A power amplifier system with a low-power dynamically optimized matching network

Abstract

The present invention relates to the field of radio frequency power amplifiers and discloses a power amplifier system with a low-power, dynamically optimized matching network. The system includes a signal sensing module, an intelligent decision-making module, a dynamic execution module, an energy recovery module, and a thermal management module. The signal sensing module extracts the time-frequency domain characteristic parameters of the input signal in real time, as well as collects thermal status data of the power amplifier chip and the heat dissipation substrate. The intelligent decision-making module generates MEMS switch state instructions and bias voltage adjustment instructions. The dynamic execution module synchronously adjusts the distributed capacitor and inductor combination of the MEMS reconfigurable matching network and the operating mode of the power amplifier chip. The energy recovery module converts the output harmonic energy into direct current to feed back the power supply system. The thermal management module dynamically adjusts the coolant flow rate of the microchannel heat dissipation substrate. The present invention achieves efficiency optimization, adaptive energy consumption control, and system-level reliability improvement for power amplifiers in multiple frequency bands, variable loads, and complex thermal environments.

Description

Technical Field

[0001] The present invention relates to the technical field of radio frequency power amplifiers, and in particular to a power amplifier system with a low-power consumption dynamically optimized matching network. Background Art

[0002] Traditional power amplifier systems face a core bottleneck in balancing efficiency and thermal stability in multi-band, highly dynamic load scenarios. Because fixed impedance matching networks are unable to adapt to the impedance variations of broadband signals in real time, the output of the power amplifier chip remains mismatched for a long time, causing significant energy reflection and heat dissipation. This efficiency degradation is particularly prominent under high peak-to-average ratio modulation signals.

[0003] In the existing technology, the power amplifier architecture based on static matching network and fixed bias voltage has difficulty in maintaining a high-efficiency operating range under variable load conditions, and lacks an effective mechanism for recovering harmonic energy, resulting in energy waste.

[0004] In addition, traditional heat dissipation solutions mostly rely on passive radiators or constant flow rate cooling systems, which cannot dynamically adjust the heat dissipation intensity according to the real-time heat load. Under high temperature and high power conditions, local heat accumulation can easily occur, leading to device performance degradation or even failure.

[0005] Although some improvement schemes attempt to introduce adjustable matching elements, they are still unable to achieve global coordinated optimization of multi-band impedance matching, thermal management and energy recovery due to the insufficient precision of discrete adjustment and the lag of control strategies. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the present invention provides a power amplifier system with a low-power dynamically optimized matching network, which solves the problems of low efficiency, high static power consumption, insufficient heat dissipation capacity and harmonic energy waste caused by impedance mismatch in traditional power amplifier systems in multi-band scenarios.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A power amplifier system with a low-power dynamic optimization matching network, comprising:

[0008] The signal sensing module includes a high-speed ADC sampling unit and a temperature sensor. The high-speed ADC sampling unit is used to extract the time-frequency domain characteristic parameters of the input signal in real time. The temperature sensor is used to collect temperature data of the power amplifier chip and the heat dissipation substrate.

[0009] an intelligent decision-making module, which is in communication with the signal sensing module, receives the time-frequency domain characteristic parameters and temperature data, and generates a control signal including a MEMS switch state instruction and a bias voltage adjustment instruction through a built-in neural network model;

[0010] a dynamic execution module, electrically connected to the intelligent decision-making module, comprising a MEMS reconfigurable matching network and a power amplifier chip, wherein the MEMS reconfigurable matching network adjusts its distributed capacitance and inductance combination according to the switch state instruction, and the power amplifier chip switches its operating mode according to the bias voltage adjustment instruction;

[0011] An energy recovery module, connected to the output end of the dynamic execution module through a coupler, is used to capture the harmonic energy in the output signal and convert it into direct current through a rectifier circuit to feed back to the power management module;

[0012] The thermal management module is embedded in the microchannel heat dissipation substrate and communicates bidirectionally with the signal sensing module and the intelligent decision-making module. It adjusts the coolant flow rate according to the temperature data and feeds back the heat dissipation status to the intelligent decision-making module to optimize the control signal.

