Method for optimizing power accuracy of signal generating device
By using random forest regression and linear regression models to filter feature state registers and dynamically adjust the power of the signal generator, the problem of unstable output in the open-loop state is solved, and higher power accuracy and stability are achieved.
Patent Information
- Application Number
- CN202510814302.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-18
- Publication Date
- 2025-11-07
Smart Images
Figure CN120911640A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of power optimization, and particularly relates to a signal generating device power accuracy optimization method. BACKGROUND
[0002] In the automatic level control (ALC) loop of a radio frequency signal source, open loop control and closed loop control are two different working modes. In the closed loop state, it is a feedback control. The system will monitor the output signal in real time, compare it with the set value, and adjust the control element through the error. It has the advantages of high precision, strong anti-interference ability, automatic compensation of the influence of environmental changes and device aging, and good stability. In the open loop state, the output level is directly set, and the actual output is not monitored for adjustment. Because there is no feedback delay, the response speed is fast. However, the disadvantages are also obvious, such as being easily affected by external interference, temperature changes, device aging and other factors, which will cause unstable output and low precision.
[0003] In the actual power calibration process, there are two important processes, one is to find the characteristic value parameter representing the current signal source working state, and the other is to obtain the power control word through a certain algorithm according to the characteristic value parameter. For the existing power calibration method, the characteristic value parameters are mostly frequency values and power values, such as establishing a two-dimensional ALC open loop power calibration array to store the calibration data. When the ALC loop works in the open loop state, the calibration data in the above array is taken out, and a suitable algorithm is adopted to compensate the current frequency and power point in real time. Or record the detection voltage under open loop and closed loop, as a compensation basis to adjust the open loop power. Some also increase the temperature to represent the working state. But for the open loop power, only frequency, power and temperature, etc. may not be sufficient to represent the working state of the instrument in some specific conditions, such as sudden temperature change, low temperature or high temperature start, and instrument not stable, etc. Therefore, a more perfect way is needed to describe the working state of the instrument and adjust the power output. SUMMARY
[0004] In view of the above problems existing in the prior art, the application provides a signal generating device power accuracy optimization method, which is reasonable in design, solves the problems of the prior art, and has good effects.
[0005] A signal generating device power accuracy optimization method, comprising the following steps:
[0006] Step 1: obtaining a data set of state power, which is divided into a training set and a test set;
[0007] Step 2: constructing a random forest regression model to screen a characteristic state register;
[0008] Step 3: constructing a linear regression model according to the screened characteristic state register;
[0009] Step 4: The values of the feature state registers at different times are read by setting a timer, input into a linear regression model, and the compensated power value is output, so that the signal generating device is optimized.
[0010] Further, in step 1, first determine the reading method of all state registers in the signal generating device, assuming that the frequency of the signal generating device = 1GHz, the power = 0dBm, under various working states of the signal generating device, read the values of all state registers v i and the power value p j under the current working state, where i = 1,...,n, n is the number of state registers, j = 1,...,m, m is the number of working state categories, and the state power data set is obtained, wherein the relationship between the power value and the state register value is:
[0011]
[0012] Where P is the power matrix, A is the control coefficient matrix, and V is the state register matrix.
[0013] Further, in step 2, a plurality of sub-sample sets are extracted from the training set with replacement by self-sampling technology, each sub-sample set has the same size as the original training set, then different decision trees are trained using the sub-sample sets respectively, and finally the prediction results of the decision trees are averaged or weighted averaged to obtain the final regression prediction value;
[0014] By creating a random forest regression model, t feature state registers are selected, and the value of the feature state register is denoted as S k , k = 1,...,t.
[0015] Further, in step 3, the linear regression model is denoted as:
[0016]
[0017] S k and the corresponding power value are used to construct a linear regression model to obtain the influence coefficient of the feature state register on the power α k , k = 1,...,t, and finally the power value to be compensated power is calculated according to formula (2).
