Multi-modal analog signal digitization method and system based on adaptive sampling

Abstract

The invention provides a multi-mode analog signal digitization method and system based on adaptive sampling, and the method comprises the steps: collecting a multi-mode analog signal, and carrying out the preprocessing of the multi-mode analog signal; obtaining a corresponding signal type based on the processed analog signal, and adaptively adjusting the sampling frequency according to the signal type; and adaptive quantization and quantization error compensation are carried out on the adjusted analog signal, and a final digital signal is output after enhancement processing is carried out on the digitized signal. The invention provides an improved analog signal and digital signal conversion method, namely a multi-modal analog signal high-precision digitalization method based on adaptive sampling and deep learning, and aims to improve conversion precision, reduce noise interference, optimize power consumption and response time and expand a dynamic range.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and in particular to a multi-modal analog signal digitization method and system based on adaptive sampling. Background Art

[0002] In fields such as industrial automation, the Internet of Things, and smart homes, various physical quantities need to be monitored, such as vibration, speed, temperature, current, voltage, and sound. These physical quantities are typically converted into analog signals by sensors, then converted to digital signals via an analog-to-digital converter (ADC), and ultimately transmitted to a host computer via a communication module for processing and analysis. While existing technologies for converting analog signals to digital signals are quite mature, they still have drawbacks and limitations in certain application scenarios. Here are some common issues:

[0003] 1. Sampling frequency limitations: According to the Nyquist sampling theorem, to reconstruct a signal without distortion, the sampling frequency must be at least twice the highest frequency component. However, in practical applications, due to hardware performance or cost limitations, it may not be possible to achieve the ideal sampling rate. This can lead to problems such as signal distortion and spectral aliasing, especially when processing high-frequency signals.

[0004] 2. Quantization error: Analog signals need to be quantized when converted to digital signals, that is, continuous analog values are mapped to a finite number of discrete levels. This process inevitably introduces quantization error, thereby reducing the accuracy and fidelity of the signal, especially in applications with low amplitude signals or high resolution requirements.

[0005] 3. Noise sensitivity: Analog-to-digital converters (ADCs) are very sensitive to environmental noise. Any external interference (such as electromagnetic interference, power supply fluctuations, etc.) may be introduced into the digital signal, thereby increasing the signal-to-noise ratio (SNR) and reducing the quality of the useful signal. Especially in low SNR environments, noise will seriously affect the validity and reliability of the data.

[0006] 4. Linearity and nonlinear distortion: The ADC used in practice may have non-ideal linearity, that is, the input-output relationship is not completely linear. In addition, nonlinear distortion may occur under overload conditions, causing signal waveform deformation, harmonic distortion, and other types of distortion, thereby affecting the accuracy of the final result.

[0007] 5. Limited dynamic range: Most ADCs have a fixed bit width (e.g., 8-bit, 12-bit, 16-bit, etc.), which determines the difference between the maximum and minimum values they can represent (i.e., the dynamic range). When the signal strength varies greatly, it may exceed the ADC's range, causing clipping distortion; or for weak signals, the resolution is insufficient, and subtle changes cannot be accurately captured.

[0008] 6. Power consumption and heat generation: High-speed, high-precision ADCs typically consume more power and generate more heat during operation. This increases system energy consumption and limits their application in portable or battery-powered devices. Overheating can also reduce device lifespan and stability.

[0009] 7. Latency and response time: Some ADC designs have a certain time delay between analog signal input and digital signal output, as well as long setup and recovery times. In real-time control systems or other latency-sensitive applications, this delay can lead to degraded control performance or system instability.

[0010] 8. High cost: High-performance ADC modules are expensive, especially those with extremely high accuracy, speed, and noise immunity. This increases the cost of the overall system and limits its widespread application in low-cost or large-scale production.

[0011] 9. Calibration complexity: To ensure ADC accuracy, complex calibration procedures are often required on a regular basis, including temperature compensation and zero drift correction. This increases maintenance difficulty and cost, especially in systems operating in harsh environments, where frequent calibration becomes a major challenge.

[0012] 10. Bandwidth and throughput limitations: Traditional ADC architectures may face throughput bottlenecks when processing broadband or multi-channel signals, failing to meet the demands of high-speed data acquisition. This limits system scalability and flexibility, making it difficult to adapt to future applications requiring higher bandwidth.

