Broadband multichannel analog-to-digital converter calibration method and device

Abstract

The invention provides a broadband multichannel analog-to-digital converter calibration method and device, and the method comprises the steps: carrying out the sampling of an input signal through a plurality of channels, and obtaining a sampled signal; constructing a data set according to the sampled signals; constructing a cross-channel time sequence correlation model through a bidirectional long short-term memory network, processing a data set, and determining an inter-channel dynamic coupling relationship between gain errors and time mismatch among different channels; the data set is processed through channel identification codes of the multi-head attention module, and channel defect information of each sampling point is determined; fusing the inter-channel dynamic coupling relationship and the channel defect information to obtain fusion features; and correcting the combined large model according to the fusion features, and calibrating an output result of the multi-channel analog-to-digital converter through the combined model. Complex nonlinear errors and time correlation can be effectively processed through the bidirectional long-short-term memory network; the multi-head attention module is connected in parallel with the bidirectional long-short-term memory network, so that more attention can be exerted on sampling data from the same channel, and the correction precision is improved.

Description

Technical Field

[0001] This invention relates to the field of mobile communication technology, and in particular to a broadband multi-channel analog-to-digital converter calibration method and apparatus. Background Technology

[0002] With the development of 5G and even future 6G communication technologies, signal bandwidth is increasing dramatically, posing unprecedented challenges to the performance of Time-Interleaved Analog-to-Digital Converters (TIADCs). In ultra-high bandwidth communication scenarios, traditional TIADC correction methods face severe tests. For example, in base station communication systems, to achieve high-speed transmission of massive amounts of data, signal bandwidth is constantly expanding. Traditional mixed-signal correction methods rely on a large number of hardware components, such as accumulators and multipliers, making system design extremely complex and hardware resource consumption grow exponentially, making it difficult to meet the real-time and high-precision requirements under ultra-high bandwidth conditions. Moreover, in multi-channel TIADC systems, the number of channels increases, and the error coupling relationships between channels become more complex. Traditional methods, which correct through simple mathematical operations—for example, averaging or multiplication—are completely unable to handle these complex nonlinear errors and time dependencies. In large-scale data and high-noise environments, error correction almost fails.

[0003] Secondly, while neural network-based correction methods, such as Long Short-Term Memory (LSTM) and Recurrent Neural Networks (RNNs), have shown some effectiveness in previous applications, they also reveal problems in new scenarios. In rapidly changing signal environments, the unidirectional data flow structure commonly used in LSTM and RNN models cannot fully capture the bidirectional temporal correlation between consecutive points. For example, in fast frequency hopping communication signal processing, signal characteristics change rapidly, and unidirectional models cannot fully grasp the signal dynamics, leading to a significant reduction in correction accuracy. Furthermore, under conditions of high precision and complex error characteristics, these neural networks have extremely high computational complexity and extremely slow convergence speed. When faced with large-scale dataset tasks requiring real-time correction, such as the massive data processing in satellite communications, they simply cannot meet the real-time requirements. In addition, they ignore the inherent relationships between sampling results within the same channel, and cannot effectively utilize this information to improve correction accuracy when processing highly correlated multi-channel data. Summary of the Invention

[0004] This invention provides a broadband multi-channel analog-to-digital converter calibration method and apparatus to solve the defects of existing technologies that cannot meet real-time requirements and have low calibration accuracy, thereby achieving a balance between accuracy and real-time performance.

[0005] This invention provides a broadband multi-channel analog-to-digital converter calibration method, comprising the following steps: The input signal is sampled through multiple channels to obtain the sampled signal; a dataset is then constructed based on the sampled signal. A cross-channel temporal correlation model is constructed using a bidirectional long short-term memory network; the dataset is then processed using the cross-channel temporal correlation model to determine the dynamic coupling relationship between gain error and time mismatch across different channels; The dataset is processed by channel identifier encoding of the multi-head attention module to determine the channel defect information of each sampling point; The dynamic coupling relationship between channels and the channel defect information are fused to obtain the fusion characteristics; the joint large model is corrected based on the fusion characteristics, and the output results of the multi-channel analog-to-digital converter are calibrated through the joint model.

[0006] According to the present invention, a broadband multi-channel analog-to-digital converter calibration method is provided, which samples the input signal through multiple channels to obtain the sampled signal; and constructs a dataset based on the sampled signal; including: The first sampled signal is obtained by sampling the input signal through multiple channels; Under error-free conditions, the input signal is sampled through multiple channels to obtain the second sampled signal; A dataset is constructed by mapping each sampled value in the first sampled signal to each sampled value in the second sampled signal.

[0007] According to the present invention, a broadband multi-channel analog-to-digital converter calibration method is provided, which determines the channel defect information of each sampling point by processing the dataset through position encoding of a multi-head attention module, including: Introduce channel identifier encoding in the multi-head attention module; Channel-specific feature vectors are embedded in the dataset by channel identifier encoding, and the channel-specific feature vectors are used to determine the channel mapping relationship of sampling points; where the channel mapping relationship of sampling points refers to the mapping relationship between each sampling point and the channel to which the sampling point belongs. Channel defect information for each sampling point is determined by using a multi-head attention module and the channel mapping relationship between sampling points.

[0008] According to the present invention, a broadband multi-channel analog-to-digital converter calibration method determines the channel defect information of each sampling point through a multi-head attention module and a channel mapping relationship between sampling points, including: By employing a multi-head attention module and a multi-head attention mechanism, the channel mismatch error is decoupled from the performance characteristics to obtain specific error information corresponding to different error types in the channel mismatch error. Among them, the channel mismatch error includes: bias error sensing head, gain error sensing head, and time mismatch sensing head. The channel defect information of different error types for each sampling point is determined by the channel mapping relationship of the sampling points and specific error information.

[0009] According to the present invention, a broadband multi-channel analog-to-digital converter calibration method is provided, wherein the joint large model includes a bidirectional long short-term memory network and a multi-head attention module; the joint large model is corrected based on fusion features, including: The loss function is used to calculate the prediction difference between the model's predicted value and the actual error value based on the fusion features of the input. The parameters of the bidirectional long short-term memory network and the multi-head attention module are updated based on the backpropagation algorithm and the prediction difference to obtain the corrected joint large model.

[0010] The present invention also provides a broadband multi-channel analog-to-digital converter calibration device, comprising the following modules: The dataset construction module is used to sample the input signal through multiple channels to obtain the sampled signal; and to construct the dataset based on the sampled signal. The module for determining the dynamic coupling relationship between channels is used to construct a cross-channel temporal correlation model through a bidirectional long short-term memory network; and to process the dataset through the cross-channel temporal correlation model to determine the dynamic coupling relationship between gain error and time mismatch between different channels. The channel defect information determination module is used to process the dataset through the channel identifier encoding of the multi-head attention module to determine the channel defect information of each sampling point; The calibration module is used to fuse the dynamic coupling relationship between channels and the channel defect information to obtain the fusion feature; the joint large model is calibrated based on the fusion feature, and the output result of the multi-channel analog-to-digital converter is calibrated through the joint model.

[0011] According to the present invention, a broadband multi-channel analog-to-digital converter calibration device is provided, wherein the dataset construction module includes: The first sampling submodule is used to sample the input signal through multiple channels to obtain the first sampled signal; The second sampling submodule is used to sample the input signal through multiple channels to obtain the second sampled signal without error. The dataset construction submodule is used to construct a dataset by matching each sampled value in the first sampled signal with each sampled value in the second sampled signal.

[0012] According to the present invention, a broadband multi-channel analog-to-digital converter calibration device is provided, wherein the channel defect information determination module includes: The mapping relationship determination submodule is used to introduce channel identifier encoding in the multi-head attention module; channel-specific feature vectors are embedded in the dataset through channel identifier encoding, and the channel-specific feature vectors are used to determine the channel mapping relationship of sampling points; where the channel mapping relationship of sampling points refers to the mapping relationship between each sampling point and the channel to which the sampling point belongs; The channel defect information determination submodule is used to determine the channel defect information of each sampling point through the multi-head attention module and the channel mapping relationship of the sampling point.

[0013] According to the present invention, a broadband multi-channel analog-to-digital converter calibration device includes a channel defect information determination submodule, which uses a multi-head attention module to perform feature decoupling on channel mismatch errors using a multi-head attention mechanism to obtain specific error information corresponding to different error types in the channel mismatch errors. The channel mismatch errors include: a bias error sensing head, a gain error sensing head, and a time mismatch sensing head. The sampling point channel mapping relationship and the specific error information determine the channel defect information of different error types at each sampling point.

