CN-122026967-A - Multi-star large-scale MIMO downlink precoding method and system considering power amplifier nonlinearity

CN122026967ACN 122026967 ACN122026967 ACN 122026967ACN-122026967-A

Abstract

The invention discloses a multi-star large-scale MIMO downlink precoding method and system considering power amplifier nonlinearity. The method comprises the steps of pre-compensating transmitting signals of satellites according to different time delays and Doppler frequency shifts of different satellite signals reaching a user terminal, modeling input and output characteristics of a nonlinear power amplifier at a satellite side by utilizing a polynomial model, establishing an optimization problem to jointly design a precoding vector and a receiving vector, considering interference and distortion from other satellites, interference and distortion in the satellite and noise items in a signal interference noise distortion ratio in the optimization problem, and converting a sum rate maximization problem into a weighted mean square error minimization problem to solve. And further, the traditional iterative updating step is mapped into a learnable network layer structure through a depth unfolding algorithm, so that the computational complexity is reduced. Compared with the traditional method, the method and the device are more in line with the actual transmission condition, can improve the system performance, reduce the online calculation cost, and are suitable for precoding design under the multi-star communication scene.

Inventors

YOU LI
GAO XIQI
ZHANG HUITING
ZHOU HUIBIN
WANG YUANSHUO
ZHU FENG
WANG YUNFEI
CAI PENGHAO
SUN CHEN
LU ANAN

Assignees

东南大学
紫金山实验室

Dates

Publication Date: 20260512
Application Date: 20260212

Claims (10)

1. A multi-star large-scale MIMO downlink precoding method considering power amplifier nonlinearity is characterized by comprising the following steps: pre-compensating the transmitted signals of all satellites by considering the difference of time delay and Doppler frequency shift of different satellite signals reaching a user terminal; Modeling the input and output characteristics of the nonlinear power amplifier by using a polynomial model in consideration of signal distortion when the satellite-side power amplifier enters a nonlinear region; establishing an optimization problem, maximizing system downlink traversal and rate, and jointly designing a precoding vector and a receiving vector, wherein the traversal and rate are calculated based on a signal interference noise distortion ratio from a satellite to a user terminal, and the denominator of the signal interference noise distortion ratio comprises interference and distortion from other satellites, interference and distortion in the satellite and noise items; The method comprises the steps of converting a sum rate maximization problem into a weighted mean square error minimization problem, obtaining a precoding vector and a receiving vector by utilizing Lagrangian multipliers and gradient descent iterative solution, wherein the receiving vector is obtained by minimizing Lagrangian functions, the precoding vector is obtained by adopting two layers of nested iterative solution, an outer layer loop updates intermediate variables of the receiving vector and a weighted mean square error algorithm, and an inner layer loop optimizes the precoding vector, the intermediate variables comprise a detector and a weighting factor, the detector is obtained by minimizing the mean square error, the weighting factor is the reciprocal of the mean square error, and an optimal detector and the weighting factor are calculated according to a closed solution through the receiving vector, the precoding vector, a channel matrix, an average power amplifier gain matrix and an interference noise distortion term in the iterative process.
2. The multi-star massive MIMO downlink precoding method considering power amplifier nonlinearity according to claim 1, wherein the iterative updating step is mapped into a learnable network layer structure by a deep expansion method, wherein each optimized iteration is mapped into a layer of neural network structure, each layer comprises an updating module for a detector, a receiving vector and a precoding vector, the step size of gradient descent of each optimized iteration is used as a trainable parameter, updating is performed by end-to-end network training learning, and the loss functions of an outer layer network and an inner layer network are respectively set as weighted minimum mean square error and Lagrangian functions.
3. The multi-star massive MIMO downlink precoding method considering power amplifier nonlinearity according to claim 1, wherein modeling the input/output characteristics of the nonlinear power amplifier using a polynomial model comprises: Modeling the input and output characteristics of the power amplifier by adopting a memory-free third-order polynomial model; Decoupling the output signal into a linear component and a nonlinear distortion component, converting the instantaneous amplification gain into an equivalent average gain, wherein the average power amplification gain matrix is expressed as The covariance matrix of the distortion term is expressed as , And Respectively represent the first The coefficients of the primary term and the tertiary term of the power amplifier of each satellite, Representing the identity matrix of the cell, Representing the total number of antennas of the satellite, , Represent the first Satellite pair number The precoding vector of the individual user terminals, Represent the first A set of user terminals for a satellite service.
4. The multi-star massive MIMO downlink precoding method considering power amplifier nonlinearity according to claim 1, wherein the precoding optimization problem that maximizes the system downlink traversal and rate is expressed as: ; Wherein the method comprises the steps of And To optimize the variables, the received vectors of all the user terminal sides and the precoding matrix of all the satellite sides are respectively represented, Represent the first Satellite to the first The downlink traversal rate at which the individual user terminals transmit signals, Represent the first Satellite pair number The precoding vector of the individual user terminals, Represent the first The user terminal pair The received vector of the individual satellites is used, Representing the transmission power budget for each satellite, Represent the first A set of user terminals for a single satellite service, Representing a set of satellites.
5. The multi-star massive MIMO downlink precoding method considering power amplifier nonlinearity of claim 4, wherein the weighted mean square error minimization problem is expressed as: ; Wherein the method comprises the steps of And The introduced optimization variables, representing the detector set and the weighting factor set respectively, The weighting factor, which is introduced, is the inverse of the mean square error, , A linear detector is shown as such, Representing the distortion term of the interference noise, Represent the first Satellites to the first The channel matrix of the individual user terminals, And Respectively represent the first The average power amplifier gain matrix of each satellite and the covariance matrix of the distortion term, , Is the variance of noise Represent the first Satellite to the first And transmitting covariance matrix of each user terminal.
6. The method for multi-star massive MIMO downlink precoding with power amplifier nonlinearity according to claim 5, wherein the iterative solution using Lagrangian multipliers and gradient descent to obtain precoding vectors and receiving vectors comprises initializing the precoding vectors in advance And receiving the vector And determining a channel matrix Gain matrix of average power amplifier Distortion matrix ; According to the formula And Calculating initial intermediate variables And ; In the first place Updating the received vector in a number of iterations by the following expression : ; Wherein the method comprises the steps of In the form of a composite channel matrix, Representing by substitution And The obtained interference and distortion from other satellites and noise are initialized to step size before starting the inner loop solving of the precoding vector The precoding vector is then updated by moving in the direction of the steepest descent of the Lagrangian function, in each iteration The updating mode of (a) is as follows: ; Wherein the method comprises the steps of Represent the first In the second iteration Step size of secondary gradient descent.
7. The multi-star massive MIMO downlink precoding method considering power amplifier nonlinearity according to claim 6, wherein the step size of gradient descent needs to satisfy the following conditions: ; Wherein the method comprises the steps of , , And is also provided with 。
8. The method of multi-star massive MIMO downlink precoding with power amplifier nonlinearity according to claim 5, wherein each WMMSE iteration is mapped into a layer of neural network structure, each layer contains an update module for the detector, the reception vector and the precoding vector, and the abduction depth of the network is set to be Corresponding to the main iteration times of the WMMSE method, the depth is expanded inwards Representing the number of iterative steps inside each layer for gradient descent update of the precoding matrix, the step length of the gradient descent as a trainable parameter, and the step length set is represented as , , Represent the first The second outer loop is expanded The updating step length of the secondary internal gradient descent module is updated through end-to-end network training learning, and the outer layer loss function and the inner layer loss function of the network are respectively set as weighted minimum mean square error and Lagrange function, and the specific expression is as follows: ; The outer layer is circulated until the iteration times are reached Or the loss function satisfies The termination is made upon the convergence condition, In order to ensure a convergence threshold value, in a network training stage, training samples are generated based on statistical channel state information, the training samples are used as input data of a deep expansion network, meanwhile, precoding results corresponding to the channel state information are obtained according to a traditional optimization algorithm method and are used as supervision information, gradient descent step length parameters corresponding to all layers in the deep expansion network are trained and optimized, in an application stage, the trained deep expansion network is called, after new channel state information is received, iterative optimization solution is not needed, and corresponding downlink precoding results can be directly calculated and output through forward propagation.
9. A computer system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the computer program when executed by the processor implements the steps of the multi-star massive MIMO downlink precoding method according to any of claims 1-8 taking into account power amplifier nonlinearities.
10. A computer program product comprising a computer program, characterized in that the computer program when executed by a processor implements the steps of the multi-star massive MIMO downlink precoding method according to any of claims 1-8, taking into account power amplifier nonlinearities.

Description

Multi-star large-scale MIMO downlink precoding method and system considering power amplifier nonlinearity Technical Field The invention relates to the field of multi-star communication and non-ideal hardware, in particular to a multi-star large-scale MIMO downlink precoding method and system considering power amplifier nonlinearity. Background The low-orbit satellite network is used as a core infrastructure of an air-ground integrated information network and is generally deployed in a near-earth orbit 200-2000 km away from the ground surface, and has the remarkable advantages of low propagation delay, small path loss, wide coverage range and the like. With the large-scale deployment of the global low-orbit satellite constellation, the ground user terminal can access the communication services of a plurality of satellites at the same time, which creates favorable conditions for the realization of the multi-satellite transmission technology. However, most of the existing researches on multi-star communication systems are based on idealized system assumptions, and technical challenges existing in actual systems are not fully considered. Due to the high-speed motion characteristics of low-orbit satellites, the signals received by the user terminal from different satellites have obvious asynchronous characteristics in the time domain and the frequency domain due to factors such as long-distance propagation between satellites and ground, terminal mobility and the like. In addition, the ubiquitous hardware impairments of on-board and off-board communication devices (e.g., power amplifier nonlinearities, phase noise, I/Q imbalance, etc.) can further degrade system performance. Therefore, the research on the downlink precoding technology of the multi-satellite large-scale MIMO system has important theoretical value and engineering significance, and particularly, a non-ideal system model comprising signal asynchronism and hardware damage needs to be established, and the research on the optimization of the system performance is carried out based on the non-ideal system model. Disclosure of Invention Aiming at the problems of signal asynchronism and hardware damage existing in the downlink transmission of a multi-star large-scale MIMO system in the prior art, particularly the signal distortion caused by a nonlinear power amplifier, the invention provides a multi-star large-scale MIMO downlink precoding method and system considering the nonlinearity of the power amplifier, which improves the system performance and reduces the calculation complexity. The technical scheme is that in order to achieve the aim of the invention, the invention adopts the following technical scheme: a multi-star large-scale MIMO downlink precoding method considering power amplifier nonlinearity comprises the following steps: pre-compensating the transmitted signals of all satellites by considering the difference of time delay and Doppler frequency shift of different satellite signals reaching a user terminal; Modeling the input and output characteristics of the nonlinear power amplifier by using a polynomial model in consideration of signal distortion when the satellite-side power amplifier enters a nonlinear region; establishing an optimization problem, maximizing system downlink traversal and rate, and jointly designing a precoding vector and a receiving vector, wherein the traversal and rate are calculated based on a signal interference noise distortion ratio from a satellite to a user terminal, and the denominator of the signal interference noise distortion ratio comprises interference and distortion from other satellites, interference and distortion in the satellite and noise items; The method comprises the steps of converting a sum rate maximization problem into a weighted mean square error minimization problem, obtaining a precoding vector and a receiving vector by utilizing Lagrangian multipliers and gradient descent iterative solution, wherein the receiving vector is obtained by minimizing Lagrangian functions, the precoding vector is obtained by adopting two layers of nested iterative solution, an outer layer loop updates intermediate variables of the receiving vector and a weighted mean square error algorithm, and an inner layer loop optimizes the precoding vector, the intermediate variables comprise a detector and a weighting factor, the detector is obtained by minimizing the mean square error, the weighting factor is the reciprocal of the mean square error, and an optimal detector and the weighting factor are calculated according to a closed solution through the receiving vector, the precoding vector, a channel matrix, an average power amplifier gain matrix and an interference noise distortion term in the iterative process. Further, the iterative updating step is mapped into a network layer structure capable of learning through a depth unfolding method, each optimized iteration is mapped into a neural network structure, each layer