Oscilloscope eye diagram diagnosis method and device, model training method and device, and oscilloscope equipment

Abstract

This application discloses an oscilloscope eye diagram diagnostic method, a model training method, an apparatus, and an oscilloscope device. The model training method, applicable to the field of electronic measurement technology, includes: acquiring multiple sets of historical eye diagram samples collected by an oscilloscope; each set of samples includes a historical eye diagram image, its signal integrity measurement parameters, and expert diagnostic text; inputting the historical eye diagram image into an initial visual encoder to obtain an image feature vector; inputting the signal integrity measurement parameters and expert diagnostic text into an initial text encoder to obtain a text feature vector; and jointly adjusting the parameters of the two encoders with the goal of reducing the distance between the image feature vector and the text feature vector of the same set of historical eye diagram samples, and increasing the distance between the image feature vector and the text feature vector of different sets of historical eye diagram samples, until a preset training termination condition is met, resulting in a pre-trained visual encoder. This pre-trained visual encoder provides a foundation for the comprehensiveness and accuracy of subsequent diagnostics.

Description

Technical Field

[0001] This application relates to the field of electronic measurement technology, and in particular to oscilloscope eye diagram diagnostic methods, model training methods, devices, and oscilloscope equipment. Background Technology

[0002] In the design and verification of high-speed digital systems, eye diagrams are a core technical means for evaluating signal integrity. By segmenting and overlaying sampled waveforms at unit intervals, eye diagrams can intuitively reflect characteristics such as amplitude noise, timing jitter, and inter-symbol interference. Traditionally, engineers would acquire eye diagrams using an oscilloscope and then rely on their personal experience to interpret physical parameters such as eye height, eye width, rise time, and jitter components, combining this with communication protocol standards to provide acceptance criteria and fault location suggestions. However, this manual diagnostic approach has significant limitations: firstly, the diagnostic quality is highly dependent on the engineer's expertise, making consistency difficult to guarantee; secondly, with the explosive growth in the number of high-speed links, the efficiency of manual interpretation can no longer meet the demands of large-scale testing.

[0003] To improve the automation of diagnosis, the industry has explored different technological approaches. One type of solution focuses on intelligent analysis of the eye diagram images themselves. For example, convolutional neural networks are used to perform end-to-end classification of eye diagram images, directly outputting a pass or fail judgment. While this method achieves automatic mapping from image to conclusion, its training process relies solely on category labels, resulting in a lack of interpretability in the diagnostic results. It cannot inform users which specific parameters of the signal quality are substandard, nor can it provide targeted investigation suggestions.

[0004] Another approach attempts to utilize large language models for signal diagnosis. For example, by acquiring the basic parameters of the signal source to be analyzed, these parameters, along with the user's question information, are input into a pre-defined large language model, and the model's diagnostic module outputs signal quality analysis conclusions. A common characteristic of these approaches is that the large language model relies solely on the input numerical parameter text for reasoning, lacking the ability to directly perceive the original waveform morphology of the signal. When faced with scenarios like eye diagrams, which contain complex waveform morphology information and where subtle anomalies often determine the diagnostic conclusion, relying solely on limited structured parameter text makes it difficult to capture key features in the waveform, fundamentally limiting the accuracy and comprehensiveness of the diagnosis. Summary of the Invention

[0005] To address the problems of existing technologies, embodiments of this application provide an oscilloscope eye diagram diagnosis method, a model training method, an apparatus, and an oscilloscope device. The technical solutions are as follows: On the one hand, a model training method for oscilloscope eye diagram diagnosis is provided, the method comprising: Acquire multiple sets of historical eye diagram samples collected by an oscilloscope; each set of historical eye diagram samples includes a historical eye diagram image, signal integrity measurement parameters corresponding to the historical eye diagram image, and expert diagnostic text. The historical eye diagram image is input into the initial visual encoder to obtain the image feature vector; and the signal integrity measurement parameters and the expert diagnostic text are input into the initial text encoder to obtain the text feature vector. The training objective is to reduce the distance between the image feature vector and the text feature vector of the same group of historical eye map samples in the latent feature space, and to increase the distance between the image feature vector and the text feature vector of different groups of historical eye map samples in the latent feature space. The parameters of the initial visual encoder and the initial text encoder are jointly adjusted until the preset training termination condition is met, and a pre-trained visual encoder is obtained.

[0006] On the other hand, a model training device for oscilloscope eye diagram diagnosis is provided, the device comprising: The sample acquisition module is used to acquire multiple sets of historical eye diagram samples collected by an oscilloscope; each set of historical eye diagram samples includes a historical eye diagram image, signal integrity measurement parameters corresponding to the historical eye diagram image, and expert diagnostic text. The vector generation module is used to input the historical eye diagram image into the initial visual encoder to obtain the image feature vector; and to input the signal integrity measurement parameters and the expert diagnostic text into the initial text encoder to obtain the text feature vector; The joint adjustment module is used to jointly adjust the parameters of the initial visual encoder and the initial text encoder with the training objective of reducing the distance between the image feature vector and the text feature vector of the same group of historical eye map samples in the latent feature space, and increasing the distance between the image feature vector and the text feature vector of different groups of historical eye map samples in the latent feature space, until a preset training termination condition is met, thereby obtaining a pre-trained visual encoder.

[0007] In one exemplary implementation, the vector generation module includes: The packaging processing module is used to package the signal integrity measurement parameters according to a preset dimension. The missing value handling module is used to fill the missing position with a preset value and generate a mask mark to identify the missing position if there is a missing value in the signal integrity measurement parameter during the packaging process. The splicing generation module is used to splice the packaged parameters with the expert diagnostic text and the mask mark to generate a parameter text sequence; The text input module is used to input the parameter text sequence into the initial text encoder to obtain the text feature vector.

[0008] On the other hand, an oscilloscope eye diagram diagnosis method is provided, the method comprising: The eye diagram under test acquired by the oscilloscope and the signal integrity measurement parameters obtained by the oscilloscope in real time measuring the eye diagram under test; The pre-trained visual encoder obtained using any of the above model training methods converts the eye map to be tested into a query feature vector. The query feature vector is used to retrieve similar historical cases from a pre-built vector database. The vector database stores multiple historical cases, each with a historical image feature vector generated by the pre-trained visual encoder as the index key, and is associated with the signal integrity measurement parameters and expert diagnostic text of the historical eye map. The signal integrity measurement parameters and expert diagnostic text associated with the similar historical cases are combined with the signal integrity measurement parameters of the eye diagram under test as prompt text and input into a pre-trained large language model to generate diagnostic results for the eye diagram under test.

