CN112562628A - Adaptive active noise control system based on deep neural network and method thereof - Google Patents

Adaptive active noise control system based on deep neural network and method thereof Download PDF

Info

Publication number
CN112562628A
CN112562628A CN202011416763.6A CN202011416763A CN112562628A CN 112562628 A CN112562628 A CN 112562628A CN 202011416763 A CN202011416763 A CN 202011416763A CN 112562628 A CN112562628 A CN 112562628A
Authority
CN
China
Prior art keywords
neural network
deep neural
control
signal
noise
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011416763.6A
Other languages
Chinese (zh)
Inventor
施麟
白宇田
唐俊
闫宏生
陈君
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou Silent Tai Technology Co ltd
Tianjin University
Original Assignee
Suzhou Silent Tai Technology Co ltd
Tianjin University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Suzhou Silent Tai Technology Co ltd, Tianjin University filed Critical Suzhou Silent Tai Technology Co ltd
Priority to CN202011416763.6A priority Critical patent/CN112562628A/en
Publication of CN112562628A publication Critical patent/CN112562628A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10K—SOUND-PRODUCING DEVICES; METHODS OR DEVICES FOR PROTECTING AGAINST, OR FOR DAMPING, NOISE OR OTHER ACOUSTIC WAVES IN GENERAL; ACOUSTICS NOT OTHERWISE PROVIDED FOR
    • G10K11/00—Methods or devices for transmitting, conducting or directing sound in general; Methods or devices for protecting against, or for damping, noise or other acoustic waves in general
    • G10K11/16—Methods or devices for protecting against, or for damping, noise or other acoustic waves in general
    • G10K11/175—Methods or devices for protecting against, or for damping, noise or other acoustic waves in general using interference effects; Masking sound
    • G10K11/178—Methods or devices for protecting against, or for damping, noise or other acoustic waves in general using interference effects; Masking sound by electro-acoustically regenerating the original acoustic waves in anti-phase
    • G10K11/1781—Methods or devices for protecting against, or for damping, noise or other acoustic waves in general using interference effects; Masking sound by electro-acoustically regenerating the original acoustic waves in anti-phase characterised by the analysis of input or output signals, e.g. frequency range, modes, transfer functions
    • G10K11/17813—Methods or devices for protecting against, or for damping, noise or other acoustic waves in general using interference effects; Masking sound by electro-acoustically regenerating the original acoustic waves in anti-phase characterised by the analysis of input or output signals, e.g. frequency range, modes, transfer functions characterised by the analysis of the acoustic paths, e.g. estimating, calibrating or testing of transfer functions or cross-terms
    • G10K11/17815—Methods or devices for protecting against, or for damping, noise or other acoustic waves in general using interference effects; Masking sound by electro-acoustically regenerating the original acoustic waves in anti-phase characterised by the analysis of input or output signals, e.g. frequency range, modes, transfer functions characterised by the analysis of the acoustic paths, e.g. estimating, calibrating or testing of transfer functions or cross-terms between the reference signals and the error signals, i.e. primary path
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10K—SOUND-PRODUCING DEVICES; METHODS OR DEVICES FOR PROTECTING AGAINST, OR FOR DAMPING, NOISE OR OTHER ACOUSTIC WAVES IN GENERAL; ACOUSTICS NOT OTHERWISE PROVIDED FOR
    • G10K2210/00—Details of active noise control [ANC] covered by G10K11/178 but not provided for in any of its subgroups
    • G10K2210/30—Means
    • G10K2210/301—Computational
    • G10K2210/3024—Expert systems, e.g. artificial intelligence
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10K—SOUND-PRODUCING DEVICES; METHODS OR DEVICES FOR PROTECTING AGAINST, OR FOR DAMPING, NOISE OR OTHER ACOUSTIC WAVES IN GENERAL; ACOUSTICS NOT OTHERWISE PROVIDED FOR
    • G10K2210/00—Details of active noise control [ANC] covered by G10K11/178 but not provided for in any of its subgroups
    • G10K2210/30—Means
    • G10K2210/301—Computational
    • G10K2210/3026—Feedback

