A surface acoustic wave filter frequency response correction system

Abstract

The present application relates to the field of frequency correction technology, in particular to a kind of surface acoustic wave filter frequency response correction system, system includes: node physical property analysis module, genetic algorithm strategy generation module, dynamic tuning process evaluation module, multidimensional fitness iteration optimization module.In the present application, the internal correlation between simulation analysis control voltage node and frequency response drift is calculated to construct sensitivity matrix, and based on this matrix, comprehensive weight coefficient and set gene bit allocation are calculated to form a complete optimization strategy set.In the iteration optimization process, not only the correction accuracy is pursued, but also the network input impedance in the tuning process is dynamically simulated and evaluated.The length of forbidden area trajectory is calculated to impose impedance path penalty, and the stability of tuning process is considered.Under the driving of multidimensional fitness evaluation and directional variation, a group of control voltage sequences that can guarantee path stability and have the best correction effect are output.

Description

Technical Field

[0001] This invention relates to the field of frequency correction technology, and in particular to a surface acoustic wave filter frequency response correction system. Background Technology

[0002] Frequency correction technology is an important branch of electronic information and communication engineering. Its core focus is on correcting frequency deviations caused by components during operation using circuit design and signal control methods, ensuring the stability and accuracy of the system. This technology encompasses methods for adjusting the characteristics of frequency-related devices such as resonators, filters, and oscillators, and is widely used in wireless communication, radar measurement and control, and precision measurement. Among these, traditional surface acoustic wave (SAW) filter frequency response correction systems address the frequency response deviation problems that easily occur in SAW filters due to manufacturing processes or operating environments. This is achieved by introducing a reference resonant circuit or an external control circuit to adjust the filter's response curve. Correction is typically accomplished by adding a compensation network to the electrode structure or introducing RC compensation components into the external circuitry.

[0003] Existing technologies heavily rely on fixed physical compensation structures, such as preset reference resonant circuits or external resistor-capacitor components. Once the design is finalized, this approach is difficult to change. When the filter experiences dynamic and complex frequency response shifts due to individual differences in manufacturing processes or environmental factors such as temperature changes and material aging, the fixed compensation network cannot adaptively adjust. For example, a compensation circuit designed for room temperature will significantly reduce its correction effect at high temperatures, or may even fail. Furthermore, its correction methods are mostly overall compensation, which cannot perform fine-grained and differentiated adjustments to different control nodes, nor can it assess the transient impedance mismatch problems that may occur during the adjustment process, posing a risk of instantaneously disrupting system stability. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose a surface acoustic wave filter frequency response correction system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a surface acoustic wave filter frequency response correction system includes:

[0006] The node physical characteristic analysis module applies voltage increments to the control voltage nodes, simulates and obtains the frequency response drift, calculates the unit voltage sensitivity, constructs the control node-frequency response index sensitivity matrix, and passes the control node-frequency response index sensitivity matrix to the genetic algorithm strategy generation module.

[0007] The genetic algorithm strategy generation module processes the control node-frequency response index sensitivity matrix, calculates the comprehensive weight coefficient, sets the gene position allocation and constructs the query logic, combines them into an optimization strategy set, and passes the optimization strategy set to the multidimensional fitness iterative optimization module.

[0008] The dynamic tuning process evaluation module acquires the control voltage sequence, simulates the tuning process of the control voltage sequence and records the network input impedance, calculates the trajectory length of the forbidden zone, obtains the impedance path penalty value, and passes the impedance path penalty value to the multidimensional fitness iterative optimization module.

[0009] The multidimensional fitness iterative optimization module calls the optimization strategy set to initialize the population, decodes individuals into control voltage sequences, calculates fitness scores by combining the impedance path penalty value, performs directional mutation, and outputs the control voltage sequence with the highest fitness score after convergence.

[0010] As a further embodiment of the present invention, the control node-frequency response index sensitivity matrix includes control node identifier, frequency response index dimension and sensitivity quantification value, the optimization strategy set specifically includes non-uniform gene position encoding rules, directional mutation triggering conditions and weight mapping relationship, the impedance path penalty value includes forbidden zone boundary threshold, trajectory overrun length and preset penalty coefficient, and the control voltage sequence includes node voltage amplitude, adjustment step order and sequence time length.

[0011] As a further aspect of the present invention, the node physical characteristic analysis module includes:

[0012] The frequency response drift simulation submodule identifies control voltage nodes in the circuit topology, applies a preset voltage increment to a single control voltage node, calls the circuit simulation kernel to obtain transient response data, performs a Fourier transform on the transient response data, extracts the frequency response curve, performs a difference calculation with the reference frequency response curve, and generates the frequency response drift.

[0013] The node sensitivity quantization submodule calls the frequency response drift amount and reads the voltage increment amplitude applied to the control voltage node, extracts the drift value of the frequency response drift amount at the target frequency point, and performs a division operation between the drift value and the voltage increment amplitude to obtain the unit voltage sensitivity value;

[0014] The sensitivity matrix construction submodule sets the node traversal sequence, repeats the aforementioned operation for all the control voltage nodes in the sequence, obtains the unit voltage sensitivity value of each control voltage node, constructs a two-dimensional data structure with node index as the row and frequency response index as the column, fills all the unit voltage sensitivity values ​​according to the index, and establishes the control node-frequency response index sensitivity matrix.

[0015] As a further aspect of the present invention, the genetic algorithm strategy generation module includes:

[0016] The weight coefficient calculation submodule traverses each row of the data vector of the control node-frequency response index sensitivity matrix, extracts all frequency response index sensitivity values ​​associated with a single control node, and performs a weighted summation operation on all the extracted frequency response index sensitivity values ​​according to the preset frequency response index importance setting to obtain the comprehensive weight coefficient.

[0017] The gene coding mapping submodule sorts all control nodes in descending order according to the value of the comprehensive weight coefficient, extracts the top-ranked control nodes as the target control nodes for optimization, assigns a unique binary gene bit to each extracted control node, establishes a mapping relationship between the binary gene bit and the control node identifier, and generates a gene coding lookup table.

[0018] The optimization strategy set construction submodule randomly generates a set of binary encoding strings that conform to the length of the binary gene bits based on the gene encoding lookup table. The binary encoding strings constitute the initial population, and each binary encoding string represents an optimization strategy. All the binary encoding strings in the initial population are aggregated and combined into an optimization strategy set.

[0019] As a further aspect of the present invention, the dynamic tuning process evaluation module includes:

[0020] The tuning process simulation submodule calls the circuit simulator and applies the timestamps and voltage values ​​in the control voltage sequence to the circuit network model as a time-varying excitation source. At each preset simulation time step, it monitors and records the instantaneous values ​​of the network input impedance of the circuit network model at the corresponding frequency point, and collects all the instantaneous values ​​of the network input impedance to form a time series, generating impedance dynamic trajectory data.

