Training data set acquisition method, wireless transmission method, device and communication equipment

Abstract

The application discloses a training data set acquisition method, a wireless transmission method, a device and communication equipment, and belongs to the technical field of communication. The training data set acquisition method comprises the following steps: determining the data amount of training data under each transmission condition based on the contribution degree of each transmission condition to a neural network optimization target; and acquiring the training data under each transmission condition based on the data amount of the training data under each transmission condition, so as to form a training data set for training the neural network; wherein the contribution degree of the transmission condition to the neural network optimization target represents the influence degree of the transmission condition on the value of the neural network optimization target.

Description

Technical Field

[0001] This application belongs to the field of communication technology, specifically relating to a training dataset acquisition method, a wireless transmission method, a device, and a communication equipment. Background Technology

[0002] Generalization refers to the ability of a neural network to produce reasonable outputs even on data not encountered during its training (learning) process. Currently, to achieve generalization capabilities for diverse wireless transmission environments, a common neural network can be trained based on mixed data, whose parameters do not need to be switched with changes in the environment. However, this neural network cannot achieve optimal performance under every transmission condition. Summary of the Invention

[0003] This application provides a training dataset acquisition method, a wireless transmission method, an apparatus, and a communication device, which can solve the problems such as insufficient generalization ability of neural networks in existing wireless transmission.

[0004] Firstly, a method for obtaining a training dataset is provided, the method comprising:

[0005] Based on the contribution of each transmission condition to the optimization objective of the neural network, the amount of training data under each transmission condition is determined.

[0006] Based on the amount of training data under each of the aforementioned transmission conditions, training data under each of the aforementioned transmission conditions is obtained to form a training dataset for training the neural network.

[0007] The contribution of the transmission conditions to the optimization objective of the neural network represents the degree of influence of the transmission conditions on the value of the optimization objective of the neural network.

[0008] Secondly, a training dataset acquisition device is provided, comprising:

[0009] The first processing module is used to determine the amount of training data under each transmission condition based on the contribution of each transmission condition to the optimization objective of the neural network.

[0010] The second processing module is used to acquire training data under each transmission condition based on the amount of training data under each transmission condition, so as to form a training dataset for training the neural network.

[0011] The contribution of the transmission conditions to the optimization objective of the neural network represents the degree of influence of the transmission conditions on the value of the optimization objective of the neural network.

[0012] Thirdly, a wireless transmission method is provided, the method comprising:

[0013] Based on a neural network model, wireless transmission calculations are performed to achieve the wireless transmission.

[0014] The neural network model is obtained by training a training dataset in advance, and the training dataset is obtained based on the training dataset acquisition method described in the first aspect.

[0015] Fourthly, a wireless transmission device is provided, comprising:

[0016] The third processing module is used to perform wireless transmission calculations based on a neural network model to realize the wireless transmission.

[0017] The neural network model is obtained by training a training dataset in advance, and the training dataset is obtained based on the training dataset acquisition method described in the first aspect.

[0018] Fifthly, a communication device is provided, the communication device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method as described in the first aspect, or implement the steps of the method as described in the third aspect.

[0019] In a sixth aspect, a communication device is provided, including a processor and a communication interface, wherein the processor is configured to determine the amount of training data under each transmission condition based on the contribution of each transmission condition to the optimization objective of the neural network; and to acquire the training data under each transmission condition based on the amount of training data under each transmission condition, so as to form a training dataset for training the neural network; wherein the contribution of the transmission condition to the optimization objective of the neural network represents the degree of influence of the transmission condition on the value of the optimization objective of the neural network.

[0020] In a seventh aspect, a communication device is provided, including a processor and a communication interface, wherein the processor is used to perform wireless transmission calculations based on a neural network model to realize the wireless transmission; wherein the neural network model is obtained by pre-training using a training dataset, and the training dataset is obtained based on the training dataset acquisition method described in the first aspect.

[0021] Eighthly, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect, or implement the steps of the method described in the third aspect.

[0022] In a ninth aspect, a chip is provided, the chip including a processor and a communication interface coupled to the processor, the processor being configured to run programs or instructions to implement the method as described in the first aspect, or to implement the method as described in the third aspect.

[0023] In a tenth aspect, a computer program / program product is provided, the computer program / program product being stored in a non-transient storage medium, the program / program product being executed by at least one processor to implement the steps of the training dataset acquisition method as described in the first aspect, or to implement the steps of the wireless transmission method as described in the third aspect.

[0024] In this embodiment of the application, when constructing a training dataset in an artificial intelligence-based communication system, data under various transmission conditions are selected in different proportions according to the contribution of data under different transmission conditions to the optimization objective (or objective function or loss function) of the neural network, and a hybrid training dataset is constructed, which can effectively improve the generalization ability of the neural network. Attached Figure Description

[0025] Figure 1 This is a structural diagram of a wireless communication system applicable to embodiments of this application;

[0026] Figure 2 A flowchart illustrating the training dataset acquisition method provided in this application embodiment;

[0027] Figure 3 This is a schematic diagram of the structure of the training dataset acquisition device provided in the embodiments of this application;

[0028] Figure 4 A flowchart illustrating the wireless transmission method provided in an embodiment of this application;

[0029] Figure 5 This is a schematic diagram illustrating the process of constructing a neural network model in the wireless transmission method provided according to an embodiment of this application;

[0030] Figure 6 This is a flowchart illustrating the process of determining the proportion of training data in the training dataset acquisition method provided in the embodiments of this application;

[0031] Figure 7 This is a schematic diagram of the neural network structure used for DMRS channel estimation in the wireless transmission method provided according to an embodiment of this application;

[0032] Figure 8 This is a schematic diagram of the structure of the wireless transmission device provided in the embodiments of this application;

[0033] Figure 9 This is a schematic diagram of the structure of a communication device provided in an embodiment of this application;

[0034] Figure 10 A schematic diagram of the hardware structure of a terminal to implement an embodiment of this application;

[0035] Figure 11 A schematic diagram of the hardware structure of an access network device according to an embodiment of this application;

[0036] Figure 12 A schematic diagram of the hardware structure of a core network device according to an embodiment of this application; Detailed Implementation

[0037] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0038] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, not limited in number; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0039] It is worth noting that the technologies described in this application are not limited to Long Term Evolution (LTE) / LTE-Advanced (LTE-A) systems, but can also be used in other wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency-Division Multiple Access (SC-FDMA), and other systems. The terms "system" and "network" in this application are often used interchangeably, and the described technologies can be used with the systems and radio technologies mentioned above, as well as with other systems and radio technologies. The following description describes New Radio (NR) systems for illustrative purposes, and the term NR is used in most of the following description; however, these technologies can also be applied to applications other than NR systems, such as 6th generation (6G) radio systems. th Generation 6G communication system.

[0040] Figure 1This diagram illustrates a structural diagram of a wireless communication system applicable to embodiments of this application. The wireless communication system includes a terminal 101 and a network-side device 102. The terminal 101 can also be referred to as a terminal device or user equipment (UE). Terminal 101 can be a mobile phone, tablet computer, laptop computer, personal digital assistant (PDA), handheld computer, netbook, ultra-mobile personal computer (UMPC), mobile internet device (MID), wearable device, vehicle-mounted device (VUE), pedestrian terminal (PUE), etc. Wearable devices include smartwatches, wristbands, headphones, glasses, etc. It should be noted that embodiments of this application do not limit the specific type of terminal 101. The network-side device 102 can be an access network device 1021, a core network device 1022, or a data network (DN) device 1023. The access network device 1021 can also be called a radio access network device or a radio access network (RAN). The access network device can be a base station or a node on the RAN side responsible for neural network training, etc. The base station can be called a node B, evolved node B, access point, base transceiver station (BTS), radio base station, radio transceiver, basic service set (BSS), extended service set (ESS), B node, evolved B node (eNB), home B node, home evolved B node, WLAN access point, WiFi node, transmitting and receiving point (TRP), or any other suitable term in the field. As long as the same technical effect is achieved, the base station is not limited to specific technical terms. It should be noted that in the embodiments of this application, only the base station in the NR system is used as an example, but the specific type of base station is not limited.The core network equipment 1022 can also be referred to as the core network (CN) or 5G core (5GC) network. The core network equipment 1022 may include, but is not limited to, at least one of the following: core network nodes, core network functions, Mobility Management Entity (MME), Access Management Function (AMF), Session Management Function (SMF), User Plane Function (UPF), Policy Control Function (PCF), Policy and Charging Rules Function (PCRF), Edge Application Server Discovery Function (EASDF), and Application Function (AF), etc. It should be noted that this embodiment only uses a core network equipment in a 5G system as an example, but it is not limited to this. Data network device 1023 may include, but is not limited to, at least one of the following: Network Data Analytics Function (NWDAF), Unified Data Management (UDM), Unified Data Repository (UDR), and Unstructured Data Storage Function (UDSF). It should be noted that this embodiment uses a data network device in a 5G system as an example, but is not limited to this.

[0041] The training dataset acquisition method, wireless transmission method, apparatus, and communication equipment provided in this application will be described in detail below with reference to the accompanying drawings and through some embodiments and application scenarios.

[0042] Figure 2 The diagram shown is a flowchart illustrating a training dataset acquisition method provided in an embodiment of this application. This method can be executed by a terminal and / or a network-side device. Specifically, the terminal can be... Figure 1 The terminal 101 shown in the diagram, specifically the network-side device, can be... Figure 1 The network-side device 102 is shown in the image. For example... Figure 2 As shown, the method includes:

[0043] Step 201: Based on the contribution of each transmission condition to the optimization objective of the neural network, determine the amount of training data under each transmission condition.

[0044] The contribution of the transmission conditions to the optimization objective of the neural network represents the degree of influence of the transmission conditions on the value of the optimization objective of the neural network.

[0045] This can be understood as follows: for each transmission condition in the target wireless transmission environment, the embodiments of this application can pre-test the degree of influence of each transmission condition on the value of the neural network optimization target, that is, by testing the degree of influence of the value of the transmission condition on the value of the optimization target, the contribution of each transmission condition to the neural network optimization target can be determined. Generally, the higher the degree of influence, the greater the contribution value of the corresponding transmission condition, and vice versa.

[0046] For example, when the optimization objective is expressed as a function of transmission conditions, if the function is an increasing function of transmission conditions (e.g., throughput is an increasing function of SNR), then the transmission conditions that result in a larger optimization objective value contribute more; if the function is a decreasing function of transmission conditions (e.g., NMSE is a decreasing function of SNR), then the transmission conditions that result in a smaller optimization objective value contribute more.

[0047] Optionally, for any transmission condition, the contribution of the transmission condition to the optimization objective can be determined based on the optimal value achievable under that transmission condition. In other words, the contribution of a given transmission condition to the optimization objective can be measured by the magnitude of the optimal value achievable under that given transmission condition. The larger the optimal value achievable, the greater the influence of the given transmission condition on the optimization objective, and the greater its contribution.

[0048] Based on the determination of the contribution of each transmission condition, this embodiment of the application can adjust the amount of training data corresponding to each transmission condition according to the magnitude of the contribution value of each transmission condition. This means that the amount of training data under each transmission condition is correlated with the contribution of that transmission condition to the optimization objective. Here, the amount of training data is a reference amount, or a set amount, of training data to be prepared for each transmission condition. This amount of data must be referenced when actually acquiring the training dataset and preparing training data for each transmission condition.

[0049] It is understandable that, in order to improve the generalization ability of neural networks to various transmission conditions, the amount of data for transmission conditions with high contribution can be reduced to decrease their influence; at the same time, the amount of data for transmission conditions with low contribution can be increased to increase their influence. That is, the impact of each transmission condition on the optimization objective is balanced by the amount of data. Here, the amount of data can refer to either the absolute value of the data or the relative proportion of the data.

