Non-linear filtering of ultrasonic signals using neural networks

Structure noise from inhomogeneous micro-structures makes the detection of flaws present in highly scattering materials difficult. Several techniques have been applied to improve the signal-to-noise ratio (SNR) in order to make flaw detection easier. Linear filtering does not provide good results because both structure noise and flaw signal concentrate energy in the same frequency band. Non-linear filtering can be used to reduce the structure noise of ultrasonic signals. Therefore, neural networks are applied in this work for this purpose. In order to use neural networks for non-linear filtering, dynamic structures must be applied. The easiest way to implement a neural network with the capability of processing temporal patterns is to consider them spatial ones, applying the signal into a tapped delay line of finite extension, that is the input of a static neural network (for example, a multi-layer perceptron). In this work, a dynamic neural network has been built to filter ultrasonic signals with structure noise, and has been trained with the real-time back-propagation algorithm, using as inputs 3000 synthetic ultrasonic signals of 896 samples each. Target signals for training are the same as the ones used as inputs but without noise. The neural network is trained in order to generate as output the target signal when the noisy input one is applied. For testing the performance of the non-linear filter, a new set of 500 noisy signals has been used. The SNR improvement is about 6 dB average. The results show that this non-linear filtering method is quite useful as pre-processing stage in flaw detection systems.     

一种基于广义Duffing振子的水中弱目标检测方法

海洋环境中,在水下目标的线谱频率未知或者目标辐射噪声的连续谱很弱时,很难实现水中弱目标的准确检测,本文提出基于广义Duffing振子检测系统的水下目标辐射噪声检测方法.通过对传统周期扰动的Duffing振子信号检测系统的分析和推广,提出了一种可输入非周期、非平稳信号的广义Duffing振子检测系统,可检测输入的无先验信息目标信号.为实现广义Duffing振子系统运动状态的精确、有效判断,提出了一种相空间图形的离散分布列计算方法,通过类网格函数实现了利用统计复杂度对系统输出的嵌入式表征,从而实现了无先验信息时的水中弱目标的嵌入式检测.相同条件下与传统检测方法仿真对比可知,本文提出的方法可以检测到更低信噪比下的目标,并能满足水中检测实时性要求.     

The effects of bandwidth on the detectability of narrow- and wideband signals

The effects of masker bandwidth on the detection of narrow- and wideband signals have been investigated. For both kinds of signals, plots of threshold as a function of masker bandwidth yielded by both narrow- and wideband signals are reasonably described with two intersecting lines. Threshold initially increase with masker bandwidth and then become independent of further increases. The rate of increase depends on the signal spectrum. The bandwidth at which the lines intersect varies with signal bandwidth and also mode of masker presentation (i.e., whether the masker is gated with the signal or is present continuously). Internal filtering is most accurate when the masker is present continuously. A model is proposed in which a listener's decisions about the presence of narrow-band signals are based upon estimates of stimulus energy within a critical band. These estimates are degraded by bandwidth-dependent processing errors. When the signal to be detected spans several critical bands (i.e., is wideband), the model forms a test statistic by summing the outputs of the relevant critical bands. The model permits the contribution of each band to the sum to vary with masker bandwidth because it incorporates a form of lateral suppression. Thresholds of narrow-band signals in gated maskers and wideband signals in gated and continuous maskers are predicted by the model. However, the model fails to account for the detectability of narrow-band signals in continuous maskers.     

Improvement of signal-to-noise ratio by stochastic resonance in sigmoid function threshold systems, demonstrated using a CMOS inverter

Stochastic resonance (SR) has become a well-known phenomenon that can enhance weak periodic signals with the help of noise. SR is an interesting phenomenon when applied to signal processing. Although it has been proven that SR does not always improve the signal-to-noise ratio (SNR), in a strongly nonlinear system such as simple threshold system, SR does in fact improve SNR for noisy pulsed signals at appropriate noise strength. However, even in such cases, when noise is weak, the SNR is degraded. Since the noise strength cannot be known in advance, it is difficult to apply SR to real signal processing. In this paper, we focused on the shape of the threshold at which SR did not degrade the SNR when noise was weak. To achieve output change when noise was weak, we numerically analyzed a sigmoid function threshold system. When the slope around the threshold was appropriate, SNR did not degrade when noise was weak and instead was improved at suitable noise strength. We also demonstrated SNR improvement for noisy pulsed voltages using a CMOS inverter, a very common threshold device. The input-output property of a CMOS inverter resembles the sigmoid function. By inputting the noisy signal voltage to a CMOS inverter, we measured the input and output voltages and analyzed the SNRs. The results showed that SNR was effectively improved over a wide range of noise strengths.     

