Neural Network-Based RIS Assisted NLoS DoA Estimation
Authors:
Yasin Azhdari,
Mahmoud Farhang
Abstract:
This paper presents a learning-based approach for Direction of Arrival (DoA) estimation using a Reconfigurable Intelligent Surface (RIS) in a Non-Line-of-Sight (NLoS) scenario. The key innovation is the employment of a novel Neural Network (NN)-based fully connected layer, referred to as the NN-based RIS layer, within a generic Multi-Layer Perceptron (MLP) structure.
The NN-based RIS layer is de…
▽ More
This paper presents a learning-based approach for Direction of Arrival (DoA) estimation using a Reconfigurable Intelligent Surface (RIS) in a Non-Line-of-Sight (NLoS) scenario. The key innovation is the employment of a novel Neural Network (NN)-based fully connected layer, referred to as the NN-based RIS layer, within a generic Multi-Layer Perceptron (MLP) structure.
The NN-based RIS layer is designed to learn the optimal RIS phase shifts that are tailored for the DoA estimation task. To achieve this, the pre-processed real-valued observations are fed into the RIS layer, which has a specialized structure. Unlike regular neural network layers, the weights of the NN-based RIS layer are constrained to be sinusoidal functions, with the phase arguments being the tunable parameters during the training process. This allows the layer to emulate the functionality of an RIS.
Accordingly, the standard feed-forward and back-propagation procedures are modified to accommodate the unique structure of the NN-based RIS layer. Numerical simulations demonstrate that the proposed machine learning-based approach outperforms conventional non-learning-based methods for DoA estimation under almost every practical SNR range in an RIS-assisted scheme.
△ Less
Submitted 26 June, 2024;
originally announced June 2024.
A deep convolutional neural network for salt-and-pepper noise removal using selective convolutional blocks
Authors:
Ahmad Ali Rafiee,
Mahmoud Farhang
Abstract:
In recent years, there has been an unprecedented upsurge in applying deep learning approaches, specifically convolutional neural networks (CNNs), to solve image denoising problems, owing to their superior performance. However, CNNs mostly rely on Gaussian noise, and there is a conspicuous lack of exploiting CNNs for salt-and-pepper (SAP) noise reduction. In this paper, we proposed a deep CNN model…
▽ More
In recent years, there has been an unprecedented upsurge in applying deep learning approaches, specifically convolutional neural networks (CNNs), to solve image denoising problems, owing to their superior performance. However, CNNs mostly rely on Gaussian noise, and there is a conspicuous lack of exploiting CNNs for salt-and-pepper (SAP) noise reduction. In this paper, we proposed a deep CNN model, namely SeConvNet, to suppress SAP noise in gray-scale and color images. To meet this objective, we introduce a new selective convolutional (SeConv) block. SeConvNet is compared to state-of-the-art SAP denoising methods using extensive experiments on various common datasets. The results illustrate that the proposed SeConvNet model effectively restores images corrupted by SAP noise and surpasses all its counterparts at both quantitative criteria and visual effects, especially at high and very high noise densities.
△ Less
Submitted 10 February, 2023;
originally announced February 2023.
