A power amplifier distortion compensation system based on deep learning

Abstract

The present invention relates to the field of audio amplification technology, and specifically discloses a power amplifier distortion compensation system based on deep learning, comprising: an input signal acquisition module, a deep learning distortion recognition module, a vacuum tube simulation module, a distortion compensation module, a PWM generation and optimization module, a power amplification module, and a feedback learning module; the system of the present invention combines a deep learning algorithm with a multi-band dynamic feature extraction method to achieve high-precision recognition of complex nonlinear distortion types such as harmonic distortion and dynamic distortion. Compared with traditional spectrum analysis or static filtering compensation, the system can dynamically adjust the compensation strategy and significantly reduce the impact of distortion on sound quality. Through a nonlinear even harmonic enhancement algorithm, the natural timbre and auditory texture of the audio signal are highly restored. Through a GAN framework, intelligent repair of signal distortion is achieved. The generator dynamically generates a high-fidelity compensation signal, and at the same time, a discriminator learns the characteristics of different distortion types to ensure that the repaired signal is close to the distortion-free target.

Description

Technical Field

[0001] The present invention belongs to the technical field of audio amplification, and in particular relates to a power amplifier distortion compensation system based on deep learning. Background Art

[0002] In the field of audio amplification, power amplifiers (PAs) are key components in audio signal transmission, responsible for amplifying low-level audio signals to a level sufficient to drive speakers. However, distortion is unavoidable during this amplification process, primarily due to the nonlinear characteristics of the equipment, thermal effects, and the physical limitations of circuit components. Distortion degrades sound quality and impacts the listener's listening experience, especially in high-fidelity audio applications.

[0003] Existing amplifier distortion compensation technologies are mostly based on traditional spectrum analysis methods or static filtering techniques. These techniques analyze the spectral characteristics of the audio signal to identify and compensate for specific distortion frequency components. However, these approaches have significant limitations. First, they are often only able to handle single or simple distortion types, such as linear distortion or static harmonic distortion. Their compensation effectiveness for complex nonlinear distortion, such as dynamic distortion or intermodulation distortion, is limited. Second, traditional methods lack the ability to dynamically adjust the compensation strategy in real time based on changes in the audio signal. This results in poor compensation results when the audio signal has a large dynamic range or complex distortion characteristics.

[0004] Furthermore, while some advanced audio processing technologies, such as digital signal processing (DSP) algorithms, can simulate the nonlinear characteristics of analog devices like vacuum tubes to a certain extent, these algorithms typically rely on fixed parameter settings and lack the ability to adapt to different audio environments and amplifier characteristics. Therefore, in practical applications, these algorithms often fail to achieve ideal sound quality restoration.

[0005] To this end, the inventors proposed a power amplifier distortion compensation system based on deep learning to solve the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide a power amplifier distortion compensation system based on deep learning to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A power amplifier distortion compensation system based on deep learning, comprising:

[0009] The input signal acquisition module is used to collect analog audio signals and convert them into digital signals through an analog-to-digital converter to obtain audio digital signals;

[0010] A deep learning distortion recognition module is used to analyze the audio digital signal using a deep learning model, identify the type of distortion produced by the power amplifier, and generate corresponding distortion compensation parameters;

[0011] a tube simulation module, configured to add an even harmonic characteristic generated by a simulated tube to the audio digital signal to generate an enhanced signal;

[0012] a distortion compensation module, configured to process the enhanced signal based on the distortion compensation parameter, eliminate distortion, restore the original characteristics of the audio signal, and generate a repaired signal;

[0013] A PWM generation and optimization module, used to convert the repair signal into a pulse width modulation signal for power amplification;

[0014] A power amplifier module, used to amplify the pulse width modulation signal and drive the speaker to produce sound and output an audio signal;

[0015] The feedback learning module is used to collect the audio signal output by the speaker, compare it with the audio digital signal, and generate feedback data for optimizing the distortion compensation capability of the deep learning model.

