Phased-array antenna system-oriented TR module packaging method

Abstract

The invention relates to the technical field of integrated circuit packaging, active array antennas, machine learning and the like, and provides a TR module packaging method for a phased-array antenna system, which comprises the following steps of: packaging electronic components in a TR module shell, arranging a liquid cooling flow channel, and connecting an external liquid circulating device through an input port and an output port to realize heat dissipation; a temperature sensor and a machine learning module are arranged in the shell, temperature data are collected in real time, the heat distribution trend is predicted, and the flow speed and flow of cooling liquid are intelligently adjusted; a power supply interface, a control interface and a radio frequency signal interface are arranged on one side of the shell, so that standardized connection is realized; a plurality of TR modules are installed on the splitter plate in an array mode, and a liquid cooling path is managed in a unified mode. According to the invention, thermal load prediction and dynamic cooling adjustment can be realized, and local overheating can be effectively prevented.

Description

Technical Field

[0001] The present invention relates to technical fields such as integrated circuit packaging, active array antennas, and machine learning, and in particular to a TR module packaging method for a phased array antenna system. Background Art

[0002] In an active phased array antenna system that integrates a TR module, a power supply module, and a thermal control module, when the TR (transmitter / receiver) module is packaged, key components such as the RF amplifier, phase shifter, and low-noise amplifier (LNA) in the TR module are packaged inside a metal shell through a stacking method or a multi-chip module (MCM) structure, forming a heat-source concentrated structure. The heat dissipation method of a heat-source concentrated structure usually relies on an external cooling substrate (liquid-cooled cold plate) or a heat conduction sheet. The TR module contacts the liquid-cooled fixing plate below through screws or elastic gaskets, relying on heat conduction at the contact interface and removing heat through the flow channel on the liquid-cooled plate. It does not have the ability to actively predict thermal loads and only relies on the external coolant system to dissipate heat at a constant flow rate. It cannot respond to hot spots or instantaneous thermal changes.

[0003] In summary, existing TR module packaging technology for phased array antenna systems lacks thermal prediction and intelligent adjustment capabilities, has a slow response, and is unable to dynamically adjust cooling strategies based on instantaneous thermal loads, making it prone to local overheating under high-power or sudden operating conditions. Summary of the Invention

[0004] To address the shortcomings of the above-mentioned existing technologies, the present invention provides a TR module packaging method and system for phased array antenna systems, which can perform thermal prediction and intelligent regulation during the TR module packaging process, dynamically adjust the cooling strategy according to the instantaneous thermal load, and avoid local overheating under high-power or sudden operating conditions.

[0005] The TR module packaging method for a phased array antenna system provided by the present invention includes:

[0006] Electronic components are encapsulated inside the housing of the TR module, and a liquid cooling channel for direct heat dissipation is provided in the housing. The liquid cooling channel has an input port and an output port provided on the housing, and the input port and the output port are respectively connected to an external liquid circulation device;

[0007] A temperature sensor and a machine learning module are provided in the housing. The temperature sensor monitors temperature data generated by the operation of electronic components in real time and transmits the monitored data to the machine learning module. The machine learning module is pre-trained with a machine learning model for predicting heat distribution trends. The machine learning model analyzes historical temperature data and current temperature data inside the housing in real time, predicts the changing trend of heat distribution inside the TR module, and automatically adjusts the liquid flow rate and flow rate of the external liquid circulation device.

[0008] A unified standard communication interface is provided on the housing. The communication interface includes a power supply interface, a control interface, and a radio frequency signal interface. The power supply interface, the control interface, and the radio frequency signal interface are uniformly arranged on one side of the housing. The power supply interface and the control interface are respectively connected to the electronic components inside the TR module to provide power and control signals to the electronic components. The radio frequency signal interface is connected to the radio frequency signal processing circuit inside the electronic components to transmit radio frequency to the power splitter and feed network of the phased array antenna.

[0009] A plurality of the TR modules are mounted in an array on a manifold plate, and an infusion channel connected to the external liquid circulation device is provided in the manifold plate. The infusion channel is sealedly connected to the liquid cooling channel input port and output port on each TR module housing through the connection port on the manifold plate.

