A millimeter wave phased array antenna communication system and method

Abstract

The application relates to the technical field of wireless communication, and discloses a millimeter wave phased array antenna communication system and method, the system comprises an environment perception and data acquisition unit, a feature coding and prediction unit, a cognitive beamforming decision unit, an adaptive hybrid beamforming control unit, a millimeter wave phased array antenna unit and a communication interface unit, real-time environment data is obtained through multi-source sensor fusion, and a deep learning model is used for feature extraction and future state prediction, the application actively adjusts the beam based on the prediction of user movement, obstacle shielding and channel change, reduces the risk of communication interruption and performance decline caused by environmental mutation, and overcomes the problems that the prior art has response lag or relies on frequent interaction to cause an increase in signaling overhead and time delay.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, in particular to a millimeter wave phased array antenna communication system and method. BACKGROUND

[0002] With the rapid development of the fifth generation (5G) and future communication technology, the millimeter wave frequency band has attracted widespread attention in the field of wireless communication due to its high bandwidth, high speed and small size. As a core component of realizing high-performance millimeter wave communication systems, the design and optimization of millimeter wave phased array antennas are of great significance to improve system capacity, coverage range and anti-interference ability. However, the existing millimeter wave phased array antenna communication systems still have deficiencies in beamforming ability, dynamic environment adaptability and hardware complexity, which limit their widespread application in practical scenarios.

[0003] In addition, the application of millimeter wave communication systems in high frequency signals under complex and variable environments such as dynamic blockages such as buildings and human bodies further exacerbates the difficulty of beam management and system optimization. This requires the system to have fast and accurate beam reconstruction capabilities, and in high-density user scenarios, how to balance the needs of multi-user access, system throughput and low latency, and ensure the real-time and adaptability of the beamforming strategy, are also challenges that need to be solved by existing technology.

[0004] In the prior art, although there are schemes to realize multi-beam multi-data stream through hybrid beamforming technology, they mainly aim at beamforming optimization in fixed scenarios, lack the ability to respond quickly to dynamic blocking environments, and require high accuracy of radio frequency links at high frequencies. Another scheme deals with dynamic blockage through a joint beam training and feedback mechanism, but frequent base station and user equipment interactions increase signaling overhead and latency, and performance is limited in high-density user scenarios, and its beam switching strategy requires high real-time processing capability of hardware, making it difficult to meet the demand for rapid beam adjustment in complex scenarios. The above problems show that the existing millimeter wave phased array antenna communication system still has significant deficiencies in dynamic environment adaptability and beam control flexibility. SUMMARY

[0005] The technical problem to be solved by the present application is to overcome the significant deficiencies of the prior art millimeter wave phased array antenna communication system in dynamic environment adaptability and beam control flexibility.

[0006] To achieve the above application purpose, the present application provides a millimeter wave phased array antenna communication system, which comprises an environment perception and data acquisition unit, a feature encoding and prediction unit, a cognitive beamforming decision unit, an adaptive hybrid beamforming control unit, a millimeter wave phased array antenna unit and a communication interface unit.

[0007] The environment perception and data collection unit is used to obtain various real-time data of the environment where the system is located, including channel state information, the position, speed, direction of the user equipment, the distribution of surrounding obstacles, the position and power of interference sources, and the running state parameters of the system itself. The channel detection module is used to measure the characteristics of the millimeter wave channel in real time, including the channel impulse response, the signal arrival angle and the signal departure angle. The environment perception and data collection unit synchronizes the time stamp, aligns the data and preliminarily fuses all the collected raw data to form a raw environment data set in a unified format. The fusion of the data collected by the sensor module adopts Kalman filtering.

[0008] The feature encoding and prediction unit is in communication connection with the environment perception and data collection unit, and is used to receive the raw environment data set and perform feature extraction, encoding and future state prediction thereon. The feature encoding and prediction unit identifies the features that have a key influence on the beamforming decision from the raw environment data set through a multi-dimensional feature extraction algorithm, including the mobile trajectory feature of the user equipment, the shielding probability feature of the obstacles, the spectrum feature of the interference sources and the time-varying feature of the channel fading. Further, the feature encoding and prediction unit encodes the extracted features in a high dimension by using a deep neural network architecture, maps the heterogeneous perception data into a unified semantic space, and forms a compact environment state feature vector. As an embodiment of the present application, the deep neural network architecture is a self-encoder or a recurrent neural network. The feature encoding and prediction unit further adopts a sequence prediction model, which adopts a long short-term memory network or a gated recurrent unit network, to predict the evolution trend of the environment state in a certain time period in the future based on the current and historical environment state feature vectors, including but not limited to the future position of the user equipment, the future moving path of the obstacles and the future change of the channel state. The sequence prediction model provides prospective information for the cognitive beamforming decision unit to realize active beam adjustment.

[0009] The cognitive beamforming decision unit is in communication connection with the feature encoding and prediction unit, configured to receive the environmental state feature vector and the predicted future environmental state information, and generate an optimized beamforming strategy based on preset optimization objectives and constraint conditions. The cognitive beamforming decision unit is specifically a decision engine based on deep reinforcement learning, which includes an environmental state encoding module, a policy generation network, a value network, and an experience replay buffer. The environmental state encoding module takes the environmental state feature vector output by the feature encoding and prediction unit as input, further abstracts and compresses it to form a state representation of the reinforcement learning agent. The policy generation network receives the state representation and outputs a plurality of selectable actions, which represent different beamforming strategies, including but not limited to beam direction, beam width, sidelobe suppression, power allocation, and millimeter wave phased array antenna element subarray configuration scheme. The value network is used to evaluate the expected long-term return of different strategies under a specific environmental state, providing guidance for the policy generation network. The experience replay buffer is used to store experience data generated by the agent during interaction with the environment, including state, action, reward, and next state, for offline training of the policy generation network and the value network. The cognitive beamforming decision unit optimizes the beamforming strategy by maximizing the long-term cumulative reward. The function expression of the cumulative reward is as follows: wherein R represents the cumulative reward; T represents the total number of time steps; t represents a time step; γ represents a discount factor, which ranges from 0 to 1; represents the system throughput at time step t; represents the communication delay at time step t; represents the received interference power at time step t; represents the total power consumption of the system at time step t; , , and represent the weight coefficients of the corresponding target items, used to balance the importance of different optimization objectives; represents the penalty term applied when the beamforming strategy violates the preset constraint condition at time step t. The cognitive beamforming decision unit adjusts the parameters of the policy generation network and the value network through an iterative learning process to find the optimal beamforming strategy in a dynamic environment. The policy optimization process effectively handles high-dimensional and nonlinear environmental state and action spaces using the nonlinear mapping capabilities of deep neural networks.

