). A window matrix or window image is formed as a data element according to the number of user equipments and a plurality of (settable) slots, and the total amount of the transmitted data corresponding to the matrix is calculated as an input of the next step.

[0233] ■ ■ Step 2: a data set based on the historical data in consecutive time window intervals is constructed (the time interval may be flexibly set). The input data constructed by the data set contains a set of the amount of transmitted data at each moment in one time window. The data set includes two classes of elements (x, y), where x is used as an input of the training data, and y is used as a value of regression prediction of the training data. For example, for x, according to the amount of transmitted data T(t-100) to T(t-1) at each moment, 100 numbers form an array, and the array is used as an input x. The y is a numerical value of the predicted input value at the moment T(t0).

[0234] ■ ■ Step 3: a neural network or regression function is constructed and trained by using the historical data, and the network parameters are modified through back propagation by using an offset (loss) generated by comparing the predicted Y value and the historical Y value, so that the trained network gradually converges (the loss value gradually decreases).

[0235] ■ Inferring of the classifier: the data amount A of the transmitted data in a period of time before the current moment (the time interval may be flexibly set, but needs to be consistent with that in training) is input into the trained network as to predict the transmitted data (the total amount of data received in a window transmitted at the next moment), and then the formula is used. For example, according to the formula and by using the amount of transmitted data inferred by the network, is classified into different classes by using an interval range of this value. Different classes may be represented as different transmission efficiency. The method is not limited to the formula-based classification, and other formulae or methods are also possible. For example, according to the number of UE accesses, the traffic types are classified first and the formulae are then classified. Or, the formulae are generated by using more information as the input condition. For example, the data such as the location and the channel condition (CQI) may also be used as the input condition. That is, all factors affecting the number of bits transmitted at the next moment can be used as the condition.

[0236] It is also possible to directly perform multi-variable regression (e.g., multi-variable LSTM, ConvLSTM, etc.) according to a matrix (x0, 0-x5, 5) of the data transmitted in the past to directly predict the matrix of X at a next moment.

[0237] Classifier embodiment 4: implementation 4 - hierarchical classifier

[0238] ■ Input parameters: the amount of transmitted data at each moment in a pervious period of time, the traffic type, and a priority.

[0239] ■ Output parameter: the class of the amount of data to be transmitted at the next moment.

[0240] ■ Structure of the classifier: an LSTM neural network (or similar AI networks that can be applied to time sequence prediction, such as RNN, GRU, ARMIA, TSMixer, PatchMixer, Transformer or other networks) + a formula classification module + a clustering classification module. In one example, the training of the classifier may include the following steps.

[0241] ■ ■ Step 1: corresponding to traffic information, the number of currently accessed user equipments and the priority of each user traffic may be directly obtained from the system.

[0242] ■ ■ Step 2: a sequence (1,5)(2,3)…(32,0) may be obtained in one scheduling window. For each element (x,y) in the sequence, x represents the number of users, and y represents the priority of the user corresponding to x. Different sequences represent different classes.

[0243] ■ ■ Step 3: a second-stage classification is performed. The second-stage classification may be performed after a first-stage classification, and the second-stage classification may be performed according to a different transmission efficiency or by the method in the above embodiment.

[0244] ■ ■ Step 4: the two classes are combined into a unified class for the generation of the conditional data.

[0245] One implementation of the embodiment is a two-stage classifier design: 1) the first stage is traffic related classification, and the classification is performed based on the following formula:

[0246] Ls=gs(S0,S1,…,SN-1),

[0247] where Sirepresents the traffic type of the user i and Si(i=0,1,…,N-1) may be one of (null, traffic 1, traffic 2, traffic S), where the "null" indicates that the user is not in the cell, and the "traffic S" represents a particular traffic (e.g., a traffic mapped as a data radio bearer). The first-stage classification can produce multiple different classes. 2) The second stage is load related classification. This classification may be performed according to the method in the above "classifier embodiment 1". This classification may be represented by the following formula:

[0248] Ll=gl(α,β,Bω),

[0249] this classification can produce at most N+4 types. By combining these two classifications, the label produced by the final classification method may be represented by the following formula:

[0250] L=(Ls-1)*(N+4)+Ll,

[0251] where Ls∈[1,…,(S+1)N], and Ll∈[1,…,N+4].