[0010] Compared with the prior art, the present invention has the following beneficial effects:

[0011] The present invention provides a TR module packaging method for a phased array antenna system, comprising: packaging electronic components inside a shell of the TR module, and providing a liquid cooling channel for direct heat dissipation inside the shell, wherein the liquid cooling channel has an input port and an output port provided on the shell, wherein the input port and the output port are respectively connected to an external liquid circulation device; providing a temperature sensor and a machine learning module inside the shell, wherein the temperature sensor monitors temperature data generated by the operation of the electronic components in real time and transmits the monitored data to the machine learning module, wherein a machine learning model for predicting heat distribution trends is pre-trained in the machine learning module, wherein the machine learning model analyzes historical temperature data and current temperature data inside the shell in real time, predicts the changing trend of heat distribution inside the TR module, and automatically adjusts the external liquid The liquid flow rate and flow of the circulation device are characterized by the following features: a unified standard communication interface is set on the shell, and the communication interface includes a power supply interface, a control interface and a radio frequency signal interface. The power supply interface, the control interface and the radio frequency signal interface are uniformly arranged on one side of the shell. The power supply interface and the control interface are respectively connected to the electronic components inside the TR module to provide power and control signals for the electronic components. The radio frequency signal interface is connected to the radio frequency signal processing circuit inside the electronic components to transmit radio frequency to the power division and feed network of the phased array antenna; multiple TR modules are installed in an array on a diverter plate, and an infusion channel connected to the external liquid circulation device is set in the diverter plate. The infusion channel is sealed and connected to the liquid cooling channel input port and output port on each TR module shell through the connection port on the diverter plate. The present invention realizes standardized connection by setting power supply interface, control interface and radio frequency signal interface on one side of the shell, and uniformly manages the liquid cooling path by installing multiple TR modules in an array on the diverter plate. By arranging temperature sensors and machine learning modules in the shell, temperature data is collected in real time and heat distribution trends are predicted, so as to intelligently adjust the coolant flow rate and flow, realize heat load prediction and dynamic cooling adjustment, and effectively prevent local overheating. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their description are used to explain the present invention and do not constitute an undue limitation of the present invention. Some specific embodiments of the present invention will be described in detail in an illustrative and non-restrictive manner with reference to the drawings. The same reference numerals in the drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the drawings:

[0013] Figure 1 The figure is a flow chart of a TR module packaging method for a phased array antenna system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0015] See also Figure 1 The embodiment of the present invention provides a TR module packaging method for a phased array antenna system, comprising the following steps:

[0016] S101. Encapsulate electronic components within a housing of a TR module, and provide a liquid cooling channel within the housing for direct heat dissipation, wherein the liquid cooling channel has an input port and an output port provided on the housing, wherein the input port and the output port are respectively connected to an external liquid circulation device;

[0017] S102. Disposing a temperature sensor and a machine learning module within the housing. The temperature sensor monitors temperature data generated by the operation of electronic components in real time and transmits the monitored data to the machine learning module. The machine learning module is pre-trained with a machine learning model for predicting heat distribution trends. The machine learning model analyzes historical temperature data and current temperature data within the housing in real time to predict a changing trend of heat distribution within the TR module, thereby automatically adjusting the liquid flow rate and flow rate of the external liquid circulation device.

[0018] S103. Disposing a unified standard communication interface on the housing, the communication interface including a power supply interface, a control interface, and a radio frequency signal interface. The power supply interface, the control interface, and the radio frequency signal interface are uniformly arranged on one side of the housing. The power supply interface and the control interface are respectively connected to electronic components inside the TR module to provide power and control signals to the electronic components. The radio frequency signal interface is connected to the radio frequency signal processing circuit inside the electronic components to transmit radio frequency to the power splitter and feed network of the phased array antenna.

[0019] S104. Install multiple TR modules in an array on a manifold. An infusion channel connected to the external liquid circulation device is provided in the manifold. The infusion channel is sealedly connected to the liquid cooling channel input port and output port on each TR module housing through the connection port on the manifold.