[0010] The adaptive hybrid beamforming control unit is communicatively connected with the cognitive beamforming decision unit and the millimeter wave phased array antenna unit, and converts the abstract strategy generated by the cognitive beamforming decision unit into hardware control instructions executable by the millimeter wave phased array antenna unit. The adaptive hybrid beamforming control unit comprises an analog beamforming control module, a digital beamforming control module, and a subarray configuration management module. The analog beamforming control module adjusts the analog phase shifter and variable gain amplifier on each radio frequency link in the phased array antenna unit according to the beam direction and width parameters output by the decision unit, so as to realize rough pointing and power adjustment of the beam. The analog phase shifter realizes phase adjustment by using a digital controlled MEMS phase shifter or a GaAs HBT phase shifter. The digital beamforming control module realizes multi-data stream transmission and fine beamforming by performing digital precoding and digital combining of signals at the baseband end. The digital precoding and digital combining are realized by a digital signal processor. The subarray configuration management module dynamically adjusts the connection relationship and resource allocation between the radio frequency link and the antenna element according to the subarray configuration scheme output by the decision unit. The dynamic subarray configuration is realized by a radio frequency switch array and a programmable interconnection network.

[0011] The millimeter wave phased array antenna unit is communicatively connected with the adaptive hybrid beamforming control unit and the communication interface unit, and is used to receive the instructions of the adaptive hybrid beamforming control unit and perform transmission and reception of millimeter wave signals. The millimeter wave phased array antenna unit comprises antenna elements, a radio frequency front end module, and a direct current power supply management module. The radio frequency front end module comprises a power amplifier, a low noise amplifier, a mixer, a filter, and a digital-to-analog converter array or an analog-to-digital converter array. The power amplifier is used to amplify millimeter wave signals on a transmission link, and the low noise amplifier is used to amplify weak millimeter wave signals on a reception link. The mixer is used to convert millimeter wave signals and intermediate frequency signals. The filter is used to filter out out-of-band spurious signals. The digital-to-analog converter array and the analog-to-digital converter array are used to realize conversion between digital signals and analog signals to cooperate with the hybrid beamforming architecture. The direct current power supply management module provides stable and adjustable direct current power supply for all radio frequency front end components. The millimeter wave phased array antenna unit accurately controls the phase and amplitude of each antenna element according to the instructions of the adaptive hybrid beamforming control unit, and forms a millimeter wave beam with a specific direction and shape.

[0012] The communication interface unit is communicatively connected with the millimeter wave phased array antenna unit, and provides millimeter wave communication services by interacting with the upper layer network protocol stack. The communication interface unit is responsible for encapsulating upper layer baseband data into millimeter wave radio frequency signals, and decapsulating received millimeter wave radio frequency signals.

[0013] According to another aspect of the present application, a millimeter wave phased array antenna communication method is provided, comprising the following steps:

[0014] S1: Obtain a plurality of real-time data of the environment in which the communication system is located (including channel state information, user equipment position, speed, direction, surrounding obstacle distribution, interference source position and power, and system self-operation state parameters), collect through a global positioning system module, an inertial measurement unit module, a millimeter wave radar module, a visual sensor module, and a channel detection module, and synchronize time stamps, align data, and preliminarily fuse all raw data to form a unified format of raw environment data set.

[0015] S2: Receive the raw environment data set, perform feature extraction, encoding, and future state prediction, generate an environment state feature vector and a predicted future environment state information.

[0016] S3: Receive the environment state feature vector and the predicted future environment state information, generate an optimal beamforming strategy based on a preset optimization target and a constraint condition.

[0017] S4: Convert the beamforming strategy into a hardware control instruction executable by the millimeter wave phased array antenna unit.

[0018] S5: The millimeter wave phased array antenna unit performs transmission and reception of millimeter wave signals according to the hardware control instruction.

[0019] S6: Real-time evaluation of system performance under the current beamforming strategy, and updating the cognitive beam strategy decision model based on the evaluation results.

[0020] Further, in S3, the decision engine generates a plurality of selectable beamforming actions according to the environment state, and evaluates the expected long-term return through a value network to select the optimal action; in S6, at least one of system throughput, communication delay, bit error rate, and signal-to-noise ratio is used as feedback information to update the strategy generation network and value network parameters of the decision engine.

[0021] Compared with the prior art, the present application has the following advantages:

[0022] Through the environment perception and prediction mechanism, the proactive response to dynamic environmental changes is realized, the millimeter wave phased array antenna communication system performs active beam adjustment based on the prediction of user movement, obstacle shielding, and channel changes, reduces the risk of communication interruption and performance decline caused by environmental mutations, and overcomes the problem of response lag or dependence on frequent interaction in the prior art, which increases signaling overhead and time delay.

[0023] The cognitive beamforming decision unit based on deep reinforcement learning can autonomously learn and optimize the beamforming strategy, maximize the long-term cumulative reward function, comprehensively consider multiple targets such as throughput, delay, interference suppression and power consumption, generate a globally optimal strategy, and make the system have high beam control flexibility and adaptability in complex scenarios, effectively cope with high-density users, strong interference and complex channels and other challenges.

[0024] The subarray configuration management module in the adaptive hybrid beamforming control unit can dynamically adjust the mapping relationship and resource allocation between the radio frequency link and the antenna array element according to the cognitive decision result, optimize the radio frequency chain utilization efficiency under the premise of ensuring the flexibility of beamforming, and flexibly adjust the number and function of digital radio frequency links according to actual needs, thereby reducing the hardware complexity and system cost.

[0025] Through continuous system performance evaluation and model iterative learning, the cognitive beamforming decision model is continuously optimized, so that the system has adaptive and self-evolution capabilities, and continuously improves the communication performance and resource utilization efficiency with time and environmental evolution.

[0026] Integrate multi-modal sensor data and perform high-dimensional feature coding and prediction to provide rich and accurate input information for cognitive decision, improve the understanding ability and decision accuracy of the system to the communication environment, and enhance its reliability and robustness in various complex environments. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is the structural block diagram of the millimeter wave phased array antenna communication system proposed by the present application;

[0028] Figure 2 is the logic flow diagram of the cognitive beamforming decision unit in the present application;

[0029] Figure 3 is the logic flow diagram of the environment data acquisition and feature coding and prediction in the present application;

[0030] Figure 4 is the logic flow diagram of the adaptive hybrid beamforming control unit in the present application;

[0031] Figure 5 is the method flow chart of the present application;

[0032] Figure 6 is the structural block diagram of the environment perception and data acquisition unit in the present application. DETAILED DESCRIPTION

[0033] In order to make the objects, technical solutions and advantages of the present application clearer, more complete, further detailed description of the present application will be made below with reference to the drawings and in conjunction with the specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, but not to constitute any form of limitation on the present application.