[0013] In summary, although existing analog-to-digital conversion technology has made great progress, there is still room for improvement in certain specific areas. Summary of the Invention

[0014] The object of the present invention is to provide a multimodal analog signal digitization method and system based on adaptive sampling, aiming to solve the above-mentioned problems in the prior art.

[0015] An embodiment of the present invention provides a multimodal analog signal digitization method based on adaptive sampling, comprising:

[0016] Collecting multimodal analog signals and preprocessing the multimodal analog signals;

[0017] Obtaining a corresponding signal type based on the processed analog signal, and adaptively adjusting the sampling frequency according to the signal type; and

[0018] The adjusted analog signal is adaptively quantized and quantization error compensated, and the digitized signal is enhanced to output the final digital signal.

[0019] An embodiment of the present invention provides a multimodal analog signal digitization system based on adaptive sampling, comprising:

[0020] A processing module, configured to collect multimodal analog signals and preprocess the multimodal analog signals;

[0021] an adaptive sampling module, configured to obtain a corresponding signal type based on the processed analog signal and adaptively adjust the sampling frequency according to the signal type; and

[0022] The adaptive quantization module is used to perform adaptive quantization and quantization error compensation on the adjusted analog signal, and output the final digital signal after enhancing the digitized signal.

[0023] An embodiment of the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the above-mentioned multimodal analog signal digitization method based on adaptive sampling.

[0024] An embodiment of the present invention further provides a computer-readable storage medium, on which a program for implementing information transmission is stored. When the program is executed by a processor, the steps of the above-mentioned multi-modal analog signal digitization method based on adaptive sampling are implemented.

[0025] The use of the embodiments of the present invention may include the following beneficial effects: The embodiments of the present invention propose an improved method for converting analog signals to digital signals, namely, a high-precision digitization method for multimodal analog signals based on adaptive sampling and deep learning, which aims to improve conversion accuracy, reduce noise interference, optimize power consumption and response time, and expand dynamic range. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 is a flow chart of a multimodal analog signal digitization method based on adaptive sampling according to an embodiment of the present invention;

[0028] Figure 2 Schematic diagram of a multimodal analog signal digitization system based on adaptive sampling according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this document.

[0030] Method Example

[0031] According to an embodiment of the present invention, a multimodal analog signal digitization method based on adaptive sampling is provided. Figure 1 FIG. 1 is a flow chart of a multimodal analog signal digitization method based on adaptive sampling according to an embodiment of the present invention. Figure 1 As shown, the multimodal analog signal digitization method based on adaptive sampling according to an embodiment of the present invention specifically includes:

[0032] Step S101, collecting multimodal analog signals and preprocessing the multimodal analog signals;

[0033] Step S102, obtaining a corresponding signal type based on the processed analog signal, and adaptively adjusting the sampling frequency according to the signal type, specifically includes:

[0034] Extract features from the processed analog signal, input the extracted features into a deep learning model for training to obtain a trained type recognition model, and input the processed analog signal into the trained type recognition model for recognition to obtain the corresponding signal type;

[0035] Using a pre-built adaptive sampling frequency prediction model to predict the multimodal analog signal to obtain an optimal sampling frequency, establishing a mapping relationship table between various types of signals and the optimal sampling frequencies, searching the corresponding optimal sampling frequency from the mapping relationship table according to the signal type, and adaptively adjusting the sampling frequency of the multimodal analog signal using the corresponding optimal sampling frequency;

[0036] Step S103, performing adaptive quantization and quantization error compensation on the adjusted analog signal, and performing enhancement processing on the digitized signal to output a final digital signal, specifically includes:

[0037] Inputting the adjusted analog signal into a pre-built neural network-based quantization model for prediction, obtaining an optimal quantization level and a quantization error estimate of the signal, adaptively quantizing the signal according to the optimal quantization level, and introducing a back-propagation error correction mechanism during the quantization process to correct the quantization error estimate in real time through the back-propagation error correction mechanism;

[0038] The digitized signal is corrected using a nonlinear correction algorithm, and the corrected digital signal is input into a pre-built deep learning enhancement model for enhancement processing to obtain the final digital signal;

[0039] The nonlinear correction algorithm is a nonlinear correction algorithm based on the Fourier transform principle and combined with an iterative optimization method.

[0040] The above technical solution of the embodiment of the present invention is described in detail below in conjunction with the specific situation of the multimodal analog signal digitization method based on adaptive sampling in the embodiment of the present invention.

[0041] The present invention aims to provide an improved method for converting analog signals into digital signals, which specifically includes the following aspects:

[0042] 1. Adaptive sampling frequency adjustment: Through intelligent algorithms, the frequency characteristics of the input signal are monitored in real time and the sampling frequency is dynamically adjusted to ensure that sufficient sampling points are obtained in different frequency bands to avoid spectrum aliasing.

[0043] The implementation method is to use a machine learning model to predict the optimal sampling frequency and fine-tune it in conjunction with a feedback mechanism. The specific steps include:

[0044] 1. Data collection and preprocessing

[0045] A. Objective: Collect enough training data to ensure that the machine learning model can accurately predict the optimal sampling frequency for different application scenarios.

[0046] B. Steps:

[0047] Collect samples: In various practical application environments, collect a large number of analog signal samples containing different frequency components and intensities; select corresponding sensors according to monitoring requirements, such as accelerometers, temperature sensors, pressure sensors, etc. The multimodal analog signals collected by the embodiment of the present invention include but are not limited to vibration, rotation, speed, acceleration, distance, angle, stress, strain, temperature, liquid level, pressure, current, voltage, resistance, potential, optical, infrared, sound, etc.

[0048] Label data: Mark each sample with its corresponding optimal sampling frequency (obtained through theoretical calculation or experimental verification); that is, set the appropriate sampling frequency based on the signal characteristics. For example, high-frequency vibration signals require a higher sampling frequency.

[0049] Feature extraction: Extract useful features from the original signal, such as spectrum distribution, instantaneous frequency, amplitude change rate, etc., as model input.

[0050] 2. Choose the right machine learning model

[0051] A. Objective: Select a machine learning model suitable for solving regression problems to predict the sampling frequency of continuous values.

[0052] B. Model Type:

[0053] Random forest regression: Suitable for nonlinear relationship modeling and has strong generalization ability.

[0054] Support Vector Machine (SVM) regression: Suitable for processing high-dimensional data and small sample sizes.

[0055] Deep Neural Network (DNN): It performs well in complex pattern recognition and nonlinear mapping.

[0056] 3. Training Machine Learning Models

[0057] A. Objective: Use the collected data to train the selected machine learning model so that it can predict the sampling frequency closest to the true value based on the input features.

[0058] B. Steps:

[0059] Divide the data set: Divide the data into training set, validation set, and test set to ensure the fairness and reliability of model evaluation.

[0060] Hyperparameter tuning: Find the optimal hyperparameter combination for the model through grid search or random search.

[0061] Model training: Use the data in the training set to fit the model parameters, while monitoring the performance indicators on the validation set to prevent overfitting.

[0062] 4. Integrated feedback mechanism

[0063] A. Goal: By monitoring the system output in real time, dynamically adjust the sampling frequency to ensure continuous performance optimization.

[0064] B. Steps:

[0065] Deploy an online learning framework: This allows the model to continuously receive new data during operation and update its own parameters accordingly.

[0066] Define feedback variables: Determine which performance indicators can reflect whether the current sampling frequency is appropriate, such as signal-to-noise ratio (SNR) and total harmonic distortion (THD).

[0067] Set thresholds and rules: When certain key indicators are detected to be outside the preset range, a reassessment process is triggered to re-predict the optimal sampling frequency using the latest data.

[0068] Smooth transition strategy: In order to avoid the negative impact of frequent switching, a gradual or weighted average method is used when adjusting the sampling frequency to ensure a smooth transition.

[0069] 5. Testing and Verification

[0070] A. Objective: To verify the stability and effectiveness of the improved ADC system under various conditions.

[0071] B. Steps:

[0072] Lab testing: In a controlled environment, we simulate different work scenarios and compare the performance of traditional methods with new methods.

[0073] Field trials: Select typical application cases for long-term follow-up observation, record actual effects, and collect user feedback.

[0074] Performance evaluation: Based on objective measurement data and subjective evaluation criteria, comprehensively evaluate the advantages and disadvantages of the new method.

[0075] 6. Continuous Improvement

[0076] A. Goal: Continuously optimize models and algorithms as technology and application scenarios evolve.

[0077] B. Steps:

[0078] Regular maintenance: Regularly review and update training data to ensure the model is always up to date.

[0079] User experience research: Gain an in-depth understanding of changes in user needs and adjust the model in a timely manner.