[0050] The transmission conditions refer to parameters such as the transmission medium, transmission signal, and transmission environment involved in the actual wireless transmission environment. Optionally, the types of transmission conditions include at least one of the following:

[0051] Signal-to-noise ratio (SNR) or signal-to-interference-plus-noise ratio (SINR);

[0052] Reference Signal Receiving Power (RSRP);

[0053] Signal strength;

[0054] Interference intensity;

[0055] Terminal movement speed;

[0056] Channel parameters;

[0057] The distance between the terminal and the base station;

[0058] Size of the community;

[0059] Carrier frequency;

[0060] Modulation order or modulation coding strategy;

[0061] Community type;

[0062] Station spacing;

[0063] Weather and environmental factors;

[0064] Antenna configuration information at the transmitting or receiving end;

[0065] Terminal capabilities or type;

[0066] Base station capabilities or types.

[0067] It can be understood that the types of transmission conditions involved in the embodiments of this application may include, but are not limited to, one or more combinations of the transmission condition types listed above.

[0068] Interference intensity can represent, for example, the intensity of co-channel interference between cells, or the magnitude of other interference; channel parameters such as path number (or LOS or NLOS scenario), delay (or maximum delay), Doppler (or maximum Doppler), angle of arrival (including horizontal and vertical) range, departure angle (including horizontal and vertical) range, or channel correlation coefficient; cell type such as indoor cell, outdoor cell, macro cell, micro cell, or pico cell; inter-station spacing such as within 200 meters, 200-500 meters, or more than 500 meters; weather and environmental factors such as temperature and / or humidity of the network environment where the training data is located; antenna configuration information of the transmitting or receiving end, such as the number of antennas and / or antenna polarization; UE capability / type, such as Redcap UE and / or normal UE.

[0069] Step 202: Based on the amount of training data under each transmission condition, obtain the training data under each transmission condition to form a training dataset for training the neural network.

[0070] This embodiment of the application, based on the acquisition of the amount of training data under each transmission condition (i.e., based on the acquisition of the amount of reference data), uses the reference data amount corresponding to each transmission condition as a reference or setting to acquire training data under each transmission condition, so that the amount of training data acquired under each transmission condition is consistent with the reference or setting. Finally, the acquired training data under each transmission condition is non-uniformly mixed to obtain a dataset, which is the training dataset. This training dataset can be used to train the aforementioned neural network or neural network model in a wireless transmission environment.

[0071] Optionally, acquiring training data under each of the transmission conditions to form a training dataset for training the neural network includes: collecting and labeling data under each of the transmission conditions based on the amount of training data under each of the transmission conditions to form a training dataset under each of the transmission conditions; or collecting a set number of data under each of the transmission conditions, and selecting and labeling a portion of the data from the set number of data based on the amount of training data under each of the transmission conditions, or supplementing and labeling the set number of data to form a training dataset under each of the transmission conditions.

[0072] This can be understood as follows: When constructing a training dataset based on the amount of training data under each determined transmission condition, the embodiments of this application can first calculate the amount of data required for each transmission condition, and then obtain the data for each transmission condition based on that amount of data. Alternatively, a large amount of data for each transmission condition can be obtained first, that is, a set amount of data can be obtained, and then the amount of data required for each transmission condition can be calculated. After that, the data can be selected or supplemented from the large amount of data obtained in advance.

[0073] In the latter case, it is assumed that the total amount of data obtained in advance under the k-th transmission condition is M. k The required amount of data, determined through calculation, is N. k If M k ≥N k Then it is necessary to start from this M k Randomly select N from the data. k Add M data points to the training dataset; if M k <N k Then N needs to be obtained again. k -M k Data points, fill in N. k After the initial data is processed, it is then added to the training dataset. The required data volume is the same as the training data volume determined under each transmission condition in the previous step.

[0074] After acquiring data under various transmission conditions, this data needs to be tagged, that is, a label needs to be added to the data under each transmission condition. The added label is the true value of the transmission environment corresponding to that data. For example, in the DMRS channel estimation scenario, the label added to the DMRS signal data under each transmission condition is the true value of the channel corresponding to that DMRS signal.

[0075] It should be noted that the embodiments of this application can be applied to any scenario where machine learning can replace the functionality of one or more modules in an existing wireless transmission network. That is, when training a neural network using machine learning, the training dataset acquisition method of the embodiments of this application can be used to construct the training dataset. Application scenarios include, for example, pilot design, channel estimation, signal detection, user pairing, HARQ, and positioning at the physical layer; resource allocation, handover, and mobility management at the higher layers; and scheduling or slicing at the network layer. The embodiments of this application do not limit the specific wireless transmission application scenarios.

[0076] In this application embodiment, when constructing a training dataset in an artificial intelligence-based communication system, data under various transmission conditions are selected in different proportions according to the contribution of data under different transmission conditions to the optimization objective (or objective function or loss function) of the neural network, and a hybrid training dataset is constructed, which can effectively improve the generalization ability of the neural network.

[0077] Optionally, determining the amount of training data under each transmission condition based on the contribution of each transmission condition to the neural network optimization objective includes: ranking the contribution of each transmission condition; and, on the basis of proportional mixing, performing at least one of the following operations: reducing the amount of training data under the transmission condition with the larger contribution in the ranking and increasing the amount of training data under the transmission condition with the smaller contribution in the ranking.

[0078] This can be understood as follows: When determining the amount of training data under each transmission condition, the embodiments of this application can first rank the contributions of data from different transmission conditions to the neural network optimization objective (or objective function, loss function) when mixed in equal proportions. Then, when constructing the mixed training dataset, the smaller and larger contributions can be determined based on the above ranking, thereby further determining the transmission conditions with lower and higher contributions. Furthermore, while ensuring sufficient data for all transmission conditions, the amount of data for transmission conditions with lower contributions can be increased, and / or the amount of data for transmission conditions with higher contributions can be decreased.

[0079] One approach is to set a threshold (i.e., a preset threshold) and ensure that the proportion of data for any transmission condition to the total data is not lower than the threshold, thereby satisfying the aforementioned premise of "ensuring sufficient data for all transmission conditions".

[0080] This application embodiment sorts the contribution of each transmission condition and adjusts the data volume of the corresponding transmission condition according to the sorting. This enables a clearer and more accurate determination of the adjustment strategy for the data volume of the corresponding transmission condition (including whether to increase or decrease the data volume, the magnitude of the increase or decrease, etc.), thereby making the efficiency higher and the results more accurate.

[0081] Optionally, performing at least one of the operations of reducing the amount of training data under the transmission conditions corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission conditions corresponding to the smaller contribution in the ranking includes: performing at least one of the operations of reducing the amount of training data under the transmission conditions corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission conditions corresponding to the smaller contribution in the ranking according to the following rules, wherein the following rules include: the larger the value of the larger contribution, the greater the reduction; the smaller the value of the smaller contribution, the greater the increase.

[0082] This can be understood as follows: when increasing the amount of data for transmission conditions with lower contribution and decreasing the amount of data for transmission conditions with higher contribution in the embodiments of this application, the goal is to ensure that the amount of training data for each transmission condition gradually decreases as the contribution of the transmission condition gradually increases. Therefore, the lower the contribution of a transmission condition, the more data is added for that transmission condition; conversely, the higher the contribution of a transmission condition, the more data is reduced for that transmission condition.

[0083] The embodiments of this application increase or decrease the amount of training data for corresponding transmission conditions proportionally according to the magnitude of the contribution value. This allows the amount of training data for transmission conditions to gradually decrease as the contribution of each transmission condition gradually increases, thereby better balancing the impact of each transmission condition on the final neural network and improving the generalization ability of the neural network.

[0084] Optionally, if the sorting result is from small to large, the amount of training data under the transmission conditions decreases in the direction of the sorting; if the sorting result is from large to small, the amount of training data under the transmission conditions increases in the direction of the sorting.

[0085] This can be understood as follows: when determining the amount of training data for each transmission condition based on the above embodiments, if the contribution ranking is from smallest to largest, the proportion of the data for the corresponding transmission condition to the total data can decrease in any way, such as linearly decreasing, arithmetically decreasing, geometrically decreasing, exponentially decreasing, or power-law decreasing. Conversely, if the contribution ranking is from largest to smallest, the proportion of the data for the corresponding transmission condition to the total data can increase in any way, such as linearly increasing, arithmetically increasing, geometrically increasing, exponentially increasing, or power-law increasing.

[0086] Optionally, at least one of the operations of reducing the amount of training data under the transmission condition corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission condition corresponding to the smaller contribution in the ranking includes: determining a reference contribution based on the ranking, and comparing the contribution of the transmission condition with the reference contribution.

[0087] Based on the comparison results, at least one of the following operations is performed, including:

[0088] If the contribution of the transmission condition is greater than the reference contribution, then the contribution of the transmission condition is determined to be the larger contribution, and the amount of training data under the transmission condition is reduced.

[0089] If the contribution of the transmission condition is not greater than the reference contribution, then the contribution of the transmission condition is determined to be the smaller contribution, and the amount of training data under the transmission condition is increased.

[0090] This can be understood as follows: when adjusting the amount of training data under various transmission conditions according to the ranking in this embodiment of the application, a reference contribution value can be determined first based on the ranking. Optionally, the reference contribution value is the median of the ranking, or the contribution value at a set position in the ranking, or the average of the contributions in the ranking, or the contribution value in the ranking that is closest to the average. The average can be an arithmetic mean, geometric mean, harmonic mean, weighted mean, squared mean, or exponential mean, etc.

[0091] Then, the contribution of each transmission condition is compared with the reference contribution in turn. If the contribution of the i-th transmission condition is greater than the reference contribution, the i-th transmission condition is determined to have a larger contribution as described in the above embodiment, and the data volume of the i-th transmission condition is reduced; otherwise, if the contribution of the i-th transmission condition is less than the median contribution, the i-th transmission condition is determined to have a smaller contribution as described in the above embodiment, and the data volume of the i-th transmission condition is increased.

[0092] This application embodiment determines an intermediate comparison reference value for contribution. By simply comparing other contribution values ​​with this comparison reference value, the amount of data to increase or decrease for the corresponding transmission conditions can be determined based on the comparison results. The algorithm is simple and has a small computational load.

[0093] Optionally, determining the amount of training data under each transmission condition based on the contribution of each transmission condition to the optimization objective of the neural network includes: determining the weighting coefficient corresponding to each transmission condition based on the probability density of each transmission condition in practical applications; and determining the amount of training data under each transmission condition based on the contribution of each transmission condition to the optimization objective, combined with the weighting coefficient.

[0094] This can be understood as follows: when determining the proportion of data under different transmission conditions to the total data volume, weighting terms can be designed based on the probability density of different transmission conditions in practice, increasing the data volume of conditions with high probability density and decreasing the data volume of conditions with low probability density. For example, suppose the probability density of the k-th SNR is p. k The corresponding weighting term is f(p) k ), f(p k ) is about p k The increasing function. Consider the data size of the k-th SNR after the weighting term is updated as f(p). k )〃N k The probability density reflects the probability of transmission conditions occurring in different environments. These transmission conditions are not equally likely to occur in different environments; some conditions have a slightly higher probability of occurrence, while others have a slightly lower probability.

[0095] Optionally, the weighting coefficients are in a functionally increasing relationship with the probability density. That is, the relationship between the weighting terms and the probability density can be any increasing functional relationship, meaning that the weighting terms with higher probability densities should be larger, and the weighting terms with lower probability densities should be smaller.

[0096] The embodiments of this application design weighting terms based on the probability density of transmission conditions in reality, which can better adapt to the actual environment.

[0097] Optionally, the method further includes: sending the training dataset to a target device, the target device being used to train the neural network based on the training dataset.

[0098] This can be understood as follows: the training dataset obtained in this application embodiment can be used to train a neural network, and the training dataset acquisition method in this application embodiment can be applied in data transmission scenarios where the data acquisition end and the neural network training end are not on the same execution end. In this application embodiment, after completing data acquisition and calibration according to a determined proportion of the mixed dataset, a training dataset is constructed, and this training dataset is fed back to other devices that need to perform the neural network training process, i.e., the target device. The target device is a second device different from the current device; it can be a terminal or a network-side device, used to complete the training of the neural network model using the obtained training dataset.

