Wireless communication resource configuration method and device and related product

Abstract

The invention provides an ultra-far coverage method and device applied to wireless communication and related products, wireless parameters and environmental parameters of a current period are obtained, the wireless parameters at least comprise signal round-trip propagation time delay, and the environmental parameters at least comprise one of terrain, weather and base station load; inputting the wireless parameters, the environmental parameters and the target coverage distance into a pre-trained first prediction model, and outputting a guard interval GP prediction value; wherein the first prediction model is obtained by training a deep learning model based on a historical wireless parameter data set, a historical environment parameter data set and a corresponding historical coverage distance label; the target coverage distance is determined based on the wireless parameters and the environmental parameters; and configuring the GP of the current wireless communication period according to the GP predicted value.

Description

Technical Field

[0001] This application relates to the field of wireless communication, and more specifically, to wireless communication resource allocation methods, apparatus, and related products. Background Technology

[0002] In 5G NR (New Ratio) Time Division Duplexing (TDD) communication technology, the frame structure includes uplink subframes, downlink subframes, and special subframes. Among these, the special subframes contain 14 Orthogonal Frequency Division Multiplexing (OFDM) symbols, including the Downlink Pilot Time Slot (DwPTS), Guard Period (GP), and Uplink Pilot Time Slot (UpPTS). The GP is primarily used for protection during the switch from downlink to uplink signals, providing a time buffer for RF transceiver switching and uplink signal propagation for long-distance users.

[0003] According to the 3rd Generation Partnership Project (3GPP) protocol specification, the length of the GP (GP Path) is configured in units of Orthogonal Frequency Division Multiplexing (OFDM) symbols, and there are only a limited number of fixed options (e.g., with a subcarrier spacing of 30kHz, it can be configured to a symbol length of 1, 2, 3, or 4). This configuration method is essentially static and step-wise. The maximum coverage distance L is related to the GP length T. GP The relationship is usually positive, meaning that the improvement of coverage distance is achieved by increasing the number of GP symbols.

[0004] This static configuration method has revealed significant drawbacks in real-world networks, especially in ultra-long-range coverage scenarios (such as vast sea areas, deserts, and grasslands): to ensure normal access for the farthest edge users, the network must configure the GP (GP Symbol) according to the maximum possible propagation delay. For example, if the farthest user is 14.5 kilometers away, the system must configure a 4-symbol GP (corresponding to approximately 19.45 kilometers of coverage capability), instead of the theoretically more precise 3.5 symbols. This leads to over-configuration of the GP for a large number of non-edge users, resulting in a waste of spectrum resources (reduced DwPTS and UpPTS symbol count) and low spectrum efficiency. Clearly, the existing resource configuration method does not meet the actual coverage requirements, and there is room for further improvement in spectrum resource utilization. Summary of the Invention

[0005] This application provides a wireless communication resource allocation method, apparatus, and related products to solve the problem that existing resource allocation methods do not match the actual coverage requirements and that spectrum resources are underutilized.

[0006] Firstly, a wireless communication resource allocation method is provided, applied to the network side, including: Obtain the wireless parameters and environmental parameters for the current period. The wireless parameters include at least the signal round-trip propagation delay, and the environmental parameters include at least one of the following: terrain, weather, and base station load. The wireless parameters, environmental parameters, and target coverage distance are input into a pre-trained first prediction model, which outputs a predicted guard interval (GP) value. The first prediction model is trained on a deep learning model based on historical wireless parameter datasets, historical environmental parameter datasets, and corresponding historical coverage distance labels. The target coverage distance is determined based on the wireless parameters and environmental parameters. Configure the GP for the current wireless communication cycle based on the GP prediction value.

[0007] Based on the wireless communication resource allocation method provided in the first aspect, a GP prediction model is trained by combining wireless parameter datasets and environmental parameter datasets. This enables the model to learn and extract the inherent rules for determining the optimal GP length from massive labeled datasets. This implementation transforms the existing "one-size-fits-all" GP setting method into an adaptive mode that precisely matches the actual channel environment, thereby achieving a dynamic optimal balance between coverage expansion and spectral efficiency improvement while ensuring basic communication quality.

[0008] In conjunction with the first aspect, in certain implementations of the first aspect, configuring the GP for the current wireless communication cycle based on the GP prediction value includes: Obtain the historical GP value of the previous communication cycle and the preset GP adjustment time interval; If the difference or rate of change between the predicted GP value and the historical GP value exceeds a preset adjustment threshold, and the time since the last GP adjustment reaches the preset GP adjustment time interval, the GP of the current wireless communication cycle is configured according to the predicted GP value. If the difference or rate of change between the predicted GP value and the historical GP value does not exceed a preset adjustment threshold, and / or the time since the last GP adjustment has not reached the preset GP adjustment time interval, the GP of the current wireless communication cycle is configured according to the historical GP value.

[0009] In conjunction with the first aspect, in some implementations of the first aspect, the preset adjustment threshold includes at least one of a relative change threshold and an absolute change threshold; wherein the relative change threshold is 10% of the historical GP value, and the absolute change threshold is 20 microseconds.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, the dynamic adjustment range of the GP length corresponding to the GP prediction value output by the first prediction model is configured to be from 350 microseconds to 750 microseconds, based on the target sea area coverage distance of 50 kilometers to 100 kilometers.

[0011] In conjunction with the first aspect, in some implementations of the first aspect, after configuring the GP of the current wireless communication cycle according to the GP prediction value, the method further includes: Obtain service parameters, which include at least one of the following: uplink / downlink traffic ratio, current number of active users, ratio of various services, average channel quality, and GP of the current wireless communication cycle; The service parameters are input into the trained second prediction model to obtain the predicted uplink and downlink time slot ratio; wherein, the second prediction model is trained on the reinforcement learning algorithm based on the service parameter dataset, and the uplink and downlink time slot ratio is the ratio of the number of symbols in the uplink to the number of symbols in the downlink. Configure the uplink and downlink time slot ratio for the current wireless communication cycle based on the predicted uplink and downlink time slot ratio.

[0012] In conjunction with the first aspect, in certain implementations of the first aspect, after configuring the uplink / downlink slot allocation for the current wireless communication cycle according to the predicted uplink / downlink slot allocation value, the following is included: The spectrum information of available communication frequency bands is obtained, and the spectrum information is input into a third prediction model to obtain interference signal prediction results. The interference signal prediction results include at least the interference category, interference intensity, interference frequency band, and interference trend. Based on the interference signal prediction results and channel quality, the available frequency band is dynamically divided into multiple logical slices; Based on the interference signal prediction results and the wireless channel parameters of each logical slice, the target logical slice is selected through the fourth prediction model. Configure the spectrum resources for the current communication cycle based on the selection result of the fourth prediction model.

[0013] In conjunction with the first aspect, in some implementations of the first aspect, the plurality of logical slices includes at least a first slice and a second slice, wherein: The first slice operates in a frequency band where the interference intensity is lower than a first reference threshold and the channel quality is higher than a first high-quality threshold; The second slice operates in a frequency band where the interference intensity is lower than the second reference threshold and the channel quality is higher than the second superior threshold; Wherein, the first benchmark threshold is less than the second benchmark threshold, and the first quality threshold is greater than the second quality threshold.

[0014] In conjunction with the first aspect, in some implementations of the first aspect, the plurality of logical slices further include a third slice and / or a fourth slice, wherein, The third slice is allocated to a frequency band with stable characteristics and corresponding suppression algorithms for persistent interference. When a service is scheduled to the third slice, the corresponding interference suppression algorithm is activated synchronously. The fourth slice is configured to operate in an anti-interference communication mode, which includes at least one of using extremely narrow bandwidth, lowest order modulation, and strongest error correction coding.

[0015] In conjunction with the first aspect, in some implementations of the first aspect, after configuring the spectrum resources for the current communication cycle according to the selection result of the fourth prediction model, the method further includes: Obtain wireless resource scheduling parameters, which include at least one of the following: real-time load of available frequency bands, terminal signal quality of available frequency bands, service demand type, terminal moving speed, and terminal carrier aggregation capability; The radio resource scheduling parameters are input into the trained fifth prediction model to obtain a resource scheduling decision. The resource scheduling decision is used to instruct at least one of the following operations: allocating an initial frequency band, triggering a frequency band handover, configuring carrier aggregation (CA), and triggering supplementary uplink (SUL) handover. The fifth prediction model is obtained by training a deep reinforcement learning algorithm based on the radio resource scheduling parameter dataset. Wireless resources are scheduled according to the resource scheduling decision.

[0016] In conjunction with the first aspect, in some implementations of the first aspect, the decision conditions for configuring carrier aggregation (CA) in the fifth prediction model include: the signal-to-interference-plus-noise ratio (SINR) of the terminal on the secondary carrier is higher than 3dB, and / or the channel quality indicator (CQI) is higher than 6, and / or the terminal's service bandwidth requirement is higher than 15Mbps.

[0017] In conjunction with the first aspect, in some implementations of the first aspect, the decision conditions for the fifth prediction model to trigger the supplementary uplink (SUL) handover include: detecting that the terminal is conducting a high-volume uplink service of a preset type, and that its uplink signal quality in the current service frequency band is lower than a preset threshold.

