A multi-core parallel processing method and device of a high-power microwave weapon

Abstract

This invention belongs to the technical field of electromagnetic countermeasures and defense, specifically disclosing a multi-core parallel processing method and device for high-power microwave weapons. The method includes: establishing a precise time protocol synchronization, and performing dynamic temporal masking based on the instantaneous power of the forward transmission of the microwave source to generate a mask scalar. Simultaneously, physical quantity normalization and fusion are performed on multi-source heterogeneous sensor data to output an initial observation vector of a unified dimension. The server digital signal processor calculates a deterministic computing power allocation threshold based on the total number of currently active targets to activate a corresponding number of client digital signal processors, and allocates the initial observation vector to a determined dedicated computing kernel through hash mapping. The purpose of this invention is to solve the problems of how to overcome the defects of existing systems when dealing with massive cluster targets, such as heterogeneous sensing self-interference, multi-core scheduling congestion, memory fragmentation, and the disconnect between information scheduling instructions and thermo-mechanical-electrical physical boundaries.

Description

Technical Field

[0001] This invention relates to the technical field of electromagnetic countermeasures defense, specifically to a multi-core parallel processing method and apparatus for high-power microwave weapons. Background Technology

[0002] With the evolution of highly dynamic and high-density UAV swarm warfare, high-power microwave weapons face constraints in both physical and information processing under complex electromagnetic environments. Heterogeneous data from distributed radar and optoelectronic sensors, due to asynchronous sampling clocks and differences in spatial coordinate references, are prone to target trajectory oscillations and spatial errors. Simultaneously, when microwave weapons emit energy pulses, antenna sidelobe leakage generates radio frequency shocks. Using traditional fixed long-term physical shielding windows to protect the radar's low-noise amplifier can lead to the loss of continuous observation data for high-speed penetrating targets, causing conventional tracking filtering algorithms to diverge or truncate tracks. When handling massive concurrent targets, existing signal processing systems generally rely on centralized task queues and thread pool scheduling based on general-purpose operating systems. This dynamic allocation mechanism, when faced with a surge of hundreds of targets, will suffer from context switching and overall... Contention for line resource locks leads to computational latency and scheduling congestion, and the frequent allocation and release of dynamic memory can easily cause memory fragmentation and stack overflow issues. In addition, at the level of firepower scheduling and execution, existing systems often process information domain calculations and physical domain boundaries separately, relying on threat level or radial distance for independent target interception sorting, without fully considering the rotational inertia loss and back EMF impact generated when the servo mechanical turntable switches scanning at large angles in space. This reduces the effective firepower interception throughput per unit time, and rarely considers hardware conditions such as the transient voltage drop of the microwave source energy storage bus voltage and the thermal tolerance of the phase shifter array, which can easily lead to thermoelectric overload of the defense platform during continuous combat. Summary of the Invention

[0003] This invention provides a multi-core parallel processing method and apparatus for high-power microwave weapons, aiming to solve the problems of how to overcome the defects of existing systems when dealing with massive cluster targets, such as heterogeneous sensing self-interference, multi-core scheduling congestion, memory fragmentation, and the disconnect between information scheduling instructions and thermo-mechanical-electrical physical boundaries.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A multi-core parallel processing method for high-power microwave weapons, deployed on a heterogeneous parallel processing array including a server digital signal processor and multiple client digital signal processors, includes the following steps: establishing a precise time protocol synchronization, performing dynamic temporal masking based on the instantaneous power of the microwave source's forward transmission to generate a mask scalar, and simultaneously normalizing and fusing physical quantities of multi-source heterogeneous sensor data to output an initial observation vector of uniform dimension; the server digital signal processor calculates a deterministic computing power allocation threshold based on the total number of currently active targets to activate a corresponding number of client digital signal processors, and allocates the initial observation vector to a determined dedicated processing array through hash mapping. The dedicated computational kernel receives the initial observation vector, combines it with the mask scalar, and performs parallel recursive extended Kalman filtering with mask compensation to extrapolate and generate the target's spatial prediction coordinates. Based on the spatial prediction coordinates, a coupled model of target arrival time and system physical preparation time is established to screen targets. Under the physical constraints of bus dynamic voltage drop and array predicted thermal tolerance, a fire strike sequence that minimizes mechanical work is generated to control the microwave radio frequency hardware to perform interception. After microwave interception, the target's hit characteristics are extracted to perform damage probability determination, and the bitmap memory of targets determined to be ineffective is automatically aged and reclaimed using a hardware-level lifetime counter.

[0005] In one aspect of the present invention, the dynamic timing masking generation of a mask scalar based on the forward transmission instantaneous power of the microwave source specifically includes: The dynamic multipath tail time is calculated based on the obtained instantaneous power of the microwave source in the forward direction and the set reference trigger power threshold constant. The dynamic multipath tail time is added to the radar protection warning time advance and the physical width of the microwave pulse to generate the hardware masking pulse width. During the high-level period of the hardware masking pulse width, the mask scalar is set to a state value that prohibits observation updates, and during other normal working periods, the mask scalar is set to a state value that allows observation updates.

[0006] In one aspect of the present invention, the calculation of the deterministic computing power allocation threshold and the allocation of the initial observation vector to a determined dedicated computing kernel through hash mapping specifically include: Calculate the theoretical minimum number of physical processing cores required based on the total number of currently active targets and the set single-core processing power constant; The total number of digital signal processor chips that need to be activated is determined by combining the theoretical minimum number of physical operation cores. The physical chip number to which the initial observation vector is mapped is determined by taking the modulus of the total number of digital signal processor chips using the target's unique identification number, and then taking the modulus again using the unique identification number within that physical chip to determine the specific dedicated computing core.

[0007] In one aspect of the invention, the parallel recursive execution of extended Kalman filtering with mask compensation to extrapolate and generate the spatial prediction coordinates of the target specifically includes: When the input mask scalar is a prohibited observation update state value, the sensor input is automatically blocked, and the spatial predicted coordinates are updated by inertial extrapolation prediction based on the physical model. When the input mask scalar is an observation-updated state value, the spatial prediction coordinates are corrected using the calculated Kalman gain matrix and the initial observation vector.

[0008] In one aspect of the invention, generating a fire strike sequence that minimizes mechanical work specifically includes: Calculate the mechanical transfer cost between any two targets in the target set to be attacked. The mechanical transfer cost includes the azimuth mechanical displacement cost and the pitch mechanical displacement cost based on the maximum rated angular velocity of the servo turntable. A Hamiltonian complete graph is constructed with the target as the node and the mechanical transfer cost as the edge. The cost function that minimizes the total mechanical work is solved by dynamic programming, and the corresponding optimal traversal sequence is generated as the fire strike sequence.

[0009] In one aspect of the present invention, the physical amplitude limiting constraint is specifically implemented as follows: During the charging of the microwave source energy storage capacitor, the instantaneous bus voltage is sampled in real time. When the predicted voltage drop rate exceeds the safety threshold, the energy supply of the target is guaranteed according to the calculated priority score from high to low. The output current of the servo turntable motor is forcibly clamped based on the minimum bus voltage constant and the equivalent internal resistance. The temperature of the phase shifter array nodes is obtained, and when the node temperature approaches the hardware safety temperature threshold, the single dwell time of the microwave beam is controlled to decrease exponentially.

[0010] In one aspect of the present invention, the extraction of the target's hit characteristics to perform damage probability determination specifically includes: A three-dimensional impact feature vector is constructed by extracting the Doppler velocity mutation rate, radar cross-section scintillation variance, and photoelectric miss distance dispersion from the output stream of the extended Kalman filter. The three-dimensional impact feature vector is weighted and summed using a nonlinear weight matrix that has been pre-fitted and trained with experimental data, and the final damage probability value is output through a mapping function.

[0011] In one aspect of the present invention, the automatic aging-up and reclamation of bitmap memory for targets determined to be invalid, combined with a hardware-level lifetime counter, specifically includes: The hardware-level lifetime counter is appended to a fixed-length data block in the on-chip static random access memory. When the target state vector is mapped into the memory block, the hardware-level lifetime counter is initialized to the maximum failure threshold cycle number. When each radar scan interruption is triggered, if the associated gate determines that no new valid observation update has been obtained, the hardware-level lifetime counter is decremented. When the hardware-level lifetime counter decrements to zero, a hardware-level atomic operation is triggered, which directly clears the memory resource allocation status bitmap flag through bitwise logical operations on the bitmap mask word to release the physical memory block.