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Soundproofing, Sound Blocking, And Sound Damping (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention relates to a self-adaptive active noise control system based on a deep neural network and a method thereof, which are technically characterized in that: the system comprises a reference microphone, a controller, an actuator and an error microphone, wherein the controller comprises a deep neural network module and a driving circuit; the reference microphone is arranged near the noise sound source and used for collecting a reference signal; the error microphone is arranged at the control point and used for collecting an error signal; the deep neural network module generates a control signal with the same amplitude and the opposite phase of the noise signal to be controlled, updates network parameters and outputs the generated control signal to the driving circuit; the driving circuit outputs a control signal to the actuator; the actuator converts the control signal into a control sound wave, and the control sound wave is superposed with the noise to be controlled at a control point to perform active noise elimination. The invention combines the RNN recurrent neural network and the MLP multilayer perceptron network, solves the defect that the minimum mean square error algorithm can not control the nonlinear noise, and improves the application range of the active noise control technology.

Description

Adaptive active noise control system based on deep neural network and method thereof
Technical Field
The invention belongs to the technical field of active noise control, and particularly relates to a self-adaptive active noise control system and a self-adaptive active noise control method based on a deep neural network.
Background
Noise pollution is a very interesting environmental problem worldwide. General noise interference can affect normal life and work of people, and even long-term exposure to high-noise environment can cause serious harm to psychological and physiological health of people. Noise control has therefore long been an important task.
In terms of strategy, the traditional noise control mainly takes a noise acoustic control method as a main part, and technical means comprise sound absorption treatment, sound insulation treatment, vibration isolation and reduction and the like. The mechanism of these methods is to cause the noise sound waves to interact with the acoustic material or structure to dissipate the sound energy, thereby achieving the goal of noise reduction, referred to as "passive" noise control. In general, the passive control method is effective in reducing medium-high frequency noise and has little effect on low-frequency noise. For this reason, the active noise control technique provides a new solution.
The active noise control technology is also called as an active noise reduction technology, and controls a secondary sound source to emit control sound waves with the same amplitude and opposite phases with the original noise through a reasonably designed adaptive filter, so that the control sound waves and the original noise are mutually superposed and offset at a control point by utilizing destructive interference of the sound waves, and the purpose of eliminating the noise is achieved. Currently, the Least Mean Square (LMS) algorithm based on the wiener filtering theory is most commonly used, and the LMS algorithm adjusts filter parameters by using the steepest gradient descent method to realize adaptive control on unknown noise.
The LMS algorithm as a linear control algorithm has the advantages of low computational complexity and good convergence on stable signals. Most active noise control systems are satisfactory to some extent because they have good linearity over a large range. But for noise with serious nonlinear characteristics, a linear control algorithm cannot achieve ideal control effect. Therefore, how to design a nonlinear control algorithm with good performance is a big difficulty on the development of the field of active noise control.
One solution is to use an Artificial Neural Network (ANN) technique. ANN is a new technology that has developed rapidly at the end of the 20 th century, and belongs to a branch of machine learning. The technology has the characteristics of good nonlinear mapping capability and self-adaptive learning capability, and provides a new idea for carrying out active control on nonlinear noise. However, the existing research related to the implementation of active noise control based on ANN still has major drawbacks: the fitting of the ANN to the nonlinear function depends on a larger hidden layer to perform high-dimensional expansion on the input vector, which undoubtedly greatly increases the calculation cost of the algorithm, and the convergence rate and the noise reduction effect are also strong when the network scale is small. These drawbacks make ANN technology difficult to dominate in the field of active noise control.
In 2006, the concept of deep learning was formally proposed on the basis of ANN. Deep learning, also known as Deep Neural Network (DNN), has surpassed the neuroscience view of previous machine learning models, resorting to the more general principle of learning "multi-level combinations," which can also be applied to machine learning frameworks that are not inspired by neuroscience. Among the various applications of DNN technology, Recurrent Neural Networks (RNNs) are a class of techniques for processing sequence data x(1)，…，x(τ)And the state H of the hidden layer is expressed by the following formula(t)，
H(t)＝f(H(t-1)，x(t))
＝g(t)(x(t)，x(t-1)，…，x(1))
