IMToolkit: An Open-Source Index Modulation Toolkit for Reproducible Research Based on Massively Parallel Algorithms
IMToolkit: An Open-Source Index Modulation Toolkit for Reproducible Research Based on Massively Parallel Algorithms
复制标题
DOI：
10.1109/access.2019.2928033
复制
发表时间：
2019-07
期刊：
影响因子：
3.9
通讯作者：
Naoki Ishikawa
中科院分区：
文献类型：
--
作者：
Naoki Ishikawa
本文提出了一个开源索引调制（IM）工具箱，它促进了可重复研究，加速了IM研究的开放创新。提出的工具包是基于大规模并行算法实现的，这些算法是为最先进的图形处理单元（gpu）设计的。由于高性能gpu可以以低成本获得，随着深度学习的深入发展，该工具包可以以低成本实现大规模但显着快速的蒙特卡罗模拟。介绍了两种基于大张量的误码率和平均互信息仿真并行算法。此外，将有源索引的设计重新表述为保证最优性的整数线性规划问题，适用于广义空间调制和子载波索引调制方案。性能比较表明，所提出的gpu辅助算法比传统的cpu辅助高效算法快145倍。此外，与广泛使用的传统方法相比，所设计的活性指标达到了理论上的最佳性能。这些设计的活跃指数的综合数据库发布在网上，可供任何研究人员使用。
This paper presents a proposal of an open-source index modulation (IM) toolkit, which facilitates reproducible research and accelerates open innovation in IM studies. The proposed toolkit is implemented based on massively parallel algorithms that are designed for state-of-the-art graphics processing units (GPUs). Since high-performance GPUs are available at low cost, along with the intensive development in deep learning, this toolkit achieves large scale but significantly fast Monte Carlo simulations at low cost. Two large-tensor-based parallel algorithms are introduced for bit error ratio and average mutual information simulations. Additionally, the design of active indices is newly formulated into an integer linear programming problem that guarantees optimality, which is applicable to the generalized spatial modulation and subcarrier-index modulation schemes. Performance comparisons demonstrated that the proposed GPU-aided algorithms were up to 145 times faster than the conventional CPU-aided efficient counterparts. Furthermore, the designed active indices achieved the theoretical optimum performance in contrast to widely used conventional methods. A comprehensive database of these designed active indices is released online and is available to any researcher.