An Adaptive Reflectionless Filter Design Method Based on Hybrid Genetic Particle Swarm Optimization

Abstract

This invention discloses an adaptive reflection-free filter design method based on hybrid genetic particle swarm optimization, comprising: setting the target filter as an initial topology; dynamically adjusting the topology and parameters of each branch of the initial topology using a hybrid genetic particle swarm optimization algorithm to minimize the loss value, forming a final topology; designing the circuit layout based on the final topology, and fine-tuning the parasitic effects in the circuit layout to form the final circuit layout as the final design scheme of the target filter. This invention can automatically select a suitable topology based on the required filter response and the available component range, and generate the value of each component in the topology; by applying planar topology constraints to the topology, the synthesized topology can be planar, eliminating the need for vias and long connectors on the signal path when designing the actual filter, thus reducing the impact of parasitic parameters of the filter layout on the reflection-free performance.

Description

Technical Field

[0001] This invention relates to electronic information technology, and more specifically to an adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization. Background Technology

[0002] A reflectionless filter is a novel type of filter, typically a lossy network, that dissipates stopband signals within the filter itself, maintaining matched port impedance over a wide frequency range. It avoids the problem of traditional filters reflecting out-of-band signals back to the port, affecting the normal operation of other devices connected to the port. In circuit design, a circuit diagram can be represented by a topology diagram. A topology diagram consists of nodes and branches. A branch connects two nodes, and a node can connect to multiple branches. Each branch requires defining a current direction, but for passive reciprocal circuits, the current direction does not affect the calculation results. Each branch can represent an element that can be described using a 2-port network, such as a capacitor, inductor, resistor, or transmission line in a lumped element configuration. When using admittance matrices for calculations, branches connected in parallel between two nodes can be combined into one branch. An incidence matrix can be used to describe the circuit's topology, and its corresponding branch admittance matrix can contain the admittance information of each branch in the circuit, thus describing the parameters of the passive circuit. By combining the correlation matrix and the branch admittance matrix, the nodal admittance matrix of the entire circuit can be calculated. Then, the nodes corresponding to the internal ports are eliminated to obtain the admittance matrix corresponding to the remaining external ports. The impedance matrix is ​​then obtained by inverting the admittance matrix, and then converted into the S-parameter matrix to realize the solution calculation of the circuit response.

[0003] However, existing topology-based design schemes for reflectionless filters typically only address a specific topology, lacking universality. Under certain design requirements (such as large relative bandwidth), existing methods may result in circuits containing unfeasible components, requiring multiple components to be combined, leading to reduced performance of the fabricated filter. Some existing methods produce non-planar structures, introducing vias and long interconnects in the signal paths, resulting in additional layout parasitics. Furthermore, these filters may require multilayer boards, increasing manufacturing costs. Existing methods do not consider the parasitic characteristics of lumped capacitance and inductance, causing significant discrepancies between the actual response of the fabricated filter and the simulation results from the design phase. Summary of the Invention

[0004] In order to at least overcome the above-mentioned shortcomings in the prior art, the purpose of this application is to provide an adaptive non-reflective filter design method based on hybrid genetic particle swarm optimization.

[0005] This application provides an adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization, including:

[0006] The target filter is set as the initial topology graph, and the loss function is set according to the constraints of the target filter; the value of the loss function is the loss value of the circuit expressed by the topology and parameters of each branch in the initial topology graph.

[0007] The topology of the initial topology and the parameters of each branch are dynamically adjusted using a hybrid genetic particle swarm optimization algorithm until the loss value is minimized, thus forming the final topology.

[0008] The circuit layout is designed based on the final topology diagram, and the parasitic effects in the circuit layout are fine-tuned to form the final circuit layout as the final design scheme of the target filter.