Obviously, the function g(t)All past inputs x may be entered(t)，x(t-1)，…，x(1)Mapping to fixed length H(t)Whereas the learned model always has a fixed input size, since it specifies the transition from one state to another. This property makes the RNN require far fewer training samples than other network architectures that are not sequence based.
And through retrieval, the application of the RNN in the field of active noise control is not found.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a self-adaptive active noise control system based on a deep neural network and a method thereof, and solves the problem that a mainstream minimum mean square error algorithm is not suitable for nonlinear noise.
The technical problem to be solved by the invention is realized by adopting the following technical scheme:
an adaptive active noise control system based on a deep neural network comprises a reference microphone, a controller, an actuator and an error microphone, wherein the controller comprises a deep neural network module and a driving circuit;
the reference microphone is arranged near the noise sound source and used for collecting a reference signal and inputting the reference signal to the receiving end of the deep neural network;
the error microphone is arranged at the control point and used for collecting an error signal and inputting the error signal to the receiving end of the deep neural network;
the deep neural network module processes the reference signal, generates a control signal with the same amplitude and the opposite phase of the noise signal to be controlled, and outputs the generated control signal to the driving circuit; the deep neural network module performs gradient calculation by using a back propagation algorithm according to the error signal, and updates network parameters;
the driving circuit outputs a control signal to the actuator;
the actuator converts the control signal into a control sound wave, and the control sound wave is superposed with the noise to be controlled at a control point to perform active noise elimination.
Further, the deep neural network module is a DNN deep neural network structure with RNN cyclic neural network nested MLP multilayer perceptron nodes, the deep neural network module is composed of 3 layers of deep networks consisting of 1 input layer, 1 hidden layer and 1 output layer, wherein the input layer comprises 5 input nodes, the hidden layer consists of 10 RNN cyclic neurons and an MLP node, and the output layer comprises an output node.
Furthermore, the controller is an integrated circuit hardware chip, and a deep neural network module and a driving circuit are integrated in the hardware chip.
A control method of an adaptive active noise control system based on a deep neural network comprises the following steps:
step 1, a reference microphone arranged near a noise sound source collects a reference signal and inputs the reference signal into a deep neural network module, and an error microphone arranged at a control point collects an error signal and inputs the error signal into the deep neural network module;
step 2, the deep neural network module performs gradient calculation and updates network parameters by adopting a back propagation algorithm according to the error signal; the deep neural network module processes the reference signal and generates a control signal with the same amplitude and the opposite phase of the noise signal to be controlled;
and 3, inputting the generated control signal to the actuator through the driving circuit by the deep neural network module.
And 4, converting the control signal into a control sound wave by the actuator, and overlapping the control sound wave with the noise to be controlled at the control point to perform active noise elimination.
Further, the specific implementation method of step 2 includes the following steps:
(1) calculating the RNN neuron state H at the current moment by using the following formula(t)，
H(t)＝XV+H(t-1)W，
Wherein H(t-1)Is the RNN neuron state at the previous time; x is an input vector; v and W are weighting coefficient matrixes;
(2) the MLP node state b is calculated using the following formula,
Figure BDA0002820323880000031
wherein U is a weighting coefficient matrix;
(3) calculating an output value of the control signal at the present time using the following formula
Figure BDA0002820323880000032
Figure BDA0002820323880000033
Wherein alpha and beta are weighting coefficient matrixes;
(4) and calculating the negative gradient of each weighting coefficient matrix by using a back propagation algorithm, wherein the calculation formula is as follows:
Figure BDA0002820323880000034
Δβ＝eb，
ΔUi＝eβXi，
ΔVij＝eαjXi，
Figure BDA0002820323880000035
wherein e is a sample value of an error signal acquired at the current moment;
(5) the network parameters are updated using the following formula,
α＝α+ηΔα，
β＝β+ηΔβ，
U＝U+ηΔU，
V＝V+ηΔV，
W＝W+ηΔW
wherein η is a predetermined learning rate.
The invention has the advantages and positive effects that:
1. as an active noise control method, the invention fully utilizes the advantages of the deep learning technology in fitting the nonlinear model, solves the defect that the traditional minimum mean square error algorithm cannot control the nonlinear noise, and greatly improves the application range of the active noise control technology.
2. The method is based on the RNN recurrent neural network, fully utilizes the advantages of the RNN recurrent neural network in the field of processing time sequence signals, reduces the network scale compared with the common neural network noise reduction method, and greatly improves the calculation speed and the convergence performance.
