Electronics Letters最新文献

Pipelined analogue-to-digital converters suffer from inter-stage gain errors and inter-stage nonlinearity errors due to gain variations and nonlinearity in residue amplifiers. While a polynomial-based calibration algorithm can address these errors, its conventional implementation demands excessive hardware resources and power consumption. This letter introduces a novel calibration algorithm that combines precomputation with a lookup table, achieving improved hardware efficiency while maintaining calibration accuracy and reducing latency.

Sparse arrays can increase the array aperture, thereby enhancing angular resolution. However, this also introduces additional computational complexity. This letter proposes a symmetric sparse array structure, where subarrays with different inter-element spacings sample distinct spatial domain signals, analogous to the use of mother wavelets at different scales in wavelet theory to process various frequency components of a signal. The root-MUSIC method can be directly applied to the proposed method, and the simulations demonstrate that it achieves direction-of-arrival estimation performance comparable to that of super-nested arrays while maintaining lower computational complexity.

Bayesian optimization (BO) is a powerful and sample-efficient method for optimizing black-box functions. In practical applications, however, it faces two major challenges: (a) scaling to high-dimensional parameter spaces and (b) optimizing in mixed spaces that encompass continuous, binary and categorical variables. This paper presents a unified approach, termed ComplexBO, to address these challenges by leveraging complex-valued representations. Our approach utilizes complex numbers to represent the variables in both high-dimensional input spaces and low-dimensional target spaces, modelling the mapping between the two spaces as a rotation operation of complex numbers. Unlike the subspace embedding methods, our ComplexBO provides an elegant solution to handle diverse types of input variables while preserving non-linear properties. Empirical evaluations show that our ComplexBO achieves competitive results compared to the state-of-the-art methods across a wide range of tasks, including machine learning benchmarks and context length extension in large language models.

Fibre-optic propagation could be harnessed to realize a photonic extreme learning machine (ELM) that accelerates classification tasks. Leveraging fibre non-linearity and dispersion during propagation is beneficial to realize high-accuracy photonic ELMs with expanding the hidden layer size. We present a scalable strategy to enlarge the ELM hidden layer size via spatial diversity enhancement by coupled-core multi-core fibres (CC-MCFs). Beyond the spatial parallelism of CC-MCFs, we transform inter-core crosstalk (IC-XT) and stochastic spatial-mode dispersion (SMD) into computational resources rather than impairments in communications. Temporally-coded sequences propagation over CC-MCFs is rigorously emulated via solving coupled nonlinear Schrodinger equations (CNSEs). Through investigating the MNIST handwritten-digit dataset as a benchmark, we demonstrate that deliberately harnessed IC-XT and SMD coefficients could boost ELM classification accuracy by 5.72% and 7.20% over SMFs, respectively. Requirements on system SNR and launched optical power are also discussed. We believe the results establish CC-MCF-enabled space-division multiplexing as a promising route toward ultra-high-accuracy, hardware-accelerated photonic ELMs.

In this paper, we reconceptualize visual tracking as a multivariate time-series forecasting (MTSF) problem. Specifically, the goal of visual tracking—predicting the target's state over time, including its $��(�x�,�y�)�$ center coordinates and scale—can be naturally framed as forecasting future states from a sequence of past observations. Viewed through this lens, visual tracking aligns with the challenges of MTSF, where the objective is to capture complex temporal dependencies among multiple variables. However, applying MTSF to visual tracking introduces new difficulties due to the inherently intricate nature of object motion, which often involves abrupt and nonlinear variations in direction, velocity, and behavioral patterns. To address these complexities, we propose a principled approach that models the dynamic nature of target motion using a regime-switching framework. This method employs an underlying Markov jump process (MJP) to govern transitions between multiple latent motion patterns, each characterized by its own stochastic differential equation (SDE). By doing so, our model adapts to diverse temporal dynamics in a data-driven manner, enabling robust and precise prediction of future target states. Experimental results demonstrate that our method outperforms state-of-the-art visual tracking approaches, particularly in scenarios where target objects exhibit diverse and dynamic motion patterns over time.

