In this work, we introduce a novel approach for implementing the multiplier of a leaky integrate and fire (LIF) spiking neuron using a single NMOS transistor, operating in the triode region to get low energy consumption. The parasitic gate-to-source capacitances of the transistors at the input of the voltage comparator stage and the parasitic drain-to-source capacitance of the reset transistor are repurposed as the integrator’s capacitance, eliminating the need for an explicit capacitor, resulting in a highly compact design and significant chip area savings. Furthermore, we introduce a novel single transistor single-pole single-throw (SPST) switch that functions as both a reset mechanism for the membrane potential when in the “ON” state and as a leakage resistor when in the “OFF” state, further minimizing area overhead. The threshold voltage of the voltage comparator is internally generated through optimal sizing of transistors, avoiding the need for external biasing and reducing the overall complexity of the neuron model. The proposed LIF neuron designed provides a higher firing rate (up to 33 MHz) compared to a biological neuron (a few Hz) to enable excellent hardware acceleration. At a supply voltage of 0.6 V and the maximum firing rate, the proposed LIF spiking neuron designed using a mixed-signal approach achieves an average energy consumption of 2.12 fJ per active synapse per input spike and 31.20 fJ per output spike and occupies an 18.5 m^2 area per synapse. To the best of our knowledge, these represent the lowest reported energy consumption and area per synapse in the literature to date. We also validated our proposed LIF spiking neuron model in the convolutional spiking neural network (CSNN) using surrogate gradient descent based supervised learning and achieved 99.33
This paper presents a novel design approach for a dual-band Doherty power amplifier (DPA), operating at 2.6 GHz and 3.5 GHz, utilizing a dual-band direct-matching method. Unlike traditional design methods, which rely on quarter-wavelength transmission lines and separate matching networks for each frequency, the proposed approach employs a compact and efficient direct dual-band synthesis network. This design ensures optimal impedance matching for both frequency bands without the need for individual matching networks, significantly reducing complexity and size while enhancing performance. A detailed theoretical analysis of the required impedance conditions for both saturation and back-off states is provided. We demonstrate the effectiveness of this method using two different types of fully utilized LDMOS-based devices to achieve optimal overall performance. Experimental results show the dual-band DPA achieves over 42 dBm saturation output power and around 60
Conventional Direction-of-Arrival (DOA) estimation methods struggle to capture useful signal features under low signal-to-noise ratio (SNR) conditions, where noise severely corrupts the Sample Covariance Matrix (SCM) spectral structure. To address this, we reframe SCM recovery from a static problem as a task of learning an adaptive recovery mapping parameterized by SNR. Our framework uses a unified network to learn this mapping, enabling it to adapt its recovery strategy across the entire SNR range. This is realized by mapping the SNR space to discrete pseudo-timesteps and designing an SNR-Conditioned Attention U-Net (SCA-UNet). The mechanism relies on Feature-wise Linear Modulation (FiLM) layers to inject the pseudo-timestep condition at multiple scales and dynamically modulate the behavior of the network. Extensive Monte Carlo simulations demonstrate that our framework achieves favorable DOA-estimation RMSE, with its advantage being most evident across the low-SNR regime.
This paper proposes a variational model for additive noise removal in images, employing a nonconvex anisotropic regularization term involving fractional derivatives. The variational problem is approximated over the region using the composite trapezoidal rule to discretize the convolution via the rectangle rule, and approximates the left-sided Riemann-Liouville fractional derivative with the left-sided Grünwald-Letnikov scheme. This ultimately derives a minimization problem in finite-dimensional spaces. Since the minimization problem is non-convex, it is transformed into a convex optimization problem using an iterative substitution method. We solve the convex optimization problem using the alternating direction method of multipliers, which decomposes the target minimization problem into four subproblems. One of the subproblems can be reduced to a generalized Sylvester matrix equation, and we solve this subproblem via a modified Jacobi-gradient iterative algorithm. Our algorithm demonstrates good denoising performance in numerical experiments, outperforming most existing methods.
The development of the discrete fractional Hankel transform (DFRHT) depends fundamentally on the generation of orthonormal eigenvectors of the symmetric kernel matrix of the discrete Hankel transform (DHT). For the DFRHT to approximate its continuous counterpart, namely, the fractional Hankel transform (FRHT), the eigenvectors in question should approximate samples of the eigenfunctions of the Hankel transform (HT), which are the products of generalized Laguerre polynomials, a Gaussian function, and a power function. Consequently, the target eigenvectors will have the descriptor "Laguerre-Gaussian-power-like (LGPL)". The recently developed techniques for generating optimal eigenvectors of any unitary kernel matrix can be classified as either indirect (in the sense of first generating initial eigenvectors as a prerequisite for generating the final optimal ones) or direct (in the sense of not requiring initial eigenvectors). Moreover, the direct techniques can be viewed as either batch (in the sense of generating a partition of the modal matrix corresponding to a single distinct eigenvalue as a whole) or sequential (in the sense of generating the columns of the partition one after another). The present paper aims to assess the performance of the recently developed direct batch and sequential generation algorithms of optimal orthonormal eigenvectors when applied to the real symmetric kernel matrix of the DHT. This research endeavor has not been undertaken before. The assessment will demonstrate the relative merits of each algorithm.