[0034] Please refer to Figure 1 With Figure 6 , the present embodiment will be described in detail around a kind of millimeter wave phased array antenna communication system and its corresponding method, to improve the dynamic response capability of system in complex and changeable environment, optimize beamforming performance, while effectively managing hardware resources.

[0035] A kind of millimeter wave phased array antenna communication system includes: environment perception and data acquisition unit, feature encoding and prediction unit, cognitive beamforming decision unit, adaptive hybrid beamforming control unit, millimeter wave phased array antenna unit and communication interface unit.These units work together to realize the cognitive and adaptive ability of system.

[0036] Environment perception and data acquisition unit is used to obtain a variety of real-time data of the environment where the communication system is located.Real-time data includes channel state information, user equipment position, user equipment speed, user equipment direction, surrounding obstacle distribution, interference source position and power and system itself operating state parameters.Environment perception and data acquisition unit realizes data acquisition by integrating a variety of sensor modules.Sensor modules include: global positioning system module, inertial measurement unit module, millimeter wave radar module, visual sensor module and channel sounding module.

[0037] Global positioning system module is responsible for obtaining the accurate geographic position information of user equipment and base station.The module obtains high-precision three-dimensional position coordinates by receiving global navigation satellite system signals.It integrates multiple radio frequency front ends, digital signal processing units and positioning algorithm modules inside.Geographic position information is output in longitude, latitude, altitude and corresponding accuracy estimate value, with accurate time stamp.These data are the basis for building geographic space environment model, and play a decisive role in user equipment tracking and beam pointing.

[0038] Inertial measurement unit module is used to obtain the attitude information of user equipment.Attitude information includes roll angle, pitch angle and yaw angle.Inertial measurement unit module is composed of three-axis accelerometer, three-axis gyroscope and three-axis magnetometer.Accelerometer measures linear acceleration, gyroscope measures angular velocity, and magnetometer measures geomagnetic field direction.Through sensor fusion algorithm (Kalman filter or extended Kalman filter), inertial measurement unit module can output the attitude angle and angular velocity of user equipment in three-dimensional space in real time.Attitude information is crucial for accurately adjusting the beam direction to cope with the tilting or rotating motion of user equipment.

[0039] The millimeter-wave radar module is used to detect obstacles in the surrounding environment, obtaining distance, angle, and relative velocity information of the obstacles. The millimeter-wave radar module operates in the millimeter-wave frequency band, using transmitted and received millimeter-wave signals to detect targets. It contains millimeter-wave front-end transceiver circuits, digital signal processors, and target detection and tracking algorithms inside. The module transmits frequency-modulated continuous waves or pulsed waves and analyzes the frequency, phase, and amplitude changes of the received echo signals to calculate the distance, angle, and radial velocity of the target. These information is used to construct a dynamic obstacle model in the environment, identify potential beam blockers, and predict their future movement paths, which are key inputs to avoid communication interruption.

[0040] The vision sensor module is used to assist in identifying static or dynamic blockers in the environment. The vision sensor module is specifically a high-resolution digital camera equipped with an image processing unit. By analyzing consecutive frames of images, the vision sensor module can identify large static obstacles and dynamic obstacles in the environment. Its internal image processing algorithms include target detection, semantic segmentation, and motion estimation. The fusion of vision data and millimeter-wave radar data can provide more comprehensive environmental perception capabilities, making up for the limitations of a single sensor in certain environments.

[0041] The channel sounding module is used to measure the characteristics of the millimeter-wave channel in real time. The characteristics include channel impulse response, signal angle of arrival, and signal angle of departure. The channel sounding module sends known sounding signals and analyzes the received signals to calculate the multipath propagation characteristics of the channel. It integrates high-speed analog-to-digital converters, digital signal processors, and channel estimation algorithms inside. The channel impulse response provides the multipath structure of the channel in the time domain, and the signal angle of arrival and signal angle of departure describe the propagation direction of the multipath signal in space. These accurate channel characteristics are key inputs for implementing high-performance beamforming, enabling the system to adapt to complex millimeter-wave propagation environments.

[0042] The environmental perception and data acquisition unit timestamps, aligns, and preliminarily fuses all collected raw data to form a unified format of raw environmental data set. Time stamp synchronization is achieved through high-precision clock sources and network time protocol or global positioning system time service, ensuring accurate correspondence of different sensor data in the time dimension. Data alignment includes spatial coordinate system conversion, data interpolation resampling, and other operations to unify heterogeneous sensor data to a common time and space reference system. Preliminary fusion includes removing redundancy, filling missing data, and preliminary anomaly detection to generate structured and consistent data sets. This data set is strictly checked to ensure its integrity and accuracy, providing reliable input for subsequent feature extraction and prediction.

[0043] The feature encoding and prediction unit is communicatively connected to the environment perception and data collection unit, configured to receive the raw environment data set and perform feature extraction, encoding and future state prediction. The feature encoding and prediction unit first identifies the features that have a key impact on the beamforming decision from the raw environment data set through a multi-dimensional feature extraction algorithm. The key features include the user equipment's moving trajectory feature, the obstacle's occlusion probability feature, the interference source's spectrum feature and the channel fading's time-varying feature. The multi-dimensional feature extraction algorithm employs principal component analysis, independent component analysis or deep learning feature extraction network. The user equipment's moving trajectory feature is obtained by trajectory fitting and pattern recognition on historical position and velocity data, such as identifying straight-line motion, turning or parking state. The obstacle's occlusion probability feature is quantitatively estimated by combining millimeter wave radar and vision sensor data to quantify the dynamic occlusion impact of obstacles on the beam path. The interference source's spectrum feature is identified and located by analyzing the spectrum distribution of the received signal, and the power level of the potential interference source. The channel fading's time-varying feature is captured by time series analysis of channel sounding results, capturing the dynamic variation of channel gain, phase and multipath components.

[0044] Further, the feature encoding and prediction unit employs a deep neural network architecture to encode the extracted features into high-dimensional, mapping heterogeneous perception data into a unified semantic space, forming a compact environment state feature vector. The deep neural network architecture is a kind of autoencoder or recurrent neural network. The autoencoder compresses high-dimensional input into low-dimensional latent representation through the encoder, and reconstructs the original input through the decoder. The high-dimensional encoding process aims to eliminate the heterogeneity between different sensor data, extract more abstract and representative semantic features, thereby reducing the processing complexity of the subsequent decision unit and improving the robustness of the decision.