[0018] In conjunction with the first aspect, in some implementations of the first aspect, after scheduling radio resources according to the resource scheduling prediction result, the method further includes: Obtain the prediction outputs of the first to the fifth prediction models; Determine whether there is a conflict between the prediction outputs of the first prediction model to the fifth prediction model; In the event of a conflict, an avoidance operation is performed according to a preset global arbitration rule, which includes one of the following: prioritizing link stability, prioritizing load balancing, and prioritizing maximizing throughput.

[0019] In conjunction with the first aspect, in some implementations of the first aspect, after scheduling radio resources according to the resource scheduling prediction result, the method further includes: Obtain key quality indicators of the network system, wherein the key quality indicators include one of total throughput, average latency, and disconnection rate; Input the key quality indicators into the first optimization model to obtain hyperparameter optimization suggestions for one or more of the first to fifth prediction models. The hyperparameters include at least one of the following: learning rate, discount factor, and optimizer type. Based on the hyperparameter optimization recommendations, set the hyperparameters of one or more prediction models from the first to the fifth prediction models.

[0020] Secondly, this application also provides a 5G ultra-long-range coverage system for marine scenarios, including: The parameter acquisition module is used to acquire wireless parameters, environmental parameters, service parameters and spectrum information, wherein the environmental parameters include at least parameters characterizing the propagation characteristics in the sea area; The intelligent time slot scheduling module is connected to the parameter acquisition module and is used to dynamically configure the protection interval (GP) and / or the time division duplex (TDD) time slot ratio based on the wireless parameters and the environmental parameters. A cognitive radio anti-interference module, connected to the parameter acquisition module, is used to perform spectrum sensing and intelligent frequency selection scheduling based on the spectrum information; A multi-band intelligent coordination module, connected to the parameter acquisition module, is used for multi-band load balancing, carrier aggregation (CA), and supplementary uplink scheduling (SUL). The collaborative optimization center is communicatively connected to the intelligent time slot scheduling module, the cognitive radio anti-interference module, and the multi-band intelligent collaborative module, respectively. It is used to aggregate the scheduling decision information of each module, perform conflict arbitration, and perform closed-loop adaptive optimization of the parameters of each module.

[0021] Thirdly, this application also provides a wireless communication resource allocation device, characterized in that it includes: The acquisition module acquires the wireless parameters and environmental parameters for the current period. The wireless parameters include at least the signal round-trip propagation delay, and the environmental parameters include at least one of the following: terrain, weather, and base station load. The prediction module is used to input the wireless parameters, environmental parameters, and target coverage distance into a pre-trained first prediction model and output a guard interval (GP) prediction value; wherein, the first prediction model is obtained by training a deep learning model based on historical wireless parameter datasets, historical environmental parameter datasets, and corresponding historical coverage distance labels; the target coverage distance is determined based on the wireless parameters and environmental parameters; The configuration module is used to configure the GP of the current wireless communication cycle based on the GP prediction value.

[0022] Fourthly, this application also provides a wireless communication resource configuration device, comprising: Memory, used to store instruction sets; A processor for invoking and executing the instruction set to implement the steps of any of the methods in the first aspect.

[0023] Fifthly, this application also provides a readable computer storage medium having a computer program stored thereon, wherein... When the computer program is executed by the processor, it implements the steps of any one of the methods in the first aspect.

[0024] In a sixth aspect, this application also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of any of the methods in the first aspect.

[0025] The wireless communication resource allocation method provided in this application combines wireless parameter datasets and environmental parameter datasets to train a GP prediction model. This enables the model to learn and extract the inherent rules for determining the optimal GP length from massive labeled datasets. This implementation transforms the existing "one-size-fits-all" GP allocation method into an adaptive mode that precisely matches the actual channel environment, thereby achieving a dynamic optimal balance between coverage expansion and spectral efficiency improvement while ensuring basic communication quality. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the data frame structure when the TDD uplink / downlink conversion period is 5ms per cycle; Figure 2 This is a diagram illustrating the different coverage distances of ultra-long-range coverage sites; Figure 3 This is a flowchart illustrating a wireless communication resource allocation method provided in this application; Figure 4This is a schematic diagram of the architecture of a 5G ultra-long-range coverage system for marine scenarios provided in this application; Figure 5 This is a schematic diagram of the structure of a wireless communication resource allocation device provided in this application; Figure 6 This is a schematic diagram of the structure of a wireless communication resource configuration device provided in this application. Detailed Implementation

[0027] The technical solutions in this application will now be described with reference to the accompanying drawings.

[0028] Unless the context otherwise requires, throughout the specification and claims, the term "comprise" and its other forms, such as the third-person singular "comprises" and the present participle "comprising," are interpreted as open-ended and encompassing, meaning "including, but not limited to." In the description of the specification, terms such as "one embodiment," "some embodiments," "exemplary embodiments," "example," "specific example," or "some examples," etc., are intended to indicate that a particular feature, structure, material, or characteristic associated with that embodiment or example is included in at least one embodiment or example of this disclosure. The illustrative representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics mentioned may be included in any suitable manner in any one or more embodiments or examples.

[0029] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of this disclosure, unless otherwise stated, "a plurality of" means two or more.

[0030] In 5G NR (New Ratio) communication technology, Time Division Duplexing (TDD) technology achieves uplink and downlink communication through time-division multiplexing, offering relatively flexible spectrum utilization and adapting to 5G application requirements. Specifically, for example... Figure 1As shown, with a 5ms single-cycle uplink / downlink transition period, the typical time slot configuration for a wireless data frame is DDDDDDDSUU (D represents downlink time slot, S represents special time slot, and U represents uplink time slot). Each time slot contains 14 Orthogonal Frequency Division Multiplexing (OFDM) symbols. One possible configuration for the special time slot S is DDDDDGGGGUUUU (D represents downlink symbol, G represents guard period GP, and U represents uplink symbol). The GP is primarily used for protection during downlink-to-uplink handover, providing transition time for stable RF transceiver switching.

[0031] In wireless communication scenarios requiring ultra-long-range coverage, such as in sea areas, deserts, and grasslands, the distance between remote users and base stations is relatively far. Typically, a larger GP (GP length) is configured to achieve coverage for these remote users, ensuring that the uplink does not interfere with the normal reception of downlink signals. According to the 3rd Generation Partnership Project (3GPP) protocol specification, a GP can be configured with four symbol lengths. Different numbers of symbols occupied by the GP correspond to different maximum coverage distances, as shown in Table 1. Table 1

[0032] in, This indicates the handover time required for the UE to switch from receiving to transmitting. Indicates the duration of the GP symbol. This indicates the maximum propagation time over the air interface that the corresponding GP configuration can support. Understandably, the larger the GP configuration, the farther the supported wireless coverage distance.

[0033] Furthermore, according to the 3GPP protocol specification, GP configuration is not directly proportional to user distance, but rather expanded by configuring the number of symbols, a step-wise configuration method. While this achieves long-distance coverage, it also results in the waste of spectrum resources. For example... Figure 2 As shown, for a distant user at a distance of 14.50 km, with a subcarrier spacing (SCS) of 30 kHz, this distance exceeds the coverage range of a GP (GP symbol) length of 3 symbols (14.10 km). Therefore, 4 GP symbols are required for this distant user. Theoretically, the downlink-to-uplink handover protection for this distant user requires 48.34 μs, but 4 GP symbols are equivalent to a protection duration of 64.84 μs, resulting in a significant waste of spectrum resources.

[0034] In some embodiments provided in this application, such as Figure 3 The diagram illustrates a flowchart of a wireless communication resource allocation method applied to the network side, comprising: S110: Obtain the wireless parameters and environmental parameters for the current period. The wireless parameters include at least the signal round-trip propagation delay, and the environmental parameters include at least one of the following: terrain, weather, and base station load. Specifically, the network side is a concept relative to the terminal side or user side. Network-side equipment can provide wireless communication coverage for a specific area and establish communication with terminal-side equipment. The network-side system can collect wireless parameters and environmental parameters in real time during the current communication cycle. Among them, wireless parameters refer to the relevant parameters after the network side and the terminal side establish communication, including one or more of the following: round-trip propagation delay, Received Signal Strength Indicator (RSSI), Channel Quality Indication (CQI), Signal-to-Noise Ratio (SNR), and even Signal-to-Interference Plus Noise Ratio (SINR). Environmental parameters can be geographical environment-related parameters between the network side and the terminal side, such as terrain (sea, desert, grassland, etc.), weather, and base station load status.

[0035] Furthermore, the raw data of wireless and environmental parameters collected in real time can be preprocessed through format conversion, noise filtering, standardization, or normalization to form structured feature vectors for subsequent analysis and processing.

[0036] S120: Input the wireless parameters, environmental parameters, and target coverage distance into the pre-trained first prediction model, and output the guard interval (GP) prediction value; wherein, the first prediction model is obtained by training a deep learning model based on the historical wireless parameter dataset, the historical environmental parameter dataset, and the corresponding historical coverage distance labels; the target coverage distance is determined based on the wireless parameters and environmental parameters.

[0037] Specifically, the first prediction model is obtained by training a deep learning model, such as a feedforward neural network (FNN), a recurrent neural network (RNN), or a variant thereof, a long short-term memory network (LSTM), or a gated recurrent unit (GRU). In this embodiment, the deep learning model is trained using a wireless parameter dataset and an environmental parameter dataset, enabling the trained model to perform propagation delay prediction tasks in ultra-long-range coverage scenarios. Based on the input wireless parameters, it outputs the corresponding GP prediction value for configuring the GP of the current communication cycle.