[0021] The restricted area trajectory recognition submodule calls the impedance dynamic trajectory data and loads the preset Smith chart restricted area boundary coordinates. It traverses each network input impedance instantaneous value in the impedance dynamic trajectory data, judges the geometric positional relationship between the complex coordinates and the restricted area boundary coordinates, filters all continuous trajectory points that geometrically fall into the restricted area, and obtains the out-of-bounds impedance trajectory sequence.

[0022] The path penalty value calculation submodule calculates the Euclidean distance between adjacent trajectory points in the sequence sequentially if the sequence is not empty, based on the out-of-bounds impedance trajectory sequence. It then sums all the calculated Euclidean distances and multiplies them by a preset penalty weighting factor to generate the impedance path penalty value.

[0023] As a further aspect of the present invention, the multidimensional fitness iterative optimization module includes:

[0024] The population initialization submodule takes the binary encoded strings in the optimization strategy set, which constitute the first generation of individuals, loads the first generation of individuals to construct the initial population, sets the upper limit of the number of iterations and the convergence judgment benchmark value, and obtains the initial state parameters of the iteration.

[0025] The individual decoding submodule, based on the iterative initial state parameters, traverses each individual in the initial population, calls the gene encoding lookup table, converts the binary gene bits of the individual into the corresponding node voltage amplitude, and combines them in a preset order to generate a control voltage sequence corresponding to the individuals in the population.

[0026] The fitness evaluation submodule calls all the control voltage sequences and, in conjunction with the impedance path penalty value, calculates the individual fitness score by substituting the impedance path penalty value into the penalty term of the fitness function according to the preset fitness function.

[0027] The iterative optimization submodule sorts the population according to the individual fitness scores, selects high-scoring individuals and performs directional mutation operations on them to generate offspring populations, and determines whether the average fitness score of the offspring populations meets the convergence judgment benchmark value. If it does not meet the benchmark value, the iteration is repeated. If it does meet the benchmark value, the individual with the highest score in the current population is selected and the corresponding control voltage sequence with the highest fitness score is output.

[0028] As a further aspect of the present invention, when the weighted summation operation is performed, the weighted coefficient calculation submodule obtains the sensitivity values ​​of all frequency response indicators associated with a single control node and the preset frequency response indicator importance values ​​corresponding to multiple frequency response indicators, and performs calculations using a preset weighted calculation formula.

[0029] The weight calculation formula is as follows:

[0030] ;

[0031] in, Representing the The comprehensive weighting coefficient of each control node. A unique identifier index for the control node. The total number of frequency response indicators. This serves as a unique identifier index for frequency response indicators. For the first The preset importance value of each frequency response indicator. The first one extracted from the control node-frequency response index sensitivity matrix The control node for the first The sensitivity quantification value of each frequency response index;

[0032] Traverse all control nodes and repeatedly perform the weighted summation operation to obtain a set of comprehensive weight coefficients corresponding to each control node.

[0033] As a further aspect of the present invention, when performing the geometric position relationship judgment, the restricted area trajectory recognition submodule parses the boundary coordinates of the Smith chart restricted area into a vertex sequence defining a closed polygon, and obtains the real and imaginary parts of the instantaneous value of the network input impedance, wherein the real and imaginary parts constitute the Cartesian coordinates of the point to be measured;

[0034] A horizontal ray is constructed from the point to be tested. The number of intersections between the ray and the sides of the closed polygon is calculated. If the number of intersections is odd, the point to be tested is determined to be inside the restricted area. If the number of intersections is even or zero, the point to be tested is determined to be outside the restricted area.

[0035] All instantaneous values ​​of continuous network input impedance that are determined to be located within the restricted area are arranged according to their original time order in the impedance dynamic trajectory data to establish an out-of-bounds impedance trajectory sequence.

[0036] As a further aspect of the present invention, the preset fitness function used by the fitness evaluation submodule is specifically calculated using the following formula:

[0037] ;

[0038] in, Individual fitness score representing a single individual. The total number of discrete target frequency points. A unique identifier index for the target frequency point. For the first Preset correction importance weights for each target frequency point After applying the current control voltage sequence, at the... The frequency response amplitude obtained from simulation at each target frequency point In the first The target amplitude of the reference frequency response curve at each target frequency point This is a preset penalty adjustment coefficient used to adjust the degree of influence of the impedance path penalty value. This is the impedance path penalty value.

[0039] As a further aspect of the present invention, when the iterative selection submodule performs the directed mutation operation, it first determines whether the improvement value of the average fitness score of the offspring population within a consecutive preset number of generations is less than a preset fitness improvement threshold. If so, the mutation operation is initiated.

[0040] Based on the individual fitness score, a roulette wheel selection method is used to select an individual to be mutated from the high-scoring individuals, and the set of comprehensive weight coefficients is called to identify the control node corresponding to the highest comprehensive weight coefficient.

[0041] In the binary encoding string of the individual to be mutated, the binary gene bit associated with the control node of the highest comprehensive weight coefficient is located, and a logical inversion operation is performed on the binary gene bit, that is, from 0 to 1 or from 1 to 0, to generate the mutated offspring individual.

[0042] The mutated offspring individuals replace the individuals with the lowest fitness scores in the original population to form a new offspring population.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0044] In this invention, the intrinsic relationship between control voltage nodes and frequency response drift is analyzed through simulation. The unit voltage sensitivity is accurately calculated to construct the Lingpei matrix. Based on this matrix, the comprehensive weight coefficient and the set gene position allocation are calculated to form a complete set of optimization strategies. In the iterative optimization process, not only is the correction accuracy pursued, but also the network input impedance in the tuning process is dynamically simulated and evaluated. By calculating the trajectory length of the forbidden zone, impedance path penalty is applied, and the stability of the tuning process is taken into consideration. Driven by multidimensional fitness evaluation and directional mutation, a set of control voltage sequences that can ensure path stability and the best correction effect are converged and output, achieving refined frequency response and adaptive correction. Attached Figure Description

[0045] Figure 1 This is a flowchart of the surface acoustic wave filter frequency response correction system of the present invention;

[0046] Figure 2 This is a flowchart of the sensitivity matrix construction process of the present invention;

[0047] Figure 3 This is a flowchart illustrating the generation process of the genetic algorithm optimization strategy set in this invention.

[0048] Figure 4 This is a flowchart illustrating the impedance path penalty value calculation process of this invention.

[0049] Figure 5 This is a flowchart of the multidimensional fitness iterative optimization process of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the software-based technical solution is described in detail below with reference to system architecture diagrams and embodiments. It should be understood that the specific embodiments described herein are only for explaining the technical solutions of this invention and do not constitute a limitation on the scope of protection.

