A signal and power amplifier based system failure recovery and data backup method

Abstract

This invention discloses a system fault recovery and data backup method based on signal and power amplifier, specifically including: establishing a causal digital twin model containing a structural causal model; defining a dynamic spatiotemporal graph and generating its temporal state; extracting features using a lightweight spatiotemporal graph neural network and outputting a health score and anomaly code; injecting the anomaly code into the digital twin for causal inference to predict the timing of faults and data risks; generating a collaborative disaster recovery instruction sequence based on the prediction results; and executing the instructions to complete link switching and state recovery. This invention effectively solves the problem of fragmented hardware fault recovery and data backup processes through intelligent perception and collaborative decision-making, significantly improving system recovery speed and data reliability.

Description

Technical Field

[0001] This invention relates to the field of intelligent fault tolerance and data security technology for high-reliability electronic systems, and in particular to a system fault recovery and data backup method based on signals and power amplifiers. Background Technology

[0002] In high-reliability applications such as communication, broadcasting, and radar, the radio frequency power amplifier (RF power amplifier) ​​is a crucial final-stage component in the signal transmission link, and its stable operation is essential for ensuring continuous system service. These systems have extremely high requirements for fault recovery and data integrity; any unplanned interruption or data loss can have serious consequences. Existing technologies primarily employ two relatively independent mechanisms to address these challenges.

[0003] In terms of system fault recovery, the primary reliance is on hardware-level redundancy design and status monitoring. Common solutions include parallel hot backup technology for primary and backup units, which uses an RF switching network to switch to a backup unit when the primary unit fails. Fault determination is mostly based on threshold monitoring of independent physical quantities such as output voltage, current, and temperature. More advanced solutions introduce health management systems to predict faults through the analysis of historical data. However, these methods focus on the rapid reconstruction of hardware paths, and their decision-making is usually limited to single, lagging physical parameter thresholds, making it difficult to capture early, latent faults predicted by the deterioration of multi-parameter spatiotemporal correlations.

[0004] The fundamental shortcoming of existing technologies lies in the temporal and logical disconnect between hardware failure recovery processes and software data backup processes. When a failure occurs, the system prioritizes hardware switching to restore signal transmission, while data backup often fails to capture the precise system state at the moment of failure. This results in a data loss window for business continuity, or the restored system state not being entirely consistent with the state before the failure. This disconnect makes it difficult to simultaneously optimize the overall system recovery time and recovery point objectives, limiting further improvements in system availability and data reliability.

[0005] Therefore, how to provide a system fault recovery and data backup method based on signals and power amplifiers to achieve intelligent coordination of fault recovery and data backup is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a system fault recovery and data backup method based on signal and power amplifiers. Addressing the technical problem in existing high-reliability RF power amplifier systems where fault emergency switching and critical operational data backup are independent and lack coordination, leading to long system recovery times and the risk of data loss, this invention proposes a complete technical solution. It abstracts the system as a dynamic spatiotemporal graph and utilizes a lightweight spatiotemporal graph neural network to extract its spatiotemporal correlation features in real time to assess health and identify abnormal patterns. Simultaneously, it constructs a digital twin containing a structural causal model to perform causal inference to accurately predict fault and data risk times, and generates a coordinated instruction sequence accordingly. This invention possesses the ability to intelligently trigger and precisely sequence data saving and hardware switching actions before hardware failure occurs through deep fusion of perception and reasoning, thereby significantly shortening the overall system recovery time, effectively avoiding critical data loss, and ultimately achieving the beneficial effect of inherent coordination and integrated execution of physical layer disaster recovery and data layer security.

[0007] A system fault recovery and data backup method based on signal and power amplifier according to an embodiment of the present invention includes the following steps:

[0008] S1: Establish a causal digital twin model of the RF power amplifier system. The causal digital twin model includes a structural causal model that describes the causal relationships between key parameters of the RF power amplifier system.

[0009] S2: Based on the physical topology definition of the RF power amplifier system, the dynamic spatiotemporal graph includes nodes representing each functional module of the RF power amplifier system, edges representing the physical connections between functional modules, and a node attribute matrix composed of real-time operating data of the functional modules.

[0010] S3: Periodically collect real-time operating data of each functional module of the RF power amplifier system, update the node attribute matrix, and generate a spatiotemporal diagram state sequence;

[0011] S4: Input the spatiotemporal graph state sequence into the lightweight spatiotemporal graph neural network, extract spatiotemporal correlation features and output the RF power amplifier system health score and abnormal pattern code;

[0012] S5: Input the abnormal mode code and the current operating status of the RF power amplifier system into the causal digital twin model, and predict the time of physical failure and the time of data processing risk through causal inference;

[0013] S6: Based on the RF power amplifier system health score, physical fault occurrence time, data processing risk time and preset recovery strategy, generate and release a collaborative disaster recovery instruction sequence that includes system operation context saving instructions, backup power amplifier module preset instructions and main / backup link switching instructions.

