阅读说明：本技术 一种基于并行深度学习网络的ofdm信道预测方法和系统 (OFDM channel prediction method and system based on parallel deep learning network ) 是由 何怡刚 隋永波 王枭 黄源 何鎏璐 程彤彤 于 2021-07-19 设计创作，主要内容包括：本发明公开了一种基于并行深度学习网络的OFDM无线通信系统信道预测方法和系统,属于无线通信自适应传输技术领域。通过并行深度学习网络对导频OFDM符号子载波的信道状态信息进行训练,实现导频OFDM符号信道状态信息的有效预测。为了提高对输入信道状态信息的泛化能力,提供一种包含数个平行的网络单元的并行深度学习网络的信道预测器,每个网络包含多个网络层。为了引入稀疏性,提供一种群组前向变量选择的输出权重估计方法。本发明公开的并行深度学习的信道预测器对导频子载波的信道状态信息具有良好的泛化能力,可以输出稀疏的输出权重矩阵,可以很好地实现OFDM信道预测,可以为自适应OFDM无线通信的自适应传输和自适应编码等提供保障。(The invention discloses a channel prediction method and a channel prediction system of an OFDM (orthogonal frequency division multiplexing) wireless communication system based on a parallel deep learning network, belonging to the technical field of wireless communication self-adaptive transmission. And training the channel state information of the pilot frequency OFDM symbol sub-carrier through a parallel deep learning network to realize effective prediction of the channel state information of the pilot frequency OFDM symbol. In order to improve generalization capability of input channel state information, a channel predictor of a parallel deep learning network comprising a plurality of parallel network elements is provided, each network comprising a plurality of network layers. In order to introduce sparsity, an output weight estimation method for group forward variable selection is provided. The channel predictor for parallel deep learning disclosed by the invention has good generalization capability on channel state information of pilot subcarriers, can output sparse output weight matrixes, can well realize OFDM channel prediction, and can provide guarantee for self-adaptive transmission, self-adaptive coding and the like of self-adaptive OFDM wireless communication.)

1. An OFDM channel prediction method based on a parallel deep learning network is characterized by comprising the following steps:

acquiring channel state information of a pilot frequency OFDM symbol through channel estimation;

training the parallel deep learning network by using the obtained channel state information of the pilot frequency OFDM symbol to obtain a well-trained parallel deep learning network predictor;

predicting the channel state information of the next sampling point of the pilot frequency subcarrier by using the well-trained parallel deep learning network predictor;

and merging the predicted channel state information of the next sampling point of each pilot frequency subcarrier to obtain the predicted channel state information of the pilot frequency OFDM symbol.

2. The method of claim 1, wherein the parallel deep learning network comprises a plurality of parallel network elements, each network element comprising a plurality of network layers.

3. The method of claim 2, wherein the training the parallel deep learning network by using the obtained channel state information of the pilot OFDM symbols comprises:

by passingObtaining the output of the 1 st network layer in the nth network unitWhere Γ (·) is the activation function, X is the input channel state information matrix,inputting a weight matrix for each network unit, wherein N is the number of the network units of the parallel deep learning network;

by passingObtaining the output of the q network layer in the n network unit For the output of the q-1 network layer in the nth network element,the connection weight of a Q-1 network layer and a Q network layer in an nth network unit is obtained, wherein Q is 1,2,3, and Q-1, and Q is the number of network layers contained in each network unit;

the output matrix of the nth network unit is formed by the output of each network layer of the nth network unitByObtaining the output matrix of each network unit, and estimating the output weight matrix W of the parallel deep learning network by a group forward variable selection method_out。

4. The method of claim 3, wherein estimating the output weight matrix W of the parallel deep learning network by group forward variable selection_outThe method comprises the following steps:

byObtaining the variable correlation of the first iterationWherein the content of the first and second substances,for the number of variables to be screened, Y is the output of the parallel deep learning network corresponding to the training data set, (. C)^TRepresenting the transpose operation of the matrix,for the output weight matrix estimated in the l-th iteration,andrepresenting data matrices with active and inactive data sets, respectively, the active data setInactive data setAnd is The number of neurons of the Q network layer in the nth network unit;

by passingObtaining an output weight matrix estimated in the first iterationLambda is a regularization coefficient, and I is an identity matrix;

correlation of variables in obtaining the first iterationThen, willFrom a data setIs removed and added to the data setPerforming the following steps;

throughAfter the sub-iteration, obtaining the output weight matrix of each iteration FromIn-process screening of optimal estimated output weight matrix

5. An OFDM channel prediction system based on a parallel deep learning network, comprising:

the channel estimation module is used for acquiring the channel state information of the pilot frequency OFDM symbol through channel estimation;

the network training module is used for training the parallel deep learning network by utilizing the acquired channel state information of the pilot frequency OFDM symbol to obtain a well-trained parallel deep learning network predictor;

the output weight matrix estimation module is used for estimating an output weight matrix of the parallel deep learning network by utilizing a group forward variable selection method;

the online prediction module is used for predicting the channel state information of the next sampling point of the pilot frequency subcarrier by the well-trained parallel deep learning network predictor; and merging the predicted channel state information of the next sampling point of each pilot frequency subcarrier to obtain the predicted channel state information of the pilot frequency OFDM symbol.

6. The system of claim 5, wherein the parallel deep learning network comprises a plurality of parallel network elements, each network element comprising a plurality of network layers.

7. The system of claim 6, wherein the network training module comprises:

a first output unit for passingObtaining the output of the 1 st network layer in the nth network unitWhere Γ (·) is the activation function, X is the input channel state information matrix,inputting a weight matrix for each network unit, wherein N is the number of the network units of the parallel deep learning network;

a second output unit for passingObtaining the output of the q network layer in the n network unitFor the output of the q-1 network layer in the nth network element,the connection weight of a Q-1 network layer and a Q network layer in an nth network unit is obtained, wherein Q is 1,2,3, and Q-1, and Q is the number of network layers contained in each network unit;

a third output unit for forming an output matrix of the nth network unit from outputs of the network layers of the nth network unitByObtaining an output matrix of each network unit;

an output weight matrix estimation module for estimating an output weight matrix W of the parallel deep learning network by a group forward variable selection method_out。

8. The system of claim 7, wherein the output weight matrix estimation module comprises:

a variable correlation calculation unit for calculating correlation between variablesObtaining the variable correlation of the first iterationWherein the content of the first and second substances,for the number of variables to be screened, Y is the output of the parallel deep learning network corresponding to the training data set, (. C)^TRepresenting the transpose operation of the matrix,andrepresenting data matrices with active and inactive data sets, respectively, the active data setInactive data setAnd is The number of neurons of the Q network layer in the nth network unit;

an output weight estimation unit for passingObtaining an output weight matrix estimated in the first iterationLambda is a regularization coefficient, and I is an identity matrix;

an output weight selection unit for obtaining the variable correlation of the first iterationThen, willFrom a data setIs removed and added to the data setPerforming the following steps; throughAfter the sub-iteration, obtaining the output weight matrix of each iteration FromIn-process screening of optimal estimated output weight matrix

9. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 4.

Technical Field

The invention belongs to the technical field of wireless communication adaptive transmission, and particularly relates to an OFDM channel prediction method and an OFDM channel prediction system based on a parallel deep learning network.

Background

With the development of technology, adaptive transmission of wireless communication is a trend in future development. In an adaptive Orthogonal Frequency Division Multiplexing (OFDM) system, channel information fed back from a receiving end to a transmitting end is easily out of date due to a rapid change of a fading channel. Channel prediction is a necessary technique to support adaptive transmission in OFDM systems.

Deep learning is a very efficient method of machine learning. In the field of wireless communication, a deep learning method has been applied to channel prediction, and an invention patent application with publication number CN112737987A discloses a time domain channel state information predictor based on deep learning. In the invention, the applicant trains and predicts the time domain channel state information of the wireless communication system by using a single deep learning network, and the generalization capability of the single deep learning network to the time domain channel state information is limited. Therefore, how to improve the deep learning network and provide a more effective OFDM channel predictor has great significance to the development of the adaptive OFDM.

