A differential signal testing system and method for detecting differential line performance.

Abstract

This invention relates to the field of differential line performance testing, and more particularly to a differential signal testing system and method for testing differential line performance. The method includes: generating multi-timescale fused clamping features by passing data acquired from pressure and temperature sensors through a time-domain branch based on an LSTM-KAN architecture; acquiring frequency-domain data of the differential signal from the differential line tested by a vector network analyzer, and generating multi-band fused differential line performance features through a frequency-domain branch based on an LSTM-KAN architecture; and generating a comprehensive differential line quality grade by passing the multi-timescale fused clamping features and the multi-band fused differential line performance features through a regression head based on a cross-modal attention mechanism architecture. This invention accurately extracts the fluctuation characteristics of the differential line clamped by the test fixture and accurately extracts multi-band differential line defect characteristics, thus improving the testing accuracy of differential line performance.

Description

Technical Field

[0001] This invention relates to the field of differential line performance testing, and more particularly to a differential signal testing system and method for testing the performance of differential lines. Background Technology

[0002] With the rapid development of high-speed digital communication technology, differential lines have become the core carrier for high-frequency signal transmission scenarios such as USB 4.0 and high-speed Ethernet due to their advantages such as strong anti-interference capability, high timing consistency, and low electromagnetic radiation. The differential impedance matching degree, insertion loss, and crosstalk suppression capability of differential lines directly determine the integrity and reliability of signal transmission. Therefore, the performance testing accuracy of differential lines is crucial for applications such as R&D verification of electronic equipment and production line quality inspection.

[0003] Differential line performance testing primarily relies on a vector network analyzer (VNA). By measuring frequency domain parameters such as differential impedance, insertion loss (Sdd21), and return loss (Sdd11), the high-frequency transmission characteristics of the differential line are evaluated. However, during differential line performance testing, the wires need to be stripped to expose the positive signal line (P-line) and negative signal line (N-line) used as test terminals. The positive and negative signal lines cannot be directly connected to the VNA; they need to be connected indirectly and stably via test fixtures.

[0004] Therefore, the reliability of the connection between the differential line and the test equipment and the accuracy of the test results depend heavily on the clamping condition of the test fixture. Insufficient clamping force of the fixture on the differential line test end will lead to increased contact resistance and damage to impedance matching. Excessive clamping force may damage the wire or change the wire spacing and introduce additional losses. Clamping jitter will cause instantaneous fluctuations in the contact state, resulting in random noise in the frequency domain test data and masking the real performance defects.

[0005] Therefore, to compensate for the impact of the clamping state of the test fixture on the test results, some solutions integrate pressure or temperature sensors into the test fixture to monitor changes in the clamping force or ambient temperature on the differential lines. However, existing methods have two main drawbacks: firstly, they employ simple statistical analysis on the time-domain data such as pressure fluctuations and temperature drift collected by the sensors, making it difficult to capture the characteristics of transient anomalies across multiple time scales, resulting in insufficient quantification accuracy of dynamic interference features of the fixture; secondly, the analysis of the frequency-domain data output by the vector network analyzer focuses on a single frequency band or static parameters, failing to correlate it with the features collected by the time-domain sensors and thus unable to reveal their inherent mapping relationships, limiting the accuracy of the comprehensive performance evaluation of the differential lines.

[0006] Therefore, how to combine the clamping condition of the test fixture on the differential line test end with the test data and use a machine learning model to determine its inherent mapping relationship in order to improve the test accuracy of differential line performance is a technical problem that needs to be solved. Summary of the Invention

[0007] To address this, the present invention provides a differential signal testing system and method for detecting differential line performance. Through the time-domain branch of the LSTM-KAN fusion architecture, it accurately extracts clamping characteristics of pressure fluctuations and temperature drift of the clamping clamp on the differential line test end. Through the frequency-domain branch of the LSTM-KAN fusion architecture, it accurately extracts differential line defect characteristics across multiple frequency bands of the test data. Furthermore, it improves the accuracy of differential line performance testing by dynamically mining the intrinsic mapping relationship of features through a cross-modal attention mechanism regression head.

[0008] To achieve the above objectives, this invention proposes a differential signal testing method for detecting the performance of differential lines. The differential lines are connected to a vector network analyzer via a port connector of a test fixture. The test fixture includes a detachable clamp for clamping and fixing the test end of the differential line that contacts the port connector. The contact surface between the clamp and the differential line is equipped with a test end pressure sensor and a temperature sensor. The differential signal testing method includes:

[0009] The acquired data from the pressure and temperature sensors at the test end are processed through a time-domain branch based on the LSTM-KAN architecture to generate multi-time-scale fused clamping features.

[0010] The differential signal frequency domain data of the differential line tested by the vector network analyzer is obtained, and the differential signal frequency domain data is processed through a frequency domain branch based on the LSTM-KAN architecture to generate multi-band fused differential line performance characteristics.

[0011] The multi-timescale fusion clamping features and the multi-band fusion differential line performance features are used to generate a comprehensive differential line quality level through a regression head based on a cross-modal attention mechanism architecture.

[0012] Furthermore, the process of generating multi-timescale fused clamping features through temporal branching includes:

[0013] The positive and negative signal line pressure balance and signal line thermal expansion are calculated based on the acquired data, and a time series input vector is constructed based on the positive and negative signal line pressure balance, the signal line thermal expansion, the acquired data, and the rate of change of the acquired data.

[0014] The temporal input vector is passed through a unidirectional LSTM-KAN layer to generate short-term transient features;

[0015] The short-term transient features are passed through a bidirectional LSTM-KAN layer to generate trend relationship features;

[0016] The trend relationship features are passed through a multi-scale KAN layer to generate the multi-timescale fused clamping features.

[0017] Furthermore, the process of generating trend relationship features in the bidirectional LSTM-KAN layer includes:

[0018] The short-term transient features are passed through a forward propagation LSTM network to generate forward propagation candidate memories and forward propagation gating parameters. The forward propagation candidate memories are then passed through a first KAN network to generate enhanced candidate memories. Forward trend relationship features are calculated based on the enhanced candidate memories and the forward propagation gating parameters.

[0019] The short-term transient features are passed through a backpropagation LSTM network to generate backpropagation cell states and backpropagation gating parameters. The backpropagation cell states are then passed through a second KAN network to generate enhanced cell states. Backward trend relationship features are calculated based on the enhanced cell states and the backpropagation gating parameters.

[0020] The forward trend relationship features and the backward trend relationship features are fused to generate the trend relationship features;

[0021] The bidirectional LSTM-KAN layer includes a forward propagation LSTM network, a first KAN network, a backward propagation LSTM network, and a second KAN network.

[0022] Furthermore, the process of generating multi-timescale fused features by multi-scale KAN layers includes:

[0023] The trend relationship features are passed through a fine-scale KAN network to generate fine-scale features;

[0024] The trend relationship features are passed through a medium-scale KAN network to generate medium-fine-scale features;

[0025] The trend relationship features are passed through a coarse-scale KAN network to generate coarse-scale features;

[0026] The fine-scale features, the medium-fine-scale features, and the coarse-fine-scale features are fused to generate the multi-time-scale fused features;

[0027] The multi-scale KAN layer includes a fine-scale KAN network, a medium-scale KAN network, and a coarse-scale KAN network, where the number of B-spline basis functions and the number of B-spline bases decrease sequentially.

[0028] In particular, the unidirectional LSTM-KAN layer can accurately characterize the nonlinear mapping relationship of contact impedance fluctuation caused by pressure change and capture transient features. The bidirectional LSTM-KAN layer can capture the cumulative trend of pressure and temperature over time, as well as the lag effect of temperature drop on the previous pressure balance. The multi-scale KAN layer focuses on microsecond-level high-frequency fluctuations, second-level medium changes, and minute-level long-term trends. The resulting multi-timescale features can cover the entire time dimension from transient disturbances to environmental drift.

[0029] Furthermore, the process of generating multi-band fused differential line performance characteristics through frequency domain branching includes:

[0030] The differential signal frequency domain data is divided into frequency bands to generate frequency domain input data for multiple frequency bands;

[0031] The frequency domain input data of multiple frequency bands are respectively passed through the corresponding LSTM-KAN network to generate multi-frequency band fusion features;

[0032] Based on the differential signal frequency domain data, an anomaly index is generated through the KAN layer.

[0033] The fusion features of the frequency domain anomaly index and the multi-band fusion features are passed through the output convolutional layer to generate the multi-band fusion differential line performance features.

[0034] The frequency domain branch includes the LSTM-KAN network, the anomaly index KAN layer, and the output convolutional layer.

[0035] Furthermore, the process of generating frequency domain anomaly indicators in the KAN layer includes:

[0036] The integral of the differential signal frequency domain data within the test frequency band is calculated to generate the actual transmission power efficiency of the differential line. The ratio of the actual transmission power efficiency to the ideal transmission power efficiency of the differential line is then passed through the bandwidth anomaly index KAN network to generate bandwidth anomaly characteristics.

[0037] The maximum test difference between the differential signal frequency domain data and the return loss threshold is calculated, and the maximum test difference is passed through the return loss anomaly KAN network to generate return loss anomaly features.

[0038] The derivative of the phase angle with respect to frequency of the differential signal frequency domain data is calculated, and the variance of the derivative within the test frequency band is calculated to generate the differential line signal delay fluctuation. The differential line signal delay fluctuation is then passed through a delay anomaly KAN network to generate high-speed differential anomaly features.

[0039] The bandwidth anomaly features, return loss anomaly features, and high-speed differential anomaly features are fused to generate the frequency domain anomaly index.

[0040] The anomaly index KAN layer includes a return loss anomaly KAN network, a bandwidth anomaly index KAN network, and a delay anomaly KAN network, where the number of B-spline basis functions and the number of B-spline basis functions increase sequentially.

[0041] Furthermore, the process of generating the overall difference line quality grade from the regression head includes:

[0042] The multi-timescale fused clamped features are passed through a query convolutional mapping layer to generate a query vector;

[0043] The multi-band fusion differential line performance characteristics are respectively passed through a key convolutional mapping layer and a value convolutional mapping layer to generate key vectors and value vectors;

[0044] The query vector, key vector, and value vector are processed through a cross-modal attention mechanism to generate cross-modal fusion features;

[0045] The cross-modal fusion features are mapped to a KAN network to generate the quality level of the integrated differential line;

[0046] The regression head includes a query convolutional mapping layer, a key convolutional mapping layer, a value convolutional mapping layer, and a quality level mapping KAN network.

[0047] Furthermore, differential signal testing methods also include:

[0048] A cross-entropy loss term is constructed based on the comprehensive difference line quality level and the true quality level of the training samples.

[0049] A false positive loss term is constructed based on the parameter that the quality level of the comprehensive difference line is unqualified and the true quality level of the training samples;

[0050] A comprehensive loss function is constructed based on the cross-entropy loss term and the false positive loss term, and the time-domain branch, frequency-domain branch, and regression head are trained and optimized using the comprehensive loss function.

[0051] This invention also proposes a differential signal testing system for detecting the performance of differential lines. The differential signal testing system includes the vector network analyzer and the test fixture, the test fixture comprising:

[0052] The test base is provided with positioning grooves, which include a wire positioning groove for embedding the sheath layer of the differential line and an insulation positioning groove for embedding the test end of the differential line. The insulation positioning groove is fixed by a screw assembly.

[0053] Two port connectors, passing through two through holes on the side wall of the insulating positioning groove, respectively, make contact with the test end of the differential line.

