Acquiring downlink channel state information (CSI) at the base station is vital for optimizing performance in massive Multiple input multiple output (MIMO) Frequency-Division Duplexing (FDD) systems. While deep learning architectures have been successful in facilitating UE-side CSI feedback and gNB-side recovery, the undersampling issue prior to CSI feedback is often overlooked. This issue, which arises from low density pilot placement in current standards, results in significant aliasing effects in outdoor channels and consequently limits CSI recovery performance. The main objective of this work is to solve this issue by introducing a new CSI upsampling framework at the gNB as a post-processing solution to address the gaps caused by undersampling. Leveraging the physical principles of discrete Fourier transform shifting theorem and multipath reciprocity, our framework effectively uses uplink CSI to mitigate aliasing effects. We further develop a learning-based method that integrates the proposed algorithm with the Iterative Shrinkage-Thresholding Algorithm Net (ISTA-Net) architecture, enhancing our approach for non-uniform sampling recovery. Our numerical results show that both our rule-based and deep learning upsampling methods significantly outperform traditional interpolation techniques or multiple state-of-the-art approaches by 8-13 dB and 2-10 dB, respectively, in terms of normalized mean square error.

Acquiring downlink channel state information (CSI) is crucial for optimizing performance in massive Multiple Input Multiple Output (MIMO) systems operating under Frequency Division Duplexing (FDD). Most cellular wireless communication systems employ codebook-based precoder designs, which offer advantages such as simpler, more efficient feedback mechanisms and reduced feedback overhead. Common codebook-based approaches include Type II and eType II precoding methods defined in the 3GPP standards. Feedback in these systems is typically standardized per subband (SB), allowing user equipment (UE) to select the optimal precoder from the codebook for each SB, thereby reducing feedback overhead. However, this subband-level feedback resolution may not suffice for frequency-selective channels. This paper addresses this issue by introducing an uplink CSI-assisted precoder upsampling module deployed at the gNodeB. This module upsamples SB-level precoders to resource block (RB)-level precoders, acting as a plug-in compatible with existing gNodeB or base stations.

Massive MIMO systems can achieve high spectrum and energy efficiency in downlink (DL) based on accurate estimate of channel state information (CSI). Existing works have developed learning-based DL CSI estimation that lowers uplink feedback overhead. One often overlooked problem is the limited number of DL pilots available for CSI estimation. One proposed solution leverages temporal CSI coherence by utilizing past CSI estimates and only sending channel state information-reference symbols (CSI-RS) for partial arrays to preserve CSI recovery performance. Exploiting CSI correlations, FDD channel reciprocity is helpful to base stations with direct access to uplink CSI. In this work, we propose a new learning-based feedback architecture and a reconfigurable CSI-RS placement scheme to reduce DL CSI training overhead and to improve encoding efficiency of CSI feedback. Our results demonstrate superior performance in both indoor and outdoor scenarios by the proposed framework for CSI recovery at substantial reduction of computation power and storage requirements at UEs.

Transmitter-side channel state information (CSI) is vital for large MIMO downlink systems to achieve high spectrum and energy efficiency. Existing deep learning architectures for downlink CSI feedback and recovery show promising improvement of UE feedback efficiency and eNB/gNB CSI recovery accuracy. One notable weakness of current deep learning architectures lies in their rigidity when customized and trained according to a preset number of antenna ports for a given compression ratio. To develop flexible learning models for different antenna port numbers and compression levels, this work proposes a novel scalable deep learning framework that accommodates different numbers of antenna ports and achieves dynamic feedback compression. It further reduces computation and memory complexity by allowing UEs to feedback segmented DL CSI. We showcase a multi-rate successive convolution encoder with under 500 parameters. Furthermore, based on the multi-rate architecture, we propose to optimize feedback efficiency by selecting segment-dependent compression levels. Test results demonstrate superior performance, good scalability, and high efficiency for both indoor and outdoor channels.

Acquiring downlink channel state information (CSI) at basestation (gNB) is crucial for optimizing performance in massive MIMO FDD systems. Deep learning (DL) architectures have shown successes in enabling UE-side CSI feedback and gNB-side recovery, but often lack flexibility and/or require volumes of customized training data for specific RF channel environments and compression ratios. This work proposes a new CSI feedback architecture called zero-replacement (ZR). ZR is free from customized training and can be directly applied to new and unseen channel scenarios without pre-training and/or customization. It is also scalable and simple to implement, making it suitable for practical massive MIMO wireless deployment. We further generalize a Select-ZR algorithm, which switches between different sparse transformation techniques to enhance recovery performance. Our numerical results demonstrate that both proposed ZR and Select-ZR algorithms achieve competitive CSI recovery accuracy and feedback efficiency across various channels against highly complex data-driven DL models.