[0099] This application embodiment enables data sharing and joint training among multiple devices by sending the acquired training dataset to a second device outside the current device, thereby effectively reducing the computational load of a single device and effectively improving computational efficiency.

[0100] Optionally, sending the training dataset to the target device includes: directly sending the training dataset to the target device, or sending the training dataset to the target device after performing a predetermined transformation, wherein the predetermined transformation includes at least one of specific quantization, specific compression, and neural network processing according to a pre-agreed or configured procedure.

[0101] This can be understood as meaning that when sending the training dataset to other devices, the sending method can be direct or indirect. Indirect sending refers to transforming the training data in the training dataset before feeding it back, for example, by using a specific quantization method, a specific compression method, or processing the training data to be sent according to a pre-agreed and configured neural network before sending.

[0102] To further illustrate the technical solutions of the embodiments of this application, examples will be given below, but the scope of protection claimed in this application is not limited.

[0103] Taking an application scenario where the target (or objective function, loss function) of the neural network optimization is the mean square error (MSE) or normalized mean square error (NMSE), and the transmission condition is the signal-to-noise ratio (SNR), as an example, consider that the data to be mixed has K different SNRs, and denote the total amount of data as N. all The k-th signal-to-noise ratio is denoted as SNR. kThe amount of data for the k-th signal-to-noise ratio is denoted as N. k .

[0104] First, sort the data for K signal-to-noise ratios:

[0105] Assume the data is mixed in equal proportions (i.e., N). k / N all =1 / K), let C be the contribution of the k-th SNR data to the indicator that needs to be minimized. k According to C k Sort the SNRs in ascending (or descending) order.

[0106] Then, determine the mixing ratio:

[0107] (1) For the k-th SNR, C k The larger N is, the more adjustments are made to make N larger. k The smaller the value of , the better. According to C... k The order from smallest to largest is such that the corresponding N k The values ​​of can be ordered from largest to smallest, and the decreasing rule can be any decreasing rule; or, according to C k The order from largest to smallest makes the corresponding N k The value of can be obtained from smallest to largest and the increment rule can be any increment rule.

[0108] (2) After adjusting the amount of data, confirm that the proportion of any SNR data to the total data is not less than the threshold γ.

[0109] Taking the optimized metrics MSE and NMSE as examples, generally speaking, low SNR data contributes more to MSE or NMSE, while high SNR data contributes less. Therefore, in the final determined mixing ratio, the proportion of low SNR data to the total data volume is the lowest, and as SNR increases, the corresponding data volume also increases.

[0110] Optionally, when determining the proportion of data under different transmission conditions to the total data volume, a weighting term can be designed based on the probability density of different transmission conditions in practice. Assume the probability density of the k-th SNR is p. k The corresponding weighting term is f(p) k ), f(p k ) is about p k The increasing function. Consider the data size of the k-th SNR after the weighting term is updated as f(p). k )〃N k .

[0111] Taking another example, where the optimization objective (or objective function, loss function) of the neural network is to maximize metrics such as signal-to-interference-plus-noise ratio (SINR), spectral efficiency, or throughput, and the transmission condition is signal-to-noise ratio (SNR), consider a scenario where the data to be mixed consists of K types of SNR, and the total amount of data is denoted as N. all The k-th signal-to-noise ratio is denoted as SNR. k The amount of data for the k-th signal-to-noise ratio is denoted as N. k .

[0112] First, sort the data for K signal-to-noise ratios:

[0113] Assume the data is mixed in equal proportions (i.e., N). k / N all =1 / K), let C be the contribution of the k-th SNR data to the aforementioned metric that needs to be maximized. k According to C k Sort the SNRs in ascending (or descending) order.

[0114] Then, determine the mixing ratio:

[0115] (1) For the k-th SNR, C k The larger N is, the more adjustments are made to make N larger. k The smaller the value of , the better. According to C... k The order from smallest to largest is such that the corresponding N k The values ​​of can be ordered from largest to smallest, and the decreasing rule can be any decreasing rule; or, according to C k The order from largest to smallest makes the corresponding N k The value of can be obtained from smallest to largest and the increment rule can be any increment rule.

[0116] (2) After adjusting the amount of data, confirm that the proportion of any SNR data to the total data is not less than the threshold γ.

[0117] Taking the optimized metrics such as SINR, spectral efficiency, or throughput as examples, generally speaking, low SNR data contributes less to SINR, spectral efficiency, and throughput, while high SNR data contributes more. Therefore, low SNR data accounts for the highest proportion of the total data volume, and as SNR decreases, the corresponding data volume also decreases.

[0118] Optionally, when determining the proportion of data under different transmission conditions to the total data volume, a weighting term can be designed based on the probability density of different transmission conditions in practice. Assume the probability density of the k-th SNR is p. k The corresponding weighting term is f(p)k ), f(p k ) is about p k The increasing function. Consider the data size of the k-th SNR after the weighting term is updated as f(p). k )〃N k .

[0119] It should be noted that the training dataset acquisition method provided in this application embodiment can be executed by a training dataset acquisition device, or by a control module within that device for executing the training dataset acquisition method. This application embodiment uses the execution of the training dataset acquisition method by a training dataset acquisition device as an example to illustrate the training dataset acquisition device provided in this application embodiment.

[0120] The structure of the training dataset acquisition device in this application embodiment is as follows: Figure 3 The diagram shown is a schematic representation of the training dataset acquisition device provided in this application embodiment. This device can be used to acquire the training dataset in the above-described training dataset acquisition method embodiments. The device includes: a first processing module 301 and a second processing module 302, wherein:

[0121] The first processing module 301 is used to determine the amount of training data under each transmission condition based on the contribution of each transmission condition to the neural network optimization objective.

[0122] The second processing module 302 is used to acquire training data under each transmission condition based on the amount of training data under each transmission condition, so as to form a training dataset for training the neural network.

[0123] The contribution of the transmission conditions to the optimization objective of the neural network represents the degree of influence of the transmission conditions on the value of the optimization objective of the neural network.

[0124] Optionally, the first processing module is configured to:

[0125] The contribution of each of the aforementioned transmission conditions is ranked.

[0126] Based on proportional mixing, at least one of the following operations is performed: reducing the amount of training data under transmission conditions corresponding to a larger contribution in the ranking and increasing the amount of training data under transmission conditions corresponding to a smaller contribution in the ranking.

[0127] Optionally, the type of transmission condition includes at least one of the following:

[0128] Signal-to-noise ratio (SNR) or signal-to-interference-plus-noise ratio (SINR);

[0129] Reference signal received power;

[0130] Signal strength;

[0131] Interference intensity;

[0132] Terminal movement speed;

[0133] Channel parameters;

[0134] The distance between the terminal and the base station;

[0135] Size of the community;

[0136] Carrier frequency;

[0137] Modulation order or modulation coding strategy;

[0138] Community type;

[0139] Station spacing;

[0140] Weather and environmental factors;

[0141] Antenna configuration information at the transmitting or receiving end;

[0142] Terminal capabilities or type;

[0143] Base station capabilities or types.

[0144] Optionally, the second processing module is used for:

[0145] Based on the amount of training data under each of the aforementioned transmission conditions, the data under each of the aforementioned transmission conditions is collected and calibrated to form a training dataset under each of the aforementioned transmission conditions.

[0146] or,

[0147] Collect a set amount of data under each of the aforementioned transmission conditions, and based on the amount of training data under each of the aforementioned transmission conditions, select a portion of the data from the set amount of data and label it, or supplement and label the set amount of data to form a training dataset under each of the aforementioned transmission conditions.

[0148] Optionally, when the first processing module performs at least one of the operations of reducing the amount of training data under the transmission conditions corresponding to larger contributions in the ranking and increasing the amount of training data under the transmission conditions corresponding to smaller contributions in the ranking, it is used to:

[0149] Perform at least one of the following operations: reducing the amount of training data under the transmission conditions corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission conditions corresponding to the smaller contribution in the ranking. The following rules include:

[0150] The larger the value of the larger contribution, the greater the decrease; the smaller the value of the smaller contribution, the greater the increase.

[0151] Optionally, if the sorting result is from small to large, the amount of training data under the transmission conditions decreases in the direction of the sorting; if the sorting result is from large to small, the amount of training data under the transmission conditions increases in the direction of the sorting.

[0152] Optionally, when the first processing module performs at least one of the operations of reducing the amount of training data under the transmission conditions corresponding to larger contributions in the ranking and increasing the amount of training data under the transmission conditions corresponding to smaller contributions in the ranking, it is used to:

[0153] Based on the sorting, a reference contribution is determined, and the contribution of the transmission condition is compared with the reference contribution.

[0154] Based on the comparison results, at least one of the following operations is performed, including:

[0155] If the contribution of the transmission condition is greater than the reference contribution, then the contribution of the transmission condition is determined to be the larger contribution, and the amount of training data under the transmission condition is reduced.

[0156] If the contribution of the transmission condition is not greater than the reference contribution, then the contribution of the transmission condition is determined to be the smaller contribution, and the amount of training data under the transmission condition is increased.

[0157] Optionally, the reference contribution is the median of the sort, or the contribution at a set position in the sort, or the average of all contributions in the sort, or the contribution in the sort that is closest to the average.

[0158] Optionally, the first processing module is further configured to:

[0159] Based on the probability density of each transmission condition in practical applications, the weighting coefficients corresponding to each transmission condition are determined.

[0160] Based on the contribution of each transmission condition to the optimization objective, and in conjunction with the weighting coefficients, the amount of training data under each transmission condition is determined.

[0161] Optionally, the weighting coefficients are in a functionally increasing relationship with the probability density.

[0162] Optionally, the device further includes:

[0163] A sending module is used to send the training dataset to a target device, which is used to train the neural network based on the training dataset.

[0164] Optionally, the sending module is configured to:

[0165] The training dataset can be sent directly to the target device, or the training dataset can be transformed according to a specified transformation and then sent to the target device. The specified transformation includes at least one of specific quantization, specific compression, and neural network processing according to a pre-agreed or configured procedure.

[0166] The training dataset acquisition device in this application embodiment can be a device, a device with an operating system, or an electronic device, or it can be a component, integrated circuit, or chip in a terminal or network-side device. This device or electronic device can be a mobile terminal or a non-mobile terminal, and can include, but is not limited to, the types of network-side devices 102 listed above. For example, a mobile terminal can include, but is not limited to, the types of terminals 101 listed above, and a non-mobile terminal can be a server, network-attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine, etc., and this application embodiment does not specifically limit the type.

[0167] The training dataset acquisition device provided in this application embodiment can achieve... Figure 2 The various processes implemented in the method embodiments achieve the same technical effect, and will not be described again here to avoid repetition.

[0168] This application also provides a wireless transmission method, which can be executed by a terminal and / or a network-side device. Specifically, the terminal can be... Figure 1 The terminal 101 shown in the diagram, specifically the network-side device, can be... Figure 1 The network-side device 102 is shown in the image. For example... Figure 4 The diagram shown is a flowchart illustrating a wireless transmission method provided in an embodiment of this application. The method includes:

[0169] Step 401: Based on the neural network model, perform wireless transmission calculations to realize the wireless transmission.

[0170] The neural network model is obtained by training a training dataset in advance, and the training dataset is obtained based on the training dataset acquisition method described in the above embodiments.

[0171] This application embodiment can be understood as follows: In advance, a training dataset can be obtained based on the above-described training dataset acquisition method embodiments (or the data proportion of each transmission condition can be obtained), and the neural network initially constructed is trained using this training dataset to obtain a neural network model. Then, this neural network model is applied to the wireless transmission operation process of this application embodiment, and the wireless transmission of this application embodiment is ultimately realized through computation.

[0172] The wireless transmission application environment described in this application can be any wireless transmission environment in which machine learning can replace the function of one or more modules in an existing wireless transmission network. That is, when training a neural network using machine learning in wireless transmission, a training dataset can be constructed using the training dataset acquisition method described above, and the neural network model can be trained using this training dataset and then used for wireless transmission. Examples of wireless transmission application environments include physical layer pilot design, channel estimation, signal detection, user pairing, HARQ, and positioning; higher layers resource allocation, handover, and mobility management; and network layer scheduling or slicing. This application does not limit the specific wireless transmission application scenario.

[0173] According to the embodiments of this application, data under different transmission conditions are selected in different proportions to construct a non-uniform mixed training dataset based on the contribution of data under different transmission conditions to the optimization objective (or objective function or loss function) of the neural network. Based on this non-uniform mixed training dataset, a common neural network is trained for wireless transmission under different actual transmission conditions, which enables the trained neural network to achieve high performance under each transmission condition.

[0174] Optionally, before performing wireless transmission calculations based on the neural network model, the wireless transmission method further includes:

[0175] Based on the training dataset, the neural network model is trained using any of the following training methods, wherein the following training methods include:

[0176] Centralized training on a single terminal;

[0177] Centralized training on a single network-side device;

[0178] Multiple terminals jointly conduct distributed training;

[0179] Multiple network-side devices jointly conduct distributed training;

[0180] A single network-side device collaborates with multiple terminals for distributed training.

[0181] Multiple network-side devices and multiple terminals jointly conduct distributed training;

[0182] Multiple network-side devices and a single terminal jointly conduct distributed training.

[0183] This can be understood as follows: before using a neural network model for wireless transmission computation, the neural network model must first be trained using a training dataset. Specifically, the training phase of the neural network can be performed offline, and the execution entity can be a network-side device, a terminal-side device, or a combination of a network-side device and a terminal-side device. Optionally, the network-side device can be an access network device, a core network device, or a data network device. That is, the network-side device in this application embodiment can include one or more of the following: network-side devices in the access network, core network devices, and data network devices (DN). The network-side device in the access network can be a base station, or a node on the RAN side responsible for AI training, or is not limited to... Figure 1 The types of access network equipment 1021 listed herein are examples. Core network equipment is not limited to... Figure 1 The core network equipment listed in section 1022 can be of various types, including data network equipment such as NWDAF, UDM, UDR, or UDSF.

[0184] When the execution entity is a network-side device, training can be centralized based on a single network-side device or distributed based on multiple network-side devices (such as federated learning). When the execution entity is a terminal-side device, training can be centralized based on a single terminal or distributed based on multiple terminals (such as federated learning). When the execution entity is a combination of network-side device and terminal-side device, it can be a single network-side device combined with multiple terminal devices, a single terminal device combined with multiple network-side devices, or multiple network-side devices combined with multiple terminal-side devices. This application does not specifically limit the execution entity of the training process.

[0185] Optionally, the wireless transmission method further includes: during the distributed training process, sharing the proportion of training data under each transmission condition among the entities performing the distributed training.

[0186] This can be understood as the proportion of training data with shared transmission conditions among the execution devices when using multiple network-side devices, multiple terminal devices, or network-side devices and terminal devices to jointly train a neural network model in the embodiments of this application.

[0187] The embodiments of this application, by sharing a certain percentage of training data among various execution entities, enable each execution entity to train a neural network model without sharing its own data. This solves the problems of insufficient computing or training capabilities of a single device, inability to share data between devices (involving privacy issues), or the very high cost of transmitting large amounts of data.

[0188] Optionally, when the distributed training is a joint distributed training of multiple network-side devices, any one of the multiple network-side devices calculates and determines the proportion of training data under each of the transmission conditions, and sends the proportion to the other network-side devices besides the aforementioned network-side device through a first set type interface signaling.

[0189] This can be understood as follows: when multiple network-side devices jointly train a distributed neural network model, all network-side devices share the same data proportion. Furthermore, one of these network-side devices can calculate and obtain the proportion of training data under various transmission conditions, and share this proportion with the other network-side devices through network-side interface signaling. This network-side interface signaling is a pre-defined first type. Optionally, the first type of interface signaling includes Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, or N22 interface signaling. Among them, base stations can use the Xn interface to share data proportions through Xn interface signaling, and core network devices can use the N1, N2, N3, N4, N5, N6, N7, N8, N9, N10, N11, N12, N13, N14, N15 or N22 interfaces between core network devices to share data proportions through corresponding interface signaling.

[0190] For example, a network-side device can calculate and determine the data percentage information, and then share the data percentage information with other network-side devices through Xn interface signaling (including but not limited to).

[0191] Optionally, when the distributed training is a joint distributed training of multiple terminals, any one of the multiple terminals calculates and determines the proportion of training data under each of the transmission conditions, and sends the proportion to the other terminals of the multiple terminals except for the aforementioned terminal through a second set type interface signaling.

[0192] This can be understood as follows: when multiple terminals jointly train a distributed neural network model, all terminals share the same data proportion. Furthermore, one of these terminals can calculate and obtain the proportion of training data under various transmission conditions, and share this proportion with the other terminals through terminal interface signaling. This terminal interface signaling is a pre-defined second configuration type. Optionally, the second configuration type interface signaling includes PC5 interface signaling or sidelink interface signaling.

[0193] For example, a terminal device can calculate and determine the data percentage information, and then share the data percentage information with other terminal devices through PC5 interface signaling (including but not limited to).

[0194] Optionally, when the distributed training is a joint distributed training of network-side devices and terminals, any one of the network-side devices or terminals calculates and determines the proportion of training data under each of the transmission conditions, and sends the proportion to other network-side devices or terminals other than the aforementioned network-side devices or terminals through a third set type signaling.

[0195] This can be understood as follows: when the execution entity is a network-side device and a terminal-side device in combination, all devices share the same data proportion. Furthermore, one terminal (or network-side device) in the combined network-side device-terminal-side device group can calculate and obtain the proportion of training data under various transmission conditions, and share this proportion with other devices in the network-side device-terminal-side device group through a configuration type interface signaling. This configuration type interface signaling is a pre-defined third configuration type interface signaling.

[0196] Optionally, the third type of signaling includes RRC, PDCCH layer 1 signaling, PDSCH, MAC CE, SIB, Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, N22 interface signaling, PUCCH layer 1 signaling, PUSCH, PRACH MSG1, PRACH MSG3, PRACH MSG A, PC5 interface signaling, or sidelink interface signaling.

[0197] In other words, multiple network-side devices and terminal-side devices participating in the training can share data proportion information through signaling such as RRC, PDCCH layer 1 signaling, PDSCH, MAC CE, SIB, Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, N22 interface signaling, PUCCH layer 1 signaling, PUSCH, PRACH MSG1, PRACH MSG3, PRACH MSG A, PC5 interface signaling, or sidelink interface signaling (including but not limited to).

[0198] Optionally, it further includes: acquiring real-time data under the transmission conditions, and adjusting the trained neural network model online based on the real-time data.

[0199] This application can be understood as follows: Based on the large amount of offline data pre-trained in the above embodiments to achieve convergence, the parameters of the pre-trained neural network are then fine-tuned using real-time collected online data from the transmission environment to adapt the neural network to the actual environment. Fine-tuning can be considered a training process that uses the parameters of the pre-trained neural network as initialization. During the fine-tuning stage, the parameters of some layers can be frozen; generally, layers closer to the input are frozen, while layers closer to the output are activated. This ensures that the network can still converge. The less data available during the fine-tuning stage, the more layers should be frozen, and only a small number of layers close to the output should be fine-tuned.

[0200] Optionally, when fine-tuning the trained neural network in a real wireless environment, the data mixing ratio under different transmission conditions during the model training phase can be maintained. Alternatively, when fine-tuning the trained neural network in a real wireless environment, data generated in the actual wireless environment can be used directly for fine-tuning without controlling the data ratio.

[0201] Based on the offline training of the neural network model, the embodiments of this application perform online fine-tuning of the trained neural network model, which can make the neural network more adaptable to the actual environment.

[0202] Optionally, acquiring real-time data under the transmission conditions and adjusting the trained neural network model online based on the real-time data includes: acquiring real-time data under each transmission condition based on the proportion of training data under each transmission condition; if the proportion of real-time data under any transmission condition is higher than the proportion of training data under any transmission condition, then during the process of adjusting the trained neural network model online, data exceeding the proportion of training data under any transmission condition in the real-time data under any transmission condition will not be input into the trained neural network model.

[0203] This can be understood as follows: according to the above embodiments, when the trained neural network model is taken to a real wireless environment for fine-tuning, the data mixing ratio under different transmission conditions during the model training phase can be retained. That is, based on the proportion of training data under each transmission condition during the training phase, the proportion or amount of real-time data under each transmission condition during the online fine-tuning phase is determined, and the corresponding amount of real-time data under each transmission condition is obtained accordingly. When using the mixing ratio from the training phase, if the proportion of data from a certain transmission condition in the actual wireless environment exceeds the proportion of that transmission condition during the model training phase, the excess data will not be input into the network for fine-tuning.

[0204] In this embodiment of the application, when using the data proportion from the training phase, data exceeding the proportion is not input into the neural network model for fine-tuning, thus avoiding the unbalanced impact of data exceeding the proportion.

[0205] Optionally, acquiring real-time data under each of the transmission conditions includes: online acquisition of data under each of the transmission conditions from at least one of the network-side devices and terminals, as real-time data under each of the transmission conditions; and online adjustment of the trained neural network model includes: based on the data under each of the transmission conditions from at least one of the network-side devices and terminals, using the network-side device or the terminal to adjust the trained neural network model online.

[0206] Similar to the training phase of a neural network, the fine-tuning phase of the neural network model in this application embodiment, if using the same data proportions as the training phase, can be performed at the input end of the neural network, or it can be a network-side device and / or a terminal-side device. In other words, when the execution subject is a network-side device, real-time data of each transmission condition under the network-side device can be obtained online; when the execution subject is a terminal device, real-time data of each transmission condition under the terminal side can be obtained online; when the execution subject includes both a network-side device and a terminal, real-time data of each transmission condition needs to be obtained for both execution subjects.

[0207] Subsequently, during actual online fine-tuning, the network-side devices or terminals perform online fine-tuning of the neural network model based on their own corresponding real-time data, and update the network parameters.

[0208] Optionally, if the network-side device or the terminal does not acquire the proportion of training data under each of the transmission conditions during the training phase, before acquiring real-time data under each of the transmission conditions based on the proportion of training data under each of the transmission conditions, the wireless transmission method further includes:

[0209] The network-side device obtains the proportion of training data under each transmission condition from the network-side device during the training phase through any one of the following interface signaling methods: Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, and N22 interface signaling.

[0210] Alternatively, the network-side device may obtain the proportion of training data under each transmission condition from the terminal during the training phase through any of the following signaling methods: PUCCH layer 1 signaling, PUSCH, PRACH MSG1, PRACH MSG3, and PRACH MSG A.

[0211] Alternatively, the terminal may obtain the proportion of training data under each transmission condition from the terminal during the training phase via PC5 interface signaling or sidelink interface signaling.

[0212] Alternatively, the terminal may obtain the proportion of training data under each transmission condition from the network-side device during the training phase via any of the following signaling methods: RRC, PDCCH layer 1 signaling, PUSCH, MAC CE, and SIB.

[0213] This can be understood as follows: during the online fine-tuning phase of the neural network model, if the execution subject did not obtain data proportion information during the training phase, then depending on the different execution subject types and the target execution subject type of the training phase, it is necessary to first obtain data proportion information through Xn interface signaling, or Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling and N22 interface signaling, or PC5 interface signaling, or sidelink interface signaling, or RRC, PDCCH layer 1 signaling, MAC CE or SIB interface signaling (including but not limited to) before fine-tuning.

[0214] In this embodiment, since the executing entity does not obtain data proportion information during the training phase, it obtains data proportion information from other executing entities through a set type interface signaling, which enables the sharing of data proportion information during the online fine-tuning phase and ensures the generalization ability of the neural network.

[0215] To further illustrate the technical solutions of the embodiments of this application, examples will be given below, but the scope of protection claimed in this application is not limited.

[0216] Figure 5 This is a schematic diagram illustrating the process of constructing a neural network model in the wireless transmission method provided according to an embodiment of this application. Figure 5 The diagram illustrates the model building process involved in the wireless transmission method proposed in this application embodiment, which can be divided into an offline training stage (shown as ① in the figure) and a fine-tuning stage in the actual transmission network (shown as ② in the figure). Before offline training, a training dataset can be constructed and obtained.

[0217] When constructing the training dataset, we can first mix the data from all transmission conditions proportionally to determine the amount or proportion of data for each transmission condition. Then, we sort all transmission conditions in the mixture according to their contribution to the neural network's optimization objective. Next, while ensuring sufficient data for all transmission conditions, we increase the amount of data from transmission conditions with lower contributions and decrease the amount of data from transmission conditions with higher contributions to determine the proportion of data for each transmission condition. Finally, we construct the mixed training dataset based on this proportion.

[0218] When determining the amount of data under different transmission conditions, "ensuring sufficient data for all transmission conditions" can mean setting a threshold, whereby the proportion of data under any transmission condition to the total data amount must not be lower than this threshold.

[0219] When "increasing the amount of data from transmission conditions with lower contribution and decreasing the amount of data from transmission conditions with higher contribution," the lower the contribution, the more data is added; conversely, the higher the contribution, the more data is reduced. If the contribution ranking is from smallest to largest, the proportion of data from each transmission condition relative to the total data can decrease in any way, such as linearly, arithmetically, geometrically, exponentially, or power-lawfully. Conversely, if the contribution ranking is from largest to smallest, the proportion of data from each transmission condition relative to the total data can increase in any way, such as linearly, arithmetically, geometrically, exponentially, or power-lawfully.

[0220] One feasible way to determine the proportion of data for each transmission condition is as follows: Figure 6 The diagram shown illustrates the process of determining the proportion of training data in the training dataset acquisition method provided in this application embodiment, mainly including:

[0221] After sorting, find the median of the contributions, and then compare the contribution of each transmission condition with this median.

[0222] If the contribution of the i-th transmission condition is greater than the median contribution, then, on the premise of ensuring that the data volume of all transmission conditions is sufficient, the data volume of the i-th transmission condition is reduced, and the reduction is proportional to the difference between the contribution of the i-th transmission condition and the median contribution.

[0223] If the contribution of the i-th transmission condition is less than the median contribution, then, on the premise that the data volume of all transmission conditions is sufficient, the data volume of the i-th transmission condition is increased, and the increase is proportional to the difference between the median contribution and the contribution of the i-th transmission condition.

[0224] After obtaining the training dataset, the neural network model is trained offline in a loop until it converges, resulting in a trained neural network model.

[0225] Then, data from the actual wireless network is collected in real time to fine-tune the parameters of the pre-trained neural network model, enabling the model to adapt to the real environment. Online fine-tuning can be considered as a retraining process using the parameters of the pre-trained neural network as initialization.

[0226] For example, taking the demodulation reference signal (DMRS) channel estimation in wireless transmission as a specific application scenario, consider a system containing N_RBs, each RB containing N_SC subcarriers and N_Sym symbols, meaning the system contains a total of N_RE = N_RB * N_SC * N_Sym time-frequency resources. N_SC_DMRS are placed in the frequency domain and N_Sym_DMRS in the time domain of each RB for channel estimation. Therefore, the DMRS occupy a total of N_RE_DMRS = N_RB * N_SC_DMRS * N_Sym_DMRS time-frequency resources. The receiver, based on the DMRS signals at the locations of the received N_RE_DMRS time-frequency resources, recovers the channel estimation for all N_RE time-frequency resources.

[0227] The structure of the neural network that implements the above process is as follows: Figure 7 The diagram shows a schematic of the neural network used for DMRS channel estimation in the wireless transmission method provided according to an embodiment of this application. The input information of the neural network is N_RE_DMRS symbols after adding noise to the DMRS channel, and the output information of the neural network is N_RE symbols, corresponding to the channel estimation results on all N_RE time-frequency resources.

[0228] When training the neural network, the training data consists of labeled DMRS information pairs. One DMRS signal sample (containing N_RE_DMRS symbols) corresponds to one label (this label is the channel truth value corresponding to the current DMRS signal sample, totaling N_RE_DMRS symbols). During training, a large number of labeled DMRS information pairs are used to adjust the parameters of the neural network, minimizing the normalized mean square error (NMSE) between the neural network output based on the DMRS signal samples and its label.

[0229] Consider that the training data contains K pairs of DMRS information obtained under different SNR values, and the k-th signal-to-noise ratio is denoted as SNR. k The data size for the k-th signal-to-noise ratio is N. k There are a total of There are training data sets. Assume the neural network is trained using federated learning, with one network-side device and multiple terminal-side devices participating in the federated learning process.

[0230] First, the proportion of data for each SNR is determined on the network-side device when constructing the hybrid training dataset.

[0231] (1) Sort the data of K signal-to-noise ratios according to the contribution of transmission conditions to the optimization objective:

[0232] Assume the data is mixed in equal proportions (i.e., N). k / N all =1 / K), let C be the contribution of the k-th SNR data to the above NMSE. k According to C k SNR is sorted in ascending (or descending) order. In this embodiment, data with low SNR contributes more to NMSE, while data with high SNR contribute less. The median of all contributions in the sorting is denoted as .

[0233] (2) Determine the mixing ratio:

[0234] 1) For the k-th SNR: If Then reduce the amount of data for the k-th SNR by the same amount. Proportional; if Then increase the amount of data for the k-th SNR, with the increase being the same as... Proportional;

[0235] 2) Assume the amount of data for the k-th signal-to-noise ratio after adjustment is N. k ′, a total of There are 10 training data points. It is confirmed that the proportion of data with any SNR after adjustment is not less than the threshold γ, that is, for all k, after adjusting the amount of data, N′... k / N′ all ≥γ.

[0236] Then, the network-side device sends the data percentage information of each SNR to all terminals participating in the joint training through interface signaling such as RRC, PDCCH layer 1 signaling, MAC CE or SIB, to perform federated learning of the neural network, that is, offline training.

[0237] After offline training of the neural network is completed, the trained neural network is fine-tuned online in a real wireless network. Since the data proportion information has already been shared with all terminals during the offline training phase, this data proportion can be used in the online fine-tuning phase. When the data proportion of a certain SNR from the actual wireless environment exceeds the proportion of that SNR during the model training phase, the excess data is not input into the neural network for fine-tuning.

[0238] The embodiments of this application can improve the generalization ability of trained neural network models in changing wireless environments.

[0239] It should be noted that the wireless transmission method provided in this application embodiment can be executed by a wireless transmission device, or by a control module within the wireless transmission device for executing the wireless transmission method. This application embodiment uses the execution of the wireless transmission method by a wireless transmission device as an example to illustrate the wireless transmission device provided in this application embodiment.

[0240] The structure of the wireless transmission device in this application embodiment is as follows: Figure 8 The diagram shown is a structural schematic of a wireless transmission device provided in an embodiment of this application. This device can be used to implement wireless transmission in the above-described wireless transmission method embodiments. The device includes:

[0241] The third processing module 801 is used to perform wireless transmission calculations based on a neural network model to realize the wireless transmission.

[0242] The neural network model is obtained by training a training dataset in advance, and the training dataset is obtained based on the training dataset acquisition method described in the above embodiments.

[0243] Optionally, the wireless transmission device further includes:

[0244] The training module is used to train the neural network model based on the training dataset using any of the following training methods:

[0245] Centralized training on a single terminal;

[0246] Centralized training on a single network-side device;

[0247] Multiple terminals jointly conduct distributed training;

[0248] Multiple network-side devices jointly conduct distributed training;

[0249] A single network-side device collaborates with multiple terminals for distributed training.

[0250] Multiple network-side devices and multiple terminals jointly conduct distributed training;

[0251] Multiple network-side devices and a single terminal jointly conduct distributed training.

[0252] Optionally, the network-side device may be an access network device, a core network device, or a data network device.

[0253] Optionally, the wireless transmission device further includes:

[0254] The fourth processing module is used to share the proportion of training data under each transmission condition among the entities performing the distributed training during the distributed training process.

[0255] Optionally, when the distributed training is a joint distributed training of multiple network-side devices, the fourth processing module is used to calculate and determine the proportion of training data under each of the multiple network-side devices under each transmission condition by any one of the multiple network-side devices, and send the proportion to the other network-side devices other than the one network-side device through a first set type interface signaling.

[0256] Optionally, the first set type interface signaling includes Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, or N22 interface signaling.

[0257] Optionally, when the distributed training is a joint distributed training of multiple terminals, the fourth processing module is used to calculate and determine the proportion of training data under each of the multiple terminals under each of the transmission conditions, and send the proportion to the other terminals of the multiple terminals except for the aforementioned terminal through a second set type interface signaling.

[0258] Optionally, the second configuration type interface signaling includes PC5 interface signaling or sidelink interface signaling.

[0259] Optionally, when the distributed training is a joint distributed training between network-side devices and terminals, the fourth processing module is used to calculate and determine the proportion of training data under each of the transmission conditions by any network-side device or terminal among the network-side devices and terminals, and send the proportion to other network-side devices or terminals other than the network-side device or terminal among the network-side devices and terminals through a third set type signaling.

[0260] Optionally, the third type of signaling includes RRC, PDCCH layer 1 signaling, PDSCH, MAC CE, SIB, Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, N22 interface signaling, PUCCH layer 1 signaling, PUSCH, PRACH MSG1, PRACH MSG3, PRACH MSG A, PC5 interface signaling, or sidelink interface signaling.

[0261] Optionally, the wireless transmission device further includes:

[0262] The fine-tuning module is used to acquire real-time data under the transmission conditions and, based on the real-time data, adjust the trained neural network model online.

[0263] Optionally, the fine-tuning module is used for:

[0264] Based on the proportion of training data under each of the aforementioned transmission conditions, real-time data under each of the aforementioned transmission conditions is obtained;

[0265] If the proportion of real-time data under any transmission condition is higher than the proportion of training data under any transmission condition, then during the online adjustment of the trained neural network model, data exceeding the proportion of training data under any transmission condition in the real-time data under any transmission condition will not be input into the trained neural network model.

[0266] Optionally, the fine-tuning module, when used for acquiring real-time data under each of the aforementioned transmission conditions, is used for:

[0267] Data is collected online from at least one of the network-side devices and terminals under each of the aforementioned transmission conditions, and is used as real-time data under each of the aforementioned transmission conditions;

[0268] The fine-tuning module, when used for the online adjustment of the trained neural network model, is used for:

[0269] Based on data under each of the transmission conditions of at least one of the network-side devices and terminals, the trained neural network model is adjusted online using the network-side device or the terminal.

[0270] Optionally, the wireless transmission device further includes:

[0271] The communication module is configured to, when the network-side device or the terminal fails to acquire the proportion of training data under each of the aforementioned transmission conditions during the training phase,

[0272] The network-side device obtains the proportion of training data under each transmission condition from the network-side device during the training phase through any one of the following interface signaling methods: Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, and N22 interface signaling.

[0273] Alternatively, the network-side device may obtain the proportion of training data under each transmission condition from the terminal during the training phase through any of the following signaling methods: PUCCH layer 1 signaling, PUSCH, PRACH MSG1, PRACH MSG3, and PRACH MSG A.

[0274] Alternatively, the terminal may obtain the proportion of training data under each transmission condition from the terminal during the training phase via PC5 interface signaling or sidelink interface signaling.

[0275] Alternatively, the terminal may obtain the proportion of training data under each transmission condition from the network-side device during the training phase via any of the following signaling methods: RRC, PDCCH layer 1 signaling, PUSCH, MAC CE, and SIB.

