ML is being applied across the spectrum of microwave material characterization—from material classification to the retrieval of dielectric and, more recently, magnetodielectric properties—using a range of sensing approaches and learning algorithms. What remains comparatively underexplored, however, is how the learning process itself is conducted and, in particular, how the reliability of the resulting models is assessed and reported. This gap motivates the present guidelines, and a detailed, methodology-focused review of works in this area is provided later in the article, once the relevant concepts have been introduced. It is worth mentioning that methodological transparency is essential to guarantee scientific rigor and allow meaningful comparisons across studies. Therefore, a well-defined and reproducible methodology strengthens the reliability of ML-based characterization and facilitates independent verification. With these questions in mind, this article provides guidelines and procedures for training and evaluating ML models in MD material characterization, with the aim of advancing the development of reliable predictive models. It begins with a description of a use case employed to illustrate the process, followed by a general overview and definitions related to ML algorithms and the training procedure. The guidelines are then introduced through the typical workflow for training ML models, with the use case developed in parallel to facilitate comprehension and to demonstrate their practical application step-by-step. This integrated approach aims to help readers connect theoretical concepts with real implementations, ensuring that the guidelines are not only explained but also exemplified in a context that is readily understandable to microwave engineers. In the last part, the article includes a brief review of ML applications in microwave material characterization, highlighting the methods reported to assess the reliability of trained models. Although some reviews have addressed ML-based methods for microwave characterization [[18], [19]](#ref-0018), they do not examine how the learning process is conducted or evaluated in terms of the models? reliability. The article concludes with some final remarks on our experience with ML and MD composites..
This article introduces a low-cost, printed circuit board (PCB)-based educational scalar reflectometer to support hands-on teaching of reflection coefficient measurement and antenna impedance matching in microwave engineering courses. The platform integrates off-the-shelf components—a broadband directional coupler, passive RF envelope detectors, configurable analog amplification stages, and a microcontroller—for real-time magnitude estimation in the 50-MHz to 1-GHz band, relevant to subgigahertz Internet of Things (IoT) applications. Inexpensive vector network analyzers (VNAs)have made reflection measurement itself widely accessible, but they remain closed instruments: The platform proposed here is low in cost and, above all, open. Unlike commercial network analyzers, the design exposes the full measurement chain, from wave sampling to reflection computation, ensuring pedagogical transparency and modularity: Students can configure analog paths, probe functional blocks, and explore directional coupler directivity, detector nonlinearity, gain tradeoffs, and calibration effects. Experimental validation demonstrates coupler directivity ranging from 8 dB to 15 dB across the target band, and a stable operating window of approximately 16 dB (from +4 dBm to +20 dBm) in the amplified configuration. All design files, firmware, and scripts are publicly available to facilitate adoption. A structured 10-stage laboratory sequence, adaptable to courses of different durations and levels, is proposed, and a pedagogical evaluation based on a student survey (n = 12) yielded an overall mean satisfaction score of 4.53 out of 5, with all respondents willing to recommend the platform in future courses.
The rapid growth of artificial intelligence, cloud computing, and hyperscale data centers is driving an unprecedented demand for communication systems with terabit-per-second throughput, ultralow latency, and massive connectivity [1], [2]. However, the limited bandwidth and increasing spectral congestion in the currently utilized microwave and millimeter-wave bands constrain further improvements in transmission capacity and data rate. By contrast, the terahertz (THz) band (generally defined as 0.1–10 THz) offers abundant spectrum resources and is the candidate band for 6G and future communication systems [3], [4]. Consequently, THz technology for high-speed interconnects and large-capacity communications has received much attention. Efficient electromagnetic wave guiding is one key challenge in the THz range. Conventional microwave transmission lines, including coaxial cables, microstrip lines (MSLs), and coplanar waveguides (CPWs), etc., suffer severe ohmic loss that is due to the metal conductors’ skin effect at THz frequencies. Standard metal waveguides, which are hollow, can provide low loss, but their submillimeter dimensions require high-precision fabrication and assembly, particularly for bends, junctions, and complex integrated circuits. Optical fibers, by contrast, provide exceptionally low transmission loss and high capacity in the optical domain but are constrained by electrooptic conversion, heterogeneous integration with THz electronic circuits, and the associated cost and system complexity [5]. By comparison, dielectric waveguides (DWGs) can effectively avoid the severe ohmic loss and thus exhibit low loss in the THz band since the electromagnetic energy is primarily confined within the dielectric rather than guided by metal conductors. Furthermore, they offer advantages of wide bandwidth, low cost, direct connection to THz electronic circuits, and compatibility with a wide range of manufacturing technologies, including polymer processing [6], printed circuit board (PCB) fabrication [7], [8], and semiconductor processes [9], [10]. Owing to these advantages, THz DWGs have attracted extensive attention for future high-speed wired interconnects and large-capacity communications. From a system-level perspective, THz DWGs serve two distinct but complementary roles. One group comprises short-range DWGs, which generally employ rigid or semirigid, planar or quasi-planar structures and are primarily intended for chip-, package-, and board-level interconnects over millimeter- to centimeter-scale distances. Moreover, their compact geometries and compatibility with planar or microfabrication technologies facilitate the incorporation of functional devices (such as transitions, filters, antennas, and multiplexers, etc.) and integration platforms [11], [12], [13], [14]. The other group comprises extended-range DWGs, which are typically flexible, elongated, and freestanding. They work as channels between physically separated modules or systems and are primarily intended for board-to-board and equipment-to-equipment links over centimeter- to meter-scale distances. Recently, further studies have been conducted to introduce them to enable wireless data distribution systems with flexible antennas [15]. This article presents a review of recent advances in THz DWG technology, aiming to provide researchers interested in this domain with an understanding of the different types of THz DWGs as well as their device- and system-level applications, as shown in Figure 1.
Radar-based vital-sign sensing has been a steadily evolving field of research for the last couple of decades. The working principle relies on a radar sensor, which monitors the relative temporal changes of the shape and small-scale vibrations of the human body, from which one can extract information on the patient`s cardiovascular activity by appropriate signal processing. After initial publications on this topic around the 1980s, the advantages of contactless vital-sign monitoring have triggered a multitude of research activities to advance this technology to potentially become a practical alternative to conventional state-of-the art contact-based measurement techniques, e.g., electrocardiography (ECG). An emerging research focus is devoted to the fundamental physical principle radar relies on: scattering and reception of electromagnetic waves. In fact, a millimeter-wave medical radar setup, in which the radar antenna is typically less than a couple of meters in front of the electrically large human body whose dimension is, in turn, comparable to the range, is an atypical radar configuration, which triggers the onset of near-field scattering phenomena. Independent of the actual extraction of vital-sign metrics, the understanding and analysis of these superordinate scattering and wave propagation effects, presented in this article, is essential for the awareness and correct interpretation of the radar channel behavior with respect to power and phase of the received signal, which deviates from classical far-field relationships.
