Article Details

Title · Authors · Abstract · Keywords · DOI · References

Research on Intelligent Generation Technology of VSIPL Code Translation Samples for Embedded Signal Processing

LI Weichao, HAN Wenjun, LI Luye, LI He, DING Linlin

(Nanjing Research Institute of Electronic Technology, Nanjing 210039, China)

Section：Computer Technology · Vol. 2026, Issue 14 · Pages：102-105,110 · 2096-4706（2026）14-0102-05 · CLC number：TP311.1；TP18 · Document code：A · DOI：10.19850/j.cnki.2096-4706.2026.14.020

Abstract

To address the scarcity of MATLAB-VSIPL parallel corpora and the high cost of manual annotation in the field of embedded signal processing, this paper proposes an LLM-driven intelligent generation method for MATLAB-VSIPL code translation samples. By designing structured translation prompt templates that integrate chain-of-thought reasoning with VSIPL domain knowledge, the method guides the model to generate MATLAB function seeds and corresponding VSIPL candidate implementations. Candidate samples then undergo progressive validation through static compilation checking and dynamic consistency testing. For samples failing validation, compilation error messages and test discrepancy feedback are leveraged to guide iterative repair by the model, thereby establishing a closed-loop MATLAB-VSIPL sample production line for generation, validation, and iteration. Finally, a MinHash-based semantic deduplication strategy is employed to eliminate redundant samples. Experimental results show that this method improves the VSIPL code test pass rate from 4% to 55%, significantly enhancing the quality of MATLAB-VSIPL sample generation.

Keywords

Large Language Model; code translation; VSIPL; chain-of-thought; sample generation

References

  1. [1] formal/14-08-01 VSIPL Specification Version 1.5 [S/OL].https://www.omg.org/spec/VSIPL/1.5/About-VSIPL.
  2. [2] Fan Chenbing, Zhu Yupeng, Pang Shuang Radar Signal Sorting and Recognition System Based on CPU+GPU Parallel Architecture [J/OL]. Telecommunications Technology, 2025:1-11 https://doi.org/10.20079/j.issn.1001-893x.250512002.
  3. [3] Deng Zhusha Software Design of Cross Platform Algorithm Middleware Based on VSIPL [J]. Telecommunications Technology, 2013, 53 (6): 773-776
  4. [4] Dai Jian, Zhang Chen, Xu Chaoyang VSIPL function library for optimizing radar signal processing [J]. Ship Electronic Warfare, 2020, 43 (5): 79-81
  5. [5] Deep search for artificial intelligence research institute DeepSeeker R1 Large Model Technology White Paper [R]. 2025
  6. [6] Feng Z Y，Guo D Y，Tang D Y，et al. CodeBERT: A Pre-Trained Model for Programming and Natural Languages [J/OL].arXiv:2002.08155 [cs.CL].https://doi.org/10.48550/arXiv.2002.08155.
  7. [7] Yu S W，Wang T，Wang J. Data Augmentation by Program Transformation [J/OL].Journal of Systems and Software，2022，190：111304.https://doi.org/10.1016/j.jss.2022.111304.
  8. [8] Long L，Wang R，XIAO R X，et al. On LLMs-Driven Synthetic Data Generation, Curation, and Evaluation: A Survey [J/OL].arXiv:2406.15126 [cs.CL].https://doi.org/10.48550/arXiv.2406.15126.
  9. [9] Du Jiale, Chen Shudong, Ye Liang, etc Overview of Thinking Chain Techniques in Large Language Models [J]. Radio Communication Technology, 2025, 51 (5): 877-887
  10. [10] Wu W，Li B，Chen L，et al.A Review for Weighted MinHash Algorithms [C]//2023 IEEE 39th International Conference on Data Engineering (ICDE)，2023：3785-3786.

About the authors

Li Weichao (1988-), male, Han ethnicity, from Zhengzhou, Henan Province, China, engineer, doctoral student, research direction: intelligent development of embedded software.