[0080] By following these steps, we can build a more intelligent, flexible, and efficient analog-to-digital conversion system that can not only automatically predict the optimal sampling frequency but also continuously optimize itself through a feedback mechanism, thereby significantly improving the quality and efficiency of signal processing.

[0081] 2. Multi-level quantization and error compensation: Multi-level quantization technology is used to map continuous analog values to discrete levels in stages, and an error compensation mechanism is introduced to correct deviations in the quantization process.

[0082] The implementation method is to design a neural network-based quantizer that can automatically select the appropriate quantization level based on the input signal characteristics and correct the error through the backpropagation algorithm. The specific steps include:

[0083] 1. Demand analysis and solution design

[0084] A. Objective: To build an intelligent quantization system that can dynamically adjust the quantization level according to the characteristics of the input signal and minimize the quantization error using the backpropagation algorithm.

[0085] B. Steps:

[0086] Determine the application scenarios and technical requirements (such as real-time performance, accuracy, etc.).

[0087] Design the overall architecture of the system, including the front-end preprocessing module, the core neural network model, and the back-end feedback adjustment mechanism.

[0088] 2. Build a dataset

[0089] A. Objective: Obtain high-quality data for training and validating neural network models.

[0090] B. Steps:

[0091] Collect or generate different types of analog signal samples, covering a wide range of application scenarios.

[0092] Each sample is carefully labeled, and its ideal quantization level and corresponding quantization error are recorded.

[0093] The dataset is divided into training, validation, and test sets to ensure that the data distribution is even and representative.

[0094] 3. Design a neural network model

[0095] A. Objective: Build an efficient neural network architecture to predict the optimal quantization level and estimate the quantization error.

[0096] B Model Type:

[0097] Multilayer Perceptron (MLP): Suitable for simple linear or nonlinear mapping problems.

[0098] Convolutional Neural Network (CNN): It works well for time series data or signals with rich frequency domain information.

[0099] Long Short-Term Memory Network (LSTM) / Recurrent Neural Network (RNN): Suitable for processing time series data and capturing historical dependencies.

[0100] C. Key components:

[0101] The input layer receives the preprocessed signal features.

[0102] The hidden layer contains multiple fully connected layers or convolutional layers, using ReLU or other activation functions.

[0103] The output layer is divided into two parts: one part outputs the predicted quantization level; the other part outputs the quantization error estimate.

[0104] 4. Model training

[0105] A. Objective: Use labeled data to train a neural network model so that it can accurately predict the quantization level and estimate the error given an input signal.

[0106] B. Steps:

[0107] Define a loss function, such as mean squared error (MSE), to measure the difference between the predicted value and the true value.

[0108] Apply optimization algorithms such as stochastic gradient descent (SGD) and Adam to update model weights.

[0109] Regularization techniques (such as Dropout and L2 regularization) are introduced during the training process to prevent overfitting.

[0110] Use the validation set to monitor model performance and adjust hyperparameters to achieve optimal results.

[0111] 5. Integrated back-propagation error correction mechanism

[0112] A. Objective: Introduce a feedback loop into the quantization process and use the backpropagation algorithm to correct the quantization error in real time.

[0113] B. Steps:

[0114] The current quantized output is compared with the expected output, and the immediate error is calculated. This error is fed back to the neural network as additional input for the next round of prediction. The learning rate is adjusted based on the error to accelerate convergence while maintaining stability. The impact of accumulated error is taken into account when updating model parameters to ensure long-term accuracy.

[0115] 6. System testing and evaluation

[0116] A.Objective: To verify the performance of the designed adaptive quantizer in practical applications.

[0117] B. Steps:

[0118] An independent test set was used to evaluate the generalization ability and robustness of the model.

[0119] Simulate the operating status under various extreme conditions and examine the system's ability to cope with sudden changes.

[0120] Gather user feedback, identify potential issues, and suggest improvements.

[0121] 7. Deployment and Maintenance

[0122] A. Goal: Apply the trained model to the real environment and continuously optimize its performance.

[0123] B. Steps:

[0124] Implement an automated monitoring system to promptly detect and resolve potential problems.

[0125] New data is collected regularly for model retraining and iterative upgrades.

[0126] Through the above steps, we build an intelligent quantizer based on a neural network. It can not only automatically select the optimal quantization level according to the characteristics of the input signal, but also continuously correct the error through the backpropagation algorithm, thereby providing more accurate and stable quantization services.