[0014] According to the present invention, a broadband multi-channel analog-to-digital converter calibration device is provided, wherein the joint large model includes a bidirectional long short-term memory network and a multi-head attention module; the calibration module is used to calculate the prediction difference between the model prediction value and the actual error value based on the fusion features of the input through a loss function; and to update the parameters of the bidirectional long short-term memory network and the multi-head attention module according to the backpropagation algorithm and the prediction difference to obtain the calibrated joint large model.

[0015] This invention provides a broadband multi-channel analog-to-digital converter (ADC) calibration device. It samples the input signal through multiple channels to obtain sampled signals; constructs a dataset based on the sampled signals; builds a cross-channel temporal correlation model using a bidirectional long short-term memory (LSTM) network; processes the dataset using the cross-channel temporal correlation model to determine the dynamic coupling relationship between gain error and time mismatch across different channels; processes the dataset using channel identifier encoding by a multi-head attention module to determine channel defect information at each sampling point; fuses the dynamic coupling relationship between channels and the channel defect information to obtain fusion features; corrects the joint large model based on the fusion features, and calibrates the output of the multi-channel ADC using the joint model. The bidirectional LSM network effectively handles complex nonlinear errors and time correlations, improving calibration accuracy while ensuring real-time performance. The parallel connection of the multi-head attention module and the bidirectional LSM network allows for greater attention to sampled data from the same channel, further improving calibration accuracy. The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the broadband multichannel analog-to-digital converter calibration method as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the broadband multi-channel analog-to-digital converter calibration method as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the broadband multichannel analog-to-digital converter calibration method as described above.

[0018] The broadband multi-channel analog-to-digital converter (ADC) calibration method and apparatus provided by this invention samples the input signal through multiple channels to obtain the sampled signal; constructs a dataset based on the sampled signal; builds a cross-channel temporal correlation model through a bidirectional long short-term memory (LSTM) network; processes the dataset through the cross-channel temporal correlation model to determine the dynamic coupling relationship between gain error and time mismatch between different channels; processes the dataset through channel identifier encoding of a multi-head attention module to determine the channel defect information of each sampling point; fuses the dynamic coupling relationship between channels and the channel defect information to obtain fusion features; corrects the joint large model based on the fusion features, and calibrates the output result of the multi-channel ADC through the joint model. The bidirectional LSM network can effectively handle complex nonlinear errors and temporal correlations, improving calibration accuracy while ensuring real-time performance. The multi-head attention module, connected in parallel with the bidirectional LSM network, can pay more attention to sampled data from the same channel, further improving calibration accuracy. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the broadband multi-channel analog-to-digital converter calibration method provided by the present invention.

[0021] Figure 2 This is the spectrum diagram of a 1GHz sinusoidal signal with error-free sampling by a four-channel TIADC 5GSPS provided by the present invention.

[0022] Figure 3 This invention provides a spectrum diagram of a four-channel TIADC 5GSPS with error sampling of a 1GHz sinusoidal signal. Figure 4 This is the design architecture diagram provided by the present invention.

[0023] Figure 5 This is the RTL verification design flowchart provided by the present invention.

[0024] Figure 6 This is a schematic diagram of the structure of the broadband multi-channel analog-to-digital converter calibration device provided by the present invention.

[0025] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] The following is combined Figures 1-7 This invention is described.

[0028] This invention mainly addresses the problems of gain error, time error, and bias error in TIADC systems, and proposes a correction method based on bidirectional long short-term memory network (BiLSTM) and attention mechanism.

[0029] Figure 1 This is one of the flowcharts illustrating the broadband multi-channel analog-to-digital converter calibration method provided by the present invention, such as... Figure 1 As shown, the method includes the following: Step 101: Sample the input signal through multiple channels to obtain the sampled signal; construct a dataset based on the sampled signal.

[0030] In step 101 above, this application constructs its own dataset to train the neural network. The dataset not only covers single-channel and multi-channel sampled data, but also includes various noise and interference conditions simulated from different transmission environments. This diversity of dataset ensures that the neural network can learn effectively in multiple environments, improving the model's generalization ability.

[0031] Optionally, step 101 includes steps A1 to A3: Step A1: Sample the input signal through multiple channels to obtain the first sampled signal.

[0032] Step A2: Under error-free conditions, sample the input signal through multiple channels to obtain the second sampled signal.

[0033] Step A3: Construct a dataset by matching each sampled value in the first sampled signal with each sampled value in the second sampled signal.

[0034] In steps A1 to A3 above, it is first necessary to model the input signal. For example, the input signal is: .

[0035] When there is When each channel samples the input signal, the sampled signal is obtained through formula (1): (1) Among them, Indicates the gain of each channel; This indicates the sampling delay for each channel; This is the bias error. In summary, these respectively contribute to gain error, time error, and bias error.

[0036] like Figure 2 As shown, when four channels sample the input signal, the displayed spectrum is a 1GHz sinusoidal signal spectrum of a four-channel TIADC 5GSPS with no error sampling.

[0037] In addition, labels for the error signals are required, therefore the input signal needs to be sampled again without errors, resulting in error-free multi-channel sampling signals. This can be achieved through formula (2): (2) like Figure 3 As shown, when the input signal is sampled by four channels without error, the spectrum of a 1GHz sine wave with error sampling is displayed by a four-channel TIADC 5GSPS.

[0038] The amplitude of the sampling points depends on the design of the input signal, and its value is usually greater than 1. Therefore, before constructing the dataset, this application normalizes the data to [0, 1], as shown in Equation (3). This helps to accelerate model training, prevent certain features from dominating the training process due to excessively large or small values, and enable the model to better learn the relationships between various features.

[0039] (3) After normalization, a dataset will be constructed based on a one-to-one correspondence between the sampled values ​​of the error-free and error-containing TIADC models.

[0040] 70% of the dataset is used as the training set, 20% as the validation set, and 10% as the test set. Each sample includes four channels of sampling data and their corresponding error labels. The dataset not only covers single-channel and multi-channel sampling data but also includes various noise and interference conditions simulated from different transmission environments. This diversity ensures that the neural network can learn effectively in multiple environments, improving the model's generalization ability.

[0041] Step 102: Construct a cross-channel temporal correlation model using a bidirectional long short-term memory network; process the dataset using the cross-channel temporal correlation model to determine the dynamic coupling relationship between gain error and time mismatch between different channels.

[0042] In step 102 above, this application proposes a BiLSTM and attention mechanism-based collaborative correction framework to address the channel mismatch error problem in TIADCs. This framework achieves accurate compensation for channel mismatch errors through multi-dimensional feature fusion.

[0043] Specifically, the input parameters for the BiLSTM are designed as follows: the data dimension input to the BiLSTM in this application is [N, T, M], where N is the channel dimension, T is the time series dimension, which is used by the BiLSTM to capture the dependencies between T sampling points in the channels. M is the size of the feature dimension of each sampling point. Since the focus of this application is on predicting the numerical value of each sampling point, M is set to 1.

[0044] Step 103: Process the dataset by encoding the channel identifiers of the multi-head attention module to determine the channel defect information of each sampling point.

[0045] In step 103 above, the system adopts a dual-modal feature extraction architecture: first, a cross-channel temporal correlation model is constructed through BiLSTM, and its bidirectional propagation characteristics are used to capture the dynamic coupling relationship between gain error and time mismatch between channels; second, a multi-head self-attention mechanism is introduced, and the channel information of the sampling point is retained through position encoding technology, effectively identifying inherent channel defects such as bias error and nonlinear distortion.

[0046] Optionally, step 103 includes steps B1 to B3: Step B1: Introduce channel identifier encoding in the multi-head attention module.

[0047] Step B2: Embed channel-specific feature vectors in the dataset through channel identifier encoding, and determine the channel mapping relationship of sampling points through the channel-specific feature vectors; where the channel mapping relationship of sampling points refers to the mapping relationship between each sampling point and the channel to which the sampling point belongs.

[0048] Step B3: Determine the channel defect information of each sampling point through the multi-head attention module and the channel mapping relationship of the sampling point.

[0049] In steps B1 to B3 above, to achieve accurate modeling of the inherent properties of the sampling channels, this application introduces channel identity encoding in the multi-head attention module. By embedding channel-specific feature vectors into the input sequence, a mapping relationship between sampling points and their source channels is established. This encoding mechanism adopts an orthogonal vector space design to ensure that the feature representations of different channels are statistically independent.

[0050] Optionally, step B3 includes steps B31 to B32: Step B31: The multi-head attention module is used to perform feature decoupling on the channel mismatch error by employing a multi-head attention mechanism to obtain specific error information corresponding to different error types in the channel mismatch error; among which, the channel mismatch error includes: bias error sensing head, gain error sensing head, and time mismatch sensing head; Step B32: Determine the channel defect information of different error types for each sampling point by using the sampling point channel mapping relationship and specific error information.

[0051] In steps B31 to B32 above, this application adopts a three-head attention mechanism (Head=3) to decouple features for three typical channel mismatch errors: bias error sensing head, gain error sensing head, and time mismatch sensing head. Each attention head contains an independent feature learning space. By jointly optimizing the mapping matrix of query-key-value triples (Q, K, V), selective attention to specific error types is achieved, specifically through the following formula (4).

[0052] (4) Where dk is the dimension of the key vector, this application sets dk = dk / Head to maintain computational stability.

[0053] Step 104: Fuse the dynamic coupling relationship between channels and the channel defect information to obtain the fusion characteristics; correct the joint large model based on the fusion characteristics, and calibrate the output results of the multi-channel analog-to-digital converter through the joint model.

[0054] In step 104 above, this application adopts a parallel design architecture for the BiLSTM network and the attention mechanism: the BiLSTM and the attention mechanism process the original input data simultaneously, and then the outputs of the two are fused. For example, the feature dimension of the BiLSTM output is... The feature dimension output by the multi-head attention mechanism is By fusing the two features, the dimension of the fused feature is: The resulting fusion incorporates both the dynamic coupling relationships between channels and the inherent defects of each channel itself.

[0055] Optionally, the joint large model includes a bidirectional long short-term memory network and a multi-head attention module. Step 104, which corrects the joint large model based on the fusion features, includes steps C1 to C2: Step C1: Calculate the prediction difference between the model prediction value and the true error value based on the fusion features of the input using the loss function.

[0056] Step C2: Update the parameters of the bidirectional long short-term memory network and the multi-head attention module according to the backpropagation algorithm and the prediction difference to obtain the corrected joint large model.

[0057] In steps C1 and C2 above, the fused features are input into subsequent fully connected layers for further processing. During training, based on the labels, a loss function, for example, the mean squared error loss function, is used to calculate the difference between the model's predicted values ​​and the true error values. Then, through the backpropagation algorithm, the parameters of the BiLSTM, the multi-head attention mechanism, and subsequent layers are updated, continuously adjusting the model to minimize the loss function.

[0058] like Figure 4 As shown, Figure 4 The diagram shown is the design architecture diagram of this application.

[0059] To verify the practical application effect of the proposed BiLSTM-Attention correction model, hardware-based RTL verification was performed. By converting the trained neural network model into a Hardware Description Language (HDL), we implemented hardware acceleration of the TIADC correction algorithm on an FPGA platform. The purpose of RTL verification is to test whether the designed hardware circuit can achieve the expected correction effect in practical applications and to verify the system's real-time performance. By inputting TIADC sampled data into the hardware platform and processing it in real time, the corrected output result was obtained and compared with the theoretical model, confirming the effectiveness of the BiLSTM model on the hardware platform.

[0060] like Figure 5 As shown, Figure 5 The RTL verification design flowchart is shown.

[0061] This application employs a parallel connection of BiLSTM and multi-head attention to correct errors in TIADC. BiLSTM focuses on the temporal sampling sequences between channels, addressing gain and timing errors; simultaneously, multi-head attention focuses on the same sampling channels, resolving inherent channel defects such as bias errors and nonlinear distortion. The extracted features are then fused for comprehensive fitting.

[0062] BiLSTM not only possesses powerful nonlinear modeling capabilities, effectively handling complex nonlinear errors and time-dependent issues, thus overcoming the shortcomings of traditional methods in effectively addressing such problems, but also exhibits low computational complexity and fast convergence speed. While improving correction accuracy, it ensures real-time performance, overcoming the problems of existing neural network methods, such as high computational complexity and slow convergence speed, which make it difficult to meet real-time correction requirements under high-precision and complex error characteristics.

[0063] BiLSTM considers the temporal correlation of both forward and backward traffic simultaneously through bidirectional data flow, enabling it to comprehensively capture the complex relationship between gain error and time error. The introduced attention mechanism, connected in parallel with BiLSTM, can pay more attention to sampled data from the same channel, taking into account the intrinsic relationship between sampled results in the same channel, which is lacking in existing neural network correction methods, thus further improving correction accuracy.

[0064] This invention provides a broadband multi-channel analog-to-digital converter (ADC) calibration method. The method involves sampling the input signal through multiple channels to obtain sampled signals; constructing a dataset based on the sampled signals; building a cross-channel temporal correlation model using a bidirectional long short-term memory (LSTM) network; processing the dataset using the cross-channel temporal correlation model to determine the dynamic coupling relationship between gain error and time mismatch across different channels; processing the dataset using channel identifier encoding by a multi-head attention module to determine channel defect information at each sampling point; fusing the dynamic coupling relationship between channels and the channel defect information to obtain fused features; correcting the joint large model based on the fused features; and calibrating the output of the multi-channel ADC using the joint model. The bidirectional LSM network effectively handles complex nonlinear errors and temporal correlations, improving calibration accuracy while ensuring real-time performance. The parallel connection of the multi-head attention module and the bidirectional LSM network allows for greater attention to sampled data from the same channels, further improving calibration accuracy.

[0065] The broadband multi-channel analog-to-digital converter calibration device provided by the present invention is described below. The broadband multi-channel analog-to-digital converter calibration device described below can be referred to in correspondence with the broadband multi-channel analog-to-digital converter calibration method described above.

[0066] Figure 6 This is one of the flowcharts of the broadband multi-channel analog-to-digital converter calibration device provided by the present invention, such as... Figure 6 As shown, the device includes the following: The dataset construction module 601 is used to sample the input signal through multiple channels to obtain the sampled signal; and to construct a dataset based on the sampled signal.

[0067] The inter-channel dynamic coupling relationship determination module 602 is used to construct a cross-channel timing correlation model through a bidirectional long short-term memory network; and to process the dataset through the cross-channel timing correlation model to determine the inter-channel dynamic coupling relationship between gain error and time mismatch in different channels. The channel defect information determination module 603 is used to process the dataset through the channel identifier encoding of the multi-head attention module to determine the channel defect information of each sampling point; The correction module 604 is used to fuse the dynamic coupling relationship between channels and the channel defect information to obtain the fusion feature; the joint large model is corrected according to the fusion feature, and the output result of the multi-channel analog-to-digital converter is calibrated through the joint model.

[0068] Optionally, the dataset construction module 601 includes: The first sampling submodule is used to sample the input signal through multiple channels to obtain the first sampled signal; The second sampling submodule is used to sample the input signal through multiple channels to obtain the second sampled signal without error. The dataset construction submodule is used to construct a dataset by matching each sampled value in the first sampled signal with each sampled value in the second sampled signal.

[0069] Optionally, the channel defect information determination module 603 includes: The mapping relationship determination submodule is used to introduce channel identifier encoding in the multi-head attention module; channel-specific feature vectors are embedded in the dataset through channel identifier encoding, and the channel-specific feature vectors are used to determine the channel mapping relationship of sampling points; where the channel mapping relationship of sampling points refers to the mapping relationship between each sampling point and the channel to which the sampling point belongs; The channel defect information determination submodule is used to determine the channel defect information of each sampling point through the multi-head attention module and the channel mapping relationship of the sampling point.

[0070] Optionally, the channel defect information determination submodule is used to perform feature decoupling on the channel mismatch error by using a multi-head attention mechanism through a multi-head attention module, so as to obtain specific error information corresponding to different error types in the channel mismatch error; wherein, the channel mismatch error includes: bias error sensing head, gain error sensing head, and time mismatch sensing head; the sampling point channel mapping relationship and specific error information determine the channel defect information of different error types of each sampling point.

[0071] Optionally, the joint large model includes a bidirectional long short-term memory network and a multi-head attention module; the correction module is used to calculate the prediction difference between the model prediction value and the true error value based on the fusion features of the input through a loss function; the parameters of the bidirectional long short-term memory network and the multi-head attention module are updated according to the backpropagation algorithm and the prediction difference to obtain the corrected joint large model.

[0072] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call logic instructions in the memory 830 to execute a wideband multi-channel analog-to-digital converter calibration method.

[0073] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program, the computer program being stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute the broadband multi-channel analog-to-digital converter calibration method provided by the above methods.

[0075] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the broadband multi-channel analog-to-digital converter calibration method provided by the methods described above.

[0076] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0077] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A calibration method for a broadband multi-channel analog-to-digital converter, characterized in that, include: The input signal is sampled through multiple channels to obtain the sampled signal; A dataset is constructed based on the sampled signal; A cross-channel temporal correlation model is constructed using a bidirectional long short-term memory network; the dataset is then processed using the cross-channel temporal correlation model to determine the dynamic coupling relationship between gain error and time mismatch across different channels. The dataset is processed by channel identifier encoding of the multi-head attention module to determine the channel defect information of each sampling point; The dynamic coupling relationship between the channels and the channel defect information are fused to obtain the fused feature; The joint large model is corrected based on the fusion features, and the output of the multi-channel analog-to-digital converter is calibrated through the joint model.

2. The broadband multi-channel analog-to-digital converter calibration method according to claim 1, characterized in that, The step of sampling the input signal through multiple channels to obtain a sampled signal and constructing a dataset based on the sampled signal includes: The first sampled signal is obtained by sampling the input signal through multiple channels; Under error-free conditions, the input signal is sampled through multiple channels to obtain the second sampled signal; A dataset is constructed by mapping each sample value in the first sampled signal to each sample value in the second sampled signal.

3. The broadband multi-channel analog-to-digital converter calibration method according to claim 1, characterized in that, The step of processing the dataset through positional encoding using a multi-head attention module to determine channel defect information for each sampling point includes: Introduce channel identifier encoding in the multi-head attention module; The channel identifier is encoded and embedded into the dataset to embed a channel-specific feature vector. The channel-specific feature vector is used to determine the channel mapping relationship of the sampling points. The channel mapping relationship of the sampling points refers to the mapping relationship between each sampling point and the channel to which the sampling point belongs. The channel defect information of each sampling point is determined by the multi-head attention module and the channel mapping relationship of the sampling points.

4. The broadband multi-channel analog-to-digital converter calibration method according to claim 1, characterized in that, The step of determining the channel defect information of each sampling point through the multi-head attention module and the channel mapping relationship of the sampling points includes: The multi-head attention module employs a multi-head attention mechanism to decouple the channel mismatch error features, thereby obtaining specific error information corresponding to different error types in the channel mismatch error; wherein, the channel mismatch error includes: bias error sensing head, gain error sensing head, and time mismatch sensing head; The channel defect information of different error types for each sampling point is determined by the channel mapping relationship of the sampling points and the specific error information.

5. The broadband multi-channel analog-to-digital converter calibration method according to claim 1, characterized in that, The joint large model includes the bidirectional long short-term memory network and the multi-head attention module; The step of correcting the joint large model based on the fusion features includes: The prediction difference between the model's predicted value and the actual error value is calculated using the loss function based on the input fusion features. The parameters of the bidirectional long short-term memory network and the parameters of the multi-head attention module are updated according to the backpropagation algorithm and the prediction difference to obtain the corrected joint large model.

6. A broadband multi-channel analog-to-digital converter calibration device, characterized in that, include: The dataset building module is used to sample the input signal through multiple channels to obtain the sampled signal; A dataset is constructed based on the sampled signal; The inter-channel dynamic coupling relationship determination module is used to construct a cross-channel temporal correlation model through a bidirectional long short-term memory network; and to process the dataset through the cross-channel temporal correlation model to determine the inter-channel dynamic coupling relationship between gain error and time mismatch in different channels. The channel defect information determination module is used to process the dataset through the channel identifier encoding of the multi-head attention module to determine the channel defect information of each sampling point; The correction module is used to fuse the dynamic coupling relationship between the channels and the channel defect information to obtain fusion features; correct the joint large model according to the fusion features, and calibrate the output results of the multi-channel analog-to-digital converter through the joint model.

7. The broadband multi-channel analog-to-digital converter calibration device according to claim 6, characterized in that, The dataset construction module includes: The first sampling submodule is used to sample the input signal through multiple channels to obtain the first sampled signal; The second sampling submodule is used to sample the input signal through multiple channels to obtain the second sampled signal without error. The dataset construction submodule is used to construct a dataset by matching each sampled value in the first sampled signal with each sampled value in the second sampled signal.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the broadband multi-channel analog-to-digital converter calibration method as described in any one of claims 1 to 5.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the broadband multichannel analog-to-digital converter calibration method as described in any one of claims 1 to 5.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the broadband multichannel analog-to-digital converter calibration method as described in any one of claims 1 to 5.

Images