Dynamic compensation method for analog-to-digital conversion and novel analog-to-digital converter

Abstract

The invention provides a dynamic compensation method for analog-to-digital conversion and a novel analog-to-digital converter, and relates to the technical field of analog-to-digital conversion. The method comprises the following steps: acquiring an analog signal through an analog signal input port; the analog signal input port is synchronously input to the analog-to-digital conversion module; training data are determined according to the analog signal, training of a deep learning model is carried out according to the training data, the deep learning model is obtained, and the deep learning model is used for determining a parameter compensation value according to the analog signal, the change period and amplitude of a digital code value output by an analog-to-digital conversion module and an ideal digital code value output deviation; and compensating the digital code value output by the analog-to-digital conversion module according to the parameter compensation value. The digital code value can be accurately compensated in real time, an accurate analog-to-digital conversion result is obtained, and the accuracy of analog-to-digital conversion is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of analog-to-digital conversion technology, and in particular to a dynamic compensation method for analog-to-digital conversion and a novel analog-to-digital converter. Background Art

[0002] In recent years, analog-to-digital conversion has been widely used in digital signal processing, analog circuit design and other fields.

[0003] Analog to Digital Converter (ADC), also known as analog-to-digital converter, is an electronic component that converts analog signals into digital signals. Usually, an analog-to-digital converter converts an input voltage signal into an output digital signal.

[0004] In actual use, it is found that as the environment changes, such as humidity, temperature or other changes, the conversion accuracy of the analog-to-digital converter will fluctuate with the change of the environment. How to improve the accuracy of the analog-to-digital converter has become an urgent problem to be solved. Summary of the invention

[0005] The present application provides a dynamic compensation method for analog-to-digital conversion and a novel analog-to-digital converter, which are used to solve the problem of low accuracy of analog-to-digital converters in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a dynamic compensation method for analog-to-digital conversion, comprising:

[0007] Acquire an analog signal through an analog signal input port; the analog signal input port is synchronously input to an analog-to-digital conversion module;

[0008] Determine training data according to the analog signal, train a deep learning model according to the training data, and obtain a deep learning model, wherein the deep learning model is used to determine a parameter compensation value according to the analog signal and a digital code value change period, amplitude, and output deviation from an ideal digital code value output by an analog-to-digital conversion module;

[0009] The digital code value output by the analog-to-digital conversion module is compensated according to the parameter compensation value.

[0010] In a second aspect, the embodiment of the present application further provides a dynamic compensation device for analog-to-digital conversion, including:

[0011] The analog signal acquisition module is used to acquire the analog signal through the analog signal input terminal; the real-time analog signal input terminal is synchronously input to the analog-to-digital conversion module;

[0012] A training module, used to determine training data according to the analog signal, train a deep learning model according to the training data, and obtain a deep learning model, wherein the deep learning model is used to determine a parameter compensation value according to the analog signal and a digital code value variation period, amplitude, and output deviation from an ideal digital code value output by the analog-to-digital conversion module;

[0013] A dynamic compensation module is used to compensate the digital code value output by the analog-to-digital conversion module according to the parameter compensation value.

[0014] In a third aspect, the embodiment of the present application further provides a novel analog-to-digital converter, including:

[0015] Analog signal input terminal, deep learning terminal, digital signal output terminal, data storage terminal, feedback configuration terminal and analog-to-digital conversion module;

[0016] The analog signal input terminal is connected to the deep learning terminal and the analog-to-digital conversion module respectively; the deep learning terminal and the analog-to-digital conversion module are connected to the digital signal output terminal respectively; the data storage terminal is connected to the deep learning terminal, and the feedback configuration terminal is connected to the deep learning terminal and the analog-to-digital conversion module respectively;

[0017] The deep learning end is used to obtain an analog signal through an analog signal input end; the analog signal input end is synchronously input into an analog-to-digital conversion module; training data is determined according to the analog signal, and a deep learning model is trained according to the training data to obtain a deep learning model, and the deep learning model is used to determine a parameter compensation value according to the analog signal and the digital code value change period, amplitude and output deviation from the ideal digital code value output by the analog signal and the analog-to-digital conversion module; the data storage end is used to provide a cache of calculation parameters for the deep learning end; and the feedback configuration end is used to store the parameters required for compensation.

[0018] In a fourth aspect, an embodiment of the present application further provides a storage medium comprising computer executable instructions, wherein the computer readable storage medium stores computer executable instructions, and when the computer executable instructions are executed by a processor, they are used to implement a dynamic compensation method for analog-to-digital conversion as shown in an embodiment of the present application.