[0014] S7: Executes a sequence of collaborative disaster recovery instructions, completes the primary / backup link switch, restores the business status from the saved system running context, and updates the causal digital twin model and background backup data.

[0015] Optionally, S1 specifically includes:

[0016] S11: Based on the physical architecture parameters, device parameters, thermodynamic model and communication protocol stack parameters of the RF power amplifier system, construct a digital twin virtual model of the RF power amplifier system;

[0017] S12: Introduce a structural causal model into the digital twin virtual model. The structural causal model represents the causal relationship between key state variables in the form of a directed acyclic graph.

[0018] S13: Use historical operating data of the RF power amplifier system to calibrate the structural causal model.

[0019] Optionally, S2 specifically includes:

[0020] S21: Determine the set of physical components of the radio frequency power amplifier system, the set including at least a power amplifier unit, a driver unit, a power supply unit, a heat dissipation unit, and a signal coupling unit;

[0021] S22: Map each physical unit to a graph node to form a node set;

[0022] S23: Determine the set of connection relationships between physical components, including electrical connections, thermal coupling connections, and control connections;

[0023] S24: Map each connection relationship to an edge connecting two corresponding graph nodes to form an edge set;

[0024] S25: Define an attribute vector for each graph node. The attribute vector contains the real-time operating parameters of the corresponding physical unit. The real-time operating parameters include at least voltage value, current value, temperature value, input signal power, output signal power, and signal quality index.

[0025] S26: Combine the attribute vectors of all graph nodes to form a node attribute matrix;

[0026] S27: Establish a dynamic spatiotemporal graph structure based on the node set, edge set, and node attribute matrix that changes over time.

[0027] Optionally, S3 specifically includes:

[0028] S31: Obtain the real-time operating parameters of each physical component of the RF power amplifier system through the data acquisition interface according to the preset sampling period;

[0029] S32: Fill the obtained real-time running parameters into the attribute vectors of the corresponding graph nodes according to the mapping relationship of the physical components;

[0030] S33: At the sampling time, arrange the attribute vectors of all graph nodes by row to form the node attribute matrix at the current sampling time;

[0031] S34: Combine the node attribute matrix at the current sampling time with a predefined graph structure to form a dynamic spatiotemporal graph at the current sampling time;

[0032] S35: Store the dynamic spatiotemporal graph at the current sampling moment into the buffer queue;

[0033] S36: Extract the dynamic spatiotemporal graph sequence within a continuous time window from the cache queue and output it as a spatiotemporal graph state sequence.

[0034] Optionally, S4 specifically includes:

[0035] S41: Input the spatiotemporal graph state sequence into the graph convolutional layer of a lightweight spatiotemporal graph neural network;

[0036] S42: Update the spatial feature representation of each graph node through the spatial feature extraction operation of the graph convolutional layer;

[0037] S43: Input the updated spatial feature representation of the graph nodes into the temporal convolutional layer of the lightweight spatiotemporal graph neural network;

[0038] S44: Extract the temporal evolution features of each graph node through one-dimensional convolution operation of the temporal convolutional layer;

[0039] S45: Perform feature fusion operation on the spatial feature representation and temporal evolution feature of each graph node;

[0040] S46: Input the fused features into the fully connected layer for global feature extraction;

[0041] S47: Input the global features into the regression output branch and the classification output branch respectively, and output the RF power amplifier system health score and abnormal mode code respectively.

[0042] Optionally, S5 specifically includes:

[0043] S51: Encode the abnormal pattern and map it to the initial perturbation value of the corresponding state variable in the causal digital twin model;

[0044] S52: Inject the current operating state parameters of the RF power amplifier system into the causal digital twin model and initialize all state variables of the model;

[0045] S53: Using the initial perturbation value as input, perform forward causal propagation iteratively in the causal digital twin model at a preset time step;

[0046] S54: In each iteration step, calculate the next time value of each state variable using the causal function in the structural causal model;

[0047] S55: Monitor the evolution of key state variables and stop iterating when preset conditions are met;

[0048] S56: Record the cumulative time corresponding to the time when the iteration stops as the time when the physical fault occurs;

[0049] S57: Based on the causal relationship chain defined in the structural causal model, trace the impact of physical failures on the data processing flow and determine the data processing risk time.

[0050] Optionally, S6 specifically includes:

[0051] S61: Determines that the health score of the RF power amplifier system is lower than the preset health threshold;

[0052] S62: The time of determining data processing risk is earlier than the time of physical failure;

[0053] S63: When the data processing risk time is earlier than the physical failure time, a first instruction sequence is generated. The first instruction sequence includes, in sequence, a system running context saving instruction, a backup power amplifier module preset instruction, and a primary / backup link switching instruction.