[0054] Furthermore, the port connector includes a fixed flange, a first center pin, a second center pin, a connecting body, a tail tube, a positioning body, and a positioning insulator;

[0055] The fixed flange is fixedly connected to the test base, the first center pin is connected to the connecting body, the tail tube is connected to the fixed flange, the positioning body is connected to the tail tube, the second center pin is connected to the tail tube, the first center pin and the second center pin are kept inserted, and the positioning insulator is connected inside the positioning body.

[0056] The first center pin is connected to the vector network analyzer, and the second center pin is in contact with the test end of the differential line.

[0057] In particular, by extracting features from frequency domain data by frequency band using an independent LSTM-KAN network, the sensitivity of extracting core performance features of each frequency band is improved. The KAN layer of the anomaly index is fitted by a kernel function to improve the recognition rate of latent defects in the differential line.

[0058] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention achieves accurate extraction of clamping characteristics of pressure fluctuation and temperature drift of the fixed clamp of the differential line test end by the time domain branch of the LSTM-KAN fusion architecture, achieves accurate extraction of differential line defect characteristics of multiple frequency bands of test data by the frequency domain branch of the LSTM-KAN fusion architecture, and improves the test accuracy of differential line performance by dynamically mining the intrinsic mapping relationship of features through the regression head of the cross-modal attention mechanism.

[0059] In particular, this invention can accurately characterize the nonlinear mapping relationship of contact impedance fluctuation caused by pressure change through a unidirectional LSTM-KAN layer, capturing transient features. Through a bidirectional LSTM-KAN layer, it can capture the cumulative trend of pressure and temperature over time, as well as the lag effect of sudden temperature drop on the previous pressure balance. Through multi-scale KAN layers, it can focus on microsecond-level high-frequency fluctuations, second-level medium changes, and minute-level long-term trends. The resulting multi-timescale features can cover the entire time dimension from transient disturbances to environmental drift.

[0060] In particular, this invention improves the sensitivity of core performance feature extraction for each frequency band by extracting features from frequency domain data through an independent LSTM-KAN network, and improves the recognition rate of latent defects in differential lines by fitting the KAN layer of the anomaly index through a kernel function. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating a differential signal testing method for detecting differential line performance according to an embodiment of the present invention.

[0062] Figure 2This is a flowchart illustrating the time-domain branch of a differential signal testing method for detecting differential line performance according to an embodiment of the present invention.

[0063] Figure 3 This is a flowchart illustrating the frequency domain branch of a differential signal testing method for detecting differential line performance according to an embodiment of the present invention.

[0064] Figure 4 This is a schematic diagram of the overall structure of the test fixture for a differential signal testing method for detecting differential line performance according to an embodiment of the present invention;

[0065] Figure 5 This is a schematic diagram of the overall structure of the test base of the test fixture for a differential signal testing method for detecting differential line performance, according to an embodiment of the present invention.

[0066] Figure 6 This is an exploded view of the test base of the test fixture for a differential signal testing method for detecting differential line performance, according to an embodiment of the present invention.

[0067] Figure 7 This is a schematic cross-sectional view of the port connector assembly of the test fixture for a differential signal testing method for detecting differential line performance, according to an embodiment of the present invention.

[0068] Figure 8 This is an exploded view of the port connector of the test fixture for a differential signal testing method for detecting differential line performance, according to an embodiment of the present invention.

[0069] Figure 9 This is a cross-sectional view of the port connector of the test fixture for a differential signal testing method for detecting differential line performance, according to an embodiment of the present invention.

[0070] In the diagram: 1. Test base plate; 2. Test base; 21. Port connector; 211. Fixing flange; 212. First center pin; 213. Second center pin; 214. Connecting body; 215. Tail tube; 2151. Tail tube fixing screw; 2152. Tail tube fixing seat; 216. Positioning body; 2161. First beveled surface; 217. Positioning insulator; 2171. Second beveled surface; 218. First insulator; 219. Second insulator; 22. Fixing clamp; 221. Upper clamp body; 222 1. Lower clamping body; 223. Clamping groove; 2231. Temperature sensor setting area; 2232. Positive signal line pressure sensor setting area; 2233. Negative signal line pressure sensor setting area; 224. Screw assembly; 24. Positioning groove; 241. Line positioning groove; 242. Insulation positioning groove; 25. Mounting surface; 3. Guide assembly; 31. Guide seat; 32. Fixing plate; 33. Locking plate; 34. Fastening groove; 35. Second guide groove; 36. Set groove; 37. Fastening screw; 4. Differential line. Detailed Implementation

[0071] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0072] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0073] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate the direction or positional relationship, are based on the direction or positional relationship shown in the drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0074] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0075] like Figures 1 to 9 As shown, this invention provides a differential signal testing system and method for detecting differential line performance. Through the time-domain branch of the LSTM-KAN fusion architecture, it accurately extracts clamping characteristics of pressure fluctuations and temperature drift of the clamping clamp on the differential line test end. Through the frequency-domain branch of the LSTM-KAN fusion architecture, it accurately extracts differential line defect characteristics across multiple frequency bands of the test data. Furthermore, it improves the testing accuracy of differential line performance by dynamically mining the intrinsic mapping relationship of features through a cross-modal attention mechanism regression head.

[0076] like Figure 1 As shown, this embodiment proposes a differential signal testing method for detecting the performance of differential line 4. Differential line 4 is connected to a vector network analyzer via a port connector 21 of a test fixture. The test fixture includes a detachable clamp 22, which clamps and secures the test end of the differential line 4 in contact with the port connector 21. The contact surface between the clamp 22 and the differential line 4 is equipped with a test end pressure sensor and a temperature sensor. The differential signal testing method includes:

[0077] The acquired data from the pressure and temperature sensors at the test end are processed through a time-domain branch based on the LSTM-KAN architecture to generate multi-time-scale fused clamping features.

[0078] The differential signal frequency domain data of the differential line tested by the vector network analyzer is obtained, and the differential signal frequency domain data is processed through a frequency domain branch based on the LSTM-KAN architecture to generate multi-band fused differential line performance characteristics.

