Flexible strategy wireless communication method based on artificial intelligence

Abstract

The invention discloses a flexible strategy wireless communication method based on artificial intelligence, which dynamically optimizes wireless communication parameters by establishing an intelligent decision-making mechanism in base station equipment, integrates collected original state parameters into a standardized vector through continuous monitoring of multi-module cooperative work, and realizes the flexible strategy wireless communication based on the standardized vector. And inputting the state vector into a pre-configured deep learning model for strategy decision, generating an action instruction containing power control, a modulation coding scheme and a resource allocation strategy, analyzing and implementing parameter adjustment by a special actuator, and carrying out data transmission and maintaining configuration by a base station based on optimized parameters until a next update period. According to the method, the limitation of a traditional static strategy is broken through, self-adaption to a dynamic wireless environment is realized through an end-to-end intelligent decision process, the resource utilization efficiency and the service quality of the system are remarkably improved, and meanwhile, stable performance is kept in a complex and changeable communication scene.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, specifically relating to a flexible strategy wireless communication method based on artificial intelligence. Background Technology

[0002] In traditional wireless communication systems, network resource allocation typically relies on predefined fixed strategies or static adjustment mechanisms based on simple thresholds. These methods often assume ideal channel conditions and steady-state service loads, and use uniform parameter settings to cope with diverse real-world communication scenarios. Due to a lack of real-time perception and understanding of the dynamic wireless environment, such methods are difficult to effectively adapt to rapid fluctuations in channel conditions, random changes in user mobility, and sudden fluctuations in service demands, resulting in low system resource utilization and unstable service quality.

[0003] With the development of artificial intelligence technology, existing research attempts to introduce data-driven methods into the field of wireless communication, optimizing network parameters by analyzing historical communication data. However, most of these methods rely on offline model training and fixed decision rules, which cannot respond to real-time changes in the communication environment. At the same time, existing solutions often require complex multi-source data fusion and cumbersome data preprocessing processes, which not only increase system overhead but also introduce significant decision delays, making it difficult to meet the stringent requirements of modern wireless communication for low latency and high reliability.

[0004] Furthermore, current AI-based communication optimization methods generally suffer from rigid strategies. Their decision-making process often relies on a closed-loop control architecture, lacking the ability to flexibly adapt to unknown communication scenarios. These methods typically require complete environmental models and accurate data labeling, and face challenges such as insufficient model generalization ability and lagging policy updates in actual deployment, which limits their practical value in complex wireless environments. Summary of the Invention

[0005] According to a first aspect of the present invention, the present invention claims protection for a flexible strategy wireless communication method based on artificial intelligence, applied to base station equipment in a wireless communication network, comprising: S1, Sensing the wireless environment status: The base station equipment continuously measures and collects wireless channel status information, system load status, and user equipment mobility status through its built-in radio frequency receiver, signal processor, and load monitoring module. The status information includes downlink channel status information and uplink channel status information. S2, prepare state data, integrate the wireless channel state information, system load state and user equipment mobility state collected in S1 into a comprehensive state vector, perform dimension alignment and format unification on the comprehensive state vector to ensure that the comprehensive state vector matches the input requirements of the artificial intelligence strategy model; S3 queries the artificial intelligence policy, inputs the comprehensive state vector generated in S2 into the pre-stored artificial intelligence policy model based on a deep learning architecture in the base station equipment, and outputs the policy action vector through forward propagation calculation of the internal neural network; S4, adjust communication parameters. The base station equipment's policy executor parses the policy action vector output in S3, adjusts the gain setting of the transmit power amplifier according to the power control action, selects the configuration parameters of the modulator and encoder according to the modulation and coding action, and allocates physical layer resource blocks to specific user equipment according to the resource allocation action. The policy executor verifies whether the adjusted parameters are within the allowable range of the equipment. If they are outside the range, the default parameters are used instead. S5, Adaptive Communication: The base station equipment uses the transmit power, modulation and coding configuration and resource allocation determined in S4 to send downlink data signals to the user equipment through the antenna array and receive uplink data signals from the user equipment to complete the wireless data transmission process. The base station equipment maintains the adjusted parameter settings until the next policy update is triggered.

[0006] Furthermore, in S1, the acquisition of the downlink channel state information includes the following sub-steps: The base station equipment receives the channel state information reference signal sent by the user equipment and converts the signal into a baseband signal through a radio frequency receiver; The channel estimator of the base station equipment processes the baseband signal, calculates the channel response matrix, and extracts the channel quality indicator, precoding matrix indicator, and rank indicator from the channel response matrix; The signal quality assessment module of the base station equipment calculates the estimated signal-to-interference-plus-noise ratio based on the extracted channel quality indicator, and combines the precoding matrix indicator and rank indicator to form downlink channel state information.

[0007] Furthermore, the artificial intelligence strategy model in S3 is a convolutional neural network architecture, including an input layer, multiple convolutional layers, pooling layers, fully connected layers, and an output layer; The process of inputting the integrated state vector into the artificial intelligence policy model includes the following detailed sub-steps: The input layer receives the comprehensive state vector and reshapes it into a two-dimensional feature map, where the row and column dimensions of the two-dimensional feature map are determined according to the number and type of state parameters. Convolutional layers perform convolution operations on two-dimensional feature maps. The convolution operation uses predefined convolution kernels to extract features and generate multiple feature maps. A convolutional layer consists of multiple convolutional sub-layers, each using a different sized convolutional kernel to capture multi-scale features; The pooling layer performs a downsampling operation on the feature map output by the convolutional layer. The downsampling operation uses the max pooling method to reduce the size of the feature map while retaining key feature information. The fully connected layer flattens the downsampled feature map output by the pooling layer into a one-dimensional vector, and performs non-linear transformation through multiple fully connected neuron layers to map the features to the policy action space; The output layer generates policy action vectors based on the output of the fully connected layer. Each element in the policy action vector corresponds to a policy action. The output layer uses the Softmax function for normalization to ensure the probability distribution of actions.