[0252] Classifier embodiment 5:the class of the amount of data transmitted at the next moment is directly predicted by using a time sequence regression network, and the data required by the network is finally generated according to the class.

[0253] ■ Input parameters: in this implementation, including the following input parameters:

[0254] ■ ■ the amount of buffered data at each moment in a previous period of time;

[0255] ■ ■ the amount of arrived data at each moment in a previous period of time;

[0256] ■ ■ a x coordinate and a y coordinate of the center of gravity of the transmitted data in the current window, where the x coordinate is calculated as follows: row number of each element of the matrix of the transmitted data is multiplied by the value of the element to obtain a number a, all elements of the matrix of the transmitted data are summated to obtain b, and the x coordinate of the center of gravity is obtained by dividing the a by the b, and the y coordinate is calculated as follows: column number of each element of the matrix of the transmitted data is multiplied by the value of the element to obtain a number a, all elements of the matrix of the transmitted data are summated to obtain b, and the coordinate y of the center of gravity is obtained by dividing the a by the b; and

[0257] ■ ■ classification labels.

[0258] ■ Output parameter: a classification value of data to be transmitted at the next moment.

[0259] ■ Structure of the classifier: an LSTM neural network (or similar AI networks that can be applied to time sequence prediction, such as RNN, GRU, ARMIA, TSMixer, PatchMixer, Transformer or other networks) + a fully connected classification network. The classifier is trained by the following steps.

[0260] ■ ■ Step 1: the historical scheduling data generated by the base station is acquired first (referring to the parameter description in the design of classifier). A window matrix or window image is formed as a data element according to the number of user equipments and a plurality of (settable) slots, and the total amount of arrived data, the total amount of initial buffered data and the total amount of transmitted data included in the matrix, the coordinate x of the center of the transmitted data in the current window and the coordinate y of the center of the transmitted data in the current window are calculated as inputs of the next step.

[0261] ■ ■ Step 2: a data set based on the historical data in consecutive time window intervals is constructed (the time interval may be flexibly set). The input data constructed by the data set contains a set of the amount of arrived data at each moment in one time window. The data set includes two classes of elements (x1,x2,x3,x4, y), where x1, x2, x3 and x4 are used as inputs of the training data, and y is used as a value of regression prediction of the training data. For example, for x1, according to the amount of arrived data A(t-100) to A(t-1) at each moment, 100 numbers form an array, and the array is used as an input x1; for x2, according to the amount of buffered data B(t-100) to B(t-1) at each moment, 100 numbers form an array, and the array is used as an input x2; x3 is the value Gx(t-100) to Gx(t-1) of the coordinate x of the center of gravity calculated according to the amount of transmitted data at each moment, 100 numbers form an array, and the array is used as an input x3; and, x4 is the value Gy(t-100) to Gy(t-1) of the coordinate y of the center of gravity calculated according to the amount of transmitted data at each moment, 100 numbers form an array, and the array is used as an input x4. The y is a classification value (which may label the historical data according to the formulae or in other ways).

[0262] ■ ■ Step 3: the LSTM network is modified as a classification neural network and trained by using the historical data, and the network parameters are modified through back propagation by using the cross entropy (loss) generated by comparing the predicted value of class Y and the historical value of class Y, so that the trained network gradually converges (the loss value gradually decreases).

[0263] ■ Inferring of the classifier: the data amount A of the arrived data in a period of time (the time interval may be flexibly set, but needs to be consistent with that in training) before the current moment as and the value β of the buffered data at the current moment are input into the network trained in the above steps, and the class of the transmitted data (the total amount of the transmitted data in the window arrived at the next moment) is directly predicted.