[0020] It should be noted that in this embodiment, a power supply interface, a control interface and a radio frequency signal interface are set on one side of the shell to achieve standardized connection, and multiple TR modules are installed in an array on the manifold to uniformly manage the liquid cooling path. By arranging temperature sensors and machine learning modules in the shell, temperature data is collected in real time and heat distribution trends are predicted, so as to intelligently adjust the coolant flow rate and flow, realize heat load prediction and dynamic cooling adjustment, and effectively prevent local overheating.

[0021] It should be noted that during operation, the TR module generates a large amount of heat due to high-power RF signal processing and frequent switching operations. If heat is not dissipated promptly and effectively, it can easily lead to excessive local temperature rise, affecting module performance and even causing device damage. Furthermore, inconsistent interface layouts and heat dissipation connection methods can lead to complex system wiring, difficult assembly and maintenance, and difficulties in achieving dynamic management of multi-module integration and heat load. In this embodiment, liquid cooling channels are pre-set within the housing and connected to an external liquid circulation device to achieve efficient heat exchange for the TR module. Temperature sensors are used to collect real-time temperature data, and heat distribution trends are predicted using a trained machine learning model to dynamically adjust the coolant flow rate and volume to avoid overcooling or local overheating, achieving on-demand cooling. Furthermore, the power supply, control, and RF signal interfaces are uniformly arranged on one side of the housing, reducing system integration complexity and improving maintainability and inter-module interchangeability. Furthermore, this embodiment supports array assembly of multiple TR modules, and centralized management of liquid cooling paths is achieved through a manifold, facilitating unified cooling system wiring and fluid regulation control.

[0022] Preferably, when the machine learning model predicts the changing trend of the heat distribution inside the TR module, the algorithm adopted is a recursive neural network model; the recursive neural network model takes the historical temperature series data as input and is expressed as: {T n-1 , T n-2 ,...,T n-m};T n-k represents the temperature sampling value at the nkth time point, and m is the number of historical backtracking steps; the recursive neural network model predicts the changing trend of the heat distribution inside the TR module based on the historical temperature series data, and is trained using the mean square error loss function to improve the prediction accuracy of the machine learning model for the heat distribution trend. The mean square error loss function is:

[0023]

[0024] Among them, T i is the actual temperature value of the i-th sample, is the temperature value predicted by the model, and N is the total number of samples.

[0025] It should be noted that when the TR module is working at high power for a long time, the internal heat distribution is time-varying and spatially non-uniform. If the cooling strategy is adjusted solely based on the current instantaneous temperature data, it is easy to cause adjustment lag or over-response, affecting the efficiency of the cooling system and the thermal safety of the module. In this embodiment, the use of recursive neural networks to model the historical temperature series can more accurately capture the temporal variation of the heat distribution and predict future temperature trends, thereby achieving higher prediction accuracy than linear regression or rule models. By predicting future heat changes in advance through the recursive neural network model, the flow rate or flow of the coolant can be dynamically adjusted before overheating occurs, effectively alleviating the thermal lag problem and preventing the generation of local hot spots. It can be understood that the recursive neural network model has generalization capabilities through training, can adapt to thermal behavior under different working modes, and realize intelligent, data-driven cooling control strategies.

[0026] Preferably, the flow regulation inside the liquid cooling channel adopts a flow control algorithm to accurately control the flow rate and flow of the coolant. The flow control algorithm is:

[0027]

[0028] In the algorithm, u(t) is the flow control output, e(t) is the temperature error, K p is the proportional gain, K i is the integral gain, K d is the differential gain.

[0029] It should be noted that in the phased array antenna system, the thermal management of the TR module under high-frequency and high-power operation is the key guarantee for the reliability and performance of the system. In the actual control process, the response of the liquid circulation device is easily affected by disturbances (such as changes in flow resistance and temperature rise inertia). If the control strategy is not accurate enough, problems such as control overshoot, response lag, and temperature fluctuation may occur. In this embodiment, the flow control algorithm can respond in time according to the size of the temperature difference, quickly adjust the liquid flow rate, and quickly intervene in the cooling process when the temperature deviates from the set value, effectively shortening the temperature recovery time. Through the flow control algorithm, the integral term can be used to compensate for the steady-state error, and the differential term can be used to suppress rapid fluctuations, making the cooling system control process smoother and preventing frequent adjustments caused by temperature fluctuations. It should be noted that in this embodiment, the flow control algorithm can serve as a control execution supplement for the machine learning model to predict thermal trends, forming a complete closed loop from prediction to decision-making to execution, greatly improving the real-time and coordination of thermal regulation.