[0008] Furthermore, S4 includes the following detailed sub-steps: The strategy executor extracts power control action values ​​from the strategy action vector. These power control action values ​​are either continuous or discrete values, representing the relative adjustment amount of the transmit power. The strategy executor converts the power control action value into the actual power amplifier gain setting. The conversion process is completed by querying a pre-stored power mapping table, which defines the correspondence between the action value and the gain setting. The policy executor extracts the modulation and coding action index from the policy action vector. The modulation and coding action index is an integer identifier that corresponds to a predefined combination of modulation and coding schemes. The strategy executor configures the modulator to use the corresponding modulation order, including quadrature amplitude modulation or phase shift keying, based on the modulation coding action index, and configures the encoder to use the corresponding coding rate, including Turbo code or LDPC code. The policy executor extracts resource allocation action patterns from the policy action vector. The resource allocation action patterns are binary or real number sequences that specify the allocation order and priority of time-frequency resource blocks. Based on the resource allocation action patterns, the policy executor dynamically allocates physical resource blocks to user equipment in the scheduler. The allocation process takes into account the number, location, and time interval of resource blocks. After completing parameter adjustment, the policy executor performs a parameter verification sub-step to check whether the adjusted transmit power, modulation and coding scheme, and resource allocation are compatible with the device hardware limitations and network protocols. If the parameters are invalid, the policy executor reverts to the default configuration and logs the error.

[0009] Furthermore, in S2, the integration process of the integrated state vector includes the following operations: The downlink channel state information and uplink channel state information in the wireless channel state information are respectively treated as vector sub-blocks; The current number of connected users, data transmission rate requirements, and buffer status in the system load state are treated as another vector sub-block; The user equipment's moving speed, moving direction, and Doppler frequency shift in the moving state are used as the last vector sub-block; All vector sub-blocks are concatenated sequentially to form a one-dimensional composite state vector. The total dimension of the composite state vector is equal to the sum of the dimensions of each sub-block. During the integration process, the state data preparation step also ensures that the timestamps of each state parameter are consistent to reflect the network state at the same moment.

[0010] Furthermore, in S5, when the base station equipment uses the adjusted parameters to perform wireless data transmission with the user equipment, it also performs the following auxiliary operations: The antenna array of the base station equipment uses beamforming technology to send signals to the user equipment in a directional manner according to the resource allocation operation mode; The base station equipment scheduler prioritizes allocating resources to high-priority user equipment or latency-sensitive services based on the resource allocation action mode. During data transmission, the base station equipment monitors the link status but does not adjust or provide feedback on the strategy in real time in order to maintain open-loop control characteristics.

[0011] Furthermore, in S1, the acquisition of the uplink channel state information also includes the following supplementary sub-steps: The base station equipment captures the uplink probe reference signal sent by the user equipment through the uplink receiver; The signal processor demodulates and analyzes the uplink detection reference signal, and calculates the signal-to-interference-plus-noise ratio and timing advance. The mobility management module estimates the distance changes of user equipment based on timing advance and derives the moving speed and direction by combining Doppler frequency shift.

[0012] Furthermore, the artificial intelligence strategy model in S3 also includes an attention mechanism module, which is integrated between the convolutional layer and the fully connected layer; The operation of the attention mechanism module includes: Attention weights are calculated on the feature map output by the convolutional layer, and these attention weights are assigned based on the importance of the features. The feature map is weighted and summed using attention weights to generate an enhanced feature representation; The enhanced feature representation is then fed into a fully connected layer for further processing.

[0013] Furthermore, in S4, the policy executor also supports multi-mode policy execution, including: If the power control action value indicates an increase in power, the strategy actuator gradually adjusts the power amplifier gain to avoid interference caused by sudden changes; If the modulation and coding action index corresponds to a higher-order modulation, the policy executor simultaneously adjusts the encoder parameters to improve error correction capability. If the resource allocation action mode involves multiple user devices, the policy executor uses a round-robin or priority scheduling algorithm to allocate resource blocks; Multi-mode strategy execution ensures smooth parameter adjustment and network stability.

[0014] Furthermore, the method also includes an initialization step, which is performed before the wireless environment state awareness step, and the initialization step includes: The base station equipment loads the pre-configured artificial intelligence strategy model into memory and verifies the integrity and compatibility of the model; The base station equipment calibration status monitoring module includes an RF receiver, a signal processor, and a load monitoring module to ensure measurement accuracy. The base station equipment sets the default parameters for the policy executor, including the default transmit power, default modulation and coding scheme, and default resource allocation mode; The initialization step is performed only once when the base station equipment is started or reset, providing basic configuration for subsequent steps.

[0015] This invention discloses a flexible strategy wireless communication method based on artificial intelligence. It dynamically optimizes wireless communication parameters by establishing an intelligent decision-making mechanism in the base station equipment. Through continuous monitoring via multi-module collaboration, the collected raw state parameters are integrated into a standardized vector. This state vector is then input into a pre-configured deep learning model for strategy decision-making, generating action instructions that include power control, modulation and coding schemes, and resource allocation strategies. A dedicated actuator parses and implements the parameter adjustments. The base station then transmits data based on the optimized parameters and maintains the configuration until the next update cycle. This invention overcomes the limitations of traditional static strategies, achieving autonomous adaptation to dynamic wireless environments through an end-to-end intelligent decision-making process. This significantly improves system resource utilization efficiency and service quality while maintaining stable performance in complex and ever-changing communication scenarios. Attached Figure Description

[0016] Figure 1 The present invention requests protection for a flowchart of a flexible strategy wireless communication method based on artificial intelligence; Figure 2 The present invention claims a second flowchart of a flexible strategy wireless communication method based on artificial intelligence. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0018] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0019] According to the first embodiment of the present invention, referring to Figure 1 This invention claims protection for a flexible strategy wireless communication method based on artificial intelligence, applied to base station equipment in a wireless communication network, comprising: S1, Sensing the wireless environment status: The base station equipment continuously measures and collects wireless channel status information, system load status, and user equipment mobility status through its built-in radio frequency receiver, signal processor, and load monitoring module. The status information includes downlink channel status information and uplink channel status information. S2, prepare state data, integrate the wireless channel state information, system load state and user equipment mobility state collected in S1 into a comprehensive state vector, perform dimension alignment and format unification on the comprehensive state vector to ensure that the comprehensive state vector matches the input requirements of the artificial intelligence strategy model; S3 queries the artificial intelligence policy, inputs the comprehensive state vector generated in S2 into the pre-stored artificial intelligence policy model based on a deep learning architecture in the base station equipment, and outputs the policy action vector through forward propagation calculation of the internal neural network; S4, adjust communication parameters. The base station equipment's policy executor parses the policy action vector output in S3, adjusts the gain setting of the transmit power amplifier according to the power control action, selects the configuration parameters of the modulator and encoder according to the modulation and coding action, and allocates physical layer resource blocks to specific user equipment according to the resource allocation action. The policy executor verifies whether the adjusted parameters are within the allowable range of the equipment. If they are outside the range, the default parameters are used instead. S5, Adaptive Communication: The base station equipment uses the transmit power, modulation and coding configuration and resource allocation determined in S4 to send downlink data signals to the user equipment through the antenna array and receive uplink data signals from the user equipment to complete the wireless data transmission process. The base station equipment maintains the adjusted parameter settings until the next policy update is triggered.