3. The invention provides a novel deep network structure by combining an RNN (neural network) and an MLP (multi-level perceptron) multi-layer perceptron network, and the structure greatly improves the stability of the algorithm and the convergence performance on the premise of not influencing the calculation speed by adding an MLP node which does not participate in circulation in a hidden layer.
Drawings
FIG. 1 is a connection diagram of an adaptive active noise control system of the present invention;
FIG. 2 is a diagram of the deep neural network architecture of the present invention;
FIG. 3 is a comparison graph of simulation results of the present invention comparing with the conventional LMS algorithm and ANN algorithm for active control of nonlinear noise; the time domain waveform of the original noise signal is (a), (b) the time domain waveform of the error signal controlled by the traditional LMS algorithm, (c) the time domain waveform of the error signal controlled by the traditional ANN algorithm, and (d) the time domain waveform controlled by the deep neural network structure provided by the invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
The design idea of the invention is as follows:
the minimum mean square error algorithm is limited by a simple linear structure and cannot effectively control nonlinear noise; the traditional ANN artificial neural network algorithm has good nonlinear mapping capability, but has high requirements on the number of nodes of a hidden layer, so that the calculation speed is slow, and the noise real-time control is not facilitated. With the continuous development of deep learning technology, the idea of utilizing a deep neural network to research a nonlinear problem is gradually expanded into different subjects and fields. Among the many applications of this technology, the best known is the research including intelligent speech recognition and natural language processing. Such studies are mostly based on RNN recurrent neural networks, with sequence-based calculation and analysis of acoustic data, similar to the form of data that needs to be processed in the field of active noise control. The success of these studies provides a new development approach to the problem of active control of nonlinear noise.
The RNN recurrent neural network is a classic deep neural network structure, and has the shape of x(1)，…，x(τ)Sequence input of (2), which hides the state H of the layer(t)Is shown as
H(t)＝f(H(t-1)，x(t))
＝g(t)(x(t)，x(t-1)，…，x(1))
Obviously, the state of the hidden layer at the current moment in the RNN needs to be represented by the hidden layer at the previous moment, and the function g is found after the loop is expanded(t)All past inputs x may be entered(t)，x(t-1)，…，x(1)Mapping to fixed length H(t)Whereas the learned model always has a fixed input size, since it specifies the transition from one state to another. Obviously, the input layer and hidden layer may be small enough in the RNN to not lose all the previous input information. This means that the number of samples required to train the model is much smaller than for other networks, and the computational cost is much lower, which is very beneficial for the model to be used for real-time control.
However, RNN recurrent neural networks also have their drawbacks, namely gradient extinction and gradient explosion problems (vanising and expanding gradient program). This is because the RNN is to repeatedly apply the same operations at each time instant to build very deep computational graphs, and model parameters are shared. Assuming that a computation graph includes a path repeatedly multiplied by the matrix W, the multiplication after t is equivalent to the multiplication by WtIf W has a characteristic value to decompose W＝V diag(λ)V-1Then it is easy to see
Wt＝(V diag(λ)V-1)t＝V diag(λ)tV-1
When the characteristic value lambdaiIf the calculated gradient is not in the vicinity of 1, it will "disappear" if the calculated gradient is smaller than 1, and "explode" if the calculated gradient is larger than 1, and the calculated gradient will be caused by diag (λ)tVary greatly. Gradient vanishing makes it difficult to know in which direction the parameter is moving to improve the cost function, while gradient explosion makes learning unstable, these changes being particularly noticeable when dealing with long time sequences.
According to the invention, the MLP multilayer perceptron node which does not enter the circulation is added outside the RNN circulation neural network, and the gradient change of the RNN is dynamically balanced by using the node, so that the problems of gradient loss and gradient explosion in the calculation process are effectively avoided, and the stability of the whole active noise control system is enhanced.
Based on the above design concept, the present invention provides an adaptive active noise control system based on a deep neural network, as shown in fig. 1, including at least one set of reference microphone 1, controller 4, actuator 5 and error microphone 6.
The reference microphone 1 is arranged near a noise sound source and used for collecting reference signals, and the error microphone 6 is arranged at a control point and used for collecting error signals; the reference signal and the error signal are converted into electric signals and input to the input terminal of the controller 4.
The controller 4 processes the reference signal and the error signal to generate a control signal with the same amplitude and the opposite phase of the noise signal to be controlled, and inputs the control signal to the input end of the actuator 5.
In the present embodiment, the controller 4 includes a deep neural network module 2 and a driving circuit 3.