[0051] In the description of this invention, the system architecture relationships or data processing flows indicated by terms such as "layer," "module," "interface," "data flow," "client," and "server" are all defined based on the architecture diagram or flowchart corresponding to the embodiments. This way of describing is only used to clearly illustrate the logical relationships between the elements in the technical solution, and not to limit the physical deployment form. The term "multiple" includes two or more technical units, including but not limited to multiple data nodes, processing threads, service instances, or functional components and other scalable elements. The specific number is determined according to the actual business scenario and needs to be specifically specified.

[0052] Please see Figure 1 and Figure 2 The present invention provides a technical solution: a surface acoustic wave filter frequency response correction system comprising:

[0053] The node physical characteristic analysis module applies voltage increments to the control voltage nodes, simulates and obtains the frequency response drift, calculates the unit voltage sensitivity, constructs the control node-frequency response index sensitivity matrix, and passes the control node-frequency response index sensitivity matrix to the genetic algorithm strategy generation module.

[0054] The control node-frequency response index sensitivity matrix includes control node identifiers, frequency response index dimensions, and sensitivity quantification values.

[0055] The node physical characteristic analysis module includes:

[0056] The frequency response drift simulation submodule identifies control voltage nodes in the circuit topology, applies a preset voltage increment to a single control voltage node, calls the circuit simulation kernel to obtain transient response data, performs Fourier transform on the transient response data, extracts the frequency response curve, performs difference calculation with the reference frequency response curve, and generates the frequency response drift.

[0057] The node sensitivity quantization submodule calls the frequency response drift and reads the voltage increment amplitude applied to the control voltage node. It extracts the drift value of the frequency response drift at the target frequency point, divides the drift value with the voltage increment amplitude, and obtains the unit voltage sensitivity value.

[0058] The sensitivity matrix construction submodule sets the node traversal sequence, repeats the aforementioned operation for all control voltage nodes in the sequence, obtains the unit voltage sensitivity value of each control voltage node, constructs a two-dimensional data structure with node index as the row and frequency response index as the column, fills all unit voltage sensitivity values ​​according to the index, and establishes the control node-frequency response index sensitivity matrix.

[0059] In this embodiment of the invention, a radio frequency surface acoustic wave (SAW) filter with a center frequency of 2.45 GHz is used as the analysis object. Its circuit topology includes three control nodes that can be tuned by external voltage, labeled Vc1, Vc2, and Vc3. These three nodes are connected to varactor diodes in the circuit. By changing the DC bias voltage, their capacitance values ​​are adjusted, thereby affecting the frequency response characteristics of the filter. The frequency response drift simulation submodule first identifies Vc1, Vc2, and Vc3 as control voltage nodes. Initially, the reference voltage for all control voltage nodes is set to 1.50V. Under this reference state, a sweeping sinusoidal signal covering the target frequency band is applied to the circuit network model of the SAW filter, and voltage and current response data are collected at its input and output ports. The S-parameters characterizing its transmission and reflection characteristics are calculated using the aforementioned data, and a reference frequency response curve is obtained. The reference values ​​for key frequency response indicators are: center frequency (…). The frequency response drift simulation was performed. The frequency response drift was 2.4500 GHz, the in-band average insertion loss (IL) was -2.10 dB, and the -3 dB bandwidth (BW) was 80.0 MHz. First, a preset voltage increment was applied to the control voltage node Vc1. The setting of this increment value was based on the following experimental verification process. The verification process involved simulations using voltage increments of 0.01V, 0.05V, 0.1V, and 0.2V, respectively, recording the frequency response drift and simulation time for each increment. Experimental data showed that a 0.01V increment resulted in too small a frequency response drift, making it susceptible to simulator noise; a 0.2V increment was too large, potentially crossing the nonlinear operating region of the component, leading to inaccurate sensitivity calculations; and a 0.1V increment provided a stable and significant frequency response drift within a reasonable timeframe. Therefore, the preset voltage increment was determined. The voltage of Vc1 is increased from 1.50V to 1.60V, while the voltages of Vc2 and Vc3 remain unchanged at 1.50V. The circuit simulation kernel is invoked to perform transient response analysis on the circuit in this state, obtaining the time-domain waveform data of the output port. The obtained transient response time-domain data sequence is converted to the frequency domain through Discrete Fourier Transform (DFT) to obtain the amplitude and phase information corresponding to each frequency point, resulting in new frequency response curve data. The amplitude difference between the new frequency response curve and the reference frequency response curve at each frequency point is calculated. For example, at the 2.4500GHz frequency point, the reference response amplitude is -2.10dB, and the new response amplitude after applying the voltage increment is -2.12dB, so the difference is -0.02dB. This process is repeated for all frequencies of interest to generate a set of frequency response drift data. Upon extraction, it was found that the center frequency drifted to 2.4495 GHz, with a drift of -0.0005 GHz (i.e., -500 kHz); the in-band average insertion loss changed to -2.13 dB, with a drift of -0.03 dB; and the -3 dB bandwidth changed to 80.2 MHz, with a drift of +0.2 MHz.

[0060] The above-mentioned preset voltage increment The experimental verification process involves determining a suitable preset voltage increment. A series of comparative experiments were conducted. The experimental object was a surface acoustic wave (SAW) filter circuit model of the same type. Under the same simulation environment, voltage increments of 0.01V, 0.05V, 0.1V, and 0.2V were applied to the control node Vc1, and the standard deviation of the center frequency drift (reflecting data stability) and the computation time of a single simulation were recorded. The experiment was repeated 50 times, and the average value was taken. The results are recorded in the table below.

[0061] Table 1. Experimental Data for Voltage Increment Selection

[0062]

[0063] As shown in Table 1, when the voltage increment is 0.01V, the standard deviation of the drift is relatively large, indicating that the simulation results are easily affected by numerical noise. When the voltage increment is 0.2V, the drift of "-1025.6kHz" deviates significantly from twice the drift of "-500.3kHz" (-1000.6kHz) at a 0.1V increment, indicating that it has entered the nonlinear region, the linear sensitivity assumption is no longer accurate, and the standard deviation slightly increases. When the voltage increment is 0.1V, the standard deviation is the smallest, and the simulation time does not increase significantly, indicating that it can achieve the highest computational efficiency while ensuring the stability and linearity of the results. Therefore, 0.1V is selected as the preset voltage increment.

[0064] The node sensitivity quantization submodule calls the previously generated frequency response drift. This is applied to the center frequency (…). This frequency response specification has a drift value of -500kHz. The voltage increment amplitude applied to the control voltage node Vc1 is read as 0.1V. A division operation is performed to calculate the unit voltage sensitivity of Vc1 to the center frequency: Similarly, calculate the unit voltage sensitivity of Vc1 to insertion loss: Calculate the voltage sensitivity of Vc1 per unit bandwidth:

[0065] .