[0045] The feature encoding and prediction unit further employs a sequence prediction model, which employs a long short-term memory network or a gated recurrent unit network. Based on the current and historical environment state feature vector, the future evolution trend of the environment state within a certain time period is predicted. The prediction content includes the future position of the user equipment, the future moving path of the obstacle and the future change of the channel state. The long short-term memory network and the gated recurrent unit network have the ability to process and predict sequence data, effectively solving the gradient vanishing or gradient explosion problem in traditional recurrent neural networks through memory cells or gating mechanisms, capturing long-term dependencies. The sequence prediction model can predict the environmental changes within a few milliseconds to a few seconds by learning the time series evolution pattern of the environment state feature vector. This prediction function provides forward-looking information for the cognitive beamforming decision unit to achieve proactive beam adjustment.

[0046] The cognitive beamforming decision unit is connected to the feature encoding and prediction unit, and is used for receiving the environmental state feature vector and the predicted future environmental state information, and generating an optimized beamforming strategy based on a preset optimization target and constraint condition. The core of the cognitive beamforming decision unit is a decision engine based on deep reinforcement learning. The decision engine includes an environmental state encoding module, a policy generation network, a value network and an experience replay buffer, wherein the environmental state encoding module adopts a 2-layer fully connected network, inputs a 16-dimensional predicted feature vector, has a 32-dimensional hidden layer, and outputs an 8-dimensional state representation s; the policy generation network adopts an "input layer-fully connected layer-fully connected layer-output layer" architecture, and each layer has 8 layers, 64 layers, 32 layers and 5 layers respectively, and outputs a 5-dimensional action vector (corresponding beam direction : 0-360°, beam width ω: 5-20°, sidelobe suppression SLR: -30 to -10 dB, power allocation P: 0.5-2W, number of subarrays: 1-8), adopts a Softmax function to output an action probability distribution, an exploration strategy is ε-greedy, ε is an exploration rate, an initial value of ε is 0.9, ε is attenuated by 0.01 every 1000 steps, and the minimum value of ε is 0.1; the value network has the same structure as the policy generation network, and outputs a state-action value Q(s, ), which represents an expected long-term cumulative reward of performing an action in the state s, and is a core evaluation index of policy optimization; the capacity of the experience replay buffer is 10000, a priority experience replay strategy is adopted, and the expression of a sampling priority is as follows: wherein represents the sampling priority; is the action vector; is a time difference error; is a constant term, and the numerical value of the constant term is , an Adam optimizer is adopted in the training process of the decision engine, an initial learning rate is 0.001, and the initial learning rate is attenuated to 0.9 times every 10000 training steps; a loss function is a mean square error; the total number of training rounds is 500000 steps, when the average cumulative reward of the policy network changes by less than 0.01 in continuous 10000 steps, the model is considered to be converged. Training data is derived from a high-precision millimeter wave communication simulation environment constructed based on Ray Tracing (a physical engine).

[0047] The environmental state encoding module takes the environmental state feature vector output by the feature encoding and prediction unit as input, further abstracts and compresses the environmental state feature vector, and forms a state representation of the reinforcement learning agent. This module maps the original feature vector to a lower-dimensional and more decision-relevant state space through a multi-layer fully connected neural network or a convolutional neural network. The state representation aims to capture all key information of the current communication environment while filtering out redundancies, so that the reinforcement learning agent can learn efficiently.

[0048] The policy generation network receives the state representation and outputs a number of alternative actions. These actions represent different beamforming strategies. Beamforming strategies include beam direction, beam width, sidelobe suppression, power allocation, and subarray configuration of the mmWave phased array antenna elements. The policy generation network employs a multi-layer perceptron or convolutional network to output a probability distribution over actions or directly output a deterministic action. Beam direction represents the pointing angle of the mmWave main beam, and beam width represents the angular range of the main beam. Sidelobe suppression aims to minimize radiation in undesired directions to reduce interference to other users. Power allocation determines the radio frequency power transmitted to different data streams or different subarrays. Subarray configuration involves dynamic adjustment of the connection between radio frequency chains and array elements in the antenna array. The policy generation network generates actions that maximize long-term rewards based on the current environmental state.

[0049] The value network is used to evaluate the expected long-term return of different strategies in a particular environmental state, providing guidance for the policy generation network. The value network is a deep neural network whose input is the same as the state representation of the policy generation network, and its output is the expected cumulative reward value of all possible actions in that state. By predicting future rewards, the value network helps the policy generation network judge the merits of the current action, thereby accelerating the learning process and improving decision-making quality.

[0050] The experience replay buffer is used to store experience data generated by the agent during interaction with the environment. These experience data include state, action, reward, and next state. The experience replay buffer is a first-in, first-out queue that provides training samples to the policy generation network and the value network in a random sampling manner. By randomly sampling from the buffer, the temporal correlation between experience data can be broken, reducing the correlation of training samples, and thus improving the stability and efficiency of training.

[0051] The cognitive beamforming decision unit optimizes the beamforming strategy by maximizing the long-term cumulative reward. The cumulative reward function is defined as follows: where R represents the cumulative reward, T represents the total number of time steps, t represents the time step; γ represents the discount factor, which ranges from 0 to 1; represents the system throughput at time step t; represents the communication delay at time step t. represents the received interference power at time step t; represents the total power consumption of the system at time step t; 、 、 and represent the weight coefficients of the corresponding target items, used to balance the importance of different optimization objectives. represents the penalty term applied at time step t if the beamforming strategy violates the preset constraint condition. When the bit error rate of the communication link is monitored to exceed a threshold , a fixed value of -10 is applied; when exceeds 90% of the maximum rated power, a linear penalty proportional to the excess power is applied, in this embodiment, = 1.0, = 0.5, = 0.3, = 0.2.

[0052] The cognitive beamforming decision unit adjusts the parameters of the strategy generation network and the value network through an iterative learning process, so as to find the optimal beamforming strategy in a dynamic environment. The strategy optimization process effectively processes high-dimensional and nonlinear environment states and action spaces by using the nonlinear mapping capability of the deep neural network. Through the gradient descent algorithm, the decision engine continuously updates its network parameters, so that the output strategy can obtain a higher cumulative reward.

[0053] The adaptive hybrid beamforming control unit is communicatively connected to the cognitive beamforming decision unit and the millimeter wave phased array antenna unit, and is used to convert the abstract strategy generated by the cognitive beamforming decision unit into hardware control instructions executable by the millimeter wave phased array antenna unit. The adaptive hybrid beamforming control unit includes an analog beamforming control module, a digital beamforming control module, and a subarray configuration management module.