[0276] The wireless transmission device in this application embodiment can be a device, a device or electronic device with an operating system, or a component, integrated circuit, or chip in a terminal or network-side device. The device or electronic device can be a mobile terminal or a non-mobile terminal, and can include, but is not limited to, the type of network-side device 102 listed above. For example, a mobile terminal can include, but is not limited to, the type of terminal 101 listed above, and a non-mobile terminal can be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine, etc., and this application embodiment does not impose specific limitations.

[0277] The wireless transmission device provided in this application embodiment can achieve... Figures 4 to 7 The various processes implemented in the wireless transmission method embodiment achieve the same technical effect, and will not be described again here to avoid repetition.

[0278] like Figure 9 As shown, this application embodiment also provides a communication device 900, including a processor 901, a memory 902, and a program or instructions stored in the memory 902 and executable on the processor 901. For example, when the communication device 900 is a terminal or a network-side device, the program or instructions, when executed by the processor 901, can implement the various processes of the above-described training dataset acquisition method embodiment and achieve the same technical effect, or implement the various processes of the above-described wireless transmission method embodiment and achieve the same technical effect. To avoid repetition, further details are omitted here.

[0279] This application embodiment also provides a communication device, which can be a terminal or a network-side device. The communication device includes a processor and a communication interface. The processor is used to determine the amount of training data under each transmission condition based on the contribution of each transmission condition to the optimization objective of the neural network; and to acquire the training data under each transmission condition based on the amount of training data under each transmission condition, to form a training dataset for training the neural network. The contribution of the transmission condition to the optimization objective of the neural network represents the degree of influence of the transmission condition on the value of the optimization objective of the neural network. It should be noted that this communication device embodiment corresponds to the above-described training dataset acquisition method embodiment. All implementation processes and methods of the above method embodiments can be applied to this communication device embodiment and can achieve the same technical effect.

[0280] This application also provides a communication device, which can be a terminal or a network-side device. The communication device includes a processor and a communication interface. The processor is used to perform wireless transmission calculations based on a neural network model to realize the wireless transmission. The neural network model is obtained by pre-training using a training dataset, which is obtained using the training dataset acquisition method described in the above embodiments. It should be noted that this communication device embodiment corresponds to the above-described wireless transmission method embodiments. All implementation processes and methods of the above method embodiments can be applied to this communication device embodiment and achieve the same technical effects.

[0281] Specifically, Figure 10This is a schematic diagram of the hardware structure of a terminal according to an embodiment of this application. The terminal 1000 includes, but is not limited to, at least some of the following components: radio frequency unit 1001, network module 1002, audio output unit 1003, input unit 1004, sensor 1005, display unit 1006, user input unit 1007, interface unit 1008, memory 1009, and processor 1010.

[0282] Those skilled in the art will understand that the terminal 1000 may also include a power supply (such as a battery) for supplying power to various components. The power supply may be logically connected to the processor 1010 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system. Figure 10 The terminal structure shown does not constitute a limitation on the terminal. The terminal may include more or fewer components than shown, or combine certain components, or have different component arrangements, which will not be elaborated here.

[0283] It should be understood that, in this embodiment, the input unit 1004 may include a graphics processing unit (GPU) 10041 and a microphone 10042. The GPU 10041 processes image data of still images or videos obtained by an image capture device (such as a camera) in video capture mode or image capture mode. The display unit 1006 may include a display panel 10061, which may be configured in the form of a liquid crystal display, an organic light-emitting diode, or the like. The user input unit 1007 includes a touch panel 10071 and other input devices 10072. The touch panel 10071 is also called a touch screen. The touch panel 10071 may include a touch detection device and a touch controller. Other input devices 10072 may include, but are not limited to, physical keyboards, function keys (such as volume control buttons, power buttons, etc.), trackballs, mice, and joysticks, which will not be described in detail here.

[0284] In this embodiment, the radio frequency unit 1001 receives downlink data from the network-side device and processes it for the processor 1010; additionally, it sends uplink data to the network-side device. Typically, the radio frequency unit 1001 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier, a duplexer, etc.

[0285] The memory 1009 can be used to store software programs or instructions and various data. The memory 1009 may primarily include a program or instruction storage area and a data storage area. The program or instruction storage area may store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, the memory 1009 may include high-speed random access memory and non-volatile memory, wherein the non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. For example, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0286] Processor 1010 may include one or more processing units; optionally, processor 1010 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications or instructions, and the modem processor mainly handles wireless communication, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into processor 1010.

[0287] The processor 1010 is configured to determine the amount of training data under each transmission condition based on the contribution of each transmission condition to the optimization objective of the neural network; and to acquire the training data under each transmission condition based on the amount of training data under each transmission condition, so as to form a training dataset for training the neural network; wherein the contribution of the transmission condition to the optimization objective of the neural network represents the degree of influence of the transmission condition on the value of the optimization objective of the neural network.

[0288] In this application embodiment, when constructing a training dataset in an artificial intelligence-based communication system, data under various transmission conditions are selected in different proportions according to the contribution of data under different transmission conditions to the optimization objective (or objective function or loss function) of the neural network, and a hybrid training dataset is constructed, which can effectively improve the generalization ability of the neural network.

[0289] Optionally, the processor 1010 is further configured to sort the contribution of each of the transmission conditions; and, on the basis of proportional mixing, perform at least one of the operations of reducing the amount of training data under the transmission conditions with larger contributions in the sort and increasing the amount of training data under the transmission conditions with smaller contributions in the sort.

[0290] This application embodiment sorts the contribution of each transmission condition and adjusts the data volume of the corresponding transmission condition according to the sorting. This enables a clearer and more accurate determination of the adjustment strategy for the data volume of the corresponding transmission condition (including whether to increase or decrease the data volume, the magnitude of the increase or decrease, etc.), thereby making the efficiency higher and the results more accurate.

[0291] Optionally, the processor 1010 is further configured to collect and calibrate data under each of the transmission conditions based on the amount of training data under each of the transmission conditions, thereby forming a training dataset under each of the transmission conditions; or, to collect a set number of data under each of the transmission conditions, and based on the amount of training data under each of the transmission conditions, to select and calibrate a portion of the data from the set number of data, or to supplement and calibrate the set number of data, thereby forming a training dataset under each of the transmission conditions.

[0292] Optionally, the processor 1010 is further configured to perform at least one of the operations of reducing the amount of training data under the transmission conditions corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission conditions corresponding to the smaller contribution in the ranking, according to the following rules, wherein the following rules include: the larger the value of the larger contribution, the greater the reduction; the smaller the value of the smaller contribution, the greater the increase.

[0293] The embodiments of this application increase or decrease the amount of training data for corresponding transmission conditions proportionally according to the magnitude of the contribution value. This allows the amount of training data for transmission conditions to gradually decrease as the contribution of each transmission condition gradually increases, thereby better balancing the impact of each transmission condition on the final neural network and improving the generalization ability of the neural network.

[0294] Optionally, the processor 1010 is further configured to determine a reference contribution based on the sorting, and compare the contribution of the transmission condition with the reference contribution. If the contribution of the transmission condition is greater than the reference contribution, the contribution of the transmission condition is determined to be the larger contribution, and the amount of training data under the transmission condition is reduced. Otherwise, the contribution of the transmission condition is determined to be the smaller contribution, and the amount of training data under the transmission condition is increased.

[0295] This application embodiment determines an intermediate comparison reference value for contribution. By simply comparing other contribution values ​​with this comparison reference value, the amount of data to increase or decrease for the corresponding transmission conditions can be determined based on the comparison results. The algorithm is simple and has a small computational load.

[0296] Optionally, the processor 1010 is further configured to determine the weighting coefficients corresponding to each of the transmission conditions based on the probability density of each of the transmission conditions in practical applications; and to determine the amount of training data under each of the transmission conditions based on the contribution of each of the transmission conditions to the optimization objective, combined with the weighting coefficients.

[0297] The embodiments of this application design weighting terms based on the probability density of transmission conditions in reality, which can better adapt to the actual environment.

[0298] Optionally, the radio frequency unit 1001 is used to send the training dataset to a target device, the target device being used to train the neural network based on the training dataset.

[0299] This application embodiment enables data sharing and joint training among multiple devices by sending the acquired training dataset to a second device outside the current device, thereby effectively reducing the computational load of a single device and effectively improving computational efficiency.

[0300] Optionally, the radio frequency unit 1001 is used to directly send the training dataset to the target device, or to send the training dataset after a specified transformation to the target device;

[0301] The processor 1010 is also configured to perform a specified transformation on the training dataset, the specified transformation including at least one of specific quantization, specific compression, and neural network processing according to a pre-agreed or configured procedure.

[0302] Optionally, the processor 1010 is further configured to perform wireless transmission calculations based on a neural network model to realize the wireless transmission; wherein the neural network model is obtained by training a training dataset in advance, and the training dataset is obtained based on the training dataset acquisition method described in the above embodiments of the training dataset acquisition method.

[0303] According to the embodiments of this application, data under different transmission conditions are selected in different proportions to construct a non-uniform mixed training dataset based on the contribution of data under different transmission conditions to the optimization objective (or objective function or loss function) of the neural network. Based on this non-uniform mixed training dataset, a common neural network is trained for wireless transmission under different actual transmission conditions, which enables the trained neural network to achieve high performance under each transmission condition.

[0304] Optionally, the processor 1010 is further configured to train the neural network model based on the training dataset using any of the following training methods:

[0305] Centralized training on a single terminal;

[0306] Centralized training on a single network-side device;

[0307] Multiple terminals jointly conduct distributed training;

[0308] Multiple network-side devices jointly conduct distributed training;

[0309] A single network-side device collaborates with multiple terminals for distributed training.

[0310] Multiple network-side devices and multiple terminals jointly conduct distributed training;

[0311] Multiple network-side devices and a single terminal jointly conduct distributed training.

[0312] Optionally, the radio frequency unit 1001 is also used to share the proportion of training data under each transmission condition among the subjects performing the distributed training during the distributed training process.

[0313] The embodiments of this application, by sharing a certain percentage of training data among various execution entities, enable each execution entity to train a neural network model without sharing its own data. This solves the problems of insufficient computing or training capabilities of a single device, inability to share data between devices (involving privacy issues), or the very high cost of transmitting large amounts of data.

[0314] Optionally, the processor 1010 is further configured to, in the case where the distributed training is a joint distributed training of multiple network-side devices, calculate and determine the proportion of training data under each of the multiple network-side devices.

[0315] The radio frequency unit 1001 is also used to send the percentage to other network-side devices among the plurality of network-side devices, excluding any one of the network-side devices, through a first set type interface signaling.

[0316] Optionally, the processor 1010 is further configured to, in the case where the distributed training is a joint distributed training of multiple terminals, calculate and determine the proportion of training data under each of the multiple terminals.

[0317] The radio frequency unit 1001 is also used to send the percentage to other terminals among the plurality of terminals other than any of the terminals via a second set type interface signaling.

[0318] Optionally, the processor 1010 is further configured to, in the case where the distributed training is a joint distributed training between network-side devices and terminals, calculate and determine the proportion of training data under each of the transmission conditions by any network-side device or any terminal among the network-side devices and terminals.

[0319] The radio frequency unit 1001 is also used to transmit the percentage to other network-side devices or terminals other than any of the network-side devices or terminals via a third setting type signaling.

[0320] Optionally, the processor 1010 is also configured to acquire real-time data under the transmission conditions and, based on the real-time data, adjust the trained neural network model online.

[0321] Based on the offline training of the neural network model, the embodiments of this application perform online fine-tuning of the trained neural network model, which can make the neural network more adaptable to the actual environment.

[0322] Optionally, the processor 1010 is further configured to acquire real-time data under each transmission condition based on the proportion of training data under each transmission condition; and if the proportion of real-time data under any transmission condition is higher than the proportion of training data under any transmission condition, then during the online adjustment of the trained neural network model, data exceeding the proportion of training data under any transmission condition in the real-time data under any transmission condition is not input into the trained neural network model.

[0323] In this embodiment of the application, when using the data proportion from the training phase, data exceeding the proportion is not input into the neural network model for fine-tuning, thus avoiding the unbalanced impact of data exceeding the proportion.