[0038] Furthermore, the training process of the first prediction model aims to minimize the mean squared error between the predicted GP length and the true GP length. Through backpropagation algorithm and optimizer (e.g., Adam), the model continuously updates its network parameters to gradually reduce the prediction error, thereby improving the accuracy of the prediction.

[0039] Furthermore, the hyperparameters of the first prediction model, such as the learning rate, the number of network layers, the number of neurons, and the selection of activation functions, can be determined through extensive offline training experiments based on datasets collected from wireless communication environments such as terrain, weather conditions, and base station load. They can also be optimized through methods such as cross-validation.

[0040] The wireless parameter dataset comprises wireless parameters from historical communication cycles, such as those from the past month or year. These parameters can include one or more parameters, such as signal round-trip propagation delay, received signal strength indication, channel quality indication, signal-to-noise ratio (SNR), and even signal-to-interference-plus-noise ratio (SINR). The environmental parameter dataset comprises environmental characteristic parameters from historical communication cycles, such as those from the past month or year. These parameters can include one or more parameters, such as terrain, temperature, humidity, weather conditions, and base station load. It is understandable that different ultra-long-range coverage scenarios, such as ocean areas, deserts, and grasslands, have different impacts on propagation delay due to varying environmental characteristics. By combining the wireless parameter dataset and the environmental parameter dataset to train a deep learning model and employing supervised learning, the model can learn and extract the inherent patterns for determining the optimal GP length from massive labeled datasets.

[0041] Furthermore, the training process of the first prediction model aims to minimize the mean squared error (MSE) loss function, as shown in the following formula: (Equation 1.1) In the above formula, M represents the number of samples in the training dataset; It is the true GP length of the i-th sample; while The model is based on input features The predicted GP length. In this implementation, the model continuously updates the weights W and biases b through a backpropagation algorithm and an optimizer (e.g., Adam) to gradually reduce the loss value, thereby improving the accuracy of the prediction.

[0042] Furthermore, the hyperparameters of the first prediction model, such as the learning rate, the number of network layers, the number of neurons, and the selection of activation functions, can be determined through extensive offline training experiments based on datasets collected from wireless communication environments such as terrain, weather conditions, and base station load. They can also be optimized through methods such as cross-validation. No restrictions are imposed here.

[0043] S130: Configure the GP for the current wireless communication cycle based on the GP prediction value.

[0044] Specifically, after the first prediction model is deployed online, it can perform GP prediction based on real-time collected wireless parameters, such as at a frequency of seconds or sub-seconds. Network-side devices, such as the MAC layer scheduler of the base station, dynamically adjust the actual GP configuration based on the GP prediction values.

[0045] Those skilled in the art will understand that, based on the wireless communication resource allocation method provided in this embodiment, by combining wireless parameter datasets and environmental parameter datasets to train a GP prediction model, the model can learn and extract the inherent rules for determining the optimal GP length from massive labeled datasets. This embodiment transforms the "one-size-fits-all" GP allocation method in the prior art into an adaptive mode that precisely matches the actual channel environment, thereby achieving a dynamic optimal balance between coverage expansion and spectrum efficiency improvement while ensuring basic communication quality.

[0046] In some embodiments provided in this application, another wireless communication resource configuration method is also provided, in which the GP of the current wireless communication cycle is configured according to the GP prediction value, including: S131: Obtain the historical GP value of the previous communication cycle and the preset GP adjustment time interval.

[0047] Specifically, the GP adjustment interval refers to the time interval between the last GP adjustment and the current communication cycle. A preset GP adjustment interval is a threshold value predetermined according to business needs, used to indicate the timing of GP adjustments, or used in conjunction with other parameters to indicate the timing of GP adjustments.

[0048] S132: If the difference or rate of change between the GP predicted value and the GP historical value exceeds the preset adjustment threshold, and the time since the last GP adjustment reaches the preset GP adjustment time interval, configure the GP of the current wireless communication cycle according to the GP predicted value.

[0049] Specifically, the difference or rate of change between the predicted GP value and the historical GP value exceeds a preset adjustment threshold. For example, the change ΔGP of the predicted GP value relative to the historical GP value exceeds a certain percentage (e.g., 10%) of the historical GP value; or the change ΔGP of the predicted GP value relative to the historical GP value exceeds a certain absolute value (e.g., 20µs). The time since the last GP adjustment reaches a preset GP adjustment time interval. For example, the GP adjustment time interval reaches a certain threshold (e.g., 5s or 10s).

[0050] S133: If the difference or rate of change between the GP predicted value and the GP historical value does not exceed the preset adjustment threshold, and / or the time since the last GP adjustment has not reached the preset GP adjustment time interval, configure the GP of the current wireless communication cycle according to the GP historical value.

[0051] Those skilled in the art will understand that the wireless communication resource configuration method provided in this embodiment achieves a smooth transition of delayed adjustment by controlling the timing of GP adjustment and GP configuration, which can effectively prevent frequent GP adjustments caused by instantaneous environmental disturbances from leading to system instability.

[0052] In some embodiments provided in this application, another wireless communication resource configuration method is also provided, in which the preset adjustment threshold includes at least one of a relative change threshold and an absolute change threshold; wherein the relative change threshold is 10% of the historical GP value and the absolute change threshold is 20 microseconds.

[0053] Specifically, the relative change threshold is a threshold set for the degree of change of the GP predicted value compared to the GP value of the previous communication cycle, and the degree of change can be determined with reference to equation (1.2). The absolute change threshold is a threshold set for the difference between the GP predicted value and the GP value of the previous communication cycle.

[0054] (Equation 1.2) In some embodiments provided in this application, another wireless communication resource configuration method is also provided, in which the dynamic adjustment range of the GP length corresponding to the GP prediction value output by the first prediction model is configured as 350 microseconds to 750 microseconds based on the target sea area coverage distance of 50 kilometers to 100 kilometers.

[0055] Specifically, in wireless communication scenarios with ultra-long-range coverage in marine areas, the coverage distance is generally between 50 and 100 kilometers, and the signal propagation delay can reach 333 to 667 microseconds. Considering the necessary processing delay and a certain system margin, the dynamic adjustment range of GP is initially set between 350 and 750 microseconds. This dynamic adjustment range is a key parameter preset based on the unique characteristics of ultra-long-range coverage in marine areas, which significantly expands the service range of 5G technology in marine environments.

[0056] In some embodiments provided in this application, another wireless communication resource configuration method is also provided, in which, after configuring the GP of the current wireless communication cycle according to the GP prediction value, the method further includes: S210: Obtain service parameters, which include at least one of the following: uplink / downlink traffic ratio, current number of active users, ratio of various services, average channel quality, and GP of the current wireless communication cycle.

[0057] Specifically, service parameters refer to the communication service operation parameters of network-side equipment within the current communication cycle, including at least one of the following: uplink / downlink traffic ratio, current number of active users, ratio of various services, average channel quality, and GP of the current wireless communication cycle. Among them, the ratio of various services refers to the proportion of different communication services such as voice services, data services, and video services.

[0058] S220: Input the service parameters into the trained second prediction model to obtain the predicted value of the uplink and downlink time slot ratio; wherein, the second prediction model is trained on the reinforcement learning algorithm based on the service parameter dataset, and the uplink and downlink time slot ratio is the ratio of the number of symbols in the uplink to the number of symbols in the downlink.

[0059] Specifically, the second prediction model is used to perform the task of predicting the uplink / downlink time slot allocation. The uplink / downlink time slot allocation is the ratio of the number of symbols in the uplink to the number of symbols in the downlink. As mentioned earlier, with a single uplink / downlink switching period of 5ms, the typical time slot allocation for a wireless data frame is DDDDDDDSUU, with each time slot containing 14 OFDM symbols. A possible allocation for a special time slot S is DDDDDGGGGUUUU. It is understandable that predicting a more suitable uplink / downlink time slot allocation using artificial intelligence methods can further improve the spectrum resource utilization efficiency of ultra-long-range coverage communication.

[0060] Furthermore, the second prediction model is obtained by training the reinforcement learning algorithm based on a service parameter dataset. The service parameter dataset refers to service parameters from historical communication cycles, such as service parameters for each communication cycle within different time periods like the past month or the past year. Specifically, it can include one or more parameters such as uplink / downlink traffic ratio, current active user count, proportion of various services, average channel quality, and GP of the current wireless communication cycle. In this embodiment, the reinforcement learning algorithm (RL) can be Q-learning or other types of reinforcement learning algorithms, without limitation.

[0061] Furthermore, the second prediction model is obtained by training a reinforcement learning algorithm. The core elements of the reinforcement learning algorithm include a state space, an action space, and a reward function. In this embodiment, the state space of the reinforcement learning algorithm is defined as service parameters, which describe network service requirements and the current system state through one or more parameters among uplink / downlink traffic ratio, current number of active users, proportion of various services, average channel quality, and GP of the current wireless communication cycle. This state space can be represented and processed using a normalized service parameter feature vector. In this embodiment, the action space of the reinforcement learning algorithm is defined as the combination of uplink / downlink time slot ratios. Specifically, this action space includes all possible TDD uplink / downlink time slot ratio schemes, such as the various downlink (DL) and uplink (UL) symbol ratio combinations defined in the LTE TDD standard configuration or the 5G NR flexible time slot format. In this embodiment, the reward function of the reinforcement learning algorithm is defined as maximizing the long-term cumulative reward of the first objective parameter. The first parameter can be one or more of the following: communication throughput improvement rate, average transmission latency reduction rate, and high-priority service quality satisfaction rate. Specifically, it can be expressed as a weighted sum of each first parameter, and the weight of each first parameter can be flexibly configured and dynamically adjusted according to the actual network operation strategy.