[0012] In one aspect of the invention, the heterogeneous parallel processing array is divided into a data plane and a control plane based on the physical architecture of the communication backplane; wherein, the server digital signal processor and multiple client digital signal processors perform large-scale transfer of the initial observation vector through the data plane; and the inter-core synchronization interrupt, the mask scalar, the bitmap reset instruction for automatic memory aging and reclamation, and the fire strike sequence are independently transmitted through the control plane.

[0013] In another aspect, the present invention also relates to a multi-core parallel processing device for a high-power microwave weapon, comprising a server digital signal processor, a client digital signal processor, a direct memory access controller, and a memory. The memory stores a computer program, which, when executed by the server digital signal processor and the client digital signal processor, implements the multi-core parallel processing method for high-power microwave weapons as described above.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention first protects the radar from high-energy beam burn-out by using physical energy-adaptive dynamic temporal masking and multi-source data spatial normalization. This not only eliminates spatial distortion and track oscillation caused by traditional fixed blind spots and asynchronous clocks, but also ensures high-precision spatiotemporal consistency of observation data under extreme electromagnetic interference. Next, the introduction of a deterministic computing power threshold and atomic-level hash mapping mechanism replaces the soft congestion bottleneck of centralized thread scheduling in traditional operating systems. This achieves zero-blocking load balancing for hundreds of concurrent targets with constant-time complexity, significantly reducing bus arbitration and context switching overhead. Finally, the implementation of masked recursive filtering and superscalar pipeline reorganization in the computation kernel maintains continuous track prediction using a pure inertial model during RF-constrained periods and compresses the solution latency of high-dimensional matrices to the hardware limit, ensuring the ultimate timeliness of spatial trajectory prediction. Next, the underlying physical boundaries, such as target convergence time, minimizing servo mechanical work using Hamiltonian optimization, and bus voltage and array thermal tolerance, are strongly coupled to avoid ineffective large-angle reciprocating losses of the mechanical turntable. This eliminates the risk of system failure due to thermal overload of the defense platform under extreme continuous firing conditions, achieving optimal and precise conversion of digital computing power into physical destructive kinetic energy. Finally, high-confidence damage assessment is performed by extracting three-dimensional kinematic impact features, and hardware-level lifetime counters and atomic bit operations are innovatively used to perform automatic aging and reclamation of physical memory. This effectively avoids memory fragmentation and stack overflow crashes under massive high-concurrency scheduling, giving the defense system continuous and stable computing power liquidity and high combat robustness when continuously resisting saturation swarm attacks. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a multi-core parallel processing method for a high-power microwave weapon according to the present invention.

[0017] Figure 2 This is a step-by-step diagram of step 1 in the multi-core parallel processing method for a high-power microwave weapon according to the present invention.

[0018] Figure 3 This is a step-by-step diagram of step 2 in the multi-core parallel processing method for a high-power microwave weapon according to the present invention.

[0019] Figure 4 This is a step-by-step diagram of step 3 in the multi-core parallel processing method for a high-power microwave weapon of the present invention.

[0020] Figure 5 This is a step-by-step diagram of step 4 in the multi-core parallel processing method for a high-power microwave weapon of the present invention.

[0021] Figure 6 This is a step-by-step diagram of step 5 in the multi-core parallel processing method for a high-power microwave weapon according to the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to embodiments. These embodiments are merely some, not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the protection scope of the present invention.

[0023] Please see Figure 1 As shown, this embodiment discloses a multi-core parallel processing method for high-power microwave weapons. In one implementation, the method includes: S1. Establish a precise time protocol for synchronization, and perform dynamic timing masking based on the instantaneous power of the forward transmission of the microwave source to generate a mask scalar. At the same time, normalize and fuse the physical quantities of the multi-source heterogeneous sensor data to output an initial observation vector with a unified dimension. S2. The server digital signal processor calculates the deterministic computing power allocation threshold based on the total number of currently active targets to activate the corresponding number of client digital signal processors, and allocates the initial observation vector to a determined dedicated computing kernel through hash mapping. S3. The dedicated computing kernel receives the initial observation vector, combines it with the mask scalar, performs parallel recursive extended Kalman filtering with mask compensation, and extrapolates to generate the spatial prediction coordinates of the target. S4. Based on the spatial predicted coordinates, a coupled model of target arrival time and system physical preparation time is established to screen targets. Under the physical constraints of bus dynamic voltage drop and array predicted thermal tolerance, a fire strike sequence that minimizes mechanical work is generated to control microwave radio frequency hardware to perform interception. S5. After microwave interception, extract the target's hit characteristics to determine the probability of damage, and combine the hardware-level lifetime counter to automatically age and reclaim the bitmap memory of targets that are determined to be ineffective.

[0024] In this embodiment, it should be noted that in S1, high-power microwave weapons operating in complex electromagnetic countermeasures environments face the technical challenges of severe self-interference from their own gigawatt-level microwave energy on the front-end radar receiver, as well as asynchronous sampling clocks from multiple sensor sources. S1 establishes a perception synchronization mechanism based on physical energy adaptation. Specifically, this step not only performs nanosecond-level precise time protocol clock alignment between the radar and optoelectronic equipment at the network level, but also introduces the forward transmission power of the microwave source as a feedforward adjustment parameter. By extracting the instantaneous transmission power, the system calculates the dynamic tail time of electromagnetic multipath scattering in real time, thereby generating a time-series masking pulse and mask scalar that accurately covers the interference cycle. This processing logic changes the conventional practice of using fixed blind zone windows in existing air defense systems, elevating simple time alignment to a collaborative approach involving both time and energy dimensions. It not only solves the problem of benchmark unification in heterogeneous data fusion, but also protects the radar's low-noise amplifier from burnout at the physical hardware level, while minimizing the radar's "blinding" time, providing a high signal-to-noise ratio and spatiotemporally consistent data source for the subsequent generation of high-precision initial observation vectors.

[0025] In S2, to address the issue of excessive single-core load overflow and context switching overhead caused by conventional task queue polling mechanisms based on general-purpose operating systems when facing drone swarm saturation attacks, S2 introduces a nonlinear topology mapping mechanism based on deterministic computing power thresholds. Specifically, instead of passively piling tasks into a single queue, the system actively monitors the total number of active targets in the airspace and, based on a pre-defined single-core processing power constant, directly determines the number of client digital signal processor physical chips that need to be woken up through algebraic operations. When establishing task mapping, centralized software distribution is abandoned, and instead, the unique identification number of the target is used as a hash key. Through hardware-level modulo operations, the target state vector is directly mapped atomically to the designated dedicated computing core. This logic reduces the scheduling time complexity of multi-core computing power to O(1), solves the bus arbitration conflict under large-scale concurrency, and realizes elastic physical coupling between computing power resources and target density.

[0026] In S3, to address the issue of track divergence and loss during target maneuvers and brief radar obfuscation, S3 implements a parallel recursive extended Kalman filter with mask compensation in conjunction with the underlying hardware pipeline. Specifically, after receiving the initial observation vector, the dedicated computational kernel does not mechanically execute the standard filtering equation, but instead reads the mask scalar generated in S1 in real time. When the mask scalar indicates that the current period is under strong interference from microwave transmission, the algorithm automatically blocks uncertain sensor inputs and relies entirely on a preset nonlinear kinematic physical model for pure inertial prediction extrapolation. Once the obfuscation period ends, the algorithm uses the accumulated covariance matrix and Kalman gain for rapid correction. Simultaneously, this solution process fully utilizes Very Long Instruction Word (VLIW) and Single Instruction Multiple Data (SIMD) extensions to achieve superscalar pipeline reassembly of matrix multiplication and addition. This calculation process solves the tracking truncation problem caused by radio frequency self-interference, ensuring the continuity of spatial predicted coordinates and hardware-level ultra-low latency output in complex interference environments.

[0027] In S4, to address the technical issues of traditional air defense systems that rely solely on independent scheduling based on "distance" or "threat level," leading to ineffective oscillations of the servo mechanical turntable and ignoring the thermodynamic limits of chassis power supply, S4 employs a comprehensive decision-making approach based on physical boundary coupling and minimizing work. Specifically, the system first converts the predicted target spatial coordinates into target arrival times and establishes a coupling model based on the system's own physical preparation time from servo deflection to microwave energy storage, accurately selecting targets within the optimal strike window. When generating the final fire strike sequence, physical attributes such as the servo motor's maximum angular velocity and moment of inertia are abstracted into mechanical transfer costs. By solving the path planning of the Hamiltonian graph, a traversal sequence minimizing overall mechanical work is derived. More importantly, before issuing the strike command, the system samples the bus voltage drop rate and phase shifter array junction temperature in real time. Once these approach physical limits, the servo current is immediately clamped according to priority scores, and the beam dwell time is exponentially limited. This scheduling logic, which deeply integrates kinematics, thermodynamics, and electricity, prevents the risk of overload and shutdown of weapon platforms during extreme firing, and effectively converts pure computing power output into effective destructive kinetic energy that conforms to physical boundaries.