[0066] The sensitivity matrix construction submodule sets the node traversal sequence to Vc1->Vc2->Vc3. For all control voltage nodes within the sequence, the aforementioned frequency response drift simulation and node sensitivity quantization operations are repeated. Specifically, Vc1 voltage is restored to 1.50V, a 0.1V increment is applied to Vc2, and its unit voltage sensitivity value is simulated and calculated; then, Vc2 voltage is restored to 1.50V, a 0.1V increment is applied to Vc3, and the same operation is performed. After traversing all nodes, the unit voltage sensitivity value of each control voltage node for each frequency response index is obtained. Finally, a two-dimensional data structure is constructed with node indices as rows (1=Vc1, 2=Vc2, 3=Vc3) and frequency response indexes as columns (fc, IL, BW), and all calculated unit voltage sensitivity values ​​are filled according to their indices to establish the control node-frequency response index sensitivity matrix.

[0067] Please see Figure 1 and Figure 3 The genetic algorithm strategy generation module processes the control node-frequency response index sensitivity matrix, calculates the comprehensive weight coefficient, sets the gene position allocation and constructs the query logic, combines them into an optimization strategy set, and passes the optimization strategy set to the multidimensional fitness iterative optimization module.

[0068] The optimization strategy set specifically includes non-uniform gene locus coding rules, directional mutation triggering conditions, and weight mapping relationships.

[0069] The genetic algorithm strategy generation module includes:

[0070] The weight coefficient calculation submodule traverses each row of the data vector in the control node-frequency response index sensitivity matrix, extracts all frequency response index sensitivity values ​​associated with a single control node, and performs a weighted summation operation on all extracted frequency response index sensitivity values ​​according to the preset frequency response index importance setting to obtain the comprehensive weight coefficient.

[0071] When performing weighted summation, the weight coefficient calculation submodule obtains the sensitivity values ​​of all frequency response indicators associated with a single control node and the preset importance values ​​of multiple frequency response indicators, and performs calculations using a preset weight calculation formula.

[0072] The formula for calculating the weight is:

[0073] ;

[0074] in, Representing the The comprehensive weighting coefficient of each control node. A unique identifier index for the control node. The total number of frequency response indicators. This serves as a unique identifier index for frequency response indicators. For the first The preset importance value of each frequency response indicator. The first one extracted from the control node-frequency response index sensitivity matrix The control node for the first The sensitivity quantification value of each frequency response index;

[0075] Traverse all control nodes and repeatedly perform the weighted summation operation to obtain a set of comprehensive weight coefficients corresponding to each control node;

[0076] The gene coding mapping submodule sorts all control nodes in descending order based on the value of the comprehensive weight coefficient, extracts the top-ranked control nodes as the target control nodes for optimization, assigns a unique binary gene bit to each extracted control node, establishes a mapping relationship between the binary gene bit and the control node identifier, and generates a gene coding lookup table.

[0077] The optimization strategy set construction submodule is based on the gene encoding lookup table. It randomly generates a set of binary encoding strings that conform to the length of binary gene bits. The binary encoding strings constitute the initial population. Each binary encoding string represents an optimization strategy. All binary encoding strings in the initial population are aggregated and combined to form an optimization strategy set.

[0078] This module receives the control node-frequency response index sensitivity matrix from the node physical characteristic analysis module and calculates the comprehensive weighting coefficients accordingly. The weighting coefficient calculation submodule first performs a weighted summation operation on the data in the sensitivity matrix based on the preset importance settings for the frequency response indexes. The importance values ​​of the frequency response indexes ( The settings were based on the stability requirements and failure mode analysis report of various performance indicators for this type of surface acoustic wave filter in practical applications. A 500-hour high-temperature and high-humidity accelerated aging test was conducted on 1000 devices from the same batch, and their performance degradation data were statistically analyzed. The experimental results showed that 62% of device failures were attributed to center frequency drift exceeding the specification range, 28% to insertion loss deterioration, and 10% to bandwidth changes. Based on this statistical data, the importance value of the frequency response indicator was set to be positively correlated with the failure rate. The center frequency (…) was set… Importance value Importance of insertion loss (IL) values Importance of bandwidth (BW) .

[0079] The accelerated aging test used to set the importance values ​​for the above frequency response indicators was conducted under the following conditions: ambient temperature 85℃, relative humidity 85%, duration 500 hours, and 1000 test samples. A complete S-parameter test was performed on all samples every 50 hours, recording the center frequency, insertion loss, and bandwidth. The failure criterion was that any indicator exceeded the range defined in the product specification (e.g., center frequency drift exceeding ±2MHz). After the experiment, the indicators that caused the first failure were statistically analyzed. The results showed that among all samples that ultimately failed, 620 failed due to the center frequency exceeding the tolerance, 280 due to the insertion loss exceeding the tolerance, and 100 due to the bandwidth exceeding the tolerance. Based on this, the importance weights of each indicator were assigned. This is set as the proportion of the number of failed samples to the total number of failed samples, i.e.

[0080] , , Then, normalization is performed to ensure that the weight sum is 1.0.

[0081] Subsequently, the weight coefficient calculation submodule iterates through each row of the data vector in the sensitivity matrix obtained using the aforementioned method. The weight calculation formula is as follows: .in, Representing the The comprehensive weighting coefficient of each control node. This is a unique identifier index for the control node, with values ​​ranging from 1 to 3. This represents the total number of frequency response indicators, with a value of 3. This is a unique identifier index for the frequency response index, with values ​​ranging from 1 to 3. For the first The preset importance values ​​for each frequency response indicator, which apply to the center frequency, insertion loss, and bandwidth. The first one extracted from the control node-frequency response index sensitivity matrix The control node for the first The sensitivity quantification value of each frequency response index. The specific calculation process is as follows:

[0082] For control node Vc1 ( ):

[0083] ;

[0084] For control node Vc2 ( ):

[0085] ;

[0086] For control node Vc3 ( ):

[0087] ;

[0088] By traversing all control nodes and repeatedly performing the weighted summation operation, the set of comprehensive weight coefficients corresponding to each control node is obtained as follows: {Vc1:3000.29,Vc2:4800.395,Vc3:900.29}.

