DON26TZ01-NV009 TITLE: Robust Universal Adaptive Denoising Technology

OUSW (R&E) CRITICAL TECHNOLOGY AREA(S): Applied Artificial Intelligence (AAI)

COMPONENT TECHNOLOGY PRIORITY AREA(S): Advanced Computing and Software;Trusted AI and Autonomy

PROJECTED CMMC LEVEL REQUIREMENT: Level 2 (Self)

OBJECTIVE: Develop robust denoising approaches that are highly adaptive and effective.

DESCRIPTION: Signal denoising has shown to be highly effective in improving performance of signal processing radio frequency and acoustic sensing systems. The main hindering signal in these applications is noise as it degrades the ability to sense low level signals masked by ambient noise sources which may be external to the sensor or generated by the sensor itself. The main goal of this SBIR topic is to develop a denoising technology that suppresses noise while preserving the underlying signal features. Traditionally, denoising methods have struggled to maintain performance when presented with highly non-stationary or complex noise patterns. The traditional approaches typically require extensive and time-consuming tuning to achieve desired performance. On the other hand many of learning-based methods have demonstrated excellent denoising performance but suffer from limited robustness. Therefore, the method’s performance will drop if the training conditions do not adequately reflect the characteristics of the operational environment. The Navy seeks improvements in denoising performance greater than 10 dB.

For such a system installed on an aircraft, it will experience both wind- and aircraft-generated noise. That noise has components that are narrow band (< 10 Hz wide) and broadband (10s to 100s of Hz wide). The spectrum of interest for sensing extends from approximately 10 Hz to 1000 Hz. When compared with more traditional active noise cancellation techniques, the denoising approach should be capable of providing 6 dB of additional cancellation and show potential to deliver 10 dB or more cancellation.

PHASE I: Develop concepts for a robust denoising approach requiring minimal training and are effective in highly non-stationary or complex noise environments. Modeling and simulation to include laboratory measurements to assess the efficacy of the approach based on an in-air or ground-based mobile acoustic sensing system is desired. Consider how the approach may be extended to a radio frequency (RF) system operating in the 1-10 GHz range. The Phase I effort will include prototype plans to be developed under Phase II.

PHASE II: Develop and demonstrate an end-to-end denoising approach on an acoustic frequency sensing system in a laboratory environment and ultimately in a representative operational environment. The prototype assessment should include narrow and broadband noise removal performance while preserving desired signal characteristics, robustness in the presence of non-stationary noise environments. At least 10 dB of noise cancellation is needed with 15 dB desired over traditional active noise cancellation techniques. Consideration for the ease of integration and fielding should be made. Demonstrating the efficacy of the denoising approach on a variety of host platforms is desired. Further refine the extension of the denoising technique to use by RF sensing systems.

PHASE III DUAL USE APPLICATIONS: Support the transition to Navy use.

A universal highly adaptive denoising approach could find applications across remote sensing, communication systems, biomedical signal processing, audio restoration, and image enhancement.

REFERENCES:

1. Sertdal, Peter; Wagner, Simon; Martin, Rainer; and Bruggenwirth, Stefan. "Radar Signal Denoising for ISAR Imaging Using Complex-valued Neural Network." 2025 IEEE International Radar Conference (RADAR), Rennes, France, 2024, pp. 1-6. https://ieeexplore.ieee.org/document/10993972
2. Cao, Linxuan et al. "An Underwater Acoustic Signal Denoising Method for Broadband Interference from Underwater Platforms Based on the Deep Neural Network Model." 2023 6th International Conference on Information Communication and Signal Processing (ICICSP), Xi'an, China, 2023, pp. 439-444. https://ieeexplore.ieee.org/document/10390690

KEYWORDS: Denoising; Signal Processing; Deep Learning; Non-Stationary; Acoustic; Radio Frequency

TPOC 1
Thomas Kreppel
thomas.j.kreppel.civ@us.navy.mil

TPOC 2
Robert Carpenter
robert.l.carpenter.civ@us.navy.mil