[0020] In this embodiment, the base station equipment continuously measures and collects wireless channel status information, system load status, and user equipment mobility status through its built-in radio frequency receiver, signal processor, and load monitoring module. Specifically, the state information of the wireless channel includes downlink channel state information and uplink channel state information. The process of acquiring downlink channel state information involves the base station equipment decoding the channel state information reference signal transmitted by the user equipment. After the signal is received by the radio frequency receiver, the signal processor performs baseband conversion and channel estimation to extract the channel quality indicator, precoding matrix indicator, and rank indicator. The channel quality indicator is obtained by calculating the received signal strength to noise-to-interference ratio (NOI), the precoding matrix indicator is determined by analyzing the eigenvectors of the channel response matrix, and the rank indicator is obtained by evaluating the rank value of the channel matrix. The process of acquiring uplink channel state information involves the base station equipment analyzing the uplink sounding reference signal transmitted by the user equipment. After the signal is captured by the radio frequency receiver, the signal processor performs demodulation and analysis to extract the signal-to-interference-plus-noise ratio (SNR) and timing advance. The SNR is calculated by measuring the ratio of the uplink signal power to the background interference, and the timing advance is obtained by estimating the difference between the signal transmission delay and the standard time offset. The process of acquiring system load status involves the load monitoring module counting the number of currently connected users, monitoring data transmission rate requirements, and checking buffer status. The number of currently connected users is obtained by maintaining and updating the user equipment connection list in real time. The data transmission rate requirements are obtained by analyzing the service demand reported by user equipment and historical transmission patterns. The buffer status is obtained by checking the queue length and overflow status of data packets to be transmitted in the base station equipment's data buffer. The process of acquiring user equipment mobility status involves the signal processor and mobility management module estimating the user equipment's mobility speed, direction of movement, and Doppler shift. Mobility speed is derived by analyzing the rate of change of uplink signal frequency offset and timing advance. Direction of movement is determined by comparing the phase difference and angle of arrival of signals received by multiple antennas. Doppler shift is calculated by measuring the drift of the uplink carrier frequency. In the wireless environment status perception step, the base station equipment performs measurements and data collection periodically. The period length is dynamically determined according to the network configuration, such as adjusting the period interval based on changes in network load or user equipment density, but no data preprocessing operations, such as filtering, normalization, or cleaning, are involved.

[0021] During state data preparation, the wireless channel state information, system load state, and user equipment mobility state collected in the wireless environment state awareness step are integrated into a comprehensive state vector. The integration process involves treating each state parameter as an independent element of the vector while preserving its original numerical format and units.

[0022] For example, channel quality indicators are expressed in decibels, precoding matrix indicators in integer indices, rank indicators in integer values, signal-to-interference-plus-noise ratios in decibels, timing lead in microseconds, the number of currently connected users in integer values, data transmission rate requirements in bits per second, buffer status in percentages, movement speed in meters per second, movement direction in radians, and Doppler shift in Hertz. The state data preparation steps also include dimensional alignment and format unification of the integrated state vector to ensure it matches the input requirements of the artificial intelligence strategy model. Dimensional alignment involves checking the dimensional consistency of each state parameter, such as ensuring all parameters have the same sampling time point, and adjusting the vector length through padding or truncation. Format unification involves converting parameters of different data types into a uniform numerical representation, such as converting Boolean values ​​to 0 or 1, and enumerated values ​​to consecutive integers. The state data preparation steps do not involve any data scaling, normalization, filtering, or cleaning; only basic structured organization is performed.

[0023] When querying artificial intelligence strategies, the comprehensive state vector generated during state data preparation is input into the artificial intelligence strategy model pre-stored in the base station equipment. The artificial intelligence strategy model is a pre-configured model based on a deep learning architecture, with multiple hidden layers and output layers, which can map the comprehensive state vector to the strategy action space.

[0024] After receiving the integrated state vector, the AI ​​policy model outputs a policy action vector through forward propagation computation via its internal neural network. The forward propagation computation includes receiving the vector at the input layer, then sequentially passing it through a weighted summation in the hidden layers, activation function transformation, and a linear combination at the output layer. The policy action vector includes power control actions, modulation and coding actions, and resource allocation actions. Power control actions represent adjustments to the transmit power, expressed in decibels and milliwatts (dW) or relative proportions; modulation and coding actions represent the selection of modulation scheme and coding rate, expressed as predefined index values; and resource allocation actions represent the allocation pattern of time-frequency resource blocks, expressed as binary sequences or arrays of real numbers. In the AI ​​policy query step, the model inference process does not involve any online learning, parameter updates, or feedback mechanisms; it only uses pre-stored model parameters for forward computation.

[0025] When communication parameters are adjusted, the base station equipment's policy executor parses the policy action vector output in the AI ​​policy query step, adjusts the gain setting of the transmit power amplifier according to the power control action, selects the configuration parameters of the modulator and encoder according to the modulation and coding action, and allocates physical layer resource blocks to specific user equipment according to the resource allocation action. The policy executor first parses the policy action vector, extracts the power control action value, and converts it into the actual power amplifier gain setting by querying a pre-stored power mapping table. The power mapping table defines the correspondence between action values ​​and gain settings; for example, a specific action value corresponds to a specific gain decibel value. Next, the policy executor extracts the modulation and coding action index and configures the modulator and encoder by querying a pre-stored modulation and coding scheme table. The modulation and coding scheme table defines the correspondence between index values ​​and modulation methods such as quadrature amplitude modulation or phase shift keying and coding rates such as Turbo codes or LDPC codes.