[0013] Preferably, the MEMS reconfigurable matching network changes the combination of distributed capacitance and inductance by switching the switch state, wherein:

[0014] When the switch is in the on state, its equivalent impedance is dominated by the on-resistance;

[0015] When the switch is in the off state, its equivalent impedance is dominated by the capacitive reactance of the off capacitor;

[0016] The combination of the capacitor and the inductor is dynamically adjusted according to the target frequency band to achieve conjugate matching of the output impedance of the power amplifier chip.

[0017] Preferably, the working modes of the power amplifier chip include:

[0018] Peak efficiency mode: the bias voltage is adjusted to be no less than a first threshold, and the MEMS reconfigurable matching network is switched to a high Q value topology;

[0019] Low-power sleep mode: the bias voltage is adjusted to be no higher than a second threshold, and the MEMS reconfigurable matching network is switched to a low-loss topology.

[0020] Preferably, the rectifier circuit includes:

[0021] Multi-stage charge pump topology, which multiplies the harmonic energy extracted by the coupler step by step through cascaded diode-capacitor units;

[0022] The DC bus path feeds the doubled DC power back to the power management module, where the voltage boost of each charge pump unit is determined by the forward voltage drop of the diode and the capacitance value of the capacitor.

[0023] Preferably, the adjustment of the coolant flow rate is achieved by PID closed-loop control, specifically satisfying:

[0024] ;

[0025] in, is the real-time temperature deviation, Indicates the preset temperature threshold, The current temperature of the heat dissipation substrate collected by the temperature sensor; 、 、 are the proportional, integral, and differential gain coefficients respectively.

[0026] Preferably, the sampling rate of the high-speed ADC sampling unit is not less than twice the instantaneous bandwidth of the input signal.

[0027] Preferably, the temperature sensor comprises a distributed thermocouple array, which is arranged on the surface of the power amplifier chip and at the inlet and outlet positions of the microchannel heat dissipation substrate.

[0028] Preferably, the generation of the control signal of the neural network model comprises the following steps:

[0029] Encoding the time-frequency domain characteristic parameters and temperature data into a multidimensional feature vector;

[0030] Extracting implicit features related to impedance matching through multi-layer nonlinear transformation;

[0031] The output includes discrete control instructions for MEMS switch state combinations and bias voltage levels.

[0032] Preferably, the neural network model is optimized by transfer learning, and its training process includes the following steps:

[0033] Utilize the historical frequency band parameter database to build a cross-band impedance matching knowledge base;

[0034] generating a multi-band training sample set based on the knowledge base;

[0035] The reinforcement learning algorithm is used to update the network weight parameters to minimize the impedance mismatch loss function.

[0036] The present invention also provides a power amplifier control method based on a dynamic optimization matching network, comprising the following steps:

[0037] The time-frequency domain characteristic parameters of the input signal are extracted in real time through the high-speed ADC sampling unit, and the temperature data of the power amplifier chip and the heat dissipation substrate are collected through the temperature sensor;

[0038] Inputting the time-frequency domain characteristic parameters and temperature data into a neural network model to generate a control signal including a MEMS switch state instruction and a bias voltage adjustment instruction;

[0039] Adjusting the distributed capacitance and inductance combination of the reconfigurable matching network according to the MEMS switch state instruction, and switching the operating mode of the power amplifier chip according to the bias voltage adjustment instruction;

[0040] The harmonic energy of the power amplifier output signal is captured by the coupler, converted into DC power by the multi-stage charge pump rectifier circuit and fed back to the power supply system;

[0041] The coolant flow rate of the microchannel heat dissipation substrate is adjusted based on the temperature data, and the heat dissipation status is fed back to the neural network model to optimize subsequent control signals.

[0042] The present invention provides a power amplifier system with a low-power dynamic optimization matching network. It has the following beneficial effects:

[0043] 1. The present invention uses a high-speed ADC to capture the time-frequency domain characteristics of the input signal in real time, combines it with the intelligent decision-making of the neural network model, and dynamically adjusts the distributed capacitor and inductor combination of the MEMS reconfigurable matching network and the power amplifier bias voltage to achieve precise impedance matching in different frequency bands, effectively solving the mismatch problem of traditional fixed matching networks in multi-band scenarios, and significantly improving the efficiency and linearity of the power amplifier.

[0044] 2. The present invention intelligently switches the amplifier operating mode based on the load status: it adopts the high peak efficiency mode (high bias voltage + high Q value matching network) during the signal peak period, and switches to the low power sleep mode (low bias voltage + low loss topology) during the low load period. By dynamically adjusting the hardware configuration and power supply strategy, it greatly reduces static power consumption and heat dissipation, thereby extending the device's battery life.

