Abstract Sound classification using neural networks has recently produced very accurate results. A large number of different applications use this type of sound classifiers such as controlling and monitoring the type of activity in a city or identifying different types of animals in natural environments. While traditional acoustic processing applications have been developed on high-performance computing platforms equipped with expensive multi-channel audio interfaces, the Internet of Things (IoT) paradigm requires the use of more flexible and energy-efficient systems. Although software-based platforms exist for implementing general-purpose neural networks, they are not optimized for sound classification, wasting energy and computational resources. In this work, we have used FPGAs to develop an ad hoc system where only the hardware needed for our application is synthesized, resulting in faster and more energy-efficient circuits. The results show that our developments are accelerated by a factor of 35 compared to a software-based implementation on a Raspberry Pi.

In this work we propose a multi-channel active noise equalizer (ANE) that can deal with multi-frequency noise signals and assigns simultaneously different equalization gains to each frequency component at each monitoring sensor. For this purpose, we state a pseudo-error noise signal for each sensor of the ANE, which has to be cancelled out in order to get the desired equalization profiles. Firstly the optimal analytic solution for the ANE filters in the case of single-frequency noise is provided, and an adaptive algorithm based on the Least Mean Squared (LMS) is proposed for the same case. We also show that this adaptive strategy reaches the theoretical solution in steady state. Secondly, we state an equivalent approach for the case of multi-frequency noise based on two alternatives: a common pseudo-error signal at each sensor for all frequencies, and a different pseudo-error signal at each sensor for each frequency. The analytic and adaptive solutions for the ANE control filters have been developed for both pseudo-error alternatives. Finally, the ability of the proposed ANE to achieve simultaneously different user-selected noise profiles in different locations has been validated by their transfer functions and simulations.

This work presents a wireless acoustic sensor network (WASN) that monitors urban environments by recognizing a given set of sound events or classes. The nodes of the WASN are Raspberry Pi devices that not only record the ambient sound, but also detect and recognize different sound events. All the signal processing tasks, from the recording to the classification carried out by a convolutional neural network (CNN), are run on Raspberry Pi devices. Due to the low cost of the proposed acoustic nodes, the system exhibits a very high potential scalability. Regarding the underlying WASN, it has been designed according to the open standard FIWARE, thus the whole system can be deployed without the need of proprietary software. Regarding the performance of the sound classifier, the proposed WASN achieves similar accuracy compared to other WASNs that make use of cloud computing. However, the proposed WASN significantly minimizes the network traffic since it does not exchange audio signals, but only contextual information in form of labels. On the other hand, most of the time the class reported by the WASN nodes is the “background” soundscape, which usually contains no event of interest. This is the case when monitoring the soundscape of big avenues, where four events have been identified: “traffic”, “siren”, “horn” and “noisy vehicles”, being the “traffic” class associated to the background soundscape. In this paper, the use of a simple pre-detection stage prior to the CNN classification is proposed, with the aim of saving computation and power consumption at the nodes. The pre-detection stage is able to differentiate the other three relevant sounds from the “traffic” and activates the classifier only when some of these three events is likely occurring. The proposed pre-detection stage has been validated through data recorded in the city of Valencia (Spain), achieving a reduction of the Raspberry Pi CPU’s usage by a factor of six.

Acoustic sensor networks (ASNs) are an effective solution to implement active noise control (ANC) systems by using distributed adaptive algorithms. On one hand, ASNs provide scalable systems where the signal processing load is distributed among the network nodes. On the other hand, their noise reduction performance is comparable to that of their respective centralized processing systems. In this sense, the distributed multiple error filtered-x least mean squares (DMEFxLMS) adaptive algorithm has shown to obtain the same performance than its centralized counterpart as long as there are no communications constraints in the underlying ASN. Regarding affine projection (AP) adaptive algorithms, some distributed approaches that are approximated versions of the multichannel filtered-x affine projection (MFxAP) algorithm have been previously proposed. These AP algorithms can efficiently share the processing load among the nodes, but at the expense of worsening their convergence properties. In this paper we develop the exact distributed multichannel filtered-x AP (EFxAP) algorithm, which obtains the same solution as that of the MFxAP algorithm as long as there are no communications constraints in the underlying ASN. In the EFxAP algorithm each node can compute a part or the entire inverse matrix needed by the centralized MFxAP algorithm. Thus, we propose three different strategies that obtain significant computational saving: 1) Gauss Elimination, 2) block LU factorization, and 3) matrix inversion lemma. As a result, each node computes only between 25% - 60% of the number of multiplications required by the direct inversion of the matrix. Regarding the performance in transient and steady states, the EFxAP exhibits the fastest convergence and the highest noise level reduction for any size of the acoustic network and any projection order of the AP algorithm compared to the DMEFxLMS and two previously reported distributed AP algorithms.