[0026] The policy executor extracts resource allocation action patterns and dynamically allocates physical resource blocks to user equipment via the scheduler. Resource allocation involves determining the number, location, and time interval of time-frequency resource blocks, for example, allocating resources based on priority identifiers in the action patterns. In the communication parameter adjustment step, the policy executor also verifies whether the adjusted parameters are within the device's allowable range, such as checking whether the transmit power exceeds the maximum limit, whether the modulation method is supported, and whether there are resource allocation conflicts. If the parameters are invalid, the policy executor reverts to default parameters and logs errors for subsequent analysis.

[0027] In adaptive communication, the base station equipment uses the transmit power, modulation and coding configuration, and resource allocation determined in the communication parameter adjustment step to send downlink data signals to the user equipment via the antenna array and receive uplink data signals from the user equipment, completing the wireless data transmission process. The downlink data signal transmission process includes: the baseband processor generating data frames, the modulator modulating the signal according to the configured modulation scheme, the power amplifier amplifying the signal according to the gain setting, and the antenna array using beamforming technology to transmit the signal directionally. The uplink data signal reception process includes: the antenna array receiving the signal, the low-noise amplifier amplifying the signal, the demodulator demodulating the signal according to the configured demodulation scheme, and the baseband processor decoding the data. In the adaptive communication step, the base station equipment maintains the adjusted parameter settings until the next policy update is triggered. The policy update trigger is based on a periodic timer or event-driven mechanism, but does not involve real-time parameter adjustment or closed-loop control.

[0028] Furthermore, in S1, the acquisition of the downlink channel state information includes the following sub-steps: The base station equipment receives the channel state information reference signal sent by the user equipment and converts the signal into a baseband signal through a radio frequency receiver; The channel estimator of the base station equipment processes the baseband signal, calculates the channel response matrix, and extracts the channel quality indicator, precoding matrix indicator, and rank indicator from the channel response matrix; The signal quality assessment module of the base station equipment calculates the estimated signal-to-interference-plus-noise ratio based on the extracted channel quality indicator, and combines the precoding matrix indicator and rank indicator to form downlink channel state information.

[0029] In this embodiment, the base station device receives the channel state information reference signal sent by the user equipment and converts the signal into a baseband signal through a radio frequency receiver.

[0030] Specifically, the radio frequency receiver first captures the wireless signal through the antenna array, then uses a mixer to downconvert the radio frequency signal to an intermediate frequency signal, and then uses an analog-to-digital converter to convert the intermediate frequency signal into a digital baseband signal. The sampling rate and resolution of the analog-to-digital converter are configured according to the signal bandwidth and dynamic range to ensure signal integrity. The converted baseband signal is stored in a buffer for subsequent processing.

[0031] The channel estimator of the base station equipment processes the baseband signal, calculates the channel response matrix, and extracts the channel quality indicator, precoding matrix indicator, and rank indicator from the channel response matrix. The channel estimator uses a least squares or least mean square error algorithm to estimate the channel response matrix, which represents the channel gain and phase on multiple antennas and subcarriers. When extracting the channel quality indicator, the channel estimator calculates the average signal-to-noise ratio (SNR) of each subcarrier and maps it to a standard channel quality indicator value. When extracting the precoding matrix indicator, the channel estimator performs singular value decomposition on the channel response matrix and selects the optimal precoding matrix index. When extracting the rank indicator, the channel estimator evaluates the rank of the channel matrix and determines the transmission layer number based on the eigenvalue distribution. All these extraction operations involve matrix operations and logical judgments of the digital signal processor.

[0032] The signal quality assessment module of the base station equipment calculates the estimated signal-to-interference-plus-noise ratio (SNR) based on the extracted channel quality indicator (CMI), and combines it with the precoding matrix indicator and rank indicator to form downlink channel state information. The module first processes the CMI value using a weighted average to obtain the overall SNR estimate. Then, it combines the precoding matrix indicator and rank indicator with the SNR estimate into a data structure, such as an array or list, representing the downlink channel state information. This data structure is then transmitted to the state data preparation step for further use.

[0033] Furthermore, the artificial intelligence strategy model in S3 is a convolutional neural network architecture, including an input layer, multiple convolutional layers, pooling layers, fully connected layers, and an output layer; The process of inputting the integrated state vector into the artificial intelligence policy model includes the following detailed sub-steps: The input layer receives the comprehensive state vector and reshapes it into a two-dimensional feature map, where the row and column dimensions of the two-dimensional feature map are determined according to the number and type of state parameters. Convolutional layers perform convolution operations on two-dimensional feature maps. The convolution operation uses predefined convolution kernels to extract features and generate multiple feature maps. A convolutional layer consists of multiple convolutional sub-layers, each using a different sized convolutional kernel to capture multi-scale features; The pooling layer performs a downsampling operation on the feature map output by the convolutional layer. The downsampling operation uses the max pooling method to reduce the size of the feature map while retaining key feature information. The fully connected layer flattens the downsampled feature map output by the pooling layer into a one-dimensional vector, and performs non-linear transformation through multiple fully connected neuron layers to map the features to the policy action space; The output layer generates policy action vectors based on the output of the fully connected layer. Each element in the policy action vector corresponds to a policy action. The output layer uses the Softmax function for normalization to ensure the probability distribution of actions.

[0034] In this embodiment, inputting the comprehensive state vector into the artificial intelligence policy model includes the following detailed sub-steps: The input layer receives the synthesized state vector and reshapes it into a two-dimensional feature map. This reshaping process involves rearranging the elements of the one-dimensional vector into a two-dimensional array, with the row and column dimensions determined by the number and type of state parameters. For example, if the state parameters include channel state, load state, and mobility state, the number of rows might correspond to the number of parameter categories, and the number of columns might correspond to the number of parameters per category. After reshaping, the two-dimensional feature map is stored in memory for processing by subsequent layers.

