Error code prediction model construction method for dynamic characteristics of LVDS (Low Voltage Differential Signaling) link

Abstract

The invention discloses an error code prediction model construction method for LVDS link dynamic characteristics, and belongs to the field of chip test algorithms. Inputting the fault type experiment data with different characteristics into the dynamic model, and extracting and processing the data characteristics by the dynamic model; dividing the time sequence data into an input group and an output group by adopting a sliding window method; the bit error rate is placed in an interval of [0, 1], and the influence of numerical range difference on model training is eliminated; a moving average filter is adopted to carry out de-noising processing on error code noise of experimental data, and the weight of a real-time updating model is designed. In order to verify the actual effect of the dynamic error code prediction model and algorithm, various changing scenes are simulated, and prediction precision and adaptive capacity tests are carried out on the model. Experiments show that the prediction model improves the accuracy of error code prediction, and has obvious advantages compared with a static model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chip test algorithm, and particularly relates to a method for constructing a bit error prediction model of dynamic characteristics of an LVDS link. BACKGROUND

[0002] In an industrial monitoring system, the use of LVDS (Low-Voltage Differential Signaling) technology can meet the requirements for real-time transmission of high-definition video. However, in the face of complex environments such as high temperature and humidity, electromagnetic interference and other uncertain factors, various bit errors may occur in the transmission of video signals by the LVDS link, and the distribution of the bit error rate may be different due to dynamic changes in the external environment. Traditional bit error test systems usually only have static detection functions and cannot predict and optimize the bit error rate under dynamic working conditions. SUMMARY

[0003] The present application relates to the technical field of chip test algorithm, and particularly relates to a method for constructing a bit error prediction model of dynamic characteristics of an LVDS link.

[0004] To solve the above technical problems, the present application provides a method for constructing a bit error prediction model of dynamic characteristics of an LVDS link, comprising:

[0005] The experimental data of different characteristic fault types are input to the dynamic model, and the dynamic model extracts and processes the data characteristics;

[0006] The time series data is divided into two groups of input and output by using a sliding window method; the bit error rate is placed in the [0, 1] interval to eliminate the influence of the difference in the numerical range on the model training;

[0007] The bit error noise of the experimental data is denoised by using a moving average filter, and the weights of the real-time updating model are designed.

[0008] In one embodiment, the real-time updating of the model weights includes: a basic learning mechanism is established, a three-layer long short-term memory network is used as the prediction model, each layer contains a plurality of hidden units, a rectified linear unit activation function is used, and a dropout layer is added after each layer to prevent overfitting;

[0009] The mean square error is used as the loss function, and the adaptive moment estimation is selected as the optimizer; the model weights after offline training are saved for dynamic updating in actual operation;

[0010] The sliding window is introduced to ensure dynamic updating of the model training data, so that the latest bit error rate distribution can be reflected;

[0011] The weight dynamic adjustment adjusts the model weights to adapt to the latest data characteristics.

[0012] In an embodiment, the last layer weight of the long short-term memory network is adjusted through online learning.

[0013] In an embodiment, the construction method further comprises: when the bit error rate prediction result exceeds a set threshold, triggering a dynamic adjustment mechanism, including reducing the transmission rate or switching to a backup link.

[0014] The application provides a bit error prediction model construction method for LVDS link dynamic characteristics, in order to verify the actual effect of the dynamic bit error prediction model and algorithm, a variety of changing scenes are simulated to test the prediction accuracy and adaptability of the model. Experiments show that the prediction model constructed by the method improves the accuracy of bit error prediction, and has obvious advantages compared with static models. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a schematic diagram of the complete design and implementation steps of the prediction model.

[0016] Figure 2 is a schematic diagram of the mean square error curve of the bit error prediction of the static training model and the dynamic training model. DETAILED DESCRIPTION

[0017] The application provides a bit error prediction model construction method for LVDS link dynamic characteristics, in order to verify the actual effect of the dynamic bit error prediction model and algorithm, a variety of changing scenes are simulated to test the prediction accuracy and adaptability of the model. Experiments show that the prediction model constructed by the method improves the accuracy of bit error prediction, and has obvious advantages compared with static models.

[0018] The application provides a bit error prediction model construction method for LVDS link dynamic characteristics, based on dynamic training and online learning mechanism, the process is as shown in Figure 1 , including the following steps:

[0019] (1) Data feature extraction and preprocessing

[0020] The bit error rate time series is the core input of the dynamic bit error prediction model. For experimental data under different working conditions, the following preprocessing method is used to extract features and construct model input.