[0028] In S5, to address the issue of static memory exhaustion leading to system memory leaks and crashes during sustained combat against saturation swarms due to false clutter or the failure to promptly clear destroyed targets, S5 proposes a closed-loop performance evaluation and hardware-level bitmap automatic aging and recycling mechanism. Specifically, after completing a single microwave interception, the system immediately extracts the target's Doppler mutation rate, radar cross-section scintillation variance, and other attack characteristics, substituting them into a pre-trained Logistic nonlinear matrix to assess the precise damage probability. For targets determined to be physically ineffective or timed out due to continuous inability to correlate, the system triggers atomic operations through an internally bound hardware-level lifetime counter, directly utilizing bitmap mask word bitwise logical operations to instantly clear and release the physical memory block it occupies within a single clock cycle. This recycling mechanism completely abandons the slow garbage collection mechanism of the operating system, eliminating memory fragmentation at the physical level and ensuring the ultimate stability of multi-core parallel processing devices during dozens of waves of high-intensity defensive operations.

[0029] like Figure 2 As shown, in one specific embodiment, S1 includes: S11. Establish a precise time protocol for synchronization. Obtain an absolute time reference through a network time protocol, measure network path delay, and perform clock alignment compensation for multi-source sensors.

[0030] S12. Dynamic timing masking is performed based on the forward transmission instantaneous power of the microwave source to generate a mask scalar. The forward instantaneous power of the microwave RF hardware is read, the dynamic multipath tail time and the hardware masking pulse width are calculated, and a mask scalar characterizing the system observation confidence is generated.

[0031] S13. Perform physical quantity normalization and fusion on the data from multi-source heterogeneous sensors. Map the polar coordinates and miss distance data of radar and photoelectric sensors to a unified Cartesian coordinate system through the Jacobian matrix, and output an initial observation vector with unified dimension.

[0032] In this embodiment, it should be noted that in S11, the high-power microwave weapon system includes a long-range search radar, a short-range tracking optoelectronic turntable, and a core server digital signal processor. Because the sampling clock crystals of each device have inherent physical temperature drift, and data transmission over gigabit Ethernet or fiber optics encounters queuing jitter from switches, directly fusing messages from different times would cause spatial distortion of the target's state. S11 achieves nanosecond-level synchronization by establishing a precise time protocol conforming to the IEEE 1588 standard.

[0033] Specifically, the server's digital signal processor sends a synchronization request message to the radar equipment and records the sending timestamp. The radar equipment receives the message and returns a response message. The system obtains the timestamp sequence through a four-way handshake and calculates the network path delay. The calculation formula is as follows:

[0034] in, This is the measured network path latency time, in seconds; The local absolute timestamp, in seconds, for sending synchronization request messages to the server's digital signal processor; The local absolute timestamp, in seconds, for radar or optoelectronic equipment receiving a synchronization request message; The local absolute timestamp, in seconds, for sending return response messages to radar or optoelectronic equipment; This is the local absolute timestamp, in seconds, for the server's digital signal processor to receive the returned response message. The constant used to calculate the round-trip average is dimensionless.

[0035] After obtaining the precise network path delay time, the server's digital signal processor performs hardware-level timing compensation on each frame of received sensor messages to calculate the aligned deterministic timestamp. The calculation formula is as follows:

[0036] in, A deterministic timestamp aligned after network and hardware compensation, in seconds; This is the original local timestamp recorded by the sensor when it acquires the transient response of the target echo, in seconds; The network path delay time calculated using the above formula is in seconds. This is the inherent processing delay constant for sensor analog-to-digital conversion (ADC) and internal bus transmission. This constant is statically calibrated at the factory and is measured in seconds.

[0037] This step eliminates the trajectory prediction time deviation caused by nondeterministic network communication through the underlying clock compensation formula, enabling heterogeneous sensor data from different physical locations with different sampling rates to be strictly aligned within the same time slice of the server's digital signal processor. This provides a time reference with extremely high physical consistency for human-machine collaboration and subsequent filtering extrapolation.

[0038] In this embodiment, it should be noted that in S12, when the high-power microwave weapon is activated, the electromagnetic energy leaked from the sidelobes of its directional antenna reaches several kilovolts per meter. If the radar equipment's receiving link is on at this moment, its core low-noise amplifier will be instantly damaged. Traditional defense systems typically set a fixed, extremely conservative long-term shielding window (e.g., a fixed shielding of 50 milliseconds). While this protects the radar, it leads to the loss of a large amount of critical observation data when dealing with high-speed penetrating drones, causing tracking filtering divergence. S12 introduces a dynamic adaptive shielding model based on real transmission energy feedback. Specifically, simultaneously with issuing the microwave activation command, the server's digital signal processor reads the voltage of the directional coupler connected in series on the microwave source waveguide feeder in real time through a hardware-level high-speed analog-to-digital converter interface, resolving the current real forward transmission instantaneous power. Subsequently, the system calculates the dynamic multipath tail time based on the physical laws of multipath scattering of electromagnetic waves in complex defense positions (such as those with building surfaces or metal structures). The calculation formula is as follows:

[0039] in, To calculate the obtained dynamic multipath tail time; The set basic environmental electromagnetic tail time constant; The electromagnetic scattering coefficient of the pre-set defensive position space; It is the natural logarithm function operator; The instantaneous power value of the microwave source during forward transmission, fed back by the directional coupler; This is the set reference trigger power threshold constant.

[0040] After acquiring the dynamic multipath trail time, the server's digital signal processor integrates the warning lead and the beamwidth itself to generate the hardware masking pulse width for ultimately controlling the RF switch of the radar transceiver (T / R) component. The calculation formula is as follows:

[0041] in, The width of the generated hardware masking pulse; The advance time for radar protection warning; This is the physical width of the microwave pulse transmitted in this operation; The dynamic multipath tail time is calculated above.

[0042] Furthermore, in order to enable the software-level Kalman filter algorithm to achieve logical handshake with the hardware-level timing masking, the server digital signal processor synchronously generates a mask scalar. Its logical segmentation formula is:

[0043] When the system clock is in During the high-level pulse period 1, other normal operating periods include... is the dimensionless scalar of the masking state output at the k-th discrete solution time; 0 means observation is prohibited from updating the state value, which means that the sensor echo at this time has been severely contaminated by strong microwaves or physically blocked by radio frequency switches, and the observation vector is unreliable; 1 means observation is allowed to update the state value, which means that the environmental electromagnetic background has been restored to a safe noise level, and the observation vector is reliable. The above calculation refers to the hardware masking pulse width. This process seamlessly transforms the pure physical energy (watts) of microwave transmission into control variables (time and Boolean logic) for digital signal processing. When performing point-fire interception at low power, the system automatically generates an extremely short masking pulse, allowing the radar to instantly regain its field of view; while during full-power saturation destruction, the system automatically extends the trailing compensation to safeguard the hardware's safety baseline. This deep coupling of energy, timing, and logic completely resolves the technical contradiction between sensing failure and processing logic paralysis in traditional equipment under strong adversarial environments.

[0044] In this embodiment, it should be noted that in S13, the high-power microwave weapon defense platform is equipped with radar equipment for wide-area search and optoelectronic equipment for precision tracking. These two types of sensors have fundamental physical differences in sampling frequency, coordinate system reference, and error characteristics. The radar equipment outputs low-frequency polar coordinate data, while the optoelectronic equipment outputs high-frequency two-dimensional angular deviation (i.e., miss distance) relative to the optical axis center. Directly adding these values ​​in polar coordinates would lead to spatial distortion and severe oscillations in the target trajectory. S13 implements a normalization and fusion of physical quantities from multi-source heterogeneous sensor data based on the Jacobian matrix.