This letter presents a fully integrated 22-MHz relaxation oscillator for ultra-low-power applications in automotive internet-of-things (IoT) systems. Fabricated using a 110-nm CMOS process, the oscillator occupies an active area of 0.05 $�{\mathrm{mm}}^2$. Empowered by the complementary input comparator featuring power efficiency and low propagation delay, the oscillator achieved an extremely low-power consumption of 5.9 $��\mu �W�$ from a 1-V power supply, resulting in an unprecedented energy efficiency of 0.27 $��\mu �\text{W/MHz}�$, enabling continuous operation for up to 58 years on a standard LR6 dry battery. The periodic jitter is 131 ps and the start-up time is within 1.2 $��\mu �s�$. It operates across a wide temperature range from −40$��{}^{\circ }�C�$ to 160$��{}^{\circ }�C�$.

In order to solve the problems of insufficient brightness, excessive noise, and loss of details in extremely dark environments, this paper proposes a novel low-light image enhancement model: WUCMamba. The model adopts a three-stage encoder-latent layer-decoder architecture, and its core component WUM module (Wavelet-UC-Net-Mamba) is responsible for processing upsampling and downsampling of each layer. Specifically, the module separates low-frequency and high-frequency information through wavelet transform. The low-frequency branch adopts the improved UC-Net and integrates the CBAM attention module to fuse the multi-layer features obtained by downsampling with the channel and spatial attention mechanism, thereby enhancing the model's ability to extract significant information; the high-frequency branch is first enhanced by residual convolution and reconstructed into a sequence, and then the sequence is processed by the channel-level Mamba module to suppress noise and enhance key textures. Finally, the sequence is spatially reconstructed back to high-frequency information. The processed high-frequency and low-frequency features are finally reconstructed back into the image through inverse wavelet transform. The proposed WUCMamba model achieves a PSNR of 22.19 dB and an SSIM value of 0.8615 on the LOLv2-real dataset, outperforming Zero-DCE, URetinex-Net, and RRDNet in terms of enhancement quality. On the LOLv2-syn dataset, the model achieves a PSNR of 24.14 dB and an SSIM value of 0.9125, again outperforming the above methods. In terms of model complexity, ours is higher than the two algorithms without reference: Zero-DCE and RUAS, but compared with the URetinex-Net algorithm which also uses reference learning, our model has less computational complexity. These results confirm that WUCMamba provides good visual quality and efficiency in both real-world and synthetic low-light scenes.

A novel light-controlled coplanar waveguide (CPW) RF attenuator is proposed. By directly metalizing a thin (<100 um) silicon substrate to form a circular meandered CPW transmission line, our device achieves an isolation/insertion-loss ratio greater than 33 dB from 2 to 18 GHz and only requires an optical bias and an electrical bias of 7 and 17.8 mW, respectively. Moreover, the proposed architecture is compatible with the thin-silicon complementary metal-oxide semiconductor (CMOS) technology which may pave the way for the large-scale integration of high-performance light-controlled and CMOS devices for demanding RF applications.

Reconfigurable devices are of paramount importance in many applications, and a method inspired by Pareto-like optimality can generate a sequence of configurations, each of which is the best trade-off between design criteria. Accordingly, an original concept of Pareto meta-front is presented: given a family of fronts depending on a variable parameter, the meta-front is defined as the curve which joins all the closest-to-utopia points. It describes the variation of the best combinations of two conflicting design criteria against a cost parameter like, for example, a material property. The case study considered in order to prove the validity of the proposed method is the optimal shape design of an interior permanent-magnet motor (IPM) that is assumed to be magnetically reconfigurable, meaning that the variable parameter indexes a set of several permanent-magnet materials, each of which is featured by its own demagnetisation curve.

This paper describes the topology and operation of a new family of asynchronous frequency dividers called fractional digital regenerative frequency dividers. The operation is explained using a digital approach in the time domain compared to the more classical analogue approach in the frequency domain. A fractional behaviour is intrinsic to this topology without the mandatory need of an external modulator. Integer ratios can also be easily obtained, for example to serve as multi-modulus dividers when required. Starting with the elementary version of this kind of divider, a generalization is made with repeated cell patterns in the closed loop for finer control of the fractional ratio. Some measurements on an FPGA-based implementation are provided to validate the division ratio of the elementary version as well as an example of a more complex configuration.