Channel Mismatch Research Articles

Increasing the robustness of CNN acoustic models using autoregressive moving average spectrogram features and channel dropout

Developing automatic speech recognition systems that are robust to mismatched and noisy channel conditions is a challenging problem, especially when the training and the test conditions are different. Here, we seek to increase the robustness of convolutional neural network (CNN) acoustic models under such circumstances by combining two methods. Firstly, we propose an improved version of input dropout, which exploits the special structure of the input time-frequency representation. Instead of just dropping out random ‘pixels’ of the spectrogram, the proposed channel dropout approach discards whole spectral channels. We expect that this dropout strategy will force the network to rely less on the whole spectrum, and make it more robust to channel mismatches and narrow-band noise. Secondly, we replaced the standard mel-spectrogram input representation with the autoregressive moving average (ARMA) spectrogram, which was recently shown to outperform the former under mismatched train-test conditions. In our experiments on the Aurora-4 database, the proposed channel dropout method attained relative word error rate reductions of 16% with ARMA features (an absolute improvement of 3%), and 20% with FBANK features (an absolute improvement of 7%) over the baseline CNN, when using the clean training scenario.

ERROR ESTIMATION AND UNAMBIGUOUS RECONSTRUCTION FOR CHINESE FIRST DUAL-CHANNEL SPACEBORNE SAR IMAGING

Abstract. Multichannel synthetic aperture radar (SAR) is a significant breakthrough to the inherent limitation between high-resolution and wide-swath (HRWS) faced with conventional SAR. Error estimation and unambiguous reconstruction are two crucial techniques for obtaining high-quality imagery. This paper demonstrates the experimental results of the two techniques for Chinese first dualchannel spaceborne SAR imaging. The model of Chinese Gaofen-3 dual-channel mode is established and the mechanism of channel mismatches is first discussed. Particularly, we propose a digital beamforming (DBF) process composed of the subspace-based error estimation algorithm and the reconstruction algorithm before imaging. The results exhibit the effective suppression of azimuth ambiguities with the proposed DBF process, and indicate the feasibility of this technique for future HRWS SAR systems.

Digital constant‐envelope modulation scheme for radar using multicarrier OFDM signals

Multicarrier signals using orthogonal frequency division multiplexing (OFDM) bring many advantages to radar such as low probability of intercept, improved target detection performance, optimisation of both radar transmitter and receiver etc. A major drawback of multicarrier OFDM signals is their high peak-to-mean envelope power ratio, which reduces the working efficiency of the used power amplifiers. To solve this problem, a digital constant-envelope modulation scheme is proposed for radar using multicarrier OFDM signals. The scheme can transmit multicarrier OFDM signals through power amplifiers being operated near saturation levels (and thus maximising power efficiency) with little spectral distortion, even in the presence of channel mismatch. Given the speed limitations of field programmable gate arrays (FPGAs), the proposed scheme is implemented in an FPGA at intermediate frequency level using a polyphase processing structure.

Demonstration of Dual-Channel TOPS SAR Imaging With Airborne C-Band Data

Multichannel in azimuth synthetic aperture radar (SAR) operating in the terrain observation by progressive scans (TOPS) acquisition mode has attracted much attention recently for its capability to achieve ultrawide-swath imaging with a high spatial resolution. In order to verify the feasibility and operability of this newly developed remote sensing concept, a C-band airborne azimuth dual-channel TOPS SAR has been designed by the Institute of Electronics, Chinese Academy of Sciences, as a test bed for future spaceborne realizations. This paper introduces the experimental SAR system and reports the data processing results of an outfield experiment conducted in late September 2014. The importance of the experiment resides in its potential to validate several important technical aspects of this novel SAR operation with real experimental data, including channel mismatch cancellation and unambiguous signal reconstruction. Besides, two kinds of processing methods are proposed to calibrate the influence of antenna phase center fluctuation occurred in the dual-channel TOPS SAR. Finally, the experimental results obtained, including the phase mismatch cancellation and the focused imageries, are presented and analyzed.