The deep neural network module 2 processes the reference signal, generates a control signal with the same amplitude and the opposite phase of the noise signal to be controlled, and outputs the generated control signal to the driving circuit 3; meanwhile, the deep neural network module 2 performs gradient calculation by using a Back Propagation (BP) algorithm according to the error signal, and updates a network parameter.
The drive circuit 3 outputs a control signal to the actuator 5.
The actuator 5 converts the control signal into a control sound wave, and the control sound wave is superposed with the noise to be controlled at the control point, so that the active noise elimination effect is achieved.
In practical applications, the controller 4 may be integrated in a hardware chip, and the deep neural network module and the driving circuit are integrated in the hardware chip.
In the invention, the deep neural network module is a novel DNN deep neural network structure with RNN cyclic neural network nested MLP multilayer perceptron nodes, as shown in FIG. 2, the DNN deep neural network structure is a 3-layer deep network composed of 1 input layer, 1 hidden layer and 1 output layer, wherein the input layer comprises 5 input nodes, the hidden layer is composed of 10 RNN cyclic neurons and an MLP node, and the output layer comprises an output node.
Based on the adaptive active noise control system based on the deep neural network, the invention also provides an adaptive active noise control method based on the deep neural network, which comprises the following steps:
step 1, a reference microphone arranged near a noise sound source collects a reference signal and inputs the reference signal into a deep neural network module, and an error microphone arranged at a control point collects an error signal and inputs the error signal into the deep neural network module.
Step 2, the deep neural network module performs gradient calculation and updates network parameters by adopting a back propagation algorithm according to the error signal; and the deep neural network module processes the reference signal and generates a control signal with the same amplitude and the opposite phase of the noise signal to be controlled.
The specific implementation method of the step comprises the following steps:
(1) calculating the state H of the RNN neuron at the current moment by using the following formula(t)，
H(t)＝XV+H(t-1)W，
Namely, it is
Figure BDA0002820323880000051
Wherein H(t-1)Is the RNN neuron state at the previous time; x is an input vector; v and W are weighting coefficient matrixes;
(2) calculating the MLP node state b using the following formula,
Figure BDA0002820323880000052
wherein U is a weighting coefficient matrix;
(3) calculating an output value of the control signal at the current time using the following formula
Figure BDA0002820323880000053
Figure BDA0002820323880000061
Wherein alpha and beta are weighting coefficient matrixes;
(4) and calculating the negative gradient of each weighting coefficient matrix by using a back propagation algorithm, wherein the calculation formula is as follows:
Figure BDA0002820323880000062
Δβ＝eb，
ΔUi＝eβXi，
ΔVij＝eαjXi，
Figure BDA0002820323880000063
wherein e is a sample value of an error signal acquired at the current moment;
(5) updating the network parameters by using the following formula,
α＝α+ηΔα，
β＝β+ηΔβ，
U＝U+ηΔU，
V＝V+ηΔV，
W＝W+ηΔW
wherein η is a predetermined learning rate.
And 3, inputting the generated control signal to the actuator through the driving circuit by the deep neural network module.
And 4, converting the control signal into a control sound wave by the actuator, and overlapping the control sound wave with the noise to be controlled at the control point to perform active noise elimination.
The adaptive active noise control system provided by the invention is simulated in a Simulink module of MATLAB, and a comparative example of the invention and LMS minimum mean square error algorithm and traditional ANN artificial neural network algorithm for actively controlling nonlinear noise is provided.
The simulation result is shown in fig. 3, and the whole graph is divided into 4 parts: (a) is the time domain waveform of the original noise signal; (b) is an error signal time domain waveform controlled by using a traditional LMS algorithm; (c) is an error signal time domain waveform controlled using a conventional ANN algorithm; (d) the learning rate in the example is 8e-3, and the value can be adjusted according to a specific noise environment or by combining other self-adaptive algorithms in actual use. Wherein, the noise to be controlled is set as a frequency conversion signal with the frequency of 200Hz to 400Hz, the sampling rate of the system is 2000Hz, and the control time is 5 seconds; the horizontal axis of the coordinate is a time axis, the vertical axis is the amplitude of the error signal, and the smaller the vibration amplitude is, the more obvious the control effect is. As can be seen from the figure, compared with the traditional algorithm, the new algorithm has obvious advantages in convergence speed and control precision, and is more suitable for research and application in a nonlinear acoustic environment.
Nothing in this specification is said to apply to the prior art.
It should be emphasized that the embodiments described herein are illustrative rather than restrictive, and thus the present invention is not limited to the embodiments described in the detailed description, but also includes other embodiments that can be derived from the technical solutions of the present invention by those skilled in the art.