[0054] The analog beamforming control module is used to adjust the analog phase shifters and variable gain amplifiers on each radio frequency link in the phased array antenna unit according to the beam direction and width parameters output by the decision unit, so as to realize the rough pointing and power adjustment of the beam. The analog phase shifter adopts a digital controlled micro-electro-mechanical system phase shifter or a gallium arsenide heterojunction bipolar transistor phase shifter. The analog phase shifter realizes the spatial pointing of the beam by controlling the phase delay of the microwave signal. The variable gain amplifier realizes the power distribution and sidelobe suppression of the beam by adjusting the amplitude of the radio frequency signal.

[0055] The digital beamforming control module is used to perform digital precoding and digital combining of signals at the baseband end, so as to realize multi-data stream transmission and fine beamforming. The digital precoding performs weighting and phase adjustment on the digital baseband signal at the sending end to form multiple mutually orthogonal or quasi-orthogonal data streams, thereby supporting multi-user multiple-input multiple-output communication. The digital combining performs weighted combining of the signals received by multiple antenna elements at the receiving end to maximize the signal-to-noise ratio of the expected signal and suppress interference.

[0056] The subarray configuration management module is configured to dynamically adjust the connection relationship and resource allocation between the radio frequency link and the antenna array element according to the subarray configuration scheme output by the decision unit. In specific embodiments, the subarray configuration management module is configured to dynamically configure according to the instructions output by the cognitive decision unit and the number of users and the rate of environmental change. Specifically, a user number threshold and an environmental change rate threshold are set. When the number of active users is less than the user number threshold and the maximum moving speed of the obstacle is less than the environmental change rate threshold, the large subarray mode is configured to obtain higher gain. When the number of active users is greater than or equal to the user number threshold or the maximum moving speed of the obstacle is greater than or equal to the environmental change rate threshold, the small subarray mode is switched to improve more user parallel communication capability and beam agility. The subarray configuration management module controls the radio frequency switch matrix and the programmable interconnection network to realize the dynamic reconstruction of the radio frequency signal path, thereby optimizing the utilization efficiency of hardware resources and the flexibility of beamforming.

[0057] The millimeter wave phased array antenna unit is connected to the adaptive hybrid beamforming control unit and the communication interface unit for receiving instructions from the adaptive hybrid beamforming control unit and performing transmission and reception of millimeter wave signals. The millimeter wave phased array antenna unit includes antenna array elements, radio frequency front-end modules, and direct current power management modules.

[0058] The antenna array elements are planar microstrip patch antennas or Vivaldi antennas arranged in a two-dimensional array. Each antenna array element is responsible for receiving or transmitting millimeter wave signals.

[0059] The radio frequency front-end module includes a power amplifier, a low-noise amplifier, a mixer, a filter, and a digital-to-analog converter array or an analog-to-digital converter array. The power amplifier is used to amplify the millimeter wave signal on the transmission link to achieve sufficient transmission power. The low-noise amplifier is used to amplify the weak millimeter wave signal on the receiving link while introducing minimal noise to improve the receiving sensitivity. The mixer is used to convert the millimeter wave signal and the intermediate frequency signal to each other to realize frequency upconversion and downconversion. The filter is used to filter out out-of-band spurious signals and image frequencies to ensure signal purity. The digital-to-analog converter array and the analog-to-digital converter array are used to convert between digital signals and analog signals to cooperate with the hybrid beamforming architecture. The digital-to-analog converter converts the digital baseband signal into an analog intermediate frequency signal, which is then transmitted after upconversion. The analog-to-digital converter converts the received analog intermediate frequency signal into a digital baseband signal for digital processing.

[0060] The direct current power management module provides stable and adjustable direct current power for all radio frequency front-end components. The module includes a voltage regulator, a current limiter, a power distribution network, and a temperature monitoring circuit. The direct current power management module accurately controls the power supply voltage and current of each radio frequency link according to the instructions of the adaptive hybrid beamforming control unit, thereby realizing power consumption management and performance optimization.

[0061] The millimeter wave phased array antenna unit controls the phase and amplitude of each antenna element accurately according to the instruction of the adaptive hybrid beamforming control unit, and forms a millimeter wave beam with a specific direction and shape. In specific embodiments, the direction of the beam can be changed by adjusting the phase of each element; the shape and sidelobe level of the beam can be controlled by adjusting the amplitude of each element.

[0062] The communication interface unit is communicatively connected to the millimeter wave phased array antenna unit, and is used to interact with the upper layer network protocol stack to provide millimeter wave communication services. The communication interface unit is responsible for encapsulating the upper layer baseband data into millimeter wave radio frequency signals, and decapsulating the received millimeter wave radio frequency signals. The unit includes a baseband modem, a channel encoder, a decoder, and a media access control layer processing module. In the transmission direction, the communication interface unit receives the original data bit stream from the upper layer protocol stack, performs channel coding, modulation, and format encapsulation, converts it into a baseband digital signal, and then transmits it to the millimeter wave phased array antenna unit for radio frequency conversion and transmission. In the receiving direction, the communication interface unit receives the digital baseband signal from the millimeter wave phased array antenna unit, performs demodulation, channel decoding, and data decapsulation, and finally delivers the recovered original data bit stream to the upper layer protocol stack. The communication interface unit ensures that the millimeter wave communication system can be seamlessly integrated into the entire communication network architecture, providing efficient and reliable data transmission services.

[0063] A millimeter wave phased array antenna communication method, which realizes intelligent and adaptive millimeter wave communication in combination with a millimeter wave phased array antenna communication system, includes the following steps:

[0064] S1: Environment data acquisition and preliminary processing

[0065] The environmental perception and data acquisition unit first activates its internal global positioning system module, inertial measurement unit module, millimeter wave radar module, vision sensor module, and channel sounding module, and performs data acquisition in parallel. The global positioning system module updates the accurate three-dimensional geographic position coordinates of the user equipment and the base station in real time by continuously receiving satellite signals; the inertial measurement unit module uses the built-in accelerometer, gyroscope, and magnetometer to obtain the linear acceleration, angular velocity, and geomagnetic field strength of the user equipment at a high sampling rate, and then calculates the roll angle, pitch angle, and yaw angle of the user equipment and other attitude information; the millimeter wave radar module periodically transmits millimeter wave pulses or frequency-modulated continuous waves, and analyzes the received echo signals to obtain the distance, azimuth angle, and relative radial velocity of the surrounding obstacles. This process uses the Doppler effect and time-of-flight measurement principle to construct an obstacle distribution map in a dynamic environment; the vision sensor module captures visual information in the environment through high-frame-rate video streams, and uses image processing algorithms to identify the type, location, and motion trend of static obstacles (such as buildings, trees) and dynamic obstacles (such as pedestrians, vehicles); the channel sounding module sends known millimeter wave sounding sequences within a specified sounding interval, and by analyzing the coherence bandwidth, correlation function, and multipath components of the received signals, it accurately measures the channel impulse response, signal arrival angle, and signal departure angle, and evaluates the channel fading, scattering, and multipath characteristics.

[0066] The collected raw data is transmitted to the environmental perception and data acquisition unit in real time. The environmental perception and data acquisition unit synchronizes the time stamps of all data. Then, the data alignment operation is performed to convert all sensor data from their respective local coordinate systems to the global geographic coordinate system or the unified system internal coordinate system. In addition, the high sampling rate inertial measurement unit data is downsampled, or the low sampling rate global positioning system data is interpolated. Finally, preliminary fusion is performed, and the fusion process includes identifying and removing duplicate data, filling in some missing values, and preliminarily identifying and removing abnormal data points through threshold detection or simple statistical methods. Finally, a unified format, time-synchronized, spatially aligned, and preliminarily cleaned raw environmental data set is formed.

[0067] S2: Environmental state feature encoding and prediction

[0068] The feature encoding and prediction unit identifies and extracts key features from the original environment dataset that have a decisive impact on the millimeter wave beamforming decision. For the user device's moving trajectory feature, by comprehensively analyzing its historical position, speed, and acceleration data, the trajectory fitting algorithm is used to predict its motion path in the short term and identify its motion pattern. In combination with millimeter wave radar and visual sensor data, the geometric position, size, material properties, and relative moving speed of obstacles on the preset beam propagation path are analyzed, and the possibility and degree of beam occlusion by obstacles in the current and future time window are calculated. The spectrum feature of the interference source is identified by spectrum analysis of the received signal, including the center frequency, bandwidth, power spectral density, and its trend over time, which classifies the type and strength of the interference. The time-varying feature of channel fading is modeled by time series modeling of the channel impulse response, signal angle of arrival, and signal angle of departure data provided by the channel probing module, and the autoregressive moving average model is used to capture the dynamic change law of channel gain, phase, and multipath components and predict their future state.

[0069] The feature encoding and prediction unit uses a deep neural network architecture to encode the extracted multi-dimensional features in a high-dimensional space. The purpose is to eliminate the format differences and semantic barriers between different feature sources and unify them into a mathematically operable vector space, providing standardized and semantically rich inputs for subsequent cognitive beamforming decisions.

[0070] Further, the feature encoding and prediction unit uses a sequence prediction model to predict the evolution trend of the environment state in a certain time period based on the current and historical environment state feature vectors. The sequence prediction model uses a long short-term memory network or a gated recurrent unit network. The model learns the dynamic pattern of the environment state feature vector sequence and extrapolates forward to predict the environment condition in the future several time steps. The prediction content includes the future precise position and direction of the user device, the future moving path and possible occlusion area of the obstacle, and the future gain, phase, and multipath structure changes of the millimeter wave channel. These prediction information has a forward-looking nature, which can provide a trend judgment of the future communication environment, enabling the cognitive beamforming decision unit to plan the beam adjustment strategy in advance, achieving proactive beam management rather than passive response to environmental changes. This prediction capability is crucial for millimeter wave communication, as millimeter wave signals are highly sensitive to occlusion and movement. Early prediction of environmental changes can significantly reduce the risk of communication interruption and performance degradation.

[0071] S3: Cognitive beam strategy decision

[0072] In this step, the cognitive beamforming decision unit receives the environmental state feature vector and the predicted future environmental state information from the feature encoding and prediction unit. These inputs comprehensively describe the various elements of the current and future communication environment. Based on these information, as well as the pre-set optimization objectives and constraints, the decision unit generates an optimized beamforming strategy. The entire decision-making process is completed by the internal deep reinforcement learning decision engine.

[0073] The core workflow of the deep reinforcement learning decision engine is as follows: The environmental state encoding module first further abstracts and compresses the received environmental state feature vector to form a state representation of the reinforcement learning agent. This representation is the internal understanding of the current communication environment by the decision engine and is the basis for decision-making. The policy generation network receives this state representation and outputs several optional actions based on its current learned policy. These actions represent a variety of possible beamforming strategies, including the precise pointing of beam direction, whether the beam width is narrow or wide, the degree of sidelobe suppression to reduce interference to other users, the reasonable allocation of power on different data streams or subarrays, and the subarray configuration scheme of the millimeter wave phased array antenna unit. The policy generation network selects the action that maximizes the long-term cumulative reward based on the current state.

[0074] The value network plays the role of evaluator in this process, it receives the same state representation and evaluates the expected long-term return of each action output or explored by the policy generation network in a specific environmental state. Through this evaluation, the value network provides guidance for the policy generation network, helping it understand which actions are superior and which actions will lead to poor performance. The experience replay buffer stores all experience data generated during the interaction between the agent and the environment, including the state, the action taken, the reward given by the environment, and the next state after taking the action. These experience data are then randomly sampled for offline training of the policy generation network and the value network, stabilizing the learning process and improving data utilization efficiency.

[0075] The cognitive beamforming decision unit continuously adjusts the internal parameters of the policy generation network and the value network through an iterative learning process. This process utilizes the nonlinear mapping capabilities of deep neural networks to effectively handle high-dimensional, nonlinear environmental states and action spaces, enabling it to learn complex decision-making rules. Through repeated interaction with the environment, collection of experience, and updating of network weights, the decision engine gradually optimizes its policy, enabling it to generate optimized beamforming strategies in various dynamic environments. Ultimately, the decision engine outputs an optimal beamforming strategy that comprehensively considers throughput maximization, latency minimization, interference suppression, and power consumption optimization.

[0076] S4: Hybrid beamforming control instruction generation

[0077] The abstract beamforming strategy generated in the cognitive beam strategy decision step is converted into hardware control instructions that can be directly executed by the millimeter wave phased array antenna unit.

[0078] The analog beamforming control module first parses the beam direction and beam width parameters in the strategy. These parameters are converted into specific adjustment values for the analog phase shifters and variable gain amplifiers on each radio frequency chain in the millimeter wave phased array antenna unit. If the decision requires the beam to be pointed in a certain elevation and azimuth angle, the analog beamforming control module calculates the required phase delay for each antenna element. Then, the phase delay is converted into a digital control code for the analog phase shifter and sent to the digitally controlled microelectromechanical system phase shifter or gallium arsenide heterojunction bipolar transistor phase shifter through a serial peripheral interface or general-purpose input-output interface to achieve precise phase adjustment. At the same time, according to the power distribution and sidelobe suppression requirements in the strategy, the gain control voltage or digital control word of the variable gain amplifier is calculated and generated to adjust the signal amplitude of each radio frequency chain, achieving rough pointing and power adjustment of the beam.