[0019] The dynamic compensation method of analog-to-digital conversion provided by the present application obtains an analog signal through an analog signal input port; the analog signal input port is synchronously input to an analog-to-digital conversion module; training data is determined according to the analog signal, and a deep learning model is trained according to the training data to obtain a deep learning model, wherein the deep learning model is used to determine a parameter compensation value according to the variation cycle, amplitude, and output deviation of the digital code value output by the analog signal and the analog-to-digital conversion module; and the digital code value output by the analog-to-digital conversion module is compensated according to the parameter compensation value. Compared with the current situation where the accuracy of analog-to-digital converter conversion is unstable, the present application can train a deep learning model in real time according to the analog information of the input port and the digital code output by the analog-to-digital conversion module, determine the parameter compensation value using the trained deep learning model, and compensate the digital code value output by the analog-to-digital conversion module using the parameter compensation value. The deep learning model can determine the parameter compensation value according to the variation cycle, amplitude, and output deviation of the digital code value output by the analog signal and the analog-to-digital conversion module, and realize the determination of real-time parameter compensation values ​​for different temperature, humidity, and other environments. After the analog-to-digital conversion module outputs the digital code value, compensation is performed according to the parameter compensation value to achieve real-time and accurate compensation of the digital code value, obtain accurate analog-to-digital conversion results, and improve the accuracy of analog-to-digital conversion. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The process of the dynamic compensation method for analog-to-digital conversion provided in the embodiment of the present application is Figure 1 ;

[0021] Figure 2 The process of the dynamic compensation method for analog-to-digital conversion provided in the embodiment of the present application is Figure 2 ;

[0022] Figure 3 is a schematic structural diagram of a dynamic compensation device for analog-to-digital conversion provided in an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of the structure of a novel analog-to-digital converter provided in an embodiment of the present application;

[0024] Figure 5 Schematic diagram of a novel analog-to-digital converter provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0026] In recent years, analog-to-digital conversion has been widely used in digital signal processing, analog circuit design and other fields. Analog-to-digital converter (ADC), also known as analog-to-digital converter, is an electronic component that converts analog signals into digital signals. Usually, an analog-to-digital converter converts an input voltage signal into an output digital signal.

[0027] In actual use, it is found that as the environment changes, such as humidity, temperature or other changes, the conversion accuracy of the analog-to-digital converter will fluctuate with the change of the environment. How to improve the accuracy of the analog-to-digital converter has become an urgent problem to be solved.

[0028] The dynamic compensation method for analog-to-digital conversion provided in the present application determines the parameter compensation value according to the currently input analog signal and the digital code value output by the analog-to-digital conversion module through a deep learning model. The parameter compensation values ​​obtained in different environments are different. After compensating the digital code value according to the parameter compensation value, a more accurate digital code value can be obtained, thereby improving the efficiency of analog-to-digital conversion.

[0029] The dynamic compensation method for analog-to-digital conversion provided in this application can be applied to the fields of digital signal processing, analog circuit design, and deep learning. It is particularly suitable for system-level chips and central processors with high requirements for power supply reliability and stability in scenarios such as high-end industrial control, communication base stations, servers, and semiconductors.

[0030] The dynamic compensation method of analog-to-digital conversion provided by the present application adopts deep learning technology to model and calibrate the nonlinear characteristics of the analog-to-digital conversion module. Through dynamic calibration of real-time collected data, the deep learning model can accurately capture and learn the nonlinear transformation law of the input signal, and realize dynamic calibration of the analog-to-digital conversion module. In this process, the background noise separation technology can effectively identify and separate the background noise signal, reducing its impact on the performance and data quality of the analog-to-digital conversion module. At the same time, the signal preprocessing technology can filter the input signal, adjust the gain and other operations, and further improve the precision and accuracy of the analog-to-digital conversion module. By solving the nonlinearity and inaccuracy problems existing in the traditional analog-to-digital conversion module, the goal of improving the precision and accuracy of the analog-to-digital conversion module is achieved.

[0031] The following is a detailed description of the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The acquisition, transmission, storage, use, and processing of data in the technical solution of the present application are in compliance with the relevant provisions of national laws and regulations.

[0032] It should be noted that in the embodiments of the present application, some existing solutions in the industry such as certain software, components, models, etc. may be mentioned, which should be considered as exemplary, and their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0033] Figure 1 The process of the dynamic compensation method of analog-to-digital conversion provided in the embodiment of the present application Figure 1 This embodiment is applicable to the case of analog-to-digital conversion in high-end industrial control, communication base stations, servers, semiconductors and other scenarios. The method can be performed by an electronic device with an analog-to-digital conversion function, especially by a deep learning module in the analog-to-digital converter provided in the embodiment of the present application. Figure 1 As shown, the dynamic compensation method for analog-to-digital conversion provided by the present application includes:

[0034] Step S101, acquiring an analog signal through an analog signal input port; the analog signal input port synchronously inputs the analog signal to an analog-to-digital conversion module.

[0035] The embodiment of the present application is used to convert an analog signal into a digital signal, which is also called a digital code value. The analog-to-digital conversion module can use any product in the prior art, and the analog-to-digital conversion module is used to convert the input analog signal into a digital code value. The analog-to-digital conversion module can sample the analog signal at a certain sampling rate and resolution, compare the sampled signal with a threshold, and convert the signal into a digital code value according to the threshold.

[0036] At present, during the transmission of analog signals, noise may occur due to environmental factors such as temperature, humidity or surrounding magnetic field, resulting in differences between the analog signals and the original analog signals. The analog-to-digital conversion module cannot identify the signal noise caused by environmental factors. The analog-to-digital conversion module is only used to convert analog signals into digital code values.

[0037] The present application obtains an analog signal while inputting the analog signal into the analog-to-digital conversion module for conversion, so that the deep learning module performs deep learning based on the analog signal to determine the noise.

[0038] Step S102: determine training data according to the simulation signal, and train a deep learning model according to the training data to obtain a deep learning model.

[0039] Among them, the deep learning model is used to determine the parameter compensation value according to the analog signal and the digital code value change period, amplitude and output deviation from the ideal digital code value output by the analog-to-digital conversion module.

[0040] The embodiment of the present application aims to provide a solution capable of performing real-time compensation on the digital code value output by the analog-to-digital conversion module. The parameter compensation value used in the real-time compensation is obtained by a deep learning model.

[0041] Optionally, determining training data according to the simulation signal, and training a deep learning model according to the training data may be implemented as follows:

[0042] Step 2021: Determine the amplitude and frequency according to the input analog signal and the data conversion code value.

[0043] By analyzing the analog signal, the frequency of the signal can be obtained. By analyzing the data conversion code value and the frequency domain noise matrix, the amplitude of the signal can be determined.

[0044] Optionally, the amplitude and frequency are determined according to the input analog signal, which can be implemented as follows:

[0045] The frequency of the input signal and the noise frequency matrix are determined by performing Fourier transformation on the waveform of the analog signal; the amplitude of the input signal and the noise amplitude are determined according to the maximum and minimum values ​​of the digital code values ​​corresponding to the analog signal and the power value of the frequency domain noise matrix.

[0046] By performing Fourier transformation on the waveform of the input analog signal, the frequency of the input analog signal and the noise frequency matrix can be obtained. Then, the amplitude of the analog signal can be determined according to the maximum and minimum values ​​of the digital code value of the analog signal. The power value of the frequency domain noise matrix can be determined according to the noise frequency matrix, and the noise amplitude can be determined according to the power value of the frequency domain noise matrix.

[0047] The above method can accurately calculate the frequency and amplitude of the analog signal itself and the noise amplitude, realize background noise separation, and improve the accurate recognition of the analog signal.

[0048] Step 2022: Determine training data according to a preset number of data points in the simulation signals under various environments.

[0049] The training data can use analog signals in different environments. Since the analog signals used as training data also need to be converted from analog to digital, in order to avoid using too many analog signals for training, which results in the deep learning model being unable to participate in the compensation of real-time digital code values ​​as soon as possible, a preset number of data points in the analog signal are used as training data in an embodiment of the present application.

[0050] Exemplarily, determining training data based on a preset number of data points in analog signals under multiple environments includes: determining training data based on a preset number of data points in analog signals under different temperatures and different voltages, wherein the preset number is at least 10^10.

[0051] The analog signal in any environment can be collected according to a preset number of data points to determine the training data. In practical applications, the analog-to-digital converter can eliminate the use of sensors to detect the environment, and instead sample the analog signal in any scene to obtain training data. The preset number can be configured based on experience and training effect feedback.

[0052] By sampling the data points of analog signals in all scenarios to obtain training data, it is possible to sample analog signals in multiple environments to obtain training data. This enables the collection of analog signals in multiple environments and improves the universality of deep learning models.

[0053] Step 2023: Train the deep learning model according to the amplitude, the frequency, and the training data to obtain parameter compensation values ​​corresponding to the training data.

[0054] Optionally, the deep learning model uses the amplitude, the frequency and the training data as input data and uses the parameter compensation value as output. The deep learning model is trained according to the amplitude, the frequency and the training data.

[0055] Optionally, training a deep learning model according to the amplitude, the frequency and the training data may be implemented as follows:

[0056] Extract the full-scale calibration points in the training data, wherein the calibration points include amplitude points, frequency points and background noise energy values; sort the extracted calibration points; calculate the ideal digital code values ​​corresponding to the analog values ​​of all calibration points according to the sorting results; perform a difference operation on the ideal digital code value of each calibration point and the analog code value of the training data one by one to obtain the parameter compensation value of each point; perform a frequency domain fast Fourier transform (FFT) on the compensated digital code value to determine the compensated frequency point and compare it horizontally with the ideal frequency point to obtain the frequency point deviation compensation value.

[0057] Extract the calibration points of the full range based on the two-point method. Sort all the extracted calibration points. When sorting, the minimum interval between data and data is 0.01LSB, and equal data points are regarded as the same data point. Calculate the ideal digital code value corresponding to the analog value of all calibration points. Perform a difference operation on the digital code value and analog code value of each data point one by one to obtain the parameter compensation value of each calibration point. According to the parameters obtained in the calibration training model as the calibration result under one working condition (such as voltage), it is also necessary to traverse the full temperature range for training, and the minimum interval of analog input is 0.01LSB.

[0058] In the above implementation, the frequency and amplitude of the input analog signal can be extracted, and a noise floor inverse waveform can be generated to separate the signal source from the noise floor. The calibration point calculated over the full range is selected and calibrated according to the amplitude. When the amplitude changes, the calibration parameters also change, thereby achieving the purpose of dynamic calibration.

[0059] Step S103: compensating the digital code value output by the analog-to-digital conversion module according to the parameter compensation value.

[0060] Furthermore, after compensating the digital code value output by the analog-to-digital conversion module according to the parameter compensation value, it also includes: determining the compensation result according to the digital code value output by the analog-to-digital conversion module; if the compensation result does not meet expectations, reselecting the standard point through the deep learning model, and determining a new parameter compensation value according to the new standard point.

[0061] The dynamic compensation method of analog-to-digital conversion provided by the present application obtains an analog signal through an analog signal input port; the analog signal input port is synchronously input to an analog-to-digital conversion module; training data is determined according to the analog signal, and a deep learning model is trained according to the training data to obtain a deep learning model, wherein the deep learning model is used to determine a parameter compensation value according to the variation cycle, amplitude, and output deviation of the digital code value output by the analog signal and the analog-to-digital conversion module; and the digital code value output by the analog-to-digital conversion module is compensated according to the parameter compensation value. Compared with the current situation where the accuracy of analog-to-digital converter conversion is unstable, the present application can train a deep learning model in real time according to the analog information of the input port and the digital code output by the analog-to-digital conversion module, determine the parameter compensation value using the trained deep learning model, and compensate the digital code value output by the analog-to-digital conversion module using the parameter compensation value. The deep learning model can determine the parameter compensation value according to the variation cycle, amplitude, and output deviation of the digital code value output by the analog signal and the analog-to-digital conversion module, and realize the determination of real-time parameter compensation values ​​for different temperature, humidity, and other environments. After the digital code value is output by the analog-to-digital conversion module, compensation is performed according to the parameter compensation value, which can realize real-time and accurate flexible compensation of the digital code value, obtain accurate analog-to-digital conversion results, and improve the accuracy of analog-to-digital conversion.

[0062] Figure 2 A dynamic compensation method for analog-to-digital conversion provided in an embodiment of the present invention, as a further explanation of the above implementation, includes:

[0063] Step S201, acquiring an analog signal through an analog signal input port; the analog signal input port synchronously inputs the analog signal to an analog-to-digital conversion module.

[0064] Step S202, determine the frequency of the input signal and the noise frequency matrix according to the Fourier transform frequency of the analog signal waveform; determine the amplitude of the input signal and the noise amplitude according to the maximum and minimum values ​​of the digital code value corresponding to the analog signal and the power value of the frequency domain noise matrix.

[0065] Step S203: determining training data according to a preset number of data points in the analog signal at different temperatures and different voltages.

[0066] Wherein, the preset number is at least 10^10.

[0067] Step S204, extracting full-scale calibration points from the training data, wherein the calibration points include amplitude points, frequency points, and background noise energy values; sorting the extracted calibration points; and calculating ideal digital code values ​​corresponding to analog values ​​of all calibration points according to the sorting results.

[0068] Step S205, perform a difference operation on the ideal digital code value of each calibration point and the analog code value of the training data one by one to obtain the parameter compensation value of each point; perform frequency domain FFT calculation on the compensated digital code value to determine the compensated frequency point and compare it horizontally with the ideal frequency point to obtain the frequency point deviation compensation value.

[0069] Step S206: Compensate the digital code value output by the analog-to-digital conversion module according to the parameter compensation value.

[0070] Step S207, determining a compensation result according to the digital code value output by the analog-to-digital conversion module; if the compensation result does not meet expectations, reselecting a standard point through the deep learning model, and determining a new parameter compensation value according to the new standard point.

[0071] The above implementation can compensate the digital code value output by the analog-to-digital conversion module in real time, realize dynamic compensation of the digital code value during the analog-to-digital conversion process, and improve the accuracy of the analog-to-digital conversion.

[0072] Figure 3The schematic diagram of the structure of the dynamic compensation device for analog-to-digital conversion provided in the embodiment of the present application. The dynamic compensation device for analog-to-digital conversion can be applied to the situation of analog-to-digital conversion in scenarios such as high-end industrial control, communication base stations, servers and semiconductors. The device can be executed by an electronic device with analog-to-digital conversion function, especially by the deep learning module in the analog-to-digital converter provided in the embodiment of the present application. Figure 3 As shown, the dynamic compensation device for analog-to-digital conversion includes: an analog signal acquisition module 31, a training module 32 and a dynamic compensation module 33.

[0073] The analog signal acquisition module 31 is used to acquire the analog signal through the analog signal input terminal; the real-time analog signal input terminal synchronously inputs the analog-to-digital conversion module;

[0074] A training module 32 is used to determine training data according to the analog signal, and to train a deep learning model according to the training data to obtain a deep learning model, wherein the deep learning model is used to determine a parameter compensation value according to the analog signal and a digital code value change period, amplitude, and output deviation from an ideal digital code value output by the analog-to-digital conversion module;

[0075] The dynamic compensation module 33 is used to compensate the digital code value output by the analog-to-digital conversion module according to the parameter compensation value.

[0076] In some embodiments, optionally, the training module 32 is used to determine training data according to the simulation signal, and train the deep learning model according to the training data, including:

[0077] Determine the amplitude and frequency according to the input analog signal and the data conversion code value;

[0078] determining training data based on a preset number of data points in simulated signals under multiple environments;

[0079] A deep learning model is trained according to the amplitude, the frequency and the training data to obtain parameter compensation values ​​corresponding to the training data.

[0080] In some embodiments, optionally, the training module 32 is used to train a deep learning model according to the amplitude, the frequency and the training data, including:

[0081] Extracting full-scale calibration points from the training data, wherein the calibration points include amplitude points, frequency points, and background noise energy values;

[0082] sorting the extracted calibration points;

[0083] According to the sorting results, the ideal digital code values ​​corresponding to the analog values ​​of all calibration points are calculated;

[0084] Perform a difference operation on the ideal digital code value of each calibration point and the analog code value of the training data one by one to obtain a parameter compensation value of each point;

[0085] The compensated digital code value is subjected to frequency domain FFT calculation to determine the compensated frequency point and is horizontally compared with the ideal frequency point to obtain the frequency point deviation compensation value.

