Abstract

Radar cross section (RCS) is the electromagnetic signature of an object and it corresponds to the effective area observable by radar. The frequency, polarization, and incident angle of the impinged radar signal as well as the geometry and composition of the object, are the main factors that affect the RCS. Reducing RCS is a significant subject in stealth technology due to its vast applications in civil and military areas. Nowadays, with the proliferation of detection systems, developing RCS reduction techniques is getting assiduous attention.

The main goal of this research is to develop fast and efficient engineered surfaces design for RCS reduction purposes using artificial intelligence (AI). Specifically, designing engineered surface
within a broad bandwidth and wide oblique incidence stability with the minimum possible RCS in all directions. Toward this end, we implement a practical approach that is based on the concept of destructive interference in the far-field. The general idea of this approach is designing a surface that reflects the impinged wave in-phase and counter-phase concurrently which subsequently cancel out
the reflected waves in the boresight direction. The design procedure includes two steps. The first step is designing at least two reflection type unit-cells with almost 180◦ reflection phase difference. The second step is arranging these unit-cells in a format to achieve field cancellation while satisfying periodic boundary conditions of the unit-cells. In this thesis, we present some approaches both in unit-cell design and arrangement step to develop efficient, low RCS engineered surfaces addressing our objectives.

In the first study, we propose a reflection type linear polarization rotator (PR) unit-cell with a polarization conversion ratio (PCR) above 90% and a very wide bandwidth of around 115%. The proposed PR unit-cell and its mirror features 180◦ reflection phase difference within its PCR
bandwidth. Predefined numbers of the same unit-cells are considered as a super-cell, and binary digits of 0 and 1 are assigned to PR super-cell and its 90◦ rotated one. Using the array theory and binary
codes, an optimization model is developed. Bi-level group search optimization (GSO) algorithm is implemented and solved for the minimum RCS in all directions. Our simulation and measurement
results demonstrate 115% bandwidth for 10 dB normalized RCS reduction.

In the second study, we focus on improving the incident angle stability and RCS reduction of the engineered surface. In this regard, we propose a new type of artificial magnetic conductor (AMC) unit-cell, which is a combination of the conventional metallic patch and perforated dielectric slab. This model provides efficient control of effective permittivity along with metallic patch geometry. The proposed AMC unit-cells (namely as 0 and 1 types) satisfy 180◦ reflection phase
difference characteristics with a wide bandwidth for the oblique incidence up to 60◦. To improve scattering properties of the engineered surface, this time arbitrary numbers of unit-cells are considered for super-cells and optimized along the type selection according to the objective of maximum RCS reduction in all directions. The new arrangement contributes to greater RCS reduction and a more uniform pattern compared to the previous arrangement applied in the first design. Moreover, the monostatic and bistatic RCS reduction are measured as a function of frequency and compared with the simulated ones. Overall, we could achieve a reasonable agreement between simulated and measured RCS reduction results.

In the two aforementioned designs, we implemented optimization algorithms for the fast and efficient arrangement of super-cells. However, the unit-cells are designed through trial and error
process using full-wave (FW) simulators. In our third study, we extend AI techniques in the first design step of engineered surface, i.e., unit-cell design, as well. We develop a methodology for unit-cell design based on state-of-the-art machine learning (ML) algorithms. Toward this purpose, a PR unit-cell model consisting of several rectangular patch segments is presented. The unit-cell inverse design recast as a regression problem and is solved through multiple data-driven algorithms, namely neural network, (NN), deep neural networks (DNNs) with multiple hidden layers, and the support vector regression (SVR). For the proposed PR unit-cell model, the geometrical dimensions of the unit-cell are predicted for the given frequency ranges in the designated radar frequency bands of X, Ku, K, and Ka with a maximum average accuracy of 95.23%. In order to measure the reliability of this approach for practical applications, we measured the resiliency of all our proposed regression models with characterizing adversarial attacks. This addresses the potential reliability concerns
of integrating ML-based models for FW simulators. Finally, we implement the proposed MLbased unit-cell model in designing low RCS engineered surfaces and investigate monostatic and bistatic RCS reduction results. We showed that the normalized 10dB RCS reduction frequency range matches well with the one given in the input vector of our proposed PR unit-cell model.