[0016] Preferably, the deep learning distortion recognition module adopts a multi-band dynamic feature extraction algorithm to extract distortion features, and adaptively generates distortion compensation parameters based on frequency bands and dynamic ranges.

[0017] Preferably, the formula of the multi-band dynamic feature extraction algorithm is:

[0018]

[0019] Where Dk: the distortion level of frequency band k;

[0020] Xk(t): the amplitude of the original signal in frequency band k;

[0021] Target undistorted signal predicted by deep learning;

[0022] T: sampling time window size;

[0023] α: a trade-off term used to control the contribution of dynamic changes to the degree of distortion;

[0024] Var(Xk(t)): Dynamic change of frequency band k.

[0025] Preferably, the electron tube simulation module simulates the nonlinear characteristics of the electron tube in real time through a digital signal processing algorithm, and adopts an even harmonic enhancement algorithm to dynamically generate even harmonics and enhance the naturalness of the signal's sound quality.

[0026] Preferably, the formula of the even harmonic enhancement algorithm is:

[0027] Y(t)=X(t)+β·X 2 (t)+γ·X 4 (t)

[0028] Where Y(t): output enhanced audio signal;

[0029] X(t): input audio signal;

[0030] β,γ: Weight parameters that control the intensity of even harmonics, dynamically adjusted based on deep learning.

[0031] Preferably, the distortion compensation module combines with a generative adversarial network to perform high-precision repair on the enhanced signal and dynamically compensate for harmonic distortion and noise distortion.

[0032] Preferably, the formula for generating the adversarial network is:

[0033] Generator objective function:

[0034] L G = -Ε[log(D(G(X)))]+λLrec

[0035] Discriminator objective function:

[0036] L D =-Ε[log(D(X))]-Ε[log(1-D(G(X)))]

[0037] Where G: generator, used to repair the distorted signal;

[0038] D: Discriminator, used to distinguish the repaired signal from the undistorted target signal;

[0039] X: input distorted signal;

[0040] Lrec: reconstruction loss, used to constrain the similarity between the repaired signal and the target signal;

[0041] λ: Weight parameter that controls the impact of reconstruction loss on the overall objective function.

[0042] Preferably, the PWM generation and optimization module optimizes PWM modulation parameters based on deep reinforcement learning;

[0043] The formula for deep reinforcement learning is:

[0044]

[0045] Where Q(st,at): the value function of performing action at in state st;

[0046] η: learning rate;

[0047] rt: reward after executing the action;

[0048] γ: Discount factor, used to balance short-term and long-term benefits.

[0049] Preferably, the feedback learning module updates the parameters of the deep learning model in real time through an online learning mechanism, adapts to different audio environments and power amplifier characteristics, and improves the compensation capability of the system.

[0050] Preferably, the formula for real-time updating of the online learning mechanism is:

[0051]

[0052] Where Li: loss function of the i-th task;

[0053] ωi: weight parameter used to balance the losses of each task;

[0054] R(Θ): Regularization term, used to prevent overfitting;

[0055] μ: regularization term weight;

[0056] Θ: Parameters of the deep learning model.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] (1) The system of the present invention combines deep learning algorithms with multi-band dynamic feature extraction methods to achieve high-precision recognition of complex nonlinear distortion types such as harmonic distortion and dynamic distortion. Compared with traditional spectrum analysis or static filtering compensation, the system can dynamically adjust the compensation strategy and significantly reduce the impact of distortion on sound quality.

[0059] (2) The present invention utilizes a tube simulation module and a nonlinear even harmonic enhancement algorithm to highly restore the natural timbre and auditory texture of the audio signal. The GAN framework is used to achieve intelligent repair of signal distortion. The generator dynamically generates a high-fidelity compensation signal. At the same time, the discriminator learns the characteristics of different distortion types to ensure that the repaired signal is close to the distortion-free target. This mechanism has strong adaptability and generalization capabilities and can handle a variety of complex audio scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a block diagram of a power amplifier distortion compensation system based on deep learning of the present invention. DETAILED DESCRIPTION

[0061] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0062] Example 1:

