ADC super-resolution reconstruction method, device and storage medium

Abstract

The present invention discloses an ADC super-resolution reconstruction method, device and storage medium. By using ADC sampling sequences at multiple different time points and utilizing convolution kernels and neural network training, the accuracy of ADC sampling can be improved. By reconstructing the ADC sampling value sequence obtained by waveform sampling, the sampling error can be reduced and the accuracy of the sampling result can be improved. And by using the trained neural network to reconstruct the third value sequence, high-precision ADC data can be obtained using a low-precision sampling chip, so that there is no need to oversample the data, reducing the burden of data storage and data transmission, and improving the upper limit of the accuracy of the ADC sampling of the original low-precision sampling chip. At the same time, the neural network is used for training and reconstruction, which can adapt to different sampling environments and noise conditions. The sampling data is made more robust, and more reliable ADC sampling results can be obtained in a complex working environment.

Description

Technical Field

[0001] The present invention relates to a field, and in particular to an ADC super-resolution reconstruction method, device and storage medium. Background Art

[0002] ADC is an analog-to-digital converter. The working principle of ADC is to sample the continuous analog signal, that is, to measure the voltage or current value of the analog signal within a certain time interval. After sampling, ADC converts the continuous analog signal into a discrete digital signal, that is, quantizes the amplitude of the analog signal into a discrete value in digital form. ADC is widely used in various fields, including communications, audio processing, instrumentation, industrial automation, etc.

[0003] In the prior art, in order to improve the sampling accuracy of ADC, it is necessary to use oversampling technology, high-speed sampling and digital filters to improve the sampling accuracy. Sigma-Delta ADC samples the signal at high speed in the high frequency band, and then extracts the required signal information through a digital filter.

[0004] However, in the prior art, the ADC sampling accuracy is related to the chip structure. High-precision applications require the use of ADC chips with more complex structures and higher costs, and the ADC sampling accuracy cannot be effectively improved through software. Summary of the invention

[0005] The present invention provides an ADC super-resolution reconstruction method, device and storage medium to solve the problem in the prior art that the ADC sampling accuracy cannot be effectively improved.

[0006] In a first aspect, the present application provides an ADC super-resolution reconstruction method, comprising:

[0007] Obtaining a third numerical sequence by waveform sampling, and reconstructing the third numerical sequence according to the trained neural network to obtain an ADC sampling numerical sequence;

[0008] Wherein, the neural network is obtained by training a training sample set; the training sample set is obtained by passing a first numerical sequence through several preset convolution kernels to obtain a convolution sequence corresponding to the convolution kernel, and is generated according to the first numerical sequence, the second numerical sequence and all convolution sequences; the first numerical sequence and the second numerical sequence are at least two ADC sampling sequences with different precisions, and the low-precision ADC sampling sequence contains several different time points.

[0009] In this way, by using multiple ADC sampling sequences with different precisions and utilizing the training of convolution kernels and neural networks, the accuracy of ADC sampling can be improved. By reconstructing the ADC sampling value sequence obtained by waveform sampling, the sampling error can be reduced and the accuracy of the sampling results can be improved. And by using the trained neural network to reconstruct the third value sequence, high-precision ADC data can be obtained using a low-precision sampling chip, so that there is no need to oversample the data, reducing the burden of data storage and data transmission, and improving the upper limit of the accuracy of the ADC sampling of the original low-precision sampling chip. At the same time, the neural network is used for training and reconstruction, which can adapt to different sampling environments and noise conditions. The sampling data is made more robust, and more reliable ADC sampling results can be obtained in a complex working environment.

[0010] Furthermore, the first numerical sequence and the second numerical sequence are at least two ADC sampling sequences with different precisions, and the low-precision ADC sampling sequence includes several different time points, including:

[0011] The waveform is generated by outputting a sine wave from 0 to a specific value and switching the sine wave frequency after a number of cycles through a chip with a DAC function;

[0012] The waveform is sampled by ADC chips with different precisions to obtain a first numerical sequence and a second numerical sequence.

[0013] In this way, by using ADC chips of different precision to sample the same waveform, a sampling sequence at multiple different time points can be obtained. In this way, more abundant information can be obtained in a single sampling process, including high-precision and low-precision sampling results. And because ADC chips of different precision may have different sampling errors and noises, by comparing the first numerical sequence and the second numerical sequence, the sampling sequence of the ADC chip can be reconstructed. By comparing the sampling results at different time points, the value of the actual waveform can be estimated more accurately and the influence of sampling errors can be reduced. The numerical sequence obtained through multiple sampling time points can be used for subsequent neural network training to improve the precision and accuracy of neural network training.

[0014] Furthermore, the first numerical sequence is passed through a plurality of preset convolution kernels to obtain a convolution sequence corresponding to the convolution kernel, including:

[0015] Using the element values ​​in the convolution kernel as weights, performing weighted summation on a number of elements in the first numerical sequence to obtain elements in the convolution sequence;

[0016] The number of the plurality of elements is the number of elements in the convolution kernel.

[0017] In this way, by using the preset convolution kernel, the feature extraction of the first numerical sequence can be performed. The weight of the convolution kernel can capture the local patterns and features in the sequence. Through weighted summation, these features can be extracted from the original sequence and form a convolution sequence. This helps to better understand and analyze the information in the original sequence. And by using different preset convolution kernels, matching and detection of different features can be achieved. Different convolution kernels can capture different patterns and features. By performing a convolution operation on the first numerical sequence, a convolution sequence corresponding to the convolution kernel can be obtained, thereby realizing the detection and extraction of specific features.

[0018] Furthermore, generating a training sample set according to the first numerical sequence, the second numerical sequence and all convolution sequences includes:

[0019] The elements in the convolution sequences and the corresponding elements in the first numerical sequence are used as convolution samples, and the moment values ​​in the second sequence are used as supervision values. The convolution samples and the corresponding supervision values ​​constitute training samples, and all the training samples constitute a training sample set.

[0020] In this way, by using such a training sample set, the neural network can learn the relationship between the convolution samples and the supervision values, so that the corresponding supervision values ​​can be predicted based on the convolution samples. In this way, when the third numerical sequence is reconstructed using the neural network, the sampling error can be corrected according to the prediction results of the neural network to improve the accuracy and precision of the ADC sampling values. And by using a training sample set consisting of multiple convolution sequences and corresponding supervision values, more sample data can be provided for neural network training. This helps to improve the generalization ability of the neural network, enabling it to better adapt to different sampling environments and noise conditions, and improve the robustness and reliability of the ADC sampling results.

[0021] Furthermore, the preset convolution kernels are specifically: a first convolution kernel and a second convolution kernel; wherein the first convolution kernel is [-1 -1 2], and the second convolution kernel is [-1 2 -1].

[0022] In this way, different convolution kernels can capture different patterns and features. By using multiple preset convolution kernels, such as [-1 -1 2] and [-1 2 -1], it is possible to match and detect different features. Each convolution kernel can correspond to a specific feature. By performing a convolution operation on the first numerical sequence, a convolution sequence corresponding to the convolution kernel can be obtained, thereby realizing the detection and extraction of specific features.

[0023] In a second aspect, the present application provides a device for improving ADC sampling accuracy, comprising: a reconstruction module;

[0024] The reconstruction module is used to obtain a third numerical sequence by waveform sampling, and reconstruct the third numerical sequence according to the trained neural network to obtain an ADC sampling numerical sequence;

[0025] Wherein, the neural network is obtained by training a training sample set; the training sample set is obtained by passing a first numerical sequence through several preset convolution kernels to obtain a convolution sequence corresponding to the convolution kernel, and is generated according to the first numerical sequence, the second numerical sequence and all convolution sequences; the first numerical sequence and the second numerical sequence are at least two ADC sampling sequences with different precisions, and the low-precision ADC sampling sequence contains several different time points.

[0026] Further, the reconstruction module includes: a waveform generation unit and an acquisition unit;

[0027] The waveform generating unit is used to generate a waveform by outputting a sine wave from 0 to a specific value through a chip having a DAC function and switching the frequency of the sine wave after a number of cycles;

[0028] The acquisition unit is used to sample the waveform through ADC chips with different precisions to obtain a first numerical sequence and a second numerical sequence.

[0029] Further, the reconstruction module includes: a convolution unit;

[0030] The convolution unit is used to use the element values ​​in the convolution kernel as weights, and perform weighted summation on a number of elements in the first numerical sequence to obtain elements in the convolution sequence;

[0031] The number of the plurality of elements is the number of elements in the convolution kernel.

[0032] Further, the reconstruction module includes: a convolution sample unit;

[0033] The convolution sample unit is used to use elements in several convolution sequences and several corresponding elements in the first numerical sequence as convolution samples, and use the moment values ​​in the second sequence as supervision values ​​to form training samples with the convolution samples and the corresponding supervision values, and all the training samples form a training sample set.

[0034] Correspondingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to perform the ADC super-resolution reconstruction method as described in any one of the above.

[0035] In this way, the same waveform is sampled by using multiple ADC chips with different precisions, and the convolution kernel and neural network training are used. The accuracy of ADC sampling can be improved by the trained neural network. By reconstructing the ADC sampling value sequence obtained by waveform sampling, the sampling error can be reduced and the accuracy of the sampling results can be improved. And by using low-precision sampling chips to obtain high-precision ADC data, there is no need to oversample the data, reducing the burden of data storage and data transmission. This can save system resources and increase the upper limit of the accuracy of the ADC sampling of the original low-precision sampling chip. At the same time, by using a training sample set composed of a preset convolution kernel and multiple convolution sequences, more sample data can be provided for neural network training. This helps to improve the generalization ability of the neural network, enabling it to better adapt to different sampling environments and noise conditions, and improve the robustness and reliability of the ADC sampling results. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 : A schematic flow chart of an embodiment of an ADC super-resolution reconstruction method provided by the present invention;

[0037] Figure 2 : A simulation diagram of a third numerical sequence of an ADC super-resolution reconstruction method provided by the present invention compared with the second numerical sequence after passing through the neural network;

[0038] Figure 3 : A module structure diagram of an embodiment of an ADC super-resolution reconstruction device provided by the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0041] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0042] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0043] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0044] Embodiment 1

[0045] Please refer to Figure 1 , an ADC super-resolution reconstruction method provided by an embodiment of the present invention includes step S1 as follows:

[0046] Step S1: obtaining a third numerical sequence by waveform sampling, and reconstructing the third numerical sequence according to the trained neural network to obtain an ADC sampling numerical sequence;

[0047] Wherein, the neural network is obtained by training a training sample set; the training sample set is obtained by passing a first numerical sequence through several preset convolution kernels to obtain a convolution sequence corresponding to the convolution kernel, and is generated according to the first numerical sequence, the second numerical sequence and all convolution sequences; the first numerical sequence and the second numerical sequence are at least two ADC sampling sequences with different precisions, and the low-precision ADC sampling sequence contains several different time points.

[0048] like Figure 2This is a simulation diagram of the comparison between the third numerical sequence and the second numerical sequence after passing through the neural network.

[0049] In an optional embodiment, the third numerical sequence is obtained through an MCU or a generally independent ADC chip, and the progress of the ADC chip can be 10 bits or 12 bits.

[0050] In a specific embodiment, the first numerical sequence and the second numerical sequence are at least two ADC sampling sequences with different precisions, and the low-precision ADC sampling sequence includes several different time points, including:

[0051] The waveform is generated by outputting a sine wave from 0 to a specific value and switching the sine wave frequency after a number of cycles through a chip with a DAC function;

[0052] The waveform is sampled by ADC chips with different precisions to obtain a first numerical sequence and a second numerical sequence.

[0053] In an optional embodiment, the chip with DAC function outputs a sine wave from 0 to a specific value and switches the sine wave frequency after several cycles to generate a waveform, specifically:

[0054] A chip with DAC function is used to output a sine wave between 0 and 3.3v, and the sine wave frequency is switched after outputting N cycles to generate a long series of waveforms.

[0055] In an optional embodiment, the waveform is sampled by ADC chips with different precisions to obtain the first numerical sequence and the second numerical sequence, specifically:

[0056] The waveform is sampled by a chip with an ADC function to obtain a first numerical sequence; wherein the accuracy of the chip with an ADC function is 10 bits or 12 bits, and the chip with an ADC function can be an MCU or a general independent ADC chip;

[0057] The second numerical sequence is obtained by sampling the waveform through a 24-bit ADC module.

[0058] In this way, by using ADC chips of different precisions to sample the same waveform, multiple sampling sequences of different precisions can be obtained. In this way, more abundant information can be obtained in one sampling process, including high-precision and low-precision sampling results. And because ADC chips of different precisions may have different sampling errors and noises, by comparing the first numerical sequence and the second numerical sequence, the sampling sequence of the ADC chip can be reconstructed. By comparing the sampling results at different time points, the value of the actual waveform can be estimated more accurately and the influence of sampling errors can be reduced. The numerical sequence obtained through multiple sampling time points can be used for subsequent neural network training to improve the precision and accuracy of neural network training.

[0059] It should be noted that the first numerical sequence is [B1B2 B3 B4 ...], and the second numerical sequence is [A1A2 A3A4 ...].

[0060] In a specific embodiment, the preset convolution kernels are: a first convolution kernel and a second convolution kernel; wherein the first convolution kernel is [-1 -1 2], and the second convolution kernel is [-1 2 -1].

[0061] In a specific embodiment, the first numerical sequence is passed through a plurality of preset convolution kernels to obtain a convolution sequence corresponding to the convolution kernels, including:

[0062] Using the element values ​​in the convolution kernel as weights, performing weighted summation on a number of elements in the first numerical sequence to obtain elements in the convolution sequence;

[0063] The number of the plurality of elements is the number of elements in the convolution kernel.

[0064] In a specific embodiment, the element values ​​in the convolution kernel are used as weights to perform weighted summation on a number of elements in the first numerical sequence to obtain elements in the convolution sequence, specifically:

[0065] Use convolution kernels [-1 -1 2] and [-1 2 -1] to convolve the first numerical sequence to obtain the elements in the convolution sequence, that is, X1=B1*(-1)+B2*(-1)+B3*(2), X2=B1*(-1)+B2*(2)+B3*(-1), and move the convolution kernel from left to right in the second sequence until all elements in the first numerical sequence have participated in the calculation, and generate the convolution value sequences X1(n) and X2(n).

[0066] In this way, different convolution kernels can capture different patterns and features. By using multiple preset convolution kernels, such as [-1 -1 2] and [-1 2 -1], it is possible to match and detect different features. Each convolution kernel can correspond to a specific feature. By performing a convolution operation on the first numerical sequence, a convolution sequence corresponding to the convolution kernel can be obtained, thereby realizing the detection and extraction of specific features.

[0067] In a specific embodiment, generating a training sample set according to the first numerical sequence, the second numerical sequence and all convolution sequences includes:

[0068] The elements in the convolution sequences and the corresponding elements in the first numerical sequence are used as convolution samples, and the moment values ​​in the second sequence are used as supervision values. The convolution samples and the corresponding supervision values ​​constitute training samples, and all the training samples constitute a training sample set.

[0069] In a specific embodiment, the convolution sample can be exemplified as: [B1, B2, B3, X1(1), X2(1)] The training sample is exemplified as: [B1, B2, B3, X1(1), X2(1), A3], where A3 is the time value corresponding to the maximum time value in the second sequence as the supervision value. The convolution sample is: [B(n), B(n+1), B(n+2), X1(n), X2(n), A(n+2)].

[0070] As shown in Table 1, the first, second and third columns are B(n), B(n+1) and B(n+2), the fourth and fifth columns are X1(n) and X2(n), and the sixth column is A(n+2).

[0071] 1270 1252 1220 -82 14 4298461 1252 1220 1193 -86 -5 4209338 1220 1193 1169 -75 -3 4121219 1193 1169 1143 -76 2 4029747 1143 1116 1089 -81 0 3843696

[0072] Table 1 is a sample table of training samples

[0073] It is understandable that by using such a training sample set, the neural network can learn the relationship between the convolution samples and the supervision values, so that the corresponding supervision values ​​can be predicted based on the convolution samples. In this way, when the third numerical sequence is reconstructed using the neural network, the sampling error can be corrected according to the prediction results of the neural network to improve the accuracy and precision of the ADC sampling values. And by using a training sample set consisting of multiple convolution sequences and corresponding supervision values, more sample data can be provided for neural network training. This helps to improve the generalization ability of the neural network, enabling it to better adapt to different sampling environments and noise conditions, and improve the robustness and reliability of the ADC sampling results.

[0074] In this way, by using the preset convolution kernel, the feature extraction of the first numerical sequence can be performed. The weight of the convolution kernel can capture the local patterns and features in the sequence. Through weighted summation, these features can be extracted from the original sequence and form a convolution sequence. This helps to better understand and analyze the information in the original sequence. And by using different preset convolution kernels, matching and detection of different features can be achieved. Different convolution kernels can capture different patterns and features. By performing a convolution operation on the first numerical sequence, a convolution sequence corresponding to the convolution kernel can be obtained, thereby realizing the detection and extraction of specific features.

[0075] It is understandable that the accuracy of ADC sampling can be improved by using multiple ADC sampling sequences of different precisions and utilizing convolution kernels and neural network training. By reconstructing the ADC sampling value sequence obtained by waveform sampling, the sampling error can be reduced and the accuracy of the sampling results can be improved. And by using the trained neural network to reconstruct the third value sequence, high-precision ADC data can be obtained using a low-precision sampling chip, so that there is no need to oversample the data, reducing the burden of data storage and data transmission, and improving the upper limit of the accuracy of the ADC sampling of the original low-precision sampling chip. At the same time, the neural network is used for training and reconstruction, which can adapt to different sampling environments and noise conditions. The sampling data is made more robust, and more reliable ADC sampling results can be obtained in a complex working environment.

[0076] Embodiment 2

[0077] Please refer to Figure 3 , an ADC super-resolution reconstruction device provided by an embodiment of the present invention includes: a reconstruction module 210;

[0078] The reconstruction module 210 is used to obtain a third value sequence by waveform sampling, and reconstruct the third value sequence according to the trained neural network to obtain an ADC sampling value sequence;

[0079] Wherein, the neural network is obtained by training a training sample set; the training sample set is obtained by passing a first numerical sequence through several preset convolution kernels to obtain a convolution sequence corresponding to the convolution kernel, and is generated according to the first numerical sequence, the second numerical sequence and all convolution sequences; the first numerical sequence and the second numerical sequence are at least two ADC sampling sequences with different precisions, and the low-precision ADC sampling sequence contains several different time points.

[0080] Furthermore, the reconstruction module 210 includes: a waveform generation unit 211 and an acquisition unit 212;

[0081] The waveform generating unit 211 is used to generate a waveform by outputting a sine wave from 0 to a specific value through a chip having a DAC function and switching the sine wave frequency after a number of cycles;

[0082] The acquisition unit 212 is used to sample the waveform through ADC chips with different precisions to obtain a first numerical sequence and a second numerical sequence.

[0083] Further, the reconstruction module 210 includes: a convolution unit 213;

[0084] The convolution unit 213 is used to use the element values ​​in the convolution kernel as weights to perform weighted summation on a number of elements in the first numerical sequence to obtain elements in the convolution sequence;

[0085] The number of the plurality of elements is the number of elements in the convolution kernel.

[0086] Further, the reconstruction module 210 includes: a convolution sample unit 214;

[0087] The convolution sample unit 214 is used to use elements in several convolution sequences and several corresponding elements in the first numerical sequence as convolution samples, and use the moment values ​​in the second sequence as supervision values ​​to form training samples with the convolution samples and the corresponding supervision values, and all training samples form a training sample set.

[0088] Embodiment 3

[0089] Accordingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to perform the ADC super-resolution reconstruction method described in any one of the above embodiments.

