Variable Precision Computing for Real-Time Applications: An Low-Complexity Markov Arithmetic-Level Approach
Kaixuan Bao, Jiehao Miao, Jindan Xu, Wei Xu · 2025
With the emerging applications of communication networks, real-time response and low-latency processing have become critical requirements for new applications under the concept of the Tactile Internet. However, the inherent increase in computational complexity of algorithms poses significant challenges to achieve low-latency computing. In this paper, we introduce a newly designed arithmetic level variable precision computing (VPC) scheme, namely AL-VPC, which dynamically allocates varying levels of computing precision to each arithmetic operation in an algorithm. This approach aims to improve computational efficiency while minimizing accuracy degradation. The proposed scheme establishes a Markov process model to optimize the computing precision allocation, incorporates low-complexity calculations to limit additional overhead. Through numerical simulations across both computing and communication applications demonstrate that the proposed new design outperforms conventional fixed-length computing (FLC) scheme while maintaining the same level of computing precision.