[0035] Convolutional layers perform convolution operations on two-dimensional feature maps. These operations use predefined convolutional kernels to extract features, generating multiple feature maps. A convolutional layer consists of multiple convolutional sub-layers, each using a different sized kernel to capture multi-scale features. For example, the first sub-layer might use a smaller kernel to capture local features, while the second sub-layer might use a larger kernel to capture global features. The convolution operation involves sliding the kernel over the two-dimensional feature map, calculating the sum of the dot products at each location, and introducing non-linearity through activation functions such as ReLU. The resulting feature maps represent different feature patterns of the input data, such as channel variation patterns or load fluctuation patterns.

[0036] Pooling layers perform downsampling on the feature map output by convolutional layers. This downsampling operation uses max pooling to reduce the size of the feature map while preserving key feature information. Max pooling involves dividing the feature map into non-overlapping regions and selecting the maximum value from each region as the output. The pooling size and stride of the pooling layer are configured based on the feature map size; for example, a 2×2 pooling window can halve the feature map size. After downsampling, the dimensionality of the feature map is reduced, but important features such as peak signal strength or anomalous load points are preserved.

[0037] Fully connected layers flatten the downsampled feature map output from the pooling layers into a one-dimensional vector, and then perform a non-linear transformation through multiple fully connected neuron layers to map the features to the policy action space. The flattening operation involves concatenating all elements of the two-dimensional feature map into a one-dimensional vector. Each fully connected neuron layer consists of multiple neurons, each performing a weighted sum of the input vector and outputting a value through an activation function such as Sigmoid or Tanh. Multiple fully connected layers are connected sequentially, progressively reducing the vector dimension, ultimately outputting a vector that matches the dimension of the policy action space.

[0038] The output layer generates policy action vectors based on the output of the fully connected layer. Each element in the policy action vector corresponds to a policy action. The output layer uses the Softmax function for normalization to ensure the probability distribution of actions. The Softmax function converts the output values ​​of the fully connected layer into probability values, representing the probability of each policy action being selected. Power control actions, modulation and coding actions, and resource allocation actions in the policy action vector each correspond to different probability distributions. Finally, the action with the highest probability or the sampling action is selected as the output.

[0039] Furthermore, S4 includes the following detailed sub-steps: The strategy executor extracts power control action values ​​from the strategy action vector. These power control action values ​​are either continuous or discrete values, representing the relative adjustment amount of the transmit power. The strategy executor converts the power control action value into the actual power amplifier gain setting. The conversion process is completed by querying a pre-stored power mapping table, which defines the correspondence between the action value and the gain setting. The policy executor extracts the modulation and coding action index from the policy action vector. The modulation and coding action index is an integer identifier that corresponds to a predefined combination of modulation and coding schemes. The strategy executor configures the modulator to use the corresponding modulation order, including quadrature amplitude modulation or phase shift keying, based on the modulation coding action index, and configures the encoder to use the corresponding coding rate, including Turbo code or LDPC code. The policy executor extracts resource allocation action patterns from the policy action vector. The resource allocation action patterns are binary or real number sequences that specify the allocation order and priority of time-frequency resource blocks. Based on the resource allocation action patterns, the policy executor dynamically allocates physical resource blocks to user equipment in the scheduler. The allocation process takes into account the number, location, and time interval of resource blocks. After completing parameter adjustment, the policy executor performs a parameter verification sub-step to check whether the adjusted transmit power, modulation and coding scheme, and resource allocation are compatible with the device hardware limitations and network protocols. If the parameters are invalid, the policy executor reverts to the default configuration and logs the error.

[0040] In this embodiment, the strategy actuator extracts power control action values ​​from the strategy action vector. These power control action values ​​are continuous or discrete numerical values ​​representing a relative adjustment in transmit power. The strategy actuator converts the power control action values ​​into actual power amplifier gain settings. This conversion process is accomplished by querying a pre-stored power mapping table. The power mapping table is a lookup table that defines the correspondence between action values ​​and gain settings; for example, a specific range of action values ​​corresponds to a specific range of gain decibel values. During the conversion, the strategy actuator searches the table for a matching gain setting based on the action value. If the action value falls within a certain range, linear interpolation is used to calculate the gain value. The converted gain setting is then sent to the power amplifier controller to adjust the amplifier's bias voltage or current to change the output power.

[0041] The policy executor extracts modulation-coding action indices from the policy action vector. Each modulation-coding action index is an integer identifier corresponding to a predefined combination of modulation and coding schemes. Based on these indices, the policy executor configures the modulator to use the appropriate modulation order, including quadrature amplitude modulation (QAM) or phase shift keying (PPS), and configures the encoder to use the appropriate coding rate, including Turbo code or LDPC code. The configuration process involves querying a pre-stored modulation-coding scheme table, which defines the correspondence between index values ​​and modulation parameters such as modulation type and order, and coding parameters such as code rate and constraint length. The policy executor inputs the index values ​​into the table to obtain the specific parameters, and then sets the modulator's constellation diagram and the encoder's generator polynomial. For example, for QAM, amplitude and phase levels are set; for Turbo code, the interleaver and component encoder are set.

[0042] The policy executor extracts resource allocation action patterns from the policy action vector. Each resource allocation action pattern is a binary or real number sequence that specifies the allocation order and priority of time-frequency resource blocks. Based on the resource allocation action patterns, the policy executor dynamically allocates physical resource blocks to user equipment in the scheduler. The allocation process considers the number, location, and time interval of resource blocks. For example, the resource allocation action pattern might instruct the priority allocation of high-frequency resource blocks to latency-sensitive users, or the balanced allocation of resource blocks among multiple users. The policy executor parses the action patterns to generate resource allocation instructions, which include the resource block index, allocation time, and user equipment identifier. The scheduler updates the resource allocation table according to the instructions and notifies the physical layer controller to execute the allocation.

[0043] After completing parameter adjustments, the policy executor performs a parameter verification sub-step to check whether the adjusted transmit power, modulation and coding scheme, and resource allocation are compatible with device hardware limitations and network protocols. Parameter verification includes: checking whether the transmit power is within the amplifier's minimum and maximum power range; checking whether the modulation scheme supports the current channel bandwidth; checking whether the coding rate matches the bit error rate requirement; and checking whether the resource allocation conflicts with other users or exceeds the total resource capacity. If the parameters are invalid, the policy executor reverts to the default configuration, such as using the default transmit power, default modulation and coding scheme, and default resource allocation mode, and logs an error, including the invalid parameter value, verification time, and error type.

