Satellite-assisted internet of things (S-IoT), especially the S-IoT based on low earth orbit (LEO) satellite, plays an important role in future wireless systems. However, the limited on-board communication and computing resource and high mobility of LEO satellites make it hard to provide satisfied service for IoT users. To maximize the task completion rate under latency constraints, collaborative computing and resource allocation among LEO networks are jointly investigated in this paper, and the joint task offloading, scheduling, and resource allocation is formulated as a dynamic mixed-integer problem. To tack the complex problem, we decouple it into two subproblems with low complexity. First, the max-min fairness is adopted to minimize the maximum latency via optimal resource allocation with fixed task assignment. Then, the joint task offloading and scheduling is formulated as a Markov decision process with optimal communication and computing resource allocation, and deep reinforcement learning is utilized to obtain long-term benefits. Simulation results show that the proposed scheme has superior performance compared with other referred schemes.

No abstract

Summary Hybrid geosynchronous earth orbit (GEO) and low earth orbit (LEO) satellite networks play an important role in future satellite‐assisted internet of things (S‐IoT). However, the limited satellite on‐board communication and computing resource poses a large challenge for the task offloading in the hybrid GEO‐LEO satellite networks. In this paper, the problem of task offloading is formulated as a cooperative user association and resource allocation problem. To tackle the problem, we model it as a Markov decision process and decompose it into two sub‐problems, which are sequential decisions for user association and resource allocation with fixed user association conditions. Deep reinforcement learning (DRL) is adopted to make sequential decisions to achieve long‐term benefits, and convex optimization method is utilized to obtain optimal communication and computing resource allocation. Simulation results show that the proposed method is superior to other referred schemes.

Total focusing method (TFM) has improved resolution and accuracy over traditional ultrasonic phased array technology. In this paper, an advanced parallel architecture in field programmable gate arrays is suggested to significantly accelerate the imaging efficiency of TFM. Several techniques are investigated, including the real-time concurrent calculation for time of flight, parallel generation of multiple pixels, and the Hilbert transform to the pixels array. This architecture achieves the real-time computation of the flight times for each pixel and the concurrent generation of double pixels for TFM imaging. Compared to conventional methods, the efficiency of TFM imaging is greatly accelerated and the impact from the increase of element and pixel number is also effectively reduced. Simulation data was used to verify the architecture, and experiment results confirmed that the efficiency was only related to the pulse repeated frequency and element number, which reaches to the physical limitation of TFM inspection. This approach also shows that high efficiency is maintained when pixel number increases, and a strict real-time imaging can be achieved in this architecture. As a result, an effective way for the fast inspection with TFM is provided.

No abstract