[0090] In this way, the same waveform is sampled by using multiple ADC chips with different precisions, and the convolution kernel and neural network training are used. The accuracy of ADC sampling can be improved by the trained neural network. By reconstructing the ADC sampling value sequence obtained by waveform sampling, the sampling error can be reduced and the accuracy of the sampling results can be improved. And by using low-precision sampling chips to obtain high-precision ADC data, there is no need to oversample the data, reducing the burden of data storage and data transmission. This can save system resources and increase the upper limit of the accuracy of the ADC sampling of the original low-precision sampling chip. At the same time, by using a training sample set composed of a preset convolution kernel and multiple convolution sequences, more sample data can be provided for neural network training. This helps to improve the generalization ability of the neural network, enabling it to better adapt to different sampling environments and noise conditions, and improve the robustness and reliability of the ADC sampling results.

[0091] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. An ADC super-resolution reconstruction method, characterized in that: include: Obtaining a third numerical sequence by waveform sampling, and reconstructing the third numerical sequence according to the trained neural network to obtain an ADC sampling numerical sequence; The neural network is obtained by training a training sample set; the training sample set is obtained by passing a first numerical sequence through a plurality of preset convolution kernels to obtain a convolution sequence corresponding to the convolution kernel, and is generated according to the first numerical sequence, the second numerical sequence and all the convolution sequences; specifically, the element values ​​in the convolution kernel are used as weights, and a plurality of elements in the first numerical sequence are weighted and summed to obtain elements in the convolution sequence; wherein the number of the plurality of elements is the number of elements in the convolution kernel; the elements in the plurality of convolution sequences and the corresponding plurality of elements in the first numerical sequence are used as convolution samples, and the convolution samples and the corresponding supervision values ​​are used as supervision values ​​according to the moment values ​​in the second sequence, and the convolution samples and the corresponding supervision values ​​constitute training samples, and all the training samples constitute a training sample set; the first numerical sequence and the second numerical sequence are at least two ADC sampling sequences with different precisions, and the low-precision ADC sampling sequence contains a plurality of different moment points, specifically: a sine wave from 0 to a specific value is output by a chip with a DAC function, and the sine wave frequency is switched after a plurality of cycles to generate a waveform; the waveform is sampled by ADC chips with different precisions to obtain the first numerical sequence and the second numerical sequence.

2. The ADC super-resolution reconstruction method according to claim 1, characterized in that: The preset convolution kernels are specifically: a first convolution kernel and a second convolution kernel; wherein the first convolution kernel is [-1-1 2], and the second convolution kernel is [-1 2-1].

3. An ADC super-resolution reconstruction device, characterized in that: include: Reconstruction module; The reconstruction module is used to obtain a third numerical sequence by waveform sampling, and reconstruct the third numerical sequence according to the trained neural network to obtain an ADC sampling numerical sequence; Wherein, the neural network is obtained by training a training sample set; the training sample set is obtained by passing a first numerical sequence through a plurality of preset convolution kernels to obtain a convolution sequence corresponding to the convolution kernel, and is generated according to the first numerical sequence, the second numerical sequence and all convolution sequences; and further includes: a convolution unit; the convolution unit is used to use the element values ​​in the convolution kernel as weights, and perform weighted summation on a plurality of elements in the first numerical sequence to obtain elements in the convolution sequence; wherein the number of the plurality of elements is the number of elements in the convolution kernel; a convolution sample unit; the convolution sample unit is used to use the elements in the plurality of convolution sequences and the corresponding plurality of elements in the first numerical sequence as convolution samples, According to the moment value in the second sequence as the supervision value, the convolution sample and the corresponding supervision value constitute the training sample, and all the training samples constitute the training sample set; the first numerical sequence and the second numerical sequence are at least two ADC sampling sequences with different precisions, and the low-precision ADC sampling sequence contains several different moment points, and also includes: a waveform generation unit and an acquisition unit; the waveform generation unit is used to generate a waveform by outputting a sine wave from 0 to a specific value through a chip with a DAC function and switching the sine wave frequency after several cycles; the acquisition unit is used to sample the waveform through ADC chips with different precisions to obtain the first numerical sequence and the second numerical sequence.

4. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the ADC super-resolution reconstruction method according to any one of claims 1 to 2.

Images