Transformer models have become a cornerstone of various natural language processing (NLP) tasks. However, the substantial computational overhead during the inference remains a significant challenge, limiting their deployment in practical applications. In this study, we address this challenge by minimizing the inference overhead in transformer models using the controlling element on artificial intelligence (AI) accelerators. Our work is anchored by four key contributions. First, we conduct a comprehensive analysis of the overhead composition within the transformer inference process, identifying the primary bottlenecks. Second, we leverage the management processing element (MPE) of the Shenwei AI (SWAI) accelerator, implementing a three-tier scheduling framework that significantly reduces the number of host-device launches to approximately 1/10 000 of the original PyTorch-GPU setup. Third, we introduce a zero-copy memory management technique using segment-page fusion, which significantly reduces memory access latency and improves overall inference efficiency. Finally, we develop a fast model loading method that eliminates redundant computations during model verification and initialization, reducing the total loading time for large models from 22 128.31 ms to 1041.72 ms. Our contributions significantly enhance the optimization of transformer models, enabling more efficient and expedited inference processes on AI accelerators.

Machine learning (ML) offers a promising solution to pathloss prediction. However, its effectiveness can be degraded by the limited availability of data. To alleviate these challenges, this paper introduces a novel simulation-enhanced data augmentation method for machine learning (ML) pathloss prediction. Our method integrates synthetic data generated from a cellular coverage simulator and independently collected real-world datasets. These datasets were collected through an extensive measurement campaign in different environments, including farms, hilly ter-rains, and residential areas. This comprehensive data collection provides vital ground truth for model training. A set of channel features was engineered, including geographical attributes derived from LiDAR datasets. These features were then used to train our prediction model, incorporating the highly efficient and robust gradient boosting ML algorithm, CatBoost. The integration of synthetic data, as demonstrated in our study, significantly improves the generalizability of the model in different environments, achieving a remarkable improvement of approximately 12 dB in terms of mean absolute error for the best-case scenario. Moreover, our analysis reveals that even a small fraction of measurements added to the simulation training set, with proper data balance, can significantly enhance the model's performance.

At the Indiana Department of Transportation, work orders are pivotal for payroll accounting, equipment utilization, and resource allocation in winter road maintenance operations. However, efficient management is hindered by the manual generation of work orders for a vast fleet of over 1000 snow plows and the associated personnel. Existing telematics research often focuses on small-scale short-term scenarios, overlooking the extended analysis required for large fleet management. Challenges in data acquisition and the unique nature of winter road maintenance further complicate the situation. This paper addresses these issues and underscores the urgent need for work order automation in winter road maintenance. It introduces work order verification and generation algorithms for operation information extraction from GPS tracks, with improved details and accuracy compared to current manual processes. The paper also demonstrates two open-source proof-of-concept programs-a Matlab implementation for algorithm development and a user-friendly web app-with real-life state-scale examples, highlighting the tangible benefits of the proposed solution.

On-demand deployment of mobile communication infrastructure has emerged as a promising solution for extending cellular coverage in rural areas. However, most current research focuses on theoretically optimizing the trajectory of unmanned aerial vehicle (UAV) relays/base stations in simplified geographic scenarios. These network-side attempts have failed to provide low-cost, practical solutions to remove today’s digital gap. This paper proposes a large-scale simulation methodology based on real-life, high-precision geographic data. With its help, we provide a user-centric approach to coverage extension using follow-me relays. We focus on the rural case exemplified by Indiana, U.S., and present quantitative coverage analyses via channel simulation. Our results show that one UAV relay added to follow the user of interest at 10m high can effectively bring over 20% more area into coverage from the user’s point of view. With the relay added at 50m, the most challenging channel the network has to deal with to guarantee 90% area-wise coverage will experience a 20dB path loss reduction. These site-specific analyses can also be applied to future wireless network planning problems, including network-side equipment deployment simulation and optimization for millimeter-wave and terahertz communications.

This paper details the design of an autonomous alignment and tracking platform to mechanically steer directional horn antennas in a sliding correlator channel sounder setup for 28 GHz V2X propagation modeling. A pan-and-tilt subsystem facilitates uninhibited rotational mobility along the yaw and pitch axes, driven by open-loop servo units and orchestrated via inertial motion controllers. A geo-positioning subsystem augmented in accuracy by real-time kinematics enables navigation events to be shared between a transmitter and receiver over an Apache Kafka messaging middleware framework with fault tolerance. Herein, our system demonstrates a 3D geo-positioning accuracy of 17 cm, an average principal axes positioning accuracy of 1.1°, and an average tracking response time of 27.8 ms. Crucially, fully autonomous antenna alignment and tracking facilitates continuous series of measurements, a unique yet critical necessity for millimeter wave channel modeling in vehicular networks. The power-delay profiles, collected along routes spanning urban and suburban neighborhoods on the NSF POWDER testbed, are used in pathloss evaluations involving the 3GPP TR38.901 and ITU-R M.2135 standards. Empirically, we demonstrate that these models fail to accurately capture the 28 GHz pathloss behavior in urban foliage and suburban radio environments. In addition to RMS direction-spread analyses for angles-of-arrival via the SAGE algorithm, we perform signal decoherence studies wherein we derive exponential models for the spatial/angular autocorrelation coefficient under distance and alignment effects.

The high cost required by traditional measurement campaigns often limits the amount of data that can be obtained, to the detriment of data-intensive modeling techniques such as machine learning. This work addresses the limitations from the measurement system and environment by changing the traditional channel modeling approach. Simulation was used as an auxiliary means to obtain data, showing the broader applicability of a site-specific model. More specifically, we explore the possibility of augmenting channel measurements with simulation predictions to acquire comprehensive sets of mm-wave channel information for improved modeling. Path loss measurements from a 28-GHz campaign in a coniferous forest were utilized in conjunction with semi-empirical statistical ray tracing simulations to evaluate the performance of measurement-based channel models beyond the specific measurement region from which they were developed. The root-mean-square deviations between model predictions and simulation results are 11.3 dB for an ITU woodland model and 6.8 dB for a site-specific model we published in a previous manuscript. Furthermore, the site-specific model was demonstrated to agree with simulation predictions at distances and locations we were unable to measure. These results show a broad applicability of our site-specific model as well as a mechanism to derive accurate models from a combination of measurement and simulation data.

With the rapid popularity of unmanned aerial vehicles (UAVs, also known as drones), UAV data relay has demonstrated potential extending wireless communication coverage, especially for rural areas. The flexibility of this approach has attracted research attention from a variety of areas, including Internet of Things, intelligent transportation systems, and digital agriculture. However, most current research effort focuses on modeling and theoretically optimizing data relay systems via UAV trajectories in simplified geographic environments, while taking advantage of UAVs for practical wireless communication networks requires large-scale quantitative performance analysis results based on real-life environment information. In this paper, we propose algorithms for generating large-scale blockage and path loss maps via terrain-based channel modeling for cellular communication systems with fixed-height relay drones. Our analyses reveal the coverage ratios for Tippecanoe County and the Wabash Heartland Innovation Network region in Indiana, with relay drones simulated at different heights. A coverage ratio gain over 40% can be achieved at a drone height of 100 m, compared to a typical pedestrian height of 1.5 m. These site-specific analyses are important in locating poorly covered spots and quantifying the coverage improvement from UAV data relay.

This paper presents millimeter wave propagation measurements at 28 GHz for a typical suburban environment using a 400-megachip-per-second custom- designed broadband sliding correlator channel sounder and highly directional 22-dBi (15° half-power beamwidth) horn antennas. With a 23-dBm transmitter installed at a height of 27m to emulate a microcell deployment, the receiver obtained more than 5000 power delay profiles over distances from 80m to 1000m at 50 individual sites and on two pedestrian paths. The resulting basic transmission losses were compared with predictions of the over-rooftop model in recommendation ITU-R P.1411-9. Our analysis reveals that the traditional channel modeling approach may be insufficient to deal with the varying site-specific propagations of millimeter waves in suburban environments. For line-of-sight measurements, the path loss exponents obtained for the close-in (CI) free space reference distance model and the alpha-beta-gamma (ABG) model are 2.00 and 2.81, respectively, which are close to the recommended site-general value of 2.29. The root mean square errors (RMSEs) for these two reference models are 9.93dB and 9.70dB, respectively, which are slightly lower than that for the ITU site-general model (10.34dB). For non-line-of-sight measurements, both reference models, with the resulting path loss exponents of 2.50 for the CI model and 1.12 for the ABG model, outperformed the site-specific ITU model by around 14dB RMSE.