[0063] See also Figure 1 As shown, a power amplifier distortion compensation system based on deep learning includes:

[0064] An input signal acquisition module is used to collect analog audio signals and convert them into digital signals through an analog-to-digital converter (ADC) to obtain audio digital signals;

[0065] A deep learning distortion recognition module is used to analyze the audio digital signal using a deep learning model, identify the type of distortion produced by the power amplifier, and generate corresponding distortion compensation parameters;

[0066] a tube simulation module, configured to add an even harmonic characteristic generated by a simulated tube to the audio digital signal to generate an enhanced signal;

[0067] a distortion compensation module, configured to process the enhanced signal based on the distortion compensation parameter, eliminate distortion, restore the original characteristics of the audio signal, and generate a repaired signal;

[0068] A PWM generation and optimization module, used to convert the repair signal into a pulse width modulation signal for power amplification;

[0069] A power amplifier module, used to amplify the pulse width modulation signal and drive the speaker to produce sound and output an audio signal;

[0070] The feedback learning module is used to collect the audio signal output by the speaker, compare it with the audio digital signal, and generate feedback data for optimizing the distortion compensation capability of the deep learning model.

[0071] Specifically, the deep learning distortion recognition module uses a multi-band dynamic feature extraction algorithm to extract distortion features and adaptively generates distortion compensation parameters based on frequency bands and dynamic ranges;

[0072] The formula of the multi-band dynamic feature extraction algorithm is:

[0073]

[0074] Where Dk: the distortion level of frequency band k;

[0075] Xk(t): the amplitude of the original signal in frequency band k;

[0076] Target undistorted signal predicted by deep learning;

[0077] T: sampling time window size;

[0078] α: a trade-off term used to control the contribution of dynamic changes to the degree of distortion;

[0079] Var(Xk(t)): dynamic change of frequency band k;

[0080] This algorithm can accurately identify nonlinear distortion (harmonic distortion, dynamic distortion) by combining the dynamic changes of frequency bands and distortion amplitude deviations. Compared with traditional spectrum analysis methods, this method can adapt to the complex structure of different audio signals and adjust the compensation strategy in real time.

[0081] Specifically, the tube simulation module simulates the nonlinear characteristics of the tube in real time through a digital signal processing (DSP) algorithm, and uses an even harmonic enhancement algorithm to dynamically generate even harmonics and enhance the naturalness of the signal's sound quality;

[0082] The formula of the even harmonic enhancement algorithm is:

[0083] Y(t)=X(t)+β·X 2 (t)+γ·X 4 (t)

[0084] Where Y(t): output enhanced audio signal;

[0085] X(t): input audio signal;

[0086] β,γ: weight parameters that control the intensity of even harmonics, dynamically adjusted based on deep learning;

[0087] The algorithm uses nonlinear transformation to generate even harmonics, enhancing the naturalness and expressiveness of the audio signal. Compared with the static tube simulation model, deep learning dynamically adjusts the parameters β and γ to optimize the harmonic distribution according to the frequency and amplitude characteristics of the signal.

[0088] Specifically, the distortion compensation module combines a generative adversarial network (GAN) to perform high-precision repair on the enhanced signal and dynamically compensate for harmonic distortion and noise distortion;

[0089] The formula for generating the adversarial network is:

[0090] Generator objective function:

[0091] L G = -Ε[log(D(G(X)))]+λLrec

[0092] Discriminator objective function:

[0093] L D =-Ε[log(D(X))]-Ε[log(1-D(G(X)))]

[0094] Where G: generator, used to repair the distorted signal;

[0095] D: Discriminator, used to distinguish the repaired signal from the undistorted target signal;

[0096] X: input distorted signal;

[0097] Lrec: reconstruction loss, used to constrain the similarity between the repaired signal and the target signal;

[0098] λ: weight parameter, which controls the impact of reconstruction loss on the overall objective function;

[0099] Through the GAN framework, the system is able to learn complex distortion characteristics and generate high-quality compensation signals. The reconstruction loss ensures the sound quality fidelity of the repaired signal while avoiding distortion amplification caused by overcompensation.