[0264] The classification may be performed according to the density of the image formed by the amount of transmitted data. Firstly, the data is labeled (for example, based on the traffic, based on the user, based on the transmission efficiency, based on the transmission efficiency upper bound, etc.) by the previous classification method, e.g., Class1 classification. Then, the density of the whole image is calculated. The calculation may be performed for the Tx of all traffics with a certain priority, the Tx of different traffics, or the Tx of different UEs. Then, according to the calculated density and in combination with the previous classification result, a corresponding label may be uniformly calculated. Next, the class of Tx at the next moment is predicted through the neural network according to the historical data (A, B, GTx, Gty) and the corresponding Label (or A, GTx and Gty are regressed for calculating the class). The class is input into the generative network to generate corresponding data.

[0265] Based on the above classifier, FIG. 5a shows a network structure of a joint classifier and C-GAN. The structure includes a structure for training and a structure for inferring. In another embodiment, in the training part, the classifier is an independent entity. In one embodiment, in the inferring part, the classifier includes a predictor and a classification determinator. The figure is applied to the implementation given in the embodiment of any one classifier. FIG. 5a will be illustrated below by the method in the classifier embodiment 1.

[0266] ■ Training part: this part mainly trains the C-GAN.

[0267] ■ ■ Training samples are scheduling images obtained by the traditional scheme, and each scheduling image is associated with one α and one β. In the training stage, three constituent parts, i.e., a classifier, a generator and a discriminator, are involved. The classifier deduces the class of each sample according to the μ calculated by the formula (25). In addition, it may further classify the samples of class 1 into different classes, i.e., class 1-1, class 1-2, and the like, according to the buffered load of each UE. Two generators are trained in the present invention, where one generator is trained for the class 1, class 2, ..., class y, and the other generator is trained for the class 1-1, …, class1-x. The structure of each generator includes an embedded module and five convolutional transposition modules. The embedded module realizes the multiplication of random noise and the label, and the convolutional transposition modules convert the input embedded matrix into a scheduling image matrix layer by layer. Finally, an output of the generator is directly connected to an input of the discriminator through a hyperbolic tangent activation function, and this function standardizes the training data as values in a range of [-1,1]. For the discriminator, the five convolutional modules are used for feature extraction, and a loss function used by the discriminator is a cross entropy function.

[0268] ■ Inferring part

[0269] ■ ■ The inferring part generates scheduling images by using the two generators obtained by the training part, and is used for scheduling of a next time window. The class of the next time window needs to be provided to the generators. In order to obtain the class of the next scheduling window, α (i.e., ) of the next scheduling window will be predicted by using the predictor. Specifically, considering the time correlation of the traffic data arrival, a long short-term memory (LSTM) structure can be used for prediction. The inputs of the LSTM are a plurality of α, for example, α1,α2,…,αM, of m scheduling window. The LSTM structure includes 5 LSTM layers. Each layer consists of a plurality of LSTM units, and the number of LSTM units is the same as the number of inputs. There is a fully connected layer after the LSTM layers, and there are 50 hidden layers. According to the output of the predictor, the classification determinator may determine the labels corresponding to the next scheduling window, and the output labels may be provided to different generators according to different labels. For example, the labels of class 1-1 to class 1-x are provided to the generator 1, and the labels of class 1 to class y are provided to the generator 2. Thus, different scheduling images are generated according to different generators.

[0270] ■ Scheduling

[0271] ■ ■ The images generated by the generator 1 or the generator 2 are transferred to the scheduler, and the scheduler performs scheduling according to the generated images. For example, the amount of transmitted data of the user corresponding to scheduling in a corresponding slot is determined according to the value of each pixel point of the scheduling image. For example, the scheduling image generated by the generator may be represented as a matrix I. In order to obtain the amount of data to be scheduled and transmitted by each user in each slot, each pixel point may be normalized, e.g., I / 255 and then, each pixel point is multiplied by the maximum number of bits (i.e., tmax) that can be transmitted by each user in one slot.

[0272] In the above method, since two generators are used, it may also be called a dual-generator network. Of course, other names are also possible. The scheduler designed according to the above method may also be called a buffer-arrival-transmission scheduler (B.A.T-scheduler).

[0273] Based on the design of the scheduler, the base station may also control the data transmission of the user according to the scheduling image generated by the scheduler, thereby achieving the effect of reducing the energy consumption of the user equipment. The following steps are included:

[0274] Step 1-0: the base station preconfigures the user. The preconfiguration functions to help the user equipment to operate effectively under the designed scheduling algorithm, for example, operating in an energy saving manner. This preconfiguration message includes at least one of the following information.