[0009] On the other hand, an oscilloscope eye diagram diagnostic device is provided, the device comprising: The information acquisition module is used to acquire the eye diagram under test collected by the oscilloscope, and the signal integrity measurement parameters obtained by the oscilloscope in real time by measuring the eye diagram under test. The eye diagram conversion module is used to convert the eye diagram to be tested into a query feature vector using a pre-trained visual encoder obtained by any of the model training methods described above. The case retrieval module is used to retrieve similar historical cases from a pre-built vector database based on the query feature vector. The vector database stores multiple historical cases, each historical case using the historical image feature vector generated by the pre-trained visual encoder as the index key, and associated with the signal integrity measurement parameters and expert diagnostic text of the historical eye map. The diagnosis generation module is used to input the signal integrity measurement parameters and expert diagnostic text associated with the similar historical cases, together with the signal integrity measurement parameters of the eye diagram under test, as prompt text into a pre-trained large language model to generate a diagnostic result for the eye diagram under test.

[0010] In one exemplary implementation, the case retrieval module includes: The similarity calculation module is used to calculate the cosine similarity between the query feature vector and the feature vectors of each historical image in the vector database; The case determination module is used to determine a preset number of similar historical cases from multiple historical cases in the vector database based on cosine similarity; the cosine similarity of any one of the preset number of similar historical cases is greater than that of any other historical case in the multiple historical cases besides the preset number of similar historical cases.

[0011] In one exemplary embodiment, the cosine similarity of any one of the preset number of similar historical cases is greater than a preset similarity threshold; the device further includes: The "No Cases" prompt module is used to generate a prompt message indicating that the vector database lacks similar historical cases if there are no historical cases in the vector database with a cosine similarity greater than the preset similarity threshold.

[0012] In one exemplary implementation, the diagnostic generation module includes: The prompt splicing module is used to splice the signal integrity measurement parameters of the eye diagram under test, the signal integrity measurement parameters of the similar historical cases, the expert diagnosis text, and the cosine similarity of each similar historical case according to a preset template to form a structured prompt text; The prompt input module is used to input the prompt text into the pre-trained large language model to generate a diagnostic result for the eye diagram to be tested.

[0013] In one exemplary embodiment, the apparatus further includes a feedback processing module, the feedback processing module comprising: A feedback receiving module is used to receive user feedback information regarding the diagnostic results; the feedback information includes an effectiveness score and / or correction text; The sample caching module is used to store the eye diagram to be tested, the corresponding signal integrity measurement parameters, the diagnostic results, and the feedback information as feedback samples in the feedback sample cache area. The incremental update module is used to incrementally fine-tune the pre-trained large language model using the feedback samples when the number of feedback samples in the feedback sample buffer reaches a preset threshold, and to store the feedback samples as new historical cases in the vector database.

[0014] In one exemplary implementation, the incremental update module includes: The hybrid extraction module is used to extract a portion of historical cases from the stored historical cases and mix them with the feedback samples to form a training batch; The incremental fine-tuning module is used to perform incremental fine-tuning on the pre-trained large language model based on the training batch.

[0015] On the other hand, an oscilloscope device is provided, comprising: The signal acquisition unit is used to acquire the eye diagram under test and measure the eye diagram under test to obtain signal integrity measurement parameters; The memory stores a pre-trained visual encoder, a pre-trained large language model, and at least one instruction or at least one program. A processor for loading and executing the at least one instruction or the at least one program segment to implement the oscilloscope eye diagram diagnostic method of any of the above aspects.

[0016] This application embodiment extracts features from historical eye diagram images, corresponding structured measurement parameters, and expert diagnostic text using an initial visual encoder and an initial text encoder, respectively. The joint parameters are adjusted with the optimization objective of narrowing the feature distance between images and texts of the same case while widening the feature distance between different cases. This process forces the visual encoder, under the semantic supervision of the parameter text, to learn to capture subtle features related to physical parameters and fault causes from the original shape of the eye diagram, providing a foundation for generating interpretable diagnostic conclusions. Simultaneously, the pre-trained visual encoder can transform the eye diagram image into an image feature vector that internalizes signal integrity semantic information, effectively compensating for the perceptual deficiencies of relying solely on limited numerical parameters to represent the full picture of the eye diagram waveform. In subsequent applications, by retrieving historical cases similar to this image feature vector, reference parameters incorporating waveform morphology influences and expert diagnostic text can be obtained as prompts for a large language model, significantly improving the comprehensiveness and accuracy of the diagnosis. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the architecture of an oscilloscope eye diagram diagnostic system provided in an embodiment of this application; Figure 2 This is a schematic flowchart of a model training method for oscilloscope eye diagram diagnosis provided in an embodiment of this application; Figure 3 This is a schematic flowchart of an oscilloscope eye diagram diagnostic method provided in an embodiment of this application; Figure 4 This is a structural block diagram of a model training device for oscilloscope eye diagram diagnosis provided in an embodiment of this application; Figure 5 This is a structural block diagram of an oscilloscope eye diagram diagnostic device provided in an embodiment of this application; Figure 6 This is a hardware structure block diagram of an oscilloscope device provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0021] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0022] In existing technologies, signal integrity diagnostics typically treats eye diagram image analysis, physical parameter measurement, and maintenance experience records as independent steps, failing to integrate historical eye diagram images, corresponding signal integrity measurement parameters, and expert diagnostic texts from similar cases. When engineers discover signal degradation issues, they still need to manually sift through numerous documents or past test reports, resulting in low diagnostic efficiency. Furthermore, when new signal degradation patterns emerge (such as new power supply noise spectra or new printed circuit board crosstalk layouts), traditional solutions require re-collecting samples and initiating a complete training process, making it impossible to utilize historical diagnostic experience in real-time at the testing site.