[0079] The digital beamforming control module focuses on fine beamforming and multi-data stream transmission requirements in the strategy. When the strategy requires simultaneous transmission of multiple data streams to multiple users, the digital beamforming control module generates corresponding digital precoding matrices based on channel state information. These precoding matrices contain coefficients for weighting and phase adjustment of baseband digital signals, implemented by a digital signal processor or a field programmable gate array. The precoding instructions are transmitted in the form of digital matrices to the digital-to-analog converter array front end of the millimeter wave phased array antenna unit through a high-speed serial interface to process the data stream that is about to be converted into an analog signal. At the receiving end, if the strategy requires digital combining to maximize the signal-to-noise ratio, the digital beamforming control module will also generate a digital combining matrix, which is executed by a digital signal processor or a field programmable gate array, to perform combining processing on multiple digital signals from the analog-to-digital converter array.

[0080] The subarray configuration management module is responsible for processing the strategy of the subarray configuration scheme of the millimeter wave phased array antenna unit. According to the instructions of the decision unit, the connection relationship and resource allocation between the radio frequency link and the antenna element are dynamically adjusted. When the strategy instruction is "split into multiple small subarrays", the subarray configuration management module will generate specific configuration codes for controlling the radio frequency switch array and the programmable interconnection network. These configuration codes are sent to the corresponding hardware through the low-speed control bus to physically or logically reconfigure the radio frequency signal path. The radio frequency switch array connects different antenna elements to different radio frequency links through switching signal paths, thereby realizing the dynamic division of subarrays. The programmable interconnection network provides more flexible connection topology reconstruction capability. This dynamic configuration ensures the optimal utilization of hardware resources, and flexibly adjusts the number and function of digital radio frequency links according to the actual scene requirements, thereby reducing the hardware complexity and system cost.

[0081] Finally, the adaptive hybrid beamforming control unit integrates these analog phase shifter adjustment values, variable gain amplifier control words, digital precoding and digital combining matrices, and configuration codes of the radio frequency switch array and the programmable interconnection network into a complete set of hardware control instruction set executable by the millimeter wave phased array antenna unit. This hardware control instruction set is transmitted to the millimeter wave phased array antenna unit to realize accurate beam coarse pointing, power adjustment, fine beamforming and dynamic subarray configuration.

[0082] S5: Millimeter wave signal transmission and reception

[0083] This step is performed by the millimeter wave phased array antenna unit according to the received hardware control instructions. The hardware control instruction set generated by the adaptive hybrid beamforming control unit and transmitted to the millimeter wave phased array antenna unit contains accurate phase and amplitude control information for each antenna element. When these instructions are received, the antenna elements, radio frequency front-end modules and direct current power supply management modules inside the millimeter wave phased array antenna unit work cooperatively to realize the transmission and reception of millimeter wave signals.

[0084] In the process of transmitting millimeter wave signals, the communication interface unit processes the upper layer baseband data into digital baseband signals and transmits them to the millimeter wave phased array antenna unit. If there are digital precoding instructions, the digital signal processor will first perform precoding processing on these digital baseband signals. Subsequently, the precoded digital baseband signals are converted into analog intermediate frequency signals by the digital-to-analog converter array, and then up-converted to the millimeter wave frequency band by the mixer in the radio frequency front-end module. The up-converted millimeter wave signals are filtered by the filter to remove spurious components, and then amplified to the required transmission power by the power amplifier. The key is that the instructions issued by the adaptive hybrid beamforming control unit accurately control the analog phase shifters and variable gain amplifiers on each radio frequency link. The analog phase shifters adjust the phase of the millimeter wave signals according to the instructions, and the variable gain amplifiers adjust their amplitudes. By finely controlling the phase and amplitude of each antenna element, the millimeter wave phased array antenna unit can accurately point the main lobe to the target user while suppressing the sidelobe level to the minimum, avoiding interference with other users. The direct current power management module provides stable and adjustable direct current power for each radio frequency front-end component according to the instructions, ensuring that it operates in the best working state while optimizing power consumption. Finally, the millimeter wave signals with accurate phase and amplitude adjustment are radiated by the antenna elements, forming a highly directional millimeter wave beam for data transmission.

[0085] In the process of receiving millimeter wave signals, the antenna elements of the millimeter wave phased array antenna unit receive millimeter wave signals from space. These weak millimeter wave signals are first amplified by the low-noise amplifier in the radio frequency front-end module to improve the reception sensitivity. The amplified millimeter wave signals are filtered by the filter to remove out-of-band noise and interference, and then down-converted to the intermediate frequency by the mixer. If there are digital combining instructions, the intermediate frequency signals are converted into digital baseband signals by the analog-to-digital converter array and processed by the digital signal processor or field programmable gate array for post-combining processing to maximize the signal-to-noise ratio of the desired signal. In this process, the analog phase shifters and variable gain amplifiers also adjust the phase and amplitude of the received signals according to the instructions to achieve accurate pointing and interference suppression of the receiving beam. The direct current power management module ensures the stable operation of the radio frequency front-end components. Finally, the processed digital baseband signals are transmitted back to the communication interface unit for demodulation and unpackaging to recover the original data. The entire transmission and reception process is highly synchronized and driven by the instructions of the adaptive hybrid beamforming control unit, ensuring the efficiency and reliability of millimeter wave communication.

[0086] S6: System performance evaluation and model iterative learning

[0087] This step is a closed-loop feedback mechanism in the millimeter wave phased array antenna communication method, aiming to achieve continuous optimization and adaptive learning of the system. After the completion of millimeter wave signal transmission and reception, the system evaluates the performance of the communication system under the current beamforming strategy in real time. The performance evaluation module continuously monitors and collects key performance indicators. The indicators include but are not limited to: system throughput, communication latency, bit error rate, and signal-to-noise ratio. In addition, other related indicators will also be monitored, including user experience quality, power consumption, and interference level to adjacent users.

[0088] The performance evaluation module transmits these real-time performance indicators as feedback information to the deep reinforcement learning decision engine of the cognitive beamforming decision unit. Inside the decision engine, the performance indicators are compared with the previously set optimization targets. If the actual throughput is lower than expected, or the latency is higher than the acceptable threshold, it indicates that the current beamforming strategy is not ideal. The performance evaluation module will convert these indicators into reward signals and use them in combination with the reinforcement learning framework to update the parameters of the policy generation network and the value network.

[0089] The update process of the policy generation network and the value network is based on the offline training mechanism of reinforcement learning. A large amount of historical interaction experience data is stored in the experience replay buffer, including past environmental states, taken beamforming actions, and rewards obtained from these actions and environmental feedback, where the rewards include the current system performance evaluation results. In each training iteration, the decision engine randomly selects a batch of experience data from the experience replay buffer, then uses these data to calculate the gradient, and updates the weights and biases of the policy generation network and the value network through optimization algorithms (deep deterministic policy gradient or approximate policy optimization). Specifically, when the system finds that a certain beamforming strategy leads to high throughput and low latency under certain environmental conditions, the policy generation network will tend to choose similar actions in the next decision. Conversely, if a certain strategy leads to high power consumption or strong interference, the value network will evaluate it as a low-return action, thereby reducing the probability of the policy generation network selecting that action.

