This paper proposes a W-band full-band waveguide bandpass filter with loaded capacitive metal columns. Theoretical analysis was first conducted, and a 9th-order offset magnetic coupling wideband filter prototype was designed. Then, by inserting metal columns into the resonant coupling cavities of the prototype, capacitance was introduced to perturb the resonant electric field within the cavities and improved the rectangular coefficient, which enhances the filter's out-of-band rejection. The measurement results show that the passband frequency range of the filter covers 70–110 GHz, with a fractional bandwidth of approximately 45% and a 30 dB rectangular coefficient of 0.824. To the best of the author's knowledge, this is the first millimeter-wave cavity filter loaded with capacitive metal columns. Compared to other research results, the proposed filter combines wideband and high out-of-band rejection characteristics, demonstrating excellent performance and suitability for use in extension modules of millimeter-wave terahertz measurement systems or high-speed communication systems.
Efficient beam training across the transition from conventional far-field massive multiple-input multiple-output (Massive MIMO) to near-field extremely large-scale MIMO (XL-MIMO) is fundamental to forth-coming sixth-generation (6G) systems, yet remains impeded by severe mobility, spherical-wave propagation and prohibitive training overhead. This paper presents a unified survey of contemporary deep learning (DL)-based beam training techniques, reviewing their evolution from early deep neural networks to more advanced architectures such as convolutional, recurrent, and other specialized models. These approaches exhibit strong adaptability and feature extraction capabilities. However, research in this area is still in its early stages, and such methods have yet to be widely applied in emerging scenarios, such as dynamically varying or multi-user interference-limited near-field communication systems. Motivated by the survey insights, we further introduce a lightweight stacked bidirectional gated recurrent unit (BiGRU) architecture equipped with residual shortcuts and adaptive normalization to capture the temporal evolution of near-field channels, referred to as the stacked residual-BiGRU (SRB). Simulations conducted on a dynamic XL-MIMO testbed demonstrate the competitive achievable rates and low training overhead of SRB, highlighting how the recurrent structure complements spatial feature extractors. Finally, we outline open challenges and future directions toward scalable, energy-efficient and robust beam management frameworks for real-world deployments.
To overcome the limitations of conventional proportional-integral-derivative (PID) controllers in handling unknown and coupled double-tank systems, this paper proposes an attention-based multi-objective reward Q-learning (AMOR-QL) algorithm. In the proposed algorithm, a three-dimensional state space integrating error, integral error, and previous error is constructed, forming the basis for the algorithm's state perception. Subsequently, equidistant binning is used to discretize the state space, and PID controller parameters are adaptively tuned through incremental actions from a finite adjustment set. Next, a multi-objective dynamic reward function that combines error suppression, control smoothness, and parameter stability is proposed. Furthermore, the AMOR-QL algorithm is developed by employing an event-triggered experience storage mechanism combined with an attention-based experience replay strategy, in which state similarity analysis guides the learning process towards context-relevant experiences. This design enables the optimal PID controller parameters to be learned using only data with guaranteed stability, while significantly improving learning efficiency. Simulation results validate the proposed method's superiority over traditional PID design methods, Q-learning, deep Q-network, and double deep Q-network across multiple performance metrics.
Accurate prediction of marine traffic flow is crucial for ensuring maritime safety. However, marine traffic flow exhibits nonlinearity and non-stationarity, posing significant challenges to the performance improvement of deep learning based prediction models. To address these challenges, we propose a novel composite model by combining Kolmogorov-Arnold network (KAN) with variational mode decomposition (VMD) and bidirectional long short-term memory network (BiLSTM), termed VMD-multi-feature BiLSTM (MFBiLSTM)-KAN model. The original non-stationary time series are decomposed by VMD. We generate multiscale features with multiscale information by concatenating original time series with stable time series generated by VMD. Then, BiLSTM captures bidirectional dependency patterns among multiscale features. KAN flexibly maps the outputs of BiLSTM to final prediction results, by using learnable activation functions as weights. We employ the VMD-MFBiLSTM-KAN model for marine traffic flow prediction and validate its performance using real-world datasets derived from automatic identification system. Experimental results demonstrate that the VMD-MFBiLSTM-KAN model achieves superior prediction accuracy compared with other traditional models by utilizing multiscale features with comprehensive information and KAN with high flexibility.
This paper investigates a near-field integrated sensing, communication, and power transfer (ISCPT) system that employs hybrid beamforming with a fully connected antenna architecture. We formulate an optimization problem to maximize the sum rate of communication users via hybrid beamforming design, subject to four constraints: the maximum Cramér-Rao bound constraint for sensing, the minimum harvested energy constraint for wireless power transfer, the base station's total transmit power constraint, and unit-modulus phase-shifter constraint on analog beamforming elements. The formulated problem is non-convex with highly coupled variables, posing significant optimization challenges. To address this, a two-stage optimization framework is proposed: The first stage optimizes the fully digital beamforming matrix to enhance algorithm performance, while the second stage decomposes the matrix into analog and digital components to achieve hybrid beamforming. Numerical simulations illustrate the influence of various constraint parameters on the sum rate and validate the effectiveness of the proposed two-stage optimization framework. Furthermore, numerical results show that by utilizing a certain number of radio frequency chains, the performance of the hybrid architecture can closely approximate that of the fully digital architecture. This hybrid architecture design enhances the economic viability of the system in practical applications.