[0086] In some embodiments, optionally, the training module 32 is used to determine the training data according to a preset number of data points in the simulation signal under various environments, including:

[0087] The training data is determined according to a preset number of data points in the analog signal at different temperatures and different voltages, wherein the preset number is at least 10^10.

[0088] In some embodiments, optionally, the training module 32 is used to determine the amplitude and frequency according to the input analog signal, including:

[0089] Determine the frequency of the input signal and the noise frequency matrix by performing Fourier transform frequency according to the waveform of the analog signal;

[0090] The amplitude of the input signal and the noise amplitude are determined according to the maximum and minimum values ​​of the digital code values ​​corresponding to the analog signal and the power value of the frequency domain noise matrix.

[0091] In some embodiments, optionally, a compensation feedback module is also included, which is used for the training module 32 to determine the compensation result according to the digital code value output by the analog-to-digital conversion module after the dynamic compensation module 33 compensates the digital code value output by the analog-to-digital conversion module according to the parameter compensation value; if the compensation result does not meet the expectation, reselect the standard point through the deep learning model, and determine the new parameter compensation value according to the new standard point.

[0092] The dynamic compensation device for analog-to-digital conversion provided by the present application includes an analog signal acquisition module 31, which is used to acquire an analog signal through an analog signal input terminal; a real-time analog signal input terminal is synchronously input into an analog-to-digital conversion module; a training module 32, which is used to determine training data according to the analog signal, train a deep learning model according to the training data, and obtain a deep learning model, wherein the deep learning model is used to determine a parameter compensation value according to the analog signal and the digital code value change cycle, amplitude, and output deviation from the ideal digital code value output by the analog-to-digital conversion module; and a dynamic compensation module 33, which is used to compensate the digital code value output by the analog-to-digital conversion module according to the parameter compensation value. Compared with the current situation where the accuracy of analog-to-digital converter conversion is unstable, the present application can train a deep learning model in real time according to the analog information of the input port and the digital code output by the analog-to-digital conversion module, determine the parameter compensation value using the trained deep learning model, and compensate the digital code value output by the analog-to-digital conversion module using the parameter compensation value. The deep learning model can determine the parameter compensation value according to the analog signal and the digital code value change cycle, amplitude and deviation from the ideal digital code value output by the analog-to-digital conversion module, and determine the real-time parameter compensation value for different temperature, humidity and other environments. After the digital code value output by the analog-to-digital conversion module is compensated according to the parameter compensation value, it can realize real-time and accurate flexible compensation of the digital code value, obtain accurate analog-to-digital conversion results, and improve the accuracy of analog-to-digital conversion.

[0093] The dynamic compensation device for analog-to-digital conversion provided in the embodiment of the present application can be used to execute the technical solution of the dynamic compensation method for analog-to-digital conversion in the above-mentioned embodiment. Its implementation principle and technical effect are similar and will not be repeated here.

[0094] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the training module 32 can be a separately established processing element, or it can be integrated in a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The function of the above training module 32. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0095] Figure 4 This is a schematic diagram of the structure of a novel analog-to-digital converter provided in an embodiment of the present application. Figure 4 As shown, the electronic device may include: an analog signal input terminal 41, a deep learning terminal 42, a digital signal output terminal 43, a data storage terminal 44, a feedback configuration terminal 45 and an analog-to-digital conversion module 46;

[0096] The analog signal input terminal 41 is connected to the deep learning terminal 42 and the analog-to-digital conversion module 46 respectively; the deep learning terminal 42 and the analog-to-digital conversion module 46 are connected to the digital signal output terminal 43 respectively; the data storage terminal 44 is connected to the deep learning terminal 42, and the feedback configuration terminal 45 is connected to the deep learning terminal 42 and the analog-to-digital conversion module 46 respectively.

[0097] The deep learning terminal 42 is used to obtain an analog signal through the analog signal input terminal 41; the analog signal input terminal 41 is synchronously input to the analog-to-digital conversion module 46; training data is determined according to the analog signal, and a deep learning model is trained according to the training data to obtain a deep learning model, and the deep learning model is used to determine a parameter compensation value according to the analog signal and the digital code value change period, amplitude and output deviation from the ideal digital code value output by the analog-to-digital conversion module 46;

[0098] The data storage end 44 is used to provide a cache of calculation parameters for the deep learning end 42;

[0099] The feedback configuration terminal 45 is used to store parameters required for compensation.

[0100] Among them, the analog signal input terminal 41 serves as the input port of the analog-to-digital conversion module 46 and the acquisition feature input port of the deep learning terminal 42. The deep learning terminal 42 is used to calculate the parameter compensation value. The deep learning terminal 42 may include an adder, a subtractor, a multiplier, and a data arrangement circuit. The above circuit components may be arranged as required to realize the functions of the above deep learning terminal 42. The digital signal output terminal 43 is the output port of the analog-to-digital conversion module 46 and the feature configuration feedback terminal of the deep learning terminal 42. The calculation process and parameters of the deep learning terminal 42 may be dynamically adjusted according to the actual adjustment effect. The data storage segment may provide a cache of calculation parameters for the deep learning terminal 42, and the calculation parameters include the calculated parameter compensation values.

[0101] Optionally, the analog-to-digital conversion module 46 is a successive approximation analog-to-digital converter (SAR ADC).

[0102] The analog signal input terminal 41 can be directly connected to the signal source to transmit the signal from the signal source to the deep learning terminal 42 and the analog-to-digital conversion module 46. If the signal quality of the signal source is poor, a corresponding module needs to be added to the analog signal input terminal 41 according to the demand. Optionally, the analog signal input terminal 41 includes a combination of one or more of a filtering module, an isolation module, or a power preprocessing module.

[0103] Among them, the filtering module may include a bandpass filtering module and a low-pass filtering module. The bandpass filtering module is used to allow waves in a specific frequency band to pass through while shielding waves in other frequency bands to achieve filtering. The low-pass filtering module is used to allow waves in a set low-frequency band to pass through, while high-frequency signals exceeding a set critical value are blocked and weakened to achieve filtering. The isolation module is used to protect the analog-to-digital conversion module 46 and the deep learning end 42 from mutual influence and interference with the signal source, such as noise generated by electromagnetic induction and electrostatic effects; the isolation module can also provide overcurrent protection. The power preprocessing module is used to control the power supply of the analog-to-digital converter.

[0104] In one implementation, the filtering module, the isolation module or the power preprocessing module included in the analog signal input terminal 41 is implemented in the form of a hardware circuit.

[0105] In another implementation, the filtering module, the isolation module or the power preprocessing module included in the analog signal input terminal 41 is implemented in the form of software calling through a processing element.

[0106] The digital signal output terminal 43 can be directly connected to an output device, or a corresponding module can be added to the digital signal output terminal 43 according to requirements. Optionally, the digital signal output terminal 43 includes a calibration module, an isolation module, or an operational amplifier module, or a combination of one or more of the above.

[0107] Among them, the calibration module is used to calibrate the digital signal based on the clock signal. The isolation module is used to protect the analog-to-digital conversion module 46 and the signal received by the deep learning terminal 42 from affecting and interfering with the output terminal, such as noise generated by electromagnetic induction and electrostatic effects; the isolation module can also provide overcurrent protection. The operational amplifier module, also known as the operational amplifier module, is used to amplify the analog or digital signal to obtain a clearer and more complete analog-to-digital conversion result.

[0108] In one implementation, the digital signal output terminal 43 includes a calibration module, an isolation module or an operational amplifier module implemented in the form of a hardware circuit.

[0109] In another implementation, the digital signal output terminal 43 includes a calibration module, an isolation module or an operational amplifier module implemented in the form of a hardware circuit or in the form of software called through a processing element.

[0110] When the analog input signal is sampled by the analog-to-digital conversion module 46 and the deep learning end 42 at the same time, phase, and frequency (following the coherent sampling theorem), the deep learning end 42 can calculate the signal-to-noise ratio, signal-to-noise ratio, integral nonlinearity (INL), differential nonlinearity (DNL), offset error, and gain error. The ideal transmission curve of analog-to-digital conversion should be a linear transmission curve, that is, the actual calculated conversion curve can be fitted with a linear calibration according to the ideal transmission curve. However, due to interference from the power supply, board signals, etc., the calibration value is a dynamically changing parameter, and there is an update cycle. The minimum update cycle time cannot be less than the time it takes for the analog-to-digital conversion module and the deep learning end to sample once.

[0111] The digital signal output terminal 43 is fed back by the deep learning terminal 42 to verify the calculation result of the deep learning terminal 42 and the analog input signal restoration comparison to confirm whether it is suitable for the analog-to-digital conversion architecture. If not, it will return to the training model library to select the calibration value.

[0112] The data storage terminal 44 is a volatile storage medium. Compared with the non-volatile storage terminal, the volatile storage medium can update the parameters according to the result of the first analog input signal sampling calculation, so that the analog-to-digital conversion can reach a nearly ideal linear state in the shortest time. The non-volatile storage circuit solution will cause the time for the analog-to-digital conversion to reach the ideal linear state to be longer.

[0113] In some embodiments, optionally, the deep learning end 42 determines training data according to the analog signal, and trains the deep learning model according to the training data, including:

[0114] Determine the amplitude and frequency according to the input analog signal and the data conversion code value;

[0115] determining training data based on a preset number of data points in simulated signals under multiple environments;

[0116] A deep learning model is trained according to the amplitude, the frequency and the training data to obtain parameter compensation values ​​corresponding to the training data.

[0117] Optionally, due to architectural design reasons, the above analog-to-digital conversion combination circuit architecture, i.e., the analog-to-digital converter, needs to discard at least the first sampling data. Optionally, the feedback needs to have at least a 3-bit area as a linear fitting calibration.

[0118] In some embodiments, optionally, the deep learning end 42 performs training of a deep learning model according to the amplitude, the frequency and the training data, including:

[0119] Extracting full-scale calibration points from the training data, wherein the calibration points include amplitude points, frequency points, and background noise energy values;

[0120] sorting the extracted calibration points;

[0121] According to the sorting results, the ideal digital code values ​​corresponding to the analog values ​​of all calibration points are calculated;

[0122] Perform a difference operation on the ideal digital code value of each calibration point and the analog code value of the training data one by one to obtain a parameter compensation value of each point;

[0123] The compensated digital code value is subjected to frequency domain FFT calculation to determine the compensated frequency point and is horizontally compared with the ideal frequency point to obtain the frequency point deviation compensation value.