[0324] Optionally, the input unit 1004 is used to collect data from at least one of the network-side devices and terminals under each of the transmission conditions online, as real-time data under each of the transmission conditions;

[0325] The processor 1010 is further configured to adjust the trained neural network model online using the network-side device or the terminal based on data under each of the transmission conditions of at least one of the network-side device and the terminal.

[0326] Optionally, when the communication device is a network-side device, the radio frequency unit 1001 is further configured to obtain the proportion of training data under each transmission condition from the network-side device in the training phase through any one of the following network-side interface signals: Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, and N22 interface signaling; or to obtain the proportion of training data under each transmission condition from the terminal in the training phase through any one of the following interface signals: RRC, PDCCH layer 1 signaling, MAC CE, and SIB.

[0327] When the communication device is a terminal, the radio frequency unit 1001 is also used to obtain the proportion of training data under each transmission condition from the terminal in the training phase through PC5 interface signaling or sidelink interface signaling, or to obtain the proportion of training data under each transmission condition from the network-side device in the training phase through any of RRC, PDCCH layer 1 signaling, MAC CE and SIB signaling.

[0328] In this embodiment, since the executing entity does not obtain data proportion information during the training phase, it obtains data proportion information from other executing entities through a set type interface signaling, which enables the sharing of data proportion information during the online fine-tuning phase and ensures the generalization ability of the neural network.

[0329] Specifically, Figure 11 A schematic diagram of the hardware structure of an access network device according to an embodiment of this application is shown. Figure 11 As shown, the access network device 1100 includes: an antenna 1101, a radio frequency (RF) device 1102, and a baseband device 1103. The antenna 1101 is connected to the RF device 1102. In the uplink direction, the RF device 1102 receives information through the antenna 1101 and transmits the received information to the baseband device 1103 for processing. In the downlink direction, the baseband device 1103 processes the information to be transmitted and transmits it to the RF device 1102. The RF device 1102 processes the received information and transmits it through the antenna 1101.

[0330] The frequency band processing device can be located in the baseband device 1103. The method executed by the network-side device in the above embodiments can be implemented in the baseband device 1103, which includes a processor 1104 and a memory 1105.

[0331] The baseband device 1103 may include, for example, at least one baseband board on which multiple chips are disposed, such as... Figure 11As shown, one of the chips, for example, is a processor 1104, which is connected to a memory 1105 to call the program in the memory 1105 and execute the network-side device operations shown in the above method embodiments.

[0332] The baseband device 1103 may also include a network interface 1106 for exchanging information with the radio frequency device 1102, such as a common public radio interface (CPRI).

[0333] Specifically, the access network device of this embodiment further includes: instructions or programs stored in memory 1105 and executable on processor 1104, wherein processor 1104 calls the instructions or programs in memory 1105 to execute. Figure 3 or Figure 8 The methods executed by each module shown achieve the same technical effect, and to avoid repetition, they will not be described in detail here.

[0334] Specifically, Figure 12 A schematic diagram of the hardware structure of a core network device according to an embodiment of this application is shown. Figure 12 As shown, the core network device 1200 includes: a processor 1201, a transceiver 1202, a memory 1203, a user interface 1204, and a bus interface, wherein:

[0335] In this embodiment, the core network device 1200 further includes: a computer program stored in the memory 1203 and executable on the processor 1201, wherein the computer program, when executed by the processor 1201, implements, as follows: Figure 3 or Figure 8 The methods executed by each module shown achieve the same technical effect, and to avoid repetition, they will not be described in detail here.

[0336] exist Figure 12 In this embodiment, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 1201 and memory represented by memory 1203 together. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be further described in this embodiment. The bus interface provides an interface. The transceiver 1202 can be multiple elements, including a transmitter and a receiver, providing a unit for communicating with various other devices over a transmission medium. For different user equipment, the user interface 1204 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0337] The processor 1201 is responsible for managing the bus architecture and general processing, and the memory 1203 can store the data used by the processor 1201 when performing operations.

[0338] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described training dataset acquisition method embodiment or the various processes of the above-described wireless transmission method embodiment, and achieve the same technical effect. To avoid repetition, these will not be described again here.

[0339] The processor mentioned above is the processor in the terminal or network-side device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0340] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described training dataset acquisition method embodiment, or to implement the various processes of the above-described wireless transmission method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0341] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0342] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0343] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0344] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for obtaining a training dataset, characterized in that, include: Based on the contribution of each transmission condition to the optimization objective of the neural network, the amount of training data under each transmission condition is determined. Based on the amount of training data under each of the aforementioned transmission conditions, training data under each of the aforementioned transmission conditions is obtained to form a training dataset for training the neural network. Wherein, the contribution of the transmission conditions to the optimization objective of the neural network represents the degree of influence of the transmission conditions on the value of the optimization objective of the neural network; The determination of the amount of training data under each transmission condition based on the contribution of each transmission condition to the neural network optimization objective includes: The contribution of each of the aforementioned transmission conditions is ranked. Based on proportional mixing, at least one of the following operations is performed: reducing the amount of training data under transmission conditions corresponding to a larger contribution in the ranking and increasing the amount of training data under transmission conditions corresponding to a smaller contribution in the ranking.

2. The training dataset acquisition method according to claim 1, characterized in that, The types of transmission conditions include at least one of the following: Signal-to-noise ratio (SNR) or signal-to-interference-plus-noise ratio (SINR); Reference signal received power; Signal strength; Interference intensity; Terminal movement speed; Channel parameters; The distance between the terminal and the base station; Size of the community; Carrier frequency; Modulation order or modulation coding strategy; Community type; Station spacing; Weather and environmental factors; Antenna configuration information at the transmitting or receiving end; Terminal capabilities or type; Base station capabilities or types.

3. The training dataset acquisition method according to claim 1, characterized in that, The step of acquiring training data under each of the aforementioned transmission conditions to form a training dataset for training the neural network includes: Based on the amount of training data under each of the aforementioned transmission conditions, the data under each of the aforementioned transmission conditions is collected and calibrated to form a training dataset under each of the aforementioned transmission conditions. or, Collect a set amount of data under each of the aforementioned transmission conditions, and based on the amount of training data under each of the aforementioned transmission conditions, select a portion of the data from the set amount of data and label it, or supplement and label the set amount of data to form a training dataset under each of the aforementioned transmission conditions.

4. The training dataset acquisition method according to claim 1, characterized in that, At least one of the operations of reducing the amount of training data under the transmission conditions corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission conditions corresponding to the smaller contribution in the ranking includes: Perform at least one of the following operations: reducing the amount of training data under the transmission conditions corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission conditions corresponding to the smaller contribution in the ranking. The following rules include: The larger the value of the larger contribution, the greater the decrease; the smaller the value of the smaller contribution, the greater the increase.

5. The training dataset acquisition method according to claim 4, characterized in that, When the sorting result is from small to large, the amount of training data under the transmission conditions decreases in the direction of the sorting; when the sorting result is from large to small, the amount of training data under the transmission conditions increases in the direction of the sorting.

6. The training dataset acquisition method according to claim 1, characterized in that, At least one of the operations of reducing the amount of training data under the transmission conditions corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission conditions corresponding to the smaller contribution in the ranking includes: Based on the sorting, a reference contribution is determined, and the contribution of the transmission condition is compared with the reference contribution. Based on the comparison results, at least one of the following operations is performed, including: If the contribution of the transmission condition is greater than the reference contribution, then the contribution of the transmission condition is determined to be the larger contribution, and the amount of training data under the transmission condition is reduced. If the contribution of the transmission condition is not greater than the reference contribution, then the contribution of the transmission condition is determined to be the smaller contribution, and the amount of training data under the transmission condition is increased.

7. The training dataset acquisition method according to claim 6, characterized in that, The reference contribution is the median of the sort, or the contribution at a set position in the sort, or the average of all contributions in the sort, or the contribution in the sort that is closest to the average.

8. The method for obtaining a training dataset according to any one of claims 1-7, characterized in that, The determination of the amount of training data under each transmission condition based on the contribution of each transmission condition to the neural network optimization objective includes: Based on the probability density of each transmission condition in practical applications, the weighting coefficients corresponding to each transmission condition are determined. Based on the contribution of each transmission condition to the optimization objective, and in conjunction with the weighting coefficients, the amount of training data under each transmission condition is determined.

9. The training dataset acquisition method according to claim 8, characterized in that, The weighting coefficients are in a functionally increasing relationship with the probability density.

10. The method for obtaining a training dataset according to any one of claims 1-7 and 9, characterized in that, The method further includes: The training dataset is sent to a target device, which is used to train the neural network based on the training dataset.

11. The training dataset acquisition method according to claim 10, characterized in that, Sending the training dataset to the target device includes: The training dataset can be sent directly to the target device, or the training dataset can be transformed according to a specified transformation and then sent to the target device. The specified transformation includes at least one of specific quantization, specific compression, and neural network processing according to a pre-agreed or configured procedure.

12. A training dataset acquisition device, characterized in that, include: The first processing module is used to determine the amount of training data under each transmission condition based on the contribution of each transmission condition to the optimization objective of the neural network. The second processing module is used to acquire training data under each transmission condition based on the amount of training data under each transmission condition, so as to form a training dataset for training the neural network. Wherein, the contribution of the transmission conditions to the optimization objective of the neural network represents the degree of influence of the transmission conditions on the value of the optimization objective of the neural network; The first processing module is used for: The contribution of each of the aforementioned transmission conditions is ranked. Based on proportional mixing, at least one of the following operations is performed: reducing the amount of training data under transmission conditions corresponding to a larger contribution in the ranking and increasing the amount of training data under transmission conditions corresponding to a smaller contribution in the ranking.

13. The training dataset acquisition device according to claim 12, characterized in that, The types of transmission conditions include at least one of the following: Signal-to-noise ratio (SNR) or signal-to-interference-plus-noise ratio (SINR); Reference signal received power; Signal strength; Interference intensity; Terminal movement speed; Channel parameters; The distance between the terminal and the base station; Size of the community; Carrier frequency; Modulation order or modulation coding strategy; Community type; Station spacing; Weather and environmental factors; Antenna configuration information at the transmitting or receiving end; Terminal capabilities or type; Base station capabilities or types.

14. The training dataset acquisition device according to claim 12, characterized in that, The second processing module is used for: Based on the amount of training data under each of the aforementioned transmission conditions, the data under each of the aforementioned transmission conditions is collected and calibrated to form a training dataset under each of the aforementioned transmission conditions. or, Collect a set amount of data under each of the aforementioned transmission conditions, and based on the amount of training data under each of the aforementioned transmission conditions, select a portion of the data from the set amount of data and label it, or supplement and label the set amount of data to form a training dataset under each of the aforementioned transmission conditions.

15. The training dataset acquisition device according to claim 12, characterized in that, When the first processing module performs at least one of the operations of reducing the amount of training data under the transmission conditions corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission conditions corresponding to the smaller contribution in the ranking, it is configured to: Perform at least one of the following operations: reducing the amount of training data under the transmission conditions corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission conditions corresponding to the smaller contribution in the ranking. The following rules include: The larger the value of the larger contribution, the greater the decrease; the smaller the value of the smaller contribution, the greater the increase.

16. The training dataset acquisition device according to claim 15, characterized in that, When the sorting result is from small to large, the amount of training data under the transmission conditions decreases in the direction of the sorting; when the sorting result is from large to small, the amount of training data under the transmission conditions increases in the direction of the sorting.

17. The training dataset acquisition device according to claim 12, characterized in that, When the first processing module performs at least one of the operations of reducing the amount of training data under the transmission conditions corresponding to the larger contribution in the ranking and increasing the amount of training data under the transmission conditions corresponding to the smaller contribution in the ranking, it is configured to: Based on the sorting, a reference contribution is determined, and the contribution of the transmission condition is compared with the reference contribution. Based on the comparison results, at least one of the following operations is performed, including: If the contribution of the transmission condition is greater than the reference contribution, then the contribution of the transmission condition is determined to be the larger contribution, and the amount of training data under the transmission condition is reduced. If the contribution of the transmission condition is not greater than the reference contribution, then the contribution of the transmission condition is determined to be the smaller contribution, and the amount of training data under the transmission condition is increased.