Personal Sound Zones (PSZ) systems aim to render independent sound signals to multiple listeners within a room by using arrays of loudspeakers. One of the algorithms used to provide PSZ is Weighted Pressure Matching (wPM), which computes the filters required to render a desired response in the listening zones while reducing the acoustic energy arriving to the quiet zones. This algorithm can be formulated in time and frequency domains. In general, the time-domain formulation (wPM-TD) can obtain good performance with shorter filters and delays than the frequency-domain formulation (wPM-FD). However, wPM-TD requires higher computation for obtaining the optimal filters. In this article, we propose a novel approach to the wPM algorithm named Weighted Pressure Matching with Subband Decomposition (wPM-SD), which formulates an independent time-domain optimization problem for each of the subbands of a Generalized Discrete Fourier Transform (GDFT) filter bank. Solving the optimization independently for each subband has two main advantages: 1) lower computational complexity than wPM-TD to compute the optimal filters; 2) higher versatility than the classic wPM algorithms, as it allows different configurations (sets of loudspeakers, filter lengths, etc.) in each subband. Moreover, filtering the input signals with a GDFT filter bank (as in wPM-SD) requires lower computational effort than broadband filtering (as in wPM-TD and wPM-FD), which is beneficial for practical PSZ systems. We present experimental evaluations showing that wPM-SD offers very similar performance to wPM-TD. In addition, two cases where the versatility of wPM-SD is beneficial for a PSZ system are presented and experimentally validated.

In this paper we deal with the estimation of the room impulse response (RIR) between each loudspeaker and each microphone of a wireless acoustic network of two nodes when used to implement a crosstalk canceller. The nodes of the network are commercial devices connected via standard wireless links, presenting low computational requirements and non-ideal synchronization between them. Moreover, the nodes can exchange information, but they cannot share their signals due to the high throughput and perfect synchronism that would be required. The proposed scheme adaptively estimates the global impulse response between the source signals and the recorded signal at each node of the network, and afterwards estimates the corresponding RIRs between each loudspeaker and the node's microphone. This scheme does not need any additional synchronism between loudspeakers. Simulations show that proportionate-type affine projection algorithms obtain good performance for order N = 4, being their cost affordable in commercial devices.

Spatial multiplexing techniques in multiuser multiple-input multiple-output orthogonal frequency- division multiplexing (MU-MIMO-OFDM) systems have increased the spectral efficiency of wireless communications systems. However, these systems require an accurate knowledge of the channel at the transmitter, which usually has to be provided by receivers through a low-rate feedback channel. In this paper, we present a channel quantization scheme based on the Karhunen-Loève (KL) transform. This scheme outperforms the scheme based on time-domain quantization in terms of mean square error for a given amount of quantization bits. The comparison has been carried out in a system that presents some unmodulated subcarriers. The computational complexity of both schemes has also been analyzed showing that, when efficient eigendecomposition is used, the scheme based on KL transform has only a slightly higher computational complexity.

Adaptive filtering schemes exhibit a compromise between convergence speed and steady-state mean square error. Trying to overcome this trade-off, convex combination of adaptive filters have been recently developed for system identification achieving better performance than traditional approaches. The purpose of this work is to apply the convex combination strategy to single-channel and multichannel active noise control systems. In these systems it is necessary to take into account the secondary path between the adaptive filter output and the error sensor and the possible unavailability of the disturbance signal, which depends on the filtering scheme considered. Even though this strategy involves a higher computational burden than the classic adaptive filters, it exhibits a good performance in terms of convergence speed and steady-state mean square error.

Fast parallel processing capability of general purpose Graphic Processing Units (GPU) can be exploited to accelerate the precoding calculation needed in spatially multiplexed wireless communication systems. In this paper, a GPU-based implementation of the well-known multiuser Tomlinson-Harashima precoding (THP) scheme combined with a lattice-reduction (LR) stage is presented. The proposed approach allows the LR stage to be switched off when user requirements are achieved by using only THP. Moreover, our GPU implementation provides scalability in the number of sub-carriers per symbol, which is a key factor in LTE and 4G wireless standards. Simulation results show that the GPU-based THP implementation performs up to 7 times faster than its CPU-equivalent whereas the LR stage implementation only achieves a speedup of 3. Despite the fact that the LR cannot be as efficiently parallelized as the THP, a speedup of nearly 6 is achieved when both are combined.