[0079] The multi-timescale fusion clamping features and the multi-band fusion differential line performance features are used to generate a comprehensive differential line quality level through a regression head based on a cross-modal attention mechanism architecture.

[0080] It is understandable that the clamping clip 22 needs to clamp the test end of the differential line 4 to ensure a stable connection between the test end and the port connector 21, thereby ensuring the stability and accuracy of the test signal of the differential line obtained by the vector network analyzer through the port connector 21. However, due to the detachable assembly of the clamping clip 22, the magnitude of its clamping force on the test end is uncertain. Insufficient force will lead to severe test signal reflection, abnormally high S11 parameter, and large test signal transmission loss; low S21 parameter; excessive force will lead to high anti-interference indicators such as near-end crosstalk (NEXT) and far-end crosstalk (FEXT) of the test signal, resulting in distorted test results. Therefore, it is necessary to combine clamping characteristics across multiple time scales to accurately determine the quality level of the differential line.

[0081] like Figure 2 As shown, the process of generating multi-timescale fused clamping features through temporal branching further includes:

[0082] The positive and negative signal line pressure balance and signal line thermal expansion are calculated based on the acquired data, and a time series input vector is constructed based on the positive and negative signal line pressure balance, the signal line thermal expansion, the acquired data, and the rate of change of the acquired data.

[0083] The temporal input vector is passed through a unidirectional LSTM-KAN layer to generate short-term transient features;

[0084] The short-term transient features are passed through a bidirectional LSTM-KAN layer to generate trend relationship features;

[0085] The trend relationship features are passed through a multi-scale KAN layer to generate the multi-timescale fused clamping features.

[0086] Specifically, the process of generating multi-timescale fused clamping features through temporal branching can be represented as:

[0087] ;

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] ;

[0093] In the formula, Represents a multi-dimensional temporal input vector, where This indicates the pressure balance between the positive and negative signal lines, where , These represent the pressure exerted on the positive and negative signal lines at the test end of the differential line after stripping at time t, respectively. This represents the thermal expansion of the signal line, where This represents the linear thermal expansion coefficient (CLTE), which is determined by the materials of the positive and negative signal lines in the differential lines. For example, the CLTE of pure copper is 16.5 × 10⁻⁶. -6 The brass is 18×10 -6 , This indicates a reference temperature, preferably 25 degrees Celsius. These represent the pressure exerted on the positive signal line, the pressure exerted on the negative signal line, and the contact temperature of the signal lines, respectively, which are used as data acquisition points. This indicates the rate of change of the collected data. These represent the short-term transient characteristics of a unidirectional LSTM-KAN layer, the cell state of a unidirectional LSTM-KAN layer, the forward trend relationship characteristics of a forward-propagating LSTM-KAN layer, and the cell state of a forward-propagating LSTM-KAN layer, respectively. These represent the backward trend relationship characteristics of backward propagation LSTM-KAN, the cell state of forward propagation LSTM-KAN, and the trend relationship characteristics, respectively. This indicates a unidirectional LSTM-KAN layer. This represents the short-term transient characteristics and unidirectional LSTM-KAN layer cell state at the previous time step. This represents a forward propagation LSTM-KAN network with bidirectional LSTM-KAN layers. These represent the forward trend characteristics of the previous time step and the forward propagation LSTM-KAN cell state, respectively. This represents a backpropagation LSTM-KAN network with bidirectional LSTM-KAN layers. These represent the backward trend characteristics at the next time step and the cell state of the backward propagation LSTM-KAN, respectively. These represent the fine-scale, medium-scale, and coarse-scale features of the multi-scale KAN layer, respectively. This indicates multi-timescale fusion clamping features. This indicates feature fusion, preferably feature splicing operation.

[0094] Preferably, the number of B-spline basis functions in the unidirectional LSTM-KAN layer is 64.

[0095] In particular, it captures the short-term temporal correlation of pressure and temperature signals and suppresses measurement noise through KAN's nonlinear adaptive capability.

[0096] Furthermore, the process of generating trend relationship features in the bidirectional LSTM-KAN layer includes:

[0097] The short-term transient features are passed through a forward propagation LSTM network to generate forward propagation candidate memories and forward propagation gating parameters. The forward propagation candidate memories are then passed through a first KAN network to generate enhanced candidate memories. Forward trend relationship features are calculated based on the enhanced candidate memories and the forward propagation gating parameters.

[0098] The short-term transient features are passed through a backpropagation LSTM network to generate backpropagation cell states and backpropagation gating parameters. The backpropagation cell states are then passed through a second KAN network to generate enhanced cell states. Backward trend relationship features are calculated based on the enhanced cell states and the backpropagation gating parameters.

[0099] The forward trend relationship features and the backward trend relationship features are fused to generate the trend relationship features;

[0100] The bidirectional LSTM-KAN layer includes a forward propagation LSTM network, a first KAN network, a backward propagation LSTM network, and a second KAN network.

[0101] Specifically, the process of generating forward trend relationship features using the forward propagation LSTM network and the first KAN network can be represented as follows:

[0102] ;

[0103] ;

[0104] ;

[0105] ;

[0106] ;

[0107] ;

[0108] ;

[0109] In the formula, , , , This represents the forget gate, input gate, output gate, and candidate cell state at the current time step t of the forward propagation gating parameters. This represents the Sigmoid function. Indicates short-term transient characteristics. This indicates feature fusion, preferably feature concatenation. This indicates the historical forward trend characteristics of the previous time step. , , , The convolutional learnable weight matrix corresponding to the gating parameters. , , , For the learnable bias term of the convolution corresponding to the gating parameters, This indicates forward propagation of candidate memories. This represents the enhanced candidate memory generated by the first KAN network. This represents the number of B-spline basis functions. Let represent the weights of the i-th learnable output layer, the activation function of the B-spline basis function, the weights of the i-th learnable input layer, the B-spline basis function, and the i-th learnable bias term, respectively. Let d represent the dimension of the hidden state (forward trend relationship feature). This indicates the historical candidate cell state at the previous time step. It represents the Hadamah accumulation. express function, This indicates a forward trend relationship.