Speech enhancement based on the discrete Gabor transform and multi-notch adaptive digital filters

This paper presents a new method to speech enhancement based on time-frequency analysis and adaptive digital filtering. The proposed method for dual-channel speech enhancement was developed by tracking frequencies of corrupting signal by the discrete Gabor transform (DGT) and implementing multi-notch adaptive digital filter (MNADF) at those frequencies. Since no a priori knowledge of the noise source statistics is required this method differs from traditional speech enhancement methods. Specifically, the proposed method was applied to the case where speech quality and intelligibility deteriorate in the presence of background noise. Speech coders and automatic speech recognition (ASR) systems are designed to act on clean speech signals. Therefore, corrupted speech signals by the noise must be enhanced before their processing. The method uses a primary input containing the corrupted speech signal while a reference input containing the noise only. In this paper, we designed MNADF instead of single-notch adaptive digital filter and used DGT to track frequencies of corrupting signal because fast filtering process and fast measure of the time-dependent noise frequency are of great importance in speech enhancement process. Therefore, MNADF was implemented to take advantage of fast filtering process. Different types of noises from Noisex-92 database were used to degrade real speech signals. Objective measures, the study of the speech spectrograms and global signal-to-noise ratio (SNR), segmental SNR (segSNR), Itakura-Saito distance measure as well as subjective listing test demonstrated consistently superior enhancement performance of the proposed method over traditional speech enhancement method such as spectral subtraction. Combining MNADF and DGT, excellent speech enhancement was obtained.     

基于振动共振的随机共振控制

分析了非线性双稳系统在高、低两种不同频率信号作用下的动力学特性，给出了高频信号参数与双稳系统输出信号的信噪比和功率谱放大率关系的解析表达式，提出了基于振动共振的随机共振控制方法.理论分析和数值仿真结果表明，通过调节高频信号的幅值或频率大小，能有效地控制双稳系统输出信号的信噪比和功率谱放大率.     

Factors influencing intelligibility of ideal binary-masked speech: implications for noise reduction

The application of the ideal binary mask to an auditory mixture has been shown to yield substantial improvements in intelligibility. This mask is commonly applied to the time-frequency (T-F) representation of a mixture signal and eliminates portions of a signal below a signal-to-noise-ratio (SNR) threshold while allowing others to pass through intact. The factors influencing intelligibility of ideal binary-masked speech are not well understood and are examined in the present study. Specifically, the effects of the local SNR threshold, input SNR level, masker type, and errors introduced in estimating the ideal mask are examined. Consistent with previous studies, intelligibility of binary-masked stimuli is quite high even at -10 dB SNR for all maskers tested. Performance was affected the most when the masker dominated T-F units were wrongly labeled as target-dominated T-F units. Performance plateaued near 100% correct for SNR thresholds ranging from -20 to 5 dB. The existence of the plateau region suggests that it is the pattern of the ideal binary mask that matters the most rather than the local SNR of each T-F unit. This pattern directs the listener's attention to where the target is and enables them to segregate speech effectively in multitalker environments.     

A new sound coding strategy for suppressing noise in cochlear implants

In the n-of-m strategy, the signal is processed through m bandpass filters from which only the n maximum envelope amplitudes are selected for stimulation. While this maximum selection criterion, adopted in the advanced combination encoder strategy, works well in quiet, it can be problematic in noise as it is sensitive to the spectral composition of the input signal and does not account for situations in which the masker completely dominates the target. A new selection criterion is proposed based on the signal-to-noise ratio (SNR) of individual channels. The new criterion selects target-dominated (SNR > or = 0 dB) channels and discards masker-dominated (SNR<0 dB) channels. Experiment 1 assessed cochlear implant users' performance with the proposed strategy assuming that the channel SNRs are known. Results indicated that the proposed strategy can restore speech intelligibility to the level attained in quiet independent of the type of masker (babble or continuous noise) and SNR level (0-10 dB) used. Results from experiment 2 showed that a 25% error rate can be tolerated in channel selection without compromising speech intelligibility. Overall, the findings from the present study suggest that the SNR criterion is an effective selection criterion for n-of-m strategies with the potential of restoring speech intelligibility.     

Auditory analysis of the periodicity of sound and its envelope: A mathematical model

A mathematical model of the auditory analysis of periodicity of sound and its envelope is proposed. The model consists of a sequence of mathematical transformations that describe the signal processing stages. The following parameters and properties of the auditory system are taken into account: the crude analysis of the input acoustic signal accurate to the width of the aural critical band; the frequency dependence of the width of the aural critical band; the spectrum of the input signal analysis by a set of 3500 filters closely spaced in frequency; the absolute audibility thresholds at a given frequency; the time-domain analysis of both the output signals of each filter and the envelope profile with the help of the periodicity function; the pulsed activity of auditory neurons; the ability of the auditory system to memorize the spectral-time images of the signal and its individual parameters; the ability of the auditory system to form the perception of the loudness of sound, to memorize and compare the loudness of sound at different time moments, and to conclude which of them is higher or lower or whether they are equal accurate to a certain threshold; the dependence of the critical modulation bandwidth on the modulation frequency; and the dependence of the audibility thresholds of amplitude modulation on the modulation frequency. By an example of processing amplitude-modulated signals with various carrier-to-envelope frequency ratios, the model is shown to give a satisfactory explanation of their pitch and auditory estimate of the envelope period.     

Binaural critical masking bands.

Current understanding of the relation monaural estimates of the critical bandwidth for masking and those obtained in binaural listening situations is poor. The present study was designed to improve this situation by obtaining estimates of critical bandwich when the signal and masker were presented: (1) monaurally (NmSm), (2) binaurally with both signal and masker in phase at the ears (NoSo), (3) binaurally with masker in phase and signals 180 degrees out of phase (NoSpi). Threshold estimates were obtained in a two-interval forced-choice paradigm as a function of masker bandwidth for signal frequencies of 500 and 2000 Hz for the three conditions mentioned above. Maskers were computer-synthesized and had essentially infinite rejection slopes. For all conditions, as masker bandwidth was narrowed from wide band, threshold remained relatively constant until some critical bandwidth was reached. Further reductions in bandwidth were followed by progressive lowering of threshold, presumably due to removal of masker energy in the critical band. For both signal frequencies, the derived critical bandwidth estimates for the NmSm and NoSo conditions were similar and were smaller than the critical bandwidth estimates obtained in the NoSpi condition.     

Informational and energetic masking effects in the perception of two simultaneous talkers

Although most recent multitalker research has emphasized the importance of binaural cues, monaural cues can play an equally important role in the perception of multiple simultaneous speech signals. In this experiment, the intelligibility of a target phrase masked by a single competing masker phrase was measured as a function of signal-to-noise ratio (SNR) with same-talker, same-sex, and different-sex target and masker voices. The results indicate that informational masking, rather than energetic masking, dominated performance in this experiment. The amount of masking was highly dependent on the similarity of the target and masker voices: performance was best when different-sex talkers were used and worst when the same talker was used for target and masker. Performance did not, however, improve monotonically with increasing SNR. Intelligibility generally plateaued at SNRs below 0 dB and, in some cases, intensity differences between the target and masking voices produced substantial improvements in performance with decreasing SNR. The results indicate that informational and energetic masking play substantially different roles in the perception of competing speech messages.     

Modeling the influence of inherent envelope fluctuations in simultaneous masking experiments

Masked thresholds are measured and simulated for bandpass-noise signals ranging in bandwidth from 4 to 256 Hz in the presence of a masking bandpass noise also ranging in bandwidth from 4 to 256 Hz. Signal and masker are centered at 2 kHz. To investigate the role of temporal processing in simultaneous masking, simulations were performed with the modulation-filterbank model by Dau et al. [J. Acoust. Soc Am. 102, 2906-2919 (1997)]. For a fixed masker bandwidth, thresholds are independent of the signal bandwidth as long as the signal bandwidth does not exceed the masker bandwidth and thresholds decrease with increasing masker bandwidth in those conditions. A simple modulation-low-pass filter (energy integrator) would be sufficient to describe the experimental results in those conditions. In contrast, the processing by a modulation filterbank is necessary to account for the conditions of "asymmetry of masking," where thresholds for signals with bandwidths larger than the masker bandwidth are much lower than those for the reversed condition. In those conditions, the modulation-filterbank model is able to use the inherent higher modulation frequencies of the signal as an additional cue.     

部分相干信息处理系统的噪声演绩(Ⅱ)——时间部分相干照明

本文研究了时间相干性对光学信息处理系统噪声演绩的影响。以输出信噪比作为信息处理系统噪声演绩的度量,系统中任意平面上的颗粒噪声及相位噪声作为主要的噪声源。结果表明,时间部分相干照明可以提高系统的输出信噪比,对系统的装置噪声有明显的抑制作用。     

Binaural masking experiments using noise maskers with frequency-dependent interaural phase differences. II: Influence of frequency and interaural-phase uncertainty

This study investigates whether binaural signal detection is improved by the listener's previous knowledge about the interaural phase relations of masker and test signal. Binaural masked thresholds were measured for a 500-ms dichotic noise masker that had an interaural phase difference of 0 below 500 Hz and of pi above 500 Hz. The thresholds for two difference 20-ms test signals were determined within the same measurement using an interleaved adaptive 3-interval forced-choice (3IFC) procedure. In each 3IFC trial, both signals could occur with equal probability (uncertainty). The two signals differed in frequency and interaural phase in such a way that one signal always had a frequency above the masker edge frequency (500 Hz) and no interaural phase difference (So), whereas the other signal frequency was below 500 Hz and the interaural phase difference was pi (S pi). The frequencies of a signal pair remained fixed during the whole 3IFC track. These two signals thus lead to two different binaural conditions, i.e., NoS pi for the low-frequency signal and N pi So for the high-frequency signal. For comparison, binaural masked thresholds were measured with the same masker for fixed signal frequency and phase. The binaural masking level differences (BMLDs) resulting from the two experimental conditions show no significant difference. This indicates that the binaural system is able to apply different internal transformations or processing strategies simultaneously in different critical bands and even within the same critical band.     

基于自适应随机共振理论的太赫兹雷达信号检测方法

太赫兹雷达系统在差频信号频谱分析过程中,干扰噪声影响其测距能力.针对上述问题,提出基于自适应随机共振理论的太赫兹雷达信号检测方法,通过对含噪差频信号进行二次采样,利用自适应随机共振系统提取信号,进行尺度恢复完成测距计算.实验数据显示,不同测量距离时,相较于快速傅里叶变换法,输出信噪比的平均增益为9.684 d B,其中测量距离为1000 mm处,差频信号初始频谱值提高了64.1倍,系统信噪比增益为11.761 d B;相较于滤波法,在测量距离为1000 mm处信噪比增益最大,提高了70.56%;输入噪声强度为1—5 V之间时,输出信噪比曲线的曲率相对于滤波法降低了86.5%,其中噪声强度为5 V时信噪比增益最大,为14.018 d B.实验表明太赫兹雷达系统的测距能力大幅提高.     

Detection of multi-frequency weak signals with adaptive stochastic resonance system

Aiming at the poor detection rate of multi-frequency weak signals under a strong background of noise, a novel method based on adaptive stochastic resonance (SR) theory is proposed in this paper. The optimal parameters can be obtained automatically via measurement by establishing an adaptive SR system model and using the reverse location method. After passing through the adaptive SR system, the spectrum values of all eight signals greatly improve, the largest spectrum value gain increases from 12.41 to 2033 when the frequency is 0.01?Hz, which is an improvement of a factor of 162.8, and the signal-to-noise ratio (SNR) gain of the whole system is 10.3134?dB. Under the condition of different input noise intensities and signal amplitudes, the mean SNR of the system increases from –13.1136 to –2.7614?dB, which is a 78.9% increase, and the largest SNR gain is 13.4702?dB when the noise intensity D?=?1.2 and signal amplitude A?=?0.11. Compared to the single optimal spectrum value, when defining multiple optimum spectrum values as the SNR criterion, the detection sensitivity is less than 0.35 when the input noise intensity is between 0.5 and 2.5, and the sensitivity value is 6.29 times higher when D?=?2.5. The system successfully realizes the adaptive detection of twelve weak signals, and the SNR gain is 7.9743?dB, which improves the channel capacity of signal detection. The experimental results demonstrate the high efficiency and strong applicability of the system, improving the signal processing efficiency and speed of signal transmission.     