[0023] Therefore, embodiments of this application provide an oscilloscope eye diagram diagnostic system, the architecture of which is as follows: Figure 1As shown, the system mainly includes a data acquisition and preprocessing unit, a bimodal vector database (knowledge base) construction unit, a retrieval augmented generation (RAG) engine, and a large language model (LLM) output unit. During the construction phase, the system acquires multiple sets of historical eye diagram samples collected by an oscilloscope through the data acquisition and preprocessing unit, forming a three-modal data foundation of "image-parameter-text" containing historical eye diagram images, signal integrity measurement parameters, and expert diagnostic text. The core of the bimodal vector database construction unit lies in employing a contrastive learning pre-training strategy to jointly adjust the visual encoder and text encoder, forcing the vector representation of the eye diagram image to align with its corresponding traditional measurement parameters in the embedding space. This ensures consistency between the visual understanding of artificial intelligence and the physical algorithm of the oscilloscope at the model structure level, making subsequent retrieval results more objective and reliable, rather than relying on subjective labels. After training, the system uses the image feature vector generated from the encoded historical eye diagrams as the index key, associating the corresponding signal integrity measurement parameters and expert diagnostic text to construct a vector database that supports efficient retrieval. In the application phase, the retrieval enhancement generation engine receives the eye diagram to be tested and its signal integrity measurement parameters acquired by an oscilloscope. It then vectorizes these parameters using a pre-trained visual encoder and retrieves similar historical cases that meet the similarity criteria from the vector database. The large language model output unit takes the parameters associated with the recalled cases, expert text, and real-time measurement parameters of the eye diagram to be tested as context, inputs them into the pre-trained large language model, generates diagnostic results supported by historical cases, and provides the retrieved similar cases to the user to enhance credibility. This system possesses lifelong learning capabilities, continuously accumulating samples through user feedback and incrementally fine-tuning the large language model, continuously expanding and optimizing the vector database. The following section details the specific methodological steps of the above technical solution.

[0024] Please see Figure 2 The diagram illustrates a model training method for oscilloscope eye diagram diagnosis provided in this application. It should be noted that while this specification provides method operation steps as shown in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only possible execution order. In actual system or product execution, the method can be executed sequentially according to the embodiments or accompanying drawings, or in parallel (e.g., in a parallel processor or multi-threaded processing environment). Specifically, as shown... Figure 2 As shown, the method may include: S201, acquire multiple sets of historical eye diagram samples collected by the oscilloscope.

[0025] Each set of historical eye diagram samples includes a historical eye diagram image, the corresponding signal integrity measurement parameters, and expert diagnostic text.

[0026] Specifically, each set of historical eye diagram samples is obtained through the following data acquisition and preprocessing process. First, historical eye diagram images are generated: the waveforms sampled by the oscilloscope are cut at unit intervals and superimposed to generate a bitmap with a fixed resolution as eye diagram data, for example, the resolution can be fixed at 751×521 pixels. Second, signal integrity measurement parameters are acquired, which include at least vertical parameters, horizontal parameters, and jitter parameters. Among them, vertical parameters include eye height and its percentage relative to the unit interval, top-eye level and bottom-eye level, and root mean square value of vertical noise; horizontal parameters include eye width and its percentage relative to the unit interval, rise time and fall time; jitter parameters include total jitter, random jitter, deterministic jitter, and duty cycle distortion at a specified bit error rate. In addition, a quality flag automatically generated by the instrument's built-in protocol standard can be acquired, which is used to give a pass or fail decision result. At the same time, expert diagnostic text is acquired, which includes the quality level marked by the personnel, the main failure modes, and the suggested troubleshooting direction. Quality levels are categorized as Excellent, Good, Borderline, or Unacceptable; primary failure modes include eye height not meeting standards, excessive jitter, severe noise, or significant inter-symbol interference; suggested troubleshooting directions include checking the transmitter's drive capability, replacing the connector, or adding an equalizer to the receiver. The above images, parameters, and text together constitute a set of historical eye diagram samples.

[0027] S203: Input the historical eye diagram image into the initial visual encoder to obtain the image feature vector; and input the signal integrity measurement parameters and expert diagnostic text into the initial text encoder to obtain the text feature vector.

[0028] The initial visual encoder refers to the untrained image feature extraction model.

[0029] The initial text encoder refers to the untrained text feature extraction model.

[0030] Specifically, the signal integrity measurement parameters are packaged according to preset dimensions. During the packaging process, if there are missing values ​​in the signal integrity measurement parameters, the missing positions are filled with preset values, and a mask mark is generated to identify the missing positions. The packaged parameters are concatenated with expert diagnostic text and mask marks to generate a parameter text sequence. The parameter text sequence is input into the initial text encoder to obtain the text feature vector.

[0031] The preset dimension refers to the length of the parameter vector predefined according to the measurement capabilities of the oscilloscope, such as 8 dimensions.

[0032] The preset value refers to the fixed fill value used to occupy the position of the missing value, such as 0.

[0033] Among them, the mask mark refers to the identifier bit that distinguishes the actual measurement value from the fill value, instructing the text encoder to perform differential processing on the two.

[0034] Specifically, during industrial field measurements, some parameters may be missing due to the test items not being enabled, and directly feeding them into the encoder would introduce noise. This application's embodiment distinguishes between "data not found" and "measured value of 0" by explicitly constructing a missing value mask. By introducing a masking mechanism, the model is allowed to explicitly know that data in a certain dimension is "unknown" rather than "zero," thus ignoring comparisons of that dimension during training and enabling incomplete data to effectively participate in model training.

[0035] Specifically, the aforementioned packaging of signal integrity measurement parameters according to a preset dimension means integrating the measured vertical parameters, horizontal parameters, jitter parameters, etc., into a parameter vector with a fixed dimension, which can be called the gold standard parameter vector. For example, the dimension of this gold standard parameter vector can be fixed at 8 dimensions to accommodate core indicators such as eye height, eye width, and various jitter parameters. During the packaging process, if there are missing values ​​in the signal integrity measurement parameters, such as a jitter parameter being missing due to measurement limitations, the corresponding missing position is filled with a preset value, such as 0, and a mask mark is generated to identify the missing position. This mask mark is used to enable the text encoder to distinguish between the missing value filling position and the actual measurement value position. Subsequently, the gold standard parameter vector obtained after packaging, the expert diagnostic text, and the mask mark are concatenated to generate a parameter text sequence, which is then input into the initial text encoder to obtain the corresponding text feature vector.

[0036] S205, with the training objective of reducing the distance between the image feature vector and the text feature vector of the same set of historical eye map samples in the latent feature space, and increasing the distance between the image feature vector and the text feature vector of different sets of historical eye map samples in the latent feature space, the parameters of the initial visual encoder and the initial text encoder are jointly adjusted until the preset training termination condition is met, and a pre-trained visual encoder is obtained.

[0037] Joint adjustment refers to updating the parameters of two encoders simultaneously and in a coordinated manner.

[0038] The latent feature space is a high-dimensional mathematical space composed of the encoder output vectors.