[0110] Specifically, the process by which the backpropagation LSTM network and the second KAN network generate backward trend relationship features can be represented as follows:

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] ;

[0116] ;

[0117] ;

[0118] In the formula, This represents the forget gate, input gate, candidate memory, and output gate for the current time step t of backpropagation of the gating parameters. This represents the Sigmoid function. Indicates short-term transient characteristics. This indicates feature fusion, preferably feature concatenation. This indicates the predictive backward trend characteristics for the next time step. The convolutional learnable weight matrix corresponding to the gating parameters. For the learnable bias term of the convolution corresponding to the gating parameters, This represents the backward propagation cell state at the current time step. This indicates the enhanced cell state generated by the second KAN network. This represents the number of B-spline basis functions. Let represent the weights of the i-th learnable output layer, the activation function of the B-spline basis function, the weights of the i-th learnable input layer, the B-spline basis function, and the i-th learnable bias term, respectively. Let d represent the dimension of the hidden state (backward trend relationship feature). This indicates the predicted candidate cell state for the next time step. It represents the Hadamah accumulation. express function, This indicates a backward trend relationship.

[0119] Specifically, the architecture of the unidirectional LSTM-KAN layer is consistent with the architecture of the forward propagation LSTM network and the first KAN network.

[0120] Preferably, the number of B-spline basis functions in the bidirectional LSTM-KAN layer is 16, and the number of B-spline bases is 6.

[0121] In particular, the current state can be predicted based on historical information by forward propagation LSTM networks, capturing the long-term trend of pressure and temperature signals, and the current state can be interpreted based on future information by backward propagation LSTM networks, analyzing the subsequent effects of changes in pressure and temperature signals.

[0122] Furthermore, the process of generating multi-timescale fused features by multi-scale KAN layers includes:

[0123] The trend relationship features are passed through a fine-scale KAN network to generate fine-scale features;

[0124] The trend relationship features are passed through a medium-scale KAN network to generate medium-fine-scale features;

[0125] The trend relationship features are passed through a coarse-scale KAN network to generate coarse-scale features;

[0126] The fine-scale features, the medium-fine-scale features, and the coarse-fine-scale features are fused to generate the multi-time-scale fused features;

[0127] The multi-scale KAN layer includes a fine-scale KAN network, a medium-scale KAN network, and a coarse-scale KAN network, where the number of B-spline basis functions and the number of B-spline bases decrease sequentially.

[0128] Specifically, the number of B-spline basis functions in the fine-scale KAN network, the medium-scale KAN network, and the coarse-scale KAN network are 24, 20, and 16, respectively, and the number of B-spline bases are 8, 6, and 4, respectively.

[0129] like Figure 3 As shown, the process of generating multi-band fused differential line performance characteristics through frequency domain branching further includes:

[0130] The differential signal frequency domain data is divided into frequency bands to generate frequency domain input data for multiple frequency bands;

[0131] The frequency domain input data of multiple frequency bands are respectively passed through the corresponding LSTM-KAN network to generate multi-frequency band fusion features;

[0132] Based on the differential signal frequency domain data, an anomaly index is generated through the KAN layer.

[0133] The fusion features of the frequency domain anomaly index and the multi-band fusion features are passed through the output convolutional layer to generate the multi-band fusion differential line performance features.

[0134] The frequency domain branch includes the LSTM-KAN network, the anomaly index KAN layer, and the output convolutional layer.

[0135] Specifically, the standard Gaussian peak model of the input differential signal is used. and reflection coefficient model This serves as the frequency domain data of the differential signal.

[0136] Specifically, the differential signal frequency domain data is divided into 5 frequency band groups. The differential signal frequency domain data in each frequency band is passed through the corresponding LSTM-KAN network to generate frequency band evaluation features. The frequency band evaluation features of all frequency bands are spliced ​​together to generate multi-frequency band fused differential line performance features. The 5 frequency band groups are low frequency DC-1GHz, mid-low frequency 1-3GHz, mid frequency 3-6GHz, mid-high frequency 6-10GHz, and high frequency 10GHz+.

[0137] In particular, through LSTM-KAN networks with multiple frequency bands, features such as increased contact resistance, material aging, impedance mismatch, abnormal reflection, dielectric loss, conductor roughness, and structural tolerance defects can be extracted.

[0138] Furthermore, the process of generating frequency domain anomaly indicators in the KAN layer includes:

[0139] The integral of the differential signal frequency domain data within the test frequency band is calculated to generate the actual transmission power efficiency of the differential line. The ratio of the actual transmission power efficiency to the ideal transmission power efficiency of the differential line is then passed through the bandwidth anomaly index KAN network to generate bandwidth anomaly characteristics.

[0140] The maximum test difference between the differential signal frequency domain data and the return loss threshold is calculated, and the maximum test difference is passed through the return loss anomaly KAN network to generate return loss anomaly features.

[0141] The derivative of the phase angle with respect to frequency of the differential signal frequency domain data is calculated, and the variance of the derivative within the test frequency band is calculated to generate the differential line signal delay fluctuation. The differential line signal delay fluctuation is then passed through a delay anomaly KAN network to generate high-speed differential anomaly features.

[0142] The bandwidth anomaly features, return loss anomaly features, and high-speed differential anomaly features are fused to generate the frequency domain anomaly index.

[0143] The anomaly index KAN layer includes a return loss anomaly KAN network, a bandwidth anomaly index KAN network, and a delay anomaly KAN network, where the number of B-spline basis functions and the number of B-spline basis functions increase sequentially.