Claims (5)

1. An adaptive active noise control system based on a deep neural network is characterized in that: the device comprises a reference microphone, a controller, an actuator and an error microphone, wherein the controller comprises a deep neural network module and a driving circuit;
the reference microphone is arranged near the noise sound source and used for collecting a reference signal and inputting the reference signal to the receiving end of the deep neural network;
the error microphone is arranged at the control point and used for collecting an error signal and inputting the error signal to the receiving end of the deep neural network;
the deep neural network module processes the reference signal, generates a control signal with the same amplitude and the opposite phase of the noise signal to be controlled, and outputs the generated control signal to the driving circuit; the deep neural network module performs gradient calculation by using a back propagation algorithm according to the error signal, and updates network parameters;
the driving circuit outputs a control signal to the actuator;
the actuator converts the control signal into a control sound wave, and the control sound wave is superposed with the noise to be controlled at a control point to perform active noise elimination.
2. The deep neural network-based adaptive active noise control system of claim 1, wherein: the deep neural network module is a DNN deep neural network structure with RNN cyclic neural network nested MLP multilayer perceptron nodes, and is composed of 3 layers of deep networks consisting of 1 input layer, 1 hidden layer and 1 output layer, wherein the input layer comprises 5 input nodes, the hidden layer consists of 10 RNN cyclic neurons and an MLP node, and the output layer comprises an output node.
3. The adaptive active noise control system based on the deep neural network according to claim 1 or 2, wherein: the controller is an integrated circuit hardware chip, and a deep neural network module and a driving circuit are integrated in the hardware chip.
4. A control method of the adaptive active noise control system based on the deep neural network as claimed in any one of claims 1 to 3, comprising the steps of:
step 1, a reference microphone arranged near a noise sound source collects a reference signal and inputs the reference signal into a deep neural network module, and an error microphone arranged at a control point collects an error signal and inputs the error signal into the deep neural network module;
step 2, the deep neural network module performs gradient calculation and updates network parameters by adopting a back propagation algorithm according to the error signal; the deep neural network module processes the reference signal and generates a control signal with the same amplitude and the opposite phase of the noise signal to be controlled;
and 3, inputting the generated control signal to the actuator through the driving circuit by the deep neural network module.
And 4, converting the control signal into a control sound wave by the actuator, and overlapping the control sound wave with the noise to be controlled at the control point to perform active noise elimination.
5. The method for controlling the adaptive active noise control system based on the deep neural network as claimed in claim 4, wherein: the specific implementation method of the step 2 comprises the following steps:
(1) calculating the RNN neuron state H at the current moment by using the following formula(t)，
H(t)＝XV+H(t-1)W，
Wherein H(t-1)Is the RNN neuron state at the previous time; x is an input vector; v and W are weighting coefficient matrixes;
(2) the MLP node state b is calculated using the following formula,
Figure FDA0002820323870000011
wherein U is a weighting coefficient matrix;
(3) calculating an output value of the control signal at the present time using the following formula
Figure FDA0002820323870000021
Figure FDA0002820323870000022
Wherein alpha and beta are weighting coefficient matrixes;
(4) and calculating the negative gradient of each weighting coefficient matrix by using a back propagation algorithm, wherein the calculation formula is as follows:
Figure FDA0002820323870000023
Δβ＝eb，
ΔUi＝eβXi，
ΔVij＝eαjXi，
Figure FDA0002820323870000024
wherein e is a sample value of an error signal acquired at the current moment;
(5) the network parameters are updated using the following formula,
α＝α+ηΔα，
β＝β+ηΔβ，
U＝U+ηΔU，
V＝V+ηΔV，
W＝W+ηΔW
wherein η is a predetermined learning rate.
CN202011416763.6A 2020-12-07 2020-12-07 Adaptive active noise control system based on deep neural network and method thereof Pending CN112562628A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011416763.6A CN112562628A (en) 2020-12-07 2020-12-07 Adaptive active noise control system based on deep neural network and method thereof