[0039] Among them, the cosine distance or Euclidean distance between distance pointers is used to measure the semantic similarity of multimodal data.

[0040] Specifically, with the comparative learning objective of reducing the distance within the same group and increasing the distance between different groups, end-to-end joint training is employed. Backpropagation simultaneously updates the parameters of both encoders, forcibly establishing alignment relationships between historical eye diagram images and information from two different modalities: signal integrity measurement parameters and expert diagnostic text, in the latent feature space. By forcibly aligning image feature vectors and parameter text feature vectors on the training target, consistency between visual understanding and physical measurement is ensured at the model structure level, enabling the direct retrieval of matching physical parameter cases based on images.

[0041] Specifically, this application employs a contrastive learning framework to jointly train historical eye diagram images and their corresponding gold standard parameter sets to the same latent feature space. The gold standard parameter set refers to the packaged signal integrity measurement parameters. The vector representations of the two in the embedding space are forcibly aligned after training, allowing direct calculation of similarity. Based on this, a vector repository containing image-gold standard parameter pairs is constructed. Regarding image data vectorization, the initial visual encoder can be implemented using an open-source visual model. For example, selectable visual models include Qwen3.6-35B-A3B or Qwen3.6-27B, used to encode eye diagram images to extract image feature vectors. Regarding gold standard parameter vectorization, the initial text encoder can be implemented using an embedding model to encode a parameterized text sequence that concatenates gold standard parameter vectors, mask markers, and expert diagnostic text to generate corresponding text feature vectors.

[0042] As can be seen from the above technical solutions of the embodiments of this application, the embodiments of this application extract features from historical eye diagram images and corresponding structured measurement parameters and expert diagnostic text through an initial visual encoder and an initial text encoder, respectively, and perform joint parameter adjustment with the optimization objective of narrowing the feature distance between the same case and widening the feature distance between different cases. This process forces the visual encoder to learn to capture subtle features related to physical parameters and fault causes from the original shape of the eye diagram under the semantic supervision of the parameter text, providing a basis for generating interpretable diagnostic conclusions. At the same time, the pre-trained visual encoder can transform the eye diagram image into an image feature vector that internalizes the semantic information of signal integrity, effectively making up for the perceptual defects of being unable to represent the full picture of the eye diagram waveform by simply relying on limited numerical parameters. In subsequent applications, by retrieving historical cases similar to the image feature vector, reference parameters and expert diagnostic text that incorporate the influence of waveform morphology can be obtained as prompts for a large language model, significantly improving the comprehensiveness and accuracy of the diagnosis.

[0043] Please see Figure 3 The diagram shown is a flowchart illustrating an oscilloscope eye diagram diagnostic method provided in an embodiment of this application. Specifically, as follows... Figure 3 As shown, the method may include: S301: Acquire the eye diagram under test acquired by the oscilloscope, and the signal integrity measurement parameters obtained by the oscilloscope in real time from the eye diagram under test.

[0044] When acquiring the eye diagram under test, its source is not limited to real-time acquisition by an oscilloscope. As a scalable implementation, the eye diagram under test can also be provided directly by the user. In other words, the eye diagram under test acquired in this step can be either a real-time measurement result directly acquired from an oscilloscope, or eye diagram data submitted by the user through local file import, historical data feedback, or other interactive methods. This allows the diagnostic system to flexibly adapt to different application scenarios such as online testing and offline analysis.

[0045] S303, using a pre-trained visual encoder obtained by any of the oscilloscope eye diagram diagnostic methods provided in the above method embodiments, the eye diagram to be tested is converted into a query feature vector.

[0046] The query feature vector refers to the image feature vector obtained after the eye diagram to be tested is input into the pre-trained visual encoder, which is used to search in the vector database.

[0047] S305: Based on the query feature vector, search the pre-built vector database to obtain similar historical cases.

[0048] The vector database stores multiple historical cases. Each historical case uses the historical image feature vector generated by the pre-trained visual encoder of the historical eye map as the index key, and is associated with the signal integrity measurement parameters of the historical eye map and expert diagnostic text.

[0049] Among them, similar historical cases refer to stored historical eye diagram cases that are sufficiently close to the eye diagram under test in the latent feature space.

[0050] Specifically, both structured and unstructured historical case data are uniformly converted into vector form to construct a bimodal index that supports efficient retrieval. This enables cross-modal mapping and retrieval between eye diagram images and corresponding gold standard parameter sets. To achieve efficient retrieval of bimodal data, a dictionary structure can be created and maintained, using the image feature vector of each historical case as the retrieval index key and the corresponding gold standard parameters and expert diagnostic text as associated attribute values. This allows for the direct retrieval of matching historical parameters and diagnostic records using the query feature vector as input. After training, the aforementioned image feature vectors and associated data are stored in a vector database for data persistence. This vector database can be, for example, Qdrant or a similar system.

[0051] Specifically, the cosine similarity between the query feature vector and the feature vectors of each historical image in the vector database is calculated; based on the cosine similarity, a preset number of similar historical cases are determined from multiple historical cases in the vector database; the cosine similarity of any one of the preset number of similar historical cases is greater than that of any other historical case among the multiple historical cases excluding the preset number of similar historical cases.

[0052] In the specific retrieval and matching process, the eye map to be tested is first parsed and vectorized to obtain the query feature vector. To maintain consistency with the similarity metric used in the aforementioned contrastive learning pre-training stage, this step also uses cosine similarity as the matching criterion in the inference stage, calculating the cosine similarity between the query feature vector and the historical image feature vector corresponding to each historical case in the vector database. As an exemplary implementation, the formula for calculating the cosine similarity is: (1) Where q represents the query feature vector. Let represent the historical image feature vector corresponding to the i-th historical case. After calculation using formula (1), the historical cases can be sorted from high to low according to their cosine similarity values, and a preset number of similar historical cases can be determined from them.

[0053] Furthermore, if the cosine similarity of any of the pre-set number of similar historical cases is greater than a pre-set similarity threshold, then if there are no historical cases in the vector database with a cosine similarity greater than the pre-set similarity threshold, a prompt message indicating that the vector database lacks similar historical cases will be generated.