[0054] S64: When the data processing risk time is later than or equal to the physical failure time, a second instruction sequence is generated. The second instruction sequence includes, in sequence, the primary / backup link switching instruction, the system running context saving instruction, and the backup power amplifier module preset instruction.

[0055] S65: Adjust the execution parameters of each instruction in the instruction sequence according to the disaster recovery priority parameters in the preset recovery strategy;

[0056] S66: Encapsulate the adjusted instruction sequence into an instruction data packet;

[0057] S67: Sends instruction data packets to each execution unit interface via the system control bus.

[0058] Optionally, S7 specifically includes:

[0059] S71: In accordance with the requirements of the system running context save instruction, write the currently running protocol stack state data, data processing buffer pointer value, and system configuration register value into the specified address area of ​​non-volatile memory;

[0060] S72: According to the requirements of the preset instructions of the backup power amplifier module, send a power-on control signal to the power management chip, read the working parameters from the digital twin model parameter library, and write the working parameters into the configuration register of the backup power amplifier module through the serial peripheral interface bus.

[0061] S73: In accordance with the requirements of the main / backup link switching instruction, send a switching control signal to the RF switch control chip to switch the RF signal input port from the output port of the main power amplifier module to the output port of the backup power amplifier module.

[0062] S74: After the RF switch control chip returns the switching completion signal, read the protocol stack status data, data processing buffer pointer value, and system configuration register value from the specified address area of ​​the non-volatile memory, and write these values ​​into the corresponding protocol stack status variables, buffer pointer register, and system configuration register respectively.

[0063] S75: The recovered protocol stack state data, data processing buffer pointer value, and system configuration register value are sent to the causal digital twin model calculation engine via the data bus to update the internal state variables of the model.

[0064] S76: Copy the system runtime context data of the specified address area of ​​the non-volatile memory to the communication buffer, and send the communication buffer data to the network storage address of the background backup server through the Ethernet controller;

[0065] S77: Based on the updated causal digital twin model state, recalculate and set the normalization parameters of the input data of the lightweight spatiotemporal graph neural network to restore the data acquisition and processing flow.

[0066] The beneficial effects of this invention are:

[0067] This invention abstracts the physical topology of an RF power amplifier system into a dynamic spatiotemporal graph and utilizes a lightweight spatiotemporal graph neural network for real-time spatiotemporal feature extraction and health assessment. This enables early and accurate identification of latent fault precursors, effectively improving the sensitivity and accuracy of fault warnings and enhancing the depth and foresight of system state perception. By constructing a causal digital twin containing a structural causal model and performing forward causal inference based on the abnormal patterns output by the lightweight spatiotemporal graph neural network, it achieves accurate prediction of the occurrence time of physical faults and the time of data processing risks, significantly improving the intelligence and accuracy of disaster recovery decision-making and providing a reliable timing basis for subsequent collaborative instruction generation. Through a collaborative disaster recovery instruction sequence based on health scores and accurate prediction times, containing strict timing logic, it effectively solves the technical problems of the disconnect and lack of coordination between hardware switching and data storage processes in existing technologies. It breaks through the bottleneck of the difficulty in simultaneously optimizing recovery time and recovery point objectives in traditional solutions, achieving a significant advancement from passive response processing to proactive preventive collaborative disaster recovery, effectively improving the overall availability and data reliability of the RF power amplifier system. Attached Figure Description

[0068] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0069] Figure 1 This is a schematic diagram of a system fault recovery and data backup method based on signal and power amplifier proposed in this invention.

[0070] Figure 2 This is a schematic diagram illustrating the principle of digital twin forward inference and intelligent collaborative decision-making based on causal inference proposed in this invention. Detailed Implementation

[0071] Combination Figures 1-2 The present invention will be described in further detail below. These accompanying drawings are simplified schematic diagrams, illustrating only the basic structure of the invention and showing the main components relevant to the invention. Figure 1 and Figure 2 The present invention provides a system fault recovery and data backup method based on signal and power amplifier, comprising the following steps:

[0072] S1: Establish a causal digital twin model of the RF power amplifier system. The causal digital twin model includes a structural causal model that describes the causal relationships between key parameters of the RF power amplifier system.

[0073] S2: Based on the physical topology definition of the RF power amplifier system, the dynamic spatiotemporal graph includes nodes representing each functional module of the RF power amplifier system, edges representing the physical connections between functional modules, and a node attribute matrix composed of real-time operating data of the functional modules.

