阅读说明：本技术 一种基于路径分多址的上行链路信道预测方法及预测系统 (path division multiple access-based uplink channel prediction method and prediction system ) 是由 张川 冀贞昊 尤肖虎 于 2019-10-25 设计创作，主要内容包括：本发明提出了一种基于路径分多址的上行链路信道预测方法和预测系统。根据大规模天线阵列中存在的空间宽带效应和频率选择效应,提出了面向使用正交频分复用技术的大规模多输入多输出天线系统的新的信道预测VLSI架构,设计了上行信道链路中信道预测模块。对于导频阶段,本发明基于流水线和脉动阵列技术,设计了预处理、预搜索、用户分组和信道特征搜索等模块；对于上行信道预测阶段,本发明设计了每个用户的上行信道估计模块。所有模块只包含复数加法、复数乘法以及寄存器,不包含其他复杂运算模块。(The invention provides uplink channel prediction methods and prediction systems based on path division multiple access, according to the space broadband effect and frequency selection effect existing in a large-scale antenna array, a new channel prediction VLSI architecture facing a large-scale multiple-input multiple-output antenna system using an orthogonal frequency division multiplexing technology is provided, and a channel prediction module in an uplink channel is designed.)

1, A method for predicting uplink channel based on path division multiple access (TDMA), comprising:

(1) pilot phase initial prediction phase: the preprocessing module processes the received signal Yp of the p-th user to obtain an initial estimation channel matrix

(2) channel characteristic prediction grouping and channel reconstruction stage: the channel characteristic searching module is used for searching the channel characteristics

(3) and an uplink channel prediction stage: and the uplink channel prediction module outputs an uplink channel prediction matrix according to the received signal Yp and the channel base vector.

2. The path division multiple access-based uplink channel prediction method of claim 1, wherein: in the step (1), the preprocessing module obtains a base vector based on an LS channel estimation modeThe base vector

3. The path division multiple access-based uplink channel prediction method of claim 1, wherein: in the step (1), the pre-search module initially estimates the channel matrix

wherein the content of the first and second substances,

4. The path division multiple access-based uplink channel prediction method of claim 1, wherein: in the step (2), the channel characteristic searching module adopts a successive binary iteration method to search the channel characteristics

5. The method for uplink channel prediction based on path division multiple access according to claim 1, wherein in step (3), the uplink channel prediction module calculates the channel gain of each paths of user p

Wherein E is_pLimiting power for the p-th user, vec (Y)_p) A received signal that is converted to a vector input; according to the channel gain obtained by calculation, the channel vector of the user p is obtained by multiplying and summing the channel gain with the corresponding channel base vector group, and then the channel vector is adjusted to a matrix form to obtain the final uplink channel prediction matrix of the p-th user

6, uplink channel prediction system based on path division multiple access, which is characterized in that the system comprises a preprocessing module, a pre-searching module, a channel characteristic grouping module, a reconstruction module and an uplink channel prediction module;

the preprocessing module is used for processing the received signal Yp of the p-th user to obtain an initial estimation channel matrix

the pre-search module is used for estimating a channel matrix according to the initial estimation

the channel characteristic searching module is used for searching the channel characteristics

the channel characteristic grouping module is used for grouping channel characteristics and storing the grouped channel characteristics into a grouping register, and the grouped channel characteristics are used for guiding the uplink channel prediction module to predict;

the reconstruction module is used for reconstructing the two-dimensional channel characteristics to obtain a channel base vector group Pp, wherein the channel base vector group comprises channel base vectors of Lp paths of the p-th user

the uplink channel prediction module is configured to output an uplink channel prediction matrix according to the received signal Yp and the channel base vector.

7. The path division multiple access based uplink channel prediction system of claim 6, whereinThe method comprises the following steps: the preprocessing module obtains a base vector based on an LS channel estimation mode

8. The path division multiple access based uplink channel prediction system of claim 6, wherein: the pre-search module is used for initially estimating a channel matrix

wherein the content of the first and second substances,

9. The path division multiple access based uplink channel prediction system of claim 6, wherein: the channel characteristic searching module adopts a successive dichotomy iteration method to search the channel characteristics

10. The uplink channel prediction system of claim 6 wherein the uplink channel prediction module calculates the channel gain for each paths for user p

Wherein E is_pLimiting power for the p-th user, vec (Y)_p) A received signal that is converted to a vector input; according to the channel gain obtained by calculation, the channel vector of the user p is obtained by multiplying and summing the channel gain with the corresponding channel base vector group, and then the channel vector is adjusted to a matrix form to obtain the final uplink channel prediction matrix of the p-th user

Wherein the content of the first and second substances,

Technical Field

The invention belongs to the technical field of -generation wireless mobile communication, and relates to a method and a system for predicting uplink channels based on path division multiple access.

Background

In 2017, Hongxiang Xie, Feifei Gao et al propose a channel prediction algorithm based on angular division multiple access for a large-scale multiple-input multiple-output system in "a Unified Transmission diversity for TDD/FDD Massive MIMO Systems With Spatial Basis Expansion Model", and according to the angular division multiple access technology, the channel shows great sparsity in an angular domain, and further channel prediction can be realized by capturing non-zero elements of the channel; and simultaneously, according to the mutual benefit of the angles, a downlink channel characteristic acquisition mode suitable for a TDD system and an FDD system is provided.

In 2019, Xiaozhen Liu et al put forward a corresponding Efficient VLSI architecture for an Angle-Division Multiple Access technology in "Efficient Channel Estimator With Angle-Division Multiple Access", and realized the prediction of uplink and downlink channels of a large-scale MIMO system of a single carrier system.

With the development of modern mobile communication Systems, orthogonal Frequency division multiplexing technology is gradually called to be widely adopted by a communication system in 2018, Bolei Wang, Feifei Gao et al in Spatial-and Frequency-Wideband efficiency in Millimeter-Wave Massive MIMO Systems, and a path division multiplexing method based on a Spatial broadband effect and a Frequency selection effect is provided, so that feasibility of channel prediction based on path division multiplexing is theoretically proved.

Disclosure of Invention

The invention aims to provide uplink channel prediction methods based on path division multiple access.

The invention provides an uplink channel prediction method based on path division multiple access, which comprises the following steps,

(1) pilot phase initial prediction phase: the preprocessing module processes the received signal Yp of the p-th user to obtain an initial estimation channel matrix

The initial estimated channel matrix

Respectively inputting the data into a pre-searching module and a channel characteristic searching module; the pre-search module estimates a channel matrix based on an initial estimate

Performing an initial search to obtainChannel characteristics of the p-th userCharacterizing channels

Respectively inputting a channel characteristic searching module and a channel characteristic grouping module, wherein l is the l-th path of a user p;

(2) channel characteristic prediction grouping and channel reconstruction stage: the channel characteristic searching module is used for searching the channel characteristics

Carrying out accurate prediction to obtain the two-dimensional channel characteristics of the p-th user and outputting the two-dimensional channel characteristics to a reconstruction module; the channel characteristic grouping module is used for grouping channel characteristics and storing the grouped channel characteristics into a grouping register, and the grouped channel characteristics are used for guiding the uplink channel prediction module to predict; the reconstruction module reconstructs the two-dimensional channel characteristics to obtain a channel base vector group Pp, wherein the channel base vector group comprises channel base vectors of Lp paths of the p-th user

The channel base vector group is output to an uplink channel prediction module;

(3) and an uplink channel prediction stage: and the uplink channel prediction module outputs an uplink channel prediction matrix according to the received signal Yp and the channel base vector.

Preferably, in the step (1), the preprocessing module obtains a base vector based on an LS channel estimation methodThe base vector

Performing serial-to-parallel conversion to obtain an initial estimation channel matrix

Preferably, in the step (1), the pre-searching module performs pre-search on the initial estimated channel matrix

Performing IFFT operation to obtain a channel prediction matrix in an angle time delay transformation domain:

wherein the content of the first and second substances,

is the transposed conjugate of an M x M dimensional fourier transform matrix,

conjugate of N x N dimensional Fourier transform matrix; for transform domain matrix

Modulus is taken, and the maximum two-dimensional coordinates of each path of the user p are compared to be used as channel characteristics obtained by initial search

Preferably, in the step (2), the channel characteristic searching module adopts a successive binary iteration method to search the channel characteristics

Neighborhood search satisfaction

Is output as a two-dimensional channel characteristic, where Ψ_M，Ψ_NThe angle domain rotation factor and the time delay domain selection factor are respectively, and theta is a phase shift matrix.

Preferably, in the step (3), the uplink channel prediction module calculates a channel gain of each paths of the user p

Wherein E is_pLimiting power for the p-th user, vec (Y)_p) A received signal that is converted to a vector input; according to the channel gain obtained by calculation, the channel vector of the user p is obtained by multiplying and summing the channel gain with the corresponding channel base vector group, and then the channel vector is adjusted to a matrix form to obtain the final uplink channel prediction matrix of the p-th user

Wherein the content of the first and second substances,

and the path set corresponding to the p-th user.

The invention also provides uplink channel prediction systems based on path division multiple access, which comprises a preprocessing module, a pre-searching module, a channel characteristic grouping module, a reconstruction module and an uplink channel prediction module;

the preprocessing module is used for processing the received signal Yp of the p-th user to obtain an initial estimation channel matrix

Respectively inputting the data into a pre-searching module and a channel characteristic searching module;

the pre-search module is used for estimating a channel matrix according to the initial estimation

Initial search is carried out to obtain the channel characteristics of the p-th userCharacterizing channels

Respectively input channel characteristic search modeA block and channel characteristic grouping module, wherein l is the l-th path of a user p;

the channel characteristic searching module is used for searching the channel characteristics

Carrying out accurate prediction to obtain the two-dimensional channel characteristics of the p-th user and outputting the two-dimensional channel characteristics to a reconstruction module;

the channel characteristic grouping module is used for grouping channel characteristics and storing the grouped channel characteristics into a grouping register, and the grouped channel characteristics are used for guiding the uplink channel prediction module to predict;

the reconstruction module is used for reconstructing the two-dimensional channel characteristics to obtain a channel base vector group Pp, wherein the channel base vector group comprises channel base vectors of Lp paths of the p-th user

Outputting the channel base vector group to an uplink channel prediction module;

the uplink channel prediction module is configured to output an uplink channel prediction matrix according to the received signal Yp and the channel base vector.

Preferably, the preprocessing module obtains a base vector based on an LS channel estimation method

The base vector

Performing serial-to-parallel conversion to obtain an initial estimation channel matrix

Preferably, the pre-search module is adapted to initially estimate the channel matrixPerforming IFFT operation to obtain a channel prediction matrix in an angle time delay transformation domain:

wherein the content of the first and second substances,

is the transposed conjugate of an M x M dimensional fourier transform matrix,

conjugate of N x N dimensional Fourier transform matrix; for transform domain matrix

Modulus is taken, and the maximum two-dimensional coordinates of each path of the user p are compared to be used as channel characteristics obtained by initial search

Preferably, the channel characteristic searching module adopts a successive binary iteration method to search the channel characteristics

Neighborhood search satisfaction

Is output as a two-dimensional channel characteristic, where Ψ_M，Ψ_NThe angle domain rotation factor and the time delay domain selection factor are respectively, and theta is a phase shift matrix.

Preferably, the uplink channel prediction module calculates the channel gain of each paths of user p

Wherein E is_pLimiting power for the p-th user, vec (Y)_p) A received signal that is converted to a vector input; according to the calculated channel gain, the user p is obtained by multiplying and summing the corresponding channel base vector groupThe channel vector is adjusted to a matrix form to obtain a final uplink channel prediction matrix of the p-th user as

Wherein the content of the first and second substances,

and the path set corresponding to the p-th user.

The invention realizes the path division multiple access channel prediction mode on hardware, does not have a high-complexity operation module, and adopts a production line and a pulse array in the framework, thereby greatly improving the efficiency of the system.

Drawings

FIG. 1 is a schematic diagram of a channel preprocessing module;

FIG. 2 is a schematic diagram of a successive binary search;

FIG. 3 is a schematic diagram of a pre-search module;

FIG. 4 is a schematic diagram of a channel feature search module;

FIG. 5 is a schematic diagram of a channel feature grouping module;

FIG. 6 shows an uplink channel prediction module for the p-th user;

FIG. 7 is a block sub-module diagram;

fig. 8 is a schematic view of the overall framework of the present invention.

Detailed Description

The invention is further illustrated in below with reference to the following figures and specific examples.

The invention provides an uplink channel prediction VLSI (very large scale integrated) framework based on path division multiple access (FDMA) under the double broadband effect considering the space broadband effect and the frequency broadband effect by using an orthogonal frequency division multiplexing technology and facing a large-scale multi-input multi-output antenna system.

As can be seen from FIG. 1, the received input data Y_pThe data buffer is divided into two paths to enter an UL estimation module (namely an uplink channel prediction module) and an LS channel estimation module respectively, and the output of the LS channel estimation module

And the signal is output to a channel characteristic searching module and a pre-searching module in a channel matrix format after passing through a serial-parallel converter. In combination with the overall prediction block diagram of fig. 8, for the pilot stage, the present invention designs modules such as preprocessing, pre-search, user grouping, and channel feature search based on the pipeline and systolic array technology; for the uplink channel prediction stage, the invention designs an uplink channel estimation module of each user, and all the modules only comprise complex addition, complex multiplication and registers and do not comprise other complex operation modules. The signals received from the 1 st user to the p-th user are M x N dimensional signals Y_pWherein M is the number of transmitting antennas, and N is the number of subcarriers of OFDM. First, Imad Barhumi et al designs a channel matrix base vector obtained by initial estimation based on an LS channel estimation method in the document "Optimal tracking Design for MIMO OFDM systems in Mobile Wireless Channels" (journal name IEEE Trans Signal Processing, published 2003)By combining the obtained basis vectors

Performing serial-to-parallel conversion to obtain corresponding initial estimation channel matrix

Thus, the channel matrix will be initially estimatedThe data is input into a pre-searching module to search the transform domain channel characteristics, and is output to a channel characteristic searching module as input data stream data to search the channel characteristics accurately.

As shown in fig. 3, the channel matrix of M × N dimensions obtained by the preprocessing moduleInput to pre-search moduleIn the block, an initial estimation value of the channel characteristics is obtained through the processing of the pre-searching module, and finally, the initial channel characteristics are output to the channel characteristic grouping module to carry out channel characteristic grouping and the channel characteristic searching module to carry out accurate searching.

When initially estimating the channel matrix

When the signal is input into a pre-search module, IFFT operation is firstly carried out to obtain a channel prediction matrix in an angle time delay transformation domain:

wherein the content of the first and second substances,is the transposed conjugate of an M x M dimensional fourier transform matrix,

is the conjugate of an N x N dimensional fourier transform matrix.

In order to reduce the complexity of hardware implementation, the invention firstly refers to an FFT architecture to perform IFFT operation on the rows respectively, secondly transposes an output matrix, inputs the transposed output matrix into an original IFFT module to perform row-column IFFT operation, and finally outputs a transform domain matrix

Subsequently, the transform domain matrix is pair by the modulus extraction module

Modulus is taken, and the maximum two-dimensional coordinates of each path of each user are compared to be used as channel characteristics obtained by initial search

l is the l-th path of the p-th user,

as an initial advance of the angular domainThe value of the measured value is measured,

is an initial predicted value of a time delay domain, and takes a channel characteristic set corresponding to each users as a channel characteristic set

Next, the channel characteristics are assembled

Respectively input into a channel characteristic searching module for accurate searching and a channel characteristic grouping module for channel characteristic grouping.

In view of the fact that two-dimensional search with fixed step length is in inverse proportion to the actual search time and step length, the invention adopts successive dichotomy feedback search based on continuously iterating every layers of the search tree to achieve higher precision search, as shown in fig. 4, in a channel feature search module, a corresponding VLSI structure is proposed for a successive dichotomy search method, and a feedback loop is controlled to continuously circulate, and search node data stored in a register 1 and a register 2 and a current step length are adjusted, in this module, ζ submodules exist, points required to be searched in this layer are searched for, in each submodule, firstly, a phase shift matrix Θ is multiplied by ( times of iterative information stored in the register) in an initial estimation channel matrix, and then the two-dimensional rotation matrix is output as a maximum iterative channel matrix, and the two-dimensional rotation information is finally output as a corresponding to a corresponding iterative channel feature output in a register , and after the initial estimation channel matrix is output as a maximum iterative information output from a register 6335, and a final iterative algorithm is performed.

Channel feature set

After being input into the channel characteristic prediction module, the channel characteristic prediction module needs to perform accurate prediction, and in combination with the iterative process of fig. 2, layer-by-layer search needs to be performed, and search is performed near the original coordinates (the obtained channel characteristic set) to satisfy the requirements

Where Ψ is_M，Ψ_NThe angle domain rotation factor and the time delay domain selection factor are respectively, and theta is a phase shift matrix. Through the successive binary iteration and the iteration hardware architecture corresponding to the successive binary iteration, the accurate channel two-dimensional set which meets the precision of the pth user can be finally obtained

As shown in FIG. 2, the searching is gradually accurate from the root node, the node bearing Pre-0 is relayed in th level, and the maximum node Max-1 in the level, the Max-1 node becomes the root node, and further the next levels are derived, each level root node derives three child nodes, wherein each child node is the average of the root node value of the upper level and the adjacent node, the calculated range is controlled in two generated nodes in each level, then the original comparison derived node and the root node are only controlled in the searching range, the searching accuracy is improved by 50% in each level from level 1 to level n, and the searching point number is greatly reduced compared with the searching in fixed length under the condition of ensuring the accuracy.

On the other hand, the channel characteristics which are accurately searched and simultaneously grouped with the channel characteristics are input into a channel characteristic grouping module, for the case of multi-user parallel input, parallel grouping is needed, firstly, sorting is carried out according to -dimensional expansion coordinates of each user, existing strategies such as parallel sorting network sorting or serial bubble sorting can be adopted according to parallel and serial input, channel information of the users after sorting is input into the channel characteristic grouping module in parallel through a designed grouping pulse array, the Euclidean distance of each path coordinate is compared with a set threshold value, and the corresponding users are output at a corresponding processing module, as shown in FIG. 5, the channel characteristics of each user are input into the channel characteristic grouping module in parallel and pass through the sorting network and the grouping module, wherein the grouping module is composed of pulse arrays composed of n grouping submodules, each submodule corresponds to the output of 5 groups, meanwhile, the grouping information of each user is stored in a grouping information register of every users, an uplink channel prediction module is guided to perform grouping prediction through a training sub-module composed of n grouping modules, and training sub-module is used for training, and the grouping information of the channel characteristics is output of n channels, and the grouping modules, and the number of the grouping modules are set to be more than n, and the threshold value, the threshold value of the number of the output of the channel characteristics.

After the initially estimated channel characteristics are input to the channel characteristic grouping module, parallel ordering is performed through an ordering network, the ordered user characteristics are input to a grouping systolic array as shown in fig. 5 and 7, in each systolic array sub-modules, the geometric distance between the coordinates of the input user channel characteristics is calculated, different users are output in different groups according to the result of comparison with a grouping threshold value Ω, grouping of the channel characteristics is further realized, grouping information is stored into each user channel characteristic grouping registers, the channel characteristics stored in the registers guide an uplink channel prediction module at a lower stage to perform grouping training, and the UL estimator in fig. 7 is the uplink channel prediction module.

The channel information accurately searched by the channel characteristic searching module enters a channel reconstruction module to obtain a reconstructed channel base vector group Pp of the p-th user, wherein the base vector group Pp comprises channel base vectors of Lp paths of the p-th user

For each paths channel basis vector:

wherein, b_p，lIs a frequency domain direction vector, a_p，lIs an angle direction matrix and reflects the channel characteristics. The set of channel basis vectors will then be correlated with the received signal Y_pThe grouping information arranges the user grouping modules of groups to for training, and can adjust the pilot training set, because the user characteristics in the same groups are not overlapped, the pilot sequence can be shared, thereby improving the training speed and adapting to the fast-changing channel.

As shown in fig. 6, each user will use uplink channel prediction modules by inputting the corresponding received signal Y_pAnd channel base vector group Pp reconstructed by the channel, and outputting the final channel prediction matrix of the uplink of the p-th user, that is, inputting signal Y_pSerial conversion is carried out to convert the expanded signal vec (Y)_p) Multiplying the channel gain of the ith path of the pth user by the transposed conjugate of the channel basis vector of each paths in the channel basis vector group Pp of the pth user to obtain the channel gain of the ith path of the pth user, then multiplying the channel gain by the corresponding channel basis vector to obtain the channel vector of the ith path of the pth user, adding Lp paths, separating and outputting the vectors to finally obtain the pth userFinal uplink channel prediction matrix for users

In the up link channel prediction module, firstly, the channel gain of each paths of every users is predicted and calculated

Wherein E is_pLimiting power for the p-th user, vec (Y)_p) Is a received signal that is converted to a vector input.

According to the calculated channel gain, the channel vector of each user is obtained by multiplying and summing the channel gain with the corresponding channel base vector group, and then the channel vector is readjusted to be in a matrix form (represented by shape) to obtain the final uplink channel prediction matrix of the p-th user as

Wherein the content of the first and second substances,

and the path set corresponding to the p-th user.

The invention designs VLSI architectures suitable for path division multiple access according to a path division multiple access mode based on a large-scale multiple-input multiple-output antenna system and combining with the current mainstream orthogonal frequency division multiplexing technology, and can efficiently realize uplink channel prediction.A channel prediction architecture suitable for the path division multiple access is designed for the first time, the path division multiple access considers the joint time delay and angle domain to perform channel prediction, and channel reconstruction is performed by accurately predicting parameters such as the direction of arrival, path gain and the like.