[0144] Specifically, the number of B-spline basis functions of the return loss anomaly KAN network, the bandwidth anomaly index KAN network, and the delay anomaly KAN network are 12, 16, and 18 respectively, and the number of B-spline bases are 3, 4, and 5 respectively.

[0145] Specifically, the return loss anomaly KAN network, the bandwidth anomaly index KAN network, and the delay anomaly KAN network can be represented as:

[0146] ;

[0147] ;

[0148] ;

[0149] In the formula, These represent the abnormal characteristics of return loss, bandwidth, and high-speed differential, respectively. This indicates that the differential signal frequency domain data and the return loss threshold are... The maximum test difference, This represents the integral of the differential signal frequency domain data within the test frequency band, i.e., the actual transmission power efficiency of the differential line. , These represent the maximum and minimum values ​​of the test frequency band, respectively. Indicates ideal transmission power efficiency. This represents the ideal transmission coefficient, which is set to 0.9 according to the industry standard for differential lines. The derivative of the phase angle with respect to frequency in the frequency domain of the differential signal is called the group delay of the differential line. This represents the variance of the derivative calculated within the test frequency band.

[0150] In particular, the bandwidth anomaly index KAN network can be used to test the matching degree between the transmission bandwidth of the differential line and the qualified bandwidth requirement. The return loss anomaly KAN network can be used to test whether the reflection impedance of the differential line is qualified. The delay anomaly KAN network can be used to test whether the waveform fidelity of the data transmission rate of the differential line is qualified.

[0151] Furthermore, the process of generating the overall difference line quality grade from the regression head includes:

[0152] The multi-timescale fused clamped features are passed through a query convolutional mapping layer to generate a query vector;

[0153] The multi-band fusion differential line performance characteristics are respectively passed through a key convolutional mapping layer and a value convolutional mapping layer to generate key vectors and value vectors;

[0154] The query vector, key vector, and value vector are processed through a cross-modal attention mechanism to generate cross-modal fusion features;

[0155] The cross-modal fusion features are mapped to a KAN network to generate the quality level of the integrated differential line;

[0156] The regression head includes a query convolutional mapping layer, a key convolutional mapping layer, a value convolutional mapping layer, and a quality level mapping KAN network.

[0157] Specifically, the process of generating the quality grade of the composite difference line from the regression head can be represented as:

[0158] ;

[0159] ;

[0160] ;

[0161] ;

[0162] ;

[0163] ;

[0164] In the formula, These represent the query vector, key vector, and value vector, respectively. These represent the multi-timescale fusion clamping features and the multi-band fusion differential line performance features, respectively. These represent the learnable convolutional weight matrices for the query convolutional mapping layer, the key convolutional mapping layer, and the value convolutional mapping layer, respectively. This represents the attention weights of the cross-modal attention mechanism. express function, The scaling factor represents the cross-modal attention mechanism. Indicates cross-modal fusion features, Indicates the quality level of the composite differential line. This represents the Sigmoid function. This indicates a quality level mapping to a KAN network.

[0165] Specifically, when the quality level of the composite differential line is less than 0.3, the differential line quality is qualified; when the quality level of the composite differential line is greater than or equal to 0.3 and less than 0.6, the differential line quality is good; when the quality level of the composite differential line is greater than or equal to 0.6 and less than 0.8, the differential line quality is poor; and when the quality level of the composite differential line is greater than or equal to 0.8, the differential line quality is extremely poor.

[0166] Furthermore, a cross-entropy loss term is constructed based on the comprehensive difference line quality level and the true quality level of the training samples;

[0167] A false positive loss term is constructed based on the parameter that the quality level of the comprehensive difference line is unqualified and the true quality level of the training samples;

[0168] A comprehensive loss function is constructed based on the cross-entropy loss term and the false positive loss term, and the time-domain branch, frequency-domain branch, and regression head are trained and optimized using the comprehensive loss function.

[0169] Specifically, the comprehensive loss function can be expressed as:

[0170] ;

[0171] ;

[0172] ;

[0173] In the formula, These represent the cross-entropy loss term, the reporting loss term, and the overall loss function, respectively. This represents the true quality level of the training samples. Its value is 1 when the difference line of the training samples is unqualified, and 0 when it is qualified. Indicates the quality level of the composite differential line. Indicates that, This represents the weighting coefficient, which is preferably 0.06.

[0174] like Figures 4 to 9 As shown, this embodiment also provides a differential signal testing system for a differential signal testing method applied to detect the performance of differential lines. The differential signal testing system includes the vector network analyzer and the test fixture, the test fixture including:

[0175] The test base 2 is provided with a positioning groove 24. The positioning groove 24 includes a wire positioning groove 241 for embedding the sheath layer of the differential line 4 and an insulation positioning groove 242 for embedding the test end of the differential line 4. The insulation positioning groove 242 is fitted with a fixing clip 22 by a screw assembly 224.

[0176] The two port connectors 21 pass through the two through holes on the side wall of the insulating positioning groove 242 and then contact the test end of the differential line 4.

[0177] like Figure 8 As shown, the port connector 21 further includes a fixed flange 211, a first center pin 212, a second center pin 213, a connecting body 214, a tail tube 215, a positioning body 216, and a positioning insulator 217.

[0178] The fixed flange 211 is fixedly connected to the test base 2, the first center pin 212 is connected to the connecting body 214, the tail tube 215 is connected to the fixed flange 211, the positioning body 216 is connected to the tail tube 215, the second center pin 213 is connected to the tail tube 215, the first center pin 212 and the second center pin 213 are kept in an insertion connection, and the positioning insulator 217 is connected inside the positioning body 216;

[0179] The first center pin 212 is connected to the vector network analyzer, and the second center pin 213 is in contact with the test end of the differential line.