[0054] In the above retrieval process, both the preset quantity and the preset similarity threshold are configurable retrieval parameters used to control the scale of the number of historical cases recalled and the minimum relevance requirement. The preset quantity refers to the number of the most similar historical cases returned, which can be called the top K historical cases. For example, K can be set to 5 or 10. The preset similarity threshold is a configurable minimum cosine similarity threshold, for example, 0.7. Cases below this threshold will not be returned. Therefore, when sorting by cosine similarity and determining the preset quantity of similar historical cases, it must be ensured that the cosine similarity of any returned similar historical case is greater than the preset similarity threshold. If no historical case in the vector database has a cosine similarity greater than the preset similarity threshold, it indicates that the vector database lacks sufficiently similar reference cases. In this case, a prompt indicating that the vector database lacks similar historical cases is generated, and low-quality matching results are no longer forcibly returned. When the similarity between the most matching case in the vector database and the test sample is still below the threshold, the system actively interrupts the subsequent generation process to avoid inputting irrelevant case text into the large language model and generating misleading diagnoses.

[0055] For each similar historical case, the search results will output the following information: the similarity score between the case and the eye diagram to be tested; the gold standard parameters associated with the case, i.e., the original physical values ​​of the signal integrity measurement parameters, such as eye height = 85mV; and the expert diagnostic text associated with the case, such as "Unqualified, insufficient eye height, connector inspection recommended". Thus, each returned similar historical case contains quantifiable matching indicators, traceable measured physical parameters, and historical diagnostic experience for reference, providing complete contextual basis for subsequent diagnostic results.

[0056] S307: The signal integrity measurement parameters associated with similar historical cases and expert diagnostic text, together with the signal integrity measurement parameters of the eye diagram under test, are used as prompt text and input into a pre-trained large language model to generate diagnostic results for the eye diagram under test.

[0057] The prompt text refers to a structured input that combines the signal integrity measurement parameters of the eye diagram under test, parameters of similar historical cases, and expert diagnostic text, and is used to guide the large language model in generating responses.

[0058] Specifically, in this embodiment, similar historical cases are first found through cross-modal retrieval, and then the reasoning and inductive ability of the large language model is used to make a comprehensive judgment by combining the old and new data to generate a diagnostic conclusion.

[0059] Specifically, the signal integrity measurement parameters and expert diagnostic text associated with similar historical cases are concatenated with the signal integrity measurement parameters of the eye diagram under test according to a preset template to assemble a prompt word containing complete retrieval information. After receiving this prompt word, the pre-trained large language model calls the model to infer and generate a diagnostic answer in natural language form. Before output, the generated answer undergoes consistency verification and formatting to ensure that its content conforms to physical standards and engineering common sense, avoiding erroneous statements that violate signal integrity principles. Regarding model selection, as a specific implementation method, the pre-trained large language model can be a lightweight instruction fine-tuning model with a parameter scale between 7 billion and 13 billion. For example, models such as Qwen-7B-Chat, Llama-3-8B-Instruct, or DeepSeek-7B can be selected. Such models can balance inference speed and diagnostic effectiveness and can be easily deployed on a single A10 or A100 graphics processor. In terms of generation control, deterministic generation parameters can be adopted, such as setting the temperature coefficient to 0.2, the kernel sampling probability threshold to 0.7, and the maximum output length to 512 tokens, to suppress randomness and ensure the stability and consistency of diagnostic conclusions. Regarding output format, the final diagnostic results output by the large language model can adopt a structured data exchange format or a markup language format, such as JSON or Markdown, to facilitate secondary parsing and interface display by downstream systems.

[0060] Specifically, following a preset template, the signal integrity measurement parameters of the eye diagram under test, the signal integrity measurement parameters of similar historical cases, expert diagnostic text, and the cosine similarity of each similar historical case are concatenated to form a structured prompt text. The prompt text is then input into a pre-trained large language model to generate diagnostic results for the eye diagram under test.

[0061] Among them, the preset template refers to a pre-designed text format that includes fixed terms and variable placeholders.

[0062] Among them, structured prompt text refers to text instructions that conform to a preset format and can be accurately understood by a large language model.

[0063] Regarding the interpretability of diagnostic output, most existing eye diagram recognition solutions only output an abstract score or classification label, failing to inform users of the physical root cause of signal failure. This application's embodiment uses a large language model to reason about the results generated by retrieval enhancement, automatically outputting a natural language diagnostic report containing complete diagnostic elements. This report uses quantitative parameters as a quantitative basis, clearly identifies the failure type, and provides specific troubleshooting suggestions, with a significantly reduced average generation time. Traditional manual analysis of an eye diagram report involves multiple steps, including locating the problem, reviewing documents, and writing conclusions, which is time-consuming and relies on personal experience. This application's embodiment automatically associates and presents similar historical cases through a retrieval enhancement generation engine, effectively avoiding repeated reviews and significantly improving diagnostic efficiency. In terms of dynamic knowledge expansion, existing static models cannot learn new cases on-site. Whenever a new signal degradation pattern (such as a new power supply noise spectrum) appears, data needs to be collected and the model retrained, resulting in a long cycle and high cost. This application's embodiment, with its unique vector database and retrieval enhancement generation architecture, allows engineers to directly store newly encountered eye diagram cases, their signal integrity measurement parameters, and manually corrected text into the vector database on-site, without retraining any model. Once a new case is added to the database, the retrieval enhancement generation engine can recall that case in subsequent searches, enabling dynamic knowledge expansion and lifelong learning for the system. Regarding hardware compatibility, this embodiment relies solely on the oscilloscope's existing output interface to acquire waveform data or eye diagram images, and signal integrity measurement parameters calculated using traditional algorithms, without requiring any modification to the oscilloscope hardware. The entire system can run as a software plugin or cloud service, is compatible with mainstream commercially available oscilloscopes, and possesses excellent versatility and deployment flexibility. In summary, this embodiment achieves a high degree of consistency between artificial intelligence semantic evaluation and physical measurement standards in the field of oscilloscope eye diagram analysis. By forcing image features and parameter features to align at the model structure level through contrastive learning, the objectivity and traceability of diagnostic conclusions are ensured. Furthermore, by constraining the output of the large language model through consistency verification and formatting processing, the technical pain points of existing models, caused by reliance on subjective labels, such as uninterpretability and disconnection from standards, are completely resolved. Simultaneously, the retrieval enhancement generation architecture endows the system with dynamic knowledge expansion capabilities, allowing it to adapt to new protocols and fault modes without retraining, providing a verifiable, traceable, and sustainably evolving general technical path for intelligent electronic measurement instruments.