[0275] ■ Configuration information of the scheduling window: the information is used to indicate a configuration used when scheduling the user, so that it is convenient for the user to determine how it is scheduled, thereby ensuring correct data transmission and reception. The information includes at least one of the following information:

[0276] ■ ■ First starting offset information, e.g., a sub-frame number, the number of sub-frames, a symbol number, the number of symbols, or the like. According to the information, the user equipment may know the starting position of the scheduling window;

[0277] ■ ■ First length information: the information indicates a length of one time window for scheduling the user;

[0278] ■ ■ First period information: the information indicates a period in which the window appears; and

[0279] ■ ■ First valid time information: the information indicates a time to continuously use the configuration information of the scheduling window. In one example, only in this valid time, the user equipment can be scheduled by the designed new scheduler.

[0280] ■ Energy saving configuration information: the information is used to provide an energy saving configuration to the user, so as to help the user to save energy. The information includes at least one of the following information:

[0281] ■ ■ Second starting offset information, e.g., a sub-frame number, the number of sub-frames, a symbol number, the number of symbols, or the like. According to the information, the user equipment may know the starting position of energy saving; and

[0282] ■ ■ Second period information: the information indicates a period in which the user monitors a downlink signal (e.g., a physical downlink control channel (PDCCH)), that is, the user equipment needs to wake up for at least one slot according to the period. For example, if the period is set as 8 slots, the user equipment needs to wake up every 8 slots. For example, if there are slots 1 to 16, the user needs to wake up in slot 1 and slot 9 to monitor the downlink signal. Further, when the user equipment receives sleep indication information (e.g., the first indication information in step 1-3 described below) transmitted by the base station, the user will not monitor the downlink signal in the remaining slots after the period.

[0283] Step 1-1: the base station generates a scheduling image (i.e., a matrix, where the row number represents the user equipment or traffic, and each column represents one slot) according to the above designed scheduler, and the base station determines, according to the scheduling image, the number of bits to be scheduled and transmitted by each user in each slot. If the number of bits is 0 in one slot, it is unnecessary to schedule this user.

[0284] Step 1-2: the base station performs physical resource allocation and data transmission according to the number of bits to be scheduled by each user in each slot obtained in the step 1-1. In one example, the base station will transmit the number of bits scheduled by each user in one slot generated in the step 1-1 at the maximum capability. Every time one slot passes, the base station will update the data in the buffer of each user. If the transmission is successful, the data will be removed from the buffer, otherwise, the data still remains in the buffer. If new data arrives in one slot, the base station will use the new arrived data as a part of the data in the buffer. In order to determine the amount of data to be finally transmitted for scheduling of the user in one slot, at least one of the following parameters needs to be taken into consideration:

[0285] ■ the amount of scheduled data: the amount of data scheduled by one user in one slot obtained in the step 1-1, e.g., the number of bits;

[0286] ■ the number of buffered data: the amount of data in the buffer of the user at the beginning of one slot; and

[0287] ■ the maximum amount of transmitted data: the amount of data that can be transmitted at most by the user equipment in one slot (it depends on the occupation of channels and radio resources of the user equipment, etc.).

[0288] The amount of data scheduled and transmitted finally is the minimum value of the above parameters.

[0289] Step 1-3: the base station transmits a first indication message of data transmission to the user. The information is used to indicate whether the user has data to be transmitted in the next slot or whether the user equipment needs to monitor a downlink signal (e.g., PDCCH) in the next slot. If the indication information indicates no data transmission, the user may select to stop the monitoring of the wireless channel (e.g., the monitoring of the PDCCH), so that the energy consumption of the user can be reduced. The method of determining the indication information by the base station comprises: by the base station, determining whether there is data in the buffer of the user equipment at the beginning of the next slot, or determining whether the smallest value in the amount of scheduled data, the amount of buffered data and the maximum amount of transmitted data is 0. If there is no data in the buffer or the smallest value in the amount of scheduled data, the amount of buffered data and the maximum amount of transmitted data is 0, the base station transmits the indication message to the user to indicate that the user has no data transmission in the subsequent slot or the user equipment can sleep in the subsequent slot. The first indication information may include at least one of the following information.

