Compute-in-memory (CIM) has emerged as a pivotal direction for accelerating workloads in the field of machine learning, such as Deep Neural Networks (DNNs). However, the effectively exploitation of sparsity in CIM systems presents numerous challenges, due to the inherent limitations in their rigid array structures. Designing sparse DNN dataflows and developing efficient mapping strategies also become more complex when accounting for diverse sparsity patterns and the flexibility of a multi-macro CIM structure. Despite these complexities, there is still an absence of a unified systematic view and modeling approach for diverse sparse DNN workloads in CIM systems. In this paper, we propose CIMinus, a framework dedicated to cost modeling for sparse DNN workloads on CIM architectures. It provides an in-depth energy consumption analysis at the level of individual components and an assessment of the overall workload latency. We validate CIMinus against contemporary CIM architectures and demonstrate its applicability in two use-cases. These cases provide valuable insights into both the impact of sparsity patterns and the effectiveness of mapping strategies, bridging the gap between theoretical design and practical implementation.

Being able to estimate monocular depth for spherical panoramas is of fundamental importance in 3D scene perception. However, spherical distortion severely limits the effectiveness of vanilla convolutions. To push the envelope of accuracy, recent approaches attempt to utilize Tangent projection (TP) to estimate the depth of$360 ^{\circ }$images. Yet, these methods still suffer from discrepancies and inconsistencies among patch-wise tangent images, as well as the lack of accurate ground truth depth maps under a supervised fashion. In this paper, we propose a geometry-aware self-supervised$360 ^{\circ }$image depth estimation methodology that explores the complementary advantages of TP and Equirectangular projection (ERP) by an asymmetric dual-domain collaborative learning strategy. Especially, we first develop a lightweight asymmetric dual-domain depth estimation network, which enables to aggregate depth-related features from a single TP domain, and then produce depth distributions of the TP and ERP domains via collaborative learning. This effectively mitigates stitching artifacts and preserves fine details in depth inference without overspending model parameters. In addition, a frequent-spatial feature concentration module is devised to simultaneously capture non-local Fourier features and local spatial features, such that facilitating the efficient exploration of monocular depth cues. Moreover, we introduce a geometric structural alignment module to further improve geometric structural consistency among tangent images. Extensive experiments illustrate that our designed approach outperforms existing self-supervised$360 ^{\circ }$depth estimation methods on three publicly available benchmark datasets.

Processing-in-memory (PIM), as a novel computing paradigm, provides significant performance benefits from the aspect of effective data movement reduction. SRAM-based PIM has been demonstrated as one of the most promising candidates due to its endurance and compatibility. However, the integration density of SRAM-based PIM is much lower than other nonvolatile memory-based ones, due to its inherent 6T structure for storing a single bit. Within comparable area constraints, SRAM-based PIM exhibits notably lower capacity. Thus, aiming to unleash its capacity potential, we propose DDC-PIM, an efficient algorithm/architecture co-design methodology that effectively doubles the equivalent data capacity. At the algorithmic level, we propose a filter-wise complementary correlation (FCC) algorithm to obtain a bitwise complementary pair. At the architecture level, we exploit the intrinsic cross-coupled structure of 6T SRAM to store the bitwise complementary pair in their complementary states$(Q/\overline {Q})$, thereby maximizing the data capacity of each SRAM cell. The dual-broadcast input structure and reconfigurable unit support both depthwise and pointwise convolution, adhering to the requirements of various neural networks. Evaluation results show that DDC-PIM yields about$2.84\times $speedup on MobileNetV2 and$2.69\times $on EfficientNet-B0 with negligible accuracy loss compared with PIM baseline implementation. Compared with state-of-the-art SRAM-based PIM macros, DDC-PIM achieves up to$8.41\times $and$2.75\times $improvement in weight density and area efficiency, respectively.

Radio tomographic imaging (RTI) has become a popular approach to reconstruct spatial loss fields (SLFs) in an area covered by a wireless network based on received signal strength (RSS) measurements. SLF images quantify the attenuation rate of the radio-frequency waves at each location in the network. The attenuation for the propagation path can be modeled as the 2-dimensional integral of SLF scaled by a weight function, which is the foundation of RTI techniques and makes the SLF reconstruction possible. In recent years, many methods, including machine-learning-based schemes, have been proposed to achieve more accurate SLF estimates. In this letter, we develop an attention neural network-augmented optimization SLF estimation scheme by taking advantage of deep learning and the traditional RTI technique. Our proposed method achieves the best reconstruction performance among the existing approaches.

In this article, we investigate spectrum allocation in vehicle-to-everything (V2X) network. We first express the V2X network into a graph, where each vehicle-to-vehicle (V2V) link is a node in the graph. We apply a graph neural network (GNN) to learn the low-dimensional feature of each node based on the graph information. According to the learned feature, multi-agent reinforcement learning (RL) is used to make spectrum allocation. Deep Q-network is utilized to learn to optimize the sum capacity of the V2X network. Simulation results show that the proposed allocation scheme can achieve near-optimal performance.

High power consumption and expensive hardware are two bottlenecks for practical massive multiple-input multiple-output (mMIMO) systems. One promising solution is to employ low-resolution analog-to-digital converters (ADCs) and digital-to-analog converters (DACs). In this paper, we consider a general multipair mMIMO relaying system with a mixed-ADC/DAC architecture, in which some antennas are connected to low-resolution ADCs/DACs, while the rest of the antennas are connected to high-resolution ADCs/DACs. Leveraging on the additive quantization noise model, both exact and approximate closed-form expressions for the achievable rate are derived. It is shown that the achievable rate can approach the unquantized one by using only 2-3 bits of resolutions. Moreover, a power scaling law is presented to reveal that the transmit power can be scaled down inversely proportional to the number of antennas at the relay. We further propose an efficient power allocation scheme by solving a complementary geometric programming problem. In addition, a tradeoff between the achievable rate and power consumption for different numbers of low-resolution ADCs/DACs is investigated by deriving the energy efficiency. Our results reveal that the large antenna array can be exploited to enable the mixed-ADC/DAC architecture, which significantly reduces the power consumption and hardware cost for practical mMIMO systems.

Channel state information (CSI) plays a vital role in wireless communication systems. However, the CSI acquisition overhead is an enormous obstacle to realize the system performance improvements promised by massive connectivity and massive multiple-input-multiple-output (MIMO). To alleviate this overhead, this paper proposes a remote channel inference framework by probing the channels occupied by a source base station (BS) and inferring the channels of target BSs at geographically separated sites. The work generalizes existing literature which mainly focuses on utilizing the CSI linear correlations of adjacent antennas, by adopting a model-free deep learning framework to investigate non-linear dependence among remote CSI. The existence of such cross-BS CSI dependence is first shown by calculating the mutual information between remote channels, and the Cramér-Rao lower bound of remote CSI inference performance based on a one-ring channel model. Inspired by this finding, modern deep learning approaches are leveraged to perform remote channel inference in heterogeneous networks for both single user and multi-user scenarios. The simulation results based on ray tracing data show evident performance advantages over conventional methods, under both homogeneous and heterogeneous frequency coverage. The proposed framework achieves beamformer inference accuracy within 4.6% of the genie-aided optimum at the cost of sweeping only two beams.

Channel state information (CSI) is of vital importance in wireless communication systems. Existing CSI acquisition methods usually rely on pilot transmissions, and geographically separated base stations (BSs) with non-correlated CSI need to be assigned with orthogonal pilots which occupy excessive system resources. Our previous work adopts a data-driven deep learning based approach which leverages the CSI at a local BS to infer the CSI remotely, however the relevance of CSI between separated BSs is not specified explicitly. In this paper, we exploit a model-based methodology to derive the Cramer-Ran lower bound (CRLB) of remote CSI inference given the local CSI. Although the model is simplified, the derived CRLB explicitly illustrates the relationship between the inference performance and several key system parameters, e.g., terminal distance and antenna array size. In particular, it shows that by leveraging multiple local BSs, the inference error exhibits a larger power-law decay rate (w.r.t. number of antennas), compared with a single local BS; this explains and validates our findings in evaluating the deep-neural-network-based (DNN-based) CSI inference. We further improve on the DNN-based method by employing dropout and deeper networks, and show an inference performance of approximately 90% accuracy in a realistic scenario with CSI generated by a ray-tracing simulator.

The practical deployment of massive multiple-input multiple-output (MIMO) in the future fifth generation (5G) wireless communication systems is challenging due to its high-hardware cost and power consumption. One promising solution to address this challenge is to adopt the low-resolution analog-to-digital converter (ADC) architecture. However, the practical implementation of such architecture is challenging due to the required complex signal processing to compensate the coarse quantization caused by low-resolution ADCs. Therefore, few high-resolution ADCs are reserved in the recently proposed mixed-ADC architecture to enable low-complexity transceiver algorithms. In contrast to previous works over Rayleigh fading channels, we investigate the performance of mixed-ADC massive MIMO systems over the Rician fading channel, which is more general for the 5G scenarios like Internet of Things. Specially, novel closed-form approximate expressions for the uplink achievable rate are derived for both cases of perfect and imperfect channel state information (CSI). With the increasing Rician K-factor, the derived results show that the achievable rate will converge to a fixed value. We also obtain the power-scaling law that the transmit power of each user can be scaled down proportionally to the inverse of the number of base station (BS) antennas for both perfect and imperfect CSI. Moreover, we reveal the tradeoff between the achievable rate and the energy efficiency with respect to key system parameters, including the quantization bits, number of BS antennas, Rician K-factor, user transmit power, and CSI quality. Finally, numerical results are provided to show that the mixed-ADC architecture can achieve a better energy-rate tradeoff compared with the ideal infinite-resolution and low-resolution ADC architectures.