[0124] In some embodiments, optionally, the deep learning end 42 determines the training data according to a preset number of data points in the simulation signal under various environments, including:

[0125] The training data is determined according to a preset number of data points in the analog signal at different temperatures and different voltages, wherein the preset number is at least 10^10.

[0126] In some embodiments, optionally, the deep learning end 42 determines the amplitude and frequency according to the input analog signal, including:

[0127] Determine the frequency of the input signal and the noise frequency matrix by performing Fourier transform frequency according to the waveform of the analog signal;

[0128] The amplitude of the input signal and the noise amplitude are determined according to the maximum and minimum values ​​of the digital code values ​​corresponding to the analog signal and the power value of the frequency domain noise matrix.

[0129] In some embodiments, optionally, after the deep learning end 42 compensates the digital code value output by the analog-to-digital conversion module 46 according to the parameter compensation value, the deep learning end 42 further includes:

[0130] Determine the compensation result according to the digital code value output by the analog-to-digital conversion module 46;

[0131] If the compensation result does not meet expectations, the standard point is reselected through the deep learning model, and a new parameter compensation value is determined based on the new standard point.

[0132] Analog signal input terminal 41, deep learning terminal 42, digital signal output terminal 43, data storage terminal 44, feedback configuration terminal 45 and analog-to-digital conversion module 46

[0133] For example, Figure 5A circuit design diagram of a novel analog-to-digital converter provided in an embodiment of the present application. The analog signal input terminal 41 is connected to the input pins of the deep learning terminal 42 and the analog-to-digital conversion module 46 respectively;

[0134] The clock signal pin Clk of the deep learning end is connected to the clock pin CLK of the trigger of the analog-to-digital conversion module and is connected to the digital signal output end. The clock signal provides a synchronization signal so that the calibration parameter compensation value obtained by the deep learning end is synchronized with the input signal, and the two are summarized in the output module for output.

[0135] The noise reduction pin noise of the deep learning end is connected to the analog signal input end. The output pin Output of the deep learning end is connected to the output pin Output of the analog-to-digital conversion module. The above output pin is connected to the digital signal output end 43. The check pin and linear calibration pin of the deep learning end point to the corresponding register address space of the feedback configuration end respectively.

[0136] Optionally, the offset calibration pin Offset-1 of the deep learning end points to the first address space of the register. The first address space is used to store the offset calibration bit Offet error. The compensation calibration pin of the deep learning end points to the second address space of the register. The second address space is used to store the compensation calibration bit Gain error. The front-end linear calibration pin of the deep learning end points to the third address space of the register; the back-end linear calibration pin of the deep learning end points to the fourth address space of the register. Exemplarily, the front section is the first 5% of the signal, and the back section is the back 95% of the signal.

[0137] Optionally, the size of the first address space is 1 bit; the size of the second address space is 1 bit; the size of the third address space is 3 bits or 2 bits; and the size of the fourth address space is 3 bits or 2 bits.

[0138] An embodiment of the present application also provides a chip for running instructions, which is used to execute the technical solution of the dynamic compensation method of analog-to-digital conversion in the above embodiment.

[0139] An embodiment of the present application also provides a computer-readable storage medium, in which computer instructions are stored. When the computer instructions are executed on a computer, the computer executes the technical solution of the dynamic compensation method for analog-to-digital conversion in the above embodiment.

[0140] An embodiment of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When at least one processor executes the computer program, it can implement the technical solution of the dynamic compensation method of analog-to-digital conversion in the above embodiment.

[0141] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0142] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A dynamic compensation method for analog-to-digital conversion, characterized in that: include: Acquire an analog signal through an analog signal input port; the analog signal input port is synchronously input to an analog-to-digital conversion module; Determine training data according to the analog signal, train a deep learning model according to the training data, and obtain a deep learning model, wherein the deep learning model is used to determine a parameter compensation value according to the analog signal and a digital code value change period, amplitude, and output deviation from an ideal digital code value output by an analog-to-digital conversion module; The digital code value output by the analog-to-digital conversion module is compensated according to the parameter compensation value.

2. The method according to claim 1, characterized in that: Determining training data according to the simulation signal, and training a deep learning model according to the training data, including: Determine the amplitude and frequency according to the input analog signal and the data conversion code value; determining training data based on a preset number of data points in simulated signals under multiple environments; A deep learning model is trained according to the amplitude, the frequency and the training data to obtain parameter compensation values ​​corresponding to the training data.

3. The method according to claim 2, characterized in that Training a deep learning model according to the amplitude, the frequency and the training data includes: Extracting full-scale calibration points from the training data, wherein the calibration points include amplitude points, frequency points, and background noise energy values; sorting the extracted calibration points; According to the sorting results, the ideal digital code values ​​corresponding to the analog values ​​of all calibration points are calculated; Perform a difference operation on the ideal digital code value of each calibration point and the analog code value of the training data one by one to obtain a parameter compensation value of each point; The compensated digital code value is subjected to frequency domain FFT calculation to determine the compensated frequency point and is horizontally compared with the ideal frequency point to obtain the frequency point deviation compensation value.

4. The method according to claim 3, characterized in that The training data is determined based on a preset number of data points from simulated signals in a variety of environments, including: The training data is determined according to a preset number of data points in the analog signal at different temperatures and different voltages, wherein the preset number is at least 10^10.

5. The method according to claim 2, characterized in that: Determine the amplitude and frequency of an analog input signal, including: Determine the frequency of the input signal and the noise frequency matrix by performing Fourier transform frequency according to the waveform of the analog signal; The amplitude of the input signal and the noise amplitude are determined according to the maximum and minimum values ​​of the digital code values ​​corresponding to the analog signal and the power value of the frequency domain noise matrix.

6. The method according to claim 1, characterized in that After compensating the digital code value output by the analog-to-digital conversion module according to the parameter compensation value, the method further includes: Determine a compensation result according to the digital code value output by the analog-to-digital conversion module; If the compensation result does not meet expectations, the standard point is reselected through the deep learning model, and a new parameter compensation value is determined based on the new standard point.

7. A dynamic compensation device for analog-to-digital conversion, characterized in that: include: An analog signal acquisition module, used for acquiring an analog signal through an analog signal input terminal; The real-time analog signal input terminal is synchronously input to the analog-to-digital conversion module; A training module, used to determine training data according to the analog signal, train a deep learning model according to the training data, and obtain a deep learning model, wherein the deep learning model is used to determine a parameter compensation value according to the analog signal and a digital code value variation period, amplitude, and output deviation from an ideal digital code value output by the analog-to-digital conversion module; A dynamic compensation module is used to compensate the digital code value output by the analog-to-digital conversion module according to the parameter compensation value.

8. A novel analog-to-digital converter, characterized in that: include: Analog signal input terminal, deep learning terminal, digital signal output terminal, data storage terminal, feedback configuration terminal and analog-to-digital conversion module; The analog signal input terminal is connected to the deep learning terminal and the analog-to-digital conversion module respectively; the deep learning terminal and the analog-to-digital conversion module are connected to the digital signal output terminal respectively; the data storage terminal is connected to the deep learning terminal, and the feedback configuration terminal is connected to the deep learning terminal and the analog-to-digital conversion module respectively; The deep learning end is used to obtain an analog signal through an analog signal input end; the analog signal input end is synchronously input into an analog-to-digital conversion module; training data is determined according to the analog signal, and a deep learning model is trained according to the training data to obtain a deep learning model, wherein the deep learning model is used to determine a parameter compensation value according to a variation period, amplitude, and output deviation of a digital code value output by the analog signal and the analog-to-digital conversion module from an ideal digital code value; The data storage end is used to provide a cache of calculation parameters for the deep learning end; The feedback configuration terminal is used to store parameters required for compensation.

9. The novel analog-to-digital converter according to claim 8, characterized in that: The analog signal input terminal includes a combination of one or more of a filtering module, an isolation module or a power preprocessing module.

10. The novel analog-to-digital converter according to claim 8, characterized in that: The digital signal output terminal includes a calibration module, an isolation module or an operational amplifier module or a combination of one or more of the above.

11. The novel analog-to-digital converter according to claim 9, characterized in that: The filtering module, the isolation module or the power preprocessing module included in the analog signal input terminal is implemented in the form of a hardware circuit; or, The filtering module, the isolation module or the power preprocessing module included in the analog signal input terminal is implemented in the form of software calling through a processing element.

12. The novel analog-to-digital converter according to claim 10, characterized in that: The digital signal output terminal includes a calibration module, an isolation module or an operational amplifier module implemented in the form of a hardware circuit; or, The digital signal output terminal includes a calibration module, an isolation module or an operational amplifier module which is implemented in the form of a hardware circuit or in the form of software calling through a processing element.

13. The novel analog-to-digital converter according to claim 8, characterized in that: The data storage end is a volatile storage medium.

14. The novel analog-to-digital converter according to any one of claims 8 to 10, characterized in that: The analog-to-digital conversion module is a successive approximation analog-to-digital converter.

15. The novel analog-to-digital converter according to claim 14, characterized in that: The analog signal input terminal is connected to the deep learning terminal and the input pin of the analog-to-digital conversion module respectively; The clock signal pin of the deep learning end is connected to the clock pin of the trigger of the analog-to-digital conversion module, and is connected to the digital signal output end; The noise reduction pin of the deep learning end is connected to the analog signal input end; The output pin of the deep learning terminal is connected to the output pin of the analog-to-digital conversion module; The verification pin and the linear calibration pin of the deep learning end respectively point to the corresponding register address space of the feedback configuration end.

16. The novel analog-to-digital converter according to claim 15, characterized in that: The offset calibration pin of the deep learning end points to the first address space of the register; the compensation calibration pin of the deep learning end points to the second address space of the register; the front-end linear calibration pin of the deep learning end points to the third address space of the register; and the back-end linear calibration pin of the deep learning end points to the fourth address space of the register.

17. The novel analog-to-digital converter according to claim 16, characterized in that: The size of the first address space is 1 bit; the size of the second address space is 1 bit; the size of the third address space is 3 bits or 2 bits; and the size of the fourth address space is 3 bits or 2 bits.

18. A storage medium comprising computer executable instructions, which when executed by a computer processor are used to perform the method according to any one of claims 1 to 6.

Images