[0100] The PWM generation and optimization module optimizes PWM modulation parameters based on deep reinforcement learning to ensure high fidelity and low distortion of pulse width modulation signals;

[0101] The formula for deep reinforcement learning is:

[0102]

[0103] Where Q(st,at): the value function of performing action at in state st;

[0104] η: learning rate;

[0105] rt: reward after executing the action;

[0106] γ: discount factor, used to balance short-term and long-term benefits;

[0107] Reinforcement learning is used to optimize PWM modulation parameters, ensuring high-fidelity output of the modulated signal in complex audio environments. Compared to traditional static modulation strategies, this method can dynamically adapt to audio characteristics, improve system efficiency, and reduce harmonic distortion.

[0108] Specifically, the feedback learning module updates the parameters of the deep learning model in real time through an online learning mechanism, adapting to different audio environments and amplifier characteristics, and improving the compensation capability of the system;

[0109] The formula for real-time updating of the online learning mechanism is:

[0110]

[0111] Where Li: loss function of the i-th task (such as distortion classification, compensation parameter generation);

[0112] ωi: weight parameter used to balance the losses of each task;

[0113] R(Θ): Regularization term, used to prevent overfitting;

[0114] μ: regularization term weight;

[0115] Θ: parameters of the deep learning model;

[0116] By combining multi-task learning with online learning mechanisms, the system can simultaneously optimize multiple tasks, such as distortion classification and compensation parameter generation, and utilize real-time feedback data to improve model adaptability. Compared with single-task learning methods, this algorithm has stronger generalization capabilities.

[0117] As can be seen from the above, the system combines deep learning algorithms and multi-band dynamic feature extraction methods to achieve high-precision identification of complex nonlinear distortion types such as harmonic distortion and dynamic distortion. Compared with traditional spectrum analysis or static filtering compensation, the system can dynamically adjust the compensation strategy, significantly reducing the impact of distortion on sound quality.

[0118] Utilizing tube simulation modules and a nonlinear even-harmonic enhancement algorithm, the natural timbre and auditory texture of audio signals are highly restored. The GAN framework enables intelligent repair of signal distortion. The generator dynamically generates a high-fidelity compensation signal, while the discriminator learns the characteristics of different distortion types to ensure that the repaired signal approaches the target of zero distortion. This mechanism has strong adaptability and generalization capabilities, capable of handling a variety of complex audio scenarios.

[0119] Reinforcement learning optimizes PWM modulation parameters to maintain low harmonic distortion and a high signal-to-noise ratio even during high-power amplification. Through a real-time feedback learning module, model parameters are dynamically updated during use, allowing the system to continuously adapt to environmental changes and user needs. This feature ensures the system's stability and efficiency over long periods of operation. Furthermore, the system demonstrates superior results in harmonic distortion control, signal-to-noise ratio enhancement, and dynamic range expansion, fully meeting a wide range of audio processing needs, from high-fidelity audio to live performances.

[0120] Example 2:

[0121] This design is specifically used in amplifier distortion compensation in high-fidelity home audio systems;

[0122] Further, system composition and parameter configuration:

[0123] Input signal acquisition module

[0124] Sampling rate: 96kHz

[0125] Quantization accuracy: 24bit

[0126] Input signal: Analog audio (20Hz–20kHz)

[0127] Deep Learning Distortion Recognition Module

[0128] Use convolutional neural network (CNN) model: 5 layers of convolution + 2 layers of full connection

[0129] Training data: including 500,000 audio data with different distortion types

[0130] Parameter configuration: Dynamically adjust the distortion compensation weight (α=0.5)

[0131] Tube simulation module

[0132] Even harmonic parameters:

[0133] β=0.03,γ=0.002

[0134] Analog Signal: Generates a signal using an adaptive harmonics algorithm to enhance the timbre.

[0135] Distortion compensation module

[0136] GAN model structure: generator (6-layer convolution + ReLU), discriminator (4-layer convolution + Sigma ID)

[0137] Reconstruction loss weight: λ = 10

[0138] Repair result: Harmonic distortion is reduced to 2% of the original value.

