The Journal of the Acoustical Society of Korea (한국음향학회지)

The Acoustical Society of Korea (한국음향학회)
권호 표지
DOI QR코드

DOI QR Code

A study on the active sonar reverberation suppression method based on non-negative matrix factorization with beta-divergence function

베타-발산 함수를 활용한 비음수 행렬 분해 기반의 능동 소나 잔향 제거 기법에 대한 연구

  • 이석진 (경북대학교 전자공학부) ;
  • 김근환 (세종대학교 해양시스템융합공학과)
  • Received : 2024.04.22
  • Accepted : 2024.07.23
  • Published : 2024.07.31
Download PDF
⟨ Previous Next ⟩

Abstract

To suppress the reverberation in the active sonar system, the non-negative matrix factorization-based reverberation suppression methods have been researched recently. An estimation loss function, which makes the multiplication of basis matrices same as the input signals, has to be considered to design the non-negative matrix factorization methods, but the conventional method simply chooses the Kullback-Leibler divergence asthe lossfunction without any considerations. In this paper, we examined that the Kullback-Leibler divergence is the best lossfunction or there isthe other loss function enhancing the performance. First, we derived a modified reverberation suppression algorithm using the generalized beta-divergence function, which includes the Kullback-Leibler divergence. Then, we performed Monte-Carlo simulations using synthesized reverberation for the modified reverberation suppression method. The results showed that the Kullback-Leibler divergence function (β = 1) has good performances in the high signal-to-reverberation environments, but the intermediate function (β = 1.25) between Kullback-Leibler divergence and Euclidean distance has better performance in the low signal-to-reverberation environments.

능동 소나 시스템에서 잔향을 제거하기 위하여 최근 비음수 행렬 분해 기법을 활용한 잔향 제거 알고리즘이 고안된 바 있다. 비음수 행렬 분해 알고리즘을 설계하기 위해서는 분해된 기저 행렬의 곱이 원본 신호와 같도록 유도하는 추정 비용 함수가 필요한데, 기존의 연구에서는 이에 대한 고찰이 없이 쿨백-라이블러 발산 함수를 활용하였다. 본 논문에서는 쿨백-라이블러 발산 함수의 선택이 좋은 선택이었는지, 혹은 성능을 개선할 수 있는 다른 추정 비용 함수가 있는지 연구하고자 하였다. 이를 위하여, 먼저 쿨백-라이블러 함수를 포함하여 일반화된 베타-발산 함수를 활용하여 수정된 잔향 제거 알고리즘을 제안하였다. 그리고 수정된 잔향 제거 알고리즘에 대해 합성된 잔향 신호를 활용한 몬테-카를로 시뮬레이션을 수행하였다. 그 결과 높은 신호대잔향비 환경에서는 쿨백-라이블러 발산 함수(β= 1)가 좋은 성능을 보이지만, 낮은 신호대잔향비 환경에서는 쿨백-라이블러 발산 함수와 유클리드 거리의 중간 특성을 가지는 함수(β= 1.25)가 더 좋은 성능을 보이는 것을 확인하였다.

Keywords

Acknowledgement

이 논문은 2024년 정부(방위사업청)의 재원으로 국방기술진흥연구소의 지원을 받아 수행된 물리데이터 기반 지능형 소나 신호 탐지 기술 연구임(No.KRIT-CT-22-052, 물리데이터 기반 지능형 소나 신호 탐지 기술 연구).

References

  1. T. Collins and P. Atkins, "Doppler-sensitive active sonar pulse designsfor reverberation processing," IEE Proceedings of Radar, Sonar, and Navigation, 145, 347-353 (1998). https://doi.org/10.1049/ip-rsn:19982434
  2. G. Kim, K. Lee, K. Yoon, and S. Lee, "A study on pulse train waveforms for high duty cycle sonar systems: optimization scheme and relationship between orthogonality and bandwidth," IEEE Access, 9, 119800-119817 (2021). https://doi.org/10.1109/ACCESS.2021.3107907
  3. S. Kay and J. Salisbury, "Improved active sonar detection using autoregressive prewhiteners," J. Acoust. Soc. Am. 87, 1603-1611 (1990). https://doi.org/10.1121/1.399408
  4. G. Ginolhac and G. Jourdain, "Principal component inverse algorithm for detection in the presence of reverberation," IEEE J. Oean. Eng. 27, 310-321 (2002). https://doi.org/10.1109/JOE.2002.1002486
  5. W. Li, Q. Zhang, X. Ma, and C. Hou, "Active sonar detection in reverberation via signalsubspace extraction algorithm," EURASIP J. Wirel. Commun. Netw. 2010, 1-10 (2010). https://doi.org/10.1155/2010/981045
  6. D. D. Lee and H. S. Seung, "Algorithmsfor non-negative matrix factorization," Proc. NeurIPS, 556-562 (2001).
  7. D. D. Lee and H. S. Seung, "Learning the parts of objects by non-negative matrix factorization," Nature, 401, 788-791 (1999). https://doi.org/10.1038/44565
  8. P. Smaragdis and J. C. Brown, "Non-negative matrix factorization for polyphonic music transcription," Proc. IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, 177-180 (2003).
  9. W. Wang, Y. Luo, J. A. Chambers, and S. Sanei, "Non-negative matrix factorization for note onset detection of audio signals," Proc. 16th IEEE Signal Processing Society Workshop on Machine Learning for Signal Processing, 447-452 (2006).
  10. M. Sun, Y. Li, J. F. Gemmeke, and X. Zhang, "Speech enhancement under low SNR conditions via noise estimation using sparse and low-rank NMF with Kullback-Leibler divergence," IEEE/ACM Trans. on Audio, Speech, and Lang. Process. 23, 1233-1242 (2015). https://doi.org/10.1109/TASLP.2015.2427520
  11. A. Ozerov and C. Fevotte, "Multichannel nonnegative matrix factorization in convolutive mixtures for audio source separation," IEEE Trans. on Audio, Speech, and Lang. Process. 18, 55-563 (2009).
  12. S. Lee and J.-S. Lim, "Reverberation suppression using non-negative matrix factorization to detect lowdoppler target with continuous wave active sonar," EURASIP J. Adv. Signal Process. 2019, 1-18 (2019). https://doi.org/10.1186/s13634-018-0596-y
  13. G. Kim, K. Lee, and S. Lee, "Linear frequency modulated reverberation suppression using non-negative matrix factorization methods, dechirping transformation, and modulo operation," IEEE Access, 8, 110720-110737 (2020). https://doi.org/10.1109/ACCESS.2020.3001865
  14. S. Lee and Y. Lee, "Improvement for non-negative matrix factorization-based reverberation suppression for bistatic active sonar" (in Korean), The J. Acoust. Soc. Kr. 41, 468-479 (2022).
  15. A. Marmin, J. H. de Morais Goulart, and C. Fevotte, "Joint majorization-minimization for nonnegative matrix factorization with the beta-divergence," Signal Processing, 209, 109048 (2023).
  16. D. Abaham and A. Lyons, "Simulation of non-rayleigh reverberation and clutter," IEEE J. Oceanic Eng. 29, 347-362 (2004). https://doi.org/10.1109/JOE.2004.828202