[0089] The gene coding mapping submodule sorts all control nodes in descending order based on the comprehensive weight coefficients {3000.29, 4800.395, 900.29}, resulting in the sorting order: Vc2 > Vc1 > Vc3. The top-ranked control nodes are then selected as the target control nodes for optimization. The selection rule is to choose nodes with weight coefficients greater than the average of all weight coefficients. The nodes are defined as follows: Vc2 and Vc1 are selected as the target control nodes for optimization. The voltage regulation range of each node is set to 0.5V to 2.5V, with a regulation accuracy of 0.1V, resulting in 21 discrete voltage levels. For binary encoding, the power of 2 closest to and greater than 21 is selected, i.e., 32 (…). Therefore, each captured control node is assigned 5 unique binary gene bits. A mapping relationship between binary gene bits and control node identifiers is established: Vc2 corresponds to gene bits 0 to 4, and Vc1 corresponds to gene bits 5 to 9. At the same time, a gene code lookup table is generated, which defines the correspondence between 5-bit binary codes (from 00000 to 10100) and specific voltage values ​​(from 0.5V to 2.5V).

[0090] The optimization strategy set construction submodule generates a set of binary encoding strings that conform to the binary gene bit length (10 bits) based on the gene encoding lookup table. For example, it randomly generates encoding strings such as 1001001101, 0011110000, and 1010000011. Each binary encoding string represents an optimization strategy (i.e., an individual). With an initial population size of 50, 50 such binary encoding strings are generated. All 50 binary encoding strings in the initial population are aggregated to form an optimization strategy set. This strategy set includes the non-uniform gene bit encoding rules (Vc2 and Vc1 are encoded, Vc3 is discarded), the weight mapping relationship that subsequent directed mutation operations will rely on (Vc2 has the highest weight), and is passed to the multidimensional fitness iterative optimization module as the initial population.

[0091] Please see Figure 1 and Figure 4 The dynamic tuning process evaluation module acquires the control voltage sequence, simulates the tuning process of the control voltage sequence and records the network input impedance, calculates the trajectory length of the forbidden zone, obtains the impedance path penalty value, and passes the impedance path penalty value to the multidimensional fitness iterative optimization module.

[0092] The impedance path penalty value includes the restricted area boundary threshold, the trajectory over-limit length, and the preset penalty coefficient;

[0093] The dynamic tuning process evaluation module includes:

[0094] The tuning process simulation submodule calls the circuit simulator and applies the timestamps and voltage values ​​within the control voltage sequence to the circuit network model as a time-varying excitation source. At each preset simulation time step, it monitors and records the instantaneous values ​​of the network input impedance of the circuit network model at the corresponding frequency point, and collects all the instantaneous values ​​of the network input impedance to form a time series, generating impedance dynamic trajectory data.

[0095] The restricted area trajectory recognition submodule calls the impedance dynamic trajectory data and loads the preset Smith chart restricted area boundary coordinates. It traverses each network input impedance instantaneous value in the impedance dynamic trajectory data, judges the geometric positional relationship between the complex coordinates and the restricted area boundary coordinates, filters all continuous trajectory points that geometrically fall into the restricted area, and obtains the out-of-bounds impedance trajectory sequence.

[0096] When performing geometric position relationship judgment, the restricted area trajectory recognition submodule resolves the boundary coordinates of the restricted area on the Smith chart into a vertex sequence that defines a closed polygon, and obtains the real and imaginary parts of the instantaneous value of the network input impedance. The real and imaginary parts constitute the Cartesian coordinates of the point to be measured.

[0097] Construct a horizontal ray from the point to be tested, and calculate the number of intersections between the ray and the sides of the closed polygon. If the number of intersections is odd, the point to be tested is determined to be inside the restricted area. If the number of intersections is even or zero, the point to be tested is determined to be outside the restricted area.

[0098] All the instantaneous values ​​of continuous network input impedance that are determined to be located within the restricted area are arranged according to their original time order in the impedance dynamic trajectory data to establish an out-of-bounds impedance trajectory sequence.

[0099] The path penalty value calculation submodule calculates the Euclidean distance between adjacent trajectory points in the sequence in turn if the sequence is not empty, based on the out-of-bounds impedance trajectory sequence. It then sums up all the calculated Euclidean distances and multiplies them by a preset penalty weight factor to generate the impedance path penalty value.

[0100] This module receives a control voltage sequence to be evaluated, generated by the individual decoding submodule of the multidimensional fitness iterative optimization module. Taking a decoded control voltage sequence as an example: the goal is to adjust the voltage of Vc2 from an initial 1.5V to 2.0V, and the voltage of Vc1 from an initial 1.5V to 1.0V. The adjustment process duration is set to 10 microseconds (µs). The tuning process simulation submodule calls the circuit simulator and applies this control voltage sequence as a time-varying excitation source to the circuit network model. Specifically, two piecewise linear (PWL) voltage sources are set in the simulator, corresponding to Vc1 and Vc2 respectively. The voltage value of Vc1 is 1.5V at 0µs and linearly decreases to 1.0V at 10µs. The voltage value of Vc2 is 1.5V at 0µs and linearly increases to 2.0V at 10µs. The simulation time step is set to 0.1µs, and a total of 101 time points are simulated. At each time step, the instantaneous value Z(t) of the network input impedance of the circuit network model at the center frequency of 2.45 GHz is monitored and recorded. For example, at t=0µs, Z(0)=50+j0Ω; at t=0.1µs, Vc1 is approximately 1.495V, Vc2 is approximately 1.505V, and the simulation yields Z(0.1)=51.2+j1.5Ω; ...; at t=10µs, Z(10)=48.5-j5.3Ω. All 101 instantaneous values ​​of the network input impedance are collected to form a time series, which is the impedance dynamic trajectory data. The restricted area trajectory identification submodule calls this impedance dynamic trajectory data. At the same time, the preset Smith chart restricted area boundary coordinates are loaded. This restricted area is set according to the stability requirements of the power amplifier matched with the filter. In this embodiment, it is set to the area on the Smith chart where the voltage standing wave ratio (VSWR) is greater than 3.0. In a standardized Smith chart, this region refers to the area outside a circle centered at the origin (0,0) with a radius of 0.5. Its boundary can be defined by a sequence of vertex coordinates. Then, each instantaneous network input impedance value in the impedance dynamic trajectory data is traversed. First, each impedance value is normalized; for example, if the system characteristic impedance is 50Ω, the instantaneous impedance value is divided by 50. For example, the instantaneous impedance Z(5.0) = 20 - j40Ω at t = 5.0µs is normalized to z(5.0) = 0.4 - j0.8. This complex coordinate is then converted to Cartesian coordinates (x, y) within the Smith chart, where x = 0.154 and y = -0.462. A geometric positional relationship determination is performed, i.e., determining whether the point is inside the restricted area. The distance from the point to the center of the circle is calculated.