[0044] Furthermore, in S2, the integration process of the integrated state vector includes the following operations: The downlink channel state information and uplink channel state information in the wireless channel state information are respectively treated as vector sub-blocks; The current number of connected users, data transmission rate requirements, and buffer status in the system load state are treated as another vector sub-block; The user equipment's moving speed, moving direction, and Doppler frequency shift in the moving state are used as the last vector sub-block; All vector sub-blocks are concatenated sequentially to form a one-dimensional composite state vector. The total dimension of the composite state vector is equal to the sum of the dimensions of each sub-block. During the integration process, the state data preparation step also ensures that the timestamps of each state parameter are consistent to reflect the network state at the same moment.

[0045] In this embodiment, the downlink channel state information and uplink channel state information in the wireless channel state information are respectively treated as vector sub-blocks. The downlink channel state information sub-block includes parameters such as channel quality indicator, precoding matrix indicator, and rank indicator, with each parameter serving as an element of the sub-block. The uplink channel state information sub-block includes parameters such as signal-to-interference-plus-noise ratio and timing advance, with each parameter serving as an element of the sub-block. The construction of the sub-blocks involves arranging the parameters in a fixed order, for example, first the channel quality indicator, then the precoding matrix indicator, and finally the rank indicator.

[0046] The system load status, including the number of currently connected users, data transfer rate requirements, and buffer status, is treated as another vector sub-block. The number of currently connected users is the first element of this sub-block, the data transfer rate requirements are the second element, and the buffer status is the third element. The data transfer rate requirements may include uplink and downlink rates, each treated as a separate element. The buffer status may include the queue lengths of multiple buffers, with each length treated as an element.

[0047] The user equipment's movement speed, direction of movement, and Doppler frequency shift are used as the last vector sub-block. The movement speed is the first element of the sub-block, the movement direction is the second element, and the Doppler frequency shift is the third element. The movement direction may be represented by two-dimensional or three-dimensional coordinates, with each coordinate component as an independent element.

[0048] All vector sub-blocks are sequentially concatenated into a one-dimensional composite state vector, the total dimension of which equals the sum of the dimensions of each sub-block. The concatenation process involves sequentially connecting the downlink channel state information sub-block, the uplink channel state information sub-block, the system load state sub-block, and the user equipment mobility state sub-block. For example, if each sub-block has 3 elements, the composite state vector will have 12 elements. After concatenation, the vector indices correspond to the parameter categories, facilitating model processing.

[0049] During the integration process, the state data preparation step also ensures that the timestamps of each state parameter are consistent to reflect the network state at the same moment. Timestamp consistency involves checking the acquisition time of each parameter. If the time deviation exceeds a threshold, the parameter value is adjusted using interpolation or prediction methods, but this is limited to basic time alignment and does not involve complex data preprocessing.

[0050] Furthermore, in S5, when the base station equipment uses the adjusted parameters to perform wireless data transmission with the user equipment, it also performs the following auxiliary operations: The antenna array of the base station equipment uses beamforming technology to send signals to the user equipment in a directional manner according to the resource allocation operation mode; The base station equipment scheduler prioritizes allocating resources to high-priority user equipment or latency-sensitive services based on the resource allocation action mode. During data transmission, the base station equipment monitors the link status but does not adjust or provide feedback on the strategy in real time in order to maintain open-loop control characteristics.

[0051] Furthermore, referring to Figure 2 In S1, the acquisition of the uplink channel state information further includes the following supplementary sub-steps: The base station equipment captures the uplink probe reference signal sent by the user equipment through the uplink receiver; The signal processor demodulates and analyzes the uplink detection reference signal, and calculates the signal-to-interference-plus-noise ratio and timing advance. The mobility management module estimates the distance changes of user equipment based on timing advance and derives the moving speed and direction by combining Doppler frequency shift.

[0052] In this embodiment, the base station equipment captures the uplink probe reference signal transmitted by the user equipment via an uplink receiver. The uplink receiver is part of a radio frequency receiver specifically designed to handle uplink frequency band signals. The capture process involves tuning the receiver to the uplink frequency, using a bandpass filter to remove out-of-band noise, and adjusting the signal level using automatic gain control.

[0053] The signal processor demodulates and analyzes the uplink detection reference signal, calculating the signal-to-interference-plus-noise ratio (SNR) and timing advance. The demodulation process involves extracting signal features using matched filters or correlators, while the analysis process involves measuring signal power, interference power, and noise power, calculating the SNR to obtain the signal-to-interference-plus-noise ratio. The timing advance is obtained by comparing the signal arrival time with the expected time difference, and an accuracy is ensured using a clock synchronization algorithm.

[0054] The mobility management module estimates the distance change of user equipment based on timing advance and derives the movement speed and direction by combining this with Doppler frequency shift. Distance change estimation involves converting the timing advance into a distance difference, calculated based on the signal propagation speed. Movement speed derivation involves differentiating the distance change and determining the direction by combining the sign of the Doppler frequency shift. The Doppler frequency shift is obtained through spectral analysis, for example, by calculating the frequency offset using a Fast Fourier Transform.

[0055] Furthermore, the artificial intelligence strategy model in S3 also includes an attention mechanism module, which is integrated between the convolutional layer and the fully connected layer; The operation of the attention mechanism module includes: Attention weights are calculated on the feature map output by the convolutional layer, and these attention weights are assigned based on the importance of the features. The feature map is weighted and summed using attention weights to generate an enhanced feature representation; The enhanced feature representation is then fed into a fully connected layer for further processing.

[0056] In this embodiment, the attention mechanism module is integrated between the convolutional layer and the fully connected layer; the operation of the attention mechanism module includes: Attention weights are calculated on the feature map output by the convolutional layer, and these weights are assigned based on the importance of the features. The calculation involves inputting the feature map into a fully connected layer or a convolutional layer to generate weight scores. These weight scores are normalized to a probability distribution using a softmax function, representing the importance of each feature location. Importance is based on the relevance of the feature value to the target policy action; for example, features with high channel variation may receive high weights.

[0057] Attention weights are used to perform a weighted summation on the feature map, generating an enhanced feature representation. The weighted summation involves multiplying the value at each feature location by its corresponding weight and then summing the results to obtain a weighted feature vector. This enhanced feature representation highlights key features and suppresses irrelevant features, such as amplifying load fluctuations or weakening noise features.

[0058] The enhanced feature representation is then fed into a fully connected layer for further processing. The fully connected layer receives the weighted feature vector, performs further nonlinear transformations, and finally outputs the policy action vector.

[0059] The attention mechanism module improves the accuracy of policy generation without introducing additional training or parameter updates. Attention weights are calculated based on pre-stored model parameters and do not adjust with changes in input data.

[0060] Furthermore, in S4, the policy executor also supports multi-mode policy execution, including: If the power control action value indicates an increase in power, the strategy actuator gradually adjusts the power amplifier gain to avoid interference caused by sudden changes; If the modulation and coding action index corresponds to a higher-order modulation, the policy executor simultaneously adjusts the encoder parameters to improve error correction capability. If the resource allocation action mode involves multiple user devices, the policy executor uses a round-robin or priority scheduling algorithm to allocate resource blocks; Multi-mode strategy execution ensures smooth parameter adjustment and network stability.

[0061] In this embodiment, if the power control action value indicates an increase in power, the strategy executor gradually adjusts the power amplifier gain to avoid interference caused by sudden changes. Gradual adjustment involves breaking down the gain change into multiple small steps, such as adjusting by 0.5 dB each time, executed at regular intervals. The adjustment step size and interval are determined based on the network configuration to ensure a smooth transition.

[0062] If the modulation-coding action index corresponds to a higher-order modulation, the policy executor simultaneously adjusts the encoder parameters to improve error correction capability. For example, for higher-order quadrature amplitude modulation, the policy executor increases the encoder's redundant bits or selects a more powerful error-correcting code, such as LDPC code, instead of Turbo code. Parameter adjustments are based on predefined rules; for example, the higher the modulation order, the lower the coding rate.

[0063] If the resource allocation action pattern involves multiple user devices, the policy executor uses either a round-robin or priority scheduling algorithm to allocate resource blocks. The round-robin algorithm involves cyclically allocating resources to each user device to ensure fairness; the priority scheduling algorithm allocates resources based on the user device's service type or service quality requirements, such as prioritizing voice users.

[0064] Multi-mode strategy execution ensures smooth parameter adjustment and network stability, avoiding parameter oscillations or resource conflicts.

[0065] Furthermore, the method also includes an initialization step, which is performed before the wireless environment state awareness step, and the initialization step includes: The base station equipment loads the pre-configured artificial intelligence strategy model into memory and verifies the integrity and compatibility of the model; The base station equipment calibration status monitoring module includes an RF receiver, a signal processor, and a load monitoring module to ensure measurement accuracy. The base station equipment sets the default parameters for the policy executor, including the default transmit power, default modulation and coding scheme, and default resource allocation mode; The initialization step is performed only once when the base station equipment is started or reset, providing basic configuration for subsequent steps.

[0066] In this embodiment, the base station device loads a pre-configured artificial intelligence strategy model into memory and verifies the model's integrity and compatibility. The loading process involves reading the model file, including the model architecture, weights, and bias parameters, from non-volatile memory such as flash memory. The verification process includes checking the model file checksum, confirming input-output dimension matching, and testing the model's inference capabilities. If verification fails, the base station device uses a backup model or the default strategy.

[0067] The base station equipment calibration status monitoring module, including an RF receiver, a signal processor, and a load monitoring module, ensures measurement accuracy. Calibrating the RF receiver involves adjusting the local oscillator frequency, calibrating the analog-to-digital converter offset and gain; calibrating the signal processor involves setting processing algorithm parameters such as filter coefficients; and calibrating the load monitoring module involves initializing counters and service quality thresholds.

[0068] The base station equipment sets default parameters for the policy executor, including default transmit power, default modulation and coding scheme, and default resource allocation mode. These default parameters are determined based on network standards or manufacturer presets; for example, the default transmit power is an intermediate value, the default modulation and coding scheme is QPSK with a 1 / 2 code rate, and the default resource allocation mode is uniform allocation.

[0069] 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, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0070] Furthermore, 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. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0071] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. An artificial intelligence based flexible policy wireless communication method applied to a base station device in a wireless communication network, characterized by, Comprise: S1, sensing the state of the wireless environment, the base station device continuously measures and collects the state information of the wireless channel, the system load state and the user equipment movement state through its built-in radio frequency receiver, signal processor and load monitoring module, the state information includes downlink channel state information and uplink channel state information; S2, preparing state data, integrating the wireless channel state information, system load state and user equipment movement state collected in S1 into a comprehensive state vector, performing dimension alignment and format unification on the comprehensive state vector, and ensuring that the comprehensive state vector matches the input requirements of the artificial intelligence strategy model; S3, querying the artificial intelligence strategy, inputting the comprehensive state vector generated in S2 into the pre-stored artificial intelligence strategy model based on the deep learning architecture in the base station device, outputting the strategy action vector through the forward propagation calculation of the internal neuron network; S4, adjusting the communication parameters, the strategy executor of the base station device parses the strategy action vector output in S3, and adjusts the gain setting of the transmit power amplifier according to the power control action, selects the configuration parameters of the modulator and encoder according to the modulation and coding action, and allocates physical layer resource blocks to specific user equipment according to the resource allocation action. The strategy executor verifies whether the adjusted parameters are within the device allowed range, and if they are out of range, uses the default parameters instead; S5, adaptive communication, the base station device uses the transmit power, modulation and coding configuration and resource allocation determined in S4 to send downlink data signals to user equipment through an antenna array and receive uplink data signals from user equipment, complete the wireless data transmission process, and the base station device maintains the adjusted parameter settings until the next strategy update trigger.