[0064] As can be seen from the above technical solutions of the embodiments of this application, the embodiments of this application utilize a visual encoder pre-trained through cross-modal contrastive learning, enabling the visual encoder to learn to map the visual features of the eye diagram to the same feature space aligned with physical parameters and fault semantics. Based on this key capability, during diagnosis, after the eye diagram under test is converted into a query feature vector by the visual encoder, similar historical cases with semantic relevance can be automatically retrieved from the pre-built vector database through vector similarity matching, eliminating the need for engineers to manually review documents. Moreover, even when facing unseen degradation patterns such as novel power supply noise spectra or printed circuit board crosstalk layouts, as long as the eye diagram morphology and parameter distribution have measurable similarity to historical cases in the feature space, it can be effectively matched without the need to recollect samples and retrain the model. Based on this, the signal integrity measurement parameters and expert diagnostic text associated with similar historical cases obtained from the retrieval are used together with the measurement parameters of the eye diagram under test as input text into the pre-trained large language model. This allows the diagnostic results to reference both measured physical quantities and historical expert experience, ultimately enabling real-time access and comprehensive analysis of long-accumulated diagnostic knowledge at the test site, significantly improving diagnostic efficiency and adaptability to unknown fault modes.

[0065] In one exemplary embodiment, after step S307 above, the following steps may also be included: The system receives user feedback on the diagnostic results. This feedback includes a validity score and / or corrected text. Specifically, the system receives feedback through a user feedback interface, which provides a scoring control, exemplarily a "helpful" or "useless" button, allowing users to rate the validity of the model's current diagnostic output. The interface also provides a text input area, allowing users to directly submit corrected text. The system stores the eye diagram to be tested, along with corresponding signal integrity measurement parameters, diagnostic results, and feedback information, as feedback samples in a feedback sample cache. When the number of feedback samples in the cache reaches a preset threshold, the pre-trained large language model is incrementally fine-tuned using these samples, and the feedback samples are stored as new historical cases in the vector database.

[0066] The preset threshold number serves as the trigger condition for initiating closed-loop learning. When the number of feedback samples accumulated in the feedback sample buffer reaches a sufficient scale, the system automatically triggers the warm-start process. As an exemplary implementation, this preset threshold number can be set to 100. After triggering the warm-start, the system adds the feedback samples as new labeled samples to the vector database to expand the case coverage of the vector database. Simultaneously, it incrementally fine-tunes the pre-trained large language model according to a regular cycle, which can be set to weekly, for example.

[0067] The incremental fine-tuning of the pre-trained large language model using feedback samples includes: extracting some historical cases from the stored historical cases and mixing them with feedback samples to form training batches; and performing incremental fine-tuning of the pre-trained large language model based on the training batches.

[0068] During incremental fine-tuning, an experience replay strategy is employed. This involves extracting a portion of historical cases from the stored dataset and mixing them with feedback samples to form a training batch. Based on this training batch, the pre-trained large language model is incrementally fine-tuned to prevent the model from forgetting previously acquired knowledge when learning new cases, thus avoiding catastrophic forgetting. Catastrophic forgetting refers to the phenomenon where a neural network almost completely loses its ability to process old knowledge after learning new knowledge.

[0069] As can be seen from the above technical solutions of the embodiments of this application, the embodiments of this application automatically collect user feedback information on diagnostic results during daily diagnosis and continuously store feedback samples containing correction text in the cache. When a preset threshold is reached, a closed-loop update is automatically triggered, so that the system can transform the latest diagnostic experience of on-site engineers into new knowledge of the model in real time without manual intervention. At the same time, in the process of incrementally fine-tuning the pre-trained large language model using feedback samples, an experience playback strategy is adopted to extract some old cases from the stored historical cases and mix them with feedback samples to form a training batch. This effectively prevents the model from catastrophically forgetting the fault diagnosis knowledge it has mastered when learning new signal degradation patterns. Thus, the system's diagnostic capability continues to evolve steadily with the increase of usage time, becoming more and more accurate with use.

[0070] Corresponding to the model training methods for oscilloscope eye diagram diagnosis provided in the above embodiments, this application also provides a model training device for oscilloscope eye diagram diagnosis. Since the model training device for oscilloscope eye diagram diagnosis provided in this application corresponds to the model training methods for oscilloscope eye diagram diagnosis provided in the above embodiments, the implementation methods of the aforementioned model training methods for oscilloscope eye diagram diagnosis are also applicable to the model training device for oscilloscope eye diagram diagnosis provided in this embodiment, and will not be described in detail in this embodiment.

[0071] Please see Figure 4 The diagram shows a schematic representation of a model training device for oscilloscope eye diagram diagnosis provided in an embodiment of this application. This model training device 400 has the function of implementing the model training method for oscilloscope eye diagram diagnosis described in the above-described method embodiments. This function can be implemented in hardware or by hardware executing corresponding software. Figure 4 As shown, the model training device 400 may include: The sample acquisition module 410 is used to acquire multiple sets of historical eye diagram samples collected by the oscilloscope; each set of historical eye diagram samples includes a historical eye diagram image, signal integrity measurement parameters corresponding to the historical eye diagram image, and expert diagnostic text. The vector generation module 420 is used to input historical eye diagram images into the initial visual encoder to obtain image feature vectors; and to input signal integrity measurement parameters and expert diagnostic text into the initial text encoder to obtain text feature vectors; The joint adjustment module 430 is used to jointly adjust the parameters of the initial visual encoder and the initial text encoder with the training objective of reducing the distance between the image feature vector and the text feature vector of the same group of historical eye map samples in the latent feature space and increasing the distance between the image feature vector and the text feature vector of different groups of historical eye map samples in the latent feature space until the preset training termination condition is met, so as to obtain the pre-trained visual encoder.

[0072] In one exemplary embodiment, the vector generation module 420 includes: The packaging processing module is used to package signal integrity measurement parameters according to preset dimensions; The missing value handling module is used to fill the missing position with a preset value and generate a mask mark to identify the missing position if there is a missing value in the signal integrity measurement parameters during the packaging process. The splicing generation module is used to splice the packaged parameters with expert diagnostic text and mask markers to generate a parameter text sequence; The text input module is used to input the parameter text sequence into the initial text encoder to obtain the text feature vector.

[0073] Corresponding to the oscilloscope eye diagram diagnostic methods provided in the above embodiments, this application also provides an oscilloscope eye diagram diagnostic device. Since the oscilloscope eye diagram diagnostic device provided in this application corresponds to the oscilloscope eye diagram diagnostic methods provided in the above embodiments, the implementation methods of the aforementioned oscilloscope eye diagram diagnostic methods are also applicable to the oscilloscope eye diagram diagnostic device provided in this embodiment, and will not be described in detail in this embodiment.

