Time–Frequency Mask-Aware Bidirectional LSTM: A Deep Learning Approach for Underwater Acoustic Signal Separation

Abstract

Underwater acoustic signal separation is a key technique for underwater communications. The existing methods are mostly model-based, and cannot accurately characterize the practical underwater acoustic communication environment. They are only suitable for binary signal separation and cannot handle multivariate signal separation. However, recurrent neural networks (RNNs) show a powerful ability to extract the features of temporal sequences. Inspired by this, in this paper, we present a data-driven approach for underwater acoustic signal separation using deep learning technology. We use a bidirectional long short-term memory (Bi-LSTM) approach to explore the features of a time–frequency (T-F) mask, and propose a T-F-mask-aware Bi-LSTM for signal separation. Taking advantage of the sparseness of the T-F image, the designed Bi-LSTM network is able to extract the discriminative features for separation, which further improves the separation performance. In particular, this method breaks through the limitations of the existing methods and not only achieves good results in multivariate separation but also effectively separates signals when they are mixed with 40 dB Gaussian noise signals. The experimental results show that this method can achieve a 97% guarantee ratio (PSR), and the average similarity coefficient of the multivariate signal separation is stable above 0.8 under high noise conditions. It should be noted that our model can only handle known signals such as test signals for calibration.

Underwater Acoustic Target Recognition in Passive Sonar Using Spectrogram and Modified MobileNet Network Classifier

“…Also, the accuracy of the proposed method for underwater acoustic target recognition compared with other introduced methods in Table 6. Wavelet Transform (Average Power Spectral Density) [10] 92.64% CNN+k-nearest neighbor Wavelet Packets [11] 90.30% Deep Convolutional Neural Network Synthetic Aperture Sonar Imagery [13] 90.89% Support Vector Machine (SVM) Competitive Deep-Belief Networks [14] 93.04% DCGAN + S-ResNet Spectrum Image [15] 83.15% Multi-Scale Residual Unit (MSRU) Spectrogram [16] 90.91% Separable Convolutional Neural Network Waveform [17] 95.22% Convolutional Neural Network Low-Frequency Analysis Recording (LOFAR) [18] 98.52% Support Vector Machine (SVM) micro-Doppler sonar [19] 94.31% ResNet-18 Fusion Features [20] 96.32% ResNet Multi-Window Spectral Analysis (MWSA) [21] 97.69% ResNet and DensNet Spectrogram [22] 94.00% Convolutional Neural Network DEMON and LOFAR [23] 97.00% Bidirectional Short-Term Memory (Bi-LSTM) Time-Frequency Diagrams [24] 96.90% Convolutional Neural Network Acoustic Spectrograms [25] 98 As shown in the table above, compared to the existing methods for performing UATR, the proposed models have high classification accuracy, which can increase the processing speed of target recognition and avoid wasting time in model training calculation operations. In addition, in this research, due to the use of the average integration method at the end of the layers of the convolutional algorithms of the proposed model, instead of the fully connected layer, it has been tried to reduce the complexity and increase calculations.…”

“…Song et al [23] by combining the lowfrequency analysis recording (LOFAR) and Envelope modulation on noise (DEMON) and CNN network have been able to achieve 94.00% recognition accuracy. Chen et al [24] proposed a method based on a bi-directional short-term memory (Bi-LSTM) to discover the features of a time-frequency mask to extract distinctive features of the underwater audio signal. Sheng and Zhu [25] proposed an underwater acoustic target detection method based on a UATR transformer to detect two classes of targets, which can capture global and local information on spectrograms, thereby improving the performance of UATR.…”

Underwater Acoustic Target Recognition Using Spectrogram ROI Approximation with Mobilenet One-dimensional and Two-dimensional Networks

“…(Song, et al, 2022) by combining the low-frequency analysis recording (LOFAR) and Envelope modulation on noise (DEMON) and CNN network have been able to achieve 94.00% recognition accuracy. (Chen, et al, 2022) proposed a method based on a bi-directional short-term memory (Bi-LSTM) to discover the features of a time-frequency mask to extract distinctive features of the underwater audio signal. (Sheng & Zhu, 2023) proposed an underwater acoustic target detection method based on a UATR transformer to detect two classes of targets, which can capture global and local information on spectrograms, thereby improving the performance of UATR.…”

A Wasserstein generative adversarial network with gradient penalty for active sonar signal reverberation suppression

“…With the development and rise of artificial intelligence in recent years, algorithms combining artificial intelligence with antireverberation technology continue to surge, such as support vector machines, CNN (Song et al, 2019), RNN (Chen et al, 2022), and GAN In the beginning, it was simply a simple addition to machine learning. For example, Zhu et al designed a feature kernel function SVM based on the non-Gaussian difference between reverberation and target echo to detect the signal in the reverberation background.…”