2. The method of claim 1, wherein, In S1, the acquisition of the downlink channel state information comprises the following sub-steps: The base station device receives the channel state information reference signal sent by the user equipment, and converts the signal into a baseband signal through the radio frequency receiver; The channel estimator of the base station device processes the baseband signal, calculates the channel response matrix, and extracts the channel quality indicator, precoding matrix indicator and rank indicator from the channel response matrix; The signal quality evaluation module of the base station device calculates the signal to interference plus noise ratio estimate value based on the extracted channel quality indicator, and forms the downlink channel state information in combination with the precoding matrix indicator and the rank indicator.

3. The method of claim 1, wherein, The artificial intelligence strategy model in S3 is a convolutional neural network architecture, including an input layer, multiple convolutional layers, pooling layers, fully connected layers and an output layer; The input of the comprehensive state vector into the artificial intelligence strategy model comprises the following detailed sub-steps: The input layer receives the comprehensive state vector and reshapes the comprehensive state vector into a two-dimensional feature map, wherein the row and column dimensions of the two-dimensional feature map are determined according to the number and type of state parameters; The convolutional layer performs convolution operation on the two-dimensional feature map, and the convolution operation uses a pre-defined convolution kernel for feature extraction to generate multiple feature maps; The convolutional layer includes multiple convolutional sub-layers, each of which uses a convolution kernel of different size to capture multi-scale features; The pooling layer performs a down-sampling operation on the feature map output by the convolutional layer, the down-sampling operation adopts a max-pooling method, and the size of the feature map is reduced while the key feature information is retained; The fully connected layer flattens the down-sampled feature map output by the pooling layer into a one-dimensional vector, and performs a non-linear transformation through multiple fully connected neuron layers to map the features to a policy action space; The output layer generates a policy action vector based on the output of the fully connected layer, each element in the policy action vector corresponds to a policy action, and the output layer uses a Softmax function for normalization processing to ensure the action probability distribution.

4. The method of claim 1, wherein, The S4 includes the following detailed sub-steps: The policy executor extracts a power control action value from the policy action vector, the power control action value is a continuous or discrete numerical value representing the relative adjustment amount of the transmit power; The policy executor converts the power control action value into an actual power amplifier gain setting, the conversion process is completed by querying a pre-stored power mapping table, and the power mapping table defines the correspondence between the action value and the gain setting; The policy executor extracts a modulation and coding action index from the policy action vector, the modulation and coding action index is an integer identifier corresponding to a predefined modulation and coding scheme combination; The policy executor configures the modulator to use the corresponding modulation order, including quadrature amplitude modulation or phase shift keying, and configures the encoder to use the corresponding coding rate, including Turbo code or LDPC code, according to the modulation and coding action index; The policy executor extracts a resource allocation action pattern from the policy action vector, the resource allocation action pattern is a binary or real number sequence that specifies the allocation order and priority of time-frequency resource blocks; The policy executor dynamically allocates physical resource blocks to user equipment in the scheduler according to the resource allocation action pattern, and the allocation process considers the number, location and time interval of resource blocks; After completing the parameter adjustment, the policy executor performs a parameter verification sub-step to check whether the adjusted transmit power, modulation and coding scheme, and resource allocation are compatible with the device hardware limitations and network protocols; If the parameters are invalid, the policy executor falls back to the default configuration and records an error log.

5. The method of claim 1, wherein, In S2, the integration process of the comprehensive state vector includes the following operations: The downlink channel state information and the uplink channel state information in the wireless channel state information are taken as vector sub-blocks respectively; The number of currently connected users, data transmission rate requirements and buffer status in the system load state are taken as another vector sub-block; The moving speed, moving direction and Doppler shift in the user equipment moving state are taken as the last vector sub-block; All vector sub-blocks are spliced into a one-dimensional comprehensive state vector in order, and the total dimension of the comprehensive state vector is equal to the sum of the dimensions of each sub-block; During the integration process, the state data preparation step also ensures that the timestamps of each state parameter are consistent to reflect the network state at the same time.

6. The method of claim 1, wherein, In S5, when the base station device uses the adjusted parameters for wireless data transmission with the user equipment, the following auxiliary operations are also performed: The antenna array of the base station device uses beamforming technology to transmit signals to the user equipment according to the resource allocation action pattern; The scheduler of the base station device prioritizes resource allocation to high-priority user devices or delay-sensitive services based on the resource allocation action mode; During data transmission, the base station device monitors the link state but does not perform real-time adjustment or feedback on the strategy to maintain open-loop control characteristics.

7. The method of claim 2, wherein, In S1, the acquisition of the uplink channel state information further includes the following supplementary sub-steps: The base station device captures the uplink sounding reference signal sent by the user device through the uplink receiver; The signal processor demodulates and analyzes the uplink sounding reference signal, calculates the signal-to-interference-plus-noise ratio and timing advance; The mobility management module estimates the distance change of the user device based on the timing advance, and deduces the moving speed and direction combined with the Doppler shift.

8. The method of claim 3, wherein, The artificial intelligence strategy model in S3 further includes an attention mechanism module integrated between the convolutional layer and the fully connected layer; The operation of the attention mechanism module includes: Calculate the attention weight based on the importance of the features for the feature map output by the convolutional layer; Weighted sum of the feature map using attention weight to generate enhanced feature representation; Input the enhanced feature representation into the fully connected layer for subsequent processing.

9. The method of claim 4, wherein, In S4, the strategy executor also supports multi-mode strategy execution, including: If the power control action value indicates power increase, the strategy executor adjusts the power amplifier gain step by step to avoid interference caused by sudden changes; If the modulation and coding action index corresponds to high-order modulation, the strategy executor adjusts the encoder parameters to improve error correction ability at the same time; If the resource allocation action mode involves multiple user devices, the strategy executor allocates resource blocks using polling or priority scheduling algorithm; Multi-mode strategy execution ensures the smoothness of parameter adjustment and network stability.

10. The method of claim 1, wherein, The method further includes an initialization step, which is performed before the wireless environment state perception step, and the initialization step includes: The base station device loads the pre-configured artificial intelligence strategy model into the memory and verifies the integrity and compatibility of the model; The base station device calibrates the state monitoring module, including the radio frequency receiver, signal processor and load monitoring module, to ensure measurement accuracy; The base station device sets the default parameters of the strategy executor, including the default transmit power, the default modulation and coding scheme and the default resource allocation mode; The initialization step is only performed once when the base station device starts or resets, providing a basic configuration for subsequent steps.

Images