[0290] ■ Sleep indication information: the information indicates that the user equipment may stop monitoring the downlink signal (e.g., PDCCH). In one example, the user equipment will stop monitoring the downlink signal in the remaining time in one configured period, for example, stopping monitoring the downlink signal in the remaining time (the remaining time is from the time when the indication information is received from the user equipment to the beginning of the next period indicted by the first period information or the second period information) in the first period information or the second period information configured in the step 1-0; and

[0291] ■ Sleep time information: the information indicates a time length in which the user equipment may stop monitoring the downlink signal, e.g., the number of slots or the number of symbols, the number of periods indicated by the "first period information" or the "second period information", or the like.

[0292] In the above method, the time of monitoring the downlink signal by the user equipment can be reduced, and the energy consumption of the user equipment can also be reduced. Moreover, since the network side (the first network node) can know that no user data needs to be transmitted in which slots, the energy consumption of the network is also saved. For example, in each scheduling window, the first network node determines, according to the "method of determining the indication information by the base station", that no user is scheduled in which slots. Thus, the network side can reduce the transmission of signals in these slots, for example, reducing the activated cells, reducing the transmission of reference signals, or the like. Thus, the energy consumption of the first network node can also be reduced.

[0293] In the design scheme of the scheduler, the design of a self-optimizer may also be included. The self-optimizer may be used to optimize the generator. FIG. 5b shows an implementation of combining training, inferring and the self-optimizer. The training and inferring parts have been given in the description of FIG. 5a and will not be repeated here. For the self-optimizer, possible implementations are as follows:

[0294] ■ The self-optimizer includes a digital twin platform for simulating the scheduling algorithm. The platform may generate scheduling data (e.g., scheduling images) based on different scheduling algorithms (e.g., proportional fairness algorithms).

[0295] ■ The self-optimizer also includes a best sample selector. The selector functions to determine samples that can assist in improving the performance of the generator, and then use these samples for iterative training in the training process, so as to optimize the generator. Specifically, one example of selecting the best samples by the selector is as follows:

[0296] ■ ■ Scheduling data 1 (e.g., scheduling image) is generated by the generator. The scheduling data 1 is used for scheduling of the user, and the user performance 1 (e.g., QoS performance) that can be achieved by the scheduling data 1 is evaluated.

[0297] ■ ■ Scheduling data 2 of other traditional scheduling algorithms (e.g., proportional fairness algorithm) is generated by the digital twin platform, and the user performance 2 (e.g., QoS performance) that can be achieve by the scheduling data 2 is evaluated in the platform.

[0298] ■ ■ In one implementation, the selector determines the scheduling data 1 having better user performance 1 than the user performance 2, and uses the scheduling data 1 as a training sample for iterative training of the generator, so as to gradually update the generator. In another implementation, the selector determines the scheduling data 2 having better user performance 2 than the user performance 1, and uses the scheduling data 2 as a training sample for iterative training of the generator, so as to gradually update the generator. In another implementation, the selector determines the best one of the user performance 1 and the user performance 2, and uses the scheduling data corresponding to the best user performance as a training sample for iterative training of the generator, so as to gradually update the generator.

[0299] According to the above method, one possible effect is that the generator obtained by the iterative training will produce better performance than other scheduling algorithms (e.g., proportional fairness algorithm). Further, a self-optimization method can also be applied in a scenario where the network condition is changed (e.g., the users served by the network are increased or decreased, the service served by the network is changed, or there are new services in the network). The self-optimizer can obtain better samples than other scheduling algorithms and use these samples for the iterative training of the generator. In an iterative training process, according to the network state, new classes are defined or the existing classes are used, and these classes are used for training the generator. Thus, in combination with the self-optimizer, the method of training the generator can adaptively train and obtain the generator according to the change of the network state, thereby increasing the range of application of the generator and even replacing the traditional scheduling algorithms.

[0300] In order to apply the scheduler designed by the present invention to commercial networks, the following different stages may be experienced.

[0301] ■ Stage 1 (initial stage): in this stage, the user is scheduled based on the traditional scheduling algorithms. A transmission matrix generated by scheduling can be used for training the B.A.T scheduler.

[0302] ■ Stage 2 (coexistence stage): in this stage, the B.A.T scheduler (e.g., B.A.T scheduler v0.0) trained in the stage 1 may schedule the user. The scheduling of the user in each slot is determined by the B.A.T scheduler v0.0 or a traditional scheduler. If the B.A.T scheduler v0.0 has better performance than the traditional scheduler (for example, the rate arrived by the user in this slot is higher), scheduling strategies of the B.A.T scheduler v0.0 are adopted; otherwise, the scheduling strategies of the traditional scheduler are adopted. A transmission matrix generated based on the method can be used for further training the B.A.T scheduler.

[0303] ■ Stage 3: in this stage, the scheduling of the user is completely determined by the B.A.T scheduler trained in the stage 2 (e.g., B.A.T scheduler v1.0). Meanwhile, a digital twin network running the traditional scheduling algorithm also operates synchronously. If scheduling strategies generated by the digital twin network are better than that of the B.A.T scheduler, the scheduling strategies generated by the digital twin network can be used for further training the B.A.T scheduler.

[0304] ■ Stage 4: in this stage, the B.A.T scheduler generated in the stage 3 (e.g., B.A.T scheduler v2.0) completely takes the place of the traditional scheduler, and the scheduling strategies generated by the B.A.T scheduler can be used for further training the scheduler.

[0305] FIG. 6 is a block diagram of a node according to an example embodiment of the present disclosure. Here, the structure and function are described by taking the node as an example. However, it should be understood that, the illustrated structure and function can also be applied to a base station (or a centralized unit of the base station, or a control panel part of the centralized unit of the base station, or a user panel part of the centralized unit of the base station, or a distributed unit of the base station, or a network node, or the like).

[0306] Referring to FIG. 6, the node 1000 includes a transceiver 1010, a controller 1020 and a memory 1030. Under the control of the controller 1020 (which may be implemented as one or more processors), the node 1000 (including the transceiver 1010 and the memory 1030) is configured to perform operations of the node described herein. Although the transceiver 1010, the controller 1020, and the memory 1030 are illustrated as separate entities, they may be implemented as a single entity, such as a single chip. The transceiver 1010, the controller 1020, and the memory 1030 may be electrically connected or coupled to each other. The transceiver 1010 may transmit signals to other network entities and receive signals from other network entities. For example, the other network entities are another node and / or UE, and the like. In one implementation, the transceiver 1010 may be omitted. In this case, the controller 1020 may be configured to execute the instructions (including computer programs) stored in the memory 1030 to control the overall operation of the node 1000, so as to implement the operations of the node described herein.

[0307] According to some implementations, the user equipment described herein may include: cellular or other communication devices with a single-line display or multi-line display or without a multi-line display; personal communication systems (PCSs) that can integrate voice, data processing, facsimile and / or data communication capabilities; personal digital assistants (PDAs), which can include RF receivers, pagers, internet networks / intranet accesses, web browsers, notepads, calendars and / or global positioning system (GPS) receivers; and conventional laptop and / or palmtop computers or other devices having and / or including a RF receiver. The "terminal" and "end device" as used herein may be portable, transportable, mountable in vehicles (aerial, maritime and / or terrestrial), or suited and / or configured to run locally and / or distributed in other places in the earth and / or space for running. The "terminal" or "end device" as used herein may also be a communication terminal, an internet terminal, a music / video player terminal. For example, it may be a PDA, a mobile Internet device (MID) and / or a mobile phone with a music / video playback function, or may be devices such as a smart TV, a set-top box, or the like.

[0308] It should be recognized by those skilled in the art that the present disclosure can be implemented in other particular forms without changing the technical ideas or basic characteristics of the present disclosure. Thus, it should be understood that the above embodiments are only examples and are not limiting. The scope of the present disclosure is defined by the appended claims, instead of the detailed description. Thus, it should be understood that all modifications or changes derived from the meaning and scope of the appended claims and their equivalents shall fall into the scope of the present disclosure.