[0062] The training of reinforcement learning algorithms is a process of continuous learning and optimization. Based on the current state, an action is selected to be executed, and the system then undergoes a state transition and generates a corresponding reward signal. The model uses these experience samples to continuously update its internal decision-making strategy, and finally obtains the trained second prediction model.

[0063] For example, when the second prediction model is obtained by training a Q-learning algorithm, the update formula for its Q value is as follows: (Equation 2.1) In the above formula, It is the learning rate, which controls the degree to which new information updates the old Q value; γ is the reward value obtained at time step t; γ is the discount factor, which indicates the degree of importance attached to future rewards. Indicates the next state s t+1 The maximum Q-value of all possible actions is calculated. In this way, the model can find a balance between exploration and exploitation, thus achieving an optimal uplink / downlink time slot allocation decision strategy. Furthermore, the learning rate is initially set to a relatively small value, such as 10. The initial discount factor is typically set to be close to 1, such as 0.95, to ensure the stability of the learning process. This encourages the model to focus on and pursue long-term rewards. The exploration strategy, such as the ε parameter in the epsilon-greedy algorithm, controls the balance between "exploring" the unknown domain and "utilizing" known optimal strategies. The initial exploration rate ε is usually set to a high value (e.g., 1.0) and gradually decays to a smaller value (e.g., 0.1) as training progresses, ensuring a smooth transition from a fully exploratory phase to a stable utilization phase of known optimal strategies.

[0064] S230: Configure the uplink and downlink time slot ratio for the current wireless communication cycle based on the predicted uplink and downlink time slot ratio.

[0065] Specifically, after the second prediction model is deployed online, it can predict the uplink / downlink time slot allocation based on real-time collected service parameters, such as at a frequency of seconds or sub-seconds. The network-side equipment dynamically adjusts the actual uplink / downlink time slot allocation settings based on the predicted values.

[0066] Those skilled in the art will understand that the wireless communication resource configuration method provided in this embodiment can intelligently and dynamically adjust the uplink and downlink time slot ratio in the TDD frame structure according to real-time changes in service requirements, thereby further improving the efficiency of spectrum resource utilization.

[0067] In some embodiments provided in this application, another wireless communication resource allocation method is also provided, which, after configuring the uplink / downlink time slot ratio of the current wireless communication cycle according to the uplink / downlink time slot ratio prediction value, further includes: S310: Obtain the spectrum information of the available communication frequency band, and input the spectrum information into the third prediction model to obtain the interference signal prediction result. The interference signal prediction result includes at least the interference category, interference intensity, interference frequency band, and interference trend.

[0068] It is understandable that in some ultra-long-range wireless communication scenarios, such as maritime scenarios, there may be radio frequency signal sources such as radar and communication signals from ships or other facilities, resulting in a complex electromagnetic environment that can interfere with wireless communication. In this embodiment, the available communication frequency band refers to the operating frequency band of the 5G system and all potential interference source frequency bands. For example, when the 5G system operates at 2.6GHz, the scanning bandwidth may need to cover the range of 2.4GHz to 2.8GHz. Wideband spectrum analysis is achieved by scanning the available communication frequency band to collect the raw spectrum information in the target frequency band, i.e., signal scanning spectrum information. This signal scanning spectrum information includes signal strength (i.e., amplitude), signal frequency, signal phase, and other information.

[0069] Furthermore, the parameter settings during the spectrum scanning process are configured and optimized based on the estimated electromagnetic environment of the target ultra-long-range coverage scenario before actual deployment. For example, the scanning bandwidth needs to cover the operating frequency band of the 5G system and all potential interference source frequency bands. For instance, when the 5G system operates at 2.6GHz, the scanning bandwidth may need to cover the range of 2.4GHz to 2.8GHz. The specific configuration can be made according to the actual business needs and is not limited here. The resolution bandwidth (RBW) can be set according to the specific characteristics of the interference signal to be identified, for example, it can be set in the range of 10kHz to 1MHz. For the scanning time or scanning period, the scanning period can be set in the tens to hundreds of milliseconds for sudden interference, and can be appropriately relaxed to the second level for continuous interference.

[0070] Furthermore, the third prediction model is used to perform the task of identifying and classifying interference signals. Based on the input signal scanning spectrum information, such as the signal strength (i.e., amplitude), signal frequency, signal phase, and transmit power of the target frequency band, it obtains the interference signal prediction result. The interference signal prediction result includes at least the interference category, interference intensity, interference frequency band, and interference trend. Among them, the interference category includes, for example, broadcast signals, shipborne radar signals, interference from other wireless communication systems, natural environmental noise, and even pseudo signals caused by multipath effects; interference intensity refers to the strength of the interference signal, which can also be defined by the absolute interference signal power value, interference-to-noise ratio (I / N), or carrier-to-interference ratio (C / I); interference frequency band refers to the frequency band in which the interference signal is located; interference trend refers to the interference intensity and interference frequency band in the future duration of the interference signal.

[0071] Furthermore, the third prediction model is obtained by training a neural network based on an interference feature dataset. This interference feature dataset is constructed through extensive actual measurements or simulation analyses of relevant ultra-long-range coverage scenarios, reflecting the complex electromagnetic environment of the corresponding ultra-long-range coverage scenarios. In this embodiment, the neural network used can be a Convolutional Neural Network (CNN). In this case, the signal scanning spectrum information and the interference feature dataset can be represented as a spectrogram for training the CNN and for prediction processing after deployment. Alternatively, the neural network used can be a Recurrent Neural Network (RNN), a Long Short-Term Memory Network (LSTM), or a Gated Recurrent Unit (GRU) network model for processing time-domain sequence features. In this case, the signal scanning spectrum information and the interference feature dataset can be represented as a sequence for training various time-series neural networks and for prediction processing after deployment. Or, the neural network used can be a Convolutional Recurrent Neural Network (CRNN). In this case, the signal scanning spectrum information and the interference feature dataset can be represented as a combination of a spectrogram and a sequence for training the CRNN and for prediction processing after deployment.

[0072] Furthermore, during the training of the neural network, supervised learning is employed to fully train it on a large labeled dataset containing various typical interference signals (such as broadcast signals, shipborne radar signals, interference from other wireless communication systems, natural environmental noise, and even pseudo-signals caused by multipath effects), as well as normal 5G communication signals. The goal is to minimize the loss function for the classification task, such as the cross-entropy loss function. For example, the loss function can be set as follows: (Equation 3.1) In the above formula, It is an indicator variable for the k-th type of interference signal in real-world conditions. The model predicts the spectral data. The probability of the k-th type of interference signal. During training, by minimizing the loss function, the neural network model can learn the interference types corresponding to different spectral features, thus enabling the trained third prediction model to achieve high interference identification accuracy.

[0073] Furthermore, regarding the setting and optimization of hyperparameters of the neural network model itself, such as the scientific setting of the decision threshold, specific performance indicators can be pursued on the validation set, such as a detection probability greater than 95%, a false alarm probability less than 2%, and an overall accuracy greater than 90%. The threshold can also be flexibly adjusted according to the tolerance of system sensitivity and false alarm rate in actual business.

[0074] S320: Based on the interference signal prediction results and channel quality, the available communication frequency band is divided into multiple logical slices.

[0075] Specifically, information such as the presence and intensity of interference in each frequency band within the available communication band can be obtained from interference signal prediction results. Channel quality is represented by the Channel Quality Indication (CQI). It can be understood that CQI is obtained by the terminal-side equipment measuring CSI-RS and reported by the terminal-side equipment under the control of the network-side equipment; it is information possessed by the network-side equipment. In this case, logical slices can be divided according to actual service needs, combining the interference situation and channel quality of each frequency band. For example, if interference is in a certain interval and channel quality is in another specific interval, that frequency band is designated as the first logical slice.

[0076] S330: Based on the interference signal prediction results and the wireless channel parameters of each logical slice, the target logical slice is selected through the fourth prediction model.

[0077] Specifically, after reasonably dividing the network into multiple logical slices, it is necessary to select appropriate logical slices for terminal-side devices / terminal users / services. The selection criteria include the wireless channel parameters of the logical slice, such as at least one of real-time channel quality indicators, signal-to-noise ratio (SNR), and interference trends. The selection can be based on whether the wireless channel parameters meet the service requirements. Among these, real-time channel quality indicators (CQI) and signal-to-noise ratio (SNR) are information already available to network-side devices.

[0078] Furthermore, the fourth prediction model can be obtained by training a time series analysis network model, which is used to intelligently select the target logical slice based on the interference signal prediction results and the wireless channel parameters of each logical slice. For example, the fourth prediction model can be constructed using an autoregressive integral moving average (ARIMA) model, or based on deep learning models such as recurrent neural networks (RNN) and long short-term memory networks (LSTM).

