Two-Layer Decision Model Based on Noise Classification

Acoustic bird species classification under low SNR and small-scale dataset conditions

This study proposes a bi-directional recurrent neural network (Bi-RNN) post-processing method for speech enhancement (SE) at low signal-to noise ratios (SNR). Current speech enhancement solutions performed badly under low SNR situations. Loizou and Kim proposed a solution to reduce speech distortion errors in time-frequency (T-F) domain but it requires the knowledge of ground truth. As ground truth is unknown in real-life applications, the current study proposes to use a Bi-RNN to implement Loizou and Kim’s solution as a post-processing method for SE engines. Our solutions do not require prior knowledge of ground truth. The effectiveness of the proposed method is investigated with a spectral subtraction (SS) SE engine, a non-negative matrix factorization (NMF) SE engine, and a deep neural network ideal ratio mask (DNN-IRM) SE engine, under matched/mis-matched noise and different SNR conditions. Experimental results demonstrate that the proposed post-processing method effectively improved both perceptual evaluation of speech quality (PESQ) and short-time objective intelligibility (STOI) for all of these SE engines, especially at low SNR conditions.

In this paper, a robust multiple observation likelihood ratio test (MOLRT) based voice activity detection (VAD) method is proposed. At the beginning of this paper, we introduce the Wiener filter to the observed signal in time domain which can help mitigating some noise. The reason why we use the Wiener filter for VAD is that the performance of the VAD is always better in high signal to noise ratio (SNR) range than in low SNR range. Then, some ideas are proposed to improve the performance of the MOLRT based VAD method. As we all know, a conventional MOLRT based method including three modules named likelihood ratio (LR) estimation, threshold setting and hangover technique. To improve the estimation accuracy of LR, we adopt the unbiased minimum mean-square error (MMSE) algorithm for noise power spectrum estimation in every frame, which is very effective for LR estimation in MOLRT-based VAD method. That is because the LR is a function of a prior and a posterior SNR and unbiased MMSE algorithm is very useful for noise estimation. In addition, to make our VAD method more robust, a dynamic threshold setting technique is proposed in our method, which is related to the minimum noise power spectrum. That is because minimum noise power spectrum can help us updating the value of threshold to a suitable level according to the denoised signal. Last but most important, a novel hangover algorithm is introduced in this paper comparing to the conventional HMM based hangover algorithm. In the novel hangover algorithm, the current frame is determined by the statistical result of the following speech/non-speech detections based on the likelihood ratio test. And the evaluation results reveal that proposed method significantly outperform the baseline result of LRT as regards VAD accuracy in both noise variations and low SNR conditions.

We focus on the sound source localization (SSL) problem of small/weak voices recorded by small humanoid robots, such as Nao. The localization of small voice on such robots is more difficult than those with large bodies because 1) the number of microphones and their positions are restricted and 2) the signal to noise ratio (SNR) between internal noise and human speech is low. Multiple Signal Classification (MUSIC) is a promising noise-robust SSL method, and it has several variants in its eigenvector decomposition process, such as generalized eigenvalue decomposition (GEVD-MUSIC) and generalized singular value decomposition (GSVD-MUSIC). However, their performances are seriously affected by noise properties, the number of microphone, and their configurations on each robot's body. Since these have been confirmed only on robots with large bodies and many microphones, we need to investigate which variants perform better on a small humanoid robot with internal noise in it. We furthermore propose another MUSIC based on transformed steering vector (TSV-MUSIC) as a new implementation of GEVD-MUSIC. The computational cost of TSV-MUSIC is reduced compared with GEVD-MUSIC by changing its matrix multiplication procedures. Experimental results using real recoded data showed that our TSV-MUSIC outperformed others in terms of localization correctness by about 10 points under low SNR condition. We also compared the properties and performances of MUSIC-based SSLs by using simulated data.

Low signal-to-noise ratio (SNR) low probability of intercept (LPI) radar signal recognition is a key technical challenge in the field of electronic reconnaissance. Under low SNR conditions, severe noise contamination hinders the extraction of time-frequency features, resulting in degraded performance for traditional recognition methods. To address this issue, this paper proposes a cascaded framework, named MAE-WGAN, which integrates a Multi-Attention Enhanced Wasserstein Generative Adversarial Network for time-frequency image denoising, followed by radar signal recognition using a Convolutional Neural Network (CNN). The proposed approach embeds channel, spatial, and self-attention mechanisms into the U-Net-based generator of WGAN, forming a multi-stage radar signal processing system. First, under the guidance of high-quality time-frequency images, MAE-WGAN adaptively learns key time-frequency features using the multi-attention modules. Next, a discriminator is trained within the WGAN framework to recover signal structures obscured by noise. Finally, a dedicated multi-level feature extraction network is employed for accurate radar signal classification. Experimental results on <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 2}$</tex> typical LPI radar signals demonstrate that MAE-WGAN achieves high recognition accuracy under low SNR conditions, exhibiting strong recognition performance and robustness across varying SNR conditions.

Target vibration introduces unexpected vibration phases in the echo signal of inverse synthetic aperture LADAR (ISAL). The vibration phases cause the imaging results to be defocused. How to estimate the vibration phases has always been one of the difficulties in ISAL imaging, and there has not been an effective solution. In this paper, two methods are proposed to estimate the vibration phases in different situations. The method based on multi-channel along-track interferometry (MCATI method) is for some cases that the target velocity vector is parallel to some baselines of ISAL. The method based on orthogonal interferometry (OI method) is for other cases that the target velocity vector is not parallel to any baseline of ISAL. In addition, a high signal to noise ratio (SNR) is essential for the MCATI and OI methods to get high estimation accuracy. Therefore, a large power-aperture product should be carefully considered in the ISAL system design. Furthermore, on the condition of low SNR, range cell accumulation is also helpful to reduce the influence of noise. Take the ISAL, which is utilized to image the satellite in 500km orbit, as an example, the estimation accuracy can satisfy the demand of synthetic aperture imaging with 0dB single-pulse SNR and 300 range cell accumulation. The proposed methods are verified with simulation and experiment.

In the last few years, the demand of robust beamforming techniques has been increased significantly to serve numerous purposes. However, beamforming under low Signal to Noise Ratio (SNR) condition has always been a challenging task, which becomes more tedious in case of multiple sources. On the other hand, there are numerous factors such as steering vector error or similar estimation error problems that confine the efficiency of major beamforming techniques. Classical methods employ Eigen-Value Decomposition (EVD) and Singular Value Decomposition (SVD) over covariance matrix of the received signal; however their efficiency is limited under low SNR and multi-source condition, and perform inferiorly in case of steering vector errors. Such approaches are limited to achieve optimal Direction-of-Arrival (DOA) estimation that degrades the overall performance of the beamformer. Considering all these limitations, in this research, the focus is made on designing a robust DOA estimation model, followed by a novel beamformer for 8-element Uniform Linear Array (ULA). The simulation work consists two parts DOA estimation and beamforming. Unlike classical MUSIC and ESPRIT algorithms popular for DOA estimation our proposed model exploits the propagation method based subspace method where at first covariance matrix is obtained for the received signal, which is followed by normalization and Cross-Correlation matrix estimation. The obtained correlation matrix is processed for Generalized Eigen-Value Decomposition (GEVD) that efficiently obtained (orthogonal) subspace and SOI separation to yield DOA information. This approach enables more efficient DOA estimation even under low-SNR and steering vector error conditions. After obtaining DOA information, a novel adaptive beamformer is designed by using MVDR and FROST in parallel. DOA information from proposed model is given as input to the adaptive MVDR FROST beamforming model. Exploiting dynamic Signal-to-Interference ratio (SINR) by each beamformer our proposed method introduces an adaptation model that intends to achieve higher SINR value. MATLAB 2018a based simulation confirmed the robustness of the proposed model in terms of DOA estimation and beamforming results.

This study proposes a high precision algorithm to estimate the Enhanced Loran skywave delay. It is based on the artificial neural network. The algorithm establishes a neural network model between the receiving Enhanced Loran signal and the skywave propagation delay. By training a large number of data, the neural network model can more accurately reflect the relationship between the receiving signal and the skywave delay. This is an innovative application of the skywave delay estimation algorithm in the Enhanced Loran receiving system, especially in low signal‐to‐noise ratio (SNR) environments. The experimental results show that the accuracy of this algorithm is about hundreds of nanoseconds in the condition of normal receiving SNR. The accuracy of the algorithm is about us in the condition of low SNR, while the previous algorithm cannot be used in this case. It has also been proved by the off‐air data.

Enhanced sparse ISAR imaging by jointly using sparsity and low-rank properties

This letter proposes the use of vowel sound detection for voice activity detection. Vowels have distinctive spectral peaks. These are likely to remain higher than their surroundings even after severe corruption. Therefore, by developing a method of detecting the spectral peaks of vowel sounds in corrupted signals, voice activity can be detected as well even in low signal-to-noise ratio (SNR) conditions. Experimental results indicate that the proposed algorithm performs reliably under various noise and low SNR conditions. This method is suitable for mobile environments where the characteristics of noise may not be known in advance.