Massive MIMO Systems With IQ Imbalance: Channel Estimation and Sum Rate Limits

In this paper, we study the impact of in-phase and quadrature imbalance (IQI) on single-cell multiuser multiple-input multiple-output systems with large antenna arrays. Moreover, we consider a time-division duplex system, where we assume uplink/downlink channel reciprocity in the downlink precoding design. First, we investigate the effect of transceiver IQI on the uplink channel estimation by deriving the linear minimum-mean-square-error estimator for the IQ-impaired model and prove that only the receiver IQI at the base station (BS) limits the estimation accuracy. Then, we study the impact of uplink/downlink channel mismatch on the downlink rate caused by different IQI at the BS and user equipments (UEs). We prove that the achievable downlink rate of each UE is limited either by the receiver IQI at the UEs or jointly by the transmit and receive IQI at the BS, when there is mismatch between the uplink and downlink channels.

Significance of Source–Filter Interaction for Classification of Natural vs. Spoofed Speech

Countermeasures used to detect synthetic and voice-converted spoofed speech are usually based on excitation source or system features. However, in the natural speech production mechanism, there exists nonlinear source–filter (S–F) interaction as well. This interaction is an attribute of natural speech and is rarely present in synthetic or voice-converted speech. Therefore, we propose features based on the S–F interaction for a spoofed speech detection (SSD) task. To that effect, we estimate the voice excitation source (i.e., differenced glottal flow waveform, $\dot{g} (t))$ and model it using the well-known Liljencrants–Fant model to get coarse structure, $g_{c}(t)$ . The residue or difference, $g_{r}(t)$ , between $\dot{g} (t)$ and $g_{c}(t)$ is known to capture the nonlinear S–F interaction. In the time domain, the $L^{2}$ norm of $g_{r}(t)$ in the closed, open, and return phases of the glottis are considered as features. In the frequency domain, the Mel representation of $g_{r}(t)$ showed significant contribution in the SSD task. The proposed features are evaluated on the first ASVspoof 2015 challenge database using a Gaussian mixture model based classification system. On the evaluation set, for vocoder-based spoofs (i.e., S1 – S9 ), the score-level fusion of residual energy features, Mel representation of the residual signal, and Mel frequency cepstral coefficients (MFCC) features gave an equal error rate (EER) of 0.017%, which is much less than the 0.319% obtained with MFCC alone. Furthermore, the residues of the spectrogram (as well as the Mel-warped spectrogram) of estimated $\dot{g} (t)$ and $g_{c}(t)$ are also explored as features for the SSD task. The features are evaluated for robustness in the presence of additive white, babble, and car noise at various signal-to-noise-ratio levels on the ASVspoof 2015 database and for channel mismatch condition on the Blizzard Challenge 2012 dataset. For both cases, the proposed features gave significantly less EER than that obtained by MFCC on the evaluation set.

Locally Normalized Filter Banks Applied to Deep Neural-Network-Based Robust Speech Recognition

This letter describes modifications to locally normalized filter banks (LNFB), which substantially improve their performance on the Aurora-4 robust speech recognition task using a Deep Neural Network-Hidden Markov Model (DNN-HMM)-based speech recognition system. The modified coefficients, referred to as LNFB features, are a filter-bank version of locally normalized cepstral coefficients (LNCC), which have been described previously. The ability of the LNFB features is enhanced through the use of newly proposed dynamic versions of them, which are developed using an approach that differs somewhat from the traditional development of delta and delta–delta features. Further enhancements are obtained through the use of mean normalization and mean–variance normalization, which is evaluated both on a per-speaker and a per-utterance basis. The best performing feature combination (typically LNFB combined with LNFB delta and delta–delta features and mean–variance normalization) provides an average relative reduction in word error rate of 11.4% and 9.4%, respectively, compared to comparable features derived from Mel filter banks when clean and multinoise training are used for the Aurora-4 evaluation. The results presented here suggest that the proposed technique is more robust to channel mismatches between training and testing data than MFCC-derived features and is more effective in dealing with channel diversity.