[0127] 3. Active noise suppression circuit: Integrates active noise suppression function and adds a noise cancellation module to the ADC front end to effectively filter out external interference signals and improve the signal-to-noise ratio.

[0128] The implementation method is to use adaptive filters and electromagnetic shielding materials to build an efficient anti-noise barrier.

[0129] 4. Nonlinear correction algorithm: To address the linear and nonlinear distortion problems that may occur in ADC, a nonlinear correction algorithm is used to make the output signal closer to the true value.

[0130] The implementation method is to establish a mathematical model based on the Fourier transform principle and find the optimal solution through iterative calculation to achieve accurate nonlinear correction. The specific steps include the following:

[0131] 1. Data collection and preprocessing

[0132] Input signal acquisition: Get digitized samples of the original analog signal from the ADC. These samples usually contain a certain degree of nonlinear distortion.

[0133] Data cleaning: Perform preliminary processing on the collected data, such as removing outliers and filtering, to ensure the effectiveness of subsequent analysis.

[0134] 2. Fourier transform analysis

[0135] Frequency domain conversion: Converting the time domain signal into the frequency domain through fast Fourier transform (FFT) helps to identify different frequency components and their corresponding amplitude and phase information.

[0136] Spectral feature extraction: Select the main frequency range of interest based on the actual application scenario and record its characteristic parameters (such as peak position, bandwidth, etc.) for subsequent modeling.

[0137] 3. Mathematical model construction

[0138] Nonlinear effect modeling: It is assumed that the nonlinear effects present in the ADC can be described by a function of a specific form (such as a polynomial or piecewise linear function). This function can capture the various types of nonlinear behaviors observed in real systems.

[0139] Parameter estimation: Use a known standard test signal (such as a sine wave or other signal with well-defined characteristics) combined with the least squares method or other optimization algorithm to determine the best fitting parameters of the above nonlinear model.

[0140] 4. Iterative Optimization Process

[0141] Initial guess: Set a reasonable set of starting parameters as the basis for the first iteration.

[0142] Residual calculation: For each given set of parameters, the difference between the model prediction and the true measurement (i.e., the residual) is calculated, and the goal is to minimize this difference.

[0143] Update rule: According to the selected optimization strategy (such as gradient descent, Newton method, etc.), adjust the current parameter vector so that the residual is smaller in the next iteration.

[0144] Convergence judgment: When the improvement after several consecutive iterations is less than the predetermined threshold, or when the maximum number of allowed iterations is reached, the optimization process is stopped.

[0145] 5. Nonlinear correction implementation

[0146] Correction coefficient generation: Based on the optimal parameter set obtained by final convergence, the correction coefficient used to compensate for nonlinear errors is calculated.

[0147] Real-time application: Whenever a new input signal arrives, these correction coefficients are used to modify the output result to obtain a more accurate digital signal representation.

[0148] 6. Performance Verification

[0149] Validation testing: A series of independent tests (with different data from the training phase) are used to evaluate whether the calibrated ADC performance meets the expected standards.

[0150] Feedback adjustment: If significant errors are still found under certain specific conditions, the model structure or parameter settings can be further fine-tuned as needed until the overall performance reaches a satisfactory level.

[0151] Through the above steps, a nonlinear correction scheme based on the Fourier transform principle and combined with iterative optimization technology is effectively established, which significantly improves the accuracy and reliability of the ADC in various complex working environments.

[0152] 5. Extended dynamic range: By introducing a variable gain amplifier (VGA) and overload protection circuit, the system can operate normally within a wider range of signal strengths, improving the ability to capture weak signals.

[0153] The implementation method is: designing the VGA structure so that it can automatically adjust the gain according to the input signal amplitude, and cooperating with the overload protection circuit to prevent clipping distortion.

[0154] 6. Low-power optimized design: Using advanced CMOS technology and power management strategies to reduce power consumption and heat generation, suitable for portable or battery-powered devices.

[0155] The implementation method is: optimizing circuit layout, adopting energy-saving mode switching technology and efficient heat dissipation solution.

[0156] 7. Zero-delay transmission architecture: Design a zero-delay transmission architecture to greatly shorten the time difference from analog signal input to digital signal output, and improve the response speed of real-time control systems.

[0157] The implementation method is to use ultra-high-speed cache technology and parallel processing units to achieve instant data transmission.

[0158] 8. Low-cost manufacturing process: Simplify the production process, reduce costs, and facilitate large-scale production.

[0159] The implementation method is: adopt standardized components and modular design concepts to reduce customized parts and improve production efficiency.

[0160] 9. Automated calibration system: Design a fully automated calibration system to reduce manual intervention and ensure long-term stability.

[0161] The implementation method is: built-in self-test program and remote maintenance interface, supporting online update and parameter adjustment.

[0162] 10. Enhanced bandwidth and throughput: Optimizes internal data flow paths to meet the needs of broadband and multi-channel signal processing, improving the overall throughput of the system.

[0163] The implementation method is: introducing a distributed data processing engine that can process data from multiple channels in parallel to meet future application requirements for higher bandwidth.

[0164] Based on the analysis of the above aspects, the present invention provides a method for high-precision digitization of multimodal analog signals based on adaptive sampling and deep learning to improve the accuracy and anti-interference capability of analog signal digitization. Specifically, the method includes the following steps:

[0165] 1. Signal preprocessing: Preprocess the collected multimodal analog signals, including denoising, amplification, filtering, etc., to improve signal quality.

[0166] 1.1 Signal Acquisition: Analog signals are acquired using various types of sensors, such as accelerometers and microphones. The conditions monitored by embodiments of the present invention include, but are not limited to, vibration, rotation, speed, acceleration, distance, angle, stress, strain, temperature, liquid level, pressure, current, voltage, potential, optical, infrared, and acoustic signals.

[0167] 1.2 Signal amplification: Use an operational amplifier to amplify the collected weak signal to improve the signal-to-noise ratio.

[0168] 1.3 Signal filtering: Use filters to remove noise from the signal, such as using a low-pass filter to remove high-frequency noise, and using a band-pass filter to remove noise of a specific frequency.

[0169] 2. Signal type recognition: Use the deep learning model to identify the type of preprocessed signal and determine whether the signal belongs to the speed, acceleration, temperature, stress, sound, etc.

[0170] 2.1 Feature extraction: Extract features from the preprocessed signal, such as time domain features (mean, variance, peak, etc.), frequency domain features (spectrum, power spectrum, etc.), time-frequency domain features (wavelet transform, short-time Fourier transform, etc.), etc.

[0171] 2.2 Model training: Use the extracted features to train deep learning models, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), to identify signal types.

[0172] 2.3 Signal Recognition: The preprocessed signal is input into the trained deep learning model to identify the signal type, such as speed, acceleration, temperature, stress, sound, etc.

[0173] 3. Adaptive sampling: Adaptively adjust the sampling frequency according to the identified signal type.

[0174] 3.1 Establish a mapping relationship: Establish a mapping relationship between signal type and optimal sampling frequency. For example, the optimal sampling frequency of speed signal is 100Hz, and the optimal sampling frequency of sound signal is 44.1kHz.

[0175] 3.2 Adaptive adjustment: According to the identified signal type, find the corresponding optimal sampling frequency from the mapping relationship and adjust the sampling frequency.

[0176] 4. Adaptive quantization: In addition to the above-mentioned adaptive quantization of signals using the network model, the embodiment of the present invention can also adaptively adjust the number of quantization bits according to the identified signal type and signal dynamic range.

[0177] 4.1 Dynamic range calculation: Calculate the dynamic range of the signal, that is, the difference between the maximum and minimum values of the signal.

[0178] 4.2 Determination of the number of quantization bits: The number of quantization bits is determined based on the dynamic range of the signal. For example, 8-bit quantization can be used for signals with a smaller dynamic range, and 16-bit quantization can be used for signals with a larger dynamic range.

[0179] 4.3 Quantization error compensation: Use nonlinear quantization, adaptive quantization and other methods to compensate for quantization errors and improve quantization accuracy.

[0180] 5. Deep learning enhancement: Use deep learning models to enhance the digitized signal.

[0181] 5.1 Model training: Use large amounts of data to train deep learning models, such as generative adversarial networks (GANs) and autoencoders (AEs), to learn the characteristics and distribution of signals.

[0182] 5.2 Signal Enhancement: The digitized signal is input into a trained deep learning model for signal enhancement, such as removing noise, compensating for distortion, and improving resolution.

[0183] 6. Data output: Output the processed digital signal to subsequent data processing and analysis systems, such as data storage systems, data transmission systems, data analysis systems, etc.

[0184] Preferably, the method further comprises: establishing a mapping relationship between the signal type and the optimal sampling frequency and quantization bit number, and performing adaptive sampling and quantization according to the mapping relationship.