Disclosure of Invention

In view of the above defects or improvement requirements of the prior art, the present invention provides an OFDM channel prediction method and system based on a parallel deep learning network, thereby solving the technical problem of how to provide a more effective channel prediction for an OFDM system based on deep learning.

To achieve the above object, according to an aspect of the present invention, there is provided an OFDM channel prediction method based on a parallel deep learning network, including:

acquiring channel state information of a pilot frequency OFDM symbol through channel estimation;

training the parallel deep learning network by using the obtained channel state information of the pilot frequency OFDM symbol to obtain a well-trained parallel deep learning network predictor;

predicting the channel state information of the next sampling point of the pilot frequency subcarrier by using the well-trained parallel deep learning network predictor;

and merging the predicted channel state information of the next sampling point of each pilot frequency subcarrier to obtain the predicted channel state information of the pilot frequency OFDM symbol.

In some optional embodiments, the parallel deep learning network comprises a plurality of parallel network elements, each network element comprising a plurality of network layers.

In some optional embodiments, the training the parallel deep learning network by using the obtained channel state information of the pilot OFDM symbol includes:

by passingObtaining the output of the 1 st network layer in the nth network unitWhere Γ (·) is the activation function, X is the input channel state information matrix,inputting a weight matrix for each network unit, wherein N is the number of the network units of the parallel deep learning network;

by passingObtaining the output of the q network layer in the n network unit For the output of the q-1 network layer in the nth network element,the connection weight of a Q-1 network layer and a Q network layer in an nth network unit is obtained, wherein Q is 1,2,3, and Q-1, and Q is the number of network layers contained in each network unit;

the output matrix of the nth network unit is formed by the output of each network layer of the nth network unitByObtaining the output matrix of each network unit, and estimating the output weight matrix W of the parallel deep learning network by a group forward variable selection method_out。

In some optional implementations, the estimating the output weight matrix W of the parallel deep learning network by the group forward variable selection method_outThe method comprises the following steps:

byObtaining the variable correlation of the first iterationWherein the content of the first and second substances,for the number of variables to be screened, Y is the output of the parallel deep learning network corresponding to the training data set, (. C)^TRepresenting the transpose operation of the matrix,andrespectively representing a plurality of data with activation dataData matrix of a set and an inactive data set, an active data setInactive data setAnd is The number of neurons of the Q network layer in the nth network unit;

by passingObtaining an output weight matrix estimated in the first iterationLambda is a regularization coefficient, and I is an identity matrix;

correlation of variables in obtaining the first iterationThen, willFrom a data setIs removed and added to the data setPerforming the following steps;

throughAfter the sub-iteration, obtaining the output weight matrix of each iteration FromIn-process screening of optimal estimated output weight matrix

According to another aspect of the present invention, there is provided an OFDM channel prediction system based on a parallel deep learning network, including:

the channel estimation module is used for acquiring the channel state information of the pilot frequency OFDM symbol through channel estimation;

the network training module is used for training the parallel deep learning network by utilizing the acquired channel state information of the pilot frequency OFDM symbol to obtain a well-trained parallel deep learning network predictor;

the output weight matrix estimation module is used for estimating an output weight matrix of the parallel deep learning network by utilizing a group forward variable selection method;

the online prediction module is used for predicting the channel state information of the next sampling point of the pilot frequency subcarrier by the well-trained parallel deep learning network predictor; and merging the predicted channel state information of the next sampling point of each pilot frequency subcarrier to obtain the predicted channel state information of the pilot frequency OFDM symbol.

In some optional embodiments, the parallel deep learning network comprises a plurality of parallel network elements, each network element comprising a plurality of network layers.

In some optional embodiments, the network training module comprises:

a first output unit for passingObtaining the output of the 1 st network layer in the nth network unitWhere Γ (·) is the activation function, X is the input channel state information matrix,inputting a weight matrix for each network unit, wherein N is the number of the network units of the parallel deep learning network;

a second output unit for passingObtaining the output of the q network layer in the n network unitFor the output of the q-1 network layer in the nth network element,the connection weight of a Q-1 network layer and a Q network layer in an nth network unit is obtained, wherein Q is 1,2,3, and Q-1, and Q is the number of network layers contained in each network unit;

a third output unit for forming an output matrix of the nth network unit from outputs of the network layers of the nth network unitByObtaining an output matrix of each network unit;

an output weight matrix estimation module for estimating an output weight matrix W of the parallel deep learning network by a group forward variable selection method_out。

In some optional embodiments, the output weight matrix estimation module comprises:

a variable correlation calculation unit for calculating correlation between variablesObtaining the variable correlation of the first iterationWherein the content of the first and second substances,for the number of variables to be screened, Y is the output of the parallel deep learning network corresponding to the training data set, (. C)^TRepresenting the transpose operation of the matrix,andrepresenting data matrices with active and inactive data sets, respectively, the active data setInactive data setAnd isThe number of neurons of the Q network layer in the nth network unit;

an output weight estimation unit for passingObtaining an output weight matrix estimated in the first iterationLambda is a regularization coefficient, and I is an identity matrix;

an output weight selection unit for obtaining the variable correlation of the first iterationAfter that, the air conditioner is started to work,will be provided withFrom a data setIs removed and added to the data setPerforming the following steps; throughAfter the sub-iteration, obtaining the output weight matrix of each iterationFromIn-process screening of optimal estimated output weight matrix

According to another aspect of the invention, a computer-readable storage medium is provided, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method of any of the above.

In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:

the OFDM wireless communication system channel prediction method provided by the invention can utilize the learning network layer in each network unit to realize the characteristic extraction of the output channel state information; in addition, the OFDM wireless communication system channel prediction method provided by the invention can screen the output variables by utilizing a group forward variable selection method, and further introduces sparsity to the output weight matrix. Therefore, the method has good prediction performance and can generate a sparse output weight solution. The method lays a foundation for future adaptive communication technologies such as adaptive coding, adaptive modulation, adaptive prediction and the like.

Drawings

Fig. 1 is a schematic structural diagram of a channel prediction system of an OFDM wireless communication system based on a parallel deep learning network according to an embodiment of the present invention;

fig. 2 is a schematic flowchart of a channel prediction method of an OFDM wireless communication system based on a parallel deep learning network according to an embodiment of the present invention;

fig. 3 shows the performance of the channel prediction method of the OFDM wireless communication system based on the parallel deep learning network in different signal-to-noise ratios according to the embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

In the present examples, "first", "second", etc. are used for distinguishing different objects, and are not used for describing a specific order or sequence.

Example one

Fig. 1 is a schematic structural diagram of a channel prediction system of an OFDM wireless communication system based on a parallel deep learning network according to an embodiment of the present invention, including:

a receiving antenna for receiving a wireless signal;

the network analyzer is used for analyzing the wireless signals obtained by the receiving antenna;

the channel estimation module is used for channel estimation to obtain the channel state information of the pilot frequency OFDM symbol;

the channel estimation module may perform channel estimation on channel state information of pilot subcarriers in the OFDM wireless communication system by using some channel estimation algorithms, such as an LS method and an MMSE method, to obtain channel state information of pilot OFDM symbols, which is used to train a channel predictor of a parallel deep learning network.

The parameter initialization module is used for initializing relevant parameters of the parallel deep learning network;

the network training module is used for training the parallel deep learning network by utilizing the acquired channel state information of the pilot frequency OFDM symbol;

the output weight matrix estimation module is used for estimating an output weight matrix of the parallel deep learning network by utilizing a group forward variable selection method;

and the online prediction module is used for predicting the trained parallel deep learning network on the channel state information of the pilot frequency subcarrier.

Fig. 2 is a schematic flow chart of an OFDM channel prediction method based on a parallel deep learning network according to an embodiment of the present invention, where the method includes the following steps:

step S1: acquiring channel state information H of a pilot frequency OFDM symbol through channel estimation;

step S2: defining a tap label i as 1;

step S3: initializing related parameters of the parallel deep learning network, such as the number N of network units of the parallel deep learning network, wherein each network unit comprises Q network layers, and the number of neurons of the Q network layer in the nth network unit isThe number of neurons contained in the output layer is M, and the input weight matrix of each network unit is initialized randomlyInitializing connection weights of a q-th network layer and a q + 1-th network layer in an nth network unit

Step S4: training the parallel deep learning network by using the channel state information of the ith pilot frequency subcarrier;

in the embodiment of the present invention, step S4 may be specifically implemented by:

step S41: calculating an output of a 1 st network layer in an nth network unit by the following formula (1)Namely:

where Γ (·) is an activation function, such as a hyperbolic tangent function tanh, and X is an input channel state information matrix.