[0074] S3: Periodically collect real-time operating data of each functional module of the RF power amplifier system, update the node attribute matrix, and generate a spatiotemporal diagram state sequence;

[0075] S4: Input the spatiotemporal graph state sequence into the lightweight spatiotemporal graph neural network, extract spatiotemporal correlation features and output the RF power amplifier system health score and abnormal pattern code;

[0076] S5: Input the abnormal mode code and the current operating status of the RF power amplifier system into the causal digital twin model, and predict the time of physical failure and the time of data processing risk through causal inference;

[0077] S6: Based on the RF power amplifier system health score, physical fault occurrence time, data processing risk time and preset recovery strategy, generate and release a collaborative disaster recovery instruction sequence that includes system operation context saving instructions, backup power amplifier module preset instructions and main / backup link switching instructions.

[0078] S7: Executes a sequence of collaborative disaster recovery instructions, completes the primary / backup link switch, restores the business status from the saved system running context, and updates the causal digital twin model and background backup data.

[0079] In this embodiment, S1 specifically includes:

[0080] S11: Based on the physical architecture parameters, device parameters, thermodynamic model and communication protocol stack parameters of the RF power amplifier system, construct a digital twin virtual model of the RF power amplifier system;

[0081] S12: Introduce a structural causal model into the digital twin virtual model. The structural causal model represents the causal relationship between key state variables in the form of a directed acyclic graph.

[0082] S13: Use historical operating data of the RF power amplifier system to calibrate the structural causal model.

[0083] In this embodiment, S2 specifically includes:

[0084] S21: Determine the set of physical components of the radio frequency power amplifier system, the set including at least a power amplifier unit, a driver unit, a power supply unit, a heat dissipation unit, and a signal coupling unit;

[0085] S22: Map each physical unit to a graph node to form a node set;

[0086] S23: Determine the set of connection relationships between physical components, including electrical connections, thermal coupling connections, and control connections;

[0087] S24: Map each connection relationship to an edge connecting two corresponding graph nodes to form an edge set;

[0088] S25: Define an attribute vector for each graph node. The attribute vector contains the real-time operating parameters of the corresponding physical unit. The real-time operating parameters include at least voltage value, current value, temperature value, input signal power, output signal power, and signal quality index.

[0089] S26: Combine the attribute vectors of all graph nodes to form a node attribute matrix;

[0090] S27: Establish a dynamic spatiotemporal graph structure based on the node set, edge set, and node attribute matrix that changes over time.

[0091] In this embodiment, S3 specifically includes:

[0092] S31: Obtain the real-time operating parameters of each physical component of the RF power amplifier system through the data acquisition interface according to the preset sampling period;

[0093] S32: Fill the obtained real-time running parameters into the attribute vectors of the corresponding graph nodes according to the mapping relationship of the physical components;

[0094] S33: At the sampling time, arrange the attribute vectors of all graph nodes by row to form the node attribute matrix at the current sampling time;

[0095] S34: Combine the node attribute matrix at the current sampling time with a predefined graph structure to form a dynamic spatiotemporal graph at the current sampling time;

[0096] S35: Store the dynamic spatiotemporal graph at the current sampling moment into the buffer queue;

[0097] S36: Extract the dynamic spatiotemporal graph sequence within a continuous time window from the cache queue and output it as a spatiotemporal graph state sequence.

[0098] In this embodiment, S4 specifically includes:

[0099] S41: Input the spatiotemporal graph state sequence into the graph convolutional layer of a lightweight spatiotemporal graph neural network;

[0100] S42: Update the spatial feature representation of each graph node through the spatial feature extraction operation of the graph convolutional layer;

[0101] S43: Input the updated spatial feature representation of the graph nodes into the temporal convolutional layer of the lightweight spatiotemporal graph neural network;

[0102] S44: Extract the temporal evolution features of each graph node through one-dimensional convolution operation of the temporal convolutional layer;

[0103] S45: Perform feature fusion operation on the spatial feature representation and temporal evolution feature of each graph node;

[0104] S46: Input the fused features into the fully connected layer for global feature extraction;

[0105] S47: Input the global features into the regression output branch and the classification output branch respectively, and output the RF power amplifier system health score and abnormal mode code respectively.

[0106] In this embodiment, S5 specifically includes:

[0107] S51: Encode the abnormal pattern and map it to the initial perturbation value of the corresponding state variable in the causal digital twin model;

[0108] S52: Inject the current operating state parameters of the RF power amplifier system into the causal digital twin model and initialize all state variables of the model;

[0109] S53: Using the initial perturbation value as input, perform forward causal propagation iteratively in the causal digital twin model at a preset time step;

[0110] S54: In each iteration step, calculate the next time value of each state variable using the causal function in the structural causal model;

[0111] S55: Monitor the evolution of key state variables and stop iterating when preset conditions are met;

[0112] S56: Record the cumulative time corresponding to the time when the iteration stops as the time when the physical fault occurs;

[0113] S57: Based on the causal relationship chain defined in the structural causal model, trace the impact of physical failures on the data processing flow and determine the data processing risk time.