[0045] Specifically, the original polar coordinate vector reported by the radar equipment is denoted as... The system first calculates the network path delay time determined in step S11. The target's Doppler velocity, measured by radar, is used to pre-estimate its position in the physical time dimension. This pre-estimated position is then projected onto the Cartesian coordinate system of the weapon's strike center using a spatial homogeneous transformation matrix. The transformation matrix equation is as follows:

[0046] in, The x-axis physical position component of the target in the sensor coordinate system, in meters; This represents the physical position component of the target's y-axis in the sensor coordinate system, in meters. This represents the physical position component of the target along the z-axis in the sensor coordinate system, in meters. Initial radial distance to the target measured by radar equipment, in meters; Radial Doppler velocity of a target measured by radar equipment, in meters per second; Network path delay time, in seconds; The first elevation angle measured for radar equipment, in radians; The first azimuth angle measured by the radar equipment, in radians; For cosine function operators; It is a sine function operator.

[0047] Subsequently, high-frequency miss distance data from the photoelectric equipment is introduced. The photoelectric miss distance is the minute angle by which the target's true position deviates from the mechanical optical axis of the photoelectric system. The system linearly maps this two-dimensional angular deviation to a position correction in a three-dimensional Cartesian coordinate system using a Jacobian matrix, ultimately synthesizing and outputting an initial observation vector of uniform dimension. The corrected equation is as follows:

[0048] in, The initial observation vector with unified dimensions is the output after fusion, which contains the precise three-dimensional position components in the Cartesian coordinate system; A rotation matrix for correcting sensor installation errors is used to eliminate rigid mechanical installation deviations between the radar phase center and the microwave weapon launch phase center. A constant matrix of dimension; , , The above calculations are used to estimate the physical location components. The Jacobian matrix for coordinate transformation is obtained by taking the partial derivative from polar coordinates to Cartesian coordinates. Dimensional transformation matrix; The azimuth angle of the target miss distance, fed back by the optoelectronic equipment, is expressed in radians. The pitch angle deviation from the target, fed back by the optoelectronic equipment, is expressed in radians.

[0049] This step, based on mathematical principles, eliminates the spatial position measurement ambiguity caused by the large beam angle of the radar. By using the Jacobian matrix, the high-precision optical angle deviation is rigorously converted into displacement compensation in the Cartesian coordinate system, ensuring that the initial observation vector input to the multi-core parallel processing array possesses centimeter-level spatial resolution sufficient for the narrow-beam strike accuracy of microwave weapons.

[0050] like Figure 3 As shown, in one specific embodiment, S2 includes: S21. The server digital signal processor calculates a deterministic computing power allocation threshold based on the total number of currently active targets to activate the corresponding number of client digital signal processors.

[0051] S22. The initial observation vector is assigned to a specific dedicated computational kernel through hash mapping. Two modulo operations are performed using the target's unique identification number as the key to achieve atomic-level physical mapping of the target state vector.

[0052] In this embodiment, it should be noted that in S21, when facing a rapidly increasing number of drone "swarm" targets with high dynamics, the traditional operating system-based task queue mechanism relies on a centralized thread scheduler. When the number of targets exceeds a certain limit, thread context switching and kernel spinlock contention can lead to nonlinear congestion and delays in the system. S21 transforms the nonlinear computational pressure into discrete wake-up of physical chips through a hardware-level deterministic computing power allocation threshold formula.

[0053] Specifically, at the beginning of each processing cycle, the server's digital signal processor hardwires the total number of active targets in the current receive buffer. Based on a pre-calibrated single-core processing capacity constant, the system calculates the theoretical minimum number of physical cores required to meet the extremely low latency requirements. The calculation formula is as follows:

[0054] in, The theoretical minimum number of physical computing cores required to meet the real-time processing requirements across the entire airspace is a positive integer; This is the floor operator; This represents the total number of active targets currently reported and confirmed as real by the radar equipment; it is a positive integer. This is a single-core processing capability constant. Its physical meaning is the maximum target processing capacity that a single dedicated computing core can stably complete the extended Kalman filter solution within a specified single control cycle, and it is a positive integer.

[0055] After obtaining the required number of physical cores, the server digital signal processor sends instructions to the hardware power management module via the control plane bus to determine and activate the required total number of digital signal processor chips. The calculation formula is as follows:

[0056] in, This represents the number of digital signal processor chips that the system currently needs to wake up and are in working condition; it is a positive integer. This represents the theoretical minimum number of physical operation cores required for the above calculations; The number of dedicated computing cores available inside the server's digital signal processor is a constant. The number of dedicated computing cores integrated within each client digital signal processor is a constant; This is to compensate for the base number by including the server's digital signal processor itself in the total.

[0057] This computational logic ensures that the allocation of computing resources strictly follows the mathematical laws of target density. When targets are sparse, the system maintains only a minimum of core operations to reduce power consumption and heat generation on the backplane; when targets are abundant, the system wakes up the physical chip within a nanosecond clock cycle, fundamentally eliminating the "soft congestion" bottleneck caused by software-level thread pool scheduling.

[0058] In S22, after completing the macroscopic activation of physical computing nodes, the system must rapidly and conflict-free transmit massive amounts of target observation vectors to each specific dedicated computing core. S22 abandons a centralized task dispatcher and implements a constant time complexity... Atomic-level hash mapping mechanism.

[0059] Specifically, the system uses the unique identification number (ID) assigned to each target by the radar equipment directly as the input variable of the hash function. First, a first-level chip-level hash is performed to determine the physical chip number to which the initial observation vector is routed. The calculation formula is as follows:

[0060] in, The index number of the target physical chip to which the initial observation vector of the target is mapped, and its value range is... to Integers; A unique identification number assigned to the target by the system, which is a positive integer; For the modulo operator; This represents the total number of currently active digital signal processor chips calculated above.

[0061] After the data packet arrives at the designated physical chip via the direct memory access controller, the general-purpose management kernel inside the chip further performs a second-level kernel-level hash to determine the dedicated computing core number that will ultimately undertake the target computation task. The calculation formula is as follows:

[0062] in, This is an index number for the dedicated computing core within the chip that performs the specific computational tasks; its value range is... to Integers; This is the floor operator; A unique identification number for the target; This represents the total number of currently active digital signal processor chips. For the modulo operator; This is a constant representing the total number of dedicated computing cores integrated within a single digital signal processor chip.

[0063] This rigorous pure algebraic mapping model decouples the dependency between tasks and the scheduler. When any data packet arrives at the backend switch, its host core physical address has been uniquely and definitively calculated. The data flow is like flowing in a hard-wired pipe, achieving non-blocking parallel load balancing.

[0064] like Figure 4 As shown, in one specific embodiment, S3 includes: S31: A dedicated computational kernel receives the initial observation vector, combines it with a mask scalar, and performs parallel recursive extended Kalman filtering with mask compensation. This constructs a nine-dimensional state-space model including position, velocity, and acceleration.

[0065] S32. Calculate the Jacobian matrix of the nonlinear observation matrix in real time, and determine the filter operating mode based on the mask scalar. When the mask scalar prohibits observation updates of state values, perform pure inertial extrapolation; when it allows state values, perform matrix correction.

[0066] S33. Utilize the Single Instruction Multiple Data (SIMD) extended instruction set to perform superscalar pipeline reconfiguration and extrapolate to generate the spatial predicted coordinates of the target.

[0067] In this embodiment, it should be noted that in S31, when dealing with highly maneuverable targets, a simple linear filtering model would cause the target to deviate from its trajectory during acceleration or sharp turns. S31 constructs a rigorous nine-dimensional physical state space model in a dedicated computing kernel.

[0068] Specifically, the target's state vector at the k-th sampling time Defined as:

[0069] in, Let be the nine-dimensional state vector describing the full motion attributes of the target at time k; These are the three-axis absolute physical position components of the target in the Cartesian coordinate system, in meters; These are the physical velocity components of the target on the three axes, in meters per second; These are the physical acceleration components of the target along the three axes, in meters per second squared. This is the matrix transpose operator.

[0070] Based on the physical mechanism of constant acceleration (CA), the system establishes a state transition matrix. Its vectorized arrangement is defined as:

[0071] in, for A system state transition matrix of dimension 1; The fixed physical time step for system calculation, in seconds; The Kronecker integral operator is used to rapidly extend a single-axis kinematic model to three-dimensional space. for An identity matrix of dimension 1.

[0072] In S32, due to the mask scalar generated in step S12 The real-time indication of the physical reliability of radar observations necessitates that a dedicated computational kernel embed this Boolean logic into tensor operations within each computational step. Furthermore, there is a strong nonlinearity between the radar's polar coordinate observations and the Cartesian state vector.