[0180] like Figure 5 and 6As shown, specifically, both ends of the differential line to be tested can be connected to the test base 2. The ends of the differential line are connected to the port connector 21 via the fixing clip 22 on the test base 2. The port connector 21 is connected to the test equipment, thus enabling the differential line to connect to the test equipment. Connecting the differential line to the test equipment via the port connector 21 and fixing clip 22 inside the test base 2 improves signal stability and reduces signal insertion loss during testing. Simultaneously, the differential line is fixed in the test base 2 by the fixing clip 22, preventing the ends of the differential line from shaking during testing and causing inaccurate test data. Using the test base 2 for testing results makes the test results more accurate and the test signal more stable, resulting in more reliable test data and improved testing efficiency. The port connector 21 is disposed on the mounting surface 25 of the test base 2.

[0181] like Figure 8 and 9 As shown, specifically, the port connector 21 also includes a first insulator 218 and a second insulator 219. Specifically, the first insulator 218 is fixedly installed on the outer wall of the first center pin 212, and the second insulator 219 is fixedly connected to the outer wall of the second center pin 213. A first groove is fixedly opened on one end of the connecting body 214, and the first insulator 218 is embedded in the first groove. One end of the first center pin 212 passes through and extends to the end of the fixing flange 211 away from the tail tube 215, so as to maintain connection with the signal line through the fixing flange 211. A second groove is provided at the end of the positioning body 216, and the second insulator 219 is embedded in the second groove. One end of the second center pin 213 is inserted into the first center pin 212, and the other end of the second center pin 213 is inserted into the positioning insulator 217.

[0182] like Figure 7 As shown, specifically, the tailpipe 215 includes a tailpipe fixing screw 2151 and a tailpipe fixing seat 2152, and the clamping groove 223 has a temperature sensor setting area 22321, a positive signal line pressure sensor setting area 2232, and a negative signal line pressure sensor setting area 2233.

[0183] like Figure 8 and 9 As shown, specifically, the diameters of the first center pin 212 and the second center pin 213 are smaller than the diameters of the fixed flange 211, the connecting body 214, the positioning body 216, and the tail tube 215, so that the first center pin 212 and the second center pin 213 are only mounted inside the fixed flange 211, the connecting body 214, the positioning body 216, and the tail tube 215 through the first insulator 218 and the second insulator 219, which can ensure the stability of signal transmission.

[0184] like Figure 7 , 8 As shown in Figure 9, specifically, the second insulator 219 outside the second center pin 213 is embedded into the second groove of the positioning body 216. Then, one end of the positioning body 216 is inserted into the connecting body 214, so that the second center pin 213 and the first center pin 212 are connected. Finally, the positioning insulator 217 is inserted into the positioning body 216, so that the first beveled surface 2161 and the second beveled surface 2171 are aligned, thus completing the assembly.

[0185] In this embodiment, the time-domain branch of the LSTM-KAN fusion architecture accurately extracts the clamping characteristics of pressure fluctuations and temperature drift of the clamping clamp on the differential line test end of the test fixture. The frequency-domain branch of the LSTM-KAN fusion architecture accurately extracts the differential line defect characteristics of multi-frequency bands in the test data. The regression head of the cross-modal attention mechanism dynamically mines the intrinsic mapping relationship of the features, improving the test accuracy of differential line performance. The unidirectional LSTM-KAN layer can accurately characterize the nonlinear mapping relationship of contact impedance fluctuations caused by pressure changes and capture transient features. The bidirectional LSTM-KAN layer can capture the cumulative trend of pressure and temperature over time, as well as the lag effect of sudden temperature drops on the previous pressure balance. The multi-scale KAN layer focuses on microsecond-level high-frequency fluctuations, second-level medium changes, and minute-level long-term trends. The finally fused multi-time-scale features can cover the entire time dimension from transient interference to environmental drift. By extracting features from frequency domain data by frequency band using an independent LSTM-KAN network, the sensitivity of core performance feature extraction for each frequency band is improved. The KAN layer of the anomaly index is fitted with a kernel function to improve the recognition rate of latent defects in the differential line.

[0186] Those skilled in the art will recognize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0187] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0188] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A differential signal testing method for detecting the performance of differential lines, characterized in that, The differential line (4) is connected to the vector network analyzer via the port connector (21) of the test fixture. The test fixture includes a detachable clamp (22) for clamping and fixing the test end of the differential line (4) in contact with the port connector (21). The contact surface between the clamp (22) and the differential line (4) is provided with a test end pressure sensor and a temperature sensor. The differential signal testing method includes: The acquired data from the pressure and temperature sensors at the test end are processed through a time-domain branch based on the LSTM-KAN architecture to generate multi-time-scale fused clamping features. The differential signal frequency domain data of the differential line tested by the vector network analyzer is obtained, and the differential signal frequency domain data is processed through a frequency domain branch based on the LSTM-KAN architecture to generate multi-band fused differential line performance characteristics. The multi-timescale fusion clamping features and the multi-band fusion differential line performance features are used to generate a comprehensive differential line quality level through a regression head based on a cross-modal attention mechanism architecture. The process of generating multi-timescale fused clamped features through temporal branching includes: The positive and negative signal line pressure balance and signal line thermal expansion are calculated based on the acquired data, and a time series input vector is constructed based on the positive and negative signal line pressure balance, the signal line thermal expansion, the acquired data, and the rate of change of the acquired data. The temporal input vector is passed through a unidirectional LSTM-KAN layer to generate short-term transient features; The short-term transient features are passed through a bidirectional LSTM-KAN layer to generate trend relationship features; The trend relationship features are passed through a multi-scale KAN layer to generate the multi-timescale fused clamping features.