[0114] In this embodiment, S6 specifically includes:

[0115] S61: Determines that the health score of the RF power amplifier system is lower than the preset health threshold;

[0116] S62: The time of determining data processing risk is earlier than the time of physical failure;

[0117] S63: When the data processing risk time is earlier than the physical failure time, a first instruction sequence is generated. The first instruction sequence includes, in sequence, a system running context saving instruction, a backup power amplifier module preset instruction, and a primary / backup link switching instruction.

[0118] S64: When the data processing risk time is later than or equal to the physical failure time, a second instruction sequence is generated. The second instruction sequence includes, in sequence, the primary / backup link switching instruction, the system running context saving instruction, and the backup power amplifier module preset instruction.

[0119] S65: Adjust the execution parameters of each instruction in the instruction sequence according to the disaster recovery priority parameters in the preset recovery strategy;

[0120] S66: Encapsulate the adjusted instruction sequence into an instruction data packet;

[0121] S67: Sends instruction data packets to each execution unit interface via the system control bus.

[0122] In this embodiment, S7 specifically includes:

[0123] S71: In accordance with the requirements of the system running context save instruction, write the currently running protocol stack state data, data processing buffer pointer value, and system configuration register value into the specified address area of ​​non-volatile memory;

[0124] S72: According to the requirements of the preset instructions of the backup power amplifier module, send a power-on control signal to the power management chip, read the working parameters from the digital twin model parameter library, and write the working parameters into the configuration register of the backup power amplifier module through the serial peripheral interface bus.

[0125] S73: In accordance with the requirements of the main / backup link switching instruction, send a switching control signal to the RF switch control chip to switch the RF signal input port from the output port of the main power amplifier module to the output port of the backup power amplifier module.

[0126] S74: After the RF switch control chip returns the switching completion signal, read the protocol stack status data, data processing buffer pointer value, and system configuration register value from the specified address area of ​​the non-volatile memory, and write these values ​​into the corresponding protocol stack status variables, buffer pointer register, and system configuration register respectively.

[0127] S75: The recovered protocol stack state data, data processing buffer pointer value, and system configuration register value are sent to the causal digital twin model calculation engine via the data bus to update the internal state variables of the model.

[0128] S76: Copy the system runtime context data of the specified address area of ​​the non-volatile memory to the communication buffer, and send the communication buffer data to the network storage address of the background backup server through the Ethernet controller;

[0129] S77: Based on the updated causal digital twin model state, recalculate and set the normalization parameters of the input data of the lightweight spatiotemporal graph neural network to restore the data acquisition and processing flow.

[0130] Example 1: To verify the feasibility and significant effects of this invention in implementation, this example selects a typical 5G cellular network base station as the application scenario. This base station is equipped with two sets of RF power amplifier modules using the Doherty architecture, operating at 3.5GHz with an output power of 60W. Such base stations need to meet a high availability requirement of 99.999% and ensure that user session data is not lost and service interruption time is minimized in the event of a power amplifier failure. Traditional solutions employ hardware hot backup of the primary and backup power amplifiers, with independent temperature control protection and periodic configuration data backup. When the temperature of the primary power amplifier exceeds 85°C or the output VSWR exceeds 1.5, the system triggers a protective switch. Simultaneously, the system software backs up context data such as the protocol stack status and user connection table to local flash memory every 30 minutes. The main problem with existing methods is that temperature or VSWR alarms are often lagging indicators of a fault already occurring. The average time from alarm to switchover completion is approximately 100 milliseconds, during which service data processing is already disrupted. Furthermore, periodic data backups cannot capture the precise system state at the moment of switching, potentially leading to the loss of some real-time user data after recovery, thus impairing the user experience. This invention aims to solve the aforementioned problem of the temporal and logical disconnect between hardware switching and data preservation.

[0131] In this embodiment, a digital twin virtual model was first constructed based on the detailed design drawings of the base station's radio frequency unit, the power amplifier tube datasheet, the heat sink thermal resistance parameters, and the 5G NR protocol specifications. This model includes physical components such as the power amplifier driver stage, power amplifier final stage, circulator, load, power supply, and heat sink fins. Within this virtual model, a structural causal model was introduced to describe the causal relationships between key state variables. For example, there is a positive causal relationship between input power, supply voltage, ambient temperature, and power amplifier tube junction temperature, and a negative causal relationship between junction temperature rise and power amplifier efficiency and output signal adjacent channel leakage ratio. This structural causal model was implemented as a directed acyclic graph, and the functional relationships within it were calibrated using historical operating data from the base station over the past six months. The calibrated model had a root mean square error (RMSE) of 0.8°C in predicting junction temperature.

[0132] Based on the physical connections of the RF unit, a dynamic spatiotemporal graph is defined. Specifically, seven physical entities—the power amplifier tube, driver chip, DC power supply chip, temperature sensor, and output coupler—are mapped as graph nodes. Power supply lines, RF signal lines, and heat conduction paths between nodes are defined as edges, forming an edge set. An attribute vector is defined for each node; for example, the attributes of the power amplifier tube node include collector voltage, collector current, base temperature, input power, output power, and the error vector amplitude value calculated through real-time spectrum analysis. During system operation, attribute data of all nodes is synchronously collected at 1-millisecond intervals via a high-speed analog-to-digital converter and a digital predistortion feedback loop. The node attribute matrix is ​​updated, and the dynamic spatiotemporal graph state for each millisecond is stored in a first-in-first-out (FIFO) buffer queue with a capacity of 100, thus forming a 100-millisecond spatiotemporal graph state sequence.

[0133] This spatiotemporal graph state sequence is input into a pre-trained lightweight spatiotemporal graph neural network. The network aggregates information from each node and its neighbors through graph convolutional layers, updating the spatial feature representation of each node. For example, the features of a power amplifier node are fused with information from its driver chip and temperature sensor. Temporal convolutional layers analyze the feature evolution patterns of each node over the past 100 milliseconds. The network fuses the spatiotemporal features of all nodes through fully connected layers, outputting a system health score between 0 and 1, and a 128-dimensional anomaly pattern encoding vector. In the tests of this embodiment, the network model size is 2.1 MB, and forward inference can be completed within 5 milliseconds on an embedded processor.

[0134] The key creative step lies in injecting the anomaly pattern encoding vector along with the system's current user connection count, transport block size, and other operational state parameters into the aforementioned causal digital twin model. The twin model uses this anomaly encoding as the initial perturbation for specific state variables, accelerating forward causal deduction in 10-microsecond increments. For example, when the anomaly encoding indicates a "minor fluctuation trend in the driver stage bias current," the model deduces how this fluctuation, under the effect of thermal accumulation, will lead to the future evolution of the power amplifier tube junction temperature, and accurately calculates the predicted time when the junction temperature exceeds the safety threshold (preset to 105°C), denoted as the physical fault occurrence time. The model continues to deduce along the causal chain: the increased junction temperature leads to a decrease in power amplifier gain and increased distortion, which in turn causes the baseband processor to increase retransmissions to maintain link quality, deducing when the queue of user data packets will reach the overflow risk point, denoted as the data processing risk time. In a real-world simulation, the model predicted the physical fault occurrence time to be 8.5 seconds later, while the data processing risk time was 7.2 seconds later, indicating that the data risk occurs earlier than a complete hardware failure.

[0135] Based on the judgment that the health score is below the threshold of 0.3 and the data processing risk occurs before the physical failure occurs, the intelligent collaborative decision engine generates the first collaborative instruction sequence. This sequence strictly stipulates: immediately issue a high-level interrupt to the baseband processor, causing it to save approximately 2KB of critical operational data, including the radio bearer context of all active users, the hybrid automatic repeat request state machine, and the scheduler pointer, in the form of a memory snapshot to a ferroelectric random access memory, which takes about 0.5 milliseconds; simultaneously, instruct the power management chip to power on the backup power amplifier module and read the optimal operating point parameters from the digital twin model, configuring its bias and predistortion tables through the serial peripheral interface, which takes about 2 milliseconds; after receiving the "context saving complete" confirmation signal from the baseband processor, instruct the RF switch to complete the signal path switching within the next protection time slot (approximately 0.2 milliseconds). The total time of the entire collaborative process from decision to switching completion is controlled within 3 milliseconds.

[0136] To quantitatively evaluate the effectiveness of this invention, we simulated five typical progressive power amplifier failure scenarios in a laboratory environment and compared them with traditional threshold alarm switching schemes. The tests recorded the total recovery time from the onset of the failure to the complete restoration of service, as well as the integrity of data services after recovery. Specific data comparisons are shown in the table below:

[0137] Fault Scenario Recovery time (milliseconds) for traditional solutions Recovery time (milliseconds) of the present invention Traditional solution data packet loss rate (%) Data packet loss rate (%) of the present invention's solution Number of incorrect handovers in the traditional solution (times) Number of erroneous switching attempts in this invention (times) Scenario 1: Cooling fan slowly fails 102.5 3.1 0.15 0 0 0 Scenario 2: Power supply ripple increases slowly 98.7 3.2 0.08 0 1 0 Scenario 3: Slight aging of the driver chip 105.3 3.4 0.22 0 0 0 Scenario 4: Slight output load mismatch 95.8 3.0 0.05 0 2 0 Scenario 5: Slow Rise in Transistor Junction Temperature 110.2 3.3 0.31 0 0 0