[0009] In one possible implementation, the final topology is formed by dynamically adjusting the topology and parameters of each branch of the initial topology graph to minimize the loss value using a hybrid genetic particle swarm optimization algorithm, including:

[0010] The initial topology graph is initialized in the solution space by initializing control variables and initial particle parameters to form an initial solution space, and the optimal solution region is set in the entire initial solution space;

[0011] Based on the optimal solution region, the movement results of the particle under different movement strategies are iteratively analyzed to form the optimal position of the particle;

[0012] The optimal solution region is updated based on the optimal position of the particle, the particles with negative lifetimes are initialized, and the newly generated particles are pre-iterated.

[0013] The process of repeatedly iterating through the optimal positions of particles in the initial solution space until a preset condition is met results in the formation of the final topology.

[0014] In one possible implementation, initializing the control variables and initial particle parameters in the solution space to form an initialized solution space includes:

[0015] Set the local optimal positions of all particles to the initial positions;

[0016] The local optimum position with the minimum loss value is taken as the global optimum position.

[0017] In one possible implementation, setting the optimal solution region within the entire initialization solution space includes:

[0018] Set the upper bound vector and lower bound vector And and The range between these two points is defined as the range of the optimal solution region; the upper bound vector... Each dimension is the upper bound of the corresponding dimension of the initialized solution space; the lower bound vector Each dimension is a lower bound of the corresponding dimension of the initialized solution space.

[0019] In one possible implementation, iterating the movement results of the particle under different movement strategies based on the optimal solution region to form the optimal position of the particle includes:

[0020] The particle is moved from its current position toward the current global optimal position by a step value to reach the first position; the step value is a step value based on a first weight.

[0021] After reaching the first position, the particle is moved from the first position to the local optimal position by a second step value until it reaches the second position; the second step value is a step value based on the second weight.

[0022] The particle's current position is superimposed with a random value based on the third weight and the optimal solution range to reach the third position;

[0023] The position with the smallest loss value among the positions generated by the particle in the random solution space is selected as the fourth position; the random solution space is a randomly selected dimension of the solution space, and multiple values ​​are randomly selected in the selected dimension, while the values ​​of the unselected dimensions remain unchanged;

[0024] The position with the smallest loss value among the first position, the second position, the third position, and the fourth position is taken as the new current position;

[0025] If the loss value of the particle's new current position is less than the loss value of the particle's local optimal position, then the new current position is taken as the new local optimal position.

[0026] If the loss value of the particle's local optimal position is less than the loss value of the global optimal position, then the local optimal position is taken as the new global optimal position.

[0027] Repeat the first position, the second position, the third position, and the fourth position, and update the corresponding current position, local optimal position, and global optimal position until the preset requirements are met.

[0028] In one possible implementation, the calculation of the first position, the second position, the third position, and the fourth position is performed according to the following formula:

[0029]

[0030]

[0031]

[0032]

[0033]

[0034] In the formula, First position, w g First weight, The optimal position globally. Current position For the second position, w l As the second weight, For the third position, w r As the third weight, It is the fourth position. Let D be the space of random solutions. This is the loss function.

[0035] In one possible implementation, updating the optimal solution region based on the optimal position of the particle includes:

[0036] Sort all particles according to their current position loss value, and select the particles with the smallest loss value as the update reference particles.

[0037] The minimum value of the particle position in each dimension is selected from the updated reference particles, and a new lower bound vector is formed by reducing a preset constant.

[0038] The maximum value of the particle position in each dimension is selected from the updated reference particles, and a preset constant is added to form a new upper bound vector.

[0039] In one possible implementation, initializing particles with negative lifetimes includes:

[0040] The result of sorting by the loss value of the particle's current position is used as the fitness of the corresponding particle, and the lifetime of the corresponding particle is reduced according to the fitness.

[0041] When the lifetime of any particle is reduced to a negative value, the particle is reinitialized in the initialization solution space.

[0042] When re-initializing particles in the initial solution space, a portion of the particles are initialized in the optimal solution region according to a preset probability, and another portion of the particles are initialized in the entire initial solution space; the number of particles initialized in the optimal solution region is greater than the number of particles initialized in the entire initial solution space.