[0073] The system first calculates the Jacobian matrix of the observation equation in real time. Observation function for:

[0074] Jacobian matrix The partial derivative expansion under the current prior prediction state is:

[0075] in, This is a nonlinear mathematical observation function that transforms Cartesian coordinates to polar coordinates; The position component in the state vector; for The nonlinear observation transformation Jacobian matrix of the dimension; The mathematical symbol for partial derivatives; It is the prior predicted state vector extrapolated from the state at the k-th time based on the state at the previous time.

[0076] Subsequently, the dedicated computing kernel executes the masked scalar. The update equation for compensation:

[0077] in, This is the posterior state vector updated at time k; Let be the prior predicted state vector at time k; The masking state scalar (value is 0 or 1) is the hardwired input of the system. For calculation Dimensional Kalman gain matrix; The initial observation vector is the fusion output of step S13; This is the Jacobian observation matrix calculated above.

[0078] Through the physical execution of this equation, when microwave high-voltage excitation leads to When, the equation degenerates into The dedicated computing kernel blocks contamination from strong electromagnetic pulses at the underlying level. The intrusion relies on a high-precision nine-dimensional dynamic model. Maintaining pure inertial extrapolation. This method of directly compiling the radio frequency physical timing into the filtering mathematical iterative equations maintains the continuity of the track.

[0079] In S33, extended Kalman filtering involves a large number of matrix multiplications and inversions. If traditional loop instructions are used, a single-objective solution would consume thousands of clock cycles. To compress the end-to-end solution latency to the hardware limit, S33 implements superscalar pipeline reconfiguration for very long instruction words.

[0080] Specifically, the dedicated computing kernel forcibly takes over the allocation of general-purpose registers through low-level hardware inline assembly. This is done during the execution of Kalman gain. With covariance matrix During an update, the kernel issues multiple instructions within a single clock cycle: it uses two load units to push floating-point numbers from memory directly into a vector register with a 128-bit width, and simultaneously uses a single-instruction multiple-data (SIMD) multiply-accumulate unit (MAC) to perform a vector dot product. Its underlying dataflow instruction logic is as follows:

[0081] in, This is an output vector register that contains the results of four independent 32-bit single-precision floating-point operations. The vector accumulator register for performing the accumulation operation; A vector register for loading the sequence of elements of the first operand matrix; A vector register for loading the sequence of elements of the second operand matrix; It is a hardware-level parallel dot multiplication operator triggered in a single clock cycle.

[0082] By employing software pipelined deployment techniques, conditional jumps and data hazards generated by traditional C language compilation are eliminated. Ultimately, each dedicated computational kernel compresses the calculation cycle of the complete matrix extrapolation for a single target to an extremely low fixed constant, outputting spatially predicted coordinates with deterministic timeliness.

[0083] like Figure 5As shown in this embodiment, it should be noted that, regarding step S4, after obtaining the spatial predicted coordinates extrapolated by the dedicated computing kernel, the system must identify effective targets with interception conditions among complex swarm targets and solve the dynamic loss problem when the servo mechanical turntable switches between multiple targets. By executing decision logic based on physical boundary coupling and work minimization, the precise conversion of computing resources into physical interception kinetic energy is achieved.

[0084] Specifically, step S4 first performs a coupled screening of target arrival time and system physical preparation time. The server digital signal processor extracts the target's predicted spatial coordinates and calculates the remaining time for the target to reach the predicted strike point. The calculation formula is as follows:

[0085] in, The time it takes for the target to reach the predicted strike point, in seconds; The three-axis components of the spatial predicted coordinates output by the dedicated computing kernel, in meters; These are the physical coordinates of the phase center of the microwave weapon's transmitting antenna, in meters. The relative radial velocity of the target along the line connecting the predicted point of impact and the transmitting antenna, expressed in meters per second.

[0086] At the same time, the system calculates its own physical preparation time in real time. The calculation formula is as follows:

[0087] in, The physical time required for the entire system to complete preparation for a single launch, in seconds; The dynamic response time required for the servo turntable to deflect to the target orientation, in seconds; The physical time required for a high-power microwave source to complete pulse energy storage and charging, expressed in seconds; This is the instruction transmission hysteresis constant for triggering pulse synchronization within the system, measured in seconds.

[0088] This calculation process solves the problem of blindly allocating firepower in traditional systems. Through comparison... and The system categorizes targets into different priority indicators based on deviations. When Much larger When the target is determined to be in a far-field monitoring state, it does not occupy the current triggering resources; when near At that time, the target is determined to be within the optimal strike window. This coupled model ensures that the limited microwave energy pulse is only delivered to physically interceptable effective tracks, significantly improving the cost-effectiveness of the weapon system.

[0089] In this embodiment, it should be noted that after determining the set of interceptable targets, in order to solve the mechanical fatigue and back EMF impact problem caused by frequent start-stop of the servo turntable when dealing with a large number of scattered swarm targets, step S4 further performs a fire strike sequence rearrangement that satisfies the minimum mechanical work.

[0090] Specifically, the system establishes a path planning model based on Hamiltonian graphs to calculate the path planning between any two objectives. and Mechanical transfer costs between :

[0091] in, To achieve the goal Switch to target The overall mechanical cost is dimensionless; and These are the kinematic weighting coefficients for azimuth and pitch angles; and These are the azimuth and elevation angles of the target in polar coordinates, in radians. It is an energy loss penalty factor; It is a function of the motor's output torque as a function of time; This represents the duration of the mechanical deflection.

[0092] The server digital signal processor solves the problem to achieve the total cost. The smallest sequence is used to generate the final fire strike sequence. This process solves the problem of reciprocating oscillations in the turntable space caused by the traditional "sorting by threat level" method, transforming discrete strike commands into a smooth mechanical scan stream, maximizing the dynamic response limit of the servo mechanism, and thus enabling the engagement of higher density targets per unit time.

[0093] like Figure 6 As shown in this embodiment, it should be noted that, regarding step S5, after the microwave interception command is issued and executed, the system must address the technical problem of how to quantitatively assess the damage effect and promptly release the hardware resources occupied by the destroyed target to prevent memory overflow. By performing hit feature extraction and hardware-level bitmap aging and recycling, real-time self-healing of system resources is achieved.

[0094] Specifically, the system first extracts the second-order kinematic mutation features of the intercepted target. A dedicated computing kernel captures the target's trajectory data in real time at the moment of impact and calculates the Doppler velocity mutation rate. :

[0095] in, The acceleration abrupt change characteristic of the target, in units of seconds; The radial velocity is measured in meters per second during the first observation period after the impact. The radial velocity is measured in meters per second during the last observation period before the impact. This represents the sampling time step of the radar equipment, measured in seconds.

[0096] Simultaneously, the system extracts the scintillation variance of the radar cross section (RCS). The calculation formula is:

[0097] in, The stability characteristics of the target reflected signal are dimensionless. This is the measured cross-sectional area of ​​the i-th sampling point; This represents the average cross-sectional area within the sliding window.

[0098] This feature construction process solves the problem of difficulty in determining the effects of "soft kill" in traditional air defense systems. By converting the physical-level velocity drop and signal flashing into quantified feature vectors, it provides crucial input for subsequent substitution into the logistic regression equation to determine the damage probability. When the damage probability exceeds the 0.85 threshold, the system immediately confirms the target's failure.

[0099] In this embodiment, it should be noted that in order to ensure that the physical memory blocks occupied by the failed target can be reclaimed with zero delay when fighting against a target with a 100-level bee swarm, step S5 implements a bitmap automatic aging and reclamation mechanism based on a hardware-level lifetime counter.

[0100] Specifically, the system hard-wires a 16-bit hardware-level lifetime counter to the header of each fixed-length data block in the on-chip static random access memory. Whenever a new target state vector is mapped into the memory pool through step S2, this counter is automatically initialized to the maximum failure threshold cycle count. In each subsequent radar scan interruption cycle, if the target does not receive a valid observation update, the background task logic of the server's digital signal processor automatically executes a decrement instruction:

[0101] in, This is an atomic-level decrement operation.

[0102] when When the value decreases to zero, or when damage assessment confirms target failure, the system immediately triggers a bitmap mask-based attack. Bitwise logical operations:

[0103] in, A bitmap variable representing the resource occupancy status of the physical memory pool; For hardware bitwise AND operation; This is the hardware bitwise NOT operator; This is the logical left shift operator; This is the physical index number of the failed target in the static partition.