[0101] Since the distance of 0.487 is less than the restricted zone boundary radius of 0.5, the point is located within the safe zone and has not entered the restricted zone. This judgment is performed on all points in the trajectory. All trajectory points that continuously fall into the restricted zone are filtered out. For example, all 18 impedance data points from t=4.5µs to t=6.2µs are located within the restricted zone. These 18 consecutive network input impedance instantaneous values ​​that are judged to be located within the restricted zone are arranged according to their original time order to establish an out-of-bounds impedance trajectory sequence. The path penalty value calculation submodule calculates the Euclidean distance between adjacent trajectory points in the sequence according to this non-empty out-of-bounds impedance trajectory sequence. For example, the distance between the normalized impedance points z(4.5) and z(4.6) corresponding to the two times t=4.5µs and t=4.6µs in the Cartesian coordinate system is calculated. Assuming the coordinates of z(4.5) are (0.18, -0.45) and the coordinates of z(4.6) are (0.17, -0.46), then the distance between the two points is...

[0102] .

[0103] The total trajectory out-of-bounds length is obtained by summing the Euclidean distances between all 17 adjacent points. Assume the summation result is... Then, this length is multiplied by a preset penalty weighting factor. Finally, the impedance path penalty value is generated. The impedance path penalty value will be passed to the multidimensional fitness iterative optimization module.

[0104] The experimental verification process for the aforementioned penalty weighting factor involved calibrating the penalty weighting factor and constructing a test system. This system included a signal source, a programmable impedance tuner, the filter under test (DUT), and a power amplifier. By controlling the impedance tuner, the filter was simulated to exhibit out-of-range trajectories of varying lengths during tuning, and the adjacent channel power ratio (ACPR) of the power amplifier's output signal was measured. The experiment set an ACPR degradation exceeding 3 dB as an unacceptable performance threshold.

[0105] Table 2 Relationship between the length of the boundary crossing trajectory and the deterioration of ACPR

[0106]

[0107] As shown in Table 2, when the length of the out-of-bounds trajectory reaches 0.20, the ACPR degradation value begins to exceed 3 dB. To impose a strong penalty on this unacceptable behavior in the fitness function, a weighting factor needs to be set such that when... When the value approaches 0.20, the penalty term's value has a comparable or stronger influence than the main frequency response error term's value. After analyzing multiple sets of typical frequency response error values, a penalty weighting factor of 200 was set. At this point, when... Penalty value This value is sufficient to significantly reduce an individual's score in fitness calculations, thereby guiding the optimization process to avoid such dangerous tuning paths.

[0108] Please see Figure 1 and Figure 5 The multidimensional fitness iterative optimization module calls the optimization strategy set to initialize the population, decodes individuals into control voltage sequences, combines the impedance path penalty value, calculates the fitness score and performs directional mutation, and outputs the control voltage sequence with the highest fitness score after convergence.

[0109] The control voltage sequence includes node voltage magnitude, adjustment step sequence, and sequence time length;

[0110] The multidimensional fitness iterative optimization module includes:

[0111] The population initialization submodule uses binary encoded strings from the optimization strategy set to form the first generation of individuals, loads the first generation of individuals to build the initial population, sets the upper limit of the number of iterations and the convergence judgment benchmark value, and obtains the initial state parameters of the iteration.

[0112] The individual decoding submodule, based on the iterative initial state parameters, traverses each individual in the initial population, calls the gene encoding lookup table, converts the individual's binary gene bits into the corresponding node voltage amplitude, and combines them in a preset order to generate control voltage sequences corresponding to multiple population individuals.

[0113] The fitness evaluation submodule calls all control voltage sequences and, in conjunction with the impedance path penalty value, calculates the individual fitness score by substituting the impedance path penalty value into the penalty term of the fitness function according to the preset fitness function.

[0114] The fitness evaluation submodule uses a preset fitness function, the specific calculation formula of which is as follows:

[0115] ;

[0116] in, Individual fitness score representing a single individual. The total number of discrete target frequency points. A unique identifier index for the target frequency point. For the first Preset correction importance weights for each target frequency point After applying the current control voltage sequence, at the... The frequency response amplitude obtained from simulation at each target frequency point In the first The target amplitude of the reference frequency response curve at each target frequency point For the first The preset normalization tolerance corresponding to each target frequency point is used to standardize the frequency response index errors of different dimensions. This is a preset penalty adjustment coefficient used to adjust the degree of influence of the impedance path penalty value. It applies to the impedance path penalty value. This is the impedance path penalty value;

[0117] The iterative optimization submodule sorts the population based on individual fitness scores, selects high-scoring individuals and performs directed mutation operations on them to generate offspring populations, and determines whether the average fitness score of the offspring population meets the convergence judgment benchmark. If it does not meet the benchmark, the iteration is repeated. If it does meet the benchmark, the individual with the highest score in the current population is selected and the corresponding control voltage sequence with the highest fitness score is output.

[0118] When performing the directional mutation operation, the iterative optimization submodule first determines whether the improvement value of the average fitness score of the offspring population within a consecutive preset number of generations is less than a preset fitness improvement threshold. If so, the mutation operation is initiated.

[0119] Based on the individual fitness score, a roulette wheel selection method is used to select an individual to be mutated from the high-scoring individuals, and the set of comprehensive weight coefficients is called to identify the control node corresponding to the highest comprehensive weight coefficient.

[0120] In the binary encoding string of the individual to be mutated, locate the binary gene bit associated with the control node of the highest comprehensive weight coefficient, perform a logical inversion operation on the binary gene bit, that is, change it from 0 to 1 or change it from 1 to 0, and generate the mutated offspring individual.

[0121] The mutated offspring replace the individual with the lowest fitness score in the original population to form a new offspring population.

[0122] This module first loads the optimization strategy set (50 10-bit binary encoded strings) from the aforementioned genetic algorithm strategy generation module into the population initialization submodule, which serves as the initial population for the first generation. Simultaneously, the operating parameters of the genetic algorithm are set: the maximum number of iterations is set to 100 generations, and the convergence criterion is set to "the average fitness score growth rate of the population over 10 consecutive generations is less than 0.05%".

[0123] The aforementioned convergence criterion and fitness enhancement threshold were set with reference to determining the upper limit of the number of iterations, the convergence criterion, and the fitness enhancement threshold (set to 0.0005) used in subsequent directed mutations. The optimization algorithm of this invention was run on three typical frequency response correction tasks with different complexities as benchmark problems. A relatively high iteration limit (e.g., 500 generations) was set for each run, and the average fitness of each generation was recorded. Analysis results showed that for all benchmark problems, the average fitness of the population reached a stable plateau between generations 80 and 120, with extremely limited subsequent improvement. Therefore, setting the upper limit of the number of iterations to 100 generations was a balance between computational cost and optimization effect. Further analysis of the stable plateau data revealed that in the 10 generations before entering the plateau, the growth rate of the average fitness score was generally below 0.05%. Therefore, this value was used as the convergence criterion. The fitness enhancement threshold of 0.0005 was set to trigger the sensitivity adjustment of mutations when convergence stalled. This value is slightly larger than the numerical noise in the simulation calculation, effectively identifying true optimization stall rather than data fluctuations.