[0018] The beneficial technical effects brought by the present application are:
[0019] The present application adopts a random forest regression model and a linear regression model, compared with other general methods, the present application adopts a more customized description method for the open-loop state of the signal source, uses the selected feature registers to dynamically adjust the power, and can more closely approximate the expected state of the signal source. BRIEF DESCRIPTION OF DRAWINGS
[0020] Figure 1 Flow chart for model establishment in the present application;
[0021] Figure 2 Flow chart for power compensation using model timing in the present application; DETAILED DESCRIPTION
[0022] The specific embodiments of the present application are further described below in conjunction with specific examples:
[0023] A signal generator power accuracy optimization method, as shown in Figure 1 and Figure 2 , includes the following steps:
[0024] Step 1: Obtain the data set of state power, divided into training set and test set;
[0025] In step 1, first determine all state register reading methods in the signal generator, assuming the frequency of the signal generator = 1GHz, power = 0dBm, read the values of all state registers v i and the power value p j under the current working state, where i = 1,...,n, n is the number of state registers, j = 1,...,m, m is the number of working state categories, to obtain the data set of state power, where the relationship between the power value and the state register value is:
[0026]
[0027] where P is the power matrix, A is the control coefficient matrix, and V is the state register matrix.
[0028] Step 2: Construct a random forest regression model and select the feature state register;
[0029] Random forest regression model is a machine learning algorithm based on ensemble learning (Ensemble Learning), which has a wide range of applications in regression problems. Random forest regression model is an ensemble model composed of multiple decision trees. The basic idea is to use multiple decision trees to predict samples, and then combine the prediction results of these decision trees to get the final prediction value. Specifically, through bootstrap sampling technology, multiple sub-sample sets are extracted from the training set with replacement, each sub-sample set has the same size as the original training set, then different decision trees are trained using the sub-sample set, and finally the prediction results of the decision trees are averaged or weighted averaged to get the final regression prediction value;
[0030] By creating a random forest regression model, t feature status registers are screened, and the value of the feature status register is denoted as S k , k = 1,..., t.
[0031] Step 3: Construct a linear regression model according to the screened feature status registers;
[0032] Linear regression is a statistical learning method for analyzing the linear relationship between independent variables (features) and dependent variables (target variables), and is one of the most basic models in machine learning. Linear regression model is simpler and has fewer parameters than other training models. The linear regression model is denoted as:
[0033]
[0034] S k and the corresponding power value are used to construct a linear regression model, and the influence coefficient of the feature status register on the power is obtained k , k = 1,..., t, and finally the power value to be compensated is calculated according to formula (2).
[0035] Step 4: The values of the feature status registers at different times are read by setting a timer, input into the linear regression model, and the compensated power value is output, so as to optimize the signal generating device.
[0036] Of course, the above description is not a limitation of the present application, and the present application is not limited to the above examples. Changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present application should also be within the scope of the present application.
Claims
1. A method of optimizing power accuracy of a signal generating device, characterized by, Comprising the following steps: Step 1: obtaining a data set of state power, divided into a training set and a test set; Step 2: constructing a random forest regression model, and screening to obtain a feature state register; Step 3: constructing a linear regression model according to the screened feature state register; Step 4: reading the values of the feature state register at different times through a timer, inputting them into the linear regression model, and outputting the compensated power value, thereby optimizing the signal generating device.
2. The method of claim 1, wherein the signal generator is a power generator. In the step 1, first determine the signal generator all state register reading method, assuming the signal generator frequency = 1GHz, target power = 0dBm, in the signal generator under a variety of working conditions, read all state register value v i And the power value p j Under the current working state, where i = 1,...,n, n is the number of state registers, j = 1,...,m, m is the number of working state categories, get the state power data set, where the relationship between the power value and the state register value is: Wherein, P is a power matrix, A is a control coefficient matrix, and V is a state register matrix.
3. The method of claim 2, wherein the signal generator is a digital-to-analog converter. In step 2, multiple sub-sample sets are extracted from the training set with replacement through a bootstrap sampling technique, each sub-sample set has the same size as the original training set, then different decision trees are trained respectively using the sub-sample sets, and finally the prediction results of the decision trees are averaged or weighted averaged to obtain the final regression prediction value; By creating a random forest regression model, t feature state registers are screened, and the value of the feature state register is denoted as S k , k = 1,..., t.
4. The method of claim 3, wherein the signal generator is a digital-to-analog converter. In step 3, the linear regression model is denoted as: S k and the corresponding power value is used to construct a linear regression model to obtain the influence coefficient α of the feature state register on the power k ,k = 1,..., t, and finally the power value power that needs to be compensated is calculated according to formula (2).