[0043] In one possible implementation, pre-iteration of the newly generated particles includes:

[0044] Based on the optimal solution region, the movement results of the newly generated particles under different movement strategies are iterated multiple times, and the current position, global optimal position, and local optimal position of the newly generated particles are updated; the number of pre-iterations of the newly generated particles is logarithmically related to the number of iterations of the optimal position of the particles in the initial solution space.

[0045] In one possible implementation, fine-tuning the parasitic effects in the circuit layout to form the final circuit layout includes:

[0046] Electromagnetic simulation is used to obtain the parasitic parameters of the circuit layout as the first parasitic parameter, and the parasitic parameters of the lumped elements are obtained as the second parasitic parameter.

[0047] Modeling is performed based on the second parasitic parameter, with the discrete parasitic parameter difference being a continuous function;

[0048] The continuous function is optimized using gradient descent.

[0049] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0050] This invention presents an adaptive reflection-free filter design method based on hybrid genetic particle swarm optimization. It automatically selects a suitable topology based on the required filter response and the available component range, and generates the value for each component in the topology. By imposing planar topology constraints on the topology, the synthesized topology can be planar. In the design of the actual filter, vias and long connectors are unnecessary in the signal path, reducing the impact of parasitic parameters of the filter layout on the reflection-free performance. By modeling the parasitic parameters of lumped components and combining them with the designed filter layout, the actual filter response can be predicted with high accuracy, enabling further optimization of component values. Attached Figure Description

[0051] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0052] Figure 1 This is a schematic diagram of the method steps in an embodiment of this application;

[0053] Figure 2 This is the node topology diagram for Embodiment 6 of this application;

[0054] Figure 3 This is a topological diagram of 13 nodes in a quasi-planar topological graph structure according to an embodiment of this application.

[0055] Figure 4 This is a schematic diagram of the parasitic parameter model of the lumped element in an embodiment of this application. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0057] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0058] Based on the above, please refer to the following: Figure 1 This is a flowchart illustrating the adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization provided in this embodiment of the invention. Further, the adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization may specifically include the contents described in steps S1-S5.

[0059] S1: Set the target filter as the initial topology graph, and set the loss function according to the constraints of the target filter; the value of the loss function is the loss value of the circuit expressed by the topology and parameters of each branch in the initial topology graph;

[0060] S2: The topology of the initial topology and the parameters of each branch are dynamically adjusted to minimize the loss value using a hybrid genetic particle swarm optimization algorithm to form the final topology.

[0061] S3: Design the circuit layout based on the final topology diagram, and fine-tune the parasitic effects in the circuit layout to form the final circuit layout as the final design scheme of the target filter.

[0062] In the implementation of this application, the synthesis method requires constraints (constraints on the target response, the range of available component values, and the topology). The basic principle of the synthesis method is to start from an initial topology graph, utilize the proposed branch model, and achieve dynamic adjustment of the topology structure through numerical optimization to obtain the value of each component in a suitable topology. The numerical optimization process is implemented by the proposed improved hybrid genetic particle swarm optimization algorithm (HGAPSO algorithm). First, the controlled parameters and particle parameters are initialized, constraints are set, and the topology graph is set as the initial graph. Then, the proposed HGAPSO algorithm is used to optimize the topology graph under the given constraints. If the result does not meet the constraints after exceeding the iteration limit, it indicates that the current topology graph cannot cover the filter structure that can achieve the target response under the constraints. The topology graph needs to be expanded, and the HGAPSO algorithm needs to be used again for optimization. If the generated result meets the requirements, the circuit layout can be designed based on the output result. When considering the influence of parasitic parameters of the layout and actual components, the calculated non-reflective performance of the filter will decrease. In this case, a gradient-based method can be used for fine-tuning. The final optimized result can then be used to fabricate an actual filter.