[0185] A specific example of analog-to-digital conversion using the embodiment of the present invention is as follows:

[0186] 1. Use a microphone to collect sound signals and preprocess the signals, including denoising, amplification, filtering, etc.

[0187] 2. Use the trained CNN model to identify the preprocessed sound signal and determine whether it is a sound signal.

[0188] 3. According to the characteristics of the sound signal, set the sampling frequency to 44.1kHz.

[0189] 4. According to the dynamic range of the sound signal, set the quantization bit number to 16 bits.

[0190] 5. Use the trained GAN model to enhance the digitized sound signal, remove quantization noise, compensate for signal distortion, etc.

[0191] 6. Output the processed digital sound signal to the subsequent speech recognition system.

[0192] System Example

[0193] According to an embodiment of the present invention, a multimodal analog signal digitization system based on adaptive sampling is provided. Figure 2 FIG is a schematic diagram of a multimodal analog signal digitization system based on adaptive sampling according to an embodiment of the present invention. Figure 2 As shown, the multimodal analog signal digitization system based on adaptive sampling according to an embodiment of the present invention specifically includes:

[0194] The processing module 20 is used to collect multimodal analog signals and pre-process the multimodal analog signals;

[0195] The adaptive sampling module 22 is used to obtain a corresponding signal type based on the processed analog signal and adaptively adjust the sampling frequency according to the signal type, specifically for:

[0196] Extract features from the processed analog signal, input the extracted features into a deep learning model for training to obtain a trained type recognition model, and input the processed analog signal into the trained type recognition model for recognition to obtain the corresponding signal type;

[0197] Using a pre-built adaptive sampling frequency prediction model to predict the multimodal analog signal to obtain an optimal sampling frequency, establishing a mapping relationship table between various types of signals and the optimal sampling frequencies, searching the corresponding optimal sampling frequency from the mapping relationship table according to the signal type, and adaptively adjusting the sampling frequency of the multimodal analog signal using the corresponding optimal sampling frequency;

[0198] The adaptive quantization module 24 is used to perform adaptive quantization and quantization error compensation on the adjusted analog signal, and to enhance the digitized signal and output the final digital signal. Specifically, it is used to:

[0199] Inputting the adjusted analog signal into a pre-built neural network-based quantization model for prediction, obtaining an optimal quantization level and a quantization error estimate of the signal, adaptively quantizing the signal according to the optimal quantization level, and introducing a back-propagation error correction mechanism during the quantization process to correct the quantization error estimate in real time through the back-propagation error correction mechanism;

[0200] The digitized signal is corrected using a nonlinear correction algorithm, and the corrected digital signal is input into a pre-built deep learning enhancement model for enhancement processing to obtain the final digital signal;

[0201] The nonlinear correction algorithm is a nonlinear correction algorithm based on the Fourier transform principle and combined with an iterative optimization method.

[0202] The embodiment of the present invention is a system embodiment corresponding to the above-mentioned method embodiment. The specific operations of each module can be understood by referring to the description of the method embodiment, which will not be repeated here.

[0203] In summary, with the rapid development of technologies such as the Internet of Things and the Industrial Internet, the demand for the perception and digitization of information in the physical world is increasing. As the main carrier of information in the physical world, the digitization accuracy of analog signals directly affects the accuracy of subsequent data processing and analysis. Traditional analog signal digitization methods mainly use fixed sampling frequencies and quantization bits, which are difficult to adapt to the characteristics of different types of analog signals, and have problems such as low accuracy and poor anti-interference ability. Therefore, an embodiment of the present invention provides a high-precision digitization method for multimodal analog signals based on adaptive sampling and deep learning to improve the accuracy and anti-interference ability of analog signal digitization, which specifically includes the following beneficial effects:

[0204] 1. The embodiment of the present invention identifies the signal type through a deep learning model and adaptively adjusts the sampling frequency and quantization bit number according to the signal type, which can effectively improve the accuracy of analog signal digitization.

[0205] 2. The embodiment of the present invention uses a deep learning model to enhance the digitized signal, which can effectively improve the signal's anti-interference ability.

[0206] 3. The method proposed in the embodiment of the present invention is applicable to various types of analog signals and has the advantages of strong versatility and wide applicability.

[0207] Device Example 1

[0208] An embodiment of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps described in the method embodiment when executed by the processor.

