CN109583575B - Processing method for improving instrument vector signal analysis performance based on deep learning - Google Patents

Processing method for improving instrument vector signal analysis performance based on deep learning Download PDF

Info

Publication number
CN109583575B
CN109583575B CN201811543273.5A CN201811543273A CN109583575B CN 109583575 B CN109583575 B CN 109583575B CN 201811543273 A CN201811543273 A CN 201811543273A CN 109583575 B CN109583575 B CN 109583575B
Authority
CN
China
Prior art keywords
layer
mlp
liquid crystal
display device
crystal display
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.)
Active
Application number
CN201811543273.5A
Other languages
Chinese (zh)
Other versions
CN109583575A (en
Inventor
蒋政波
刘景鑫
洪伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Shanghai TransCom Instruments Co Ltd
Original Assignee
Southeast University
Shanghai TransCom Instruments Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southeast University, Shanghai TransCom Instruments Co Ltd filed Critical Southeast University
Priority to CN201811543273.5A priority Critical patent/CN109583575B/en
Publication of CN109583575A publication Critical patent/CN109583575A/en
Application granted granted Critical
Publication of CN109583575B publication Critical patent/CN109583575B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B17/00Monitoring; Testing
    • H04B17/20Monitoring; Testing of receivers
    • H04B17/21Monitoring; Testing of receivers for calibration; for correcting measurements
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B17/00Monitoring; Testing
    • H04B17/20Monitoring; Testing of receivers
    • H04B17/29Performance testing
    • YGENERAL 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
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02DCLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00Reducing energy consumption in communication networks
    • Y02D30/70Reducing energy consumption in communication networks in wireless communication networks

Abstract

The invention relates to a processing method for improving instrument vector signal analysis performance based on deep learning, which comprises the steps of (1) carrying out ADC sampling on a distortion signal passing through a non-ideal analyzer radio frequency receiver to obtain a sequence y, introducing a feedforward neural network to compensate hardware distortion of y (2) analyzing the data sequence y according to the modulation type of the signal and the position of a corresponding constellation diagram, estimating symbol sequence data s to be transmitted by the signal, and estimating error epsilon according to a multi-layer perceptron neural network MLP with L layers; (3) Obtaining an MLP output layer through an activation function according to the multi-layer perceptron neural network MLP with the L layer; (4) And correcting the output result of the MLP output layer by y to obtain a final measurement symbol. (5) The weighting set of the MLP is obtained by training the data set and training using the back propagation algorithm BP. By adopting the method, the mathematical characteristics of the distortion of the receiver of the instrument are fitted, and the distortion is corrected, so that the characteristics of the original input signal can be reserved.