2. The differential signal testing method for detecting differential line performance according to claim 1, characterized in that, The process of generating trend relationship features in a bidirectional LSTM-KAN layer includes: The short-term transient features are passed through a forward propagation LSTM network to generate forward propagation candidate memories and forward propagation gating parameters. The forward propagation candidate memories are then passed through a first KAN network to generate enhanced candidate memories. Forward trend relationship features are calculated based on the enhanced candidate memories and the forward propagation gating parameters. The short-term transient features are passed through a backpropagation LSTM network to generate backpropagation cell states and backpropagation gating parameters. The backpropagation cell states are then passed through a second KAN network to generate enhanced cell states. Backward trend relationship features are calculated based on the enhanced cell states and the backpropagation gating parameters. The forward trend relationship features and the backward trend relationship features are fused to generate the trend relationship features; The bidirectional LSTM-KAN layer includes a forward propagation LSTM network, a first KAN network, a backward propagation LSTM network, and a second KAN network.

3. The differential signal testing method for detecting differential line performance according to claim 1, characterized in that, The process of generating multi-timescale fused clamped features using multi-scale KAN layers includes: The trend relationship features are passed through a fine-scale KAN network to generate fine-scale features; The trend relationship features are passed through a medium-scale KAN network to generate medium-fine-scale features; The trend relationship features are passed through a coarse-scale KAN network to generate coarse-scale features; The fine-scale features, the medium-fine-scale features, and the coarse-scale features are fused to generate the multi-time-scale fused clamping features; The multi-scale KAN layer includes a fine-scale KAN network, a medium-scale KAN network, and a coarse-scale KAN network, where the number of B-spline basis functions and the number of B-spline bases decrease sequentially.

4. The differential signal testing method for detecting differential line performance according to claim 1, characterized in that, The process of generating multi-band fused differential line performance characteristics through frequency domain branching includes: The differential signal frequency domain data is divided into frequency bands to generate frequency domain input data for multiple frequency bands; The frequency domain input data of multiple frequency bands are respectively passed through the corresponding LSTM-KAN network to generate multi-frequency band fusion features; Based on the differential signal frequency domain data, an anomaly index is generated through the KAN layer. The fusion features of the frequency domain anomaly index and the multi-band fusion features are passed through the output convolutional layer to generate the multi-band fusion differential line performance features. The frequency domain branch includes the LSTM-KAN network, the anomaly index KAN layer, and the output convolutional layer.

5. The differential signal testing method for detecting differential line performance according to claim 4, characterized in that, The process of generating frequency domain anomaly indicators using the KAN layer includes: The integral of the differential signal frequency domain data within the test frequency band is calculated to generate the actual transmission power efficiency of the differential line. The ratio of the actual transmission power efficiency to the ideal transmission power efficiency of the differential line is then passed through the bandwidth anomaly index KAN network to generate bandwidth anomaly characteristics. The maximum test difference between the differential signal frequency domain data and the return loss threshold is calculated, and the maximum test difference is passed through the return loss anomaly KAN network to generate return loss anomaly features. The derivative of the phase angle with respect to frequency of the differential signal frequency domain data is calculated, and the variance of the derivative within the test frequency band is calculated to generate the differential line signal delay fluctuation. The differential line signal delay fluctuation is then passed through a delay anomaly KAN network to generate high-speed differential anomaly features. The bandwidth anomaly features, return loss anomaly features, and high-speed differential anomaly features are fused to generate the frequency domain anomaly index. The anomaly index KAN layer includes a return loss anomaly KAN network, a bandwidth anomaly index KAN network, and a delay anomaly KAN network, where the number of B-spline basis functions and the number of B-spline basis functions both increase sequentially.

6. The differential signal testing method for detecting differential line performance according to claim 1, characterized in that, The process of generating the composite difference line quality grade from the regression head includes: The multi-timescale fused clamped features are passed through a query convolutional mapping layer to generate a query vector; The multi-band fusion differential line performance characteristics are respectively passed through a key convolutional mapping layer and a value convolutional mapping layer to generate key vectors and value vectors; The query vector, key vector, and value vector are processed through a cross-modal attention mechanism to generate cross-modal fusion features; The cross-modal fusion features are mapped to a KAN network to generate the quality level of the integrated differential line; The regression head includes a query convolutional mapping layer, a key convolutional mapping layer, a value convolutional mapping layer, and a quality level mapping KAN network.

7. The differential signal testing method for detecting differential line performance according to any one of claims 1 to 6, characterized in that, Also includes: A cross-entropy loss term is constructed based on the comprehensive difference line quality level and the true quality level of the training samples. A false positive loss term is constructed based on the parameter that the quality level of the comprehensive difference line is unqualified and the true quality level of the training samples; A comprehensive loss function is constructed based on the cross-entropy loss term and the false positive loss term, and the time-domain branch, frequency-domain branch, and regression head are trained and optimized using the comprehensive loss function.

8. A differential signal testing system applying the differential signal testing method for detecting differential line performance as described in any one of claims 1 to 7, characterized in that, The differential signal testing system includes the vector network analyzer and the test fixture, the test fixture comprising: The test base (2) is provided with a positioning groove (24), the positioning groove (24) includes a wire positioning groove (241) for embedding the sheath layer of the differential line (4) and an insulation positioning groove (242) for embedding the test end of the differential line (4). The insulation positioning groove (242) is fitted with a fixing clip (22) by a screw assembly (224). Two port connectors (21) pass through two through holes on the side wall of the insulating positioning groove (242) and then contact the test end of the differential line (4).

9. The differential signal testing system for detecting differential line performance according to claim 8, characterized in that, The port connector (21) includes a fixed flange (211), a first center pin (212), a second center pin (213), a connecting body (214), a tail tube (215), a positioning body (216), and a positioning insulator (217); The fixed flange (211) is fixedly connected to the test base (2), the first center pin (212) is connected to the connecting body (214), the tail tube (215) is connected to the fixed flange (211), the positioning body (216) is connected to the tail tube (215), the second center pin (213) is connected to the tail tube (215), the first center pin (212) and the second center pin (213) are kept in the insertion, and the positioning insulator (217) is connected inside the positioning body (216); The first center pin (212) is connected to the vector network analyzer, and the second center pin (213) is in contact with the test end of the differential line.

Images