[0079] The training objective of the fourth prediction model is to learn the mapping relationship between historical interference trends, channel quality sequences, and optimal slice selection. Its input features typically include historical and current interference categories, intensities, and trends output by the third prediction model, as well as historical and real-time channel quality indices (CQI) and signal-to-noise ratio (SNR) for each logical slice. The output is an availability and quality score for each logical slice within a future prediction time window; the system then selects the slice with the highest score as the target logical slice.

[0080] During training, the fourth prediction model is optimized to maximize the accuracy of selecting the correct slice (i.e., the slice that provides stable, high-quality communication) or minimize the probability of communication interruption. After online deployment, the fourth prediction model can combine real-time predicted interference trends to achieve intelligent selection of logical slices and proactive avoidance of interference.

[0081] Understandably, the selection of logical slices affects the development of wireless communication services within a certain period of time. It is not only related to wireless channel parameters, but also depends on the predicted interference trend. By training time series analysis network models, it is beneficial to intelligently select suitable network models and actively avoid interference.

[0082] S340: Configure the spectrum resources for the current communication cycle based on the selection result of the fourth prediction model.

[0083] Specifically, after the network-side equipment is deployed online according to the fourth prediction model, it predicts the logical slice selection results in real time and actually performs spectrum resource configuration for the current communication cycle, that is, configures the communication frequency band for service implementation.

[0084] Those skilled in the art will understand that the wireless communication resource allocation method provided in this embodiment achieves accurate and rapid perception and identification of various interference signals through signal scanning and artificial intelligence methods. Based on interference identification and wireless channel quality, the available communication frequency band is divided into multiple logical slices. Finally, by combining the interference identification results and wireless channel quality, a timing prediction method is used to achieve intelligent selection of logical slices and active interference avoidance, thereby enhancing the anti-interference capability and robustness of the entire communication system in complex and ever-changing ultra-long-range coverage electromagnetic environments.

[0085] In some embodiments provided in this application, another wireless communication resource allocation method is also provided. In this method, multiple logical slices include at least a first slice and a second slice. The first slice operates in a frequency band where the interference intensity is lower than a first reference threshold and the channel quality is higher than a first superior threshold. The second slice operates in a frequency band where the interference intensity is lower than a second reference threshold and the channel quality is higher than a second superior threshold. The first reference threshold is lower than the second reference threshold, and the first superior threshold is higher than the second superior threshold. This allocation arrangement results in less interference and higher channel quality in the frequency band where the first slice is located.

[0086] Furthermore, by using the first slice as the preferred spectrum slice and the second slice as the backup spectrum slice, when the first slice is unavailable or resources are severely insufficient, services can be automatically switched to the second slice based on service priority, ensuring service continuity.

[0087] Furthermore, the values ​​of the first baseline threshold, the second baseline threshold, the first high-quality threshold, and the second high-quality threshold can be set according to the needs of preferred and backup slices in actual business operations. Alternatively, initial values ​​can be assigned and dynamically adjusted through an adaptive optimization mechanism. For example, the initial values ​​can be obtained through offline modeling and simulation. First, a simulation environment is constructed that includes a channel propagation model of an overloaded coverage communication scenario and the characteristics of typical interference sources (such as broadcasting and shipborne radar). In this environment, 5G communication performance (such as throughput and block error rate) under different signal-to-noise ratios and interference intensities is simulated. Combined with the channel quality requirements of different QoS levels in the 5G protocol (such as the CQI standard), a set of initial thresholds that can meet different performance objectives are calibrated (e.g., defining an initial CQI threshold for "high-quality" and an initial upper limit for "tolerant" interference power). This ensures that the system has a scientifically reasonable operating baseline at startup. For example, the adaptive optimization mechanism for each threshold can continuously monitor the long-term correlation between the overall network's macro KPIs (such as total throughput, drop rate, and handover success rate) and the actual performance of each spectrum slice, and make real-time adjustments through heuristic search algorithms or meta-learning techniques. Understandably, if the first high-quality threshold is set too strictly (resulting in an excessively small "preferred spectrum slice" resource pool, squeezing a large number of services to "backup slices," and causing overall performance degradation), the mechanism will automatically initiate a slow-cycle adjustment to the "first high-quality threshold." Conversely, if the second baseline threshold is too lenient (resulting in frequent substandard communication quality in "backup slices"), the mechanism will also automatically tighten this upper limit, thereby making the allocation of logical slices more in line with system operation needs.

[0088] Furthermore, this implementation does not limit the number of the first slice and the second slice, and can be reasonably set according to actual service needs and / or channel quality.

[0089] In some embodiments provided in this application, another wireless communication resource configuration method is also provided, in which the multiple logical slices further include a third slice and / or a fourth slice, wherein, The third slice is allocated to a frequency band with stable characteristics and corresponding suppression algorithms for persistent interference. When a service is scheduled to the third slice, the corresponding interference suppression algorithm is activated synchronously. The fourth slice is configured to operate in an anti-jamming communication mode, which includes at least one of using extremely narrow bandwidth, lowest order modulation, and strongest error correction coding.

[0090] Specifically, the third slice, as a tolerable spectrum slice, is allocated to frequency bands with stable interference characteristics and corresponding suppression algorithms. Stable interference characteristics mean that the strength, frequency, power, and other characteristics of the interference signal do not fluctuate significantly over the duration of the interference; for example, this is measured by the ratio of extreme differences to the mean not exceeding a certain percentage. Suppressible interference means that network-side devices or systems possess the means to suppress this type of interference, such as adaptive filtering and interference cancellation techniques. The strength, frequency, power, and other characteristics of the interference signal, as well as the type of interference, can be obtained from interference signal prediction results. It is understandable that for ultra-long-range wireless communication scenarios, due to the large coverage area and relatively complex electromagnetic environment, dividing the spectrum into tolerable slices allows for maximizing spectrum utilization at a controllable performance cost, greatly improving the efficiency of spectrum resource utilization and avoiding the complete abandonment of a portion of the spectrum due to a small amount of manageable interference.

[0091] Furthermore, when a service is scheduled to the third slice, the corresponding interference suppression algorithm, such as adaptive filtering or interference cancellation technology, is activated simultaneously to maximize the efficiency of spectrum resource utilization while effectively ensuring the quality of communication services.

[0092] Furthermore, the fourth slice, serving as an emergency communication slice, operates in an anti-interference communication mode. This anti-interference communication mode could be, for example, narrowband communication, or low-order modulation or error-correcting coding. Narrowband mode restricts signal transmission to a relatively narrow frequency range, providing strong anti-interference capabilities. Low-order modulation typically refers to modulation methods where each symbol has only two states: 0 or 1, such as binary amplitude shift keying (2ASK) or binary phase shift keying (2PSK), offering strong anti-interference capabilities and suitable for scenarios with poor channel environments. Error-correcting coding refers to techniques that add redundant symbols to the information sequence to detect and correct transmission errors, thereby improving communication reliability and reducing the bit error rate. It is understandable that in ultra-long-range wireless coverage scenarios, there may be extreme electromagnetic environment conditions. The fourth slice can prioritize the establishment and maintenance of links for core emergency communications (such as basic voice or critical control signaling).

[0093] Furthermore, in this embodiment, the logical slice further includes a third slice and / or a fourth slice, meaning that the logical slice further includes a third slice, or the logical slice further includes a fourth slice, or the logical slice further includes both a third slice and a fourth slice.

[0094] It is important to note that the GP length determined by the first prediction model is independently adapted to the signal round-trip propagation delay (i.e., coverage distance) of a specific end user, with the fundamental purpose of preventing uplink and downlink time-domain interference in TDD. The spectrum slice selected by the fourth prediction model, however, is a frequency domain resource dynamically selected based on real-time interference identification results and channel quality. When scheduling a specific offshore user, the system must use the GP length calculated by the first prediction model and apply the TDD frame structure containing this GP to any spectrum slice (whether preferred, backup, or tolerant) selected by the fourth prediction model for data transmission. Both are orthogonally coordinated in the time and frequency domains and are indispensable, thereby ensuring reliable communication on optimal frequency domain resources under ultra-long-distance coverage.

[0095] In some embodiments provided in this application, another wireless communication resource allocation method is also provided. After configuring the spectrum resources for the current communication cycle according to the selection result of the fourth prediction model, the method further includes: S410: Obtain radio resource scheduling parameters, which include at least one of the following: real-time load of available frequency bands, load warning threshold and congestion threshold, terminal signal quality of available frequency bands, service demand type, terminal moving speed, and terminal carrier aggregation capability.

[0096] Specifically, the real-time load of available frequency bands can be the current load of each available frequency band, such as the number of users and traffic volume. The load warning threshold and congestion threshold are preset load thresholds for each frequency band. For example, for the 700MHz frequency band, the initial warning threshold can be set to 60% and the congestion threshold to 80%, and for the 2.6GHz frequency band, the warning threshold can be set to 70% and the congestion threshold to 85%. These thresholds serve as the state input and decision boundary of the fifth prediction model (trained based on deep reinforcement learning), and can be dynamically adjusted according to the actual network load and service mode through a closed-loop adaptive optimization mechanism, thereby achieving forward-looking congestion management and intelligent load balancing.