[0309] In the embodiments of the present disclosure, all operations and messages can be selectively performed or omitted. In addition, the operations in each embodiment are not necessarily performed sequentially, and the order of the operations can be changed. The messages are not necessarily transmitted sequentially, and the transmission order of the messages may be changed. Each operation and the transmission of each message can be performed independently.

[0310] Although the present disclosure has been illustrated and described with reference to various embodiments thereof, it should be understood by those skilled in the art that, various changes can be made in forms and details without departing from the spirit and scope of the present disclosure as defined by the appended claims and their equivalents.

Claims

1.A method performed by a first network node in a wireless communication system, comprising:transmitting, to a first node, a first configuration message including at least one of configuration information of a scheduling window and energy saving configuration information;generating, based on a scheduling module trained by a generative network, scheduling information of at least one first node in at least one time unit; andtransmitting, to the first node, a first indication message for instructing the first node to perform downlink signal monitoring,wherein the scheduling information is obtained according to a matrix or an image generated by the scheduling module.2.The method of claim 1, whereinthe scheduling module trained by the generative network comprises a classifier, at least one generator and a scheduler,the classifier is configured to generate different types of indication information about input data,the generator generates the scheduling information according to the different types of indication information generated by the classifier, andthe scheduler schedules data transmission of the first node according to the scheduling information.3.The method of claim 1 or 2, whereinfor one slot of one of the first nodes, the scheduling information further comprises at least one of bit number information, resource allocation information, and indication information of the type of transmission traffic.4.The method of claim 1,wherein the configuration information of the scheduling window comprises at least one of first starting offset information, first length information, first period information, and first valid time information.5.The method of claim 1, whereinthe first indication message comprises at least one of sleep indication information and sleep time information.6.The method of claim 2, whereinthe input data of the classifier comprises at least one of information related to the amount of buffered data, information related to the amount of arrived data, information related to the amount of transmitted data, information related to channel quality, information related to physical resources, information related to a user's location, and fingerprint information of a wireless environment, andoutput data of the classifier is the different types of indication information, and comprises at least one of a numerical value of a class, a vector of the class, a sequence of the class, and a token of the class.7.The method of claim 2, whereinthe classifier comprises a predictor configured to predict information related to a next scheduling window, andan output of the predictor is used by the classifier to generate the output data.8.The method of claim 2, whereinthe at least one generator is selected according to the output data of the classifier to generate the scheduling information.9.The method of claim 2, whereinthe classifier further generates first classification information and second classification information, andthe first classification information is used by a first generator in the at least one generator to generate the scheduling information, and the second classification information is used by a second generator in the at least one generator to generate the scheduling information.10.The method of claim 2, whereinthe at least one generator is trained according to a conditional generative adversarial network.11.A method performed by a first node in a wireless communication system, comprising:receiving, from a first network node, a first configuration message including at least one of configuration information of a scheduling window and energy saving configuration information;scheduling by the first network node, the scheduling being performed by generating, based on a scheduling module trained by a generative network, scheduling information of at least one first node in at least one time unit; andreceiving, from the first network node, a first indication message for instructing the first node to perform downlink signal monitoring,wherein the scheduling information is obtained according to a matrix or an image generated by the scheduling module.12.The method of claim 11, whereinthe scheduling module trained by the generative network comprises a classifier, at least one generator and a scheduler,the classifier is configured to generate different types of indication information about input data,the generator generates the scheduling information according to the different types of indication information generated by the classifier, andthe scheduler schedules data transmission of the first node according to the scheduling information.13.The method of claim 11, whereinfor one slot of one of the first nodes, the scheduling information further comprises at least one of bit number information, resource allocation information, and indication information of the type of transmission traffic.14.The method of claim 11,wherein the configuration information of the scheduling window comprises at least one of first starting offset information, first length information, first period information, and first valid time information.15.A first network node or first node in a wireless communication system, comprising:a transceiver configured to transmit and receive signals; anda controller coupled to the transceiver and configured to perform the method of one of the preceding corresponding claims.

Images