Sequential approximate optimisation for statistical analysis and yield optimisation of circularly polarised antennas

Abstract

The study addresses a problem of low‐cost statistical analysis and yield optimisation of circularly polarised (CP) microstrip antennas. For the sake of computational efficiency, auxiliary response surface approximation models are utilised to establish a fast representation of the expensive full‐wave electromagnetic (EM)‐simulation antenna model. The surrogate permits feasible Monte Carlo (MC) analysis using a large number of random samples (necessary to maintain low variance of yield estimation). The sequential approximate optimisation is then utilised to carry out a robust design, where at each iteration, a new surrogate model is established in a relocated domain. The design objectives are set for the in‐band axial ratio and impedance bandwidth. For illustration purposes, a planar CP antenna with two feed line stubs is considered. Accurate yield estimation is obtained at the cost of only a hundred of EM antenna simulations despite a large number of independent geometry parameters. Yield improvement from the initial value of 63.5% at the nominal design to 99% is achieved at the cost of 300 EM analyses. Comparison with conventional MC analysis is also provided demonstrating the accuracy of the presented approach.

Machine Learning-assisted Antenna Design optimization: A Review and the State-of-the-art

“…For a robust design of antennas and to ensure a full design closure, statistical analysis is required for the quantification of the fabricated antenna deviations from its nominal design values. This procedure is referred to as process variation-aware or yield-driven design and it is aimed at maximizing the probability that a fabricated prototype will meet the performance specifications within the range of assumed statistical deviations from its nominal design [16], [42].…”

“…Several SBO methods employing local optimization methods and/or global optimization methods as their search engines have been proposed for the single-objective, multifidelity, multi-objective and process variation-aware or yield optimization of antennas [11], [13]- [16]. As a result, there is a variety of SBO paradigms for the machine learning-assisted optimization of antennas.…”

Rapidtolerance‐awaredesign of miniaturized microwave passives by means ofconfined‐domainsurrogates

“…However, the overall cost of setting up the surrogates within all iterations of the sequential algorithm is low. Furthermore, these algorithms can be viewed as illustrations of the two conceptual approaches, whereas the literature offers a number of well refined techniques, typically following either of these paradigms (eg, [15][16][17][18][19][20][21]31 ). In this work, the aim is to develop a technique that maintains the simplicity of the one-shot approach while constructing the surrogate at a reasonable cost.…”

“…30 However, due to a high cost of setting up the surrogate valid within broader ranges of the system parameters, otherwise necessary to conduct the optimization process, iterative methods seem to be more economical. Sequential approximate optimization (SAO) 31 is a generic concept, where the surrogate is constructed in the domain being a small vicinity of the current design, locally optimized, and relocated into the new domains defined along the optimization path. Here, the necessity of reconstructing the model is compensated for by a significantly lower cost of training data acquisition within each region considered during the process.…”

Expedited Feature-Based Quasi-Global Optimization of Multi-Band Antenna Input Characteristics With Jacobian Variability Tracking

“…Even more importantly, in real-world engineering applications, for the model to be truly useful for design purposes, it has to cover wide ranges of parameters [26], [27]. Satisfying this demand is challenging as the characteristic features of the responses, such as frequency allocation of the resonances, change rapidly across the design space [28], [29].…”