This paper proposes a hybrid 3D placement approach which can reduce the number of TSVs while getting short wirelength. Although existing 3D analytical placement approaches can get short wirelength, they usually use a large number of TSVs. Moreover, because cells may be allocated to two tiers by their analytical placement formulations, placement utilization cannot be calculated precisely which may increase the difficulty in cell legalization. To get a less number of TSVs, our approach first allocates cells to tiers by a partitioning algorithm and maintains the result in the following stages. More importantly, we propose a novel 2D-2-3D analytical placement approach according to the space transformation technique. This approach can spread cells over 3D space while optimizing 3D wirelength by a 2D analytical placement formulation. Moreover, placement utilization can be calculated accurately in this approach since cells and TSVs are spread over a 2D plane instead of a 3D space. Experimental results show that our approach can obtain short wirelength and significantly fewer TSVs than previous works.

In this paper, we consider the automotive orthogonal frequency division modulation-radar in millimeter wave band. To avoid interference between different radar systems, resources need to be split and then used by different radar systems. This thus degrades the radar performance as compared to the radar system having full resources (FRs). To mitigate this issue, we develop a deep learning-based range-Doppler (R-D) map reconstruction approach along with a time-frequency resource allocation scheme. In the reconstruction approach, we propose a deep learning-based convolutional neural network to reconstruct the R-D map such that the reconstructed R-D map can be close to the R-D map under FRs. In the resource allocation scheme, we propose a block-wise interleaved method that can facilitate the proposed reconstruction approach. Simulation results show that our proposed approach can effectively mitigate the performance degradation of radar systems when resources are shared among users.

Recently, non-contact vital sign estimating devices, which are used for health monitoring, have gradually gained interest among researchers. However, most of these devices have the disadvantages of high power consumption and high cost, which limit their practicality. Therefore, a less-expensive radar-based system is suggested for long-term health monitoring. Existing radar-based vital sign estimating schemes introduce unacceptable estimating errors. In order to improve the precision and stability, we employ Newtonized Orthogonal Matching Pursuit (NOMP) algorithm. NOMP provides better estimating results compared to existing schemes in vital sign estimation tasks. However, the performance of NOMP deteriorates severely under conditions of low signal-to-noise ratio, which causes poor power efficiency. In this study, we propose deep learning (DL)-aided NOMP schemes to tackle the aforementioned issue. Our simulation results and over the air measurements suggest that DL-aided NOMP schemes are superior to existing schemes.

The Society of Automotive Engineers (SAE) emphasizes that radars have become a critical technology. With the consideration of application scenarios and cost, increasing importance has been attributed to millimeter Wave (mmWave) radars. Because of the unique features of simultaneous detection and communication, Orthogonal Frequency- Division Modulation (OFDM) radars have been discussed frequently in the literature. In this paper, under the architecture of an OFDM radar, we propose a novel protocol which divides different radar users (i.e., different vehicles) into different logical channels by time and frequency. Following the rules of the protocol, a non- cooperative interference avoidance logical channel selection method is also proposed to choose a logical channel with the least inter-carrier interferences (ICIs). Simulation results show that the proposed logical channel selection method can choose a nearly optimal channel in a short time with a high probability.

The well-known cell-averaging constant false alarm rate (CA-CFAR) scheme and its variants suffer from masking effect in multi-target scenarios. Although order-statistic CFAR (OS-CFAR) scheme performs well in such scenarios, it is compromised with high computational complexity. To handle masking effects with a lower computational cost, in this paper, we propose a deep-learning based CFAR (DL- CFAR) scheme. DL-CFAR is the first attempt to improve the noise estimation process in CFAR based on deep learning. Simulation results demonstrate that DL-CFAR outperforms conventional CFAR schemes in the presence of masking effects. Furthermore, it can outperform conventional CFAR schemes significantly under various signal-to-noise ratio conditions. We hope that this work will encourage other researchers to introduce advanced machine learning technique into the field of target detection.