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011416763.6A CN112562628A (en) 2020-12-07 2020-12-07 Adaptive active noise control system based on deep neural network and method thereof

Publications (1)

Publication Number Publication Date
CN112562628A true CN112562628A (en) 2021-03-26

Family

ID=75059066

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011416763.6A Pending CN112562628A (en) 2020-12-07 2020-12-07 Adaptive active noise control system based on deep neural network and method thereof

Country Status (1)

Country Link
CN (1) CN112562628A (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107229227A (en) * 2017-06-30 2017-10-03 邢优胜 A kind of active noise controlling method based on fuzzy neural network, system and the tank helmet
US10361673B1 (en) * 2018-07-24 2019-07-23 Sony Interactive Entertainment Inc. Ambient sound activated headphone
CN110321603A (en) * 2019-06-18 2019-10-11 大连理工大学 A kind of depth calculation model for Fault Diagnosis of Aircraft Engine Gas Path
CN110970010A (en) * 2019-12-03 2020-04-07 广州酷狗计算机科技有限公司 Noise elimination method, device, storage medium and equipment
CN111091805A (en) * 2019-11-15 2020-05-01 佳禾智能科技股份有限公司 Feedback type noise reduction method based on neural network

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107229227A (en) * 2017-06-30 2017-10-03 邢优胜 A kind of active noise controlling method based on fuzzy neural network, system and the tank helmet
US10361673B1 (en) * 2018-07-24 2019-07-23 Sony Interactive Entertainment Inc. Ambient sound activated headphone
CN110321603A (en) * 2019-06-18 2019-10-11 大连理工大学 A kind of depth calculation model for Fault Diagnosis of Aircraft Engine Gas Path
CN111091805A (en) * 2019-11-15 2020-05-01 佳禾智能科技股份有限公司 Feedback type noise reduction method based on neural network
CN110970010A (en) * 2019-12-03 2020-04-07 广州酷狗计算机科技有限公司 Noise elimination method, device, storage medium and equipment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
张小平: "循环神经网络自适应有源噪声对消研究", 电讯技术, no. 5, pages 72 - 75 *

Similar Documents

Publication Publication Date Title
CN107301864B (en) Deep bidirectional LSTM acoustic model based on Maxout neuron
Back et al. FIR and IIR synapses, a new neural network architecture for time series modeling
Haykin et al. Nonlinear adaptive prediction of nonstationary signals
CN107703486A (en) A kind of auditory localization algorithm based on convolutional neural networks CNN
Er et al. Adaptive noise cancellation using enhanced dynamic fuzzy neural networks
Shi et al. Transferable latent of cnn-based selective fixed-filter active noise control
Li et al. Densely connected network with time-frequency dilated convolution for speech enhancement
CN109379652A (en) A kind of the secondary channel offline identification method and system of earphone Active noise control
Ren et al. An improved artificial bee colony algorithm for model-free active noise control: algorithm and implementation
Luo et al. Delayless generative fixed-filter active noise control based on deep learning and bayesian filter
Yu et al. An adaptive activation function for multilayer feedforward neural networks
CN112562628A (en) Adaptive active noise control system based on deep neural network and method thereof
Al-Allaf Removing noise from speech signals using different approaches of artificial neural networks
Tahaabdulsadda et al. Acoustics recognition with expert intelligent system
Pavisic et al. Active noise control with dynamic recurrent neural networks.
Englehart et al. Myoelectric signal classification using a finite impulse response neural network
Wan et al. Incremental extreme learning machine based on cascade neural networks
Sha et al. On-line adaptive learning rate BP algorithm for MLP and application to an identification problem
CN111862925B (en) Adaptive active noise control system and method based on inertia learning
Karayiannis Hybrid learning schemes for fast training of feed-forward neural networks
Xiang et al. Channel Function Modeling of ANC Based on Deep Learning network
Uwate et al. Chaotically Oscillating Sigmoid Function in Feedforward Neural Network
Liu et al. GPU Implementation of a Fast Multichannel Wiener Filter Algorithm for Active Noise Control
JP2654686B2 (en) neural network
Nasr et al. On the training of recurrent neural networks

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination