The Open Radio Access Network (O-RAN) represents a significant advancement in cellular networks, promoting openness, intelligence, and flexibility in 5G deployment. However, designing a scheduler for 5G O-RAN presents significant challenges due to its unique network architecture, large scheduling space, and stringent timing requirements of various control loops. Existing model-based and model-free schedulers both have inherent drawbacks that hinder their performance and adoption in 5G O-RAN. Model-based schedulers struggle because accurately modeling wireless system is often impossible, and they usually suffer from high complexity due to the NP-hard problem structure and large scheduling space. On the other hand, model-free schedulers often face convergence issue under a large scheduling space, and they usually cannot provide performance guarantees or even satisfy constraints. In this paper, we present MOGUL—a MOdel-GUided Learning approach that retains the strengths of both model-based and model-free methods while avoiding their pitfalls. MOGUL employs a model-based optimization problem to derive a reduced yet promising scheduling space, which is then used as the action space for model-free online Deep Multi-agent Reinforcement Learning (DMARL) to determine the final scheduling decision. Moreover, MOGUL is specifically tailored to the O-RAN architecture, allowing seamless integration into various control loops while meeting their stringent timing requirements. Experimental results demonstrate that MOGUL outperforms both state-of-the-art model-based and model-free algorithms.

The scarcity of spectrum and the proliferation of data-intensive applications in 5G/Next-G networks call for innovations of new techniques that are capable of offering UE-level data rate guarantee with minimum required spectrum usage. This is a challenging problem due to the complexity of mechanisms involved in the process, such as Resource Block (RB) allocation, modulation and coding scheme (MCS) selection, and MU-MIMO beamforming (BF) design. Further complicating the problem is the random, unknown nature of Channel State Information (CSI) and the errors involved in its estimation. In this paper, we present Rudra, which offers a comprehensive solution to these challenges. Rudra formulates the bandwidth minimization problem by incorporating probabilistic data rate guarantee through a chance constraint, which embeds RB allocation, MCS selection, and MU-MIMO BF mechanisms. The CSI uncertainty problem is addressed through a novel error-embedded (EE)-Wasserstein ambiguity set based on a small set of data samples. We show that the solution by Rudra meets our design objective and outperforms a modified state-of-the-art algorithm.

Synchronous Localization and Mapping (SLAM) and autonomous navigation technology are fundamental to the intelligence and automation of robots and they represent important research directions in intelligent transportation. This paper proposes an enhanced mapping algorithm based on Cyber-Physical Systems (CPS) to address the high precision demands of indoor robot autonomous navigation. The proposed system integrates a Kalman filter to fuse output data from multiple sensors, including a laser odometer, Inertial Measurement Unit (IMU), and wheel odometer, achieving precise multi-sensor fusion for localization. Additionally, the system combines the A$^{\ast }$algorithm with the pure pursuit autonomous navigation algorithm to optimize autonomous navigation performance. Visual information processing incorporates Yolov5 for data augmentation and scene recognition, thereby improving the system’s environmental perception. Experimental validation demonstrates an average mapping accuracy of 0.032 meters and an average navigation accuracy of 0.05 meters. The CPS architecture is highly effective for high-precision mapping and autonomous navigation in unknown and complex indoor environments, demonstrating the ability to improve localization and mapping accuracy in complex indoor environments.

MU-MIMO beamforming is a key technology for 5G networks, relying on Channel State Information (CSI). However, in practice, the estimated CSI in reality is prone to uncertainty. Further, a MU-MIMO beamforming solution must be derived within a millisecond to be useful for real-time 5G applications. We present ReDBeam—a real-time data-driven beamforming solution for MU-MIMO using limited CSI data samples. The main novelties of ReDBeam are a parallel algorithm and an optimized GPU implementation. ReDBeam delivers a MU-MIMO beamforming solution within 1 millisecond to meet the probabilistic data rate requirements from the users, and minimize a base station’s power consumption. Through extensive experiments, we show that ReDBeam consistently meets the stringent 1-millisecond real-time requirement and is orders of magnitude faster than other state-of-the-art algorithms. ReDBeam conclusively demonstrates that MU-MIMO beamforming with data rate requirements can be achieved in real-time using only limited CSI data samples.

Open Radio Access Network (O-RAN) offers a new paradigm for the design and deployment of future RANs. The unique architecture of O-RAN presents two main challenges when designing a scheduler. First, it is impractical to obtain accurate and full Channel State Information (CSI) due to estimation errors and limited bandwidth of the fronthaul link between Open Radio Unit (O-RU) and Open Distributed Unit (O-DU). Second, the large-scale processing at an O-DU introduces difficulties in meeting the stringent time requirement in O-RAN, especially in the real-time (RT) control loop. To address these challenges, we propose R3—a real-time robust Multi-user, Multiple Input, Multiple Output (MU-MIMO) scheduler for O-RAN. R3 serves as a comprehensive scheduling solution encompassing RB allocation, MCS selection, and beamforming calculation. Most notably, R3 utilizes a limited number of CSI samples to offer probabilistic QoS guarantees. To meet the timing requirements of O-RAN, R3 decomposes the scheduling problem into two distinct sub-problems and integrates them into separate control loops. Moreover, each sub-problem is designed with a parallel structure, utilizing a reduced search space, and implemented on a GPU platform to accelerate the computation time. Experimental results demonstrate that R3 offers competitive throughput performance as the state-of-the-art while simultaneously fulfilling the QoS guarantees. Further, R3 meets the timing requirements of various control loops in O-RAN over a wide range of operating conditions.

