CN114285526A - 5G base station AMC solution based on SVM kernel transformation method - Google Patents

Abstract

The invention provides a 5G base station AMC solution based on an SVM kernel conversion method, SINR and CRC (A/N) carried in ULSCH.ind information received by a base station are mapped to corresponding weight factors, the two weight factors are accumulated and summed to obtain a composite weight factor, the composite weight factor is a sample point of an SVM, the base station receives more and more sample points along with the change of TTI, can train the sample points so as to obtain a decision plane, the decision plane is described by a linear equation, when burst occurs, an outlier can cause the movement of a hyperplane to reduce the Large Margin, at the moment, a relaxation factor is introduced to allow an individual outlier to violate a limiting condition, convert a low-dimensional inseparable problem into a high-dimensional scene separable, a common Gaussian kernel function is adopted for conversion, the SINR value and the CRC condition carried in the ULSCH.ind which should be received by the next TTI are predicted according to the decision plane, mapping to the MCS. The invention can effectively restrict the influence of the difference point caused by the change of the channel quality on the judgment of the MCS, and has higher adaptability.

Description

5G base station AMC solution based on SVM kernel transformation method

Technical Field

The invention relates to the technical field of 5G base station scheduling, in particular to a 5G base station AMC solution based on an SVM kernel transformation method.

Background

The traditional AMC function adopts the design principle of the interaction of the inner loop and the outer loop, and counts CRC feedback based on the design mode of BLER statistical window (actually counting a/N points). The statistics has the problems of untimely MCS adjustment, jumping of adjustment amplitude and poor adaptability when few sparse service sample points exist; meanwhile, the two statistical windows are independent, if more error codes occur near the junction between the statistical windows, but the NACK point positioned in each window is not enough to reduce the order, the MCS deviation is larger at the moment; if the window statistics are too long, the MCS change is too slow and inflexible.

Disclosure of Invention

The invention aims to provide a 5G base station AMC solution based on an SVM kernel transformation method.

In order to achieve the purpose, the invention is realized by the following technical scheme: the 5G base station AMC solution based on the SVM kernel transformation method comprises the following steps:

(1) mapping SINR and CRC (A/N) carried in ULSCH.ind information received by a base station to corresponding weight factors, accumulating and summing the two weight factors to obtain a composite weight factor, wherein the composite weight factor is used as a sample point of an SVM (support vector machine), and the base station receives more and more sample points along with the change of TTI (transmission time interval), so that the sample points can be trained, and a decision plane is obtained;

(2) the decision boundary requires Large Margin, and the decision plane is described by a linear equation: omega^Tx + b is 0, wherein omega is a normal vector and determines the direction of the hyperplane; b is displacement, determines the distance between the hyperplane and the origin, and the decision variables are supported and determined by the points on the boundary, and the boundary points are support vectors;

(3) in the case of a burst, the outliers cause the hyperplane to move, making the Large Margin smaller, the model is very sensitive to noise, and a relaxation factor is introduced, allowing individual outliers to violate the constraints, the relaxation factor being: y is_i(ω·x_i+ b) is greater than or equal to 1-epsilon, at which time the new objective function is

The limiting conditions are as follows: y is_i(ω^Tx_i+b)≥1-ε_i，ε_iNot less than 0, introduced epsilon_iThe non-negative parameter being a relaxation variable, selected to be hyperplaneWhen the surfaces are processed, selecting a hyperplane which enables the number of outliers to be minimum, wherein C is the weight of the outliers, the larger C is, the larger the influence of the outliers on the target function is, and performing a Lagrange multiplier method on the target function:

(4) aiming at the problem that data sample points cannot be linearly segmented, the problem of inseparability in a low dimension is converted into a separable scene in a high dimension;

(5) a decision surface is used for meeting the calculation requirement and is mapped to a high-dimensional space, the calculation result is calculated in a low-dimensional space, the mapping from the low dimension to the high dimension is taken as a kernel function, and a Gaussian kernel function Radial Basis Function (RBF) function is adopted for transformation;

(6) after a decision plane is obtained, predicting the SINR value and the CRC (A/N) condition carried in the ULSCH. ind which is received by the next TTI according to the decision plane, and if the data received by the next TTI belongs to an outlier relative to the decision plane, discarding the data;

(7) mapping to MCS according to the decision plane.

The invention provides a 5G base station AMC solution based on an SVM kernel transformation method. The method has the following beneficial effects:

(1) the decision plane curve obtained by adopting the SVM kernel transformation mode can effectively restrict the influence of the difference point caused by the channel quality change on the judgment of the MCS, and the condition that the adjustment amplitude of the MCS has jumping and the condition that the adjustment deviation of the MCS is large can not occur.

(2) The adaptability is high because the sample points are updated along with the TTI, and the decision plane is also updated accordingly.

Drawings

FIG. 1 is a schematic flow chart of the operation of the present invention;

FIG. 2 is a diagram of SINR segment mapping to weight factors and CRC mapping to weight factors according to the present invention;

fig. 3 is a schematic diagram illustrating a change of the complex weight factor with TTI after the 5G base station receives ulsch.ind according to the present invention;

FIG. 4 is a schematic diagram of the present invention illustrating the movement of the outliers in a hyperplane;

FIG. 5 is a diagram illustrating the linear inseparability of the data according to the present invention;

FIG. 6 is a high-dimensional space effect diagram of the present invention;

FIG. 7 is a diagram illustrating mapping of a decision surface to MCS in accordance with the present invention.

Detailed Description

The invention is further described with reference to the following drawings and detailed description:

as shown in fig. 1-7, the solution of 5G base station AMC based on SVM kernel transform method includes the following steps:

(1) mapping SINR and CRC (A/N) carried in ULSCH.ind information received by a base station to corresponding weight factors, accumulating and summing the two weight factors to obtain a composite weight factor, wherein the composite weight factor is used as a sample point of an SVM (support vector machine), and the base station receives more and more sample points along with the change of TTI (transmission time interval), so that the sample points can be trained, and a decision plane is obtained;

(2) it can be seen from fig. 3 that the decision boundary requires Large Margin, and then finding the point closest to the decision boundary finds the decision boundary, and the decision plane is described by a linear equation: omega^Tx + b is 0, wherein omega is a normal vector and determines the direction of the hyperplane; b is displacement, determines the distance between the hyperplane and the origin, and the decision variables are supported and determined by points on the boundary, which are support vectors, and only the support vectors influence the result;

(3) it can be seen from fig. 4 that in the case of a burst, the outliers cause the hyperplane shift, reducing the Large Margin, and the model is very sensitive to noise, and a relaxation factor is introduced to allow individual outliers to violate the constraint, where the relaxation factor is: y is_i(ω·x_i+ b) is greater than or equal to 1-epsilon, at which time the new objective function is

The limiting conditions are as follows: y is_i(ω^Tx_i+b)≥1-ε_i，ε_iNot less than 0, introduced epsilon_iThe nonnegative parameter is a relaxation variable, when selecting a hyperplane, the hyperplane which enables the number of outliers to be minimum is selected, C is the weight of the outliers, the larger C is, the larger the influence of the outliers on the target function is, and the Lagrange multiplier method is carried out on the target function:

(4) it can be seen from fig. 5 that the data sample points at this time cannot be linearly divided, and at this time, the approach shown in fig. 5 is adopted to convert the low-dimensional inseparable problem into a high-dimensional separable scene;

(5) it can be seen from fig. 6 that a decision surface can be found to meet the requirements at this time. In this case, only the high-dimensional space is mapped, but actually, the result to be calculated is calculated in the low-dimensional space. The mapping from low dimension to high dimension is a kernel function, and a common Gaussian kernel function (radial basis RBF function) is adopted for transformation;

(6) after a decision plane is obtained, predicting the SINR value and the CRC (A/N) condition carried in the ULSCH. ind which is received by the next TTI according to the decision plane, and if the data received by the next TTI belongs to an outlier relative to the decision plane, discarding the data;

(7) mapping to MCS according to the decision plane.

The decision plane curve obtained by adopting the SVM kernel transformation mode can effectively restrict the influence of the difference point caused by the channel quality change on the judgment of the MCS, the condition that the adjustment amplitude of the MCS jumps does not occur, the condition that the adjustment deviation of the MCS is large does not occur, and the decision plane is also updated along with the update of the TTI due to the fact that the sample point is updated along with the change of the TTI, so that the adaptability is high.

In practical application, the scheme is applied to software, a certain number of original sample points are considered to be needed for counting of the SVM, so that prior samples are very important, a set of samples with strong adaptability can be provided according to experience for training to obtain a decision plane curve, after the UE is accessed, along with the change of the TTI, the sample points can be updated accordingly, a new decision curve can adapt to the situation of the UE at the moment, the adjustment timeliness of the MCS is high, the adjustment amplitude is in smooth transition, the adjustment deviation is low, and fig. 1 is a software operation flow chart.

Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will understand that various changes, modifications and substitutions can be made without departing from the spirit and scope of the invention as defined by the appended claims. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (1)

1. The AMC solution of the 5G base station based on the SVM kernel transformation method is characterized by comprising the following steps:

(1) mapping SINR and CRC (A/N) carried in ULSCH.ind information received by a base station to corresponding weight factors, accumulating and summing the two weight factors to obtain a composite weight factor, wherein the composite weight factor is used as a sample point of an SVM (support vector machine), and the base station receives more and more sample points along with the change of TTI (transmission time interval), so that the sample points can be trained, and a decision plane is obtained;

(2) the decision boundary requires Large Margin, and the decision plane is described by a linear equation: omega^Tx + b is 0, wherein omega is a normal vector and determines the direction of the hyperplane; b is displacement, determines the distance between the hyperplane and the origin, and the decision variables are supported and determined by the points on the boundary, and the boundary points are support vectors;

(3) in the case of a burst, the outliers cause a hyperplane shift, which reduces the largemgin size, and the model is very sensitive to noise, and a relaxation factor is introduced, which allows individual outliers to violate constraints, the relaxation factor being: y is_i(ω·x_i+ b) is greater than or equal to 1-epsilon, at which time the new objective function is

The limiting conditions are as follows: y is_i(ω^Tx_i+b)≥1-ε_i，ε_iNot less than 0, introduced epsilon_iThe nonnegative parameter is a relaxation variable, when selecting a hyperplane, the hyperplane which enables the number of outliers to be minimum is selected, C is the weight of the outliers, the larger C is, the larger the influence of the outliers on the target function is, and the Lagrange multiplier method is carried out on the target function:

(4) aiming at the problem that data sample points cannot be linearly segmented, the problem of inseparability in a low dimension is converted into a separable scene in a high dimension;

(5) a decision surface is used for meeting the calculation requirement and is mapped to a high-dimensional space, the calculation result is calculated in a low-dimensional space, the mapping from the low dimension to the high dimension is taken as a kernel function, and a Gaussian kernel function Radial Basis Function (RBF) function is adopted for transformation;

(6) after a decision plane is obtained, predicting the SINR value and the CRC (A/N) condition carried in the ULSCH. ind which is received by the next TTI according to the decision plane, and if the data received by the next TTI belongs to an outlier relative to the decision plane, discarding the data;

(7) mapping to MCS according to the decision plane.

Images