[0097] Available frequency band terminal signal quality refers to the signal quality indicators of terminal devices on various available frequency bands; service demand types include different services such as voice, video, and data; terminal movement speed refers to the degree of change in the position of terminal devices during communication; carrier aggregation (CA) refers to the technology of integrating wireless channel resources of multiple frequency bands to improve data transmission rate and network efficiency. Currently, not all terminal devices support carrier aggregation capabilities.

[0098] S420: Input the radio resource scheduling parameters into the trained fifth prediction model to obtain the resource scheduling decision. The resource scheduling decision is used to instruct at least one of the following operations: allocate an initial frequency band, trigger frequency band switching, and configure carrier aggregation. The fifth prediction model is obtained by training a deep reinforcement learning algorithm based on the radio resource scheduling parameter dataset.

[0099] Specifically, the fifth prediction model is used to make selection decisions for wireless resource scheduling strategies. Based on the input wireless resource scheduling decision parameters, it obtains resource scheduling prediction results. The fifth prediction model is obtained by training a deep reinforcement learning algorithm on a wireless resource scheduling parameter dataset. The resource scheduling prediction results include one of the following: initial frequency band allocation, frequency band switching, or carrier aggregation configuration. The wireless resource scheduling parameter dataset is pre-constructed and contains wireless resource scheduling parameters for historical communication periods, such as the past month, the past year, or a specific historical time period. In the resource scheduling prediction results, initial frequency band allocation refers to allocating an initial frequency band for newly accessed terminal devices; triggering frequency band switching refers to switching the communication frequency band with other terminals for the current terminal based on the resource scheduling decision; and configuring carrier aggregation refers to enabling carrier aggregation communication mode for the current terminal.

[0100] Furthermore, the fifth prediction model is obtained by training a reinforcement learning algorithm. The core elements of reinforcement learning include state space, action space, and reward function. In this embodiment, the state space of the reinforcement learning algorithm is defined as wireless resource scheduling parameters, that is, at least one parameter among available frequency band real-time load, available frequency band terminal signal quality, service demand type, terminal movement speed, and terminal carrier aggregation capability, which characterizes the current state of the wireless resource scheduling system; the action space of the reinforcement learning algorithm is defined as executable scheduling operations such as allocating initial frequency bands, triggering frequency band switching, and configuring carrier aggregation; the reward function is defined as one of maximizing communication throughput, minimizing average transmission latency, and minimizing the load balancing pressure of each frequency band.

[0101] Furthermore, the training of reinforcement learning algorithms is a continuous learning and optimization process. Based on the current state, an action is selected and executed, causing a state transition and generating a corresponding reward signal. The model then uses these experience samples to continuously update its internal decision-making strategy, ultimately obtaining the trained fifth prediction model. The hyperparameters of its deep reinforcement learning networks (such as Deep Q-Network, DQN, or Actor-Critic architecture), including learning rate, discount factor, and exploration rate, need to be fine-tuned through a combination of offline simulation and online learning.

[0102] S430: Schedules wireless resources based on resource scheduling decisions.

[0103] Specifically, wireless resources are scheduled according to resource scheduling decisions, that is, wireless resource scheduling and configuration are actually carried out according to the resource scheduling decisions output by the fifth prediction model; for example, initial frequency bands are allocated to newly accessed terminal devices; or frequency band switching is performed between the current terminal and other terminals; or carrier aggregation communication mode is enabled for the current terminal.

[0104] In some embodiments provided in this application, another wireless communication resource configuration method is also provided, in which the decision conditions for configuring carrier aggregation (CA) by the fifth prediction model include: the signal-to-interference-plus-noise ratio (SINR) of the terminal on the secondary carrier is higher than 3dB, and / or the channel quality indicator (CQI) is higher than 6, and / or the terminal's service bandwidth requirement is higher than 15Mbps.

[0105] It is understandable that the decision logic of carrier aggregation (CA) is intelligently and uniformly managed by the fifth prediction model, but its specific decision-making behavior is still affected by a series of pre-set CA decision boundary conditions. These boundary conditions can be regarded as configurable parameters of the system. Specifically, in this embodiment, these boundary conditions mainly include minimum threshold requirements for the signal-to-noise ratio (SINR) and channel quality indication (CQI) of the secondary carrier, as well as the bandwidth demand threshold of the end user. For example, during initial configuration, the SINR threshold of the secondary carrier can be set to 3-5 dB, and the CQI threshold can be set to 6-8. In addition, for video services that require high bandwidth support, a bandwidth demand threshold of, for example, 15-20 Mbps can be set. The system will only consider enabling CA when the end user's demand exceeds this threshold.

[0106] In some embodiments provided in this application, another wireless communication resource configuration method is also provided. In this method, the decision conditions for the fifth prediction model to trigger the supplementary uplink (SUL) handover include: detecting that the terminal is carrying out a high-volume uplink service of a preset type, and that its uplink signal quality in the current service frequency band is lower than a preset threshold.

[0107] For example, when a terminal performs high-bandwidth uplink services such as uploading high-definition video or streaming live, and the uplink signal-to-noise ratio of the current service frequency band is lower than a preset threshold (e.g., 10dB), the fifth prediction model will decide to trigger SUL switching, migrating the uplink to a supplementary uplink frequency band to improve uplink transmission capability and user experience.

[0108] Specifically, high-volume uplink services can be live streaming, video streaming, or similar services. By detecting when a terminal is conducting a pre-defined type of high-volume uplink service, and the uplink signal quality in the current service band is below a pre-defined threshold, a decision to trigger a supplementary uplink SUL handover can be made, effectively ensuring the quality of important communication services.

[0109] In some embodiments provided in this application, another wireless communication resource allocation method is also provided, which, after scheduling wireless resources according to the resource scheduling prediction result, further includes: S510: Obtain the outputs of the first to fifth prediction models.

[0110] Specifically, the outputs of the first to fifth prediction models include GP prediction values, uplink / downlink time slot ratio prediction values, interference signal prediction results, logical slice selection results, and resource scheduling prediction results.

[0111] S520: Determine whether there are any conflicts in the outputs of the first to fifth prediction models.

[0112] For example, the fourth prediction model predicts the selection of spectrum slices, which involves the allocation of frequency domain resources to improve resource utilization. The fifth prediction model predicts resource scheduling strategies, involving decisions such as initial frequency band allocation, frequency band switching, and carrier aggregation, also involving the allocation of frequency domain resources to achieve load balancing. There is a possibility of conflict between the output decisions of the fourth and fifth prediction models. For instance, the fifth prediction model might output carrier aggregation, aggregating users to the 2.6GHz band to improve throughput, while the interference prediction results from the third prediction model show high-intensity interference in the 2.6GHz band.

[0113] Specifically, network-side devices can determine whether a conflict exists based on the outputs of the first to fifth prediction models, from the perspectives of time-domain resource allocation, frequency-domain resource allocation, frequency-domain resource allocation, and interference signals.

[0114] S530: In the event of a conflict, perform avoidance operations according to preset rules, which include one of the following: prioritizing link stability, prioritizing load balancing, and prioritizing maximizing throughput.

[0115] Specifically, when the preset rule prioritizes link stability, the network-side device will prioritize executing the output of the fourth prediction model, i.e., the logical slice selection result, and will not execute the output of the fifth prediction model, so as to ensure that the communication link operates in a frequency band that meets the requirements for interference level and channel quality, thus ensuring the stability of the communication link. When the preset rule prioritizes load balancing, the network-side device will prioritize executing the output of the fifth prediction model, i.e., allocating an initial frequency band to the terminal, triggering frequency band switching, or configuring carrier aggregation, and will not execute the output of the fourth prediction model, so as to ensure load balancing among the available frequency bands. When the preset rule prioritizes maximizing throughput, the network-side device will prioritize executing the output of the fifth prediction model, i.e., enabling carrier aggregation for the terminal, and will not execute the output of the fourth prediction model, so as to maximize throughput.

[0116] Those skilled in the art will understand that, based on the wireless communication resource allocation method provided in this embodiment, the key state parameters and decision outputs of the first to fifth prediction models are periodically aggregated, which can achieve strategy optimization and conflict avoidance according to the preset network optimization objectives, thus ensuring the agility and real-time performance of the entire network system.

[0117] In some embodiments provided in this application, another method for allocating wireless communication resources is also provided, which, after scheduling wireless resources according to resource scheduling prediction results, further includes: S610: Obtain key quality indicators of the network system, including at least one of the following: total throughput, average latency, dropout rate, and user experience quality. Specifically, Key Quality Indicators (KPIs) are observable metrics for evaluating the quality of communication services. In this embodiment, at least one of total throughput, average latency, and call drop rate is obtained for subsequent processing steps. Total throughput refers to the amount of data successfully transmitted per unit time, typically expressed in bits per second (bps) or bytes per second (Bps); latency refers to the time required for data blocks (such as messages, packets, bit streams, etc.) to travel from one end of the network to the other; and call drop rate refers to the probability of unexpected connection interruption during communication.

[0118] Among them, the Quality of Experience (QoE) is quantitatively evaluated by comprehensively considering factors such as signal strength, data transmission rate, and packet loss rate, and is used to reflect the end user's subjective perception of the quality of communication services. As one of the core inputs of the closed-loop optimization mechanism, when the regional QoE is detected to be continuously lower than the preset acceptable range, the system will automatically trigger the adjustment of relevant prediction model parameters, forming an autonomous optimization loop of "experience evaluation - parameter correction".