[0030] Preferably, the infusion channel includes a liquid separation channel and a liquid collection channel, and the flow relationship between the liquid separation channel and the liquid collection channel satisfies: Q 集 is the total flow rate of the liquid collecting channel, Q 分，iis the flow rate of the i-th liquid separation channel, and n is the number of TR modules. It should be noted that in the scenario where multiple TR module arrays are installed on the same manifold, the liquid circulation device needs to distribute the coolant to each TR module through the liquid separation channel, and then uniformly recover the hot liquid through the liquid collection channel. However, if the flow is unbalanced, it is easy for some modules to be over-cooled and some modules to be under-cooled, affecting the overall thermal balance of the phased array antenna system, and even causing local overheating and decreased system efficiency due to inconsistent flow resistance. In this embodiment, the flow relationship between the liquid separation channel and the liquid collection channel is set to meet the following requirements: Q 集 is the total flow rate of the liquid collecting channel, Q 分，i is the flow rate of the i-th liquid distribution channel, and n is the number of TR modules, thereby ensuring that the sum of the coolant input flow rate of each TR module in the liquid cooling system is equal to the total coolant confluence flow rate, satisfying the continuity equation of fluid mechanics and achieving flow conservation.

[0031] Preferably, the TR module housing comprises two upper and lower housing components, each with a groove on its inner surface. When the two housing components are assembled, the grooves interlock to form a liquid-cooling channel, and the housing is sealed by brazing. It should be noted that existing TR module liquid cooling systems often use external cold plates or independent liquid-cooling structures bonded to the housing surface to achieve heat dissipation. This not only results in complex structure, cumbersome assembly, and increased thermal resistance, but also presents problems such as unreliable sealing performance, redundant volume, and detours in the liquid cooling path. In highly integrated, compactly arranged phased array antenna systems, achieving efficient, compact, and tightly sealed liquid-cooling channel integration without increasing volume is a key challenge in optimizing the packaging structure of such TR modules. In this embodiment, the liquid-cooling channel is designed as a groove structure on the inner surfaces of each of the upper and lower housing components. When assembled, it forms a completely enclosed liquid-cooling channel. This eliminates the need for additional piping or cold plates; the housing structure itself performs the function of the cooling channel, achieving structural and functional integration. Furthermore, brazing connects the upper and lower housings, ensuring good airtightness and liquid-tightness at the junction, ensuring the safe transmission of high-pressure coolant within the channel and preventing leakage. Furthermore, the size and path of the groove are designed based on the simulation analysis results of the steady-state or quasi-steady-state thermal power density distribution generated by the heat source element when the TR module is in operation, so as to ensure that the shell has an optimized heat conduction path and improve the overall heat dissipation capacity of the TR module; the heat source element includes a power amplifier, a low-noise amplifier, a phase shifter, and a transceiver switch. It should be noted that in the highly integrated and high-power-density operating environment of the TR module, the heat generated by core components such as the power amplifier (PA), low-noise amplifier (LNA), phase shifter, and transceiver switch at different locations is unevenly distributed. If the liquid-cooling channel structure remains unchanged or these non-uniform heat source distributions are not fully considered, it may lead to insufficient local cooling, accumulation of hot spots, and waste of cooling resources in low heat density areas. The overall temperature rise of the TR module cannot be effectively controlled, thereby affecting performance and life. In this embodiment, simulation analysis technology is used to obtain the thermal power distribution of different electronic devices inside the TR module under stable or quasi-stable working conditions. When designing the flow path and dimensions, the flow density and flow rate of the coolant flowing through key high-heat-generating areas such as power amplifiers, low-noise amplifiers, phase shifters, and transceiver switches can be prioritized based on the simulation analysis results, and thermal load matching can be achieved by adjusting the groove position, depth, and width.