[0124] The individual decoding submodule iterates through each individual in the initial population. Taking individual 1001001101 as an example, it calls the gene coding lookup table for decoding. The first four bits of this individual are 10010, corresponding to the voltage Vc2. Looking up the table, it represents the 18th voltage level, i.e., 2.2V. The fifth to ninth bits of this individual are 01101, corresponding to the voltage Vc1. Looking up the table, it represents the 13th voltage level, i.e., 1.7V. Therefore, the control voltage sequence generated after decoding this individual is {Vc2: 2.2V, Vc1: 1.7V}. This process is repeated for all 50 individuals in the population, generating 50 sets of corresponding control voltage sequences.

[0125] The fitness evaluation submodule calls upon these 50 sets of control voltage sequences and, in conjunction with the impedance path penalty value calculated by the dynamic tuning process evaluation module, calculates the fitness score for each individual based on a preset fitness function. The fitness function formula is: Example: Taking the above sequence {Vc2:2.2V, Vc1:1.7V} as an example. First, the frequency response after applying this voltage is obtained through simulation: =2.4508GHz, =-2.14dB, =79.5MHz. The benchmark target value is: =2.4500GHz, =-2.10dB, =80.0MHz. Meanwhile, assume the impedance path penalty value obtained from the dynamic tuning process of this sequence is... Preset correction importance weights. , , Preset penalty adjustment coefficient The preset normalization tolerance is set according to the allowable deviation range of each indicator in the product specification, such as the center frequency tolerance. For 2MHz (i.e., 0.002GHz), insertion loss tolerance 0.5dB, with tolerance It is 5MHz.

[0126] Calculate the weighted frequency response error term:

[0127] ;

[0128] Calculate the penalty item: ;

[0129] Calculate individual fitness score: The result indicates that the individual's overall evaluation value is 0.1135.

[0130] The fitness evaluation submodule mentioned above uses a preset normalization tolerance. The purpose of determining its value is to convert frequency response index errors of different physical units and different numerical ranges (such as frequency error in GHz, loss error in dB, and bandwidth error in MHz) into dimensionless values ​​of a unified scale, so that they are comparable in the fitness function. The value of is usually determined by referring to the maximum allowable deviation or tolerance range of the k-th frequency response index in the filter product's design specifications or production testing standards. This normalization process can more evenly reflect the deviation of each frequency response index, avoiding the dominance of the entire fitness score by the excessively large or small value range of individual indices.

[0131] The iterative optimization submodule sorts the population in descending order based on the fitness scores calculated from all 50 individuals. A roulette wheel selection method is used to select parent individuals from high-scoring individuals to generate the offspring population. During the iteration process, it is determined whether the growth rate of the average fitness score of the offspring population over 10 consecutive generations is less than a preset fitness stagnation threshold (e.g., 0.01%). If so, a targeted mutation operation is initiated. Based on the comprehensive weight coefficient set {Vc1:3000.29,Vc2:4800.395,Vc3:900.29} calculated in the second step, the control node with the highest comprehensive weight coefficient is identified as Vc2. In the binary encoding string of the individual to be mutated, the gene position (0 to 4 positions) associated with Vc2 is located, and one position is randomly selected to perform a logical inversion operation. For example, if the gene corresponding to Vc2 in the individual to be mutated is 10010, the second position (counting from the left) is randomly selected and inverted, becoming 11010. The mutated offspring replaces the individual with the lowest fitness score in the original population, forming a new offspring population. This iterative process of fitness evaluation, selection, and mutation is repeated until the upper limit of the number of iterations or the convergence criterion is met. Finally, the individual with the highest score in the current population is selected, and its corresponding control voltage sequence {Vc2:2.1V, Vc1:1.8V} is output as the optimal frequency response correction scheme.

[0132] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the scope of protection defined by the claims of the present invention.

Claims

1. A surface acoustic wave filter frequency response correction system, characterized in that, The system includes: The node physical characteristic analysis module applies voltage increments to the control voltage nodes, simulates and obtains the frequency response drift, calculates the unit voltage sensitivity, constructs the control node-frequency response index sensitivity matrix, and passes the control node-frequency response index sensitivity matrix to the genetic algorithm strategy generation module. The genetic algorithm strategy generation module processes the control node-frequency response index sensitivity matrix, calculates the comprehensive weight coefficient, sets the gene position allocation and constructs the query logic, combines them into an optimization strategy set, and passes the optimization strategy set to the multidimensional fitness iterative optimization module. The dynamic tuning process evaluation module acquires the control voltage sequence, simulates the tuning process of the control voltage sequence and records the network input impedance, calculates the trajectory length of the forbidden zone, obtains the impedance path penalty value, and passes the impedance path penalty value to the multidimensional fitness iterative optimization module. The dynamic tuning process evaluation module includes: The tuning process simulation submodule calls the circuit simulator and applies the timestamps and voltage values ​​in the control voltage sequence to the circuit network model as a time-varying excitation source. At each preset simulation time step, it monitors and records the instantaneous values ​​of the network input impedance of the circuit network model at the corresponding frequency point, and collects all the instantaneous values ​​of the network input impedance to form a time series, generating impedance dynamic trajectory data. The restricted area trajectory recognition submodule calls the impedance dynamic trajectory data and loads the preset Smith chart restricted area boundary coordinates. It traverses each network input impedance instantaneous value in the impedance dynamic trajectory data, judges the geometric positional relationship between the complex coordinates and the restricted area boundary coordinates, filters all continuous trajectory points that geometrically fall into the restricted area, and obtains the out-of-bounds impedance trajectory sequence. The path penalty value calculation submodule calculates the Euclidean distance between adjacent trajectory points in the sequence in turn if the sequence is not empty, based on the out-of-bounds impedance trajectory sequence. It then sums up all the calculated Euclidean distances and multiplies them by a preset penalty weighting factor to generate the impedance path penalty value. The multidimensional fitness iterative optimization module calls the optimization strategy set to initialize the population, decodes individuals into control voltage sequences, calculates fitness scores by combining the impedance path penalty value, and performs directional mutation. After convergence, it outputs the control voltage sequence with the highest fitness score. The control node-frequency response index sensitivity matrix includes control node identifier, frequency response index dimension and sensitivity quantification value. The optimization strategy set specifically includes non-uniform gene position encoding rules, directional mutation triggering conditions and weight mapping relationship. The impedance path penalty value includes forbidden zone boundary threshold, trajectory overrun length and preset penalty coefficient. The control voltage sequence includes node voltage amplitude, adjustment step order and sequence time length.

