CN114024586B

CN114024586B - Intelligent beam prediction method, device, equipment and medium for nonlinear track

Info

Publication number: CN114024586B
Application number: CN202111310209.4A
Authority: CN
Inventors: 孟帆; 黄永明; 尤肖虎
Original assignee: Network Communication and Security Zijinshan Laboratory
Current assignee: Network Communication and Security Zijinshan Laboratory
Priority date: 2021-11-04
Filing date: 2021-11-04
Publication date: 2022-10-04
Anticipated expiration: 2041-11-04
Also published as: CN114024586A

Abstract

The invention discloses an intelligent beam prediction method, a device, equipment and a medium for a nonlinear track, which are used for acquiring received pilot signals and measurement signals fed back by a mobile terminal at a plurality of observation moments, wherein the motion track of the mobile terminal is a nonlinear track; respectively calculating a mobile terminal projection position estimation value and a speed estimation value based on the received pilot signal and a mobile terminal projection position estimation value and a speed estimation value based on the measurement signal according to an estimation method and a mapping table, and fusing a neural network model according to data obtained by pre-training to obtain a mobile terminal projection position and speed output by the model; calculating the projection position of the mobile terminal at the predicted moment according to the projection position and the projection speed of the mobile terminal output by the model and a mapping table; and realizing beam prediction according to the projection position of the mobile terminal at the prediction moment. The invention is beneficial to greatly reducing the beam training overhead and the instruction issuing time delay in beam alignment and tracking and improving the frequency spectrum efficiency.

Description

Intelligent beam prediction method, device, equipment and medium for nonlinear track

Technical Field

The invention relates to the technical field of millimeter wave MIMO wireless mobile communication, in particular to an intelligent beam prediction method, device, equipment and medium for nonlinear tracks.

Background

Millimeter wave multi-user multiple-input multiple-output (MIMO) wireless communication uses large-scale antennas and beam forming technology to solve the path loss problem under a high frequency band, space division multiplexing is achieved, and spectrum efficiency is improved. However, in mobile scenarios with large-scale antennas, especially in high-speed railway scenarios, beam alignment and tracking have a large amount of beam training overhead and significant command issue delays. Therefore, reducing the beam training overhead and the command issue delay are key issues in mobile wireless communications. The existing beam management framework still has a great space for improving the problem solving aspect, and the beam prediction technology of long-time fine-time granularity can greatly reduce the beam training overhead and avoid the time delay caused by instruction issuing.

The method based on model driving in wireless communication has good theoretical guarantee and interpretability, but when the method faces a nonlinear complex scene containing an ambiguous prior, the model driving method cannot effectively solve the problem, and the performance is obviously reduced.

Disclosure of Invention

The technical purpose is as follows: aiming at the defects in the prior art, the invention discloses an intelligent beam prediction method, device, equipment and medium for a nonlinear track, which are beneficial to greatly reducing beam training overhead and instruction issuing time delay in beam alignment and tracking and improving spectrum efficiency.

The technical scheme is as follows: in order to realize the technical purpose, the invention adopts the following technical scheme: an intelligent beam prediction method for a nonlinear trajectory, comprising the steps of:

acquiring received pilot signals and measurement signals fed back by a mobile terminal at a plurality of observation moments, wherein the motion track of the mobile terminal is a nonlinear track;

respectively calculating a projection position estimation value and a speed estimation value of the mobile terminal based on the received pilot signal and a projection position estimation value and a speed estimation value of the mobile terminal based on the measurement signal according to a parameter estimation method of a probability theory and a pre-constructed mapping table about the projection position and the absolute path length of the mobile terminal, and fusing a neural network model according to pre-trained data to obtain the projection position and the speed of the mobile terminal output by the model;

calculating the projection position of the mobile terminal at the predicted moment according to the projection position and the projection speed of the mobile terminal output by the model and by combining the mapping table;

and realizing beam prediction according to the projection position of the mobile terminal at the prediction moment.

Further, the parameter estimation method of the probability theory includes a likelihood estimation method and/or a Bayesian estimation method.

Further, according to a parameter estimation method of probability theory and the received pilot signal, determining a probability function about the projection position and the speed of the mobile terminal based on the received pilot signal, and combining the mapping table to iteratively calculate the projection position estimation value and the speed estimation value of the mobile terminal based on the received pilot signal;

and determining a probability function about the projection position and the speed of the mobile terminal based on the measurement signal according to a parameter estimation method of probability theory and the measurement signal, and combining the mapping table to iteratively calculate the projection position estimation value and the speed estimation value of the mobile terminal based on the measurement signal.

Further, the method for constructing the mapping table includes:

the method comprises the steps of constructing a piecewise function fitting nonlinear track, dividing projection position independent variables in the piecewise function at equal intervals, calculating absolute path length corresponding to each divided projection position of the mobile terminal, and recording the corresponding relation between the projection position of the mobile terminal and the corresponding absolute path length into a mapping table;

the absolute path length is the integral path length of the mobile terminal on the nonlinear track by taking the initial projection position as a starting point.

Further, in both iterative calculations, the following are included: obtaining the projection position of the mobile terminal at any other observation time except the last observation time according to the mapping table, wherein the method comprises the following steps:

searching the absolute path length corresponding to the projection position of the mobile terminal at the last observation moment from the mapping table, and recording the absolute path length as a first absolute path length;

calculating the relative path length between the projection position of the mobile terminal at any other observation time and the projection position at the last observation time, and recording the relative path length as a second relative path length;

adding the first absolute path length and the second relative path length to obtain a third absolute path length;

and searching and obtaining the projection position of the mobile terminal at any other observation moment in the mapping table according to the third absolute path length.

Further, the data fusion neural network model comprises a position network and a speed network; the position network is used for outputting the variance of the projection position estimation values in the two groups of estimation values and the deviation of the projection position estimation values; the speed network is used for outputting the variance of the speed estimation values and the deviation of the speed estimation values in the two groups of estimation values;

correcting the two projection position estimation values of the mobile terminal according to the deviation of the projection position estimation values, and then distributing the weight of the projection position estimation values to the corrected estimation values to obtain the projection position of the mobile terminal output by the model; the weight of the projection position estimation value is obtained according to the variance of the projection position estimation value;

correcting the two speed estimation values of the mobile terminal according to the deviation of the speed estimation values, and then distributing the weight of the speed estimation values to the corrected estimation values to obtain the speed of the mobile terminal output by the model; the weight of the velocity estimation value is obtained according to the variance of the velocity estimation value.

Further, the position network and the speed network both comprise a variance sub-network and a bias sub-network, and the variance sub-network and the bias sub-network are both neural network models comprising two hidden layers;

the variance sub-network is used for outputting variance, and the bias sub-network is used for outputting deviation.

Further, the calculating the projection position of the mobile terminal at the predicted time according to the projection position and the projection speed of the mobile terminal output by the model and by combining the mapping table includes:

searching the absolute path length corresponding to the projection position of the mobile terminal output by the model from the mapping table, and recording as a fourth absolute path length;

calculating the relative path length between the projection position of the mobile terminal at the prediction time and the projection position of the mobile terminal output by the model according to the speed of the mobile terminal output by the model, and recording as a fifth relative path length;

adding the fourth absolute path length and the fifth relative path length to obtain a sixth absolute path length which is used as the absolute path length corresponding to the projection position of the mobile terminal at the prediction time;

and searching the projection position of the mobile terminal at the predicted moment in the mapping table according to the sixth absolute path length.

Further, the beam prediction is realized according to the projection position of the mobile terminal at the prediction time, and the method comprises the following steps:

according to the projection position of the mobile terminal at the predicted time, obtaining the departure angle of a channel LOS of a mobile terminal receiving end, and further obtaining the simulated pre-coding of a base station transmitting end and the mobile terminal receiving end;

obtaining a virtual channel according to the projection position of the mobile terminal at the predicted time and the starting angle of a LOS (line of sight) channel of the receiving end of the mobile terminal, and obtaining digital precoding of the transmitting end of the base station according to the virtual channel;

the base station transmitting end analog precoding and the base station transmitting end digital precoding are both used for the base station to transmit data signals, and the mobile terminal receiving end analog precoding is used for the mobile terminal to receive the data signals transmitted by the base station.

An intelligent beam prediction apparatus for nonlinear trajectories, comprising:

the mobile terminal comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring received pilot signals and measurement signals fed back by the mobile terminal at a plurality of observation moments, and the motion track of the mobile terminal is a nonlinear track;

the position and speed estimation module is used for respectively calculating a mobile terminal projection position estimation value and a speed estimation value based on the received pilot signal and the measurement signal according to a parameter estimation method of probability theory and a pre-constructed mapping table about the projection position and the absolute path length of the mobile terminal, and fusing a neural network model according to data obtained by pre-training to obtain a mobile terminal projection position estimation value and speed output by the model;

the position prediction module is used for calculating the projection position of the mobile terminal at the prediction moment according to the projection position and the projection speed of the mobile terminal output by the model and by combining the mapping table;

and the beam prediction module is used for realizing beam prediction according to the projection position of the mobile terminal at the prediction moment.

Further, the method for constructing the mapping table includes:

the method comprises the steps of fitting a nonlinear track by constructing a piecewise function, dividing projection position independent variables in the piecewise function at equal intervals, calculating absolute path length corresponding to each divided projection position of the mobile terminal, and recording the corresponding relation between the projection position of the mobile terminal and the corresponding absolute path length into a mapping table;

An apparatus comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements any of the above intelligent beam prediction methods for non-linear trajectories when executing the program.

A computer-readable storage medium having stored thereon computer-executable instructions for performing any of the above-described intelligent beam prediction methods for non-linear trajectories.

Has the beneficial effects that: the method is applied to a scene that the mobile terminal does nonlinear track motion, and the received pilot signals and the measurement signals fed back by the mobile terminal at a plurality of observation moments are obtained; according to a parameter estimation method of a probability theory and a pre-constructed mapping table about the projection position and the absolute path length of the mobile terminal, respectively calculating the estimation values of two groups of projection positions and speeds of the mobile terminal, inputting a pre-trained data fusion neural network model, and obtaining the projection position and speed of the mobile terminal output by the model; combining with a mapping table to obtain the projection position of the mobile terminal at the predicted moment; realizing beam prediction according to the predicted projection position; the invention can greatly reduce the beam training overhead and the instruction issuing time delay in beam alignment and tracking, improve the frequency spectrum efficiency and obviously improve the performance.

Drawings

FIG. 1 is a flow chart of a beam prediction method according to an embodiment of the present invention;

FIG. 2 is a flow diagram illustrating beam prediction in one embodiment;

FIG. 3 is a schematic diagram of a communication scenario in a linear rail environment according to an embodiment;

FIG. 4 is a plot of projected position versus projected position estimate MSE for a non-linear trajectory in one embodiment;

FIG. 5 is a projection position-velocity estimation MSE plot under a non-linear trajectory in one embodiment;

FIG. 6 (a) is a graph of projected position-projected position estimate MSE using linear estimation for a non-linear trajectory in one embodiment;

FIG. 6 (b) is a plot of projected position-velocity estimation MSE using linear estimation for a non-linear trajectory in one embodiment;

FIG. 7 is a projection position-SE plot under a non-linear trajectory in one embodiment;

FIG. 8 is a schematic structural diagram of the apparatus of the present invention.

Detailed Description

The method, apparatus, device and medium for intelligent beam prediction for nonlinear trajectory according to the present invention will be further described and explained with reference to the accompanying drawings and embodiments.

Example 1:

as shown in fig. 1 and 2, an intelligent beam prediction method for a non-linear trajectory includes the following steps:

the method comprises the following steps of S1, obtaining a plurality of receiving pilot signals and measuring signals fed back by a mobile terminal at an observation moment, wherein the motion track of the mobile terminal is a nonlinear track;

specifically, at each observation time, a base station transmitting end transmits a group of pilot signals to the mobile terminal, and the base station receives received pilot signals and corresponding measurement signals fed back by the mobile terminal;

the measurement signal comprises a Doppler frequency and a relative communication time delay, wherein the Doppler frequency can be obtained by carrier offset frequency estimation (CFO), and the relative communication time delay is obtained by dividing a linear distance between the base station and the mobile terminal by the speed of light and by millimeter wave ranging; and the mobile terminal feeds back the measurement signal at the equal time interval.

Step S2, respectively calculating a mobile terminal projection position estimation value and a speed estimation value based on the received pilot signal and a mobile terminal projection position estimation value and a speed estimation value based on the measurement signal according to a parameter estimation method of a probability theory and a pre-constructed mapping table about the mobile terminal projection position and the absolute path length, and fusing a neural network model according to data obtained by pre-training to obtain a mobile terminal projection position and speed output by the model;

Further, according to a parameter estimation method of probability theory and the received pilot signal, determining a probability function about the projection position and the projection speed of the mobile terminal based on the received pilot signal, and combining the mapping table to iteratively calculate the projection position estimation value and the projection speed estimation value of the mobile terminal based on the received pilot signal;

according to a parameter estimation method of probability theory and the measurement signal, determining a probability function about the projection position and the speed of the mobile terminal based on the measurement signal, and combining the mapping table to iteratively calculate the projection position estimation value and the speed estimation value of the mobile terminal based on the measurement signal;

Further, the method for constructing the mapping table includes:

the method comprises the steps of constructing a piecewise function fitting nonlinear track, dividing projection position independent variables in the piecewise function at equal intervals, calculating absolute path length corresponding to each divided projection position of the mobile terminal, and recording the corresponding relation between the projection position of the mobile terminal and the corresponding absolute path length into a mapping table.

S3, calculating the projection position of the mobile terminal at the predicted moment according to the projection position and the projection speed of the mobile terminal output by the model and by combining the mapping table;

the projection position of the mobile terminal at the predicted moment can be calculated according to the projection position and the speed estimation value of the mobile terminal output by the model by combining the nonlinear motion of the mobile terminal;

the calculating the projection position of the mobile terminal at the predicted moment according to the projection position and the projection speed of the mobile terminal output by the model and the mapping table comprises the following steps:

calculating the relative path length between the projection position of the mobile terminal at the prediction moment and the projection position of the mobile terminal output by the model according to the speed of the mobile terminal output by the model, and recording the relative path length as a fifth relative path length;

And S4, realizing beam prediction according to the projection position of the mobile terminal at the prediction moment.

The method specifically comprises the following steps:

The method is applied to a scene that the mobile terminal does nonlinear track motion, and the received pilot signals and the measurement signals fed back by the mobile terminal at a plurality of observation moments are obtained; according to a parameter estimation method of probability theory and a mapping table which is constructed in advance and is related to the projection position and the absolute path length of the mobile terminal, the projection position and the speed estimation value of the mobile terminal based on the received pilot signal and the measurement signal are calculated respectively, a data fusion neural network model obtained through pre-training is input, and the projection position and the speed estimation value of the mobile terminal output by the model are obtained; further calculating the projection position of the mobile terminal at the predicted time; obtaining the pre-coding of the base station and the receiving end according to the predicted projection position to realize beam prediction; the invention can greatly reduce the beam training overhead and the instruction issuing time delay in beam alignment and tracking, improve the frequency spectrum efficiency and obviously improve the performance.

Example 2:

as shown in fig. l and 2, in this embodiment, an intelligent beam prediction method for nonlinear trajectory is provided, which is applied in an MU-MIMO millimeter wave communication system, and includes the following steps:

as shown in fig. 3, a transmitting end of a base station (solid point Tx in fig. 3) transmits a set of pilot signals to each mobile terminal at equal time intervals, each mobile terminal performs nonlinear track motion, a receiving end of the mobile terminal receives the pilot signals transmitted by the transmitting end of the base station, and the mobile terminal feeds back the received pilot signals and measurement signals to the base station;

step S2, respectively calculating a mobile terminal projection position estimation value and a speed estimation value based on the received pilot signal and a mobile terminal projection position estimation value and a speed estimation value based on the measurement signal according to a parameter estimation method of probability theory and a pre-constructed mapping table about the mobile terminal projection position and the absolute path length, and fusing a neural network model according to data obtained by pre-training to obtain a mobile terminal projection position and speed output by the model;

consider a link level MU-MIMO millimeter wave communication system comprising a plurality of MMIC chips, one of which has N _t Root antenna and N _rf Base Station (BS) of a radio unit, and N _rf Each has N _r A root antenna and a Mobile Terminal (MT) of 1 radio unit. Wherein, the base station transmitting end and the mobile terminal receiving end both use Discrete Fourier Transform (DFT) codebooks for the analog precoder, that is, the base station transmitting end and the mobile terminal receiving end both use DFT codebooks

And

wherein A is _t，n Is an analog precoding vector of a base station transmitting end radio unit n,

for the ith column vector, A, of the DFT matrix at the transmitting end of the base station _t，n And

the subscript t in (a) denotes Transmitter, i.e. the Transmitter corresponding to the transmitting end of the base station,A _r is an analog precoding vector at the receiving end of the mobile terminal,

for the jth column vector, A, of the DFT matrix at the receiving end of the mobile terminal _r And

the subscript r in (1) indicates Receiver, i.e. corresponds to the Receiver at the receiving end of the mobile terminal. The antenna domain received signal for the u-th MT is modeled as:

y _u ＝H _u A _t Ds+n _u (1)

wherein u is more than or equal to 1 and less than or equal to N _rf ，H _u Is the channel matrix from the base station to the mobile terminal u, A _t Is an analog precoding vector at the transmitting end of the base station,

is the base station transmitting side digital precoding matrix, s is the baseband signal,

is the additive Gaussian noise of mobile terminal u, the subscript u indicating for the u-th mobile terminal.

Which represents the variance of the noise, is,

a unit array with dimension N;

for any mobile terminal, the millimeter wave channel H consists of K major paths, so the discrete-time narrowband channel matrix is:

wherein alpha is _k Is the complex gain of a path K ', K' is more than or equal to 1 and less than or equal to K,

respectively a horizontal arrival angle AOA of a receiving end path k 'and a departure angle AOD of a transmitting end path k';

are respectively related to the receiving end

Antenna response and transmitting end of

The calculation formula thereof refers to formula (3); superscript H represents the conjugate transpose of the matrix; phi is a ^r And phi ^t The Angle of Arrival (AOA) and Angle of Departure (AOD) are horizontal angles of Arrival (AOA), respectively.

For AOA and AOD, when the antenna spacing is half of the carrier wavelength, the antenna responses of the base station transmitting end and the mobile terminal receiving end are respectively:

wherein,

for the receiving end of a mobile terminal

The antenna response of (a) is to be,

for the transmitting end of the base station

The antenna response of (a) is determined,

for angle, j in formula (3) is an imaginary number, and formula (3) is suitable for calculation of two angles, namely AOA and AOD.

The invention is suitable for nonlinear track motion, taking a high-speed rail scene as an example, prior information under the high-speed rail scene is helpful for simplifying beam prediction, wherein the prior is summarized as:

(1) the channel always contains a direct path (LOS).

(2) The channel LOS power is much higher than the Non-LOS (LOS).

(3) The MT moves at a constant velocity along the non-linear rail track at a certain velocity v.

(4) LOS departure angle AOD phi ^t And its projection x on the x-axis is a bijection.

From apriori (1) and (2), the channel of equation (2) can be simplified to:

where α denotes the complex gain of the channel LOS and φ denotes the departure angle φ of the channel LOS ^t Thus, the channel may be described by a set of parameters α, φ.

And (4) estimating the projection position and the projection speed of the MT on the nonlinear rail through a parameterized nonlinear track motion model established in a priori (3) according to the received pilot signal and the measurement signal at each observation moment received by the base station. The number of observations was L and the observation time interval was Δ t. At each observation instant, the BS transmits all horizontal pilot beams. First, let the projection position of the first observation time be x _l The receiving pilot signal of the mobile terminal about the beam i is:

wherein, y _l，i For the received pilot signal, s, of beam i at observation time l _p Is a pilot sent by a base stationA vector of frequency symbols is generated by a frequency-domain vector,

is a base station transmitting end digital precoding matrix, alpha _l Complex gain at the channel LOS for the l-th observation instant, a _r (Φ(x _l ) -pi) is the antenna response of the receiving end of the mobile terminal, and is calculated by formula (3), wherein phi (x) _l ) For the projection position in x _l Calculating an arc tangent function, wherein the formula refers to formula (8); a is _t (Φ(x _l ) Is calculated by equation (3) for the base station transmitting end antenna response,

to define an equation, z _i (x _l ) For the l-th observation time with respect to the projection position x _l And the received pilot signal estimate of beam i, n _l，i Additive Gaussian noise for beam i at the l-th observation instant.

Constructing a parameterized nonlinear track motion model: as shown in fig. 3, when the rail trajectory is described as an arbitrary non-linear projected distance function f (·), and the motion priors (3) and (4) are still satisfied, the expression for the projected position of the mobile terminal (solid point Rx in fig. 3) is defined as:

F(x _l ，x)＝(l-L)vΔt (6)

wherein, F (x) _l X) is the projection position x and x of the mobile terminal at the first observation moment _l Relative path length, x, between _l The projection position of the mobile terminal at the observation time L, x the projection position of the last observation time (L = L) of the mobile terminal, v the mobile terminal speed, L the total number of observations, and Δ t the observation time interval. F (x) _l X) can in turn be defined as:

where f' (w) is the first derivative with respect to the argument projection position. f (w) is an unknown non-linear trajectory function, even if f (w) is known, usually x _l The closed-form solution of (a) is difficult to obtain. However, F (x) _l X) with respect to x _l Is monotonically decreasing, thus, x _l It can be obtained from the mapping table regarding the projected position of the mobile terminal and the absolute path length by a binary search.

Further, the method for constructing the mapping table includes:

Specifically, the method comprises the following steps: constructing a piecewise function fitting nonlinear track, which specifically comprises the following steps:

(1) The deterministic non-linear trajectory function f (w) is unknown. Thus, f (w) can be parameterized by a function g (w; Θ) _g ) Implementation, wherein theta _g Is a learnable parameter set. Learnable data set Θ _g Can be acquired by off-line geometric measurements such as aerial photography and satellite photography.

(2) A general multi-layer perceptron model can approximate an arbitrary function with arbitrary accuracy, but this requires a large number of redundant learnable parameters and the learning model lacks interpretability. Therefore, in the present invention, a piecewise function is constructed to fit the nonlinear trajectory function f (·), wherein the piecewise function is: parameterized function g (w; Θ) _g ) The method comprises the following steps:

where w is the argument of the projection position, N _w Is the number of segments, o is the segment number, w _o Is a piecewise function ofProjected position origin value, being an unlearable parameter, y _o Is w _o The corresponding projection distance is a learnable parameter,

is a learnable set of parameters;

wherein the rectangular window function is pi (w, w) _o ，w _o+1 ) Comprises the following steps:

f (w) support set w e [ -r [ ] _max ，r _max ]Is equally divided into N _w Segment wherein r _max Is the BS maximum service radius. Thus, w _o The expression of (a) is:

wherein,

is the distance interval. Function g (w; theta) _g ) Is continuous over its expanse.

(3) Training of piecewise functions, i.e. parameterized functions g (w; Θ) _g ) The training of (2) is implemented at the base station side, in the form of supervised learning. The training data set is acquired by off-line geometric measurements (e.g., aerial and satellite photography). The loss function is defined as:

L(Θ _g )＝||f _n (w)-g(w；Θ _g )||

wherein, f _n (w) is an observed value for f (w), indicating that the observation of f (w) is noisy. The value of f (w) is the projection distance corresponding to the projection position w, and the parameter set theta _g The update was iterated with MBGD (Small Batch Gradient Descences) until convergence. Due to the certainty of the non-linear trajectory, the function g (w; Θ) _g ) No wire trimming or periodic updating is required after the training deployment.

Constructing a mapping table about the projection position and the absolute path length of the mobile terminal according to the fitted piecewise function; from the mapping table, a projection position and a projection distance under the nonlinear trajectory are determined using a binary search.

The method comprises the steps of dividing projection position independent variables in a piecewise function at equal intervals, calculating absolute path length corresponding to each divided projection position of the mobile terminal, and recording the corresponding relation between the projection position of the mobile terminal and the corresponding absolute path length into a mapping table, wherein the specific steps are as follows:

using the piecewise function g (w; theta) to f (w) in the formula (7) _g ) Replacing; x is the number of _l By w _o＝1 Replacing; subdividing the projection position w of the mobile terminal to obtain a subdivided projection position set omega, wherein the element of the subdivided projection position set omega is marked as w'; the subdivided w' is used in place of x in equation (7); calculating to obtain the absolute path length corresponding to each w': f (x) _l ＝w _o＝1 X = w '), i.e. absolute path length, recording the correspondence between the projection position w' of the mobile terminal and the corresponding absolute path length into a mapping table;

Specifically, the method comprises the following steps:

when the projection position x of the mobile terminal and the speed v of the mobile terminal corresponding to the last observation time L are known, the projection position x of the mobile terminal at the ith observation time is obtained according to the mapping table _l The process of (2) is as follows:

firstly, searching for the absolute path length corresponding to the projection position x of the mobile terminal at the last observation moment from a mapping table through binary search, and recording the absolute path length as a first absolute path length F; next, the relative path length F (x) on the left side of formula (6) is obtained from formula (6) _l X), denoted as second relative path length F'; thirdly, adding the first absolute path length F and the second relative path length F ' to obtain a third absolute path length F ', wherein F ' is the projection position x of the mobile terminal at the ith observation moment _l A corresponding absolute path length; finally, according to the third absolute path length F', obtaining the projection position x of the mobile terminal at the l-th observation moment in the mapping table by binary search _l 。

The parameter estimation method of the probability theory adopted in the embodiment is a likelihood estimation method and a bayesian estimation method, and certainly, the parameter estimation may be performed based on only one of the estimation methods, and the present invention is not limited to the estimation method adopted in the embodiment.

Specifically, the method comprises the following steps:

note that v is the positive direction as MT moves from left to right. The arctangent function Φ is defined as:

equation (8) is a transcendental equation, and a numerical solution w can be obtained by a one-dimensional search.

When the prior function p (v) is related to the velocity _p ) When known, with respect to the received pilot signal y _l，i And a set of dependent variables Θ based on the received pilot signal _p The probability formula of (a) is:

wherein x is _p ，v _p Respectively, a projected position and velocity estimate, p (y), of the mobile terminal based on the received pilot signal _l，i ；Θ _p ) For receiving a pilot signal y _l，i A posterior probability of (x) _p，l For the projected position of the mobile terminal based on the received pilot signal at the l-th observation instant, p (v) _p ) Is velocity v _p A prior function of (a); sigma _n Is an additive noise standard deviation, alpha _l Complex gain of channel LOs for the l-th observation time, z _i (x _p，l ) Has been defined in equation (5), based on a set of dependent variables of the received pilot signal

The LOS complex gain set is a channel LOS complex gain set for L observations.

Thus, the overall probability formula for the received pilot signal is:

wherein equation (10) is also a probability function for the projected position and velocity of the mobile terminal based on the received pilot signal at the observation time, and is related to the projected position x _p A posteriori function of, also, the velocity v _p Bayes function of (a).

According to Bayes criterion(Bayesian criterion) to estimate the set of dependent variables Θ based on the received pilot signal _p Velocity v in _p And estimating the set of dependent variables Θ based on the received pilot signal using a maximum likelihood criterion _p Of the remaining parameters x _p And

alternate iterative estimation of Θ using the block-alternating-descent method _p All the parameters in (1) can obtain: a mobile terminal projection position and velocity estimate based on the received pilot signal; it should be noted that, in the present embodiment, the above-mentioned criterion and iterative method are used for estimation, but the present invention is not limited to the above-mentioned criterion and iterative method;

the initialization parameters are as follows:

wherein,

respectively obtaining a projection position initial value, a channel complex gain initial value and a speed initial value of the mobile terminal at an observation time l based on a received pilot signal;

to take on variable x _l The function of the maximum value of (c),

is y _l，i The conjugate transpose of (a) is performed,

is composed of

Conjugate transpose of (max) _v To take a function of the maximum value, alpha, of the variable mobile terminal speed v _l With closed expressions and with respect to x _p And v _p Is non-convex and the probability function of (a),one-dimensional search (minimum under euclidean distance criterion) is used for parameter optimization. The expression for the kth iteration is:

the parameters are iteratively updated in equation (12) until a convergence condition is satisfied.

The detailed estimation algorithm for the projected position and velocity of the mobile terminal based on the received pilot signal is as follows:

above { x _p ，v _p Projected position and velocity estimates based on the received pilot signal corresponding to the last observation instant.

The mobile terminal feeds back a measuring signal corresponding to the pilot signal to the base station, wherein the measuring signal comprises relative communication time delay and Doppler frequency; relative communication time delay tau of base station obtaining measuring signal _m And Doppler frequency f _d，m . At observation time i, the observed value is:

τ _m，l ＝τ _l +n _τ，l ， (13)

wherein n is _τ，l In order to observe additive Gaussian noise (i.e. measurement error noise of the ith observation time relative to the communication delay) of the relative communication delay at the ith observation time,

additive Gaussian noise of observed Doppler frequency for the ith observation time (i.e., measurement error noise of Doppler frequency at the ith observation time); tau. _m，l Measuring the relative communication time delay in the signal at the ith observation time, wherein the relative communication time delay is an observed value and is a known quantity; tau is _l The relative communication time delay (namely the linear distance between the base station and the mobile terminal divided by the light speed) of the l-th observation time is a true value; f. of _d，m，l Measuring the Doppler frequency in the signal for the l-th observation time, wherein the Doppler frequency is an observed value and is a known quantity; f. of _d，l The Doppler frequency at the l-th observation time is a true value; the subscript m represents the measurement,

is the measurement error noise of relative communication time delay with variance of

Is the measurement error noise of Doppler frequency with a variance of

The variance of the measurement error is modeled as:

where c is the speed of light, B is the bandwidth, f _c Is the carrier frequency, T _c Is the cumulative time of the day,

is the residual carrier frequency ratio.

Therefore, the projection position and velocity estimation problem is: given a received pilot signal

Estimating a set of parameters { x ] for the projection position and velocity of the MT at the observation time L _p ，v _p }; the given measurement signal includes a relative communication time delay

And Doppler frequency

Estimating a set of parameters { x ] for the projection position and velocity of the MT at the observation time L _m ，v _m }。

According to the geometrical relation between BS and MT, the projection position x of mobile terminal at the l observation time _l Set of variables { tau } _l ，f _d，l The method is as follows:

where the function Ψ is:

equation (16) is a transcendental equation. When x is _l When known, the formula (16) is related only to the positive or negative characteristic of v. X is to be _l And v is substituted into equation (15) to obtain: tau is _l And f _d，l 。

When p (v) _m ) When known (p (v) _m ) Is exactly p (v) _p ) Probability functions of the relative communication delay and Doppler frequency measurements at the ith observation time are:

p(τ _m，l ，v _m ；x _m ) Is based on τ _m，l P (f) is a probability function of _d，m，l ，v _m ；x _m ) Is based on f _d，m，l Of the probability function of (c) (-) _m ＝{x _m ，v _m Is a set of variables based on the measurement signal.

The overall probability function for a set of measurement signals is:

wherein the formula (18) is also a probability function of the projection position and velocity of the mobile terminal based on the measurement signal at the observation time, with respect to the projection position x _m A posteriori function of, also, the velocity v _m Bayes function of (a).

Under the nonlinear track, a dependent variable set theta based on a measurement signal _m The middle initialization parameters are as follows:

where sign (·) is a sign function. Updating the parameters by using a block coordinate alternative descent method, wherein the expression of the kth iteration is as follows:

the parameters are iteratively updated in equation (20) until a convergence condition is satisfied. The detailed mobile terminal projection position and velocity estimation algorithm based on the measurement signal is as follows:

above { x _m ，v _m Projected position and velocity estimates based on the measurement signal corresponding to the last observation instant.

The position network and the speed network have the same topological structure, the position network and the speed network both comprise a variance sub-network and a bias sub-network, and the variance sub-network and the bias sub-network are both neural network models comprising two hidden layers;

Specifically, the method comprises the following steps:

the parameter set estimated from the received pilot signal is { x } _p ，v _p An estimated set of parameters from the measurement signal is { x } _m ，v _m }。{x _p ，v _p And { x } _m ，v _m All are Neural Network (NN) models h (·;. Theta.) _h ) Wherein Θ is an input of _h ＝{Θ _x ，Θ _v The parameters of the data fusion neural network model are included in the position network h _x Model parameter theta of _x And speed network h _v Parameter theta of _v . Location network h _x And speed network h _v Have the same topology and each network consists of a variance sub-network and a bias sub-network. The variance sub-network comprises an input layer, two hidden layers and an output layer, and the bias sub-network comprises an input layer, two hidden layers and an output layer. The topology of the variance and bias subnetworks is shown in table 1,

wherein 'ReLU' is a modified Linear Unit (ReLU), 'BN' is Batch Normalization (BN), and 'Linear' is a Linear function. The number is the number of computational units of the layer. The expressions for the location network and the speed network are:

wherein,

respectively a variance based on the measurement signal projection position estimate and a variance based on the measurement signal velocity estimate,

the variance based on the received pilot signal projection position estimation and the variance based on the received pilot signal velocity estimation are respectively; b _x，p 、b _v，p Deviation of the projection position estimation based on the received pilot signal and deviation of the velocity estimation based on the received pilot signal, respectively, b _x，m 、b _v，m Respectively, the deviation estimated based on the projection position of the measurement signal and the deviation estimated based on the velocity of the measurement signal, theta _x 、Θ _v Are respectively a location network h _x Trainable parameters and speed network h of _v The trainable parameters of (a).

Network h according to said location _x And speed network h _v Set of variances of output

And a set of deviations

The final output mobile terminal projection position and speed of the data fusion neural network model are as follows:

wherein,

representing weights estimated based on the measured signal projection positions,

representing weights estimated based on the projected positions of the received pilot signals,

representing weights estimated based on the measured signal velocity,

representing weights estimated based on the received pilot signal velocity.

Compared with a general NN (neural network), the data fusion neural network model is light in weight, can effectively resist overfitting, and has good interpretability.

The training mode of the data fusion neural network model is as follows: data fusion neural network model h (·; Θ) _h ) The training of (2) is performed in a supervised learning manner, the input samples comprise estimated values of the projected position and velocity of the mobile terminal based on the received pilot signals and measurement signals, and the output samples are tags of the projected position and velocity of the mobile terminal. The loss function is defined as:

wherein the subscript (.) _tar Representing tag data, x _tar Indicating projected position tag, v _tar Indicating a speed tag. Parameter set Θ _h And (5) iteratively updating by a small batch gradient descent method until convergence.

And S3, calculating the projection position of the mobile terminal at the predicted moment according to the projection position and the projection speed of the mobile terminal output by the model and by combining the mapping table, wherein the calculation comprises the following steps:

adding the fourth absolute path length and the fifth relative path length to obtain a sixth absolute path length which is used as the absolute path length corresponding to the projection position of the mobile terminal at the prediction moment;

Specifically, the method comprises the following steps:

when nonlinear parallel rails are considered, the u th MT projection position x at the q th prediction moment is obtained according to the projection position and the speed estimation value of the mobile terminal output by the model _q，u The formula is:

F _y (x _u ，x _q，u )＝v _u (q-1)Δt _p (24)

wherein x is _u Projected position, v, of mobile terminal u output for model _u Velocity estimate, Δ t, for mobile terminal u output by the model _p Is a prediction time interval;

solving the projection position x in equation (24) by binary search according to the mapping table _q，u ；

Specifically, the method comprises the following steps:

knowing x _u And v _u ：

First, the projection position x of the mobile terminal u output by the model is searched from the mapping table by binary search _u The corresponding absolute path length is marked as a fourth absolute path length F;

next, the u-th MT projection position x at the q-th prediction time is obtained by the formula (24) _q，u Relative path length F _y (x _u ，x _q，u ) Denoted as a fifth relative path length F';

thirdly, adding the fourth absolute path length F and the fifth relative path length F ' to obtain a sixth absolute path length F ', wherein F ' is the projection position x of the mobile terminal at the qth prediction moment _q，u A corresponding absolute path length;

finally, according to the sixth absolute path length F', the projection position x of the mobile terminal at the qth prediction moment is obtained by binary search in the mapping table _q，u 。

Step S4, realizing beam prediction according to the projection position of the mobile terminal at the prediction moment and the projection position of the mobile terminal at the prediction moment, and comprising the following steps:

according to the projection position of the mobile terminal at the prediction moment, obtaining a departure angle of a channel LOS of a receiving end of the mobile terminal, and further obtaining the simulated pre-coding of a transmitting end of the base station and the receiving end of the mobile terminal;

Further, according to the projection position of the mobile terminal at the predicted time, the starting angle of a channel LOS of a receiving end of the mobile terminal is obtained, and then the simulated pre-coding of a transmitting end of the base station and the receiving end of the mobile terminal is obtained;

specifically, the method comprises the following steps:

according to formula (8), let x _q，u = w, obtained by _q，u And = Φ (w) is the departure angle AOD of the channel LOS at the u-th MT receiving end at the q-th prediction time.

Therefore, the analog precoding vectors of the u-th MT receiving end and the base station transmitting end are:

the analog precoding matrix of the base station transmitting end is

Analog precoding for all MT receivers

Further, according to the projection position of the mobile terminal at the predicted time and the starting angle of the LOS of the channel of the receiving end of the mobile terminal, a virtual channel is obtained, and according to the virtual channel, digital precoding of the transmitting end of the base station is obtained;

specifically, the method comprises the following steps:

at the qth prediction moment, the base station transmitting terminal digital precoding matrix D _q From N _rf A precoding vector component, i.e.

Wherein V _q，u A digital precoding vector for the mobile terminal u at the qth predicted time instant. Wherein, the digital precoding vector set is obtained by the following optimization problem:

wherein,

is the square of the vector 2-norm, P _t，max Is BS maximum transmission power, equivalent low dimensional channel

a _r，q，u For the antenna response (also analog precoding vector of the u-th MT receiving end) of the u-th MT receiving end at the q-th prediction time, H _q，u Channel matrix for the receiving end of the mobile terminal u at the qth predicted time, A _t，q Is the analog precoding matrix at the qth prediction time at the transmitting end of the base station. In long-term prediction, instantaneous channel state information (I-CSI) and even Statistical channel state information (I-CSI) are difficult to obtain. Thus, the present invention characterizes the actual channel by a virtual channel:

wherein, phi (x) _q，u ) To relate to x _q，u The function of formula (8), a is a formula of magnitude about the projection position, and an estimation expression can be obtained through a formula of path loss in 3gpp TR 38.901. The virtual low-dimensional equivalent channel is

According to Zero Forcing (ZF) precoder, the base station transmitting end digital precoding matrix is:

wherein,

is dimension of N _rf And (4) unit array.

In this embodiment, at the predicted time q, precoding a is simulated according to the base station transmitting end _t，q And base station digital precoding D _q The base station transmits signals with hybrid precoding (a combination of analog precoding and digital precoding). Meanwhile, the receiving end of the mobile terminal simulates precoding A according to the receiving end _r，q To receive the signal.

According to the base station transmitting end analog precoding and the base station transmitting end digital precoding, the base station transmits data signals by mixed precoding, wherein a base station radio frequency unit realizes digital precoding, a phase shifter of a base station antenna realizes analog precoding, and the radio frequency unit is fully connected with the phase shifter; meanwhile, the mobile terminal receives the data signal according to the mobile terminal receiving end analog precoding.

In the current single-user beam alignment and tracking scheme, the high-speed railway beam dwell period is 10ms, and the base station and the mobile terminal both scan 3 horizontal beams, thus 900 beams need to be scanned within 1 s. In the invention, one beam prediction period is 1.25s, the base station transmits 3 times of pilot signals in each period, each time the pilot signals are transmitted by 8 beams, and only 24 beams are scanned in 1.25 s. In a multi-user scenario, the beam overhead grows linearly with the number of users. First, the present invention helps to greatly reduce the beam training overhead, especially in a scenario of multiple mobile users. Secondly, the beam tracking needs to feed back the measured value to the base station side, which introduces about 25ms of instruction issue delay, and the invention avoids the instruction issue process through the beam prediction, so the delay is zero. In conclusion, the invention has smaller beam overhead and zero time delay in a multi-user mobile scene, thereby greatly improving the spectrum efficiency.

Example 3:

as shown in fig. 8, an intelligent beam prediction apparatus for nonlinear trajectory comprises:

the position and speed estimation module is used for respectively calculating a mobile terminal projection position estimation value and a speed estimation value based on the received pilot signal and the measurement signal according to a parameter estimation method of probability theory and a pre-constructed mapping table about the projection position and the absolute path length of the mobile terminal, and fusing a neural network model according to data obtained by pre-training to obtain the projection position and the speed of the mobile terminal output by the model;

according to a parameter estimation method of probability theory and the measuring signal, determining a probability function about the projection position and the speed of the mobile terminal based on the measuring signal, and combining the mapping table to iteratively calculate the projection position estimation value and the speed estimation value of the mobile terminal based on the measuring signal;

further, the method for constructing the mapping table includes:

obtaining a virtual channel according to the projection position of the mobile terminal at the predicted time and the departure angle of a channel LOS of a receiving end of the mobile terminal, and obtaining digital precoding of a transmitting end of the base station according to the virtual channel;

The invention is applied to a scene that the mobile terminal does nonlinear track motion, can greatly reduce the beam training overhead and the instruction issuing time delay in beam alignment and tracking, improves the frequency spectrum efficiency and obviously improves the performance.

Example 4:

an apparatus comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor when executing the program implementing a method for intelligent beam prediction for non-linear trajectories as described in any of the above. The memory can be various types of memory, such as random access memory, read only memory, flash memory, and the like. The processor may be various types of processors, such as a central processing unit, a microprocessor, a digital signal processor, or an image processor.

A computer-readable storage medium storing computer-executable instructions for performing a method for intelligent beam prediction for non-linear trajectories as described in any one of the above. The storage medium includes: various media capable of storing program codes, such as a U disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

Example 5:

the invention provides an intelligent beam prediction method for a nonlinear track, and in order to verify the performance advantage of the method, an example process of the invention is given below.

The millimeter wave channel model is 3GPP TR 38.901 UMa LOS. Base station carrier frequency f _c =30GHz, bandwidth B =80MHz, maximum transmission power P of base station _t，max =30dBm, 8 horizontal antennas, 4 radio frequency units and 200m distance between the base stations. The noise power spectral density was-174 dBm/Hz. There are 4 mobile terminals, each having 4 horizontal antennas and 1 radio frequency unit. The terminal speed follows Laplacian distribution, mean 256 (km/h) ² Variance 18 (km/h) ² . The nonlinear rail function is modeled as

Prediction of temporal granularity at _p =1.25ms, observation period 1.25s, observation interval Δ t =100ms, and observation number L =3. Cumulative time T _c =12.5ms, residual carrier fraction 1ppm.

In one embodiment, as shown in fig. 4, the measured location estimate is more accurate when the MT is far away from the BS, and less accurate when the MT is near the BS. This phenomenon is caused by the fact that the relative communication delay contains information on the distance between the BS and the MT, but it cannot be inferred that the MT is located on the right or left side of the BS. In addition, however, when the Doppler frequency is heavily contaminated by noise or the velocity component is small, the estimation performance cannot be improved. Therefore, as shown in fig. 4 and fig. 5, the pilot signal represents a method based on receiving the pilot signal, that is, only the pilot signal is used to obtain the projection position and velocity estimation value of the mobile terminal, and the beam prediction is performed by using the projection position and velocity estimation value; the measurement signal represents a method based on the measurement signal, namely, the measurement signal is only used for obtaining the projection position and speed estimated value of the mobile terminal, and the beam prediction is carried out according to the projection position and speed estimated value. When the MT is far away from the BS, neither the projected position nor velocity estimates based on the measurement signals are accurate. Meanwhile, as the BS and MT distances become smaller, the estimation based on the received pilot signal is more accurate. This is because the path loss becomes smaller and the SNR improves. In addition, the AOD of the MT is also easily distinguished within this range. The accuracy of the estimation drops dramatically when the MT moves away from the BS.

In general, when the MT is far from the BS, the estimation accuracy based on the measurement signal is higher than the estimation based on the received pilot signal, and when the MT is near the BS, the estimation accuracy based on the received pilot signal is higher than the estimation based on the measurement signal. The estimation method of the present invention has the highest accuracy in projection position and velocity estimation. The simulation result verifies the effectiveness of the method.

In the present embodiment, as shown in fig. 6 (a) and 6 (b), the estimation result of the nonlinear trajectory is obtained by using a linear estimation algorithm (assuming that the rail is linear), that is, the mapping table and the piecewise function are not used. First, the estimation results of the linear estimation method and the nonlinear estimation method of the present invention, i.e., the projection position estimation in fig. 4 and 6 (a), and the velocity estimation in fig. 5 and 6 (b), are compared. Simulation results show that the performance of the estimation method is remarkably improved, and the effectiveness of the mapping table and the piecewise function in a nonlinear rail scene is verified. The reason for the poor performance of the linear estimation method is that: for non-linear trajectories, the linear estimation method produces model mismatch that results in estimation bias. In addition, simulation results also show that the data fusion neural network model is obviously superior to a method based on a measurement signal and a method based on a received pilot signal. The data fusion neural network model can reduce deviation and improve estimation performance. Meanwhile, the simulation result verifies the effectiveness of the data fusion neural network model.

In this embodiment, as shown in fig. 7, a relationship between Spectral Efficiency (SE) and a projection position is shown in the figure. Among them, the beam prediction method based on I-CSI and S-CSI (which is different from the method of the present invention in the acquisition of the channel) is known as the upper bound of the performance of the inaccessible SE. The inventive method is superior in SE performance to methods based on measurement signals or pilot signals only. In addition, table 2 shows the average and rate performance (bps/Hz), N, of the different methods _rf The average sum rate of the MTs is listed in table 2.

In one embodiment, the beam training overhead of beam alignment/tracking (BA/T) in the prior art and beam prediction in the present invention grows linearly with the MT number, taking into account MT specific pilot signals. We define the overhead cost ratio as the proportion of overhead to time-frequency resources. As shown in table 3, the overhead ratio and average effective sum rate of different methods are given, compared with the BA/T algorithm, the beam prediction training overhead is close to zero, and when the MT number is greater than 10, the effective sum rate of the beam prediction is higher than BA/T. Further, the command issuance delay of BA/T is about 20ms, and the beam prediction delay is 0. The simulation result verifies the effectiveness of the beam prediction method.

The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.

Claims

1. An intelligent beam prediction method for a nonlinear trajectory, comprising the steps of:

acquiring a plurality of receiving pilot signals and measuring signals fed back by a mobile terminal at a plurality of observation moments, wherein the motion track of the mobile terminal is a nonlinear track;

calculating the projection position of the mobile terminal at the predicted moment according to the projection position and the projection speed of the mobile terminal output by the model and the mapping table;

realizing beam prediction according to the projection position of the mobile terminal at the prediction moment;

the construction method of the mapping table comprises the following steps: the method comprises the steps of fitting a nonlinear track by constructing a piecewise function, dividing projection position independent variables in the piecewise function at equal intervals, calculating absolute path length corresponding to each divided projection position of the mobile terminal, and recording the corresponding relation between the projection position of the mobile terminal and the corresponding absolute path length into a mapping table; the absolute path length is the integral path length of the mobile terminal on the nonlinear track by taking the initial projection position as a starting point;

the data fusion neural network model comprises a position network and a speed network; the position network is used for outputting the variance of the projection position estimation values in the two groups of estimation values and the deviation of the projection position estimation values; the speed network is used for outputting the variance of the speed estimation values and the deviation of the speed estimation values in the two groups of estimation values;

2. The method of claim i, wherein: the parameter estimation method of the probability theory comprises a likelihood estimation method and/or a Bayesian estimation method.

3. The intelligent beam prediction method for nonlinear trajectory according to claim 1, characterized by:

according to a parameter estimation method of probability theory and the received pilot signal, determining a probability function about the projection position and the speed of the mobile terminal based on the received pilot signal, and combining the mapping table to iteratively calculate the projection position estimation value and the speed estimation value of the mobile terminal based on the received pilot signal;

and determining a probability function about the projection position and the projection speed of the mobile terminal based on the measurement signal according to a parameter estimation method of probability theory and the measurement signal, and combining the mapping table to iteratively calculate the projection position estimation value and the projection speed estimation value of the mobile terminal based on the measurement signal.

4. The intelligent beam prediction method for nonlinear tracks in claim 3, wherein:

in both iterative calculations, including: obtaining the projection position of the mobile terminal at any other observation time except the last observation time according to the mapping table, wherein the method comprises the following steps:

5. The method of claim i, wherein:

the position network and the speed network both comprise a variance sub-network and a bias sub-network, and the variance sub-network and the bias sub-network are both neural network models comprising two hidden layers;

6. The intelligent beam prediction method for nonlinear trajectory according to claim 1, characterized by:

7. The intelligent beam prediction method for nonlinear trajectory according to claim 1, characterized by:

according to the projection position of the mobile terminal at the prediction moment, beam prediction is realized, and the method comprises the following steps:

8. An intelligent beam prediction apparatus for nonlinear trajectory, characterized by: the method comprises the following steps:

the position prediction module is used for calculating the projection position of the mobile terminal at the prediction moment according to the projection position and the projection speed of the mobile terminal output by the model and the mapping table;

the beam prediction module is used for realizing beam prediction according to the projection position of the mobile terminal at the prediction moment;

the construction method of the mapping table comprises the following steps:

the absolute path length is the integral path length of the mobile terminal on the nonlinear track by taking the initial projection position as a starting point;

9. An intelligent beam prediction device for a non-linear trajectory, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the intelligent beam prediction method for a non-linear trajectory according to any one of claims 1 to 7.

10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out a method for intelligent beam prediction for non-linear trajectories according to any one of claims 1 to 7.