Speaker Model Clustering to Construct Background Models for Speaker Verification

Abstract Conventional speaker recognition systems use the Universal Background Model (UBM) as an imposter for all speakers. In this paper, speaker models are clustered to obtain better imposter model representations for speaker verification purpose. First, a UBM is trained, and speaker models are adapted from the UBM. Then, the k-means algorithm with the Euclidean distance measure is applied to the speaker models. The speakers are divided into two, three, four, and five clusters. The resulting cluster centers are used as background models of their respective speakers. Experiments showed that the proposed method consistently produced lower Equal Error Rates (EER) than the conventional UBM approach for 3, 10, and 30 seconds long test utterances, and also for channel mismatch conditions. The proposed method is also compared with the i-vector approach. The three-cluster model achieved the best performance with a 12.4% relative EER reduction in average, compared to the i-vector method. Statistical significance of the results are also given.

Robust blind separation for MIMO systems against channel mismatch using second-order cone programming

To improve the deteriorated capacity gain and source recovery performance due to channel mismatch problem, this paper reports a research about blind separation method against channel mismatch in multiple-input multiple-output (MIMO) systems. The channel mismatch problem can be described as a channel with bounded fluctuant errors due to channel distortion or channel estimation errors. The problem of blind signal separation/extraction with channel mismatch is formulated as a cost function of blind source separation (BSS) subject to the second-order cone constraint, which can be called as second-order cone programing optimization problem. Then the resulting cost function is solved by approximate negentropy maximization using quasi-Newton iterative methods for blind separation/extraction source signals. Theoretical analysis demonstrates that the proposed algorithm has low computational complexity and improved performance advantages. Simulation results verify that the capacity gain and bit error rate (BER) performance of the proposed blind separation method is superior to those of the existing methods in MIMO systems with channel mismatch problem.

Parameter Estimation Using Partly Calibrated Vector Antennas

This letter presents a parameter estimation method using an imperfectly calibrated array of vector antennas. Based on a series of separation of different types of model errors and a subsequent iterative and alternate self-calibration procedure, the proposed method works well in the case of coexistence of channel mismatch, antenna misorientation, and element position error. The mutual coupling effect among antennas is reduced by spreading sparsely the array elements. Simulation results show that the proposed method has a superior performance over the conventional uncalibrated counterpart in terms of incident angle and polarization estimation accuracy.

A Study on The Performance of Multiple-beam Antenna Satellite Receiving System

High quality, high capability and wide overlay are desired in modern satellite communication. Many new technologies have been used in communication satellite, and multiple-beam antenna was one of them. Discussing the amplitude and phase inconsistency of the RF multiple-receiver is the most important part for the multiple-beam antenna in practical application. The influence of the channel mismatch on the satellite multiple-beam antenna system has been discussed and simulated in this paper. We have given the signal to noise ratio (SNR) to bit error rate (BER) curves at last. And the discussion results have a certain practical application value for designing the satellite system.

Optimal and Robust Beamforming for Secure Transmission in MISO Visible-Light Communication Links

This work considers secure downlink transmission in indoor multiple-input, single-output visible-light communication (VLC) links. In particular, we study the design of transmit beamformers that maximize the achievable secrecy rate subject to amplitude constraints imposed by the limited dynamic range of the light-emitting diodes (LEDs). Such constraints render the design problem nonconvex and difficult to solve. We show, however, that this nonconvex problem can be transformed into a solvable quasiconvex line search problem. We also consider the more realistic case of imperfect channel information regarding the receiver's and eavesdropper's links. We tackle the worst-case secrecy rate maximization problem, again subject to amplitude constraints. In our treatment, uncertainty in the receiver's channel is due to limited feedback, and is modeled by spherical sets. On the other hand, there is no feedback from the eavesdropper, and the transmitter shall utilize the line-of-sight channel gain equation to map the eavesdropper's nominal location and orientation into an estimate of the channel gain. Thus, we derive uncertainty sets based on inaccurate information regarding the eavesdropper's location and orientation, as well as the emission pattern of the LEDs. We also consider channel mismatches caused by the uncertain non-line-of-sight components. We provide numerical examples to demonstrate the performance gain of the optimal beamformer compared with the suboptimal schemes, and the robust beamformer compared with its nonrobust counterparts. We also evaluate the worst-case secrecy rate performance of the robust beamformer in a typical VLC scenario along with the aforementioned uncertainty sources.