18. The training dataset acquisition device according to claim 17, characterized in that, The reference contribution is the median of the sort, or the contribution at a set position in the sort, or the average of all contributions in the sort, or the contribution in the sort that is closest to the average.

19. The training dataset acquisition device according to any one of claims 12-18, characterized in that, The first processing module is further configured to: Based on the probability density of each transmission condition in practical applications, the weighting coefficients corresponding to each transmission condition are determined. Based on the contribution of each transmission condition to the optimization objective, and in conjunction with the weighting coefficients, the amount of training data under each transmission condition is determined.

20. The training dataset acquisition device according to claim 19, characterized in that, The weighting coefficients are in a functionally increasing relationship with the probability density.

21. The training dataset acquisition device according to any one of claims 12-18, 20, characterized in that, The device further includes: A sending module is used to send the training dataset to a target device, which is used to train the neural network based on the training dataset.

22. The training dataset acquisition device according to claim 21, characterized in that, The sending module is used for: The training dataset can be sent directly to the target device, or the training dataset can be transformed according to a specified transformation and then sent to the target device. The specified transformation includes at least one of specific quantization, specific compression, and neural network processing according to a pre-agreed or configured procedure.

23. A wireless transmission method, characterized in that, include: Based on a neural network model, wireless transmission calculations are performed to achieve the wireless transmission. The neural network model is obtained by training a training dataset in advance, and the training dataset is obtained based on the training dataset acquisition method described in any one of claims 1-11.

24. The wireless transmission method according to claim 23, characterized in that, Before performing wireless transmission calculations based on the neural network model, the wireless transmission method further includes: Based on the training dataset, the neural network model is trained using any of the following training methods, wherein the following training methods include: Centralized training on a single terminal; Centralized training on a single network-side device; Multiple terminals jointly conduct distributed training; Multiple network-side devices jointly conduct distributed training; A single network-side device collaborates with multiple terminals for distributed training. Multiple network-side devices and multiple terminals jointly conduct distributed training; Multiple network-side devices and a single terminal jointly conduct distributed training.

25. The wireless transmission method according to claim 24, characterized in that, The network-side equipment can be access network equipment, core network equipment, or data network equipment.

26. The wireless transmission method according to claim 24 or 25, characterized in that, The wireless transmission method further includes: During the distributed training process, the proportion of training data under each transmission condition is shared among the entities performing the distributed training.

27. The wireless transmission method according to claim 26, characterized in that, In the case where the distributed training is a joint distributed training by multiple network-side devices, any one of the multiple network-side devices calculates and determines the proportion of training data under each of the transmission conditions, and sends the proportion to the other network-side devices besides the first network-side device through a first set type interface signaling.

28. The wireless transmission method according to claim 27, characterized in that, The first set type of interface signaling includes Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling or N22 interface signaling.

29. The wireless transmission method according to claim 26, characterized in that, In the case where the distributed training is a joint distributed training of multiple terminals, any one of the multiple terminals calculates and determines the proportion of training data under each of the transmission conditions, and sends the proportion to the other terminals of the multiple terminals except for the aforementioned terminal through a second set type interface signaling.

30. The wireless transmission method according to claim 29, characterized in that, The second type of interface signaling includes PC5 interface signaling or sidelink interface signaling.

31. The wireless transmission method according to claim 26, characterized in that, In the case where the distributed training is a joint distributed training between network-side devices and terminals, any one of the network-side devices or terminals calculates and determines the proportion of training data under each of the transmission conditions, and sends the proportion to other network-side devices or terminals other than the aforementioned network-side devices or terminals through a third set type signaling.

32. The wireless transmission method according to claim 31, characterized in that, The third type of signaling includes RRC, PDCCH layer 1 signaling, PDSCH, MAC CE, SIB, Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, N22 interface signaling, PUCCH layer 1 signaling, PUSCH, PRACH MSG1, PRACH MSG3, PRACH MSG A, PC5 interface signaling, or sidelink interface signaling.

33. The wireless transmission method according to any one of claims 24, 25, and 27-32, characterized in that, Also includes: Acquire real-time data under the transmission conditions, and adjust the trained neural network model online based on the real-time data.

34. The wireless transmission method according to claim 33, characterized in that, The step of acquiring real-time data under the transmission conditions and adjusting the trained neural network model online based on the real-time data includes: Based on the proportion of training data under each of the aforementioned transmission conditions, real-time data under each of the aforementioned transmission conditions is obtained; If the proportion of real-time data under any transmission condition is higher than the proportion of training data under any transmission condition, then during the online adjustment of the trained neural network model, data exceeding the proportion of training data under any transmission condition in the real-time data under any transmission condition will not be input into the trained neural network model.

35. The wireless transmission method according to claim 34, characterized in that, The acquisition of real-time data under each of the aforementioned transmission conditions includes: Data is collected online from at least one of the network-side devices and terminals under each of the aforementioned transmission conditions, and is used as real-time data under each of the aforementioned transmission conditions; The online adjustment of the trained neural network model includes: Based on data under each of the transmission conditions of at least one of the network-side devices and terminals, the trained neural network model is adjusted online using the network-side device or the terminal.

36. The wireless transmission method according to claim 35, characterized in that, If the network-side device or the terminal does not acquire the proportion of training data under each of the transmission conditions during the training phase, before acquiring real-time data under each of the transmission conditions based on the proportion of training data under each of the transmission conditions, the wireless transmission method further includes: The network-side device obtains the proportion of training data under each transmission condition from the network-side device during the training phase through any one of the following interface signaling methods: Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, and N22 interface signaling. Alternatively, the network-side device may obtain the proportion of training data under each transmission condition from the terminal during the training phase through any of the following signaling methods: PUCCH layer 1 signaling, PUSCH, PRACH MSG1, PRACH MSG3, and PRACH MSG A. Alternatively, the terminal may obtain the proportion of training data under each transmission condition from the terminal during the training phase via PC5 interface signaling or sidelink interface signaling. Alternatively, the terminal may obtain the proportion of training data under each transmission condition from the network-side device during the training phase via any of the following signaling methods: RRC, PDCCH layer 1 signaling, PUSCH, MAC CE, and SIB.

37. A wireless transmission device, characterized in that, include: The third processing module is used to perform wireless transmission calculations based on a neural network model to realize the wireless transmission. The neural network model is obtained by training a training dataset in advance, and the training dataset is obtained based on the training dataset acquisition method described in any one of claims 1-11.

38. The wireless transmission device according to claim 37, characterized in that, The wireless transmission device also includes: The training module is used to train the neural network model based on the training dataset using any of the following training methods: Centralized training on a single terminal; Centralized training on a single network-side device; Multiple terminals jointly conduct distributed training; Multiple network-side devices jointly conduct distributed training; A single network-side device collaborates with multiple terminals for distributed training. Multiple network-side devices and multiple terminals jointly conduct distributed training; Multiple network-side devices and a single terminal jointly conduct distributed training.

39. The wireless transmission device according to claim 38, characterized in that, The network-side equipment can be access network equipment, core network equipment, or data network equipment.

40. The wireless transmission device according to claim 38 or 39, characterized in that, The wireless transmission device also includes: The fourth processing module is used to share the proportion of training data under each transmission condition among the entities performing the distributed training during the distributed training process.

41. The wireless transmission device according to claim 40, characterized in that, In the case where the distributed training is a joint distributed training of multiple network-side devices, the fourth processing module is used to calculate and determine the proportion of training data under each of the multiple network-side devices, and send the proportion to the other network-side devices other than the first network-side device through a first set type interface signaling.

42. The wireless transmission device according to claim 41, characterized in that, The first set type of interface signaling includes Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling or N22 interface signaling.

43. The wireless transmission device according to claim 40, characterized in that, In the case where the distributed training is a joint distributed training of multiple terminals, the fourth processing module is used to calculate and determine the proportion of training data under each of the multiple terminals under each transmission condition by any one of the multiple terminals, and send the proportion to the other terminals of the multiple terminals except for the one terminal mentioned above through the second set type interface signaling.

44. The wireless transmission device according to claim 43, characterized in that, The second type of interface signaling includes PC5 interface signaling or sidelink interface signaling.

45. The wireless transmission device according to claim 40, characterized in that, In the case where the distributed training is a joint distributed training between network-side devices and terminals, the fourth processing module is used to calculate and determine the proportion of training data under each transmission condition by any network-side device or terminal among the network-side devices and terminals, and send the proportion to other network-side devices or terminals other than the network-side device or terminal through a third set type signaling.

46. ​​The wireless transmission device according to claim 45, characterized in that, The third type of signaling includes RRC, PDCCH layer 1 signaling, PDSCH, MAC CE, SIB, Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, N22 interface signaling, PUCCH layer 1 signaling, PUSCH, PRACH MSG1, PRACH MSG3, PRACH MSG A, PC5 interface signaling, or sidelink interface signaling.

47. The wireless transmission device according to any one of claims 38, 39, and 41-46, characterized in that, The wireless transmission device also includes: The fine-tuning module is used to acquire real-time data under the transmission conditions and, based on the real-time data, adjust the trained neural network model online.

48. The wireless transmission device according to claim 47, characterized in that, The fine-tuning module is used for: Based on the proportion of training data under each of the aforementioned transmission conditions, real-time data under each of the aforementioned transmission conditions is obtained; If the proportion of real-time data under any transmission condition is higher than the proportion of training data under any transmission condition, then during the online adjustment of the trained neural network model, data exceeding the proportion of training data under any transmission condition in the real-time data under any transmission condition will not be input into the trained neural network model.

49. The wireless transmission device according to claim 48, characterized in that, The fine-tuning module, when used to acquire real-time data under each of the aforementioned transmission conditions, is used for: Data is collected online from at least one of the network-side devices and terminals under each of the aforementioned transmission conditions, and is used as real-time data under each of the aforementioned transmission conditions; The fine-tuning module, when used for the online adjustment of the trained neural network model, is used for: Based on data under each of the transmission conditions of at least one of the network-side devices and terminals, the trained neural network model is adjusted online using the network-side device or the terminal.

50. The wireless transmission device according to claim 49, characterized in that, The wireless transmission device also includes: The communication module is configured to, when the network-side device or the terminal fails to acquire the proportion of training data under each of the aforementioned transmission conditions during the training phase, The network-side device obtains the proportion of training data under each transmission condition from the network-side device during the training phase through any one of the following interface signaling methods: Xn interface signaling, N1 interface signaling, N2 interface signaling, N3 interface signaling, N4 interface signaling, N5 interface signaling, N6 interface signaling, N7 interface signaling, N8 interface signaling, N9 interface signaling, N10 interface signaling, N11 interface signaling, N12 interface signaling, N13 interface signaling, N14 interface signaling, N15 interface signaling, and N22 interface signaling. Alternatively, the network-side device may obtain the proportion of training data under each transmission condition from the terminal during the training phase through any of the following signaling methods: PUCCH layer 1 signaling, PUSCH, PRACH MSG1, PRACH MSG3, and PRACH MSG A. Alternatively, the terminal may obtain the proportion of training data under each transmission condition from the terminal during the training phase via PC5 interface signaling or sidelink interface signaling. Alternatively, the terminal may obtain the proportion of training data under each transmission condition from the network-side device during the training phase via any of the following signaling methods: RRC, PDCCH layer 1 signaling, PUSCH, MAC CE, and SIB.

51. A communication device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein when the program or instructions are executed by the processor, they implement the steps of the training dataset acquisition method as described in any one of claims 1 to 11, or the steps of the wireless transmission method as described in any one of claims 23 to 36.

52. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the training dataset acquisition method as described in any one of claims 1-11, or the steps of the wireless transmission method as described in any one of claims 23-36.

Images