Distributed congestion control (DCC) improves system performance by lowering channel congestion in vehicular environments with high vehicle density. The 3rd Generation Partnership Project standard defines the related metrics of channel busy ratio (CBR) and introduces possible rate and power control mechanisms to mitigate channel congestion in cellular vehicle-to-everything (C-V2X) sidelink. However, the DCC of C-V2X is not sufficiently specified to implement these controls. In this letter, we propose a novel DCC algorithm based on deep reinforcement learning (DRL) to improve congestion control performance in C-V2X sidelink. The proposed algorithm allows the DRL agent to observe a CBR state and select the packet transmission rate that can maximize the reward of packet delivery rate (PDR) while maintaining higher channel utilization. Simulation results show that the proposed algorithm provides performance gain in terms of PDR and sidelink throughput compared with the existing DCC method.

Recently, research into autonomous driving and traffic safety has been drawing a great deal of attention. To realize autonomous driving and solve traffic safety problems, wireless access in vehicular environments (WAVE) technology has been developed, and IEEE 802.11p defines the physical (PHY) layer and medium access control (MAC) layer in the WAVE standard. However, the IEEE 802.11p frame structure, which has low pilot density, makes it difficult to predict the properties of wireless channels in a vehicular environment with high vehicle speeds; thus, the performance of the system is degraded in realistic vehicular environments. The motivation for this paper is to improve the channel estimation and tracking performance without changing the IEEE 802.11p frame structure. Therefore, we propose a channel estimation technique that can perform well over the entire SNR range of values by changing the method of channel estimation accordingly. The proposed scheme selectively uses two channel estimation schemes, each with outstanding performance for either high-SNR or low-SNR signals. To implement this, an adaptation algorithm based on a preamble is proposed. The preamble is a signal known to the transmitter–receiver, so that the receiver can obtain channel estimates without demapping errors, evaluating performance of the channel estimation schemes. Simulation results comparing the proposed method to other schemes demonstrate that the proposed scheme can selectively switch between the two schemes to improve overall performance.

Wireless access in vehicular environments to support wireless communication between vehicles has been developed to provide road safety and infotainment services. In vehicular environments where the channel changes rapidly, channel estimation is very important in improving the reliability of wireless communication. Therefore, numerous channel estimation schemes have been proposed; however, none of the schemes proposed so far can perform well over the entire signal-to-noise ratio (SNR) region. In this paper, we propose a novel channel estimation scheme that selectively uses the better scheme between two channel estimation schemes on a symbol-by-symbol basis. The results show that the proposed scheme performs symbol-by-symbol selection of the better channel estimation scheme within a packet, and thus shows excellent performance over the entire SNR region in vehicular environments in terms of the bit error rate and packet error rate.

본 논문에서는 전술 데이터 링크 표준인 Link-16의 정밀 시간 동기 과정에서 발생하는 JU(Joint Tactical Information Distribution System Unit)들 간의 충돌 확률을 분석하고 안정적인 통신을 위해 JU가 각 타임 슬롯에서 정밀 시간 동기를 시도할 확률과 전체 JU의 수에 따른 정밀 시간 동기 성공 확률을 계산한다. JU가 정밀 시간 동기에 성공할 확률, 여러 차례 정밀 시간 동기를 시도하였을 때 한 번 이상 성공할 확률, 그리고 모든 JU가 각각 일정 횟수 이내에 정밀 시간 동기를 성공할 확률을 유도하고, 유의수준 α에서 정밀 동기 성공확률이 1- α 이상이 되기 위해 필요한 파라미터 조건을 계산한다. 유도된 확률식을 컴퓨터 모의실험을 통해 검증하고 타당성을 입증하여, Link-16의 정밀 시간 동기 시스템의 안정성과 효율성을 최적화할 수 있는 정밀시간 동기 충돌 확률 분석 방법을 제안한다.

The tremendous recent progress in analyzing the training dynamics of overparameterized neural networks has primarily focused on wide networks and therefore does not sufficiently address the role of depth in deep learning. In this work, we present the first trainability guarantee of infinitely deep but narrow neural networks. We study the infinite-depth limit of a multilayer perceptron (MLP) with a specific initialization and establish a trainability guarantee using the NTK theory. We then extend the analysis to an infinitely deep convolutional neural network (CNN) and perform brief experiments.