Step S42: calculating an output of a q-th network layer in the n-th network unit by the following formula (2)Namely:

wherein the content of the first and second substances,for the output of the q-1 network layer in the nth network element,the connection weight of the (q-1) th network layer and the (q) th network layer in the nth network unit is given;

step S43: combining the outputs of the network layers of the nth network element obtained in the steps S41 and S42 into an output matrix of the nth network elementTo collect the output matrices of the individual network elements, i.e.:

step S5: estimating an output weight matrix W using a group forward variable selection method_out；

In the embodiment of the present invention, step S5 may be specifically implemented by:

step S51: defining activation data setsInactive data setAnd is

Step S52: the variable correlation for the l iteration is calculated by:

wherein the content of the first and second substances,and Y is the output of the parallel deep learning network corresponding to the training data set (namely the channel state information of the acquired pilot OFDM symbols) for the number of variables needing to be screened. Here, Y is specifically defined as channel state information of the pilot subcarriers during the training process, and the output is the output of the parallel deep learning network, that is, the prediction target corresponding to the training set. (.)^TRepresenting the transpose operation of the matrix,andrepresenting a data matrix with an active data set and an inactive data set, respectively.For output weight matrices estimated in the first iteration, i.e.

Wherein, λ is a regularization coefficient, and I is an identity matrix;

step S53: after calculating the variable correlation of the first iteration, it willFrom a data setIs removed and added to the data setPerforming the following steps;

step S54: throughAfter the iteration, obtaining an output weight matrix of each iteration, namely:

step S55: based onScreening optimal estimation output weight matrix by utilizing Akaike information standard

Step S6: predicting the channel state information of the next sampling point of the ith pilot frequency subcarrier by using the trained parallel deep learning network channel predictor;

step S7: judging whether channel prediction of all subcarriers is finished or not, namely judging whether i is larger than K, wherein K is the number of pilot subcarriers; if not, then i equals i +1 and jumps back to step S3;

step S8: and outputting the channel state information of the predicted pilot OFDM symbol.

In order to verify the effectiveness of the invention, the related parameters of the pilot frequency OFDM symbols are set according to the 2MHz mode in the IEEE802.11ah standard, and a single-transmitting single-receiving antenna is set. Fig. 3 is a comparison of predicted performance at different signal-to-noise ratios. Among them, invention 1 is "a channel prediction system and method for an OFDM wireless communication system" (publication No. CN 110995379A). It can be seen that the channel prediction method disclosed by the invention has better performance.

In another embodiment of the present invention, there is also provided a computer readable storage medium, on which program instructions are stored, which program instructions, when executed by a processor, implement the channel prediction method for an OFDM-oriented wireless communication system as described above.

The above-described method according to the present invention can be implemented in hardware, firmware, or as software or computer code storable in a recording medium such as a CD ROM, a RAM, a floppy disk, a hard disk, or a magneto-optical disk, or as computer code originally stored in a remote recording medium or a non-transitory machine-readable medium and to be stored in a local recording medium downloaded through a network, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware such as an ASIC or FPGA. It will be appreciated that the computer, processor, microprocessor controller or programmable hardware includes memory components (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code that, when accessed and executed by the computer, processor or hardware, implements the processing methods described herein. Further, when a general-purpose computer accesses code for implementing the processes shown herein, execution of the code transforms the general-purpose computer into a special-purpose computer for performing the processes shown herein.

It should be noted that, according to the implementation requirement, each step/component described in the present application can be divided into more steps/components, and two or more steps/components or partial operations of the steps/components can be combined into new steps/components to achieve the purpose of the present invention.

It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.