[0032] Preferably, the temperature sensor utilizes a thermistor array, with each thermistor located corresponding to a heat source element within the TR module. The thermistor array monitors the temperature data generated by the heat source element in real time and transmits the monitored data to the machine learning module. It should be noted that within the TR module, multiple heating elements (such as power amplifiers, low-noise amplifiers, phase shifters, and transceiver switches) typically have discrete spatial distributions, asynchronous thermal load changes, and complex heating dynamics. Using only a single-point temperature sensor may not accurately sense the real-time temperature of each key element. Temperature data sampling delays or omissions can affect the prediction model's judgment and prevent rapid early warning and regulatory response to local overheating risks. In this embodiment, the temperature sensor utilizes a thermistor array, with each thermistor acting as an independent temperature measurement point. These thermistors offer small size, high sensitivity, and fast response, making them suitable for placement close to components to achieve point-to-point temperature sensing. Each thermistor's physical location corresponds one-to-one with a major heating element within the TR module (PA, LNA, phase shifter, etc.), ensuring that the actual temperature rise of the heat source is captured, rather than the average or diffracted temperature away from the heat source. The thermistor array uploads multi-point temperature data to the machine learning module in real time. Multi-point temperature data input can improve the machine learning model's ability to spatially understand changes in heat distribution and enhance the model's prediction accuracy for local overheating trends.

[0033] Preferably, the arrangement density and position of the temperature sensor array are designed based on the simulation analysis results of the steady-state or quasi-steady-state thermal power density distribution generated by the heat source element to ensure the response speed and overall heat dissipation efficiency of high-precision heat dissipation control. It should be noted that in TR modules with high power density and uneven heat source distribution, there are significant differences in the amount and spatial distribution of heat generated by different electronic devices during operation. If the temperature sensor array is not arranged in an appropriate position, it will cause a decrease in monitoring accuracy, the temperature rise in key hot zones cannot be perceived in time, the error in temperature prediction results will increase, the input of the machine learning model will be distorted, the cooling system will not respond in time, and it is easy to cause problems such as overheating or energy waste. In this embodiment, the thermal power density distribution (i.e., heat source intensity + spatial position) of each heating element inside the TR module under steady-state or quasi-steady-state conditions is obtained through simulation analysis technology (such as finite element simulation, CFD calculation, etc.), and the temperature measurement points are customized based on the thermal density distribution. More sensors are set in high heat density areas to improve the thermal response monitoring density. Sensors are configured in areas with drastic thermal gradient changes to improve the sensitivity of overheating trend identification. The number of points can be appropriately reduced in low heat areas to achieve system resource optimization, improve the spatial sensitivity and timeliness of the system as a whole to thermal changes, and provide an accurate and real-time data basis for cooling regulation and thermal prediction models.

[0034] Preferably, the machine learning module uses a gradient descent optimization strategy to perform online fine-tuning on the model parameters of the machine learning model during operation. The gradient descent optimization strategy has a gradient update method with a momentum term. The update formula of the gradient update method with a momentum term is: θ t represents the model parameters of the tth iteration, η is the learning rate, γ is the momentum factor, is the current gradient. It should be noted that when the recursive neural network model based on historical temperature series data is used to predict the heat distribution trend, factors such as the working environment, heat source load, and liquid cooling efficiency of the TR module will change dynamically, making it difficult for the original training model to maintain high prediction accuracy for a long time, and easily causing the model to have prediction deviations under new working conditions, and environmental drift causing changes in thermal behavior. In this embodiment, a gradient descent optimization algorithm with a momentum term is used for online fine-tuning training of the machine learning model after deployment. During operation, the model parameters are slightly adjusted according to the new input data and the actual observation error, rather than offline batch training. At the same time, the gradient update strategy with a momentum term can accelerate convergence, reduce oscillations, and to a certain extent jump out of the local optimum, so that during the actual operation of the TR module, the machine learning model parameters can be continuously and dynamically fine-tuned and optimized, thereby improving its prediction robustness and generalization ability in a changing environment.

[0035] Furthermore, during operation, the machine learning module uses an adaptive momentum gradient descent optimization strategy with a local thermal anomaly weighting term to fine-tune the parameters of the thermal prediction model online based on the local temperature change trends collected in real time by the multi-point temperature sensor array inside the TR module. In each round of parameter update, the optimization algorithm dynamically adjusts the momentum factor and learning rate according to the temperature gradient fluctuation amplitude in the target area to improve the prediction sensitivity in areas with high thermal disturbances. The update formula in the adaptive momentum gradient descent optimization strategy is:

[0036]

[0037] in, Represents the learning rate that is dynamically adjusted as the local temperature gradient changes; represents the momentum factor that is adaptively adjusted as the variance of the local temperature disturbance changes; Predict the gradient for the current temperature, is the standard deviation of the temperature prediction residual in the recent time window; η0 is the baseline learning rate, γ0 is the momentum factor, and β and δ are the adjustment coefficients. In the learning rate It means that when the temperature change rate of a certain area is greater, that is, when a sudden thermal disturbance occurs, the learning rate will increase accordingly, thereby prompting the model to converge to the new thermal prediction state more quickly. middle, Represents the standard deviation of the local temperature prediction residual. If the prediction error in this area fluctuates greatly, it means that the thermal behavior in this area is unstable. In this way, the momentum term is increased to improve the prediction inertia and buffer adjustment ability to reduce oscillations. In this way, this embodiment not only improves the rapid response capability of the prediction model in hot spots, but also improves the stability and adaptability of the entire thermal prediction system in a dynamic environment. Compared with the fixed hyperparameter gradient descent algorithm, the update formula in the adaptive momentum gradient descent optimization strategy can form a locally differentiated model update mechanism based on the spatial variation characteristics of the heat source inside the TR module and the real-time performance feedback of the prediction model, thereby achieving a high degree of coupling between structured thermal control and intelligent temperature control strategy, and avoiding the model from falling into local optimality or response hysteresis.

[0038] It should be noted that in this embodiment, based on the temperature information collected in real time by the multi-point temperature sensor array inside the TR module, the standard gradient descent algorithm with a fixed learning rate and momentum factor is no longer used. Instead, an adaptive momentum gradient descent optimization strategy driven by local thermal disturbance perception is introduced to achieve dynamic adjustment of the model update process. Through the adaptive momentum gradient descent optimization strategy, in each round of model parameter update, the learning rate and momentum factor can be dynamically adjusted based on the temperature gradient fluctuation amplitude (i.e., temperature change rate) and the variance of the temperature prediction error (i.e., the standard deviation of the prediction residual) in the target area. This makes the model more responsive in areas with high thermal disturbances, thereby adapting to the changing trend of local heat distribution more quickly.

[0039] Preferably, the target coolant flow rate Q of each liquid cooling channel sub-segment of the TR module is i The target coolant flow rate Q is automatically adjusted according to the predicted regional temperature gradient function. i The resource optimization allocation formula is satisfied to achieve optimal allocation of local area cooling resources. The resource optimization allocation formula is:

[0040]

[0041] In the formula, Represents the predicted regional temperature gradient function at position x i The value of , α is the adjustment coefficient.

[0042] It should be noted that in a TR module with multiple heat sources and high heat flux density, the heat distribution is highly uneven. Even if the liquid cooling system has good heat dissipation capabilities as a whole, if the flow rate configuration of the coolant at different positions is unreasonable (such as: insufficient flow rate in the high-temperature gradient area), it may still lead to low system heat dissipation efficiency, high energy consumption or excessive temperature rise. Traditional fixed flow or constant flow rate designs are difficult to accurately adapt to dynamic thermal changes. In this embodiment, through the resource optimization allocation formula, the position with more drastic temperature changes (larger gradient) can be allocated a larger coolant flow rate, and the position with a small temperature gradient (thermal stable area) can reduce the flow rate to save energy, thereby solving the problems of uneven distribution of cooling resources, lag in control and low energy efficiency inside the TR module, realizing dynamic local regulation driven by heat distribution, and building a more efficient, intelligent and real-time responsive liquid cooling system, which is of great value to improving the stability and life of the TR module in a complex thermal environment.

[0043] Furthermore, the liquid separation channel and the liquid collection channel of the liquid infusion channel in the manifold are respectively connected to the liquid cooling channel input port and output port of multiple TR modules, and the cross-sectional size A of each liquid separation channel is i Satisfy the preset proportional relationship, the preset proportional relationship is: A i ∝ρ i ; Among them, ρ i is the heat source density in the i-th TR module, and the cross-sectional area of ​​the cross section of the liquid distribution channel is positively correlated with the heat source power density to improve the heat dissipation capacity of the high heat source area. It should be noted that in the liquid cooling system of the TR module array, the heat source power distribution is inconsistent among multiple modules. Some modules (such as modules equipped with high-power amplifiers) have high heat loads, while other modules may have lower heat loads. If the liquid distribution channel is designed uniformly in structure, the coolant flow distribution is difficult to reflect the "heat load difference", which will cause insufficient cooling in high heat density areas, easy formation of hot spots, waste of cooling resources in low heat areas, reduced system energy efficiency, and decreased overall system thermal management performance. In this embodiment, the liquid distribution channel and the liquid collection channel of the infusion channel in the diverter plate are respectively connected to the liquid cooling channel input port and output port of multiple TR modules, and the cross-sectional size A of each liquid distribution channel is large. i Satisfy the preset proportional relationship, the preset proportional relationship is: A i ∝ρ i ,ρ i is the heat source density in the i-th TR module. The cross-sectional area of ​​the cross section of the liquid separation channel is positively correlated with the heat source power density. Therefore, the cross-sectional size of the liquid separation channel corresponding to the module with higher heat source density is enlarged so that it can obtain more coolant. For the module with lighter heat load, the cross-sectional area is reduced accordingly to reduce the waste of cooling resources. The liquid separation channel structure is differentiated according to the heat source power density of each module, and the heat dissipation resource optimization matching at the structural level is achieved.

[0044] Furthermore, the five-fold cross validation method was used to verify the generalization ability of the machine learning model and to verify the error function E val for:

[0045]

[0046] Among them, K is the number of cross-validation folds, N k is the number of samples in the k-th fold, is the actual temperature, To predict temperature.

[0047] It should be noted that in the TR module intelligent thermal management system, machine learning models (such as RNN models for heat distribution trend prediction) play a key decision-making role, and their prediction accuracy directly affects the response quality of cooling regulation and the accuracy of thermal control. However, in actual design, there is a model that may perform well on training data, but fail to predict data in a new environment (overfitting), and there is a lack of a unified standard to measure the generalization performance of different models or hyperparameter combinations; if the model generalization ability is poor, it may lead to cooling strategy errors, causing local overheating or energy waste. In this embodiment, the verification error function E val It can provide the mean prediction error of the machine learning model on unseen data, avoiding misjudging the model's capabilities by relying solely on training errors. Through comprehensive analysis of the training and verification results of multiple subsets, it ensures that the model has stable prediction capabilities under various working conditions. With improved prediction accuracy, the liquid cooling flow rate adjustment of the drive will be more precise, effectively preventing overcooling or overheating, and improving the operating stability of the TR module in high-power applications. This allows for a systematic evaluation of the model's generalization capabilities under unknown working conditions before deployment, ensuring the stability and reliability of the TR module thermal prediction system.

[0048] It should be pointed out that the above embodiments are only preferred specific implementation methods of the present invention, and the protection scope of the present invention is not limited thereto. Any changes or replacements that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. The protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A TR module packaging method for a phased array antenna system, characterized in that: include: Electronic components are encapsulated inside the housing of the TR module, and a liquid cooling channel for direct heat dissipation is provided in the housing. The liquid cooling channel has an input port and an output port provided on the housing, and the input port and the output port are respectively connected to an external liquid circulation device; A temperature sensor and a machine learning module are provided in the housing. The temperature sensor monitors temperature data generated by the operation of electronic components in real time and transmits the monitored data to the machine learning module. The machine learning module is pre-trained with a machine learning model for predicting heat distribution trends. The machine learning model analyzes historical temperature data and current temperature data inside the housing in real time, predicts the changing trend of heat distribution inside the TR module, and automatically adjusts the liquid flow rate and flow rate of the external liquid circulation device. A unified standard communication interface is provided on the housing. The communication interface includes a power supply interface, a control interface, and a radio frequency signal interface. The power supply interface, the control interface, and the radio frequency signal interface are uniformly arranged on one side of the housing. The power supply interface and the control interface are respectively connected to the electronic components inside the TR module to provide power and control signals to the electronic components. The radio frequency signal interface is connected to the radio frequency signal processing circuit inside the electronic components to transmit radio frequency to the power splitter and feed network of the phased array antenna. A plurality of the TR modules are mounted in an array on a manifold plate, and an infusion channel connected to the external liquid circulation device is provided in the manifold plate. The infusion channel is sealedly connected to the liquid cooling channel input port and output port on each TR module housing through the connection port on the manifold plate.