[0104] This recycling logic solves the problems of uncontrollable fragmentation and poor time determinism caused by the operating system's dynamic memory management (malloc / free) under high-concurrency scheduling. By flipping pure hardware bits, the system can complete the logical release of resources within a single clock cycle, ensuring that the computing pool of multi-core parallel processing devices maintains a highly efficient, clean, and stable physical state when dealing with continuous "swarm" attacks.

[0105] After the dedicated computing kernel completes the parallel recursive extended Kalman filtering and outputs the spatial predicted coordinates of the target, the system must identify the real targets with physical interception conditions from the massive "swarm" of targets and solve the dynamic and thermodynamic loss problem when the servo mechanical turntable switches between multiple targets. Step S4 realizes the deterministic transformation of computing power scheduling instructions into microwave radio frequency hardware interception kinetic energy by implementing physical boundary coupling and Hamiltonian path optimization.

[0106] Specifically, such as Figure 5 As shown, step S41: Establish a coupled model of target arrival time and system physical preparation time.

[0107] In order to eliminate invalid far-field targets and lock the interception window, the server's digital signal processor uses extrapolated spatial prediction coordinates to calculate the time it takes for each target to arrive at the predicted strike point. The calculation formula is as follows:

[0108] in, The remaining physical time, in seconds, for the target to reach the predicted impact point; These are the three-axis components of the target space predicted coordinates output by the dedicated computing kernel, in meters; These are the absolute fixed coordinates of the phase center of the microwave weapon's transmitting antenna, in meters; The relative radial velocity component of the target along the line connecting the predicted point of impact and the transmitting antenna, expressed in meters per second.

[0109] Meanwhile, based on the servo turntable parameters and microwave source status, the system calculates in real time the physical time required for the system to complete preparation for a single launch. The calculation formula is as follows:

[0110] in, The physical preparation time for the system, in seconds; The dynamic response time required for the servo turntable to deflect to the target orientation and stabilize, in seconds; The physical time required for a high-power microwave source to complete pulse energy storage and capacitor charging, in seconds; The hardware hysteresis constant for instruction transmission that triggers pulse synchronization within the system, measured in seconds.

[0111] The server digital signal processor compares and The algebraic difference is used to rigorously screen targets. When the absolute value of the difference is less than the set hit window constant, the target is marked as a candidate to enter the optimal strike window and enters the next fire sequence rearrangement queue.

[0112] Step S42: Physical limiting constraints of bus dynamic voltage drop and array predicted thermal tolerance.

[0113] Under extreme transmission conditions, energy extraction from the microwave source can cause a sudden drop in the bus voltage, while continuous transmission can lead to thermal breakdown of the phase shifter array. The system incorporates these two dimensions of underlying physical data as forced interruption conditions into the command delivery link.

[0114] During the charging of the microwave source energy storage capacitor, the high-speed analog-to-digital converter acquires the instantaneous bus voltage at a microsecond-level sampling rate. The system calculates and predicts the voltage sag rate. When the predicted sag rate exceeds a set safety threshold, the system immediately prioritizes the energy supply to the core targets according to pre-calculated priority scores, from highest to lowest, and forcibly clamps the output current of the servo turntable motor, setting a current upper limit. The calculation formula is:

[0115] in, This is the upper limit of the forced output current of the servo turntable motor, in amperes. It is the minimum bus voltage constant required for a microwave source to normally generate pulses, expressed in volts; The instantaneous bus voltage sampled is in volts. This represents the equivalent internal resistance of the energy storage circuit in the microwave platform, expressed in ohms.

[0116] Simultaneously, the system acquires the temperature of the phase shifter array nodes. When the node temperature approaches the hardware safety temperature threshold, the single dwell time of the microwave beam is controlled. This generates an exponentially decaying limit, and its update formula is:

[0117] in, The beam dwell time after amplitude limiting adjustment, in seconds; The baseline kill calibration dwell time is expressed in seconds. It is a natural exponential function; This is the physical thermal decay control constant; This represents the real-time temperature of the array nodes, in degrees Celsius. This refers to the hardware safety temperature threshold, expressed in degrees Celsius. This represents the limit junction temperature threshold of the hardware material, expressed in degrees Celsius.

[0118] Step S43: Generate a fire strike sequence that minimizes mechanical work.

[0119] In the target set that has passed the above physical limit screening, in order to squeeze out the dynamic response limit of the servo turntable and avoid mechanical damage caused by frequent retrace, the system transforms the arrangement of the strike order into a Hamiltonian Complete Graph optimization problem in graph theory.

[0120] The server's digital signal processor calculates any two targets in the target set to be attacked. and Mechanical transfer costs between :

[0121] in, The total mechanical cost of switching from target i to target j is dimensionless; and is the kinematic weighting coefficient for azimuth and elevation angles; and are the azimuth and elevation angles of the target in polar coordinates, respectively, in radians; The time penalty weight constant; This is the maximum rated angular velocity of the servo turntable, expressed in radians per second.

[0122] Using the selected targets as nodes, and with the cost of mechanical transfer... Construct a complete graph with edges, and solve the cost function that minimizes the total mechanical work using a dynamic programming algorithm. :

[0123] in, The value of the total mechanical work done after minimization; Solving for the minimum value; The total number of targets to be attacked; This is the index of the k-th target in the traversal sequence; This is the index of the (k+1)th adjacent target. The value between the two.

[0124] The final output is the optimal traversal sequence. As the underlying firepower sequence of a high-power microwave weapon, the data is directly sent to the servo controller, which controls the microwave radio frequency hardware to perform smooth and continuous mechanical deflection interception.

[0125] Fifth Implementation Example: Detailed Implementation of Closed-Loop Performance Evaluation and Hardware-Level Ghost Target Aging and Recycling Phase (Step S5) After the microwave radio frequency hardware transmits a pulse to execute the interception action, the defense system faces a complex performance evaluation and feedback problem. At the same time, if the failed data structure under extremely high concurrency is not released in time, it will cause a static random access memory (SRAM) stack overflow and system crash. Step S5 assesses the damage probability by constructing a three-dimensional impact feature vector and uses hardware-level counters and bit operations that are independent of the operating system to perform absolutely deterministic memory reclamation.

[0126] Specifically, such as Figure 6 As shown, step S51: Extract impact features to construct a three-dimensional impact feature vector; After the target is irradiated by a high-power microwave pulse, the server's digital signal processor extracts second-order physical mutation features from the state stream continuously output by the extended Kalman filter.

[0127] The first dimension extracts the target Doppler velocity mutation rate. To measure the transient deceleration characteristics caused by the failure of an aircraft's aerodynamic layout or propulsion system:

[0128] in, Characterized by Doppler velocity mutation rate; The radial absolute velocity during the first radar observation cycle after the interception maneuver ends; The radial absolute velocity is the last valid observation before the interception action is triggered; This refers to the sampling update cycle step size for radar equipment.

[0129] The second dimension extracts the scintillation variance of the radar cross section (RCS). To measure the characteristics of the violent reversal of the echo polarization surface caused by the uncontrolled tumbling of the target:

[0130] in, The scintillation variance characteristic of radar cross section; This is the set constant for the continuous echo sampling sliding window length; This represents the actual measured value of the cross-sectional area obtained for the i-th time within the window; This is the arithmetic mean of the radar cross-section within the sliding window.

[0131] Third dimension extraction of photoelectric target miss dispersion It is the mean of the sum of squares of the miss distance error of the photoelectric tracking turntable in N consecutive video frames, which characterizes the degree of disordered random motion of the target deviating from the smooth extrapolated trajectory.

[0132] These three types of features are combined to construct a three-dimensional impact feature vector. .

[0133] Step S52: Calculate the damage probability using a nonlinear weight matrix mapping; After obtaining the three-dimensional impact feature vector, the system does not use a simple manual fixed threshold for judgment, but instead uses a nonlinear weight matrix trained from massive amounts of microwave impact test data from test ranges. .

[0134] The system inputs the feature vectors into the Logistic mapping function for weighted summation and activation, and outputs the damage probability value. :

[0135] in, The final normalized damage probability value takes a range of values. ; The exponentiation operator is the base of the natural logarithm; for The constant of the performance evaluation weight matrix for each dimension; To extract the constructed 3D impact feature vector; This is the system's decision bias constant.

[0136] When calculated When the damage exceeds a preset damage confirmation threshold (e.g., 0.85), the target is determined by the system to be completely physically failed.