[0139] PWM generation and optimization module

[0140] PWM modulation frequency: 400kHz

[0141] Reinforcement learning optimization: γ = 0.9, learning rate η = 0.01

[0142] Feedback Learning Module

[0143] Real-time feedback interval: optimize model parameters every 5 minutes

[0144] Task weights: distortion classification (ω1=0.6), compensation parameter generation (ω2=0.4)

[0145] Test data and results

[0146] Test environment: 20 square meters living room, speaker frequency response range 30Hz–22kHz

[0147] Input signal: High-fidelity audio file (distortion-free reference standard)

[0148] Output signal measurement: using an audio analyzer (distortion, frequency response)

[0149] As shown in Table 1 below

[0150] Test items Before optimization After optimization Total harmonic distortion (THD) 3.5％ 0.5％ Signal-to-noise ratio (SNR) 85dB 96dB Tone restoration 80％ 98％ User satisfaction medium to high Very high

[0151] Table 1

[0152] As can be seen from the above, by implementing this system, the sound quality of high-fidelity audio equipment has been greatly improved, the distortion has been significantly reduced, and the user's feedback on the bass and mid-high frequencies has become more natural.

[0153] Example 3:

[0154] Another aspect of this design is applied to amplifier distortion compensation in live performance audio processing;

[0155] Among them, system composition and parameter configuration:

[0156] Input signal acquisition module

[0157] Sampling rate: 48kHz

[0158] Quantization accuracy: 16bit

[0159] Input signal: Live mixed audio signal (large dynamic range)

[0160] Deep Learning Distortion Recognition Module

[0161] Using the Transformer Model

[0162] Training data: includes 100,000 samples of common audio distortions in live performances

[0163] Parameter configuration: dynamic range compression ratio α = 0.7

[0164] Tube simulation module

[0165] Even harmonic parameters:

[0166] β=0.05,γ=0.005

[0167] Simulates the distortion characteristics of real tubes and enhances the mid-frequency expression.

[0168] Distortion compensation module

[0169] GAN model structure: generator (7 layers of convolution + BatchNorm), discriminator (3 layers of full connection + Dropout) reconstruction loss weight: λ = 5

[0170] Restoration result: Dynamic distortion reduced to 1%.

[0171] PWM generation and optimization module

[0172] PWM modulation frequency: 500kHz

[0173] Reinforcement learning optimization: γ = 0.8, learning rate η = 0.02

[0174] Feedback Learning Module

[0175] Real-time feedback interval: Optimize model parameters every 30 seconds Task weights: distortion classification (ω1=0.5), compensation parameter generation (ω2=0.5)

[0176] Test data and effects Test environment: medium-sized performance venue with 500 people, maximum output power of the amplifier is 1000W

[0177] Input signal: multi-track mixed input signal (drums, guitar, vocals, bass)

[0178] Output signal measurement: Frequency response and distortion testing using on-site audio test equipment

[0179] As shown in Table 2 below

[0180] Test items Before optimization After optimization Total harmonic distortion (THD) 7.0％ 1.0％ Signal-to-noise ratio (SNR) 75dB 88dB Sound pressure level uniformity ±6dB ±2dB User satisfaction medium high

[0181] Table 2

[0182] From the above, we can see that the system effectively solves the dynamic distortion problem commonly found in live high-power amplification, while optimizing sound uniformity and dynamic performance. It performs particularly well in processing vocal and instrument reverberation, significantly enhancing the audience's auditory experience.