[0074] Please see Figure 5 The diagram shown is a structural schematic of an oscilloscope eye diagram diagnostic device provided in an embodiment of this application. This oscilloscope eye diagram diagnostic device 500 has the function of implementing the oscilloscope eye diagram diagnostic method described in the above-described method embodiments. This function can be implemented in hardware or by hardware executing corresponding software. Figure 5 As shown, the oscilloscope eye diagram diagnostic device 500 may include: The information acquisition module 510 is used to acquire the eye diagram under test collected by the oscilloscope, and the signal integrity measurement parameters obtained by the oscilloscope in real time by measuring the eye diagram under test. The eye diagram conversion module 520 is used to convert the eye diagram to be tested into a query feature vector using a pre-trained visual encoder obtained by any of the model training methods described above. The case retrieval module 530 is used to retrieve similar historical cases from a pre-built vector database based on the query feature vector. The vector database stores multiple historical cases, each with a historical image feature vector generated by a pre-trained visual encoder of the historical eye map as the index key, and is associated with the signal integrity measurement parameters of the historical eye map and expert diagnostic text. The diagnosis generation module 540 is used to input the signal integrity measurement parameters associated with similar historical cases and expert diagnostic text, together with the signal integrity measurement parameters of the eye diagram under test, as prompt text into a pre-trained large language model to generate diagnostic results for the eye diagram under test.

[0075] In one exemplary implementation, the case retrieval module 530 includes: The similarity calculation module is used to calculate the cosine similarity between the query feature vector and the feature vectors of each historical image in the vector database; The case determination module is used to determine a preset number of similar historical cases from multiple historical cases in the vector database based on cosine similarity; the cosine similarity of any similar historical case among the preset number of similar historical cases is greater than that of any historical case other than the preset number of similar historical cases among the multiple historical cases.

[0076] In one exemplary embodiment, the cosine similarity of any similar historical case among a preset number of similar historical cases is greater than a preset similarity threshold; the oscilloscope eye diagram diagnostic device 500 further includes: The "No Cases" module is used to generate a message indicating that the vector database lacks similar historical cases if no historical cases with a cosine similarity greater than a preset similarity threshold exist in the vector database.

[0077] In one exemplary embodiment, the diagnostic generation module 540 includes: The prompt splicing module is used to splice the signal integrity measurement parameters of the eye diagram under test, the signal integrity measurement parameters of similar historical cases, the expert diagnosis text, and the cosine similarity of each similar historical case according to a preset template to form a structured prompt text; The prompt input module is used to input prompt text into a pre-trained large language model to generate diagnostic results for the eye diagram under test.

[0078] In one exemplary embodiment, the oscilloscope eye diagram diagnostic device 500 further includes a feedback processing module, which includes: The feedback receiving module is used to receive user feedback on the diagnostic results; the feedback information includes effectiveness scores and / or correction text. The sample caching module is used to store the eye diagram to be tested, as well as the corresponding signal integrity measurement parameters, diagnostic results, and feedback information as feedback samples in the feedback sample cache area. The incremental update module is used to incrementally fine-tune the pre-trained large language model using feedback samples when the number of feedback samples in the feedback sample buffer reaches a preset threshold, and to store the feedback samples as new historical cases in the vector database.

[0079] In one exemplary implementation, the incremental update module includes: The hybrid extraction module is used to extract a portion of historical cases from the stored historical cases and mix them with the feedback samples to form a training batch. The incremental fine-tuning module is used to perform incremental fine-tuning on pre-trained large language models based on training batches.

[0080] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0081] This application provides an oscilloscope device, which includes: The signal acquisition unit is used to acquire the eye diagram under test and measure the eye diagram under test to obtain signal integrity measurement parameters; The memory stores a pre-trained visual encoder, a pre-trained large language model, and at least one instruction or at least one program. A processor is configured to load and execute the at least one instruction or the at least one program segment to implement any of the oscilloscope eye diagram diagnostic methods provided in the above method embodiments.

[0082] Memory is used to store software programs and modules. The processor executes these stored software programs and modules to perform various functional applications and data processing. Memory can primarily consist of a program storage area and a data storage area. The program storage area stores the operating system, application programs required for functionality, etc.; the data storage area stores data created based on device usage, etc. Furthermore, memory can include high-speed random access memory (RAM) and non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, memory can also include a memory controller to provide the processor with access to the memory.

[0083] The method embodiments provided in this application can be executed in a computer terminal, server or similar computing device, that is, the above-mentioned oscilloscope device may include a computer terminal, server or similar computing device. Figure 6 This is a hardware structure block diagram of a computer device for running an oscilloscope eye diagram diagnostic method provided in an embodiment of the present invention, such as... Figure 6 As shown, the internal structure of this computer device may include, but is not limited to, a processor, a network interface, and a memory. The processor, network interface, and memory within the computer device can be connected via a bus or other means, as illustrated in the embodiments of this specification. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0084] The processor (or CPU, Central Processing Unit) is the computing and control core of the computer device. The network interface may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, mobile communication interface, etc.). The memory is the storage device in the computer device used to store programs and data. It is understood that the memory here can be a high-speed RAM storage device, or a non-volatile memory device, such as at least one disk storage device; optionally, it can also be at least one storage device located remotely from the aforementioned processor. The memory provides storage space, which stores the operating system of the oscilloscope device, including but not limited to: Windows (an operating system), Linux (an operating system), Android (a mobile operating system), iOS (a mobile operating system), etc., which are not limited in this invention; and the storage space also stores one or more instructions suitable for loading and execution by the processor, which can be one or more computer programs (including program code). In the embodiments of this specification, the processor loads and executes one or more instructions stored in the memory to implement the oscilloscope eye diagram diagnostic method provided in the above method embodiments.

[0085] Embodiments of this application also provide a computer-readable storage medium that can be disposed in an oscilloscope device to store at least one instruction or at least one program related to implementing an oscilloscope eye diagram diagnosis method. The at least one instruction or the at least one program is loaded and executed by the processor to implement any of the oscilloscope eye diagram diagnosis methods provided in the above-described method embodiments.