The acoustic pyrometer system uses physical properties of the gas and registers the perceivable temperature without the effects of radiation, and the measuring accuracy of acoustic wave flight time is the major factor in its application. The traditional correlation algorithm could not overcome convolution interference and calculate the time delay of acoustic signals accurately, under the low SNR (Signal Noise Ratio) and complex reverberation noise conditions. This paper presents a generalized cross power spectrum algorithm to filter the product and convolution interferences, and improves the performances of anti-noise and anti-convolution by whitening the sample signals and adjusting the weighted value of cross power spectrum algorithm with the variation of SNR. The experimental result and theoretical analysis showed that the new generalized cross power spectrum algorithm compared to the traditional correlation function analytical algorithm, can overcome the convolution interferences from reverberation noise, and sharpen peak value, thereby estimate the time delays of signals accurately.

Unmanned aerial vehicles (UAVs) offer a new approach to wireless communication, leveraging their high mobility, flexibility, and visual communication capabilities. Ambient backscatter communication enables Internet of Things devices to transmit data by reflecting and modulating ambient radio waves, eliminating the need for additional wireless channels, and reducing energy consumption and cost for sensors. However, passive ambient backscatter communication has limitations such as limited range and poor communication quality. By utilizing UAVs as communication nodes, these limitations can be overcome, expanding the communication range and improving the quality of communication. Although some research has been conducted on combining UAVs and ambient backscatter, several challenges remain. One key challenge is the strong direct link interference in ambient backscatter under UAV conditions, which significantly affects communication quality. In this paper, we propose an intelligent backward and forward straight link interference cancellation algorithm based on NOMA decoding technique to enhance ambient backscatter communication quality under UAV conditions and extract more ambient energy for UAV energy supply. The paper includes theoretical derivation, algorithm description, and simulation analysis to validate the error bit rate of labeled information bits. The results demonstrate that the forward algorithm reduces the error bit rate by approximately 20% under low signal-to-noise ratio (SNR) conditions, while the backward algorithm reduces the error bit rate under high SNR conditions. The combination of the forward and backward algorithms reduces the error bit rate under both high and low SNR conditions. The proposed method contributes to improving the quality of ambient backscatter communication in UAV settings.

In order to solve the problems of low recognition accuracy and high computation complexity of radar signal processing under low signal-to-noise ratio (SNR), this paper studies a neural network based on 1-D convolution for radar intra-pulse modulation recognition. First, we generate seven different types of radar continuous FM signals and phase coded signals (LFM, NLFM, Frank, P1, P2, P3, P4). Then, the three-dimensional convolutional network suitable for image processing is improved into a one-dimensional convolutional network which will process the generated radar signal in an end-to-end approach. The simulation results show that the recognition algorithm using one-dimensional (1-D) MobileNetV2 network can effectively extract the essential characteristics of radar signal under the condition of low SNR, avoid the huge calculation caused by using time-frequency method, and achieve over 85% recognition accuracy at −6dB.

Regulation of drones is already an important research topic. The drone identification method based on the radio frequency (RF) signal analysis technology is an efficient approach to regulate mainstream civilian drones. However, existing research on drone identification has been conducted under high signal-to-noise ratio (SNR) conditions. In fact, long-range drone identification in urban environments is performed under low SNR conditions. In this article, a global context (GC)-based threshold strategy is proposed to solve the above problem. The threshold is automatically determined during the deep architecture optimization. Meanwhile, the unimportant features in the network are removed by threshold denoising. First, the weight of each feature channel is obtained by modeling cross-channel dependencies, and then a weighted average is performed for each position along the feature channel on the feature map to obtain the thresholds. The method reduces the impact of unimportant features on performance without complex data preprocessing. In the experiments, the performance and complexity of the proposed method are extensively evaluated by 11 different drone RF signals. The experimental results show that in comparison with other deep learning-based threshold calculation methods, the proposed method has a more outstanding performance and lower computational complexity for drone RF signal identification under the low SNR condition.

LoRa (Long Range) technology is the cornerstone for many Low-Power Wide-Area Network (LPWAN) applications, demanding reliable signal detection for efficient spectrum utilization and robust communication. This paper introduces LoraNet, a Convolutional Neural Network (CNN) specifically designed for the binary classification of LoRa signals—distinguishing LoRa transmissions from Non-LoRa signals (e.g., noise or other interference) using their spectrogram representations. A comprehensive dataset was curated, featuring LoRa signals generated with diverse parameters including spreading factor and bandwidth. These signals, along with Non-LoRa samples consisting of single-tone sine waves, were embedded in additive white Gaussian noise across a challenging signal-to-noise ratio (SNR) range of -30 dB to +10 dB. LoraNet leverages a deep CNN architecture to learn discriminative features from these spectrograms. Experimental results demonstrate strong performance, with LoraNet achieving an overall accuracy of 89.88% on the test set. Notably, it achieves 51.20% accuracy at -30dB SNR, 98.20% at -20dB SNR, and 100.00% accuracy at SNRs of -10 dB and above. This highlights its robustness and effectiveness, particularly in low SNR conditions where traditional signal processing techniques often struggle. This study highlights the efficacy of deep learning in reliably detecting LoRa signal presence within complex radio environments.