CN116248202A - Method for realizing radio frequency channel calibration based on deep learning - Google Patents
Method for realizing radio frequency channel calibration based on deep learning Download PDFInfo
- Publication number
- CN116248202A CN116248202A CN202211722002.2A CN202211722002A CN116248202A CN 116248202 A CN116248202 A CN 116248202A CN 202211722002 A CN202211722002 A CN 202211722002A CN 116248202 A CN116248202 A CN 116248202A
- Authority
- CN
- China
- Prior art keywords
- layer
- data
- channel
- nodes
- radio frequency
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 24
- 238000013135 deep learning Methods 0.000 title claims abstract description 13
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000012360 testing method Methods 0.000 claims abstract description 12
- 238000013528 artificial neural network Methods 0.000 claims abstract description 10
- 238000010606 normalization Methods 0.000 claims abstract description 4
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000012545 processing Methods 0.000 claims abstract description 4
- 210000002569 neuron Anatomy 0.000 claims description 21
- 230000004913 activation Effects 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000004891 communication Methods 0.000 abstract description 2
- 238000012827 research and development Methods 0.000 abstract description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 239000002699 waste material Substances 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/10—Monitoring; Testing of transmitters
- H04B17/11—Monitoring; Testing of transmitters for calibration
- H04B17/14—Monitoring; Testing of transmitters for calibration of the whole transmission and reception path, e.g. self-test loop-back
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/391—Modelling the propagation channel
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Physics & Mathematics (AREA)
- Electromagnetism (AREA)
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Monitoring And Testing Of Transmission In General (AREA)
Abstract
The invention discloses a method for realizing radio frequency channel calibration based on deep learning, which belongs to the technical field of wireless communication equipment research and development, and comprises the steps of numbering data according to channels, and inputting data by taking each channel as a category; selecting 50 representative points for each channel; each piece of data has 5 dimensions of frequency, power and radio frequency switch state according to the calibration condition, and the value of each dimension of the data is readjusted on normalization processing to enable the value to be in a [0,1] interval; dividing the training set and the test set, and selecting 7:3, dividing the proportion, randomly selecting each channel, and finally integrating the channels into a training set and a testing set to finish data preprocessing; the network structure of the training model adopts a four-layer neural network structure, the first layer of input layer totally comprises 2 nodes, the second layer and the third layer are middle hidden layers which are respectively 16 nodes and 8 nodes, and the fourth layer is an output layer which comprises 3 nodes.
Description
Technical Field
The invention discloses a method for realizing radio frequency channel calibration based on deep learning, and belongs to the technical field of research and development of wireless communication equipment.
Background
Conventional calibration methods generally require calibration by external instruments. Setting the same frequency of the channel simulator and an external instrument, after the configuration is completed, recording instrument data of the external instrument and storing the instrument data as calibration data in the state of the channel, acquiring full-band calibration data by traversing frequency points of the radio frequency channel, and storing the calibration data into a database for loading when the channel simulator simulates and operates after the calibration is completed, wherein the traditional radio frequency channel calibration has the defects of resource waste, high time cost and lower accuracy.
Disclosure of Invention
The invention discloses a method for realizing radio frequency channel calibration based on deep learning, which solves the problems of resource waste, high time expenditure and lower accuracy of the traditional radio frequency channel calibration in the prior art.
A method for implementing radio frequency channel calibration based on deep learning, comprising:
s1, numbering data according to channels, and inputting data by taking each channel as a category during training;
s2, selecting characteristic values, and selecting 50 representative points from each channel according to the characteristics of the radio frequency device;
s3, normalizing, wherein each piece of data has 5 dimensions of frequency, power and radio frequency switch state according to the calibration condition, and readjusting the value of each dimension of the data on normalization processing to enable the value of each dimension of the data to be in a [0,1] interval;
s4, dividing the training set and the testing set, wherein the dividing ratio is 7:3, dividing the proportion, randomly selecting each channel, and finally integrating the channels into a training set and a testing set to finish data preprocessing;
s5, adopting a four-layer neural network structure on the network structure of the training model, wherein the first layer of input layer totally comprises 2 nodes, the second layer and the third layer are middle hidden layers which are respectively 16 nodes and 8 nodes, and the fourth layer is an output layer which comprises 3 nodes.
The calculation process of one neuron of the neural network is as follows:
representing the connection weight from the kth neuron in layer (l-1) of the network to the jth neuron in layer l,/o>Represents the deviation of the j-th neuron in the first layer,/->Representing the linear result of the j-th neuron in the first layer,/and>representing the activation function output of the jth neuron in the first layer, the activation function being represented by σ, the activation of the jth neuron in the first layer being:
based on the formula, the whole propagation process is to multiply the weight matrix W by the input layer matrix A and finally add a bias matrix B.
The invention has the beneficial effects that: the invention uses the trained calibration model, and only a small number of points need to be calibrated, so that the calibration value of the full frequency band of the channel can be calculated, and the time is saved. Fewer instruments are used for completing calibration work, so that instrument resources and manpower are saved; by learning the radio frequency characteristics, the requirement of higher power accuracy can still be met on the power accuracy of the non-calibration point; the use is more flexible, and the transplanting is more convenient.
Drawings
FIG. 1 is a neural network structure of a training model of the present invention.
Detailed Description
The following is a further description of embodiments of the invention, in conjunction with the specific examples:
a method for implementing radio frequency channel calibration based on deep learning, comprising:
s1, numbering data according to channels, and inputting data by taking each channel as a category during training;
for each channel, there is a slight difference in each rf channel due to process errors of the rf link devices, device uniformity errors, etc., but after the rf link is determined, the rf characteristics of the channel have been determined.
S2, selecting characteristic values, and selecting 50 representative points from each channel according to the characteristics of the radio frequency device;
s3, normalizing, wherein each piece of data has 5 dimensions of frequency, power and radio frequency switch state according to the calibration condition, and readjusting the value of each dimension of the data on normalization processing to enable the value of each dimension of the data to be in a [0,1] interval;
s4, dividing the training set and the testing set, wherein the dividing ratio is 7:3, dividing the proportion, randomly selecting each channel, and finally integrating the channels into a training set and a testing set to finish data preprocessing;
s5, as shown in FIG. 1, a four-layer neural network structure is adopted on the network structure of the training model, wherein the first layer of input layer totally comprises 2 nodes, the second layer and the third layer are middle hidden layers, respectively comprise 16 nodes and 8 nodes, and the fourth layer is an output layer and comprises 3 nodes.
The calculation process of one neuron of the neural network is as follows:
representing the connection weight from the kth neuron in layer (l-1) of the network to the jth neuron in layer l,/o>Represents the deviation of the j-th neuron in the first layer,/->Representing the linear result of the j-th neuron in the first layer,/and>representing the activation function output of the jth neuron in the first layer, the activation function being represented by σ, the activation of the jth neuron in the first layer being:
based on the formula, the whole propagation process is to multiply the weight matrix W by the input layer matrix A and finally add a bias matrix B;
the invention adopts BP neural network, which also includes reverse process, which is the process of correcting weight and deviation, so that the error function value is minimum, and the weight and deviation of the network are continuously adjusted according to gradient descent method, finally achieving the purpose of convergence. The BP neural network is a multi-layer feedforward network trained according to error back propagation, the basic idea is a gradient descent method, the gradient search technology is utilized to minimize the error mean square error of the actual output value and the expected output value of the network, and the BP algorithm comprises two processes of forward propagation of signals and back propagation of errors. That is, the calculation of the error output is performed in the direction from the input to the output, and the adjustment of the weight and the threshold value is performed in the direction from the output to the input. During forward propagation, an input signal passes through an implicit layer scope output node, and is subjected to nonlinear transformation to generate an output signal, and if the actual output does not accord with the expected output, the error direction propagation process is shifted.
The back transmission is to reverse the output error layer by layer to the input layer through the hidden layer, and to distribute the error to each layer unit, to use the error signal obtained from each layer as the basis for adjusting the weight of each unit, to reduce the error along the gradient direction by adjusting the connection strength of the input node and the hidden layer node and the link strength and threshold value of the hidden layer node and the output system , and to determine the network parameter corresponding to the minimum error, namely the weight and threshold value, through repeated learning and training, and to stop the training.
In the whole process, the effects of the selected different activation functions are tested respectively, and after training, different effects are achieved on the final convergence accuracy respectively, so that the following table comparison is obtained:
table 1 comparison of three activation functions
Activation function | Number of iterations | Accuracy rate of |
Sigmoid | 350 | 97.6% |
tanh | 230 | 86.5% |
ReLU | 190 | 91.7% |
On the last data comparison, the activation function Sigmoid is more suitable for this network model.
The data accuracy of the test set can reach 98% according to the generated calibration model, the use requirement of the data is met, and the index requirement of the instrument can be met in industrial personal computer software integrated to the channel simulator.
Through actual testing, the test results are shown in the table:
table 2 comparison of deep learning mode and conventional mode
Calibration mode | In a conventional manner | Deep learning mode |
Time overhead | 5h | 0.25h |
Calibration point power accuracy | ±0.5dB | ±0.5dB |
Non-calibration point power accuracy | ±1dB | ±0.5dB |
The above table shows that: the deep learning mode can be changed from original 1000 points to 50 points, so that the time of 20 times is saved in the calibration time, and the calibration time is saved from original 5 hours to 0.25 hour. In terms of calibration accuracy, the method is mainly divided into calibration point power accuracy and non-calibration point accuracy, and the method is more beneficial to the calibration of a channel simulator through a deep learning mode by means of actual test representation.
It should be understood that the above description is not intended to limit the invention to the particular embodiments disclosed, but to limit the invention to the particular embodiments disclosed, and that the invention is not limited to the particular embodiments disclosed, but is intended to cover modifications, adaptations, additions and alternatives falling within the spirit and scope of the invention.
Claims (2)
1. A method for implementing radio frequency channel calibration based on deep learning, comprising:
s1, numbering data according to channels, and inputting data by taking each channel as a category during training;
s2, selecting characteristic values, and selecting 50 representative points from each channel according to the characteristics of the radio frequency device;
s3, normalizing, wherein each piece of data has 5 dimensions of frequency, power and radio frequency switch state according to the calibration condition, and readjusting the value of each dimension of the data on normalization processing to enable the value of each dimension of the data to be in a [0,1] interval;
s4, dividing the training set and the testing set, wherein the dividing ratio is 7:3, dividing the proportion, randomly selecting each channel, and finally integrating the channels into a training set and a testing set to finish data preprocessing;
s5, adopting a four-layer neural network structure on the network structure of the training model, wherein the first layer of input layer totally comprises 2 nodes, the second layer and the third layer are middle hidden layers which are respectively 16 nodes and 8 nodes, and the fourth layer is an output layer which comprises 3 nodes.
2. The method for implementing radio frequency channel calibration based on deep learning as claimed in claim 1, wherein the calculation process of a neuron of the neural network is as follows:
representing the connection weight from the kth neuron in layer (l-1) of the network to the jth neuron in layer l,/o>Represents the deviation of the j-th neuron in the first layer,/->Representing the jth neuron in the first layerIs a linear result of->Representing the activation function output of the jth neuron in the first layer, the activation function being represented by σ, the activation of the jth neuron in the first layer being:
based on the formula, the whole propagation process is to multiply the weight matrix W by the input layer matrix A and finally add a bias matrix B.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211722002.2A CN116248202A (en) | 2022-12-30 | 2022-12-30 | Method for realizing radio frequency channel calibration based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211722002.2A CN116248202A (en) | 2022-12-30 | 2022-12-30 | Method for realizing radio frequency channel calibration based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116248202A true CN116248202A (en) | 2023-06-09 |
Family
ID=86635575
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211722002.2A Pending CN116248202A (en) | 2022-12-30 | 2022-12-30 | Method for realizing radio frequency channel calibration based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116248202A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117200906A (en) * | 2023-11-07 | 2023-12-08 | 成都嘉晨科技有限公司 | Radio frequency channel calibration method based on deep learning |
-
2022
- 2022-12-30 CN CN202211722002.2A patent/CN116248202A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117200906A (en) * | 2023-11-07 | 2023-12-08 | 成都嘉晨科技有限公司 | Radio frequency channel calibration method based on deep learning |
CN117200906B (en) * | 2023-11-07 | 2024-01-23 | 成都嘉晨科技有限公司 | Radio frequency channel calibration method based on deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11030528B1 (en) | Convolutional neural network pruning method based on feature map sparsification | |
WO2020156348A1 (en) | Structural damage identification method based on ensemble empirical mode decomposition and convolution neural network | |
CN110084221B (en) | Serialized human face key point detection method with relay supervision based on deep learning | |
CN112580588B (en) | Intelligent flutter signal identification method based on empirical mode decomposition | |
CN112464713A (en) | Communication radiation source radio frequency fingerprint identification method based on deep learning | |
CN116248202A (en) | Method for realizing radio frequency channel calibration based on deep learning | |
CN111912611B (en) | Method and device for predicting fault state based on improved neural network | |
CN108647807B (en) | River flow prediction method | |
CN114091766B (en) | CEEMDAN-LSTM-based space load prediction method | |
CN107919983B (en) | Space-based information network efficiency evaluation system and method based on data mining | |
CN104915515A (en) | BP neural network based GFET modeling method | |
US6317730B1 (en) | Method for optimizing a set of fuzzy rules using a computer | |
CN112596016A (en) | Transformer fault diagnosis method based on integration of multiple one-dimensional convolutional neural networks | |
CN113297994A (en) | Pilot behavior analysis method and system | |
CN110309904A (en) | A kind of neural network compression method | |
CN114584230A (en) | Predictive channel modeling method based on countermeasure network and long-short term memory network | |
CN114626769B (en) | Operation and maintenance method and system for capacitor voltage transformer | |
CN113919221A (en) | Fan load prediction and analysis method and device based on BP neural network and storage medium | |
CN110110406B (en) | Slope stability prediction method for achieving LS-SVM model based on Excel computing platform | |
CN113189513B (en) | Ripple-based redundant power supply current sharing state identification method | |
CN116911169A (en) | EDFA test parameter prediction method based on BP neural network | |
Li et al. | CTLE adaptation using deep learning in high-speed SerDes link | |
CN112749784A (en) | Computing device and neural network acceleration method | |
CN108538301B (en) | Intelligent digital musical instrument based on neural network audio technology | |
CN115900802A (en) | Universal dynamic compensation method for multi-type sensors |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |