Microwave solid-state power amplifier thermal-aware dynamic bias adjustment method and implementation system

Abstract

The application discloses a microwave solid-state power amplifier thermal perception dynamic bias adjustment method and an implementation system, and belongs to the technical field of microwave solid-state power amplifiers. The method comprises the following steps: collecting active region thermal image time sequence data and electrical parameter time sequence data of the microwave solid-state power amplifier in real time; establishing a thermal gradient field according to the thermal image time sequence data, and performing time-frequency decomposition to extract thermal inertia components corresponding to multiple thermal memory depths; constructing a thermal-electric coupling state space model with the thermal inertia components as state inputs; taking the electrical parameter time sequence data as excitation inputs, recursively estimating a transient thermal impedance matrix at the current moment through Kalman filtering; based on the transient thermal impedance matrix and a preset junction temperature safety boundary, solving a multi-step prediction control sequence of bias adjustment amounts; and synchronously adjusting a gate bias voltage and a drain bias voltage according to a first step adjustment amount in the sequence. The system implements the above method.

Description

Technical Field

[0001] This invention relates to the field of microwave solid-state power amplifier technology, specifically to a method and system for thermal sensing dynamic bias adjustment of microwave solid-state power amplifiers. Background Technology

[0002] The junction temperature variation of microwave solid-state power amplifiers directly affects their electrical performance and reliability during operation. Most existing bias adjustment techniques rely on temperature sensors to collect case temperature or local temperature data, using a fixed mapping relationship to compensate for the bias. This open-loop approach struggles to respond to rapid, non-uniform changes in thermal state. Some solutions use simplified thermal models to predict junction temperature, but these typically assume the thermal system has only a single time constant, neglecting the coexistence of multiple thermal inertia phenomena caused by different thermal paths within the transistor's active region. This results in delayed junction temperature prediction and significant errors when the signal envelope fluctuates rapidly or the duty cycle changes.

[0003] Current technologies lack methods to separate thermal memory features at different time scales from the spatiotemporal distribution of thermal images, making it impossible to construct dynamic models that can simultaneously characterize short-term and long-term thermal inertia effects. When local hotspots rapidly form in thermal images, single-inertial-loop models cannot capture their steep temperature rise gradients and subsequent slow residual temperature decay processes, leading to a mismatch between bias adjustment actions and the actual thermal state. Furthermore, junction temperature thermal management largely relies on current detection values ​​to trigger protection, lacking advance constraints on future junction temperature trends. Especially under deteriorating heat dissipation conditions or continuous high-power excitation, junction temperatures are prone to exceeding safety boundaries.

[0004] In microwave solid-state power amplifiers, the gate and drain bias adjustments under thermal shock exhibit coupled effects. Existing technologies often adjust these biases independently or only one bias at a time, failing to fully consider their combined impact on power dissipation and gain. The challenges in this field include: how to extract multi-depth thermal inertia components from multi-point thermal gradient sensing and establish a state-space model describing thermal memory; and how to generate a multi-step bias adjustment sequence that proactively constrains junction temperature and coordinates gate and drain synchronization based on the real-time identified transient thermal impedance matrix. Summary of the Invention

[0005] This invention provides a method and system for dynamic bias adjustment of microwave solid-state power amplifier based on thermal sensing. The purpose is to extract multi-timescale thermal inertial components from the spatiotemporal information of the active region thermal image, establish a thermoelectric coupling model reflecting the thermal memory effect, and implement multi-step predictive coordinated bias adjustment by adaptively estimating the transient thermal impedance matrix, so that the junction temperature is constrained within the safe boundary.

[0006] The objective of this invention can be achieved through the following technical solutions: This invention provides a thermal sensing dynamic bias adjustment method for a microwave solid-state power amplifier, comprising: real-time acquisition of thermal imaging time-series data and electrical parameter time-series data of the active region of the microwave solid-state power amplifier; establishing a thermal gradient field based on the thermal imaging time-series data, and performing time-frequency decomposition on the thermal gradient field to extract thermal inertial components corresponding to multiple thermal memory depths; constructing a thermoelectric coupling state-space model with the thermal inertial components as state inputs; using the electrical parameter time-series data as excitation inputs to the thermoelectric coupling state-space model, and recursively estimating the transient thermal impedance matrix at the current moment through Kalman filtering; solving a multi-step predictive control sequence for the bias adjustment amount based on the transient thermal impedance matrix and a preset junction temperature safety boundary; and synchronously adjusting the gate bias voltage and drain bias voltage of the microwave solid-state power amplifier according to the first adjustment amount in the multi-step predictive control sequence.

[0007] By jointly analyzing high thermal spatial resolution infrared thermal images with electrical parameters, thermal inertial components with different decay characteristics are separated from the thermal gradient field. This enables the accurate capture of multi-timescale thermal memory effects in the active region, from the chip level to the package level, providing key state variables for establishing a thermal model that closely reflects the physical essence. A thermoelectric coupled state-space model is constructed by concatenating short-time and long-time thermal inertial components to form a state vector and using dissipated power as input. The state transition matrix is ​​formed by diagonalizing the reciprocal of the thermal time constant corresponding to each thermal inertial component, directly characterizing the natural decay behavior of heat at adjacent time points. This gives the junction temperature prediction output clear thermophysical meaning and high dynamic accuracy.

[0008] In the recursive estimation stage of the thermal state, Kalman filtering is used, and the state estimate is updated online based on the actual junction temperature measurement residuals. The transient thermal impedance matrix is ​​extracted in real time from the posterior state estimate. As a preferred approach, when initializing the state estimate covariance matrix, the vectorized value of the steady-state thermal impedance matrix under rated operating conditions is used as the diagonal element. The diagonal elements of the process noise covariance matrix are set as the product of the variance estimate of the corresponding thermal inertia component and the process noise scaling factor. This scaling factor dynamically increases with the current operating frequency, enabling the noise model to adaptively match the increased thermal disturbances under high-frequency operating conditions. During the filtering process, adaptive noise covariance matching technology is also used to dynamically correct the process noise covariance matrix, thereby improving the robustness of the thermal impedance estimate to variable power and variable frequency operating conditions and ensuring the accuracy of transient thermal characteristic identification.

[0009] A multi-step predictive optimization problem is constructed based on the identified transient thermal impedance matrix and junction temperature safety boundary. Substituting the transient thermal impedance matrix into the thermal impedance transfer function yields the junction temperature evolution expression in the future prediction time domain. With the predicted junction temperature not exceeding the safety boundary as a hard constraint, a quadratic programming objective function is established with the bias adjustment as the decision variable. In this objective function, the adjustment step size of the gate bias voltage is given a larger penalty weight, while the adjustment step size of the drain bias voltage is given a smaller penalty weight. This makes the optimization process more inclined to adjust the drain bias voltage, thereby effectively suppressing junction temperature rise while maintaining gate bias stability and reducing disturbances to power gain and linearity. The solved multi-step predictive control sequence only uses the first step adjustment to perform rolling optimization of the bias voltage, avoiding overshoot or oscillation easily caused by traditional single-step adjustment.

[0010] During adjustment, the target changes in gate bias voltage and drain bias voltage from the first adjustment step are superimposed on the current bias voltage. Two control voltages are simultaneously output via a digital-to-analog converter, driving the gate bias adjustment circuit and the drain bias adjustment circuit respectively, ensuring that the gate and drain bias voltages change synchronously within the same adjustment cycle. This thermoelectrically coupled multi-step prediction and synchronous execution mechanism ensures rapid thermal protection response and no overshoot, keeping the junction temperature within safe limits under complex modulation signals and large dynamic power changes, and maintaining stable RF output performance of the microwave solid-state power amplifier.

[0011] This invention also provides a system for implementing thermal sensing dynamic bias adjustment of a microwave solid-state power amplifier, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the thermal sensing dynamic bias adjustment method for the microwave solid-state power amplifier. This system can be integrated into the power amplifier control unit to perform real-time online thermal imaging data analysis, thermal inertia feature extraction, transient thermal impedance estimation, and bias voltage prediction control, providing the power amplifier with full-condition thermal sensing protection and adaptive bias optimization capabilities.

[0012] The beneficial effects of this invention are: A thermal gradient field is established by acquiring time-series thermal imaging data of the active region and performing wavelet packet time-frequency decomposition. Based on the energy decay time constant, the frequency band subsequence is divided into short-memory thermal inertial components and long-memory thermal inertial components, thereby extracting thermal inertial features corresponding to different thermal path depths. These short-time and long-time thermal inertial components are concatenated into a state vector, and a thermoelectric coupled state-space model containing a state transition matrix and an input matrix is ​​constructed. The diagonal elements of the state transition matrix are the reciprocals of the thermal time constants corresponding to each thermal inertial component, thus naturally incorporating both fast and slow thermal relaxation processes into the model. Using this structure, the junction temperature prediction output can reflect the cross-scale thermal dynamics from microsecond-level hotspot growth to second-level substrate temperature rise in real time, eliminating the overestimation or underestimation of junction temperature that occurs in single-time-constant models under transient load conditions.

[0013] Using time-series electrical parameter data as the excitation input to the aforementioned model, a Kalman filter is used to recursively estimate the transient thermal impedance matrix at the current moment. The residual between the prior state estimate and the actual junction temperature measurement is used to update the posterior state value, and elements of the transient thermal impedance matrix are extracted from this residual. An adaptive noise covariance matching method is used during the filtering process to adjust the process noise covariance matrix in real time based on the residual statistical characteristics of the thermal imaging time-series data, ensuring the estimator maintains its ability to track the time-varying characteristics of thermal imaging measurement noise. After obtaining the transient thermal impedance matrix, it is substituted into the thermal impedance transfer function to generate junction temperature prediction expressions for multiple future prediction time-domain steps. A quadratic programming objective function is constructed with the junction temperature not exceeding the safety boundary as a constraint. Different penalty weights are assigned to the gate bias voltage adjustment step size and the drain bias voltage adjustment step size in the objective function, making the adjustment process more inclined to prioritize changing the drain bias. Solving this quadratic programming yields a multi-step predictive control sequence, and only the first step of the sequence is implemented to synchronously update the gate and drain bias voltages. This predictive control strategy based on real-time thermal impedance identification enables bias adjustment to act in advance before the thermal shock is fully manifested, thus avoiding the junction temperature from reaching the preset safety boundary and reducing the disturbance to gain flatness by coordinating the two-pole bias. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a flowchart of the thermal sensing dynamic bias adjustment method for microwave solid-state power amplifiers. Figure 2 This is a flowchart of the construction of the thermal gradient field and the extraction of thermal inertial components from time-frequency decomposition of thermal imaging time-series data; Figure 3 This is a flowchart of the construction process of a thermoelectric coupling state-space model based on thermal inertial components; Figure 4 This is a flowchart of junction temperature prediction and bias adjustment optimization control. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] See Figure 1 This invention provides a thermal sensing dynamic bias adjustment method for microwave solid-state power amplifiers. The overall implementation scheme of this method is as follows: real-time acquisition of thermal image time-series data and electrical parameter time-series data of the active region of the microwave solid-state power amplifier; establishment of a thermal gradient field based on the thermal image time-series data, and time-frequency decomposition of the thermal gradient field to extract thermal inertial components corresponding to multiple thermal memory depths; construction of a thermoelectric coupling state-space model with thermal inertial components as state input; using the electrical parameter time-series data as the excitation input of the thermoelectric coupling state-space model, and recursively estimating the transient thermal impedance matrix at the current moment through Kalman filtering; solving a multi-step predictive control sequence of bias adjustment amount based on the transient thermal impedance matrix and a preset junction temperature safety boundary; and synchronously adjusting the gate bias voltage and drain bias voltage of the microwave solid-state power amplifier according to the first step adjustment amount in the multi-step predictive control sequence.

[0018] Example 1: In specific implementation, refer to Figure 2 The process of establishing a thermal gradient field based on thermal imaging time-series data is as follows: For each frame of the thermal image in the time-series data, pixel-level temperature values ​​are extracted, and temperature values ​​are read pixel by pixel to form a pixel temperature matrix. The temperature difference between two horizontally adjacent pixels in the pixel temperature matrix is ​​used as the horizontal gradient magnitude to generate a horizontal gradient matrix; the temperature difference between two vertically adjacent pixels in the pixel temperature matrix is ​​used as the vertical gradient magnitude to generate a vertical gradient matrix. The horizontal and vertical gradient matrices are combined to form a thermal gradient vector field. The thermal gradient vector corresponding to each pixel position in the thermal gradient vector field is composed of the horizontal gradient value of the horizontal gradient matrix at that pixel position and the vertical gradient value of the vertical gradient matrix at that pixel position. The direction of the thermal gradient is determined by the ratio of the horizontal gradient value to the vertical gradient value.

[0019] The process of performing time-frequency decomposition on the thermal gradient field and extracting the thermal inertial components corresponding to multiple thermal memory depths is as follows: The time series of the thermal gradient field is input into a pre-constructed wavelet packet decomposition filter bank, which contains multiple sub-band filters. Multi-level wavelet packet decomposition is performed on the time series of the thermal gradient field, outputting multiple frequency band sub-sequences. For each frequency band sub-sequence, the corresponding energy decay time constant is calculated. The energy decay time constant is calculated by fitting an exponential function to the energy change of the frequency band sub-sequence over time, and extracting the decay time constant from the exponential function as the energy decay time constant for that frequency band sub-sequence.

[0020] The active region semiconductor material of a microwave solid-state power amplifier has a thermal diffusivity and a characteristic dimension, where the characteristic dimension is the minimum linewidth of the key structure in the active region. The first threshold is dynamically calculated based on the thermal diffusivity and characteristic dimension, using the following formula: in: Indicates the first threshold. This represents the preset scaling factor. The value range is from 0.1 to 10. The specific values ​​are determined based on the empirical thermal response characteristics of the semiconductor materials used. Indicates the feature size, This represents the thermal diffusivity.

[0021] The energy decay time constant of each frequency band subsequence is compared with a first threshold. If the energy decay time constant of a frequency band subsequence is less than the first threshold, the frequency band subsequence is classified as a short-memory thermal inertial component; if the energy decay time constant of a frequency band subsequence is greater than or equal to the first threshold, the frequency band subsequence is classified as a long-memory thermal inertial component. The frequency band subsequences corresponding to the classified short-memory thermal inertial components are reconstructed using inverse wavelet packets to synthesize a short-time thermal inertial component sequence; the frequency band subsequences corresponding to the classified long-memory thermal inertial components are reconstructed using inverse wavelet packets to synthesize a long-time thermal inertial component sequence. The inverse wavelet packet reconstruction uses a reconstruction filter bank matched with the wavelet packet decomposition filter bank, so that the reconstructed short-time thermal inertial component sequence and long-time thermal inertial component sequence are consistent with the time series of the original thermal gradient field in terms of time scale.

[0022] Example 2: In specific implementation, refer to Figure 3The thermoelectric coupling state-space model with thermal inertia components as state input is constructed as follows: The short-time thermal inertia component sequence and the long-time thermal inertia component sequence obtained from Example 1 are concatenated to obtain the state vector. The short-time thermal inertia component sequence contains multiple short-time thermal inertia components, and the long-time thermal inertia component sequence contains multiple long-time thermal inertia components. At each sampling moment, the component values ​​of the short-time thermal inertia component sequence and the component values ​​of the long-time thermal inertia component sequence are concatenated end-to-end according to a pre-set arrangement order to form the state vector corresponding to that sampling moment. subscript Indicates the first Each sampling time point. The arrangement order can be: the first part of the state vector corresponds to all short-time thermal inertia components, and the second part corresponds to all long-time thermal inertia components.

[0023] The drain current and drain voltage of the microwave solid-state power amplifier are sampled. The instantaneous values ​​of the drain current and drain voltage are multiplied to obtain the instantaneous value of the dissipated power. This instantaneous value of the dissipated power is then used as the input vector of the thermoelectric coupling state-space model. The input vector is a one-dimensional vector.

[0024] A state transition matrix is ​​defined to characterize the decay relationship of the thermal inertial component between adjacent time steps. For each thermal inertial component, the corresponding thermal time constant is extracted. The thermal time constant is obtained by fitting an exponential function to the time series of the thermal inertial component; specifically, the amplitude-time variation curve of the thermal inertial component is obtained using a function of the form shown in the figure. The exponential model was fitted, where For time variables, Let be the undetermined thermal time constant. The thermal time constant of each thermal inertial component is obtained after fitting. The reciprocal of the thermal time constant of each thermal inertial component is used as the corresponding decay coefficient. The thermal time constant of short-time thermal inertial components is smaller than that of long-time thermal inertial components. All decay coefficients are placed on the main diagonal of the diagonal matrix in the same order as the thermal inertial components in the state vector, with all off-diagonal elements set to zero. The resulting diagonal matrix is ​​the state transition matrix. .matrix The dimension of the matrix is ​​the same as the dimension of the state vector. The Line 1 Column elements Equal to the th in the state vector The attenuation coefficient of each thermal inertial component.

[0025] Set the input matrix Characterizes the excitation relationship of dissipated power to each thermal inertial component. Input matrix As column vectors, input matrix Each element in the matrix corresponds to a thermal inertial component, and the element value represents the excitation intensity of that thermal inertial component under unit power dissipation input. Input matrix The values ​​of each element were obtained through finite element thermal simulation combined with measured thermal imaging data calibration: First, a three-dimensional thermal model of the microwave solid-state power amplifier was established, a unit step dissipation power excitation was applied, and the modal coordinate responses corresponding to each thermal inertial component were extracted. The response amplitude was used as the input matrix. The initial values ​​of the corresponding elements are then determined; finally, the initial values ​​are corrected using least squares with the measured short-time and long-time thermal inertia component sequences to obtain the final input matrix. .

[0026] A thermoelectric coupling state-space model is constructed using state equations and output equations. The state equations are: in: Indicates the first The state vector at each sampling time. Represents the state transition matrix. Indicates the first The state vector at each sampling time. Represents the input matrix, Indicates the first The input vector at each sampling time point.

[0027] The output equation is a weighted sum of the output variables and the state vectors. The output variables are the predicted junction temperatures, and the weighting coefficients of the weighted sum form the output matrix. Output matrix Given row vectors, output matrix Each element in the matrix corresponds to a thermal inertia component, and the element value represents the contribution coefficient of that thermal inertia component to the predicted junction temperature. Output matrix The elements are obtained through steady-state thermal imaging calibration: when the microwave solid-state power amplifier reaches thermal steady state, the actual measured value of the junction temperature and the steady-state value of each thermal inertial component are recorded, and the contribution coefficient of each thermal inertial component is solved by multiple linear regression.

[0028] The thermoelectric coupling state-space model consists of state equations and output equations. The state vector is recursively updated based on the state transition matrix and the input matrix. The output variable is obtained by mapping the updated state vector through the output matrix to obtain the predicted junction temperature value.

[0029] Example 3: In specific implementation, the initialization of the state estimation covariance matrix and the process noise covariance matrix is ​​as follows: Obtain the steady-state thermal impedance matrix under the rated operating conditions of the microwave solid-state power amplifier. The rated operating conditions are determined by the typical drain voltage and drain current specified in the microwave solid-state power amplifier's technical specifications. Expand the steady-state thermal impedance matrix into a vector by rows or columns, with the expansion order strictly corresponding to the arrangement order of the thermal inertia components in the state vector. Use the values ​​of each element of the vectorized steady-state thermal impedance matrix as the initial diagonal elements of the state estimation covariance matrix. All off-diagonal elements of the state estimation covariance matrix are initialized to zero. The state estimation covariance matrix is ​​used to characterize the uncertainty of the state vector estimation.

[0030] The process noise covariance matrix is ​​set as a diagonal matrix. The time series corresponding to each thermal inertia component is obtained, and the time series variance of each thermal inertia component is calculated. This time series variance is used as the variance estimate of that thermal inertia component. Each diagonal element of the process noise covariance matrix is ​​equal to the variance estimate of the corresponding thermal inertia component multiplied by the process noise scaling factor. The process noise scaling factor is dynamically adjusted according to the current operating frequency of the microwave solid-state power amplifier. Specifically, it is adjusted in advance through offline experiments, performing Kalman filtering recursion at different operating frequency points and recording the process noise scaling factor values ​​that minimize the state estimation error variance, forming a mapping table between the operating frequency and the process noise scaling factor. During online recursion, the current operating frequency of the microwave solid-state power amplifier is acquired in real time, and the corresponding process noise scaling factor is obtained by looking up the table. The higher the operating frequency of the microwave solid-state power amplifier, the stronger the uncertainty of electromagnetic and thermal coupling, and the larger the value of the process noise scaling factor.

[0031] After initialization, the recursive loop begins. Using the posterior state estimate from the previous time step and the state transition matrix, the prior state estimate for the current time step is predicted. In this embodiment, the state transition matrix obtained in Example 2 is used, and the input matrix is ​​also obtained in Example 2. The formula for predicting the prior state estimate is: in: Indicates the first The prior state estimate at each sampling time; Indicates the first The state transition matrix at each sampling time is a diagonal matrix composed of the attenuation coefficients of each thermal inertial component. Indicates the first The posterior estimate of the state at each sampling time; Indicates the first The input matrix at each sampling time point is a column vector representing the excitation intensity of the dissipated power on each thermal inertial component. Indicates the first The input vector at each sampling moment is composed of the instantaneous value of the dissipated power obtained by multiplying the drain current and drain voltage of the microwave solid-state power amplifier.

[0032] After obtaining the state prior estimate, the output prediction value at the current time is calculated using the state prior estimate and the input vector. The output matrix is ​​the output matrix obtained in Example 2, and the output prediction value is the product of the output matrix and the state prior estimate. The output variable corresponds to the junction temperature prediction value.

[0033] The system acquires the actual junction temperature measurement at the current moment, obtained via an infrared thermal imager or an on-chip integrated temperature sensor. It then calculates the residual between the predicted and actual junction temperature measurements, which is the difference between the predicted and actual junction temperature values.

[0034] Calculate the Kalman gain. Calculating the Kalman gain requires first calculating the state prior estimation covariance matrix. The state prior estimation covariance matrix is ​​obtained by jointly calculating the state transition matrix, the state posterior estimation covariance matrix from the previous time step, and the process noise covariance matrix. The Kalman gain is obtained through matrix operations using the state prior estimation covariance matrix, the output matrix, and the measurement noise covariance matrix. The measurement noise covariance matrix is ​​preset to a fixed value based on the measurement noise characteristics of the temperature sensor.

[0035] The posterior state estimate is updated using the residuals and Kalman gain. The posterior state estimate equals the prior state estimate plus the product of the Kalman gain and the residuals. Simultaneously, the posterior state estimate covariance matrix is ​​updated for recursion in the next time step.

[0036] The elements of the transient thermal impedance matrix are extracted from the posterior state estimates. Each component of the posterior state estimate corresponds to the posterior estimate of a thermal inertia component at the current moment. These components are then recombine according to the arrangement of thermal resistance and thermal capacitance elements in the thermal impedance network topology to form the transient thermal impedance matrix. The transient thermal impedance matrix characterizes the dynamic response of the junction temperature to the dissipated power pulse at the current moment.

[0037] The Kalman filter recursion employs an adaptive noise covariance matching method to adjust the process noise covariance matrix in real time to match the residual statistical characteristics of the current thermal imaging time series data. Specifically, a sliding window is maintained, storing the most recent... The residual value at each sampling time. The window length is preset. At each sampling time, the sample variance of the residual sequence within the sliding window is calculated. Simultaneously, based on the prior estimation of the covariance matrix, output matrix, and measurement noise covariance matrix of the current state, the theoretical variance of the residuals is calculated. The sample variance is compared with the theoretical variance. If the sample variance is greater than the theoretical variance, the diagonal elements of the process noise covariance matrix are increased proportionally; if the sample variance is less than the theoretical variance, the diagonal elements of the process noise covariance matrix are decreased proportionally. The scaling factor is set to the ratio of the sample variance to the theoretical variance, enabling the process noise covariance matrix to adaptively track changes in the statistical characteristics of the residuals.

[0038] Example 4: In specific implementation, refer to Figure 4 Substituting the transient thermal impedance matrix into the thermal impedance transfer function yields the junction temperature prediction expressions for multiple future prediction time steps. The thermal impedance transfer function is expressed by the state transition matrix, input matrix, and output matrix of the thermoelectric coupled state-space model, and the elements of the transient thermal impedance matrix correspond to the unit impulse response sequence of the thermal impedance transfer function. At the sampling time... Using the current state estimate And the future bias adjustment step size sequence, through recursion, yields the predicted time domain length. The predicted junction temperature value for each prediction step. The junction temperature prediction expression is: in: Indicates at the sampling time The predicted first The junction temperature at the time of sampling; The output matrix is ​​composed of the contribution coefficients of each thermal inertia component to the predicted junction temperature. The state transition matrix is ​​represented by a diagonal matrix consisting of the attenuation coefficients of each thermal inertial component. Indicates at the sampling time The state posterior estimate formed by vectorizing the transient thermal impedance matrix; The input matrix represents the excitation intensity of the dissipated power on each thermal inertial component. Indicates the prediction step number. The range of values ​​is to ; Indicates the summation index; This represents the gain factor of the gate bias voltage on the power dissipation. The results were obtained by linearizing the transconductance characteristic curve of the drain current versus gate bias voltage of the microwave solid-state power amplifier near the operating point. The value is the measured transconductance value multiplied by the current drain bias voltage value; This represents the gain factor of the drain bias voltage on the power dissipation. The output conductance characteristic curve of the microwave solid-state power amplifier as a function of the drain current versus the drain bias voltage was obtained by linearizing the curve near the operating point. The value is the measured output conductance multiplied by the current drain bias voltage plus the static drain current at the bias point. Indicates at the sampling time The first planning The gate bias voltage adjustment step size; Indicates at the sampling time The first planning The step size for adjusting the drain bias voltage.

[0039] Preset junction temperature safety boundary The maximum allowable junction temperature is determined based on the active region semiconductor material of the microwave solid-state power amplifier. This maximum allowable junction temperature is obtained from a table in the reliability datasheet provided by the device manufacturer. At the sampling time... For each step in the prediction time domain, a constraint is applied that the predicted junction temperature does not exceed the junction temperature safety boundary; that is, for all... , constraint is .

[0040] Construct a quadratic programming objective function with respect to the bias adjustment. In the quadratic programming objective function, the time domain length is controlled. The weighted sum of the squares of all gate bias voltage adjustment steps and all drain bias voltage adjustment steps is used as a penalty term. The objective function expression is: in: This represents the value of the objective function in a quadratic programming problem. This represents the first weight value. This represents the second weight value; This indicates the control time domain length; other symbols are defined as above.

[0041] First weight value Second weight value The specific settings are as follows: Set to greater than The value of this value causes the quadratic programming objective function to impose a heavier penalty on the gate bias voltage adjustment step size than on the drain bias voltage adjustment step size. In one implementation, Values , Values . Value greater than The rationale is that the gate bias voltage of a microwave solid-state power amplifier is much more sensitive to the control of the drain current than the drain bias voltage. Even a small gate bias voltage adjustment step may cause a significant deterioration in the linearity of the microwave solid-state power amplifier. Therefore, a larger penalty weight needs to be given to the objective function to limit the adjustment range of the gate bias voltage, so that the obtained bias adjustment sequence will preferentially use the drain bias voltage adjustment to achieve the purpose of thermal management.

[0042] Find the optimal solution for the quadratic programming objective function under the junction temperature safety boundary constraints. Substitute the junction temperature prediction expression into the constraints to form a unified set of inequalities. Represent the objective function and the set of constraints as a standard quadratic programming problem and solve it online using the effective set method or the interior point method. The specific steps of the effective set method are as follows: First, assume a set of inequality constraints as effective constraints and solve the corresponding equal-constraint quadratic programming subproblems. Then, check whether the remaining constraints are violated. If they are violated, add constraints to the effective set; if they are exceeded, remove constraints from the effective set. Iterate until convergence. The optimal solution obtained by solving the quadratic programming objective function includes the values ​​from the current sampling time. Rise together A sequence of bias adjustment values ​​with a control time-domain step size, i.e., a sequence of gate bias voltage adjustment steps. and drain bias voltage adjustment step sequence .

[0043] Example 5: In specific implementation, the process of synchronously adjusting the gate bias voltage and drain bias voltage of the microwave solid-state power amplifier according to the first step adjustment amount in the multi-step predictive control sequence is as follows. The multi-step predictive control sequence is extracted from the optimal solution obtained by solving the quadratic programming objective function in Example 4. The multi-step predictive control sequence includes the gate bias voltage adjustment step size sequence and the drain bias voltage adjustment step size sequence. The gate bias voltage adjustment step size sequence is calculated from the current sampling time... The gate bias voltage adjustment step size is composed of several control time-domain steps, and the drain bias voltage adjustment step size sequence is composed of a total of steps starting from the current sampling time. It consists of a drain bias voltage adjustment step size that controls the time domain step size. Indicates the length of the control time domain. Based on the thermal response speed of the microwave solid-state power amplifier. The value of makes the control time domain cover the main time interval of the junction temperature response to bias adjustment.

[0044] The first step adjustment is extracted from the multi-step predictive control sequence. This first step adjustment includes the target change values ​​for the gate bias voltage and the drain bias voltage. The target change value for the gate bias voltage is equal to the first element in the gate bias voltage adjustment step sequence, which corresponds to the gate bias voltage adjustment step size at the current sampling time. Similarly, the target change value for the drain bias voltage is equal to the first element in the drain bias voltage adjustment step sequence, which corresponds to the drain bias voltage adjustment step size at the current sampling time.

[0045] The target gate bias voltage value is obtained by superimposing the target change in gate bias voltage onto the current gate bias voltage. The current gate bias voltage is acquired in real-time by a gate bias voltage sampling circuit, which consists of a precision resistor divider network and an analog-to-digital converter (ADC). The ADC quantizes the divided voltage and sends it to the processor. The superposition operation is performed by the processor, and the target gate bias voltage value equals the current gate bias voltage value plus the target change in gate bias voltage. Similarly, the target drain bias voltage value is obtained by superimposing the target change in drain bias voltage onto the current drain bias voltage. The current drain bias voltage is acquired in real-time by a drain bias voltage sampling circuit, which consists of a high-voltage divider network and an ADC. The superposition operation is performed by the processor, and the target drain bias voltage value equals the current drain bias voltage value plus the target change in drain bias voltage.

[0046] Two control voltages are simultaneously output via a digital-to-analog converter (DAC). The DAC has at least two independent conversion channels: the first channel corresponds to the gate bias control voltage output, and the second channel corresponds to the drain bias control voltage output. The processor writes the target gate bias voltage value to the input register of the first channel and the target drain bias voltage value to the input register of the second channel. A unified data latch signal is set within the DAC. After both channel input registers are written, the processor sends the data latch signal to the DAC. Upon receiving the data latch signal, the DAC initiates the digital-to-analog conversion of both channels simultaneously, ensuring that the analog voltage signals output by the first and second channels are updated at the same time. The analog voltage signal output by the first channel drives the gate bias adjustment circuit, and the analog voltage signal output by the second channel drives the drain bias adjustment circuit. The gate bias adjustment circuit is a controllable voltage source circuit that receives the analog voltage signal output by the first channel of the DAC as a reference voltage, amplifies it, and low-pass filters it before outputting the gate bias voltage. The drain bias adjustment circuit is a controllable high-voltage power supply circuit. It receives the analog voltage signal output from the second channel of the digital-to-analog converter as a reference voltage, and outputs the drain bias voltage after high-voltage amplification and energy storage filtering. The voltage settling time of the gate bias adjustment circuit and the drain bias adjustment circuit are both less than the preset adjustment period, so that the gate bias voltage and the drain bias voltage change simultaneously within the same adjustment period. The adjustment period is generated by the system timer and is consistent with the sampling period of the Kalman filter recursion. The value of the adjustment period is determined based on the highest sampling rate of the thermal imaging timing data of the microwave solid-state power amplifier. The adjustment period is set so that the processor can complete the Kalman filter recursion, quadratic programming solution, and bias voltage command issuance within a single cycle.

[0047] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for thermally sensing dynamic bias adjustment of a microwave solid-state power amplifier, characterized in that, The method includes: Real-time acquisition of active region thermal imaging time-series data and electrical parameter time-series data of microwave solid-state power amplifiers; A thermal gradient field is established based on the thermal imaging time series data, and the thermal gradient field is decomposed into time and frequency to extract thermal inertial components corresponding to multiple thermal memory depths. Construct a thermoelectric coupling state-space model with the aforementioned thermal inertial components as state inputs; The electrical parameter time series data is used as the excitation input of the thermoelectric coupling state space model, and the transient thermal impedance matrix at the current moment is estimated recursively through Kalman filtering. Based on the transient thermal impedance matrix and the preset junction temperature safety boundary, the multi-step predictive control sequence of the bias adjustment is solved, specifically as follows: Substituting the transient thermal impedance matrix into the thermal impedance transfer function, we obtain the junction temperature prediction expression for multiple future prediction time steps. With the constraint that the value of the junction temperature prediction expression does not exceed the preset junction temperature safety boundary, a quadratic programming objective function for the bias adjustment is constructed. In the objective function of the quadratic programming, the weighted sum of squares of the change step size of the bias adjustment amount is used as a penalty term, and the adjustment step size of the gate bias voltage and the adjustment step size of the drain bias voltage have different penalty weights. Find the optimal solution of the quadratic programming objective function under the constraints. The optimal solution contains a sequence of bias adjustment values ​​for multiple control time-domain steps starting from the current time. According to the adjustment amount in the first step of the multi-step predictive control sequence, the gate bias voltage and drain bias voltage of the microwave solid-state power amplifier are simultaneously adjusted, specifically as follows: The first step adjustment amount is extracted from the multi-step predictive control sequence, the first step adjustment amount including the target change value of the gate bias voltage and the target change value of the drain bias voltage; The target change in gate bias voltage is superimposed on the current gate bias voltage to obtain the target gate bias voltage value. The target change in drain bias voltage is added to the current drain bias voltage to obtain the target drain bias voltage value. Two control voltages are output simultaneously by a digital-to-analog converter, which drive the gate bias adjustment circuit and the drain bias adjustment circuit respectively, so that the gate bias voltage and the drain bias voltage change simultaneously within the same adjustment cycle.

2. The microwave solid-state power amplifier thermal sensing dynamic bias adjustment method according to claim 1, characterized in that, A thermal gradient field is established based on the aforementioned thermal imaging time-series data, specifically as follows: Pixel-level temperature values ​​are extracted from each frame of the thermal image in the thermal imaging time series data; The temperature difference between two adjacent pixels is used as the gradient magnitude, and horizontal and vertical gradient matrices are constructed according to the horizontal and vertical directions, respectively. The horizontal gradient matrix and the vertical gradient matrix are combined to form a thermal gradient vector field, wherein the direction of the thermal gradient at each location is determined by the ratio of the horizontal gradient value to the vertical gradient value.

3. The microwave solid-state power amplifier thermal sensing dynamic bias adjustment method according to claim 1, characterized in that, The thermal gradient field is decomposed into time-frequency components to extract multiple thermal memory depths, specifically: The time series of the thermal gradient field is input into a wavelet packet decomposition filter bank to obtain multiple frequency band sub-sequences; For each frequency band subsequence, calculate its energy decay time constant. Classify the frequency band subsequences with energy decay time constants less than the first threshold as short-memory thermal inertial components, and classify the frequency band subsequences with energy decay time constants greater than or equal to the first threshold as long-memory thermal inertial components. Inverse wavelet packet reconstruction is performed on the short-memory thermal inertial component and the long-memory thermal inertial component respectively to obtain the short-time thermal inertial component sequence and the long-time thermal inertial component sequence.

4. The microwave solid-state power amplifier thermal sensing dynamic bias adjustment method according to claim 3, characterized in that, The first threshold is dynamically calculated based on the thermal diffusivity and characteristic size of the active region semiconductor material of the microwave solid-state power amplifier.

5. The microwave solid-state power amplifier thermal sensing dynamic bias adjustment method according to claim 3, characterized in that, Constructing a thermoelectric coupling state-space model with the aforementioned thermal inertial components as state inputs, specifically as follows: The short-time thermal inertial component sequence and the long-time thermal inertial component sequence are concatenated into a state vector; The product of the drain current and drain voltage of the microwave solid-state power amplifier is used as the dissipated power, and the dissipated power is used as the input vector of the state-space model. The state transition matrix is ​​set to characterize the decay relationship of the thermal inertial component in adjacent time steps, and the input matrix is ​​set to characterize the excitation relationship of the dissipated power on the thermal inertial component. The thermoelectric coupling state-space model is constructed using state equations and output equations, where the output variable of the output equation is the predicted junction temperature.

6. The microwave solid-state power amplifier thermal sensing dynamic bias adjustment method according to claim 1, characterized in that, The method of recursively estimating the transient thermal impedance matrix at the current moment using Kalman filtering is as follows: Initialize the state estimation covariance matrix and the process noise covariance matrix; Using the state estimate and state transition matrix from the previous time step, predict the prior state estimate for the current time step; Using the prior state estimate and the input vector, calculate the predicted output value at the current time. Collect the actual junction temperature measurement value at the current moment, calculate and output the residual between the predicted value and the actual junction temperature measurement value; The residuals and Kalman gain are used to update the posterior state estimate, and the elements of the transient thermal impedance matrix are extracted from the posterior state estimate.

7. The microwave solid-state power amplifier thermal sensing dynamic bias adjustment method according to claim 6, characterized in that, The Kalman filter recursion employs an adaptive noise covariance matching method to adjust the process noise covariance matrix in real time to match the residual statistical characteristics of the current thermal imaging time series data.

8. A microwave solid-state power amplifier thermal sensing dynamic bias adjustment implementation system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the microwave solid-state power amplifier thermal sensing dynamic bias adjustment method according to any one of claims 1 to 7.

Images