2. The surface acoustic wave filter frequency response correction system according to claim 1, characterized in that, The node physical characteristic analysis module includes: The frequency response drift simulation submodule identifies control voltage nodes in the circuit topology, applies a preset voltage increment to a single control voltage node, calls the circuit simulation kernel to obtain transient response data, performs a Fourier transform on the transient response data, extracts the frequency response curve, performs a difference calculation with the reference frequency response curve, and generates the frequency response drift. The node sensitivity quantization submodule calls the frequency response drift amount and reads the voltage increment amplitude applied to the control voltage node, extracts the drift value of the frequency response drift amount at the target frequency point, and performs a division operation between the drift value and the voltage increment amplitude to obtain the unit voltage sensitivity value; The sensitivity matrix construction submodule sets the node traversal sequence, repeats the aforementioned operation for all the control voltage nodes in the sequence, obtains the unit voltage sensitivity value of each control voltage node, constructs a two-dimensional data structure with node index as the row and frequency response index as the column, fills all the unit voltage sensitivity values ​​according to the index, and establishes the control node-frequency response index sensitivity matrix.

3. The surface acoustic wave filter frequency response correction system according to claim 2, characterized in that, The genetic algorithm strategy generation module includes: The weight coefficient calculation submodule traverses each row of the data vector of the control node-frequency response index sensitivity matrix, extracts all frequency response index sensitivity values ​​associated with a single control node, and performs a weighted summation operation on all the extracted frequency response index sensitivity values ​​according to the preset frequency response index importance setting to obtain the comprehensive weight coefficient. The gene coding mapping submodule sorts all control nodes in descending order according to the value of the comprehensive weight coefficient, extracts the top-ranked control nodes as the target control nodes for optimization, assigns a unique binary gene bit to each extracted control node, establishes a mapping relationship between the binary gene bit and the control node identifier, and generates a gene coding lookup table. The optimization strategy set construction submodule randomly generates a set of binary encoding strings that conform to the length of the binary gene bits based on the gene encoding lookup table. The binary encoding strings constitute the initial population, and each binary encoding string represents an optimization strategy. All the binary encoding strings in the initial population are aggregated and combined into an optimization strategy set.

4. The surface acoustic wave filter frequency response correction system according to claim 3, characterized in that, The multidimensional fitness iterative optimization module includes: The population initialization submodule takes the binary encoded strings in the optimization strategy set, which constitute the first generation of individuals, loads the first generation of individuals to construct the initial population, sets the upper limit of the number of iterations and the convergence judgment benchmark value, and obtains the initial state parameters of the iteration. Based on the iterative initial state parameters, the individual decoding submodule traverses each individual in the initial population, calls the gene encoding lookup table, converts the binary gene bits of the individual into the corresponding node voltage amplitude, and combines them in a preset order to generate a control voltage sequence corresponding to the individuals in the population. The fitness evaluation submodule calls all the control voltage sequences and, in conjunction with the impedance path penalty value, calculates the individual fitness score by substituting the impedance path penalty value into the penalty term of the fitness function according to the preset fitness function. The iterative optimization submodule sorts the population according to the individual fitness scores, selects high-scoring individuals and performs directional mutation operations on them to generate offspring populations, and determines whether the average fitness score of the offspring populations meets the convergence judgment benchmark value. If it does not meet the benchmark value, the iteration is repeated. If it does meet the benchmark value, the individual with the highest score in the current population is selected and the corresponding control voltage sequence with the highest fitness score is output.

5. The surface acoustic wave filter frequency response correction system according to claim 4, characterized in that, When performing the weighted summation operation, the weight coefficient calculation submodule obtains the sensitivity values ​​of all frequency response indicators associated with a single control node and the preset frequency response indicator importance values ​​corresponding to multiple frequency response indicators, and performs calculations using a preset weight calculation formula. The weight calculation formula is as follows: ； in, Representing the The comprehensive weighting coefficient of each control node. A unique identifier index for the control node. The total number of frequency response indicators. This serves as a unique identifier index for frequency response indicators. For the first The preset importance value of each frequency response indicator. The first one extracted from the control node-frequency response index sensitivity matrix The control node for the first The sensitivity quantification value of each frequency response index; Traverse all control nodes and repeatedly perform the weighted summation operation to obtain a set of comprehensive weight coefficients corresponding to each control node.

6. The surface acoustic wave filter frequency response correction system according to claim 1, characterized in that, When performing the geometric position relationship judgment, the restricted area trajectory recognition submodule resolves the boundary coordinates of the Smith chart restricted area into a vertex sequence defining a closed polygon, and obtains the real and imaginary parts of the instantaneous value of the network input impedance. The real and imaginary parts constitute the Cartesian coordinates of the point to be measured. A horizontal ray is constructed from the point to be tested. The number of intersections between the ray and the sides of the closed polygon is calculated. If the number of intersections is odd, the point to be tested is determined to be inside the restricted area. If the number of intersections is even or zero, the point to be tested is determined to be outside the restricted area. All instantaneous values ​​of continuous network input impedance that are determined to be located within the restricted area are arranged according to their original time order in the impedance dynamic trajectory data to establish an out-of-bounds impedance trajectory sequence.

7. The surface acoustic wave filter frequency response correction system according to claim 4, characterized in that, The fitness evaluation submodule uses a preset fitness function, the specific calculation formula of which is as follows: ； in, Individual fitness score representing a single individual. The total number of discrete target frequency points. A unique identifier index for the target frequency point. For the first Preset correction importance weights for each target frequency point After applying the current control voltage sequence, at the... The frequency response amplitude obtained from simulation at each target frequency point In the first The target amplitude of the reference frequency response curve at each target frequency point This is a preset penalty adjustment coefficient used to adjust the degree of influence of the impedance path penalty value. This is the impedance path penalty value.

8. The surface acoustic wave filter frequency response correction system according to any one of claims 4-5, characterized in that, When performing the directional mutation operation, the iterative optimization submodule first determines whether the improvement value of the average fitness score of the offspring population within a consecutive preset number of generations is less than a preset fitness improvement threshold. If so, the mutation operation is initiated. Based on the individual fitness score, a roulette wheel selection method is used to select an individual to be mutated from the high-scoring individuals, and the set of comprehensive weight coefficients is called to identify the control node corresponding to the highest comprehensive weight coefficient. In the binary encoding string of the individual to be mutated, the binary gene bit associated with the control node of the highest comprehensive weight coefficient is located, and a logical inversion operation is performed on the binary gene bit, that is, from 0 to 1 or from 1 to 0, to generate the mutated offspring individual. The mutated offspring individuals replace the individuals with the lowest fitness scores in the original population to form a new offspring population.

Images