Development of a hybrid algorithm for optimizing noise-protection structures with modal decomposition

Problem statement. Modern radioelectronic equipment (REE) is subject to significant risks of malfunction due to electromagnetic interference (EMI). Conductive interference penetrating through power supply circuits and signal lines poses a particular danger, which can lead to degradation of REE characteristics or their partial (breakdown) or complete (component burnout) failures. One of the most dangerous types of conductive EMI are ultrashort pulses (USP), which have a high amplitude and short duration. To protect against them, modal filters (MF) are developed that attenuate USP by decomposing them into a sequence of pulses of lower amplitude due to different arrival times of the modes. Meanwhile, the design of MF requires solving a complex optimization problem due to multi-extreme objective functions (OF) (when optimizing a number of parameters according to several criteria). MFs are usually optimized using various classes of algorithms: evolutionary, stochastic, deterministic, etc. Each of them, in addition to advantages, has two main disadvantages: significant computational and time costs and a tendency to localize in local extremes of the OF. The purpose of the work is to develop and test a new algorithm for global optimization of noise-protection structures with modal decomposition, providing an effective balance between accuracy, time costs, resistance to local extremes and computational efficiency. Results. The developed adaptive section pruning (ASP) algorithm was tested on two test functions with different landscapes (complexity levels), as a result of which the features of its operation were revealed, and recommendations for its basic settings were formulated. When optimizing the MF based on a 3-conductor microstrip line using the ASP algorithm by the amplitude criterion, it was possible to achieve a minimum output voltage level of 146.9 mV. The obtained results were compared with the results of optimization using a genetic algorithm (GA), an evolutionary strategies, and a random search, where the obtained results differ by no more than 1%. At the same time, the time spent on a single calculation of the objective function using the ASP algorithm is 10.7% less than that of the GA. The developed algorithm maintains a balance between the accuracy of calculations, time costs, and resistance to local extremes. Practical significance. The proposed ASP algorithm expands the toolkit of engineers involved in the development and modification of protective devices.