We present FedKit, a federated learning (FL) system tailored for cross-platform FL research on Android and iOS devices. FedKit pipelines cross-platform FL development by enabling model conversion, hardware-accelerated training, and cross-platform model aggregation. Our FL workflow supports flexible machine learning operations (MLOps) in production, facilitating continuous model delivery and training. We have deployed FedKit in a real-world use case for health data analysis on university campuses, demonstrating its effectiveness. FedKit is open-source at https://github.com/FedCampus/FedKit.

Over-the-air (OTA) validation is an important step before deploying new techniques in real world for NextG (5G, 6G and beyond) and wireless Internet of Things (IoT). However, most existing NextG and IoT testbeds are limited in scenario diversity, availability of computing and hardware resources, as well the flexibility to access and operate the resources. To alleviate these limitations, in this work we introduce UnionLabs, an AWS-based unified platform for remote access and sharing of OTA testbeds for NextG and IoT. We will demonstrate UnionLabs by i) federating six heterogeneous wireless testbeds deployed at University at Buffalo and the University of Utah, spanning ground, underwater and underground domains; ii) scheduling and conducting real-time OTA experiments over these testbeds, and iii) engaging with public repositories for user-generated code and datasets. Through UnionLabs, we aim to create a grassroots campaign to democratize access to wireless research testbeds with heterogeneous hardware resources and network environments.

Addressing network fault tolerance tasks with various scenarios and domains is a crucial step towards achieving autonomous networks. Despite the abundance of artificial intelligence models designed for specific network scenarios and fault tasks, they may encounter challenges in delivering optimal performance across all environments. Considering the exhibited exceptional abilities of large language models (LLMs) in content generation and task planning, we advocate that LLMs can act as automated schedulers to manage existing fault tolerance models. Through the current network scenarios and task prompts, appropriate models are invoked to orchestrate optimization strategies. By leveraging pre-optimized knowledge and tools, we propose an LLM-based edge network fault-tolerant paradigm (LLM-ENFT), which autonomously manages the full cycle of perceiving, diagnosing, and recovering network faults. Experiments conducted on the edge network testbed based on Mininet and Ryu show that LLM-ENFT exhibits strong resilience in the face of network congestion and link failure.

Future wireless sensing applications heavily depend on the analysis of wide-band channel state information (CSI) acquired from extremely dense device deployments. However, a gap exists in practical evaluations within existing literature. To bridge this gap, we introduce SensingWall, a highly adaptable, cost-effective solution leveraging IEEE 802.11 and the readily available ESP32 platform. SensingWall simplifies the process for application designers by abstracting away intricate technical details, offering visualization tools, and enabling direct access to a centralized database housing real-time CSI data. Our demonstration includes a functional prototype showcasing two key features. Firstly, real-time visualization of CSI time-series data from both uplink and downlink. Secondly, a demonstration of custom CSI retrieval through SQL-like query operations.

Many molecular communication (MC) prototypes documented in the existing literature necessitate intrusion into the communication channel during the signal detection process, potentially posing harm to the organism. Addressing this concern, we present a non-invasive MC testbed in this demonstration employing the capacitive sensing mechanism to capture chemical signals. The chemical substances utilized in our testbed are soybean oil and water, exhibiting distinct permittivity characteristics. To the best of our knowledge, our testbed has achieved the highest attainable data rate, reaching up to 7.14 bit/s, surpassing other non-invasive MC platforms.

Neither inaccurate empirical methods nor time consuming simulations would meet the growing demands of providing fast optimization for the Robotic Mobile Fulfilment System (RMFS). In this demonstration, we introduce a rapid yet accurate consultation tool for tailored RMFS, leveraging queueing theory to expedite steady-state performance prediction and optimize resource specifications to such interconnected automation processes. We construct a novel shared token multi-class semi-open queueing network based on multiple stages of the system's operation processes which encompasses various order properties. Optimal system parameter combinations including the number of required robots, the number of picking stations, the number of charging station can be instantly computed. Numerical results benchmarked against simulation-based method demonstrates that the estimated steady-state performance derived from the queueing network achieves over 95% accuracy. Additionally, computation time is reduced by more than 10,000 times in the testing scenarios.

In this paper, we design and implement a body-aware advertising signage system, called BodyAd, with activity-specific interaction based on Internet of Things (IoT) technologies. In the BodyAd system, an IoT signage is adopted to transmit multimedia contents to smartphone users on demand in a click-and-drag manner. In addition, approaching audiences can be sensed and interacted with the IoT signage using image sensors. Furthermore, deep learning techniques are explored to recognize the activities of audiences and synchronize to the on-screen doll on the IoT signage. In particular, commodity dances/advertising movements can be learned and played by audiences in front of the IoT signage for getting attention/bonuses. Moreover, based on the skeleton of body components (e.g., hands, arms, legs, etc.) constructed by BodyAd, the audience's body activities can be recognized and applied to correct his/her posture in learning the commodity dance/dvertising movement. This paper demonstrates our current prototype consisting of the smartphone App, BodyAd server, and IoT signage.

Channel knowledge map (CKM) is fundamental for mobile communication system design, optimization, and performance evaluation. In this paper, an unmanned aerial vehicle (UAV) -based real-time channel knowledge mapping system including the aerial platform and mobile ground terminal is designed and implemented. We propose a field programmable gate array (FPGA) -based data processing algorithm to extract the measured channel impulse responses (CIRs) and calculate the channel knowledge labeled by GPS location information. It can run in real-time and greatly reduces the processing complexity, time consumption, and data storage size. Meanwhile, we recover the channel knowledge of the positions other than the sampling trajectory via data completion. Finally, the whole CKM in the measured region is constructed. The developed mapping system can be used to measure and visualize the mapping between the channel characteristics and the three-dimensional environment. The constructed map is expected to benefit diverse applications such as network optimization, physical layer algorithms, link budget, signal coverage, node placement, etc.

With the evolution towards 5G-A and 6G, cloud-native cellular network orchestration is expected to play a key role in network performance optimization, yet systems research in this domain is still nascent. With a view to furthering the conversation on cellular orchestration, through this paper, we introduce ORCA, a comprehensive demo on the cloud-native orchestration and automation of end-to-end cellular network functions and slices. Set against the backdrop of a lab-scale, over-the-air experimental testbed, key highlights of ORCA include Nephio-driven RAN and core orchestration, automated network slice creation and deployment, and O-RAN-driven closed loop slice assurance.

Satellite networks, as an important part of future space-ground integrated network, face problems such as rapid changes in network topology as well as complex and variable service requirements. Intent-driven networks are a promising new network paradigm for addressing the aforementioned issues. Therefore, we propose an intent-based reliable routing algorithm that considers the remaining bandwidth of the link, the variation of the link's acceptance of services with load, and the user's service preference for the link. We also developed a related system for simulation experiments to achieve network-centric and usercentric trade-offs and provide reliable routing guarantees for users.