[0119] S620: Input key quality indicators into the pre-trained first optimization model to obtain hyperparameter optimization suggestions for one or more prediction models from the first to the fifth prediction models. The hyperparameters include at least one of the following: learning rate, discount factor, and optimizer type. Specifically, the first optimization model can be an advanced heuristic search algorithm, such as simulated annealing, genetic algorithms, particle swarm optimization, etc., or it can be a meta-learning technique, etc., to achieve hyperparameter tuning for the first to fifth prediction models. Hyperparameters include at least one of the following: learning rate, discount factor, and optimizer type. The learning rate controls the degree to which the model adjusts its parameters in each optimization step; the discount factor calculates the present value of future rewards to reflect the importance placed on future rewards; and the optimizer is an algorithm used to adjust model parameters to minimize the loss function, such as Adam, RMSprop, Adagrad, etc. In addition, hyperparameters can also include batch size, number of training epochs, etc.

[0120] S630: Based on the hyperparameter optimization recommendations, set the hyperparameters of one or more prediction models from the first to the fifth prediction models.

[0121] Specifically, based on the hyperparameter optimization results, the network-side system actually adjusts the hyperparameters of one or more prediction models from the first to the fifth prediction models to achieve hyperparameter tuning of the first to the fifth prediction models.

[0122] Those skilled in the art will understand that the wireless communication resource allocation method provided in this embodiment can organically integrate the capabilities of various prediction models, thereby enabling the overall effect of the entire system to continuously and dynamically tend towards the global optimal state.

[0123] In some embodiments provided in this application, such as Figure 4 As shown, a 5G ultra-long-range coverage system for marine scenarios is also provided, including: The parameter acquisition module 410 is used to acquire wireless parameters, environmental parameters, service parameters and spectrum information, wherein the environmental parameters include at least parameters characterizing the propagation characteristics in the sea area.

[0124] Specifically, parameters that characterize the propagation characteristics of a sea area can include, for example, topography (types such as islands, harbors, and sea surfaces), weather, temperature, and humidity.

[0125] The intelligent time slot scheduling module 420 is connected to the parameter acquisition module and is used to dynamically configure the protection interval (GP) and / or time division duplex (TDD) time slot ratio based on wireless parameters and environmental parameters.

[0126] Specifically, the dynamic configuration of the guard interval (GP) and time division duplex (TDD) time slot ratio can be achieved using machine learning or artificial intelligence technologies. For example, based on the collected wireless and environmental parameters, machine learning or artificial intelligence techniques can be used to predict the guard interval (GP) and time division duplex (TDD) time slot ratio.

[0127] The cognitive radio anti-interference module 430 is connected to the parameter acquisition module and is used to perform spectrum sensing and intelligent frequency selection scheduling based on the spectrum information.

[0128] Specifically, spectrum information can be interference signal information and / or channel quality information for each available communication frequency band. Spectrum sensing refers to analyzing interference signals in each available communication frequency band to determine the interference status of each available communication frequency band, such as the type, frequency, amplitude (intensity), and phase of the interference, or to determine the communication channel quality of each available communication frequency band. Intelligent frequency selection scheduling refers to intelligently selecting the corresponding communication frequency band for terminal services based on the spectrum information of each available communication frequency band.

[0129] The multi-band intelligent coordination module 440 is connected to the parameter acquisition module and is used for multi-band load balancing, carrier aggregation (CA), and supplementary uplink scheduling (SUL).

[0130] The collaborative optimization center 450 is communicatively connected to the intelligent time slot scheduling module, the cognitive radio anti-interference module, and the multi-band intelligent collaborative module, respectively. It is used to aggregate the scheduling decision information of each module, perform conflict arbitration, and perform closed-loop adaptive optimization of the parameters of each module.

[0131] Specifically, the 5G ultra-long-range coverage system for marine scenarios provided in this embodiment can achieve, for example... Figure 3 For details of the method steps in the illustrated embodiment, please refer to [link / reference]. Figure 3 The wireless communication resource configuration method of the illustrated embodiment will not be described in detail here.

[0132] In some embodiments provided in this application, such as Figure 5 As shown, a wireless communication resource configuration device is also provided, comprising: The acquisition module 510 is used to acquire the wireless parameters and environmental parameters of the current period. The wireless parameters include at least the signal round-trip propagation delay, and the environmental parameters include at least one of the following: terrain, weather, and base station load. Prediction module 520 is used to input the wireless parameters, environmental parameters, and target coverage distance into a pre-trained first prediction model and output a guard interval (GP) prediction value; wherein, the first prediction model is obtained by training a deep learning model based on historical wireless parameter datasets, historical environmental parameter datasets, and corresponding historical coverage distance labels; the target coverage distance is determined based on the wireless parameters and environmental parameters; Configuration module 530 is used to configure the GP of the current wireless communication cycle according to the GP prediction value.

[0133] Specifically, the wireless communication resource allocation device provided in this embodiment can achieve the following: Figure 3 For details of the method steps in the illustrated embodiment, please refer to [link / reference]. Figure 3 The wireless communication resource configuration method of the illustrated embodiment will not be described in detail here.

[0134] Those skilled in the art will understand that, based on the wireless communication resource allocation device provided in this embodiment, by combining wireless parameter datasets and environmental feature datasets to train a GP prediction model, the model can learn and extract the inherent rules for determining the optimal GP length from massive labeled datasets. This embodiment transforms the "one-size-fits-all" GP setting method in the prior art into an adaptive mode that precisely matches the actual channel environment, thereby achieving a dynamic optimal balance between coverage expansion and spectrum efficiency improvement while ensuring basic communication quality.

[0135] In some embodiments provided in this application, such as Figure 6 As shown, a wireless communication resource configuration device is also provided, including: Memory 610 is used to store instruction sets; The processor 620 is used to call and execute the instruction set, and to implement the steps and corresponding technical effects of the method described in any of the foregoing embodiments by executing the instruction set.

[0136] In some embodiments provided in this application, a readable computer storage medium is also provided, on which a computer program is stored, wherein, When the computer program is executed by the processor, it implements the steps of the method described in any of the foregoing embodiments and the corresponding technical effects.

[0137] In some embodiments provided in this application, a computer program product is also provided, including a computer program / instruction that, when executed by a processor, implements the steps of the method described in any of the foregoing embodiments and the corresponding technical effects.

[0138] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0139] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0140] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0141] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0142] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0143] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0144] It should be understood that the training and prediction processes of the AI ​​models involved in the various embodiments of this specification all adhere to multiple legal and compliant principles, including legal data sources, compliant data content, compliant data governance, compliant training objectives and schemes, compliant training processes, compliant training environments and tools, and compliant ethical verification of training results, and comply with the requirements of Article 5 of the Patent Law. Among them: All datasets used for AI model training were obtained through legal means, covering three categories: publicly authorized data, data authorized by partners, and self-collected compliant data. Publicly authorized data comes from compliant data sources following open-source licenses such as Apache 2.0, with complete copyright attribution and authorization scope clearly marked, and no unauthorized open-source code or data reuse. Data authorized by partners has been subject to formal data usage agreements, clearly defining the scope, duration, and confidentiality obligations, and possessing a complete authorization chain. For self-collected data involving personal information, strict informed consent procedures have been followed, and anonymization processes (including but not limited to field masking, feature anonymization, and differential privacy technology applications) have been implemented to remove personally identifiable information, fully complying with the requirements of relevant laws and regulations such as the "Interim Measures for the Administration of Generative Artificial Intelligence Services" and the "Personal Information Protection Law."

[0145] The AI ​​model's dataset undergoes multiple screenings and cleaning processes to remove all content that may violate social morality or harm public interests. It contains no obscene, pornographic, violent, discriminatory, or information that endangers national or public safety, nor does it involve the illegal acquisition or use of genetic resources. For data in sensitive fields (such as healthcare and finance), an additional privacy-preserving computation module (including federated learning and secure multi-party computation technologies) ensures that the data is "usable but not visible," avoiding compliance risks during the original data transmission process and ensuring that the data application scenarios and uses comply with public order and good morals and industry regulatory requirements.

[0146] A complete data traceability system is established during the AI ​​model training process to automatically record the source, collection time, annotation process, cleaning rules, and permission allocation of training data, generating traceable compliance reports to ensure that the data is verifiable throughout its entire lifecycle. The dataset annotation process for AI models is completed by a professional human R&D team, clearly defining the proportion of human creative contributions and avoiding reliance on AI-generated data that has not undergone substantial human modification, thus meeting the examination requirements for "human main contributions" in AI patent applications.

[0147] The AI ​​model training objective focuses on the intelligent allocation and interference management optimization of wireless communication network resources. The training scheme and the final output results do not violate any mandatory provisions of laws and administrative regulations, do not harm the public interest or the legitimate rights and interests of others, and do not pose any potential risks of being used for illegal activities, privacy infringement, or public safety disruption. It strictly adheres to the ethical principle of "intelligent for good".