Description

Processing method for improving instrument vector signal analysis performance based on deep learning
Technical Field
The invention relates to the field of instruments and meters, in particular to the field of error correction of instruments and meters, and specifically relates to a processing method for improving analysis performance of instrument vector signals based on deep learning.
Background
The invention belongs to the technical field of instruments and meters. The method mainly relates to the relevant application fields of correcting the systematic errors of vector signal analysis instruments to improve the measurement accuracy. The method is particularly used for correcting and balancing the vector measurement result of the measuring instrument based on a deep learning method, so that the influence of the system error of the measuring instrument on the vector measurement performance is made up, and the accuracy of the vector signal analysis measurement result is improved.
Measuring instruments to which the present invention is applicable include, but are not limited to, vector signal analyzers (Vector Signal Analyzer), vector measurements including, but not limited to, error vector magnitude (Error Vector Magnitude, EVM).
Vector signal analysis is widely used in wireless communication device testing. Instruments supporting vector signal analysis functions typically include a radio frequency receiving module (e.g., amplifier, mixer, filter, etc.), an analog-to-digital converter (ADC), and a digital signal processing unit (e.g., DSP, FPGA, CPU, GPU). The most typical instrument with the vector signal analysis function is a vector signal analyzer, and communication instruments such as a terminal comprehensive tester and a base station comprehensive tester generally have the vector signal analysis function.
In general, a digital modulation signal is demodulated to obtain two components of a real part (denoted as I) and an imaginary part (denoted as Q), symbol information included in the signal is obtained through processing such as symbol synchronization, phase difference correction, frequency difference correction, and the like, and the symbols are plotted on an orthogonal coordinate diagram with the I component as the abscissa and the Q component as the ordinate, thereby obtaining a so-called constellation diagram. Fig. 1 and fig. 2 are constellation diagrams of a QPSK modulation mode, where each data point on the constellation diagram is concentrated near 4 reference points, as shown in fig. 1, under the conditions of better signal quality and smaller distortion; when the signal quality is poor and the distortion is serious, the data points may deviate from the reference points, and the positions are relatively scattered, for example, as shown in fig. 2. We can determine the quality of the signal by analyzing the vector distance between the measurement point and the reference point, usually using the error vector magnitude (Error Vector Magnitude, EVM) to represent the vector difference between the actual signal and the reference point. The EVM measures an important parameter of the quality of the modulated signal, and includes both the amplitude error and the phase error of the signal.
Distortion of the modulated signal typically results from imperfections in the radio frequency circuit itself, such as IQ imbalance, frequency response non-flatness within the signal passband, phase noise, etc. The vector signal analyzer is one of the most commonly used instruments for measuring the EVM of the signal, however, the radio frequency receiving circuit of the vector signal analyzer itself also has the problems described above, and brings additional distortion to the measured signal, so that the measurement result of the EVM is affected, and especially for the 5G and the ultra-large bandwidth signals of the future communication system, the effects are remarkable. In order to measure the accurate EVM index, the vector signal analyzer is very difficult to design and optimize the rf receiving hardware, so that adding a compensation algorithm for the distortion of the rf hardware in the process of digital signal processing becomes a necessary task.
In order to compensate the signal distortion condition as shown in fig. 2, it is common practice to use a comb spectrum generator to transmit a multitone signal with a plurality of sine waves superimposed to a signal receiver, the receiver receives and collects the multitone signal and compares the multitone signal with the original signal to obtain power and phase variation at each sine wave frequency point, the radio frequency circuit of the receiver is modeled by means of the data to obtain frequency response and phase variation in the passband of the radio frequency circuit, and then the analyzer receiving circuit is equalized and compensated. However, this method has the disadvantage that in order to obtain the frequency response variation everywhere in the passband of the receiver, the power and phase condition of each spectral line of the comb spectrum need to be known accurately, which is generally difficult to obtain, and is generally used in measurement calibration with particularly high requirements on accuracy and traceability, and has high complexity and high calibration cost.
Because of the complexity and computational difficulties of current distortion compensation correction methods, there is a need to find a simpler method. The Neural Network (NN) is a mathematical model built by imitating a biological nervous system connection structure, is a very popular mathematical tool which is widely applied at present, can solve a plurality of complex problems, and has very outstanding information processing capability, in particular capability of processing inaccurate information and fuzzy information. The behavior of a neural network is determined by a series of parameters, which are typically obtained through training and learning. In recent years, with the development of deep learning technology, neural networks have begun to play an increasing role in various fields. The invention adopts a neural network to correct vector measurement errors of the instrument based on an algorithm architecture of deep learning.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a processing method which is used for avoiding excessive correction, keeping the original characteristics unchanged, and improving the analysis performance of the instrument vector signal based on deep learning, and is convenient and practical.
In order to achieve the above object, the processing method for improving the instrument vector signal analysis performance based on deep learning according to the present invention comprises the following steps:
the processing method for improving the instrument vector signal analysis performance based on deep learning is mainly characterized by comprising the following steps of:
(1) Performing ADC (analog-to-digital converter) sampling on a distortion signal passing through a non-ideal analyzer radio frequency receiver to obtain a sequence y, and introducing a feedforward neural network to compensate hardware distortion of the sequence y;
(2) Analyzing the data sequence y according to the modulation type of the signal and the position of the corresponding constellation diagram, estimating symbol sequence data s to be transmitted by the signal, and estimating errors according to a multi-layer perceptron neural network MLP with L layers
Figure SMS_1
(3) Obtaining an MLP output layer through an activation function according to the multi-layer perceptron neural network MLP with the L layer;
(4) Correcting y to obtain a final measurement symbol according to the output result of the MLP output layer;
(5) The weighting set of the MLP is obtained by training the data set and training using the back propagation algorithm BP.
Preferably, the output layer of the first layer in the obtained MLP in the step (2) is specifically:
the output layer of the first layer in the MLP is derived according to the following formula:
Figure SMS_2
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_3
for the output vector of each layer, +.>
Figure SMS_4
For all parameters within the MLPSet of->
Figure SMS_5
Preferably, the second layer to the L-1 layer in the obtained MLP in the step (2) are hidden layers of the network.
Preferably, the step (2) of obtaining the second layer to the L-1 layer in the MLP specifically includes:
the second layer to L-1 layer in the MLP is derived according to the following equation:
Figure SMS_6
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_7
for the output vector of each layer, +.>
Figure SMS_8
For the set of all parameters within the MLP, +.>
Figure SMS_9
Preferably, the activation function in the step (3) is
Figure SMS_10
Preferably, the step (3) of obtaining the MLP output layer specifically includes:
the MLP output layer is derived according to the following formula:
Figure SMS_11
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_12
for the output vector of each layer, +.>
Figure SMS_13
For the set of all parameters within the MLP, +.>
Figure SMS_14
Is of each layerWeights of the inter-linear connection.
Preferably, the said
Figure SMS_15
Providing a weight of +.>
Figure SMS_16
Is a linear transformation of (a).
Preferably, the final measurement symbol obtained in the step (4) is specifically:
the final measurement symbol is obtained according to the following formula:
Figure SMS_17
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_18
,/>
Figure SMS_19
,/>
Figure SMS_20
preferably, the training of the weight set of the MLP by the M training data set in the step (5) is specifically:
training the weight set of the MLP by the M training data set according to the following formula:
Figure SMS_21
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_22
and->
Figure SMS_23
Representing the real part and the imaginary part of the mth distortion-containing data, respectively, +.>
Figure SMS_24
And->
Figure SMS_25
Representing the real part and the imaginary part of the mth ideal transmission data, respectively,/->
Figure SMS_26
And->
Figure SMS_27
Representing the real and imaginary parts of the error of the mth data, respectively, M representing the size of the training data set.
By adopting the processing method for improving the analysis performance of the instrument vector signal based on the deep learning, the invention realizes the correction of the distortion characteristic of the instrument. By means of the good fitting capacity of the neural network, mathematical characteristics of the distortion of the receiver of the instrument are fitted, and the distortion is corrected. Excessive correction of the signal is avoided. Only the distortion caused by the instrument receiver is corrected, and the characteristics of the original input signal can be reserved.
Drawings
Fig. 1 is a QPSK constellation diagram with better signal quality and dense data point distribution according to the processing method for improving the instrument vector signal analysis performance based on deep learning.
Fig. 2 is a QPSK constellation diagram with poor signal quality and scattered data points according to the processing method for improving the vector signal analysis performance of the apparatus based on deep learning of the present invention.
Fig. 3 is a schematic diagram of a correction method of a vector signal analyzer based on a processing method for improving the analysis performance of an instrument vector signal based on deep learning.
Fig. 4 is a schematic diagram of a neural network constructed by the processing method for improving the instrument vector signal analysis performance based on deep learning.
Fig. 5 is a schematic diagram of a convergence process of neural network training based on the processing method for improving the instrument vector signal analysis performance based on deep learning.
Fig. 6 is a real machine test result of a constellation diagram without neural network correction based on the processing method for improving the instrument vector signal analysis performance based on deep learning.
Fig. 7 is a real machine test result of a constellation diagram corrected by a neural network based on the processing method for improving the instrument vector signal analysis performance by deep learning.
Detailed Description
In order to more clearly describe the technical contents of the present invention, a further description will be made below in connection with specific embodiments.
The processing method for improving the instrument vector signal analysis performance based on deep learning comprises the following steps:
(1) Performing ADC (analog-to-digital converter) sampling on a distortion signal passing through a non-ideal analyzer radio frequency receiver to obtain a sequence y, and introducing a feedforward neural network to compensate hardware distortion of the sequence y;
(2) Analyzing the data sequence y according to the modulation type of the signal and the position of the corresponding constellation diagram, estimating symbol sequence data s to be transmitted by the signal, and estimating errors according to a multi-layer perceptron neural network MLP with L layers
Figure SMS_28
(3) Obtaining an MLP output layer through an activation function according to the multi-layer perceptron neural network MLP with the L layer;
(4) Correcting y to obtain a final measurement symbol according to the output result of the MLP output layer;
(5) The weight set of the MLP is obtained by training the data set using the back propagation algorithm BP.
As a preferred embodiment of the present invention, the output layer of the first layer in the derived MLP in the step (2) is specifically:
the output layer of the first layer in the MLP is derived according to the following formula:
Figure SMS_29
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_30
for the output vector of each layer, +.>
Figure SMS_31
For the set of all parameters within the MLP, +.>
Figure SMS_32
As a preferred embodiment of the present invention, the second layer to the L-1 layer in the derived MLP in the step (2) are hidden layers of the network.
As a preferred embodiment of the present invention, the second layer to the L-1 layer in the derived MLP in the step (2) is specifically:
the second layer to L-1 layer in the MLP is derived according to the following equation:
Figure SMS_33
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_34
for the output vector of each layer, +.>
Figure SMS_35
For the set of all parameters within the MLP, +.>
Figure SMS_36
As a preferred embodiment of the present invention, the activation function in the step (3) is
Figure SMS_37
As a preferred embodiment of the present invention, the obtaining the MLP output layer in the step (3) specifically includes:
the MLP output layer is derived according to the following formula:
Figure SMS_38
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_39
output for each layerVector (S)>
Figure SMS_40
For the set of all parameters within the MLP, +.>
Figure SMS_41
Is the weight of the linear connection between the layers.
As a preferred embodiment of the present invention, the
Figure SMS_42
Providing weights as
Figure SMS_43
Is a linear transformation of (a).
As a preferred embodiment of the present invention, the final measurement symbol obtained in the step (4) is specifically:
the final measurement symbol is obtained according to the following formula:
Figure SMS_44
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_45
,/>
Figure SMS_46
,/>
Figure SMS_47
as a preferred embodiment of the present invention, the training of the weight set of the MLP by the M training data set in the step (5) specifically includes:
training the weight set of the MLP by the M training data set according to the following formula:
Figure SMS_48
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_49
and->
Figure SMS_50
Representing the real part and the imaginary part of the mth distortion-containing data, respectively, +.>
Figure SMS_51
And->
Figure SMS_52
Representing the real part and the imaginary part of the mth ideal transmission data, respectively,/->
Figure SMS_53
And->
Figure SMS_54
Representing the real and imaginary parts of the error of the mth data, respectively, M representing the size of the training data set.
In the specific embodiment of the invention, the neural network is used for correcting the vector error of the measuring instrument, and the method is convenient and fast to realize. Meanwhile, the signal is not excessively corrected, the original characteristics of the input signal are kept unchanged, and the purpose of measurement is achieved.
The invention provides a method for obtaining instrument vector measurement errors by using a neural network and eliminating the errors based on deep learning. Taking a signal analyzer as an example, constructing a neural network, during training, fitting a modulating signal input to the signal analyzer to the distortion of a receiving circuit of the signal analyzer, inputting current and previous continuous time data to the neural network, training the neural network to fit the mathematical characteristics of the signal distortion of a radio frequency circuit, and outputting an error vector of the time. When in actual use, the error vector of the instrument is obtained to the neural network according to the input current time signal, and then the error vector is subtracted from the original signal to restore to obtain the original input signal.
Neural network-based vector signal analysis correction scheme as shown in fig. 3, signals originating from a device under test (Device under test, DUT)
Figure SMS_55
Through irrational treatmentThe wanted analyzer radio frequency receiving circuit is distorted, then the sequence y is obtained after ADC acquisition, and the feedforward neural network is introduced to compensate the hardware distortion of y.
Based on the modulation type of the signal and the position of the corresponding constellation, the data sequence y is first parsed to estimate symbol data s to be transmitted by the signal, as shown in fig. 4, the symbol sequence s is provided to a multi-layer perceptron (MLP) neural network having L layers to estimate errors
Figure SMS_58
. We use +.>
Figure SMS_59
To represent the mapping relation between input and output, and input real number vector
Figure SMS_63
Then L layers of operations are carried out to obtain real number vector output +.>
Figure SMS_57
Wherein->
Figure SMS_60
Comprising N 0 Real and imaginary parts of the complex values, +.>
Figure SMS_61
Comprising a real part and an imaginary part of a complex value. We use +.>
Figure SMS_62
To represent a set of all parameters within the MLP, and
Figure SMS_56
to represent the output vector of each layer, the output layer of the first layer:
Figure SMS_64
wherein the method comprises the steps of
Figure SMS_65
. This model needs to take into accountN before the current data 0 -1 symbol. The second layer to L-1 layer are hidden layers of the network, their output definition:
Figure SMS_66
wherein the method comprises the steps of
Figure SMS_67
,/>
Figure SMS_68
Representing the mapping from layer L-1 to layer L. The form of this map is as follows:
Figure SMS_69
wherein the method comprises the steps of
Figure SMS_70
Is the weight of the linear connection between the layers, +.>
Figure SMS_71
Is an additional bias of the respective connection node, +.>
Figure SMS_72
It is the activation function of the individual nodes that introduces a non-linear transformation. In this model, bias is set>
Figure SMS_73
Selecting an activation function as a sigmoid function +.>
Figure SMS_74
The L-th layer, i.e., the MLP output layer, is denoted as:
Figure SMS_75
wherein the method comprises the steps of
Figure SMS_76
Providing a weight of +.>
Figure SMS_77
Finally we use the result of the MLP output to correct y and then get the final measured sign
Figure SMS_78
Wherein the method comprises the steps of
Figure SMS_79
,/>
Figure SMS_80
,/>
Figure SMS_81
Note that each of the magnitudes described above contains real and imaginary parts of complex valued symbols, with the two component instruments representing the positions of the constellation points.
Weighting set of MLP
Figure SMS_82
Is updated continuously during the training process by M training data sets, wherein the training process is to make the loss function L (W) continuously converged
Figure SMS_83
The back propagation algorithm (BP) is used here, and the update of the weight is calculated based on gradient descent, which is an algorithm commonly used in deep learning.
The method is used for correcting the data of the measuring instrument, firstly, the data with smaller distortion is input to the measuring instrument for training, and as the training frequency increases, the EVM value gradually decreases, the signal distortion is smaller and smaller, and the output of the neural network gradually converges. After training, the neural network is checked for correction, and compared with the uncorrected result, fig. 6 shows the uncorrected constellation, fig. 7 shows the corrected constellation, and it can be seen that the distortion is removed and the signal quality is improved.
By adopting the processing method for improving the analysis performance of the instrument vector signal based on the deep learning, the invention realizes the correction of the distortion characteristic of the instrument. By means of the good fitting capacity of the neural network, mathematical characteristics of the distortion of the receiver of the instrument are fitted, and the distortion is corrected. Excessive correction of the signal is avoided. Only the distortion caused by the instrument receiver is corrected, and the characteristics of the original input signal can be reserved.
In this specification, the invention has been described with reference to specific embodiments thereof. It will be apparent, however, that various modifications and changes may be made without departing from the spirit and scope of the invention. The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense.

Claims (3)

1. The processing method for improving the analysis performance of the instrument vector signal based on the deep learning is characterized by comprising the following steps of:
(1) Performing ADC (analog-to-digital converter) sampling on a distortion signal passing through a non-ideal analyzer radio frequency receiver to obtain a sequence y, and introducing a feedforward neural network to compensate hardware distortion of the sequence y;
(2) Analyzing the data sequence y according to the modulation type of the signal and the position of the corresponding constellation diagram, estimating symbol sequence data s to be transmitted by the signal, and estimating errors according to a multi-layer perceptron neural network MLP with L layers
Figure QLYQS_1
(3) Obtaining an MLP output layer through an activation function according to the multi-layer perceptron neural network MLP with the L layer;
(4) Correcting y to obtain a final measurement symbol according to the output result of the MLP output layer;
(5) Training a weight set of the MLP through a training data set and using a back propagation algorithm BP;
Figure QLYQS_2
for the output vector of each layer, +.>
Figure QLYQS_3
Is a set of all parameters within the MLP;
the output layer of the first layer in the obtained MLP in the step (2) specifically is:
the output layer of the first layer in the MLP is derived according to the following formula:
Figure QLYQS_4
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure QLYQS_5
,/>
Figure QLYQS_6
inputting the number of symbols of data for an input layer;
the second layer to the L-1 layer in the MLP obtained in the step (2) are hidden layers of the network;
the second layer to the L-1 layer in the MLP obtained in the step (2) are specifically:
the second layer to L-1 layer in the MLP is derived according to the following equation:
Figure QLYQS_7
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure QLYQS_8
the activation function in step (3) is
Figure QLYQS_9
Where u is the argument of the activation function σ (u);
the MLP output layer obtained in the step (3) is specifically:
the MLP output layer is derived according to the following formula:
Figure QLYQS_10
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure QLYQS_11
is the weight of the linear connection between the layers, +.>
Figure QLYQS_12
The input number of the first layer is the number of network nodes of the first layer;
the said process
Figure QLYQS_13
Providing a weight of +.>
Figure QLYQS_14
Is a linear transformation of (a).
2. The method for improving the analysis performance of the instrument vector signal based on the deep learning according to claim 1, wherein the final measurement symbol obtained in the step (4) is specifically:
the final measurement symbol is obtained according to the following formula:
Figure QLYQS_15
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure QLYQS_16
is->
Figure QLYQS_17
,/>
Figure QLYQS_18
Is->
Figure QLYQS_19
,/>
Figure QLYQS_20
Is->
Figure QLYQS_21
3. The method for improving the analysis performance of the instrument vector signal based on the deep learning according to claim 1, wherein the training of the weight set of the MLP by the M training data set in the step (5) is specifically:
training the weight set of the MLP by the M training data set according to the following formula:
Figure QLYQS_22
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure QLYQS_23
and->
Figure QLYQS_24
Representing the real part and the imaginary part of the mth distortion-containing data, respectively, +.>
Figure QLYQS_25
And->
Figure QLYQS_26
Representing the real part and the imaginary part of the mth ideal transmission data, respectively,/->
Figure QLYQS_27
And->
Figure QLYQS_28
Representing the real and imaginary parts of the error of the mth data, respectively, M representing the size of the training data set.
CN201811543273.5A 2018-12-17 2018-12-17 Processing method for improving instrument vector signal analysis performance based on deep learning Active CN109583575B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811543273.5A CN109583575B (en) 2018-12-17 2018-12-17 Processing method for improving instrument vector signal analysis performance based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811543273.5A CN109583575B (en) 2018-12-17 2018-12-17 Processing method for improving instrument vector signal analysis performance based on deep learning

Publications (2)

Publication Number Publication Date
CN109583575A CN109583575A (en) 2019-04-05
CN109583575B true CN109583575B (en) 2023-06-23

Family

ID=65930500

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811543273.5A Active CN109583575B (en) 2018-12-17 2018-12-17 Processing method for improving instrument vector signal analysis performance based on deep learning

Country Status (1)

Country Link
CN (1) CN109583575B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111259861A (en) * 2020-02-18 2020-06-09 西北工业大学 Underwater acoustic communication signal modulation mode identification method under data set unbalanced condition
CN111999587B (en) * 2020-08-31 2023-07-28 中电科思仪科技股份有限公司 Iterative vector error correction method for modulation domain network parameter test
CN112187266B (en) * 2020-09-29 2024-03-08 中国电子科技集团公司第三十六研究所 Nonlinear correction method and device of analog-to-digital converter and electronic equipment

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170228646A1 (en) * 2016-02-04 2017-08-10 Qualcomm Incorporated Spiking multi-layer perceptron
CN108037520A (en) * 2017-12-27 2018-05-15 中国人民解放军战略支援部队信息工程大学 Direct deviations modification method based on neutral net under the conditions of array amplitude phase error
CN108959728A (en) * 2018-06-12 2018-12-07 杭州法动科技有限公司 Radio-frequency devices parameter optimization method based on deep learning

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170228646A1 (en) * 2016-02-04 2017-08-10 Qualcomm Incorporated Spiking multi-layer perceptron
CN108037520A (en) * 2017-12-27 2018-05-15 中国人民解放军战略支援部队信息工程大学 Direct deviations modification method based on neutral net under the conditions of array amplitude phase error
CN108959728A (en) * 2018-06-12 2018-12-07 杭州法动科技有限公司 Radio-frequency devices parameter optimization method based on deep learning

Also Published As

Publication number Publication date
CN109583575A (en) 2019-04-05

Similar Documents

Publication Publication Date Title
CN109583575B (en) Processing method for improving instrument vector signal analysis performance based on deep learning
EP1829314B1 (en) Calibrating amplitude and phase imbalance and DC offset of an analog I/Q modulator in a high-frequency transmitter
US10042029B2 (en) Calibration of test instrument over extended operating range
CN109617560A (en) A kind of I/Q signal compensation for calibrating errors method
US20050129152A1 (en) Method and sytem for noise reduction in measurement receivers using automatic noise subtraction
KR101294771B1 (en) Filter equalization using magnitude measurement data
US7161511B2 (en) Linearization system and method
CN106534038B (en) Ultra-wideband complex format vector modulation error correction method based on two-stage predistortion
EP1152563B1 (en) Method for displaying signal quality measurement
CN110336572A (en) A kind of gain flatness compensation method of transceiver
CN109067677B (en) Adjustable EVM vector signal generation method based on Gaussian white noise
US10379162B1 (en) System for performing modulation analysis without using a modulated signal
US11796621B2 (en) Fast convergence method for cross-correlation based modulation quality measurements
CN101518014A (en) Calibrating DC offset and I/Q imbalance of analogue I/Q-modulator of transmitter
US10841019B1 (en) Cross-correlation measurements for modulation quality measurements
CN114301552A (en) Digital modulation signal testing method and system
KR101325658B1 (en) Characterization of a frequency response for a frequency translation device
CN104040982B (en) The mechanism of correction is detracted for I/Q and detracts measurement using the transmitter for offseting local oscillator
CN107064610A (en) Being averaged based on group delay
CN112051532A (en) Intermediate frequency calibration method based on vector network analyzer
CN107728101B (en) Angular precision calibration method for microwave landing simulator
CN112130011A (en) Satellite-borne power amplifier EVM testing method
Qiu et al. Calibration of a gigahertz-bandwidth vector signal generation and analysis system
Fu et al. Test bed for power amplifier behavioral characterization and modeling
CN112888025B (en) Method and system for processing received WIFI data based on extended bandwidth

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
GR01 Patent grant
GR01 Patent grant