Pose-guided person image generation aims to transfer reference images to target poses while preserving the source appearance. Recent approaches achieve considerable improvement by using spatial transformation modules such as attention operation. However, the commonly used vanilla attention tends to generate a dense correlation matrix which means that the value of a target position is the weighted sum of many source positions, resulting in blurry appearance. In this paper, we propose a novel model named Flow-guided Attention Deformation (FAD) to perform the spatial transformation. Our model first establishes the correlation between sources and targets with a flow-guided attention operation. Then, with the obtained correlation matrix, we perform an accurate deformation for source features to generate the predicted image. Extensive results demonstrate the superiority of the proposed method, outperforming state-of-the-art methods quantitatively and qualitatively. Ablation studies clarify the efficiency of the proposed modules and verify our hypothesis.

Multi-User (MU)-MIMO technology plays a vital role in 5G NR. Under MU-MIMO transmission, multiple users can share the same time-frequency resources simultaneously. For 5G MU-MIMO systems, it is challenging to design a scheduler. The scheduler needs to determine resource block (RB) allocation, the number of data streams and modulation and coding scheme (MCS) for each user in each transmission time interval (TTI). In particular, multiple users can be co-scheduled on the same RB for MU-MIMO transmission. In addition, it is necessary for the scheduler to find a scheduling solution within each TTI to be useful. In this paper, we present mCore$+$, a novel design and implementation that can achieve$\sim$500$\mu$s timing performance for 5G MU-MIMO systems. mCore$+$is meticulously designed with a multi-phase optimization and heavily leverages large-scale parallel computation. In each phase, mCore$+$either decomposes the optimization problem into a number of independent sub-problems, or reduces the search space into a smaller but most promising subspace, or both. mCore$+$is validated on a commercial-off-the-shelf GPU platform. Experimental results show that mCore$+$can offer a scheduling solution in$\sim$500$\mu$s for up to 100 RBs, 100 users, 29 MCS levels and$4 \times 12$MIMO systems. Also, mCore$+$can achieve better or comparable throughput performance compared to other state-of-the-art algorithms.

MU-MIMO technology enables a base station (BS) to transmit signals to multiple users simultaneously on the same frequency band. It is a key technology for 5G NR to increase the data rate. In 5G specifications, an MU-MIMO scheduler needs to determine RBs allocation and MCS assignment to each user for each TTI. Under MU-MIMO, multiple users may be coscheduled on the same RB and each user may have multiple data streams simultaneously. In addition, the scheduler must meet the stringent real-time requirement (~1 ms) during decision making to be useful. This paper presents mCore, a novel 5G scheduler that can achieve ~1 ms scheduling with joint optimization of RB allocation and MCS assignment to MU-MIMO users. The key idea of mCore is to perform a multi-phase optimization, leveraging large-scale parallel computation. In each phase, mCore either decomposes the optimization problem into a number of independent sub-problems, or reduces the search space into a smaller but most promising subspace, or both. We implement mCore on a commercial-off-the-shelf GPU. Experimental results show that mCore can offer the best scheduling performance for up to 100 RBs, 100 users, 29 MCS levels and 4 × 12 antennas when compared to other state-of-the-art algorithms. It is also the only algorithm that can find its scheduling solution in ~1 ms.

Pose-guided person image synthesis aims to synthesize person images by transforming reference images into target poses. In this paper, we observe that the commonly used spatial transformation blocks have complementary advantages. We propose a novel model by combining the attention operation with the flow-based operation. Our model not only takes the advantage of the attention operation to generate accurate target structures but also uses the flow-based operation to sample realistic source textures. Both objective and subjective experiments demonstrate the superiority of our model. Meanwhile, comprehensive ablation studies verify our hypotheses and show the efficacy of the proposed modules. Besides, additional experiments on the portrait image editing task demonstrate the versatility of the proposed combination.

In vehicular ad-hoc network (VANET), due to its unique characteristics such as high nodes' speed, dynamic network topology and variable nodes' density, end-to-end data transmission faces many challenges. To address these challenges, many routing protocols specially for VANET have been proposed. But a chief part of them are designed only in network layer and independently of other layers. Besides, few of them consider the influence of interference. In this paper, we propose a new cross-layer routing with interference constraint for VANET which combines technologies of network layer and MAC layer. First, we adopt a location-based resource allocation algorithm to constrain interference between vehicles. Then on the basis of it, relay is selected depending on their geographical zones. In addition, we also use the relative distance among vehicles and the reception Signal to Interference plus Noise Ratio (SINR) of hello messages to optimize the relay selection scheme. By simulation, we compare it to other protocols and experimental results show that our routing protocol outperforms existing solutions in terms of packet delivery ratio (PDR).