[0148] A closed-loop training framework is adopted to ensure compliance and controllability of the training process. The specific process is as follows: First, training samples are obtained through compliant data sources. After the aforementioned data cleaning and desensitization, they are input into the neural network model to generate preliminary training results. Second, an expert system is introduced to verify the preliminary results. Based on preset rules and human expert experience, the feasibility of the results is evaluated, and outputs that may pose ethical risks or compliance hazards are corrected (such as removing decision-making logic that violates public order and good morals, and adjusting model parameters that do not comply with safety regulations). Finally, the loss function weights are dynamically optimized based on expert system feedback to strengthen the model's learning of compliant results, avoid overfitting errors or non-compliant labels, and form a closed-loop control of "data input - model training - expert verification - parameter optimization - result feedback" to ensure that the entire training process complies with A5 ethical review requirements.

[0149] AI model training is implemented using nationally licensed chips and a compliant training platform. All open-source frameworks and components used in the training process have obtained their corresponding licenses, and copyright statements and patent citation information are fully retained, with no instances of infringement or reuse. The training environment is built using virtual devices (containers / virtual machines) with fixed random seeds and initial parameter configurations to ensure the reproducibility of the training process. Furthermore, through access control and operation log recording, risks such as data leakage and parameter tampering during training are prevented, ensuring the security and compliance of the training process.

[0150] After the model is trained, it undergoes additional third-party ethical compliance assessment and algorithm filing review to verify that the model output does not violate social morality or harm public interests. For potentially sensitive scenarios (such as public services and intelligent decision-making), a special result verification mechanism is established to ensure that the model always complies with Article 5 of the Patent Law and relevant laws and regulations in practical applications.

[0151] In summary, the data and training process used in the AI ​​model of this specification strictly comply with the relevant provisions of Article 5 of the Patent Law and the Patent Examination Guidelines (2023 Edition), and there are no violations of laws, social ethics, public interests, or illegal use of genetic resources. It fully meets the compliance requirements for patent authorization.

[0152] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for allocating wireless communication resources, characterized in that, Applied to the network side, including: Obtain the wireless parameters and environmental parameters for the current period. The wireless parameters include at least the signal round-trip propagation delay, and the environmental parameters include at least one of the following: terrain, weather, and base station load. The wireless parameters, environmental parameters, and target coverage distance are input into a pre-trained first prediction model, which outputs a predicted guard interval (GP) value. The first prediction model is trained on a deep learning model based on historical wireless parameter datasets, historical environmental parameter datasets, and corresponding historical coverage distance labels. The target coverage distance is determined based on the wireless parameters and environmental parameters. Configure the GP for the current wireless communication cycle based on the GP prediction value.

2. The method according to claim 1, characterized in that, The step of configuring the GP for the current wireless communication cycle based on the GP prediction value includes: Obtain the historical GP value of the previous communication cycle and the preset GP adjustment time interval; If the difference or rate of change between the predicted GP value and the historical GP value exceeds a preset adjustment threshold, and the time since the last GP adjustment reaches the preset GP adjustment time interval, the GP of the current wireless communication cycle is configured according to the predicted GP value. If the difference or rate of change between the predicted GP value and the historical GP value does not exceed a preset adjustment threshold, and / or the time since the last GP adjustment has not reached the preset GP adjustment time interval, the GP of the current wireless communication cycle is configured according to the historical GP value.

3. The method according to claim 2, characterized in that, The preset adjustment threshold includes at least one of a relative change threshold and an absolute change threshold; wherein the relative change threshold is 10% of the historical GP value, and the absolute change threshold is 20 microseconds.

4. The method according to any one of claims 1-3, characterized in that, The dynamic adjustment range of the GP length corresponding to the GP prediction value output by the first prediction model is configured to be from 350 microseconds to 750 microseconds, based on the target sea area coverage distance of 50 kilometers to 100 kilometers.

5. The method according to claim 1, characterized in that, After configuring the GP of the current wireless communication cycle according to the GP prediction value, the method further includes: Obtain service parameters, which include at least one of the following: uplink / downlink traffic ratio, current number of active users, ratio of various services, average channel quality, and GP of the current wireless communication cycle; The service parameters are input into the trained second prediction model to obtain the predicted uplink and downlink time slot ratio; wherein, the second prediction model is trained on the reinforcement learning algorithm based on the service parameter dataset, and the uplink and downlink time slot ratio is the ratio of the number of symbols in the uplink to the number of symbols in the downlink. Configure the uplink and downlink time slot ratio for the current wireless communication cycle based on the predicted uplink and downlink time slot ratio.

6. The method according to claim 1, characterized in that, After configuring the uplink / downlink time slot ratio for the current wireless communication cycle based on the predicted uplink / downlink time slot ratio, the method further includes: The spectrum information of available communication frequency bands is obtained, and the spectrum information is input into a third prediction model to obtain interference signal prediction results. The interference signal prediction results include at least the interference category, interference intensity, interference frequency band, and interference trend. Based on the interference signal prediction results and channel quality, the available frequency band is dynamically divided into multiple logical slices; Based on the interference signal prediction results and the wireless channel parameters of each logical slice, the target logical slice is selected through the fourth prediction model. Configure the spectrum resources for the current communication cycle based on the selection result of the fourth prediction model.

7. The method according to claim 6, characterized in that, The plurality of logical slices includes at least a first slice and a second slice, wherein: The first slice operates in a frequency band where the interference intensity is lower than a first reference threshold and the channel quality is higher than a first high-quality threshold; The second slice operates in a frequency band where the interference intensity is lower than the second reference threshold and the channel quality is higher than the second superior threshold; Wherein, the first benchmark threshold is less than the second benchmark threshold, and the first quality threshold is greater than the second quality threshold.

8. The method according to claim 7, characterized in that, The plurality of logical slices further includes a third slice and / or a fourth slice, wherein, The third slice is allocated to a frequency band with stable characteristics and corresponding suppression algorithms for persistent interference. When a service is scheduled to the third slice, the corresponding interference suppression algorithm is activated synchronously. The fourth slice is configured to operate in an anti-interference communication mode, which includes at least one of using extremely narrow bandwidth, lowest order modulation, and strongest error correction coding.

9. The method according to claim 7, characterized in that, After configuring the spectrum resources for the current communication cycle based on the selection result of the fourth prediction model, the method further includes: Obtain wireless resource scheduling parameters, which include at least one of the following: real-time load of available frequency bands, terminal signal quality of available frequency bands, service demand type, terminal moving speed, and terminal carrier aggregation capability; The radio resource scheduling parameters are input into the trained fifth prediction model to obtain a resource scheduling decision. The resource scheduling decision is used to instruct at least one of the following operations: allocating an initial frequency band, triggering a frequency band handover, configuring carrier aggregation (CA), and triggering supplementary uplink (SUL) handover. The fifth prediction model is obtained by training a deep reinforcement learning algorithm based on the radio resource scheduling parameter dataset. Wireless resources are scheduled according to the resource scheduling decision.

10. The method according to claim 9, characterized in that, The decision conditions for configuring carrier aggregation (CA) in the fifth prediction model include: the signal-to-interference-plus-noise ratio (SINR) of the terminal on the secondary carrier is higher than 3dB, and / or the channel quality indicator (CQI) is higher than 6, and / or the terminal's service bandwidth requirement is higher than 15Mbps.

11. The method according to claim 9, characterized in that, The decision conditions for the fifth prediction model to trigger supplementary uplink (SUL) handover include: detecting that the terminal is carrying out a high-volume uplink service of a preset type, and that its uplink signal quality in the current service frequency band is lower than a preset threshold.

12. The method according to claim 1, characterized in that, After scheduling radio resources based on the resource scheduling prediction result, the method further includes: Obtain the prediction outputs of the first to the fifth prediction models; Determine whether there is a conflict between the prediction outputs of the first prediction model to the fifth prediction model; In the event of a conflict, an avoidance operation is performed according to a preset global arbitration rule, which includes one of the following: prioritizing link stability, prioritizing load balancing, and prioritizing maximizing throughput.

13. The method according to claim 12, characterized in that, After scheduling radio resources based on the resource scheduling prediction result, the method further includes: Obtain key quality indicators of the network system, wherein the key quality indicators include one of total throughput, average latency, and disconnection rate; Input the key quality indicators into the first optimization model to obtain hyperparameter optimization suggestions for one or more of the first to fifth prediction models. The hyperparameters include at least one of the following: learning rate, discount factor, and optimizer type. Based on the hyperparameter optimization recommendations, set the hyperparameters of one or more prediction models from the first to the fifth prediction models.

14. A 5G ultra-long-range coverage system for marine scenarios, characterized in that, include: The parameter acquisition module is used to acquire wireless parameters, environmental parameters, service parameters and spectrum information, wherein the environmental parameters include at least parameters characterizing the propagation characteristics in the sea area; The intelligent time slot scheduling module is connected to the parameter acquisition module and is used to dynamically configure the protection interval (GP) and / or the time division duplex (TDD) time slot ratio based on the wireless parameters and the environmental parameters. A cognitive radio anti-interference module, connected to the parameter acquisition module, is used to perform spectrum sensing and intelligent frequency selection scheduling based on the spectrum information; A multi-band intelligent coordination module, connected to the parameter acquisition module, is used for multi-band load balancing, carrier aggregation (CA), and supplementary uplink scheduling (SUL). The collaborative optimization center is communicatively connected to the intelligent time slot scheduling module, the cognitive radio anti-interference module, and the multi-band intelligent collaborative module, respectively. It is used to aggregate the scheduling decision information of each module, perform conflict arbitration, and perform closed-loop adaptive optimization of the parameters of each module.

Images