Affine projection (AP) algorithms have been proposed in recent years for use in active noise control systems. This is due to their potential high convergence speed along with their robustness and moderate computational cost. However, these algorithms can exhibit an excessive computational cost for high projection orders (just when higher convergence speed is achieved). Thus, computationally efficient versions of these algorithms have been proposed. For the particular case of the AP algorithms applied to active noise control, the use of the conventional filtered-x structure instead of the commonly used modified filtered-x method can be understood as an efficient strategy, since it needs fewer operations to update the adaptive filter coefficients. However, the use of this structure implies different algorithm behavior for the following two reasons: the signals needed in the coefficient updates do not correspond exactly to the AP algorithm and this structure introduces a delay between the update of the adaptive filter coefficients and its effect on the noise signal. In practice, this dual effect mainly affects convergence of the algorithms in the transient regime. This correspondence presents a mathematical model so that the transient behavior of the conventional filtered-x AP algorithm can be predicted from the reference signal statistics and algorithm parameters.

This paper provides an analysis of the steady-state behavior of the filtered-x affine projection algorithm (FXAP). This efficient affine projection (AP) algorithm for active noise control (ANC) applications is based on the filtered-x scheme, unlike most AP algorithms based on the more computationally demanding modified filtered-x scheme. This study depends on energy conservation arguments and does not require an specific signal distribution. The theoretical expressions derived for the mean square error (MSE) allowed to accurately predict the steady-state performance of the FXAP for meaningful practical cases. Simulation results of a single-channel ANC system validate the analysis and the theoretical expressions derived.

In recent years, affine projection algorithms have been proposed for adaptive system applications as an efficient alternative to the slow convergence speed of least mean square (LMS)-type algorithms. Whereas much attention has been focused on the development of efficient versions of affine projection algorithms for echo cancellation applications, the similar adaptive problem presented by active noise control (ANC) systems has not been studied so deeply. This paper is focused on the necessity to reduce even more the computational complexity of affine projection algorithms for real-time ANC applications. We present some alternative efficient versions of existing affine projection algorithms that do not significantly degrade performance in practice. Furthermore, while in the ANC context the commonly used affine projection algorithm is based on the modified filtered-x structure, an efficient affine projection algorithm based on the (nonmodified) conventional filtered-x structure, as well as efficient methods to reduce its computational burden, are discussed throughout this paper. Although the modified filtered-x scheme exhibits better convergence speed than the conventional filtered-x structure and allows recovery of all the signals needed in the affine projection algorithm for ANC, the conventional filtered-x scheme provides a significant computational saving, avoiding the additional filtering needed by the modified filtered-x structure. In this paper, it is shown that the proposed efficient versions of affine projection algorithms based on the conventional filtered-x structure show good performance, comparable to the performance exhibited by the efficient approaches of modified filtered-x affine projection algorithms, and also achieve meaningful computational savings. Experimental results are presented to validate the use of the algorithms introduced in the paper for practical applications.

The problem of designing joint power control and optimal beamforming (JPCOB) algorithms for the downlink of a coordinated multi-cell WCDMA system is considered throughout this paper. In this case, the JPCOB design is formulated as the problem of minimizing the total transmitted power in the coordinated multi-cell system, subject to a certain quality of service requirement for each user. In this paper, the performance of two JPCOB algorithms based on different beamforming approaches is compared over the coordinated multi-cell system. The first one, obtains local beamformers by means of the well-known virtual uplink-downlink duality. In contrast, the second algorithm implements multi-base beamformers, taking into account match filter equalizers at the receivers. Moreover, realistic system parameters, such as per-base station power constraints or the asynchronous nature of the signals arriving at the receivers, are taken into account. Simulation results show that the algorithm based on multi-base beamforming presents attractive properties, such as an inherent multi-base scheduling technique or a decreasing total transmitted power as the degree of coordination between base stations is increased.

An efficient version of the affine projection (AP) algorithm, which is called FXAP-AC throughout the paper, is presented. This algorithm uses the filtered-x structure instead of the modified structure which is required for the implementation of the AP algorithm when the desired signal is not available, for instance in active noise control (ANC) applications. The conventional AP algorithm for ANC, which is based on the modified structure (MFXAP), provides good performance. However the filtered-x structure is advisable for practical cases since it achieves a meaningful computational saving avoiding the additional filtering needed by the modified structure without significantly affect the performance in practice. In this paper, an efficient computation of coefficient update is developed for the FXAP resulting in the FXAP-AC. This low-cost strategy already proposed in the conventional AP algorithm can be suitably combined with others previously reported to achieve very fast and robust FXAP algorithms. Experimental results validate the use of the proposed algorithm for practical applications