[0021] 1.1 Sliding window construction input data

[0022] In order to capture the historical characteristics of time series, sliding window method is used to divide time series into input and output pairs. The input of sliding window is the bit error rate of the past n time points, and the output is the bit error rate of the next time point:

[0023] X = [Et-n , E t-n+1 , ···, E t-1 ], Y = E t ( Formula 1 )

[0024] where E t is the error rate at the t-th time point, and n is the window size.

[0025] 1.2 Data Normalization

[0026] To eliminate the influence of the difference in the numerical range on model training, the error rate is normalized to the interval [0, 1]:

[0027]

[0028] where E min and E max are the minimum and maximum values of the error rate, respectively.

[0029] 1.3 Denoising Processing

[0030] The random error noise in the experimental data is smoothed by using a moving average filter, and the smoothed error rate is represented as follows.

[0031]

[0032] (2) Model Design and Optimization

[0033] The core of the dynamic training and online learning mechanism is to adapt to the dynamic changes of the error rate by updating the weights of the model in real time. The overall design is divided into the following three stages:

[0034] 2.1 Initial Model Training

[0035] Based on the experimental data, offline training is performed to construct a basic prediction model. The input is the historical error rate time series, and the output is the error rate prediction value at the next time point. A three-layer LSTM (Long Short-Term Memory) network is used as the prediction model, each layer contains 50 hidden units, uses the ReLU (Rectified Linear Unit) activation function, and adds a Dropout (Dropout) layer (proportion 0.2) after each layer to prevent overfitting. The mean square error (MSE) is used as the loss function, and the optimizer selects Adam (Adaptive Moment Estimation). The model weights after offline training will be saved for dynamic updating in actual operation.

[0036] 2.2 Sliding Window Management

[0037] A sliding window is used to dynamically maintain the latest real-time data, ensuring that the dataset for online learning always remains up-to-date. Error rate data is collected every fixed time interval (e.g., 0.5 seconds) and added to the sliding window for real-time updates. When the window data exceeds the maximum capacity, the oldest data is removed, ensuring that the data remains up-to-date. The introduction of the sliding window ensures dynamic updating of the model training data, enabling it to reflect the latest error rate distribution.

[0038] 2.3 Online Fine-tuning of Weights

[0039] Online learning adapts to the latest data characteristics by fine-tuning the model weights. To reduce computational burden, online learning only adjusts the weights of the last layer of the LSTM network. When the error rate prediction result exceeds a set threshold, a dynamic adjustment mechanism is triggered, including reducing the transmission rate or switching to a backup link.

[0040] The above description is only a description of the preferred embodiments of the present application, and is not any limitation on the scope of the present application. Any modification or modification made by a person skilled in the art according to the above disclosure is within the protection scope of the claims.

Claims

1. A method for constructing a bit error prediction model based on the dynamic characteristics of an LVDS link, characterized in that, include: Experimental data on fault types with different characteristics are input into the dynamic model, which then extracts and processes the data features. The sliding window method is used to divide the time series data into two groups: input and output. The bit error rate is placed in the interval [0,1] to eliminate the influence of numerical range differences on model training. A moving average filter was used to denoise the bit error noise in the experimental data, and the weights of the model were designed to be updated in real time.

2. The method for constructing a bit error prediction model based on the dynamic characteristics of an LVDS link as described in claim 1, characterized in that, The weights of the real-time updated model include: establishing a basic learning mechanism, using a three-layer long short-term memory network as the prediction model, with each layer containing several hidden units, using a modified linear unit activation function, and adding a dropout layer after each layer to prevent overfitting. Mean squared error is used as the loss function, and adaptive moment estimation is selected as the optimizer; the model weights after offline training are saved for dynamic updates during actual operation. Introducing a sliding window ensures dynamic updates to the model training data, enabling it to reflect the latest bit error rate distribution; Dynamic weight adjustment adapts to the latest data characteristics by fine-tuning the model weights.

3. The method for constructing a bit error prediction model based on the dynamic characteristics of an LVDS link as described in claim 2, characterized in that, The weights of the last layer of the Long Short-Term Memory network are adjusted through online learning.

4. The method for constructing a bit error prediction model based on the dynamic characteristics of an LVDS link as described in claim 1, characterized in that, The construction method further includes triggering a dynamic adjustment mechanism when the bit error rate prediction result exceeds a set threshold, including reducing the transmission rate or switching to a backup link.

Images