A reexamination of forward masking in the auditory nerve

Forward masking, as measured behaviorally, is defined as an increase in a signal's detection threshold resulting from a preceding masker. Previously, forward masking in the auditory nerve has been measured as a reduction in the neural response to a signal when preceded by a masker. However, detection threshold depends on both the magnitude of the response to the signal and the variance of the response. Thus changes in detectability cannot be inferred from response reduction alone. Relkin and Pelli (1987) have described a two-interval forced-choice procedure that may be used to measure the threshold for the detection of a probe signal in recordings of spike counts in single auditory neurons. These methods have been used to study the forward masking of characteristic frequency probe tones by characteristic frequency maskers as masker intensity was varied. Although the masker does reduce the detectability of the probe tone, it was found that the threshold shifts are much less than those observed behaviorally, particularly for intense maskers. In part, the small threshold shifts can be attributed to the reduction in response variance following the masker, which is the result of the adaptation of spontaneous activity. These results imply that behavioral forward masking must result from suboptimal processing of spike counts from auditory neurons at a location central to the auditory nerve.     

Signal area estimation based on deep learning

In many practical application scenarios, radio communication signals are commonly represented as a spectrogram, which represents the signal strength measured at multiple discrete time instants and frequency points within a specific time interval and frequency band, respectively. In the context of spectrum occupancy measurements, the notion of Signal Area (SA) is defined as the rectangular region in the time–frequency domain where a signal is assumed to be present. Signal Area Estimation (SAE) is an important functionality in spectrum-aware wireless systems where spectrum usage monitoring is required. However, the conventional approaches to SAE have a limited estimation accuracy, in particular at low SNR. In this work, a novel technique for SAE is proposed using Deep Learning based on Artificial Neural Network (DL-ANN) for enhanced extraction of SA information from radio spectrograms. The performance of the proposed DL-ANN method is evaluated both with software simulations and hardware experiments, and the results are compared with several conventional methods from the literature, showing significant performance improvements. A key feature of the proposed method is the improvement in the SAE accuracy compared to other existing methods (in particular in the low SNR regime) and the capability to extract the location of the detected SAs automatically. Overall, the proposed technique is a promising solution for the automatic processing of radio spectrograms in spectrum-aware wireless systems.     

Auditory filter shape derived from binaural masking experiments

The shape of the auditory filter was calculated from binaural masking experiments. Two different types of maskers were used in the study, a masker that was interaurally in phase at all frequencies (No), and a masker with an interaural phase difference of 0 below 500 Hz and of pi above 500 Hz. The test-signal frequency varied between 200 and 800 Hz, and the test signal was presented either monaurally (Sm) or binaurally in antiphase (S pi). By comparing the masked thresholds from the two experimental conditions, the following conclusion can be drawn: The threshold of the test signal is only affected by the masker phase within a narrow frequency range around the test frequency. Thus, for test-signal frequencies well above or below 500 Hz, no influence of the phase transition on the BMLD is observed, and normal masked thresholds for No and N pi maskers are obtained. For test frequencies around 500 Hz, the step in interaural phase difference leads to a decrease in the interaural correlation of the masker within the critical band around the test-signal frequency. This results in strong threshold changes for both monaural and binaural signals. A calculation of the auditory filter shape from the masked threshold values was performed under the assumption that the masked threshold is only dependent on the interaural cross correlation of the masker within the filter band. Using the formula of the EC theory for the relation between masker correlation and BMLD, the experimental data are well described by a trapezoidal filter with an equivalent rectangular bandwidth of 80 to 84 Hz.