Very Deep Convolutional Neural Networks for Noise Robust Speech Recognition

Although great progress has been made in automatic speech recognition, significant performance degradation still exists in noisy environments. Recently, very deep convolutional neural networks (CNNs) have been successfully applied to computer vision and speech recognition tasks. Based on our previous work on very deep CNNs, in this paper this architecture is further developed to improve recognition accuracy for noise robust speech recognition. In the proposed very deep CNN architecture, we study the best configuration for the sizes of filters, pooling, and input feature maps: the sizes of filters and poolings are reduced and dimensions of input features are extended to allow for adding more convolutional layers. Then the appropriate pooling, padding, and input feature map selection strategies are investigated and applied to the very deep CNN to make it more robust for speech recognition. In addition, an in-depth analysis of the architecture reveals key characteristics, such as compact model scale, fast convergence speed, and noise robustness. The proposed new model is evaluated on two tasks: Aurora4 task with multiple additive noise types and channel mismatch, and the AMI meeting transcription task with significant reverberation. Experiments on both tasks show that the proposed very deep CNNs can significantly reduce word error rate (WER) for noise robust speech recognition. The best architecture obtains a 10.0% relative reduction over the traditional CNN on AMI, competitive with the long short-term memory recurrent neural networks (LSTM-RNN) acoustic model. On Aurora4, even without feature enhancement, model adaptation, and sequence training, it achieves a WER of 8.81%, a 17.0% relative improvement over the LSTM-RNN. To our knowledge, this is the best published result on Aurora4.

A Calibration Method for Nonlinear Mismatches in M-Channel Time-Interleaved Analog-to-Digital Converters Based on Hadamard Sequences

The time-interleaved analog-to-digital converter (TIADC) is an architecture used to achieve a high sampling rate and high dynamic performance. However, estimation and compensation methods are required to maintain the dynamic performance of the constituent analog-to-digital converters (ADCs) due to channel mismatches. This paper proposes a blind adaptive method to calibrate the nonlinear mismatches in M-channel TIADCs (M-TIADCs). The nonlinearity-induced error signal is reconstructed by the proposed multiplier Hadamard transform (MHT) structure, and the nonlinear parameters are estimated by the filtered-X least-mean square (FxLMS) algorithm. The performance of cascade calibration is also analyzed. The numerical simulation results show that the proposed method consumes much less hardware resources while maintaining the calibration performance.

Calibration of Time-Interleaved Errors in Digital Real-Time Oscilloscopes.

A channel mismatch calibration method is proposed for use in time-interleaved digital real-time oscilloscope (DRTO) applications. Linear equations are derived using Fourier transforms of the separated signals from each of the time-interleaved analog-to-digital converters (TIADCs). Thus the errors in the TIADCs can be easily calibrated by inversion of a matrix, as opposed to most previous work where additional filters are employed. The calibration accuracy of the proposed method is limited only by the noise produced after the TIADC circuitry, while other methods depend on the filter design. A transfer function measurement method is then proposed for application to commercially available DRTOs. Two-tone signals are measured using DRTOs from various suppliers to validate the proposed method. The occurrence of signals at spurious frequencies is considerably reduced, as demonstrated by the calibrated results.

Investigating and predicting the perceptibility protect of channel mismatch in stereoscopic video

The degree of visual discomfort caused by watching stereoscopic scenes with channel mismatch is investigated and predicted. A scene with channel order mismatch is one in which the right and left views are swapped. A way of finding channelmismatch is used to analyze 105 3D films; the scenes found in this analysis are used for experimental study of the visual discomfort caused by channelmismatch. The experimental results are used to construct samplingwith a reference pattern. This sampling is used to learn different regression analysis algorithms, and the bestway of predicting visual discomfort caused by channel mismatch is chosen.

