High-precision and low-latency wireless localization is a key objective for future networks. The majority of existing localization methods rely on multi-antenna arrays (MAAs) to estimate directions of arrival (DoAs), but the size and cost of MAAs limit their application in portable electronic devices. Virtual antenna arrays (VAAs), constructed from signals received at different positions by a moving single-antenna receiver, offer a promising alternative. However, VAA-based localization is challenged by the local oscillator frequency offset (LOFO) in transceivers, the irregular array geometry caused by the receiver’s movement, and the non-line-of-sight (NLoS) environments due to physical blockage. To address the above challenges, this paper proposes a VAA-based online localization approach. Specifically, we employ a manifold separation technique based on the Jacobi-Anger expansion to maintain a uniform linear array-like channel representation for the irregular VAA geometry. This enables us to transform the joint estimation of multi-path DoAs, times of arrival (ToAs), and LOFO into a modified two-dimensional atomic norm minimization problem, which is solved using an alternating convex search algorithm. Additionally, we introduce an unscented Kalman filter-based simultaneous localization and mapping algorithm for real-time position tracking in NLoS environments. Extensive simulations validate the effectiveness of our approaches, showing a 70.15% reduction in DoA estimation error and a 54.02% reduction in localization error compared to benchmark methods.

Previous studies on time-of-arrival (ToA) estimation with phase ambiguity have largely relied on correlation matrices constructed from multi-shot signals. However, only a single-shot signal with binary phase ambiguity is available in Bluetooth low energy (BLE) ranging systems, rendering traditional methods ineffective. To address the above limitation, this paper proposes an efficient method for the joint estimation of ToA and phase ambiguity in BLE ranging systems. By leveraging the idea of beamforming among multiple frequency channels, the proposed method is comprised of three phases. In the first phase, we extract the one-way time differences of arrival (TDoAs) from the two-way channel frequency response measurements based on a modified atomic norm minimization algorithm by further addressing a component-dominating issue. In the second phase, we solve the matching problem related to the topological order of TDoAs by employing an alternating optimization framework and the Hungarian matching algorithm. In the final phase, we cope with the remaining first-arriving delay estimation problem by solving a series of low-rank MaxCut semidefinite programs. In particular, we prove the uniqueness and the global optimality of the obtained solution under certain conditions that can be easily satisfied in practice. Simulation shows that the proposed method can achieve the nanosecond-level accuracy in ToA estimation at a signal-to-noise ratio of 25 dB, with approximately 90% of ToA estimations exhibiting errors of less than 18 nanoseconds.

Dropout [1], a standard regularization technique in Large Language Models, incurs extra computational overhead, particularly due to its repeated application during the training process. To address this issue, we propose Accelerated Dropout, an algorithm that revolutionizes the traditional dropout method by employing a bitmask, rather than a float mask, to significantly expedite the training process. In the theoretical proof, we demonstrate that accelerated dropout is exponentially convergent to the probability of retaining neurons. Our extensive experimental analysis, conducted on 13 benchmark datasets and 9 deep learning models, confirms that accelerated dropout outperforms traditional dropout in terms of training efficiency and generalization performance. The experimental results indicate that, compared to torch dropout, our accelerated dropout achieves 8x speedup on a single dropout operator. With the use of accelerated dropout, the average training speed of each model increased by 1.0624x. The training speed improvements ranged from a 1.046x increase for ProphetNet on the CNN/DailyMail dataset to a 1.077x increase for ViT-B on the ImageNet dataset.

Data harmonization is critical for establishing generalizable model on multicenter medical data. Traditional data harmonization strategies aim to align data distributions from different sources, but often lack mechanisms to learn hidden complementary and discriminative manifestations from multicenter data. To this end, we proposed a methodology for harmonizing multicenter data by matching their first and second order statistics in a shared space, which is framed in an optimization architecture to learn harmonized and discriminative latent features for downstream classification modeling. The developed method integrated representation learning, feature dimension reduction and selection within a unified framework. Several relational regularizations such as data attribute preservation and feature-task correlation have been explored and incorporated to encourage learning potential associations inherent in multicenter data. Extensive evaluations on three independent clinical datasets have demonstrated the efficacy of the proposed method in producing harmonized and distinguishing data for multicenter medical prediction modeling.

Conventional direction of arrival (DoA) estimation techniques require multi-antenna arrays at the transceiver, leading to high hardware costs. To address this issue, this paper investigates the joint DoAs and time of arrivals (ToAs) estimation in a single-input single-output (SISO) multi-carrier system. Specifically, the DoAs are estimated using a virtual antenna array (VAA), which is constructed by integrating signals received at different positions of a moving receiver. Since the receiver trajectory is irregular and there is a relatively large time interval between two neighboring positions, mainstream estimation methods in VAA face two challenges: i) The VAA lacks the advantageous antenna geometry of a uniform linear array (ULA) for signal processing, and ii) There exist phase errors caused by the oscillator frequency offset between the transmitter and receiver. We overcome the first challenge using an array manifold separation technique, which transforms the irregular VAA to the ULA geometry via the Jacobi-Anger expansion. After manifold separation, we accomplish super-resolution recovery of DoAs and ToAs by formulating an atomic norm minimization (ANM) problem. To address the second challenge, we propose an alternating optimization approach that iteratively solves the ANM and frequency offset problems. In each iteration, the frequency offset is solved through a bounded univariate minimization problem given the DoAs and ToAs. Numerical experiments consolidate that the proposed estimators for DoAs and ToAs exhibit superior performance, and their errors are close to the Cramer-Rao Lower Bound (CRLB) when SNR ≥ 5dB.