[0209] Device Example 2

[0210] An embodiment of the present invention provides a computer-readable storage medium, on which a program for implementing information transmission is stored. When the program is executed by a processor, the steps described in the method embodiment are implemented.

[0211] The computer-readable storage medium in this embodiment includes, but is not limited to, ROM, RAM, magnetic disk, or optical disk.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multimodal analog signal digitization method based on adaptive sampling, characterized in that include: Collecting multimodal analog signals and preprocessing the multimodal analog signals; Obtaining a corresponding signal type based on the processed analog signal, and adaptively adjusting the sampling frequency according to the signal type; as well as The adjusted analog signal is adaptively quantized and quantization error compensated, and the digitized signal is enhanced to output the final digital signal.

2. The method according to claim 1, characterized in that The corresponding signal types obtained based on the processed analog signal specifically include: The processed analog signal is subjected to feature extraction, and the extracted features are input into a deep learning model for training to obtain a trained type recognition model, and the processed analog signal is input into the trained type recognition model for recognition to obtain the corresponding signal type.

3. The method according to claim 2, characterized in that Adaptively adjusting the sampling frequency according to the signal type specifically includes: The multimodal analog signal is predicted using a pre-built adaptive sampling frequency prediction model to obtain the optimal sampling frequency, a mapping relationship table is established between each type of signal and the optimal sampling frequency, the corresponding optimal sampling frequency is found from the mapping relationship table according to the signal type, and the sampling frequency of the multimodal analog signal is adaptively adjusted using the corresponding optimal sampling frequency.

4. The method according to claim 1, wherein Adaptive quantization and quantization error compensation of the adjusted analog signal specifically include: The adjusted analog signal is input into a pre-built neural network-based quantization model for prediction to obtain the optimal quantization level and quantization error estimation value of the signal. The signal is adaptively quantized according to the optimal quantization level, and a back-propagation error correction mechanism is introduced in the quantization process to perform real-time correction on the quantization error estimation value through the back-propagation error correction mechanism.

5. The method according to claim 4, characterized in that The output of the final digital signal after the digitized signal is enhanced includes: The digitized signal is corrected using a nonlinear correction algorithm, and the corrected digital signal is input into a pre-built deep learning enhancement model for enhancement processing to obtain the final digital signal; The nonlinear correction algorithm is a nonlinear correction algorithm based on the Fourier transform principle and combined with an iterative optimization method.

6. A multimodal analog signal digitization system based on adaptive sampling, characterized in that include: A processing module, configured to collect multimodal analog signals and preprocess the multimodal analog signals; An adaptive sampling module, configured to obtain a corresponding signal type based on the processed analog signal and adaptively adjust the sampling frequency according to the signal type; as well as The adaptive quantization module is used to perform adaptive quantization and quantization error compensation on the adjusted analog signal, and output the final digital signal after enhancing the digitized signal.

7. The system according to claim 6, characterized in that The adaptive sampling module is specifically used for: Extract features from the processed analog signal, input the extracted features into a deep learning model for training to obtain a trained type recognition model, and input the processed analog signal into the trained type recognition model for recognition to obtain the corresponding signal type; The multimodal analog signal is predicted using a pre-built adaptive sampling frequency prediction model to obtain the optimal sampling frequency, a mapping relationship table is established between each type of signal and the optimal sampling frequency, the corresponding optimal sampling frequency is found from the mapping relationship table according to the signal type, and the sampling frequency of the multimodal analog signal is adaptively adjusted using the corresponding optimal sampling frequency.

8. The system according to claim 6, wherein: The adaptive quantization module is specifically used for: Inputting the adjusted analog signal into a pre-built neural network-based quantization model for prediction, obtaining an optimal quantization level and a quantization error estimate of the signal, adaptively quantizing the signal according to the optimal quantization level, and introducing a back-propagation error correction mechanism during the quantization process to correct the quantization error estimate in real time through the back-propagation error correction mechanism; The digitized signal is corrected using a nonlinear correction algorithm, and the corrected digital signal is input into a pre-built deep learning enhancement model for enhancement processing to obtain the final digital signal; The nonlinear correction algorithm is a nonlinear correction algorithm based on the Fourier transform principle and combined with an iterative optimization method.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the multimodal analog signal digitization method based on adaptive sampling as described in any one of claims 1 to 5 are implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores an information transmission implementation program, and when the program is executed by the processor, the steps of the multimodal analog signal digitization method based on adaptive sampling as described in any one of claims 1 to 5 are implemented.

Images