The deep learning era endows immense opportunities for image restoration by leveraging big data generated from widespread sensors and ever-growing computing capabilities. Driven by the advances in deep learning, deep neural networks (DNNs) have been thriving with great success and popularity in tackling problems across diverse topics, including direction-of-arrival (DOA) estimation, synthetic-aperture radar (SAR) imaging, unmanned aerial vehicle (UAV) based signal processing, computer vision, robot control system and autonomous driving. With the boost of superior accuracy, the complete neural network architecture becomes deeper and wider.

Moreover, the increased capabilities of data-driven DNN systems also come along with challenges for efficiently analyzing and processing generated big data, and these challenges are expected to be exacerbated. This is because DNNs are almost always trained on power-hungry modern day supercomputers with costly expensive graphics processing units (GPUs), leading to the limitations of running them on low-powered embedded applications with inadequate hardware resources, such as vehicle-mounted sensors/radars, mobile and wearable devices. Recently, heterogeneous neural network compression and acceleration techniques are proposed to reduce the hardware dependency, such as parameter pruning, quantization, and sparsification. However, the DNNs with compact architectures are vulnerable to significant performance degradation in general.

This Special Issue is devoted to computing and accelerating DNNs efficiently for multidimensional signal processing (MSP), e.g., multiple-input multiple-output (MIMO) localization and UAV, with the goal to highlight novel research developments in deep learning. Although recent research works have shown state-of-the-art performance gains over traditional approaches, compelling research challenges remain to be addressed, especially when encountering the trade-off between complexity and accuracy. We welcome original research and review articles to this effect.

Potential topics include but are not limited to the following:

• Efficient dataset preprocessing techniques for MSP
• Efficient encoding and feature extraction strategies of DNNs for MSP
• Efficient DNN learning methods for MSP
• Efficiently computational models and new architectures of DNNs for MSP, such as spiking neural networks (SNNs), graph neural networks (GNNs), capsule networks, deep unfolding networks, and their variants
• Efficient DNN systems for MSP applications, such as SAR imaging, MIMO/massive MIMO, frequency diverse array (FDA)-MIMO radar, UAV, antenna array, computer vision, robot control system, and others
• Efficient and effective tricks for training DNNs for MSP, such as sparsification, quantization, binarization, thresholding, pruning, and matrix/tensor decomposition