[0063] For example, the specific scheme for the initial topology and constraint settings is as follows:

[0064] Please see Figure 2 First, set several nodes in the current topology as ports and grounding points, and use the remaining nodes as voltage reference nodes, current reference nodes, or other reference nodes that can be used as nodes in the topology. Figure 2 Figure 'a' shows the node arrangement of a complete graph with 6 nodes, where node 1 is set as port 1 and node 2 is set as port 2. Figure 2 In step b, the dynamic topology adjustment function of this method is implemented by general branches, that is, each branch in the topology graph consists of an inductor, a capacitor, and a resistor. Therefore, each branch can be fitted to the following three cases:

[0065] (1) An open branch: Lij→∞, Cij→0, and Rij→∞

[0066] (2) A short-circuit branch: Lij→0, Cij→∞, or Rij→0

[0067] (3) A normal branch: cases other than (1) and (2)

[0068] To facilitate optimization, and considering that the distribution of actually available component values ​​follows an exponential function, the following function is introduced in this design to map the numerical values:

[0069]

[0070] Where, x i p(x) is the independent variable before mapping. i ) represents the mapped element value, a and b are the segmentation boundaries, m1 and m2 are the out-of-range mapping coefficients (where m1 needs to be set to a positive number close to 0, and m2 needs to be set to a sufficiently large positive number), and k i and s i The mapping coefficients related to the component range can be calculated using the following formula:

[0071]

[0072] Among them, V i,min and V i,max These are the lower and upper bounds of the i-th element, respectively. Using p(x) i The function can map an independent variable with a range of [0,1] to the component value in different states, thereby enabling dynamic adjustment of the component through numerical optimization.

[0073] See another example. Figure 3 This example illustrates the case of a planar structure with via grounding constraints. Specifically, by setting the topology to a planar structure, it is guaranteed that the generated result will be a planar structure, meaning all components can be placed on the same side of the PCB without the need for vias and long traces for connection. When a planar structure is required for the synthesis result, both the initial and extended topologies can be set to planar topologies. Furthermore, for PCBs with metal layers on both sides, vias can be used to connect the ground plane, thus allowing the use of the proposed quasi-planar structure to fully utilize the metal layers on the back of the PCB. This example has 13 nodes, where node 1 is port 1, node 2 is port 2, and node 13 is a ground node. This topology first arranges all nodes except the ground node in a grid and connects them to each other. Finally, all nodes are connected to the ground node by a separate branch.

[0074] Before optimizing the topology using the HGAPSO algorithm, the designed filter response needs to be constrained, and the loss function f needs to be set accordingly. The loss function f constructed in this invention sets a defined upper and lower bound for each frequency point of each S-parameter. For frequency points in the response that exceed the defined limits, the difference between the response value and the boundary value is used as a weighted sum. Therefore, if a response curve falls within the defined range, the loss function f will be 0. Otherwise, the loss function f will be the total distance between the response curve and the defined boundary.

[0075] In this embodiment, after constructing the objective function, the circuit represented by the topology and parameters of each branch can be evaluated, and the corresponding loss value, i.e., the value of the loss function f, can be obtained. The goal of the HGAPSO algorithm is to find suitable parameters to minimize the loss value. Simultaneously, due to the introduction of the general branch model, the circuit topology is dynamically adjusted as the parameters are adjusted. This algorithm combines genetic algorithms and particle swarm optimization, and embeds random neighborhood search and random dimension search to enhance metaheuristic capabilities. Furthermore, a classifier is introduced to construct the optimal solution region to improve convergence speed. In this implementation, the optimal solution region is set as a hyperrectangle with the same dimensions as the solution space. In each iteration, the hyperrectangle is updated to cover a subset of the best particles.

[0076] In one possible implementation, the final topology is formed by dynamically adjusting the topology and parameters of each branch of the initial topology graph to minimize the loss value using a hybrid genetic particle swarm optimization algorithm, including:

[0077] The initial topology graph is initialized in the solution space by initializing control variables and initial particle parameters to form an initial solution space, and the optimal solution region is set in the entire initial solution space;

[0078] Based on the optimal solution region, the movement results of the particle under different movement strategies are iteratively analyzed to form the optimal position of the particle;

[0079] The optimal solution region is updated based on the optimal position of the particle, the particles with negative lifetimes are initialized, and the newly generated particles are pre-iterated.

[0080] The process of repeatedly iterating through the optimal positions of particles in the initial solution space until a preset condition is met results in the formation of the final topology.

[0081] In one possible implementation, initializing the control variables and initial particle parameters in the solution space to form an initialized solution space includes:

[0082] Set the local optimal positions of all particles to the initial positions;

[0083] The local optimum position with the minimum loss value is taken as the global optimum position.

[0084] In one possible implementation, setting the optimal solution region within the entire initialization solution space includes:

[0085] Set the upper bound vector and lower bound vector And and The range between these two points is defined as the range of the optimal solution region; the upper bound vector... Each dimension is the upper bound of the corresponding dimension of the initialized solution space; the lower bound vector Each dimension is a lower bound of the corresponding dimension of the initialized solution space.

[0086] In the implementation of this application embodiment, during initialization, the control variables and initial parameters of all particles are initialized, and the optimal solution region is set in the entire solution space. The range of the optimal solution region is represented by two vectors, namely... and Each dimension of the vector represents the upper and lower bounds of each dimension of the knowledge space. Furthermore, the local optimal positions of all particles are set to their initial positions, and the global optimal position is set to the local optimal position that minimizes the loss value.

[0087] For example, it is constructed as follows: and

[0088]

[0089] Particle position Generated by random values ​​between the upper and lower bound vectors:

[0090]

[0091] Local optimal position Set in the initial position:

[0092]

[0093] The local optimum position with the minimum loss value is taken as the global optimum position.

[0094]

[0095] In one possible implementation, iterating the movement results of the particle under different movement strategies based on the optimal solution region to form the optimal position of the particle includes:

[0096] The particle is moved from its current position toward the current global optimal position by a step value to reach the first position; the step value is a step value based on a first weight.

[0097] After reaching the first position, the particle is moved from the first position to the local optimal position by a second step value until it reaches the second position; the second step value is a step value based on the second weight.

[0098] The particle's current position is superimposed with a random value based on the third weight and the optimal solution range to reach the third position;

[0099] The position with the smallest loss value among the positions generated by the particle in the random solution space is selected as the fourth position; the random solution space is a randomly selected dimension of the solution space, and multiple values ​​are randomly selected in the selected dimension, while the values ​​of the unselected dimensions remain unchanged;

[0100] The position with the smallest loss value among the first position, the second position, the third position, and the fourth position is taken as the new current position;

[0101] If the loss value of the particle's new current position is less than the loss value of the particle's local optimal position, then the new current position is taken as the new local optimal position.

[0102] If the loss value of the particle's local optimal position is less than the loss value of the global optimal position, then the local optimal position is taken as the new global optimal position.

[0103] Repeat the first position, the second position, the third position, and the fourth position, and update the corresponding current position, local optimal position, and global optimal position until the preset requirements are met.

[0104] In one possible implementation, the calculation of the first position, the second position, the third position, and the fourth position is performed according to the following formula:

[0105]

[0106]

[0107]

[0108]

[0109]

[0110] In the formula, First position, w g First weight, The optimal position globally. Current position For the second position, w l As the second weight, For the third position, w r As the third weight, It is the fourth position. Let D be the space of random solutions. This is the loss function.

[0111] In the implementation of this application embodiment, each particle will move according to a certain movement strategy, and the optimal position obtained from several strategies will be selected as the result of this iteration. In this application embodiment, four movement strategies are used for particle movement, among which... To approach the current global optimum based on a weight w g The step value, In order to be in Based on the local optimum of the current particle, it approaches a weighted solution based on w. l The step value. For overlay based on the current position and weight w r The random value is determined by the corresponding dimension of the optimal solution region, and the component in each dimension is determined by the corresponding dimension of the optimal solution region. Therefore, as the algorithm converges, the range of the optimal solution region will shrink accordingly after particle aggregation. The time-stacked perturbations are also reduced, thus enabling a more accurate search. For selection The location with the smallest loss value among the generated locations. Defined as randomly selecting one dimension from all D dimensions of the solution space, and... The values ​​of other dimensions remain unchanged, while several values ​​are randomly selected in this dimension to generate a vector. In the calculation... to After calculating the loss value, the position with the minimum loss value will be updated to... If the loss value at a given position is less than the local optimum loss value for that particle, then that position will be updated to the local optimum. After processing each particle, if it is found that the loss value at a particle's local optimum position is less than the loss value at the global optimum position, then that local optimum position will be updated to the global optimum position.

[0112] In one possible implementation, updating the optimal solution region based on the optimal position of the particle includes:

[0113] Sort all particles according to their current position loss value, and select the particles with the smallest loss value as the update reference particles.

[0114] The minimum value of the particle position in each dimension is selected from the updated reference particles, and a new lower bound vector is formed by reducing a preset constant.

[0115] The maximum value of the particle position in each dimension is selected from the updated reference particles, and a preset constant is added to form a new upper bound vector.

[0116] In the implementation of this application embodiment, after completing one round of particle iteration, the optimal solution region will be updated. This step first requires sorting the loss values ​​corresponding to the current positions of all particles in ascending order, and then selecting the top K particles from the sorted results. The minimum value of all particle positions in each dimension is selected for generation. At the same time, the maximum value of each dimension is selected for generation. To avoid the solution space being too small and easily converging to a local optimum, Each dimension will be reduced by a constant. Each dimension will be incremented by a constant, which is defined in The sorting is based on the following formula:

[0117]

[0118] {SV,SI} represents the sorting results. Obtained using the following formula:

[0119]

[0120]

[0121] In the formula, K is the number of selected particles. A vector composed of constants.

[0122] In one possible implementation, initializing particles with negative lifetimes includes:

[0123] The result of sorting by the loss value of the particle's current position is used as the fitness of the corresponding particle, and the lifetime of the corresponding particle is reduced according to the fitness.

[0124] When the lifetime of any particle is reduced to a negative value, the particle is reinitialized in the initialization solution space.

[0125] When re-initializing particles in the initial solution space, a portion of the particles are initialized in the optimal solution region according to a preset probability, and another portion of the particles are initialized in the entire initial solution space; the number of particles initialized in the optimal solution region is greater than the number of particles initialized in the entire initial solution space.

[0126] In the implementation of this application, after updating the optimal solution region, all particles are sorted according to their loss values. The sorting result is used as the particle's fitness to reduce its lifetime. The further down the sorting result is, the more its lifetime value will decrease. When a particle's lifetime decreases to a negative value, the particle will be reinitialized. This mechanism avoids particles from excessively searching regions with high loss values ​​in the solution space, improving the efficiency of iteration. When reinitializing particles, most particles (following a set probability p) will be initialized in the optimal solution region. Therefore, these particles can be considered a variant of the optimal particles. Simultaneously, allowing a small number of particles to be initialized throughout the entire solution space allows these particles to search regions far from the particle swarm, reducing the probability of the algorithm converging to a local minimum.

[0127] In one possible implementation, pre-iteration of the newly generated particles includes:

[0128] Based on the optimal solution region, the movement results of the newly generated particles under different movement strategies are iterated multiple times, and the current position, global optimal position, and local optimal position of the newly generated particles are updated; the number of pre-iterations of the newly generated particles is logarithmically related to the number of iterations of the optimal position of the particles in the initial solution space.

[0129] In the implementation of this application embodiment, to prevent newly generated particles from having a higher loss value than particles that have already undergone several iterations and to avoid these particles being completely unable to catch up with existing particles, a pre-optimization mechanism is introduced in this method, namely, pre-itererating the newly generated particles W times. The process described in sub-process 1 is reused in the pre-iteration. The value of W increases with the iteration loop count (in this implementation, the value of W is the logarithm of the iteration count, IterationCount). During the pre-optimization process, there may be situations where the loss value of a local optimum is lower than the loss value of the current global optimum, in which case the global optimum needs to be updated. After completing the above process, one iteration is completed, and the iteration counter is incremented by 1. When the loss value of the global optimum reaches a specified value, or the iteration counter reaches a set value, the algorithm's iteration ends, and the global optimum is returned as the result.

[0130] In one possible implementation, fine-tuning the parasitic effects in the circuit layout to form the final circuit layout includes:

[0131] Electromagnetic simulation is used to obtain the parasitic parameters of the circuit layout as the first parasitic parameter, and the parasitic parameters of the lumped elements are obtained as the second parasitic parameter.

[0132] Modeling is performed based on the second parasitic parameter, with the discrete parasitic parameter difference being a continuous function;

[0133] The continuous function is optimized using gradient descent.

[0134] In the implementation of this application embodiment, after the numerical optimization synthesis process is completed, the filter circuit can be designed based on the topology diagram and the returned parameter values. After selecting the dielectric substrate, component package, and interface, the resulting layout can be obtained. Electromagnetic simulation can be used to obtain the parasitic parameters of the layout. For the parasitic parameters of lumped components, the device datasheet of the selected manufacturer can be consulted. After considering the parasitic characteristics of the layout and components, the filter's response characteristics will deteriorate, thus requiring further optimization. Since the distribution of the actual usable values ​​of lumped components is discrete, this method models the parasitic parameters of lumped inductors and capacitors and uses interpolation to convert discrete data points into continuous functions. In this application embodiment, when adjusting the second parasitic parameter, it can be adjusted together with the first parasitic parameter until the filter exhibits the set response characteristics, ensuring that the actual filter response is consistent with expectations.

[0135] Please see Figure 4 The diagram illustrates the parasitic parameter models of lumped elements. The parasitic parameters of an inductor can be modeled as an ideal inductor connected in series with a metal resistor, and then in parallel with a packaged capacitor. The parasitic parameters of a capacitor can be modeled as an ideal capacitor connected in series with a metal resistor and a high-frequency parasitic inductance. After converting the discrete component parasitic parameters obtained from the datasheet into continuous functions using interpolation methods, gradient descent can be used for further optimization. During this process, the same loss function constructed during synthesis can be used to ensure that the final result matches the expected result.

[0136] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0137] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices or units, or may be electrical, mechanical or other forms of connection.

[0138] The units described as separate components may or may not be physically separate. As will be apparent to those skilled in the art, the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0139] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0140] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or grid device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0141] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization, characterized in that, include: The target filter is set as the initial topology graph, and the loss function is set according to the constraints of the target filter; the value of the loss function is the loss value of the circuit expressed by the topology and parameters of each branch in the initial topology graph. The topology of the initial topology and the parameters of each branch are dynamically adjusted using a hybrid genetic particle swarm optimization algorithm until the loss value is minimized, thus forming the final topology. The circuit layout is designed based on the final topology diagram, and the parasitic effects in the circuit layout are fine-tuned to form the final circuit layout as the final design scheme of the target filter. The final topology is formed by dynamically adjusting the topology and parameters of each branch of the initial topology graph using a hybrid genetic particle swarm optimization algorithm to minimize the loss value. The initial topology graph is initialized in the solution space by initializing control variables and initial particle parameters to form an initial solution space, and the optimal solution region is set in the entire initial solution space; Based on the optimal solution region, the movement results of the particle under different movement strategies are iteratively analyzed to form the optimal position of the particle; The optimal solution region is updated based on the optimal position of the particle, the particles with negative lifetimes are initialized, and the newly generated particles are pre-iterated. The process of repeatedly iterating through the optimal positions of particles in the initial solution space until those positions satisfy a preset condition forms the final topological graph. Setting the optimal solution region within the entire initialization solution space includes: Set an upper bound vector. and lower bound vector and will and The range between these two points is defined as the range of the optimal solution region; the upper bound vector... Each dimension is the upper bound of the corresponding dimension of the initialized solution space; the lower bound vector Each dimension is a lower bound of the corresponding dimension of the initialized solution space; Based on the optimal solution region, iteratively determining the particle's optimal position using different movement strategies for the particle's movement results includes: The particle is moved from its current position toward the current global optimal position by a step value to reach the first position; the step value is a step value based on the first weight. After reaching the first position, the particle is moved from the first position to the local optimal position by a second step value until it reaches the second position; the second step value is a step value based on the second weight. The particle's current position is superimposed with a random value based on the third weight and the optimal solution range to reach the third position; The position with the smallest loss value among the positions generated by the particle in the random solution space is selected as the fourth position; the random solution space is a randomly selected dimension of the solution space, and multiple values ​​are randomly selected in the selected dimension, while the values ​​of the unselected dimensions remain unchanged; The position with the smallest loss value among the first position, the second position, the third position, and the fourth position is taken as the new current position; If the loss value of the particle's new current position is less than the loss value of the particle's local optimal position, then the new current position is taken as the new local optimal position. If the loss value of the particle's local optimal position is less than the loss value of the global optimal position, then the local optimal position is taken as the new global optimal position. Repeat the first position, the second position, the third position, and the fourth position, and update the corresponding current position, local optimal position, and global optimal position until the preset requirements are met.

2. The adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization according to claim 1, characterized in that, The initialization of the initial topology graph in the solution space, including the initialization of control variables and initial particle parameters, forms the initial solution space. Set the local optimal positions of all particles to the initial positions; The local optimum position with the minimum loss value is taken as the global optimum position.

3. The adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization according to claim 1, characterized in that, The calculation of the first position, the second position, the third position, and the fourth position is performed according to the following formula: In the formula, As the first position, As the first weight, The optimal position globally. For the local optimal position, Current position For the second position, As the second weight, It is the third position. As the third weight, It is the fourth position. For a random solution space, D Let be the dimension of the random solution space. This is the loss function.

4. The adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization according to claim 1, characterized in that, Updating the optimal solution region based on the optimal position of the particle includes: Sort all particles according to their current position loss value, and select the particles with the smallest loss value as the update reference particles. The minimum value of the particle position in each dimension is selected from the updated reference particles, and a new lower bound vector is formed by reducing a preset constant. The maximum value of the particle position in each dimension is selected from the updated reference particles, and a preset constant is added to form a new upper bound vector.

5. The adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization according to claim 4, characterized in that, Initialization for particles with negative lifetimes includes: The result of sorting by the loss value of the particle's current position is used as the fitness of the corresponding particle, and the lifetime of the corresponding particle is reduced according to the fitness. When the lifetime of any particle is reduced to a negative value, the particle is reinitialized in the initialization solution space. When re-initializing particles in the initial solution space, a portion of the particles are initialized in the optimal solution region according to a preset probability, and another portion of the particles are initialized in the entire initial solution space; the number of particles initialized in the optimal solution region is greater than the number of particles initialized in the entire initial solution space.

6. The adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization according to claim 1, characterized in that, Pre-iteration of newly generated particles includes: Based on the optimal solution region, the movement results of the newly generated particles under different movement strategies are iterated multiple times, and the current position, global optimal position, and local optimal position of the newly generated particles are updated; the number of pre-iterations of the newly generated particles is logarithmically related to the number of iterations of the optimal position of the particles in the initial solution space.

7. The adaptive reflectionless filter design method based on hybrid genetic particle swarm optimization according to claim 1, characterized in that, Fine-tuning the parasitic effects in the circuit layout to form the final circuit layout includes: Electromagnetic simulation is used to obtain the parasitic parameters of the circuit layout as the first parasitic parameter, and the parasitic parameters of the lumped elements are obtained as the second parasitic parameter. Modeling is performed based on the second parasitic parameter, where the discrete parasitic parameter difference is a continuous function; The continuous function is optimized using gradient descent.

Images