[0183] In the description of this specification, the reference terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0184] In the drawings of the embodiments disclosed in the present invention, only the structures related to the embodiments disclosed in the present invention are involved. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0185] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A power amplifier distortion compensation system based on deep learning, characterized in that: include: The input signal acquisition module is used to collect analog audio signals and convert them into digital signals through an analog-to-digital converter to obtain audio digital signals; A deep learning distortion recognition module is used to analyze the audio digital signal using a deep learning model, identify the type of distortion produced by the power amplifier, and generate corresponding distortion compensation parameters; a tube simulation module, configured to add an even harmonic characteristic generated by a simulated tube to the audio digital signal to generate an enhanced signal; a distortion compensation module, configured to process the enhanced signal based on the distortion compensation parameter, eliminate distortion, restore the original characteristics of the audio signal, and generate a repaired signal; A PWM generation and optimization module, used to convert the repair signal into a pulse width modulation signal for power amplification; A power amplifier module, used to amplify the pulse width modulation signal and drive the speaker to produce sound and output an audio signal; The feedback learning module is used to collect the audio signal output by the speaker, compare it with the audio digital signal, and generate feedback data for optimizing the distortion compensation capability of the deep learning model.

2. The power amplifier distortion compensation system based on deep learning according to claim 1, characterized in that: The deep learning distortion recognition module uses a multi-band dynamic feature extraction algorithm to extract distortion features and adaptively generates distortion compensation parameters based on frequency bands and dynamic ranges.

3. The power amplifier distortion compensation system based on deep learning according to claim 2, characterized in that: The formula of the multi-band dynamic feature extraction algorithm is: Where Dk: the distortion level of frequency band k; Xk(t): the amplitude of the original signal in frequency band k; Target undistorted signal predicted by deep learning; T: sampling time window size; α: a trade-off term used to control the contribution of dynamic changes to the degree of distortion; Var(Xk(t)): Dynamic change of frequency band k.

4. The power amplifier distortion compensation system based on deep learning according to claim 1, characterized in that: The tube simulation module simulates the nonlinear characteristics of the tube in real time through a digital signal processing algorithm, and adopts an even harmonic enhancement algorithm to dynamically generate even harmonics and enhance the naturalness of the signal's sound quality.

5. The power amplifier distortion compensation system based on deep learning according to claim 4, characterized in that: The formula of the even harmonic enhancement algorithm is: Y(t)=X(t)+β·X 2 (t)+γ1·X 4 (t) Where Y(t): output enhanced audio signal; X(t): input audio signal; β,γ1: Weight parameters that control the intensity of even harmonics, dynamically adjusted based on deep learning.

6. The power amplifier distortion compensation system based on deep learning according to claim 1, characterized in that: The distortion compensation module combines with the generative adversarial network to perform high-precision repair on the enhanced signal and dynamically compensate for harmonic distortion and noise distortion.

7. The power amplifier distortion compensation system based on deep learning according to claim 6, characterized in that: The formula for generating the adversarial network is: Generator objective function: L G =-Ε[log(D(G(X)))]+λLrec Discriminator objective function: L D =-Ε[log(D(X))]-Ε[log(1-D(G(X)))] Where G: generator, used to repair the distorted signal; D: Discriminator, used to distinguish the repaired signal from the undistorted target signal; X: input distorted signal; Lrec: reconstruction loss, used to constrain the similarity between the repaired signal and the target signal; λ: Weight parameter that controls the impact of reconstruction loss on the overall objective function.

8. The power amplifier distortion compensation system based on deep learning according to claim 1, characterized in that: The PWM generation and optimization module optimizes PWM modulation parameters based on deep reinforcement learning; The formula for deep reinforcement learning is: Where Q(st,at): the value function of performing action at in state st; η: learning rate; rt: reward after executing the action; γ2: Discount factor, used to balance short-term and long-term benefits.

9. The power amplifier distortion compensation system based on deep learning according to claim 1, characterized in that: The feedback learning module updates the parameters of the deep learning model in real time through an online learning mechanism, adapting to different audio environments and power amplifier characteristics, and improving the compensation capability of the system.

10. The power amplifier distortion compensation system based on deep learning according to claim 9, characterized in that: The formula for real-time updating of the online learning mechanism is: Where Li: loss function of the i-th task; ωi: weight parameter used to balance the losses of each task; R(Θ): Regularization term, used to prevent overfitting; μ: regularization term weight; Θ: Parameters of the deep learning model.

Images