CN-122001447-A - Self-adaptive sum-difference channel phase compensation method based on double-flow feature fusion

CN122001447ACN 122001447 ACN122001447 ACN 122001447ACN-122001447-A

Abstract

The invention relates to the field of wireless communication networks, and provides a self-adaptive and difference channel phase compensation method based on double-flow feature fusion, which comprises the steps of receiving non-cooperative target star signals, constructing a space network pre-training data set by combining antenna beam pointing, and pre-training a space network model to output a static phase difference compensation predicted value; the method comprises the steps of constructing a time network pre-training data set based on historical phase deviation cached by a register and a temperature difference value of an environment sensor, pre-training a time network model to output a phase difference change trend predicted value, carrying out feature fusion on a static phase difference compensation predicted value and the phase difference change trend predicted value, training through a fusion network model, outputting a final phase compensation value of a sum channel and a final phase compensation value of a difference channel, and driving a digital phase shifter to carry out real-time phase correction. By means of heterogeneous feature separation of the spatial stream and the time stream, phase compensation accuracy and environmental robustness are greatly improved, and accurate output of sum and difference channel phase compensation is guaranteed particularly in a complex dynamic scene.

Inventors

PAN KUNBEI
HUANG BO
HUANG YU
CHENG QINGLIN
PENG YIWEN
CHEN XIN
ZHANG GUOJIN
CAI LULU

Assignees

上海航天测控通信研究所

Dates

Publication Date: 20260508
Application Date: 20260206

Claims (10)

1. A self-adaptive and difference channel phase compensation method based on double-flow characteristic fusion is characterized by comprising the following steps of, Step S1, receiving a non-cooperative target star signal, combining antenna beam pointing to construct a space network pre-training data set, and pre-training a space network model based on the space network pre-training data set to output a local optimal space feature vector as a static phase difference compensation predicted value; S2, constructing a time network pre-training data set based on the historical phase deviation cached by a register and the temperature differential value of an environmental sensor, and pre-training a time network model through the time network pre-training data set to output a local optimal time feature vector as a phase difference change trend predicted value in a future time domain window; and S3, carrying out feature fusion on the static phase difference compensation predicted value and the phase difference change trend predicted value, constructing a fusion network training data set, training through a fusion network model, outputting a final phase compensation value of a sum channel and a difference channel, and driving a digital phase shifter to carry out real-time phase correction so as to realize self-adaptive tracking of a non-cooperative target star.
2. The adaptive and bad channel phase compensation method based on dual-stream feature fusion of claim 1, wherein in step S1, receiving non-cooperative target star signals, constructing a spatial network pre-training data set in combination with antenna beam pointing, comprises: step S11, receiving the non-cooperative target star signal, extracting the working frequency point and the received signal strength, and respectively carrying out normalization processing, wherein, The normalization formula of the working frequency point is that, ; In the formula, For the normalized operating frequency point, As a result of the frequency of operation, For the lowest operating frequency of the self-tracking receiver, The highest operating frequency of the self-tracking receiver; the normalization formula of the received signal strength is that, ; In the formula, For the normalized received signal strength, For the received signal strength of the signal, For self-tracking receiver minimum enabled power, Maximum saturated power for the self-tracking receiver; Step S12, constructing a pre-training data set of the space network according to the antenna beam pointing acquisition direction angle and pitch angle and combining the normalized working frequency point and the normalized received signal strength as input data of the space network, and recording as , wherein, For the direction angle, EL is the pitch angle, And And the working frequency point and the received signal strength after normalization are respectively obtained.
3. The adaptive and bad channel phase compensation method based on dual-flow feature fusion of claim 2, wherein in step S1, pre-training a spatial network model based on the spatial network pre-training dataset comprises: Step S13, constructing a space network pre-training loss function, wherein the space network pre-training loss function is used for evaluating a phase compensation error between a space network predicted phase difference compensation value and a real phase difference compensation value, and the formula is as follows: ; step S14, optimizing the spatial network model by taking the phase compensation error in the minimized static scene as an optimization target; step S15, in back propagation, the gradient of the spatial network pre-training loss function is calculated as, And pretraining a loss function based on the spatial network Is trained by an Adam optimizer, and each training round is used for spatial network parameters Updating is carried out, the updating formula is that, ; In the formula, Represent the first The number of training rounds is the same as the number of training rounds, Is the learning rate of the spatial network parameters; i is a sample index, which corresponds to the ith sample in batch processing; is a spatial network output representing the network predicted phase difference compensation value for the self-tracking receiver and the difference channel for the ith sample; is the true phase compensation value representing the measured phase difference compensation value of the self-tracking receiver and difference channel of the ith sample; Is a spatial network parameter representing a trainable weight of the CNN; is a regularization coefficient representing the intensity of the suppressed overfitting.
4. The adaptive sum and difference channel phase compensation method based on dual-stream feature fusion according to claim 3, wherein in step S1, a locally optimal spatial feature vector is output as a static phase difference compensation predicted value, comprising: step S16, in the pre-training process of the space network model, when the loss value corresponding to the current training round of the space network pre-training loss function is lower than a preset space network pre-training loss threshold value or the pre-training loss change rate between two adjacent training rounds is smaller than a preset space network pre-training loss change rate threshold value, terminating the pre-training process of the space network model, outputting the corresponding local optimal 256-dimensional space feature vector, and recording as As the static phase difference compensation predicted value.
5. The adaptive and differential channel phase compensation method based on dual-stream feature fusion of claim 1, wherein in step S2, constructing a time network pre-training dataset based on the register-cached historical phase bias and the temperature differential value of the environmental sensor comprises: step S21, acquiring a historical phase deviation sequence according to the register cache: ; step S22, acquiring a historical radio frequency front end temperature sequence according to the environmental sensor: ; Step S23, constructing the time network pre-training data set , , wherein, The historical phase bias, normalized, is expressed as , The window mean value is represented as such, The standard deviation of the window is indicated, Representing a numerical stability term parameter; Normalized to the temperature differential value, expressed as , 。
6. The adaptive and bad channel phase compensation method based on dual-flow feature fusion of claim 3, wherein in step S2, pre-training the time network model through the time network pre-training data set comprises: Step S24, constructing the time network pre-training loss function, taking the phase compensation error of the self-tracking receiver and the difference channel, which is introduced by phase drift under the minimized dynamic environment, as the optimization target to optimize the time network model, wherein the formula of the time network pre-training loss function is as follows, ; Step S25, during said counter-propagating, the gradient of said time network pre-training loss function is calculated as, And pre-training a loss function based on the time network Is trained by the Adam optimizer, and time network parameters are calculated for each time step Updating is carried out, the updating formula is that, ; In the formula, T is the length of the time sequence; Representing the phase compensated prediction value of the time network model at sample i time t, Is the true phase compensation value and, Is the learning rate of the time network parameter; Wherein Is of first order momentum, with The initial value is zero; Is the first order decay rate; Wherein Is of second order momentum, with The initial value is zero; Is the second order decay rate; Is the gradient of the time network pretraining loss function, and has 。
7. The adaptive and difference channel phase compensation method based on dual-stream feature fusion according to claim 6, wherein in step S2, a locally optimal time feature vector is output as a phase difference variation trend prediction value in a future time domain window, comprising: Step S26, in the training process of the time network model, when the change rate of the time network pre-training loss in a preset round window is smaller than a preset time network pre-training loss change rate threshold value, terminating the training process of the time network model, outputting a corresponding local optimal 512-dimensional time feature vector, and recording as As the phase difference variation trend prediction value in the future time domain window.
8. The adaptive and bad channel phase compensation method based on dual-stream feature fusion according to claim 7, wherein in step S3, feature fusion is performed on the static phase difference compensation predicted value and the phase difference variation trend predicted value, and a fused network training data set is constructed, comprising: For the locally optimal 256-dimensional spatial feature vector And the locally optimal 512-dimensional temporal feature vector And the retrograde feature alignment and normalization are used as the fusion network training data set to respectively solve the spatial feature dimension difference and the temporal feature dimension difference: ; ; in the formula, Is a space feature adapter, obeys Kaiming normal distribution, and carries out the local optimum 256-dimensional space feature vector Projecting to a fusion space; Is a time feature adapter, obeys the uniform distribution of the Xavier, and carries out the local optimum 512-dimensional time feature vector Projected to the fusion space.
9. The adaptive and bad channel phase compensation method based on dual-flow feature fusion of claim 8, wherein in step S3, training is performed by fusing a network model, comprising: Step S31, constructing a fusion network pre-training loss function based on the static phase difference compensation predicted value and the phase difference change trend predicted value, wherein the fusion network pre-training loss function is that, ; Step S32, in the process of training the fusion network model, parameter updating is carried out on the spatial feature adapter and the time feature adapter in the fusion network based on the fusion network pre-training loss function so as to reduce the phase compensation error of the fusion network output, The gradient update formula of the spatial signature adapter is, ; The gradient update formula of the time feature adapter is, ; In the formula, Is that The maximum singular value left singular vector, Is that I is the sample index, and corresponds to the ith sample in batch processing; And The local optimal 256-dimensional space feature vector and the local optimal 512-dimensional time feature vector are respectively projected to a fused space and then normalized, And The spatial feature adapter and the temporal feature adapter respectively, 、 Representing the phase compensation predicted value and the true phase compensation value of the sample i respectively, And The locally optimal 256-dimensional spatial feature vector and the locally optimal 512-dimensional temporal feature vector, respectively, wherein, In the process of training the fusion network model, regarding the gradient descent strategy of the spatial feature adapter, training the fusion network model regarding spatial features through the Adam optimizer, and updating parameters of the spatial feature adapter in each training round, wherein an updating formula is that, In which, in the process, Represent the first The number of training rounds is the same as the number of training rounds, Representing the learning rate of the converged network parameters; In the process of training the fusion network model, regarding the gradient descent strategy of the time feature adapter, training the fusion network model regarding the time feature through the Adam optimizer, updating parameters of the time feature adapter in each training round, wherein an updating formula is that, In which, in the process, Represent the first The number of training rounds is the same as the number of training rounds, Representing the learning rate of the converged network parameters; The training of the local optimal 256-dimensional space feature vector after training and fusion can be respectively output by the fusion network model and the fusion network And the locally optimal 512-dimensional temporal feature vector As a final phase difference compensation value.
10. The adaptive sum and difference channel phase compensation method based on dual-stream feature fusion according to claim 9, wherein outputting the final phase compensation value of the sum and difference channels and driving the digital phase shifter for real-time phase correction in step S3 comprises: In the training process of the fusion network model, when the change rate of the fusion network pre-training loss is smaller than the preset fusion network pre-training loss change rate threshold value in the preset training round times, terminating the training process of the fusion network model and outputting the final phase compensation value of the corresponding sum and difference channel The digital phase shifter is used for directly driving the digital phase shifter to carry out real-time phase correction.

Description

Self-adaptive sum-difference channel phase compensation method based on double-flow feature fusion Technical Field The invention relates to the field of wireless communication networks, in particular to a self-adaptive and difference channel phase compensation method based on double-flow feature fusion. Background With the iterative evolution of wireless communication technology and the breakthrough progress of semiconductor technology, satellite development has entered into the fast traffic lane since the 21 st century, and current satellite communication systems are widely used in the fields of civil communication, navigation positioning, resource exploration, electronic countermeasure, etc. In the future 6G space-earth integrated network architecture, satellite communication is taken as an important component, is complementary with the traditional ground network, and can eliminate digital gaps and support global seamless connection. Because the satellite-ground link has the characteristics of high-speed relative motion, frequent working condition change and the like, the high-precision antenna tracking capability is a key basis for guaranteeing the stability of the satellite communication link and reliable data transmission. In the existing satellite communication system, a single-pulse self-tracking technology is generally adopted to realize high-precision pointing control of a target satellite. According to the technology, a sum channel and difference channel signal is formed through a multi-feed antenna, an angle error signal of an azimuth angle and a pitch angle direction is calculated in real time, and a servo system is driven to carry out closed-loop control, so that dynamic tracking of an antenna beam to a target star is realized. However, the single-pulse self-tracking technology has high requirements on phase consistency between the sum and difference channels, and when an uncompensated phase difference exists between the sum and difference channels, the angle error signal calculation result deviates from the real direction, so that tracking accuracy is reduced. In the prior art, aiming at the phase difference problem of the sum and difference channels, a stepping adjustment or ground manual calibration mode is generally adopted for compensation. However, in a complex dynamic environment on the satellite, the characteristics of the front end of the radio frequency are easy to drift due to the influence of factors such as temperature change, device aging, working frequency point switching and the like, so that the phase difference is changed along with time. When the phase difference is large, the angle error signal may have direction reversal, so that the stepping adjustment process is difficult to converge, and even the ping-pong effect of the antenna pointing to swing back and forth is caused, and the stability of the system is seriously affected. For a cooperative target star, the prior art can realize on-board calibration by means of a beacon signal, but under a non-cooperative target star scene, an effective beacon cannot be obtained, only manual calibration on the ground can be relied on, dynamic environment changes are difficult to adapt, and the maintenance cost is high. Therefore, in the on-board application scenario of the non-cooperative target star, the problems that the phase compensation depends on manpower, the dynamic adaptability is insufficient, the tracking stability is difficult to guarantee and the like generally exist, and a technical scheme capable of realizing the phase adaptive compensation of the sum and difference channel under a complex dynamic environment is needed. Disclosure of Invention Aiming at the problems that the neutralization channel phase of a self-tracking receiver is easy to drift, depends on manual calibration and is difficult to stably compensate in a non-cooperative target star and on-satellite complex dynamic environment, the invention provides a self-adaptive and difference channel phase compensation method based on double-current feature fusion, which improves the tracking stability and the pointing precision of a single-pulse self-tracking system, and the invention can be realized by the following technical scheme: the invention provides a self-adaptive and difference channel phase compensation method based on double-flow characteristic fusion, which comprises the following steps of, Step S1, receiving a non-cooperative target star signal, combining antenna beam pointing to construct a space network pre-training data set, and pre-training a space network model based on the space network pre-training data set to output a local optimal space feature vector as a static phase difference compensation predicted value; S2, constructing a time network pre-training data set based on the historical phase deviation cached by the register and the temperature differential value of the environmental sensor, and pre-training a time network model through the time