Optimal downlink beamforming for statistical CSI with robustness to estimation errors

We consider the problem of worst-case robust downlink beamforming in a multiuser Multiple-input–single-output (MISO) network with statistical channel state information (CSI) that is erroneous. In previous beamforming techniques, robustness is usually incorporated by employing generic error models, in which the mismatch between the presumed and the true instantaneous channel vectors, or the corresponding channel covariance matrices, is bounded by the Frobenius norm. A shortcoming of these methods is that these strongly rely on the assumption that the exact noise variance of the terminals and bounds on the channel mismatch are available. If this assumption is violated, then the beamforming approach may result in either insufficient robustness or overly conservative solutions. This assumption is particularly difficult as different sources of CSI error are generally not considered. In our proposed robust approach, we consider CSI mismatch from estimation errors in the noise power and the channel plus noise covariance matrix. Addressing both estimation errors separately allows us to derive meaningful threshold values for the uncertainty sets. In contrast to previous robust approaches based on Frobenius norm, we propose to use confidence intervals with given confidence levels to model the uncertainty sets. The simulation results verify the effectiveness of the proposed techniques.

Underwater acoustic communication in time-varying channel environment based on passive time reversal

Channels estimated from probe signal or training symbols will mismatch with the true channels in passive time reversal (TR) underwater acoustic (UWA) communication when the ocean environment is time-varying, leading to poor channel equalization performance. In order to solve this problem, this paper studies Block-based TR processing structure. The received data are divided into small blocks and the channel estimated from previously decoded symbols is used to match filter the current block to suppress the channel mismatch. Decision feedback equalizer (DFE) is also combined with Block-based TR to remove the residual inter symbol interference (ISI) and Doppler effect (Block-based TR-DFE). Mobile UWA communication experiment was conducted near Xiao Changshan island, Dalian in July, 2015. Though the received data are preprocessed by resampling, residual Doppler effect still exists due to ununiform motion between the receiver and transmitter, making the channel time-varying fastly. Block-based TR realized effective tracking and compensation for time-varying channels. The output signal-to-noise ratio (SNR) after equalization is greatly improved and the bit error rate (BER) is significantly reduced.

Considerations for Design and Implementation of a RF Emitter Localization System with Array Antennas

In this paper, design and implementation issues for a network-oriented RF emitter localization system with array antenna are discussed. For hardware, the problem of array mismatch and RF/IF channel mismatch are introduced and the calibration schemes for solving those problems are also provided. For software, it is explained how to overcome the drawback of conventional MUltiple Signal Identification and Classification (MUSIC) algorithm in a point of identifying the number of received signals and problems such as Data Association Problem and Ghost Node Problem in regard to multiple emitter localization are presented with some approaches for getting around those problems. Finally, for implementation, a criterion for arranging each of sensors and a requirement for alignment of array antenna' orientation are also given.

Analysis on the adaptive filter based on LMS algorithm

This article focuses on the application of adaptive filter based on the LMS algorithm. An adaptive filter of the closed-loop system is introduced, including the elimination of interference signal, the prediction of useful signal, and the approximation of expected signal. LMS (Least Mean Square) algorithm is used to meet the optimum norm of error between estimated signal and expected signal. The structure of LMS algorithm is presented and the simulation of LMS algorithm is carried out. The results indicate that the convergence performances of LMS algorithm are prefect, and the input signal can converge to the expected signal. The application of adaptive filtering technology in this article includes the correction of channel mismatch by an adaptive linear filter, the improvement of system performance by an adaptive equalizer, and the filter of frequency signal by an adaptive notch filter. The analysis on adaptive linear filter shows that the constant channel mismatch can be corrected quite well by the correction algorithm. The analysis on adaptive equalizer shows that the error rate of system with an adaptive equalizer has significant improvement gains over that of system without an adaptive equalizer. The smaller the error rate, the larger the SNR. The relationship between error rate and multi-path loss show that the error rate is largest when the loss factor is 0.5. The analysis on adaptive notch filter shows that the interference signal with two different known frequencies can be eliminated effectively by the adaptive notch filter. The filtered signals accord with the corresponding useful signals very well.