[0137] Step S53: Automatic aging and reclamation of bitmap memory based on hardware-level lifetime counter; For targets deemed ineffective after being destroyed, and for radar ghost traces generated by false correlations in complex and cluttered backgrounds, these must be instantly cleared after they leave the tracking gate to prevent memory fragmentation on computing nodes. The system introduces a hardware-level lifetime counter in the underlying static memory scheduler controller of the server's digital signal processor. With bitmap mask mechanism.

[0138] In the initial stage, when the state vector of a brand new target is entered into a fixed-length data block through the hash mapping in step S2, the storage controller hardware initializes the lifetime counter associated with that block:

[0139] in, Index of the physical memory block where the target is located The current value of the corresponding hardware-level lifetime counter; The maximum failure threshold cycle number constant set for the system (i.e., the maximum number of radar scan frames that the target is allowed to survive without new observations).

[0140] Upon each radar scan interruption, the hardwired logic in the control plane performs a parallel scan of the associated gate determination states of all active memory blocks. If a target fails to receive a valid new round of observation updates (e.g., it flies out of bounds, is destroyed, or is determined to be a false alarm), its corresponding lifetime counter executes an unconditional low-level decrement instruction.

[0141] when When the value decreases to absolute zero, or when the target is determined to be invalid in step S5.2, the general management kernel immediately triggers a hardware-level atomic operation. This operation directly overwrites the bitmap mask word managing the memory allocation state by invoking assembly-level leading zero counting and bitwise logical operations. :

[0142] in, A 32-bit or 64-bit wide register variable used to identify the allocation of a global physical memory block; This refers to the assembly-level bitwise AND logical operator; This is the assembly-level bitwise NOT logical operator; This is a logical left shift operator; The index number of the target physical memory block that needs to be aged and recycled.

[0143] Through the aforementioned Boolean logic operations, the memory block allocation flag occupied by the target is cleared and released within a single machine clock cycle (ns). The address of this physical memory block immediately becomes an allocatable idle state, and will be completely overwritten by newly captured target data in subsequent probing cycles.

[0144] This automatic aging mechanism completely removes target lifecycle management from the heavy burden of operating system stack maintenance, sinking it entirely to the underlying register bit operations. From both mathematical and hardware perspectives, it eliminates the basis for memory fragmentation, giving this multi-core parallel processing device the computational power fluidity to maintain stable operation even under extreme saturation attacks.

[0145] In some embodiments, a typical combat scenario is presented: a high-power microwave weapon defense position countering a saturation attack by a swarm of hundreds of drones. During the perception synchronization phase, the defense platform first sends a synchronization request to the front-end radar via the PTP protocol and obtains the timestamp through actual measurement. for Radar reception for Radar response for The server receives for Substituting into the network path delay formula, we can accurately calculate:

[0146] When the radar is at the local original timestamp for Real-time target acquisition, combined with inherent device latency for (Right now Substituting these values ​​into the deterministic timestamp formula yields the system-wide alignment:

[0147] Immediately afterwards, the microwave source fired, and the directional coupler measured the instantaneous power of the forward transmission. for Set a reference threshold for Basic trailing time for Environmental scattering coefficient for Substituting into the dynamic multipath tailing time formula, we get:

[0148] Combined with the warning time for With pulse physical width for Substitute the values ​​into the hardware masking pulse formula to generate the accurate hardware masking pulse width:

[0149] During this 190μs high-level period, the masking state scalar Forced to be At this point, the state update equation degenerates into The system shields itself from strong electromagnetic interference at the underlying mathematical logic level, relying solely on the model for pure inertial physical extrapolation. After 190 μs... Restore to The closed-loop observation and correction were restarted instantly.

[0150] In the process of normalizing and fusing physical quantities from heterogeneous data, radar equipment measures the radial distance of a target in polar coordinates. for Doppler velocity for First azimuth for First pitch angle for Combining known information for (Right now Substituting the spatial mapping transformation matrix, the position components in the sensor coordinate system are calculated:

[0151]

[0152]

[0153] Meanwhile, the optoelectronic equipment feeds back high-frequency sub-millisecond miss distance. for , for Let the sensor rotation matrix be... For the standard identity matrix approximation, the Jacobian matrix... The specific partial derivative algebraic values ​​are respectively Substituting into the fusion correction equation, we obtain... Matrix compensation amount is The final output is an initial observation vector with uniform dimensions. :

[0154] Centimeter-level homogeneous spatial normalization of heterogeneous sensor data was achieved.

[0155] Faced with a sudden influx of large numbers of targets, the server's digital signal processor performs a non-linear topology mapping to detect the total number of currently active targets. for One, preset single-core processing power constant. for Substituting into the computing power threshold formula, we obtain the theoretical minimum number of physical computing cores:

[0156] Known number of server cores for Number of cores per client for Substituting the values ​​into the chip activation formula, the number of digital signal processor chips that the system can wake up in nanoseconds is:

[0157] With a unique identification number assigned by the system for Taking the target as an example, substituting it into the first-level hash mapping formula, we obtain that the target is assigned to the physical chip number [number missing]. Client chip:

[0158] Then, substituting the second-level kernel-level hash formula into the chip, the result is... :

[0159] At this point, for The target data packet passes through the hardware without going through any software routing queuing. Complexity is precisely pushed to The first chip The dedicated computing kernel enables physically isolated load balancing.

[0160] Within the computational kernel, a nine-dimensional state-space model is constructed for this highly maneuverable target. Let the radar sampling physical time step be... for Substitute the values ​​into the state transition matrix formula to generate a sparse matrix for target prediction:

[0161] Extrapolation yields the predicted coordinates of the target. for , for , for Fixed coordinates of microwave weapon transmitting antenna Target relative radial velocity for Substituting the values ​​into the remaining time coupling formula, the time it takes for the target to reach the predicted strike point is calculated. Approximately:

[0162] Simultaneously, the system acquires the dynamic response time of the servo turntable. for Microwave source pulse charging time for Triggering instruction synchronization delay for Substituting into the system physical preparation time formula yields... for:

[0163] because Much larger The target is temporarily placed in the secondary queue, while those... near The most dangerous targets were immediately drawn into the fire support roster.

[0164] For those selected that are in the optimal attack window The core objective is for the system to perform Hamiltonian graph path optimization that minimizes the work done by the machines. (Set the objective...) With the goal polar coordinate azimuth They are respectively and Pitch angle They are respectively and directional weight for Pitch weight for Time weight for servo maximum angular velocity for Substituting into the formula for mechanical transfer cost, we can calculate:

[0165] The system exhaustively constructs a cost matrix for all nodes and substitutes it into the dynamic programming objective function. Finally, the total cost is calculated. Minimum value is traversal sequence During interception following this sequence, the microwave source deload causes a momentary drop in bus voltage. Sampled instantaneous bus voltage Fall to Below the minimum maintenance constant of internal resistance for Substituting into the current clamping formula, the output current of the servo motor is forcibly clamped:

[0166] Meanwhile, the real-time temperature of the array nodes Temperatures rise to 85°C, reaching the safety threshold. The limiting junction temperature is 70℃. At 120℃, the thermal decay coefficient for Baseline dwell time for Substituting into the exponential thermal limiting formula, the new microwave dwell duty cycle limit is forcibly compressed to:

[0167] After the microwave energy pulse irradiation is completed, the damage closed-loop evaluation phase begins. The radar measures the radial absolute velocity of the target during the last effective cycle before impact. for Radial velocity in the first cycle after being hit sudden drop Sampling step size for Substitute the values ​​into the acceleration mutation characteristic formula to calculate:

[0168] The system captures continuous A sliding window ( )of The value is used to calculate the root mean square error and obtain the scintillation variance characteristic of the radar cross section. for Simultaneously, the dispersion of photoelectric target miss distance. Measured as The numerical combination of these three dimensions constructs a three-dimensional impact feature vector. The system retrieves the pre-trained offline performance evaluation weight matrix. Determine the bias constant for Substitute The mapping activation formula calculates the final probability:

[0169]

[0170] Due to the probability of damage Strictly greater than Upon reaching the confirmation threshold, the system determines that the target has physically failed. At this point, the physical memory block of the target (assuming an index) is... for The bound hardware-level lifetime counter is atomically decremented each time it is no longer associated. Since the system has been confirmed destroyed, it does not need to wait for the counter to return to zero naturally; it directly triggers hardware-level atomic operations, substituting the bitmap logic operation formula:

[0171] Within a single machine clock cycle, the control is forced to... The allocation bit of memory block number is flipped to... This instantly releases the static storage resources that were originally occupied, preventing system memory leaks during high-concurrency operations.