[0045] 3. The present invention recovers the output harmonic energy through a coupler and a multi-stage charge pump rectifier circuit, converts it into direct current and feeds it back to the power management module, thereby realizing closed-loop utilization of the system's internal energy, reducing dependence on external power supplies, and reducing overall system power consumption.

[0046] 4. The present invention adjusts the coolant flow rate of the microchannel heat dissipation substrate in real time based on temperature sensor data, and dynamically optimizes the heat dissipation strategy through PID algorithm and neural network feedback mechanism to ensure the stable operation of the power amplifier chip under high temperature and high load conditions, and avoid performance degradation or device damage due to local overheating.

[0047] 5. The neural network model of the present invention coordinates signal characteristics, thermal status and energy recovery data to achieve global linkage optimization of matching network, bias voltage and heat dissipation control, enhance the adaptability of the system in complex electromagnetic environments and variable temperature conditions, and at the same time improve the reliability of long-term operation through hardware status self-check and protection mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1Schematic diagram of the system structure of the present invention;

[0049] Figure 2 Schematic diagram of the circuit topology of the charge pump in the harmonic rectification circuit of the present invention;

[0050] Figure 3 Schematic diagram of the MEMS reconfigurable matching network of the present invention;

[0051] Figure 4 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments 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 are within the scope of protection of the present invention.

[0053] Please see the attached Figure 1 -Attached Figure 3 The present invention provides a power amplifier system with a low-power dynamic optimization matching network. By real-time sensing of signal characteristics and environmental parameters, combined with intelligent decision-making and multi-module collaborative control, it realizes multi-band impedance matching, energy efficiency optimization and integrated thermal management.

[0054] The system includes a signal perception module, an intelligent decision-making module, a dynamic execution module, an energy recovery module and a thermal management module. The following is a detailed description of each module.

[0055] The detailed description of the signal perception module is as follows:

[0056] In this embodiment, the signal perception module is composed of a high-speed ADC sampling unit and a distributed temperature sensing network, which is used to capture the time-frequency domain characteristic parameters of the input RF signal and the temperature data of key parts of the system in real time, providing dynamic perception input for the subsequent intelligent decision-making module.

[0057] The high-speed ADC sampling unit uses a broadband analog-to-digital conversion architecture, and its sampling rate is dynamically configured based on the instantaneous bandwidth of the input signal. Specifically, the operating clock frequency of the ADC sampling unit is synchronized to the center frequency of the input signal through a phase-locked loop, ensuring that the sampling rate satisfies the following relationship:

[0058] ;

[0059] in, is the actual sampling rate, Represents the instantaneous bandwidth of the input signal. Preferably, the ADC sampling unit incorporates a built-in digital down-conversion (DDC) processing core, which extracts the time- and frequency-domain parameters of the baseband signal through orthogonal mixing and decimation filtering. In the signal processing flow, the input signal is pre-processed by an anti-aliasing filter before being sampled and quantized by the ADC unit, generating a digitized data stream containing the amplitude and phase of the time-domain waveform and the frequency-domain spectral characteristics.

[0060] The temperature sensing network consists of an array of high-precision micro-thermocouples, whose placement and density are dynamically adjusted based on the thermally sensitive areas. Preferably, the thermocouple array is distributed in a matrix across the surface of the power amplifier chip, with the spacing between adjacent nodes no greater than 1 / 5 of the chip's thermal diffusion length. This array is used to capture the temperature gradient distribution on the chip surface. Furthermore, redundant temperature measurement nodes are placed at the coolant inlet and outlet of the microfluidic heat sink substrate, allowing differential calculations to determine the substrate's real-time thermal conductivity efficiency. The thermocouple output signal is conditioned by a low-noise amplifier and then switched by a multiplexer to an analog-to-digital conversion channel, generating a digital value that is linearly related to the temperature value.

[0061] In terms of data fusion, the signal perception module integrates a feature parameter packaging unit to timestamp-align and spatially correlate the time-frequency domain data with the temperature data. Specifically, the time-frequency domain parameters output by the ADC sampling unit, including the signal's instantaneous bandwidth, modulation type, and peak-to-average ratio, are converted into frequency-domain spectral feature vectors via a sliding-window Fourier transform (FFT). The temperature data is then encoded into a two-dimensional thermal distribution matrix based on spatial coordinates. Preferably, the data fusion process employs the following normalization:

[0062] ;

[0063] in, represents raw data (such as temperature value or signal amplitude), and The normalized data is transmitted to the intelligent decision module via a high-speed serial interface (such as JESD204B).

[0064] In the dynamic calibration mechanism, the working mode of the ADC sampling unit can be switched adaptively according to the characteristics of the input signal. For example, when the input signal is detected to be constant envelope modulation (such as FSK), the oversampling mode is enabled ( ) to improve time-domain resolution. When the input signal is high-peak-to-average-ratio modulation (such as OFDM), it switches to differential sampling mode, extending the dynamic range through alternating sampling of the two ADC channels. Preferably, the calibration mechanism is implemented through a table lookup method, pre-stored tables of optimal sampling parameter configurations for different signal types, which are dynamically called by the intelligent decision module based on real-time signal analysis results.

[0065] The detailed description of the intelligent decision-making module is as follows:

[0066] In this embodiment, the intelligent decision-making module builds a neural network model based on a deep learning architecture. By integrating time-frequency signal characteristics with thermal state data, it generates dynamic control instructions to achieve coordinated optimization of the MEMS matching network and power amplifier bias. The core of the module is to map real-time sensor data into discrete control strategies and continuously improve decision-making accuracy through transfer learning and reinforcement learning mechanisms.

[0067] The neural network model architecture adopts a hierarchical encoding-decoding structure, consisting of a feature encoding layer, a spatiotemporal attention layer, and a control output layer. The feature encoding layer receives multi-source input data from the signal perception module, including time-frequency domain feature parameters (instantaneous bandwidth, modulation type, peak-to-average ratio) and temperature data (chip surface temperature distribution, heat sink inlet and outlet temperature difference). Preferably, the feature encoding layer normalizes the multidimensional input into a 128-dimensional feature vector through a fully connected network, which is mathematically expressed as:

[0068] ;

[0069] in, represents the input feature matrix, and are the coding layer weights and bias terms, is the activation function (such as ReLU).

[0070] The spatiotemporal attention layer is used to dynamically assign feature weights to address the issue of differences in the correlation between signals in different frequency bands and temperature parameters. This layer calculates the attention score between the input feature vector and the hidden state:

[0071] ;

[0072] in, and are the query matrix and the key matrix respectively, which are generated by linear transformation of the eigenvectors; is the vector dimension scaling factor. The implicit features that are strongly related to impedance matching are obtained through weighted summation. , whose dimension is consistent with the encoding layer output.

[0073] The control output layer maps implicit features into discrete control instructions. For each MEMS switch state combination, a binary classifier is used to output the on / off state of each branch switch. For the bias voltage level, a piecewise linear function is used to generate discrete voltage range instructions. Preferably, the output layer introduces a temperature compensation factor β(T) to adjust the decision threshold of the control instruction based on real-time temperature data:

[0074] ;

[0075] in, and They are respectively the minimum and maximum operating temperatures allowed by the power amplifier chip.

[0076] The training and optimization method adopts a hybrid strategy of transfer learning and reinforcement learning. In the pre-training stage, a cross-band knowledge base is constructed using the historical multi-band impedance matching database, and enhanced training samples are generated through the following steps:

[0077] 1. Extract frequency bands from the knowledge base of Parameter Matrix ;

[0078] 2. Generate cross-band virtual samples based on interpolation ,in is the random interpolation coefficient;

[0079] 3. Mix virtual samples with real-time environmental data to form a multi-band training set.

[0080] In the online learning phase, a policy gradient-based reinforcement learning algorithm is used to update the network weights. The impedance mismatch loss function is defined as:

[0081] ;

[0082] in, Indicates the mismatch between the measured output impedance and the target conjugate impedance, is the policy entropy regularization term, is the entropy weight coefficient. Preferably, the proximal policy optimization (PPO) algorithm is used to perform gradient updates on the policy network to ensure training stability.

[0083] In terms of real-time performance, the embedded inference chip optimizes inference latency through a dynamic allocation of computing resources. Under low-load conditions, weight pruning and low-bit quantization (e.g., 8-bit integer inference) are enabled to reduce power consumption. In scenarios with high real-time requirements (e.g., transient frequency band switching), multi-core parallel computing is activated, assigning each neural network layer to independent computing units. The model preferably employs an inter-layer pipeline scheduling strategy, enabling overlapping execution of feature encoding, attention calculation, and control instruction generation.

[0084] During model deployment and inference, the neural network model is quantized, compressed, and loaded into the embedded AI accelerator. Preferably, dynamic computational graph optimization technology is used to adaptively select the neural network subgraph execution path based on the input signal bandwidth: for narrowband signals, a lightweight subnetwork is enabled to reduce latency; for broadband signals, the full network is activated to ensure decision accuracy.

[0085] The detailed description of the dynamic execution module is as follows:

[0086] In this embodiment, the dynamic execution module realizes dynamic adaptation of the power amplifier output impedance and intelligent switching of the working mode through the linkage control of the MEMS reconfigurable matching network and the power amplifier chip. Its core lies in converting the control instructions generated by the intelligent decision-making module into physical state adjustments of the reconfigurable hardware to ensure efficient energy transmission under multi-band signals.

[0087] The MEMS reconfigurable matching network is composed of a distributed capacitor-inductor branch array. Each branch contains a series inductor, a parallel capacitor, and a RF MEMS switch. The on and off states of the MEMS switch directly determine the equivalent impedance characteristics of the branch: when the switch is in the on state, the branch equivalent impedance is composed of the on resistance Inductive reactance Series dominance, its mathematical expression is:

[0088] ;

[0089] When the switch is in the off state, the branch equivalent impedance is composed of the capacitive reactance of the off capacitor. With parasitic resistance Parallel dominance, its mathematical expression is:

[0090] ;

[0091] By combining the switching states of different branches, the matching network can be reconfigured into a π-type, T-type or L-type topology, and its overall impedance Satisfy the conjugate matching conditions between the amplifier output impedance and the load impedance in the target frequency band:

[0092] ;

[0093] in, is the load impedance, The MEMS switch preferably uses an aluminum nitride packaging structure with an on-resistance of ≤0.8Ω, an off-capacitance of ≤50fF, and a switching time of ≤2μs to reduce insertion loss and improve response speed.

[0094] The power amplifier chip supports multi-mode bias voltage regulation, and its working state is dynamically linked with the matching network topology. In the peak efficiency mode, the bias voltage is increased to the first threshold At the same time, the MEMS matching network switches to a high Q value topology (Q ≥ 40), which improves output efficiency by reducing resonant circuit losses. In low-power sleep mode, the bias voltage is reduced to the second threshold. Next, the matching network switches to a low-loss topology (insertion loss ≤ 0.6dB) to reduce static power consumption by optimizing the current path. Preferably, the ratio of the first threshold to the second threshold ranges from 2.5:1 to 3:1 to ensure a balance between efficiency and stability during mode switching.

[0095] Regarding the control logic and drive circuits, the dynamic execution module receives binary switching commands and bias voltage level codes from the intelligent decision-making module. The switching commands drive the MEMS switch array via a level conversion circuit, with the on / off state of each branch independently configured by the corresponding control bit. The bias voltage adjustment commands generate a reference voltage via a digital-to-analog converter (DAC), which drives the gate bias circuit of the power amplifier chip via a high-voltage operational amplifier. The drive circuit preferably integrates an overvoltage protection module. When a bias voltage exceeds the limit or a switch state conflict is detected, it automatically triggers a hardware reset and feeds a fault code back to the intelligent decision-making module.

[0096] The dynamic impedance matching process specifically includes the following steps: According to the center frequency of the target frequency band Calculate the equivalent reactance value of the required matching network, traverse the pre-stored topology library to select the optimal branch combination, and fine-tune the capacitor and inductor parameters through iterative algorithms to ensure that the matching network is consistent across the frequency band. The impedance matching error is ≤5%. Preferably, the iterative algorithm is based on the Smith chart admittance transformation principle and optimizes the reactance parameters by the least squares method:

[0097] ;

[0098] in, are discrete sampling points within the frequency band, is the target impedance value.

[0099] The detailed description of the energy recovery module is as follows:

[0100] In this embodiment, the energy recovery module realizes efficient capture and DC feedback of harmonic energy based on a multi-stage charge pump rectifier circuit. Its core lies in converting the high-frequency harmonic components at the output end of the power amplifier into stable DC power, and realizing the recycling of internal energy of the system through the power management module.

[0101] The multi-stage charge pump topology consists of cascaded diode-capacitor units, with each unit containing a rectifier diode and a storage capacitor. The charge pump input is connected to the power amplifier output via a directional coupler to extract harmonic energy in a specific frequency band (such as the second and third harmonics). When an RF signal is input, each unit performs half-wave rectification and charge accumulation operations in sequence, achieving a step-by-step voltage multiplication. The voltage boost of a single-stage charge pump is determined by the following formula:

[0102] ;

[0103] in, is the peak voltage of the input RF signal, represents the forward voltage drop of the diode, is the signal frequency, is the capacitance of the energy storage capacitor, is the load current. Preferably, the diode should be a Schottky type with a low on-state voltage drop (such as the BAT54 series) to reduce rectification losses; the energy storage capacitor should be a ceramic capacitor with a low equivalent series resistance (ESR) to ensure rapid charge and discharge at high frequencies.

[0104] The cascade voltage doubling mechanism achieves an approximate N-fold increase in output voltage by connecting N charge pump units in series. For example, when the input signal is a peak voltage When the sine wave is high, the theoretical DC output voltage after N-stage voltage multiplication is:

[0105] ;

[0106] Preferably, the number of charge pump stages is dynamically configured according to the target feedback voltage, and the inter-stage connection mode is switched by relays or analog switches to adapt to the input intensity of different harmonic energies.

[0107] The DC bus path consists of a low-dropout linear regulator (LDO) and a storage capacitor array. The pulsating DC output of the charge pump is filtered by the capacitor array and then stabilized by the LDO to the system supply voltage (such as 5V or 12V). An isolation diode is used to prevent power backflow. Preferably, the bus path integrates a maximum power point tracking (MPPT) algorithm to dynamically adjust the load impedance of the charge pump. , so that the following formula holds true to maximize the energy conversion efficiency:

[0108] ;

[0109] In terms of harmonic selection and coupling design, the directional coupler utilizes a stripline structure, with its coupling and directivity optimized based on the target harmonic frequency band. Preferably, the coupler's output is connected to a bandpass filter with its center frequency set to the target harmonic component (such as the second harmonic) to suppress the fundamental wave and other spurious signals from entering the rectifier circuit. The filter's insertion loss is ≤0.5dB, and its bandwidth is ≥20% of the center frequency, ensuring broadband energy capture.

[0110] The detailed description of the thermal management module is as follows:

[0111] In this embodiment, the thermal management module realizes dynamic adjustment and optimization of the thermal state of the power amplifier system through the synergy of the closed-loop control algorithm and the microchannel heat dissipation substrate. The core of the thermal management module is to accurately control the coolant flow rate according to the real-time temperature data and feed back the heat dissipation status to the intelligent decision-making module to form a closed-loop control link.

[0112] The coolant flow rate is regulated using a proportional-integral-differential (PID) closed-loop control algorithm, which is mathematically expressed as:

[0113] ;

[0114] in, Indicates the real-time temperature deviation, is the preset temperature threshold, The current temperature of the heat dissipation substrate collected by the temperature sensor; 、 、 They are proportional, integral, and differential gain coefficients, respectively. Preferably, the proportional coefficient is greater than the integral coefficient, and the integral coefficient is greater than the differential coefficient, to ensure a balance between rapid response and steady-state accuracy of the control system.

[0115] The microfluidic heat dissipation substrate is made of high thermal conductivity ceramic materials (such as aluminum nitride) and is embedded with a serpentine microchannel network. The cross-sectional area of ​​the channel is inversely proportional to the coolant flow rate, and its design must satisfy the following heat conduction equation:

[0116] ;

[0117] in, is the heat dissipation power, is the convective heat transfer coefficient, is the flow channel surface area, is the coolant density, is the specific heat capacity, is the flow rate, Preferably, the cross section of the flow channel is rectangular, and the width-to-depth ratio is optimized to 1:2 to 1:3 to enhance the turbulence effect and reduce the flow resistance.

[0118] The coolant selection and circulation system uses low-viscosity ethylene glycol aqueous solution as the working fluid, and its ratio is dynamically adjusted according to the operating temperature range. Preferably, the coolant circulation pump uses a brushless DC micro-turbine pump, and the speed is adjusted by PWM signal to achieve linear control of the flow rate. The driving voltage of the pump is related to the output signal of the PID controller. Through the connection of digital-to-analog converters (DACs), a closed loop of action is formed from algorithm instructions to physical execution.

[0119] The temperature data feedback mechanism will heat the substrate inlet and outlet temperature difference The maximum temperature of the power amplifier chip surface is encoded as a thermal state vector and transmitted to the intelligent decision module through the SPI interface. ,in is the hysteresis threshold), the weights of the neural network model are adaptively adjusted, and the power amplifier bias voltage and matching network topology are optimized in a linked manner.

[0120] In general, the present invention achieves efficient adaptation of multi-band signals and coordinated optimization of system resources by integrating intelligent reasoning chips, high-speed ADC sampling units, neural network models, MEMS reconfigurable matching networks and microfluidic heat dissipation substrates. The system uses high-speed ADCs to extract the time-frequency domain characteristic parameters of the input signal in real time, combines the thermal state data collected by the temperature sensor network, generates dynamic control instructions through a pre-trained neural network model, and synchronously adjusts the distributed capacitor and inductor combination of the MEMS switch array and the bias voltage of the power amplifier chip, switching to a high peak efficiency mode during signal peaks and entering a low-power sleep mode during low-load periods; at the same time, the output harmonic energy is recovered through a coupler and a multi-stage charge pump rectifier circuit and converted into a DC feedback power supply system, and the coolant flow rate of the microfluidic heat dissipation substrate is dynamically controlled in combination with a PID closed-loop algorithm, forming a full-link closed-loop control of "signal perception-intelligent decision-making-dynamic execution-energy circulation-thermal management feedback", solving the problem of coordinated optimization of multi-band impedance mismatch and power consumption and heat consumption.

[0121] Please see the attached Figure 4 The present invention also provides a power amplifier control method based on a dynamic optimization matching network, comprising the following steps:

[0122] S1. Signal sensing and temperature acquisition: The high-speed ADC unit samples the input signal in real time, extracting time-domain waveform characteristics (such as amplitude and phase) and frequency-domain parameters (such as center frequency and bandwidth). Distributed temperature sensors are used to simultaneously obtain temperature data on the surface of the power amplifier chip and key nodes of the heat dissipation substrate.

[0123] S2. Intelligent decision generation: The above data is input into a pre-trained neural network model. Through feature fusion and policy reasoning, the model outputs a discrete control instruction containing a combination code of the MEMS switch on / off state and the power amplifier bias voltage level.

[0124] S3, dynamic execution and mode switching: According to the control instructions, the MEMS reconfigurable matching network switches the distributed capacitor and inductor combination, and at the same time adjusts the bias voltage of the power amplifier chip to achieve dynamic switching between high-efficiency mode during peak hours and low-power sleep mode;

[0125] S4, Harmonic Energy Recovery: Use a directional coupler to capture the residual harmonic energy at the power amplifier output, convert it into stable DC power through a multi-stage charge pump rectifier circuit, and feed it back to the system power management module for energy reuse;

[0126] S5. Thermal closed-loop feedback optimization: The coolant flow rate of the microchannel heat dissipation substrate is adjusted based on real-time temperature data, and the heat dissipation state parameters are fed back to the neural network model for iterative optimization of subsequent control strategies.

[0127] This method achieves adaptive and stable operation of the power amplifier in multiple frequency bands, variable loads and complex thermal environments through a closed-loop control chain of "perception-decision-execution-recovery-feedback".

[0128] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A power amplifier system with a low-power dynamically optimized matching network, characterized in that: include: The signal sensing module includes a high-speed ADC sampling unit and a temperature sensor. The high-speed ADC sampling unit is used to extract the time-frequency domain characteristic parameters of the input signal in real time. The temperature sensor is used to collect temperature data of the power amplifier chip and the heat dissipation substrate. an intelligent decision-making module, which is in communication with the signal sensing module, receives the time-frequency domain characteristic parameters and temperature data, and generates a control signal including a MEMS switch state instruction and a bias voltage adjustment instruction through a built-in neural network model; a dynamic execution module, electrically connected to the intelligent decision-making module, comprising a MEMS reconfigurable matching network and a power amplifier chip, wherein the MEMS reconfigurable matching network adjusts its distributed capacitance and inductance combination according to the switch state instruction, and the power amplifier chip switches its operating mode according to the bias voltage adjustment instruction; An energy recovery module, connected to the output end of the dynamic execution module through a coupler, is used to capture the harmonic energy in the output signal and convert it into direct current through a rectifier circuit to feed back to the power management module; a thermal management module embedded in the microchannel heat dissipation substrate and communicating bidirectionally with the signal sensing module and the intelligent decision-making module, adjusting the coolant flow rate according to the temperature data, and simultaneously feeding back the heat dissipation status to the intelligent decision-making module to optimize the control signal; The generation of control signals of the neural network model comprises the following steps: Encoding the time-frequency domain characteristic parameters and temperature data into a multidimensional feature vector; Extracting implicit features related to impedance matching through multi-layer nonlinear transformation; The output includes discrete control instructions for MEMS switch state combinations and bias voltage levels.

2. The power amplifier system with a low-power dynamic optimization matching network according to claim 1, characterized in that: The MEMS reconfigurable matching network changes the combination of distributed capacitance and inductance by switching the switch state, wherein: When the switch is in the on state, its equivalent impedance is dominated by the on-resistance; When the switch is in the off state, its equivalent impedance is dominated by the capacitive reactance of the off capacitor; The combination of the capacitor and the inductor is dynamically adjusted according to the target frequency band to achieve conjugate matching of the output impedance of the power amplifier chip.

3. The power amplifier system with a low-power dynamic optimization matching network according to claim 1, characterized in that: The working modes of the power amplifier chip include: Peak efficiency mode: the bias voltage is adjusted to be no less than a first threshold, and the MEMS reconfigurable matching network is switched to a high Q value topology; Low-power sleep mode: the bias voltage is adjusted to be no higher than a second threshold, and the MEMS reconfigurable matching network is switched to a low-loss topology.

4. The power amplifier system with a low-power dynamic optimization matching network according to claim 1, characterized in that: The rectifier circuit comprises: Multi-stage charge pump topology, which multiplies the harmonic energy extracted by the coupler step by step through cascaded diode-capacitor units; The DC bus path feeds the doubled DC power back to the power management module, where the voltage boost of each charge pump unit is determined by the forward voltage drop of the diode and the capacitance value of the capacitor.

5. The power amplifier system with a low-power dynamic optimization matching network according to claim 1, characterized in that: The coolant flow rate is regulated by PID closed-loop control, specifically satisfying: Where, e(t) = T set -T(t) is the real-time temperature deviation, T set represents the preset temperature threshold, T(t) is the current temperature of the heat dissipation substrate collected by the temperature sensor; K p , K i , K d are the proportional, integral, and differential gain coefficients respectively.

6. The power amplifier system with a low-power dynamic optimization matching network according to claim 1, characterized in that: The sampling rate of the high-speed ADC sampling unit is not less than twice the instantaneous bandwidth of the input signal.

7. The power amplifier system with a low-power dynamic optimization matching network according to claim 1, characterized in that: The temperature sensor comprises a distributed thermocouple array, which is arranged on the surface of the power amplifier chip and at the inlet and outlet positions of the microchannel heat dissipation substrate.

8. The power amplifier system with a low-power dynamic optimization matching network according to claim 1, characterized in that: The neural network model is optimized through transfer learning, and its training process includes the following steps: Utilize the historical frequency band parameter database to build a cross-band impedance matching knowledge base; generating a multi-band training sample set based on the knowledge base; The reinforcement learning algorithm is used to update the network weight parameters to minimize the impedance mismatch loss function.

9. A power amplifier control method based on a dynamic optimization matching network, according to the system according to any one of claims 1 to 8, characterized in that: The following steps are involved: The time-frequency domain characteristic parameters of the input signal are extracted in real time through the high-speed ADC sampling unit, and the temperature data of the power amplifier chip and the heat dissipation substrate are collected through the temperature sensor; Inputting the time-frequency domain characteristic parameters and temperature data into a neural network model to generate a control signal including a MEMS switch state instruction and a bias voltage adjustment instruction; Adjusting the distributed capacitance and inductance combination of the reconfigurable matching network according to the MEMS switch state instruction, and switching the operating mode of the power amplifier chip according to the bias voltage adjustment instruction; The harmonic energy of the power amplifier output signal is captured by the coupler, converted into DC power by the multi-stage charge pump rectifier circuit and fed back to the power supply system; The coolant flow rate of the microchannel heat dissipation substrate is adjusted based on the temperature data, and the heat dissipation status is fed back to the neural network model to optimize subsequent control signals.

Images