[0086] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0087] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0088] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0089] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0090] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A model training method for oscilloscope eye diagram diagnosis, characterized in that, include: Acquire multiple sets of historical eye diagram samples collected by an oscilloscope; each set of historical eye diagram samples includes a historical eye diagram image, signal integrity measurement parameters corresponding to the historical eye diagram image, and expert diagnostic text. The historical eye diagram image is input into the initial visual encoder to obtain the image feature vector; and the signal integrity measurement parameters and the expert diagnostic text are input into the initial text encoder to obtain the text feature vector. The training objective is to reduce the distance between the image feature vector and the text feature vector of the same group of historical eye map samples in the latent feature space, and to increase the distance between the image feature vector and the text feature vector of different groups of historical eye map samples in the latent feature space. The parameters of the initial visual encoder and the initial text encoder are jointly adjusted until the preset training termination condition is met, and a pre-trained visual encoder is obtained.

2. The training method according to claim 1, characterized in that, The step of inputting the signal integrity measurement parameters and the expert diagnostic text into the initial text encoder to obtain the text feature vector includes: The signal integrity measurement parameters are packaged according to preset dimensions; During the packaging process, if there are missing values ​​in the signal integrity measurement parameters, the missing positions will be filled with preset values, and a mask mark will be generated to identify the missing positions. The packaged parameters are concatenated with the expert diagnostic text and the mask marker to generate a parameter text sequence; The parameter text sequence is input into the initial text encoder to obtain the text feature vector.

3. An oscilloscope eye diagram diagnostic method, characterized in that, include: The eye diagram under test acquired by the oscilloscope and the signal integrity measurement parameters obtained by the oscilloscope in real time measuring the eye diagram under test are also obtained. Using the pre-trained visual encoder obtained by the training method as described in any one of claims 1 to 2, the eye diagram to be tested is converted into a query feature vector; The query feature vector is used to retrieve similar historical cases from a pre-built vector database. The vector database stores multiple historical cases, each with a historical image feature vector generated by the pre-trained visual encoder as the index key, and is associated with the signal integrity measurement parameters and expert diagnostic text of the historical eye map. The signal integrity measurement parameters and expert diagnostic text associated with the similar historical cases are combined with the signal integrity measurement parameters of the eye diagram under test as prompt text and input into a pre-trained large language model to generate diagnostic results for the eye diagram under test.

4. The diagnostic method according to claim 3, characterized in that, The step of retrieving similar historical cases from a pre-built vector database based on the query feature vector includes: Calculate the cosine similarity between the query feature vector and the feature vectors of each historical image in the vector database; A preset number of similar historical cases are determined from multiple historical cases in the vector database based on cosine similarity; the cosine similarity of any one of the preset number of similar historical cases is greater than that of any other historical case in the multiple historical cases besides the preset number of similar historical cases.

5. The diagnostic method according to claim 4, characterized in that, The method further includes: The cosine similarity of any one of the preset number of similar historical cases is greater than a preset similarity threshold; If there are no historical cases in the vector database with a cosine similarity greater than the preset similarity threshold, a prompt message indicating that the vector database lacks similar historical cases will be generated.

6. The diagnostic method according to claim 3, characterized in that, The step of using the signal integrity measurement parameters and expert diagnostic text associated with the similar historical cases, together with the signal integrity measurement parameters of the eye diagram under test, as prompt text, and inputting them into a pre-trained large language model to generate a diagnostic result for the eye diagram under test, includes: According to the preset template, the signal integrity measurement parameters of the eye diagram to be tested, the signal integrity measurement parameters of the similar historical cases, the expert diagnosis text, and the cosine similarity of each similar historical case are spliced ​​together to form a structured prompt text; The prompt text is input into the pre-trained large language model to generate a diagnostic result for the eye diagram under test.

7. The diagnostic method according to any one of claims 3 to 6, characterized in that, The method further includes: Receive user feedback on the diagnostic results; the feedback includes an effectiveness score and / or corrected text. The eye diagram to be tested, the corresponding signal integrity measurement parameters, the diagnostic results, and the feedback information are used as feedback samples and stored in the feedback sample cache area. When the number of feedback samples in the feedback sample buffer reaches a preset threshold, the pre-trained large language model is incrementally fine-tuned using the feedback samples, and the feedback samples are stored as new historical cases in the vector database.

8. The diagnostic method according to claim 7, characterized in that, The incremental fine-tuning of the pre-trained large language model using the feedback samples includes: A portion of historical cases are extracted from the stored historical cases and mixed with the feedback samples to form a training batch; Incremental fine-tuning is performed on the pre-trained large language model based on the training batches.

9. A model training device for oscilloscope eye diagram diagnosis, characterized in that, include: The sample acquisition module is used to acquire multiple sets of historical eye diagram samples collected by an oscilloscope; each set of historical eye diagram samples includes a historical eye diagram image, signal integrity measurement parameters corresponding to the historical eye diagram image, and expert diagnostic text. The vector generation module is used to input the historical eye diagram image into the initial visual encoder to obtain the image feature vector; and to input the signal integrity measurement parameters and the expert diagnostic text into the initial text encoder to obtain the text feature vector; The joint adjustment module is used to jointly adjust the parameters of the initial visual encoder and the initial text encoder with the training objective of reducing the distance between the image feature vector and the text feature vector of the same group of historical eye map samples in the latent feature space, and increasing the distance between the image feature vector and the text feature vector of different groups of historical eye map samples in the latent feature space, until a preset training termination condition is met, thereby obtaining a pre-trained visual encoder.

10. An oscilloscope eye diagram diagnostic device, characterized in that, The device includes: The information acquisition module is used to acquire the eye diagram under test collected by the oscilloscope, and the signal integrity measurement parameters obtained by the oscilloscope in real time by measuring the eye diagram under test. The eye diagram conversion module is used to convert the eye diagram to be tested into a query feature vector using a pre-trained visual encoder obtained by the training method as described in any one of claims 1 to 2. The case retrieval module is used to retrieve similar historical cases from a pre-built vector database based on the query feature vector. The vector database stores multiple historical cases, each historical case using the historical image feature vector generated by the pre-trained visual encoder as the index key, and associated with the signal integrity measurement parameters and expert diagnostic text of the historical eye map. The diagnosis generation module is used to input the signal integrity measurement parameters and expert diagnostic text associated with the similar historical cases, together with the signal integrity measurement parameters of the eye diagram under test, as prompt text into a pre-trained large language model to generate a diagnostic result for the eye diagram under test.

11. An oscilloscope device, characterized in that, include: The signal acquisition unit is used to acquire the eye diagram under test and measure the eye diagram under test to obtain signal integrity measurement parameters; The memory stores a pre-trained visual encoder, a pre-trained large language model, and at least one instruction or at least one program. A processor for loading and executing the at least one instruction or the at least one program segment to implement the oscilloscope eye diagram diagnostic method as described in any one of claims 3 to 8.

Images