[0172] It should be noted that, in this embodiment, the preset thresholds and constant parameters involved in various embodiments of the present invention (including but not limited to the reference trigger power threshold constant Pref, the basic environmental electromagnetic tail time constant Tbase, the single-core processing capability constant Ccore, the damage confirmation threshold, and the performance evaluation weight matrix constant Weval, etc.) are not subjectively arbitrarily set, but are obtained offline through statistical distribution models and machine learning algorithms based on a limited number of historical range test data and dynamic bench experiments. The specific value selection process is as follows: Before system deployment, at least 5000 sets of system physical state data including normal interception, extreme continuous firing, and different power level excitation are collected; the dataset is smoothed by Kalman filtering and extreme value removal is performed, and then the feature value set under normal working conditions is fitted using a normal distribution function, and the boundary value corresponding to its one-sided 95% confidence interval is taken in combination with engineering safety redundancy as the preset threshold. Taking the values ​​of "efficiency assessment weight matrix constant Weval" and "damage confirmation threshold 0.85" as examples, the researchers extracted historical samples of radar Doppler velocity mutation rate, RCS scintillation variance, and miss distance after 2000 UAV targets were irradiated by microwaves. These were used as input features, and whether the UAV ultimately crashed (1 or 0) was used as the true label. The cross-entropy loss function was used to train the Logistic regression model using gradient descent. The training results showed that when a typical hit feature vector [46.67, 12.5, 8.4]T was input, the model output probability was approximately 0.963. In the entire validation set, when the model output probability was greater than 0.85, the accuracy of the UAV's actual physical failure reached 99.2%. Therefore, the weight matrix Weval in this scenario was fixed offline as [0.05, 0.1, 0.2] and the damage confirmation threshold was scientifically calibrated to 0.85. This parameter processing method, which combines historical data distribution characteristics with machine learning, ensures that all preset parameters have a rigorous physical basis and reliable implementation effect in actual complex electromagnetic warfare environments.

[0173] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-core parallel processing method for high-power microwave weapons, deployed on a heterogeneous parallel processing array comprising a server digital signal processor and multiple client digital signal processors, characterized in that, Includes the following steps: Establish a precise time protocol for synchronization, and perform dynamic temporal masking based on the instantaneous power of the forward transmission of the microwave source to generate a mask scalar. At the same time, normalize and fuse the physical quantities of multi-source heterogeneous sensor data to output an initial observation vector with a unified dimension. The server digital signal processor calculates a deterministic computing power allocation threshold based on the total number of currently active targets to activate a corresponding number of client digital signal processors, and allocates the initial observation vector to a determined dedicated computing kernel through hash mapping; The dedicated computing kernel receives the initial observation vector, combines it with the mask scalar, performs parallel recursive extended Kalman filtering with mask compensation, and extrapolates to generate the spatial prediction coordinates of the target. Based on the spatial predicted coordinates, a coupled model of target arrival time and system physical preparation time is established to screen targets. Under the physical constraints of bus dynamic voltage drop and array predicted thermal tolerance, a fire strike sequence that minimizes mechanical work is generated to control microwave radio frequency hardware to perform interception. After microwave interception, the target's hit characteristics are extracted to determine the probability of damage. Combined with a hardware-level lifetime counter, the bitmap memory of targets determined to be ineffective is automatically aged and reclaimed.

2. The multi-core parallel processing method for a high-power microwave weapon according to claim 1, characterized in that, The dynamic timing masking generated by the forward transmission instantaneous power of the microwave source to produce a mask scalar specifically includes: The dynamic multipath tail time is calculated based on the obtained instantaneous power of the microwave source in the forward direction and the set reference trigger power threshold constant. The dynamic multipath tail time is added to the radar protection warning time advance and the physical width of the microwave pulse to generate the hardware masking pulse width. During the high-level period of the hardware masking pulse width, the mask scalar is set to a state value that prohibits observation updates, and during other normal working periods, the mask scalar is set to a state value that allows observation updates.

3. The multi-core parallel processing method for a high-power microwave weapon according to claim 1, characterized in that, The calculation of the deterministic computing power allocation threshold and the allocation of the initial observation vector to a specific dedicated computing kernel through hash mapping specifically include: Calculate the theoretical minimum number of physical processing cores required based on the total number of currently active targets and the set single-core processing power constant; The total number of digital signal processor chips that need to be activated is determined by combining the theoretical minimum number of physical operation cores. The physical chip number to which the initial observation vector is mapped is determined by taking the modulus of the total number of digital signal processor chips using the target's unique identification number, and then taking the modulus again using the unique identification number within that physical chip to determine the specific dedicated computing core.

4. The multi-core parallel processing method for a high-power microwave weapon according to claim 1, characterized in that, The parallel recursive execution of extended Kalman filtering with mask compensation, extrapolating to generate the spatial prediction coordinates of the target, specifically includes: When the input mask scalar is a prohibited observation update state value, the sensor input is automatically blocked, and the spatial predicted coordinates are updated by inertial extrapolation prediction based on the physical model. When the input mask scalar is an observation-updated state value, the spatial prediction coordinates are corrected using the calculated Kalman gain matrix and the initial observation vector.

5. The multi-core parallel processing method for a high-power microwave weapon according to claim 1, characterized in that, The generation of the fire strike sequence that satisfies the minimization of mechanical work specifically includes: Calculate the mechanical transfer cost between any two targets in the target set to be attacked. The mechanical transfer cost includes the azimuth mechanical displacement cost and the pitch mechanical displacement cost based on the maximum rated angular velocity of the servo turntable. A Hamiltonian complete graph is constructed with the target as the node and the mechanical transfer cost as the edge. The cost function that minimizes the total mechanical work is solved by dynamic programming, and the corresponding optimal traversal sequence is generated as the fire strike sequence.

6. The multi-core parallel processing method for a high-power microwave weapon according to claim 1, characterized in that, The specific implementation method of the physical amplitude limiting constraint is as follows: During the charging of the microwave source energy storage capacitor, the instantaneous bus voltage is sampled in real time. When the predicted voltage drop rate exceeds the safety threshold, the energy supply of the target is guaranteed according to the calculated priority score from high to low. The output current of the servo turntable motor is forcibly clamped based on the minimum bus voltage constant and the equivalent internal resistance. The temperature of the phase shifter array nodes is obtained, and when the node temperature approaches the hardware safety temperature threshold, the single dwell time of the microwave beam is controlled to decrease exponentially.

7. The multi-core parallel processing method for a high-power microwave weapon according to claim 1, characterized in that, The process of extracting the target's hit characteristics to determine the probability of damage specifically includes: A three-dimensional impact feature vector is constructed by extracting the Doppler velocity mutation rate, radar cross-section scintillation variance, and photoelectric miss distance dispersion from the output stream of the extended Kalman filter. The three-dimensional impact feature vector is weighted and summed using a nonlinear weight matrix that has been pre-fitted and trained with experimental data, and the final damage probability value is output through a mapping function.

8. The multi-core parallel processing method for a high-power microwave weapon according to claim 1, characterized in that, The step of automatically aging up and reclaiming bitmap memory for targets determined to be invalid by combining hardware-level lifetime counters specifically includes: The hardware-level lifetime counter is appended to a fixed-length data block in the on-chip static random access memory. When the target state vector is mapped into the memory block, the hardware-level lifetime counter is initialized to the maximum failure threshold cycle number. When each radar scan interruption is triggered, if the associated gate determines that no new valid observation update has been obtained, the hardware-level lifetime counter is decremented. When the hardware-level lifetime counter decrements to zero, a hardware-level atomic operation is triggered, which directly clears the memory resource allocation status bitmap flag through bitwise logical operations on the bitmap mask word to release the physical memory block.

9. The multi-core parallel processing method for high-power microwave weapons according to claim 1, characterized in that, The heterogeneous parallel processing array is divided into a data plane and a control plane based on the physical architecture of the communication backplane. The server digital signal processor and multiple client digital signal processors perform large-scale transfer of the initial observation vector through the data plane. The control plane independently transmits inter-core synchronization interrupts, the mask scalar, bitmap reset instructions for automatic memory aging and reclamation, and the fire strike sequence.

10. A multi-core parallel processing device for a high-power microwave weapon, characterized in that, This includes server digital signal processors, client digital signal processors, direct memory access controllers, and memory; The memory stores a computer program, which, when executed by the server digital signal processor and the client digital signal processor, implements the multi-core parallel processing method for high-power microwave weapons as described in any one of claims 1 to 9.

Images