[0138] Analysis of the data in the table above shows that the overall system recovery time of the proposed solution is consistently around 3.2 milliseconds under five different fault scenarios, representing an improvement of over 30 times compared to the average recovery time of approximately 102.5 milliseconds for the traditional solution. This is mainly attributed to the predictive capability of the proposed solution based on causal inference, which enables the collaborative recovery process to be initiated before hardware performance deteriorates to the point of triggering traditional thresholds, achieving "pre-fault" handling. Regarding data integrity, the proposed solution achieved zero data packet loss in all five tests, while the traditional solution, due to the loss of state during switching, resulted in an average packet loss rate of 0.162%, which is unacceptable for high-reliability services. Furthermore, in scenarios 2 and 4, the traditional solution triggered unnecessary protection switching (false switching) due to instantaneous fluctuations in power ripple and load. The proposed solution, through accurate identification of abnormal patterns using a spatiotemporal graph neural network and causal twin-based fault development inference, effectively avoids such false alarms, achieving zero false switching. In summary, this embodiment fully verifies that the present invention, through deep perception, intelligent inference, and collaborative execution, can significantly improve the recovery speed, data preservation capability, and decision-making accuracy of a high-reliability RF power amplifier system, and solves the fundamental problem of the separation of the two core processes in the prior art.

Claims

1. A system fault recovery and data backup method based on signal and power amplifier, characterized in that, Includes the following steps: S1: Establish a causal digital twin model of the RF power amplifier system. The causal digital twin model includes a structural causal model that describes the causal relationships between key parameters of the RF power amplifier system. S2: Based on the physical topology definition of the RF power amplifier system, the dynamic spatiotemporal graph includes nodes representing each functional module of the RF power amplifier system, edges representing the physical connections between functional modules, and a node attribute matrix composed of real-time operating data of the functional modules. S3: Periodically collect real-time operating data of each functional module of the RF power amplifier system, update the node attribute matrix, and generate a spatiotemporal diagram state sequence; S4: Input the spatiotemporal graph state sequence into the lightweight spatiotemporal graph neural network, extract spatiotemporal correlation features and output the RF power amplifier system health score and abnormal pattern code; S5: Input the abnormal mode code and the current operating status of the RF power amplifier system into the causal digital twin model, and predict the time of physical failure and the time of data processing risk through causal inference; S6: Based on the RF power amplifier system health score, physical fault occurrence time, data processing risk time and preset recovery strategy, generate and release a collaborative disaster recovery instruction sequence that includes system operation context saving instructions, backup power amplifier module preset instructions and main / backup link switching instructions. S7: Executes a sequence of collaborative disaster recovery instructions, completes the primary / backup link switch, restores the business status from the saved system running context, and updates the causal digital twin model and background backup data.

2. The system fault recovery and data backup method based on signal and power amplifier according to claim 1, characterized in that, S1 specifically includes: S11: Based on the physical architecture parameters, device parameters, thermodynamic model and communication protocol stack parameters of the RF power amplifier system, construct a digital twin virtual model of the RF power amplifier system; S12: Introduce a structural causal model into the digital twin virtual model. The structural causal model represents the causal relationship between key state variables in the form of a directed acyclic graph. S13: Use historical operating data of the RF power amplifier system to calibrate the structural causal model.

3. The system fault recovery and data backup method based on signal and power amplifier according to claim 1, characterized in that, S2 specifically includes: S21: Determine the set of physical components of the radio frequency power amplifier system, the set including at least a power amplifier unit, a driver unit, a power supply unit, a heat dissipation unit, and a signal coupling unit; S22: Map each physical unit to a graph node to form a node set; S23: Determine the set of connection relationships between physical components, including electrical connections, thermal coupling connections, and control connections; S24: Map each connection relationship to an edge connecting two corresponding graph nodes to form an edge set; S25: Define an attribute vector for each graph node. The attribute vector contains the real-time operating parameters of the corresponding physical unit. The real-time operating parameters include at least voltage value, current value, temperature value, input signal power, output signal power, and signal quality index. S26: Combine the attribute vectors of all graph nodes to form a node attribute matrix; S27: Establish a dynamic spatiotemporal graph structure based on the node set, edge set, and node attribute matrix that changes over time.

4. The system fault recovery and data backup method based on signal and power amplifier according to claim 1, characterized in that, S3 specifically includes: S31: Obtain the real-time operating parameters of each physical component of the RF power amplifier system through the data acquisition interface according to the preset sampling period; S32: Fill the obtained real-time running parameters into the attribute vectors of the corresponding graph nodes according to the mapping relationship of the physical components; S33: At the sampling time, arrange the attribute vectors of all graph nodes by row to form the node attribute matrix at the current sampling time; S34: Combine the node attribute matrix at the current sampling time with a predefined graph structure to form a dynamic spatiotemporal graph at the current sampling time; S35: Store the dynamic spatiotemporal graph at the current sampling moment into the buffer queue; S36: Extract the dynamic spatiotemporal graph sequence within a continuous time window from the cache queue and output it as a spatiotemporal graph state sequence.

5. A system fault recovery and data backup method based on signal and power amplifier according to claim 1, characterized in that, S4 specifically includes: S41: Input the spatiotemporal graph state sequence into the graph convolutional layer of a lightweight spatiotemporal graph neural network; S42: Update the spatial feature representation of each graph node through the spatial feature extraction operation of the graph convolutional layer; S43: Input the updated spatial feature representation of the graph nodes into the temporal convolutional layer of the lightweight spatiotemporal graph neural network; S44: Extract the temporal evolution features of each graph node through one-dimensional convolution operation of the temporal convolutional layer; S45: Perform feature fusion operation on the spatial feature representation and temporal evolution feature of each graph node; S46: Input the fused features into the fully connected layer for global feature extraction; S47: Input the global features into the regression output branch and the classification output branch respectively, and output the RF power amplifier system health score and abnormal mode code respectively.

6. The system fault recovery and data backup method based on signal and power amplifier according to claim 1, characterized in that, S5 specifically includes: S51: Encode the abnormal pattern and map it to the initial perturbation value of the corresponding state variable in the causal digital twin model; S52: Inject the current operating state parameters of the RF power amplifier system into the causal digital twin model and initialize all state variables of the model; S53: Using the initial perturbation value as input, perform forward causal propagation iteratively in the causal digital twin model at a preset time step; S54: In each iteration step, calculate the next time value of each state variable using the causal function in the structural causal model; S55: Monitor the evolution of key state variables and stop iterating when preset conditions are met; S56: Record the cumulative time corresponding to the time when the iteration stops as the time when the physical fault occurs; S57: Based on the causal relationship chain defined in the structural causal model, trace the impact of physical failures on the data processing flow and determine the data processing risk time.

7. The system fault recovery and data backup method based on signal and power amplifier according to claim 1, characterized in that, S6 specifically includes: S61: Determines that the health score of the RF power amplifier system is lower than the preset health threshold; S62: The time of determining data processing risk is earlier than the time of physical failure; S63: When the data processing risk time is earlier than the physical failure time, a first instruction sequence is generated. The first instruction sequence includes, in sequence, a system running context saving instruction, a backup power amplifier module preset instruction, and a primary / backup link switching instruction. S64: When the data processing risk time is later than or equal to the physical failure time, a second instruction sequence is generated. The second instruction sequence includes, in sequence, the primary / backup link switching instruction, the system running context saving instruction, and the backup power amplifier module preset instruction. S65: Adjust the execution parameters of each instruction in the instruction sequence according to the disaster recovery priority parameters in the preset recovery strategy; S66: Encapsulate the adjusted instruction sequence into an instruction data packet; S67: Sends instruction data packets to each execution unit interface via the system control bus.

8. A system fault recovery and data backup method based on signal and power amplifier according to claim 1, characterized in that, Specifically, S7 includes: S71: In accordance with the requirements of the system running context save instruction, write the currently running protocol stack state data, data processing buffer pointer value, and system configuration register value into the specified address area of ​​non-volatile memory; S72: According to the requirements of the preset instructions of the backup power amplifier module, send a power-on control signal to the power management chip, read the working parameters from the digital twin model parameter library, and write the working parameters into the configuration register of the backup power amplifier module through the serial peripheral interface bus. S73: In accordance with the requirements of the main / backup link switching instruction, send a switching control signal to the RF switch control chip to switch the RF signal input port from the output port of the main power amplifier module to the output port of the backup power amplifier module. S74: After the RF switch control chip returns the switching completion signal, read the protocol stack status data, data processing buffer pointer value, and system configuration register value from the specified address area of ​​the non-volatile memory, and write these values ​​into the corresponding protocol stack status variables, buffer pointer register, and system configuration register respectively. S75: The recovered protocol stack state data, data processing buffer pointer value, and system configuration register value are sent to the causal digital twin model calculation engine via the data bus to update the internal state variables of the model. S76: Copy the system runtime context data of the specified address area of ​​the non-volatile memory to the communication buffer, and send the communication buffer data to the network storage address of the background backup server through the Ethernet controller; S77: Based on the updated causal digital twin model state, recalculate and set the normalization parameters of the input data of the lightweight spatiotemporal graph neural network to restore the data acquisition and processing flow.

Images