2. The TR module packaging method for a phased array antenna system according to claim 1, characterized in that: When the machine learning model predicts the changing trend of the heat distribution inside the TR module, the algorithm adopted is a recursive neural network model; the recursive neural network model takes the historical temperature series data as input and is expressed as: {T n-1 , T n-2 ,...,T n-m };T n-k represents the temperature sampling value at the nkth time point, and m is the number of historical backtracking steps; the recursive neural network model predicts the changing trend of the heat distribution inside the TR module based on the historical temperature series data, and is trained using the mean square error loss function to improve the prediction accuracy of the machine learning model for the heat distribution trend. The mean square error loss function is: Among them, T i is the actual temperature value of the i-th sample, is the temperature value predicted by the model, and N is the total number of samples.

3. The TR module packaging method for a phased array antenna system according to claim 1, characterized in that: The flow regulation inside the liquid cooling channel adopts a flow control algorithm to accurately control the flow rate and flow of the coolant. The flow control algorithm is: In the algorithm, u(t) is the flow control output, e(t) is the temperature error, K p is the proportional gain, K i is the integral gain, K d is the differential gain.

4. The TR module packaging method for a phased array antenna system according to claim 1, wherein: The infusion channel includes a liquid separation channel and a liquid collection channel, and the flow relationship between the liquid separation channel and the liquid collection channel satisfies: Q 集 is the total flow rate of the liquid collecting channel, Q 分，i is the flow rate of the i-th liquid separation channel, and n is the number of TR modules.

5. The TR module packaging method for a phased array antenna system according to claim 1, wherein: The TR module housing includes two upper and lower housing parts. The inner surface of each housing part is provided with a groove. When the two housing parts are combined, the grooves are buckled to form a liquid cooling channel, and the housing is sealed by brazing.

6. The TR module packaging method for a phased array antenna system according to claim 5, characterized in that: The size and path of the groove are designed based on the simulation analysis results of the steady-state or quasi-steady-state thermal power density distribution generated by the heat source element when the TR module is in operation, so as to ensure that the housing has an optimized heat conduction path and improve the overall heat dissipation capacity of the TR module; the heat source element includes a power amplifier, a low-noise amplifier, a phase shifter, and a transceiver switch.

7. The TR module packaging method for a phased array antenna system according to claim 6, characterized in that: The temperature sensor uses a thermistor array, and each thermistor position corresponds to the heat source element inside the TR module. It monitors the temperature data generated by the heat source element in real time and transmits the monitored data to the machine learning module.

8. The TR module packaging method for a phased array antenna system according to claim 7, characterized in that: The arrangement density and position of the temperature sensor array are designed based on the simulation analysis results of the steady-state or quasi-steady-state thermal power density distribution generated by the heat source element to ensure the response speed and overall heat dissipation efficiency of high-precision heat dissipation control.

9. The TR module packaging method for a phased array antenna system according to claim 1, characterized in that: The machine learning module uses a gradient descent optimization strategy to perform online fine-tuning on the model parameters of the machine learning model during operation. The gradient descent optimization strategy has a gradient update method with a momentum term. The update formula of the gradient update method with a momentum term is: θ t represents the model parameters of the tth iteration, η is the learning rate, γ is the momentum factor, is the current gradient.

10. The TR module packaging method for a phased array antenna system according to claim 1, characterized in that: The target coolant flow rate Q of each liquid cooling channel sub-segment of the TR module i The target coolant flow rate Q is automatically adjusted according to the predicted regional temperature gradient function. i The resource optimization allocation formula is satisfied to achieve optimal allocation of local area cooling resources. The resource optimization allocation formula is: In the formula, Represents the predicted regional temperature gradient function at position x i The value of , α is the adjustment coefficient.

Images