An algorithmic variant of the conventional active noise equalizer (ANE), which independently controls some given frequencies of the primary signal, has been developed and extended to the multichannel case. The modified version of the ANE is named common-error multiple-frequency ANE. A detailed analysis of both multichannel equalizers has been carried out. From a convergence analysis in the frequency-domain, the significance of transducer locations in the behavior of a practical system can be predicted through the matrix of secondary path responses at each frequency. The ANEs steady-state transfer functions from the primary input signal to the noise output have also been developed and compared for different parameter settings and for accurate and inaccurate secondary path estimation. Furthermore, the multichannel extension of both equalizers has been implemented in a real-time active system inside a listening room for multifrequency noise. Useful-size zones of equalization have been binaurally measured by using a head and torso simulator. It was found that the common-error multiple-frequency ANE performs better than the conventional equalizer because it achieves a saving in computational complexity and has smaller overshoot. It can also be implemented in a real controller more easily than the conventional ANE and without showing meaningful differences in the practical results provided.

joint power control and optimal beamforming (JPCOB) algorithms have been proposed in the literature as promising solutions to increase the capacity of wireless multimedia networks. A new development of one of such JPCOB techniques is presented in this paper, whose main feature is the possibility of downlink cooperative transmission between all the base stations involved. Some practical applications under adverse conditions of deep fading are discussed, and different realistic scenarios have been simulated in order to analyze the new algorithm performance. Finally, some interesting conclusions are stated.

A novel approach to the widespread problem of evaluating the “quality” of noise of a car engine is presented. Traditionally some psychoacoustic parameters of the sound are calculated and used to generate a certain Noise Quality Index (NQI) according to a previous jury assessment. Once the NQI is validated, new noise recordings can be evaluated and automobile engineers can estimate how the perceived quality varies as engine working conditions change. In this paper we relate well-known psychoacoustic features - which usually contribute to generate the NQI - to Time-Frequency (TF) characteristics of diesel noise, in order to find relations between both kinds of parameters. These relations could help engine designers to predict the noise quality just analysing the spectral changes of the noise waveform. Some promising results are showed involving a diesel engine belonging to the luxury class of an European car manufacturer.

This work is a preliminary study which focuses on subjective improvements of vehicle interior noise by using active noise control techniques. A local active noise control system has been mounted inside a room in order to reduce synthesized and real engine noise in the area around the headrest of a typical car seat. Recordings before and after cancellation have been made using a head acoustics mannequin positioned in the seat with two calibrated microphones at the ear-canals. Two methods were used to evaluate the comfort improvements achieved by active noise control. First, a prediction model of comfort based on four psychoacoustic descriptors (loudness, roughness, sharpness and tonality) was applied. Secondly, the subjective evaluation of controlled signals was carried out by means of a jury test. The reduction of noise levels does not necessarily reduce the annoyance of car engine noise that depends on spectral characteristics of noise.

Acoustic pressure waves are non-linearly superimposed when a periodic gas flow is perturbing the propagating medium, as it arises in exhaust systems. In such an environment, to estimate a certain set of source parameters from the pressure measures taken by an array of sensors is non-trivial. In this paper two possible linearization methods for the wave superposition are given. Moreover, a new problem formulation including the noise corruption of the source data is presented. Based on this linear approach, three different sound speed estimates are compared for a numerical experiment generated by thermodynamics modeling software. Results have shown that good performance of sound speed estimates mainly depends on the limited array aperture imposed by the linear model feasibility, being ML-like estimates more robust than those based in subspace decomposition methods.

A parametric method based on spatial filter techniques (beamforming) is proposed to estimate the propagation speed of acoustic waves. The propagation speed estimate is analyzed for the case of narrowband signals and compared to the maximum likelihood estimate (MLE) of the propagation speed. It is shown that for an array of 3 sensors our estimate coincides with the ML estimate but its performance analysis is simpler and its computational cost is much more reduced. The proposed estimate is also applied to the wideband waves propagating along a car exhaust. It is shown that the signal-to-noise ratio and the magnitude of the relative aperture (distance between the array sensors respect to the wavelength) for each frequency could limit the good performance of the speed estimator. Good results have been achieved when these limitations have been taken into account.

We present a new interdisciplinary application of digital signal processing and its results in comparison with the application of classical methods. The problem is the separation of two acoustic waves propagating along an exhaust pipe. It has been proved that the use of the array signal processing technique-signals are registered by an array of sensors placed along the pipe and processed by means of the LCMV beamformer-has achieved very good results under bad conditions such as uncertainty in the knowledge of the propagation speed. The array technique has also been applied to the separation of two pressure waves generated by the discharge of a fuel injection pump. In this case the results have been good but a preprocessing must be done due to the pulse characteristics of these pressure waveforms.