[0090] This iterative learning process ensures that the system can adapt to long-term environmental changes and continuously improve performance. Over time and as the environment evolves, the system achieves self-optimization and adjustment of its strategy by continuously interacting with the real communication environment, collecting new experiences, evaluating performance, and updating the decision model. This adaptive and self-evolutionary capability enables the system to continuously learn and evolve better beamforming strategies in complex and variable millimeter wave propagation environments such as high-density users, dynamic obstacle blocking, and rapidly changing channels, thereby significantly improving the reliability, efficiency, and user experience quality of millimeter wave communication. Through this closed-loop learning and optimization mechanism, the millimeter wave phased array antenna communication system realizes truly intelligent communication management and can cope with the growing needs of future communication networks.

[0091] The above merely describes the preferred embodiments of the present application, but is not intended to limit the present application in other forms. Any person skilled in the art can make changes or modifications to the above disclosed technical contents into equivalent embodiments with equivalent changes, and apply to other fields. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application without departing from the technical solution content of the present application still belongs to the protection scope of the present application technical solution.

Claims

1. A millimeter wave phased array antenna communication system, characterized by, The application relates to a cognitive beamforming system for a communication system, comprising: an environment perception and data collection unit for acquiring various real-time data of an environment where the communication system is located; a feature coding and prediction unit in communication connection with the environment perception and data collection unit, for receiving the real-time data and performing feature extraction, coding and future state prediction on the real-time data to obtain an environment state feature vector and predicted future environment state information, wherein the feature coding and prediction unit adopts a sequence prediction model and predicts future positions of user equipment, future moving paths of obstacles and future changes of channel states in a future time period based on current and historical environment state feature vectors to obtain the predicted future environment state information; a cognitive beamforming decision unit in communication connection with the feature coding and prediction unit, for receiving the environment state feature vector and predicted future environment state information and generating an optimized beamforming strategy based on preset optimization objectives and constraint conditions; an adaptive hybrid beamforming control unit in communication connection with the cognitive beamforming decision unit and a millimeter wave phased array antenna unit, for converting the optimized beamforming strategy into hardware control instructions executable by the millimeter wave phased array antenna unit, wherein the adaptive hybrid beamforming control unit comprises a subarray configuration management module which dynamically adjusts a connection relationship and resource allocation between radio frequency links and antenna elements according to the beamforming strategy; and a communication interface unit in communication connection with the millimeter wave phased array antenna unit, for receiving the hardware control instructions and performing transmission and reception of millimeter wave signals, and for performing data interaction with an upper-layer network protocol stack. The real-time data comprises channel state information, user equipment positions, user equipment speeds, user equipment directions, surrounding obstacle distribution, interference source positions and powers and system self-operation state parameters, and the environment perception and data collection unit further performs time stamp synchronization, data alignment and preliminary fusion on all collected original data to form an original environment data set in a unified format.

2. The millimeter-wave phased array antenna communication system of claim 1, wherein: The cognitive beamforming decision unit is specifically a decision engine based on deep reinforcement learning, and the cognitive beamforming decision unit optimizes the beamforming strategy by maximizing long-term cumulative rewards.

3. The mmWave phased array antenna communication system of claim 1, wherein: The function expression of the cumulative rewards is as follows:

4. The mmWave phased array antenna communication system of claim 3, wherein: The millimeter wave phased array antenna unit comprises antenna elements, a radio frequency front end module and a direct current power management module; the antenna elements are planar microstrip patch antennas or Vivaldi antennas; the radio frequency front end module comprises a power amplifier, a low-noise amplifier, a mixer, a filter and a digital-to-analog converter array; and the direct current power management module provides stable and adjustable direct current power for the radio frequency front end module. wherein R denotes the cumulative reward, T denotes the total number of time steps, t denotes a time step, γ denotes a discount factor, and the denotes the system throughput at time step t; denotes the communication latency at time step t, denotes the received interference power at time step t; denotes the total power consumption of the system at time step t; and denote the weight coefficients of the corresponding target terms, and the denotes the penalty term imposed when the beamforming strategy violates the preset constraint condition at time step t; the beamforming strategy includes beam direction, beam width, sidelobe suppression, power allocation, and subarray configuration scheme of millimeter wave phased array antenna elements.​​ 5. The mmWave phased array antenna communication system of claim 1, wherein, The application further relates to a cognitive beamforming method for a communication system, comprising the following steps: 6.A millimeter wave phased array antenna communication method, characterized by, ​ S1: Obtain multiple real-time data of the environment where the communication system is located, and synchronize the time stamp, align the data, and preliminarily fuse the original data to form a unified format of the original environment data set; S2: Receive the original environment data set, and perform feature extraction, encoding, and future state prediction to obtain an environment state feature vector and predicted future environment state information, wherein, in the future state prediction, a sequence prediction model is used, and based on the current and historical environment state feature vectors, the future position of the user equipment, the future moving path of the obstacles, and the future change of the channel state in a certain time period are predicted; S3: Receive the environment state feature vector and the predicted future environment state information, and generate an optimized beamforming strategy based on a preset optimization target and constraint condition; S4: Convert the beamforming strategy into hardware control instructions executable by the millimeter wave phased array antenna unit, wherein the conversion step includes dynamically adjusting the connection relationship and resource allocation between the radio frequency link and the antenna array element according to the beamforming strategy; S5: The millimeter wave phased array antenna unit performs transmission and reception of millimeter wave signals according to the hardware control instructions; S6: Real-time evaluation is performed on the communication system performance under the current beamforming strategy, and the decision model of the cognitive beam strategy decision step is updated based on the evaluation result.

7. The millimeter-wave phased array antenna communication method of claim 6, wherein: In S1, the real-time data is collected by a global positioning system module, an inertial measurement unit module, a millimeter wave radar module, a visual sensor module, and a channel detection module, and the real-time data includes channel state information, user equipment position, speed, direction, surrounding obstacle distribution, interference source position and power, and system itself running state parameters.

8. The millimeter-wave phased array antenna communication method of claim 7, wherein: In S3, the decision engine generates several selectable beamforming actions according to the environment state, and evaluates the expected long-term return of these actions through a value network to select the optimal action; in S6, at least one of system throughput, communication delay, bit error rate, and signal-to-noise ratio is used as feedback information to update the strategy generation network and value network parameters of the decision engine.

Images