CN109067688B - Dual-drive OFDM (orthogonal frequency division multiplexing) receiving method of data model - Google Patents

Dual-drive OFDM (orthogonal frequency division multiplexing) receiving method of data model Download PDF

Info

Publication number
CN109067688B
CN109067688B CN201810743534.1A CN201810743534A CN109067688B CN 109067688 B CN109067688 B CN 109067688B CN 201810743534 A CN201810743534 A CN 201810743534A CN 109067688 B CN109067688 B CN 109067688B
Authority
CN
China
Prior art keywords
channel estimation
neural network
deep neural
result
frequency domain
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810743534.1A
Other languages
Chinese (zh)
Other versions
CN109067688A (en
Inventor
金石
高璇璇
张静
温朝凯
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Original Assignee
Southeast University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Southeast University filed Critical Southeast University
Priority to CN201810743534.1A priority Critical patent/CN109067688B/en
Publication of CN109067688A publication Critical patent/CN109067688A/en
Application granted granted Critical
Publication of CN109067688B publication Critical patent/CN109067688B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L27/00—Modulated-carrier systems
    • H04L27/26—Systems using multi-frequency codes
    • H04L27/2601—Multicarrier modulation systems
    • H04L27/2647—Arrangements specific to the receiver only
    • H04L27/2655—Synchronisation arrangements
    • H04L27/2689—Link with other circuits, i.e. special connections between synchronisation arrangements and other circuits for achieving synchronisation
    • H04L27/2695—Link with other circuits, i.e. special connections between synchronisation arrangements and other circuits for achieving synchronisation with channel estimation, e.g. determination of delay spread, derivative or peak tracking
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202—Channel estimation
    • H04L25/024—Channel estimation channel estimation algorithms
    • H04L25/025—Channel estimation channel estimation algorithms using least-mean-square [LMS] method
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202—Channel estimation
    • H04L25/024—Channel estimation channel estimation algorithms
    • H04L25/0254—Channel estimation channel estimation algorithms using neural network algorithms
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L27/00—Modulated-carrier systems
    • H04L27/26—Systems using multi-frequency codes
    • H04L27/2601—Multicarrier modulation systems
    • H04L27/2647—Arrangements specific to the receiver only
    • H04L27/2655—Synchronisation arrangements
    • H04L27/2689—Link with other circuits, i.e. special connections between synchronisation arrangements and other circuits for achieving synchronisation
    • H04L27/2692—Link with other circuits, i.e. special connections between synchronisation arrangements and other circuits for achieving synchronisation with preamble design, i.e. with negotiation of the synchronisation sequence with transmitter or sequence linked to the algorithm used at the receiver

Abstract

The invention discloses a data model dual-drive OFDM receiving method, which comprises the following steps: training a deep neural network in a channel estimation and signal detection module in a receiver respectively; for the channel estimation module, inputting received frequency domain pilot frequency and local frequency domain pilot frequency, estimating by adopting a least square estimation method to obtain a least square channel estimation result, taking a real part and an imaginary part out, and inputting the real part and the imaginary part in series into a deep neural network for improvement and outputting the channel estimation result; for the signal detection module, inputting the received frequency domain pilot frequency and the channel estimation result obtained by the channel estimation module, obtaining a zero-forcing equalization result by adopting a zero-forcing equalization method, and inputting a deep neural network for improvement and obtaining output; and carrying out hard decision on the obtained deep neural network output and a set threshold, obtaining a decision result, and obtaining the estimation of the transmitted bit stream according to the decision result. The invention is optimized and improved by the neural network, the time consumption of network training is less, the detection performance is high, and the channel information can be quickly obtained.

Description

Dual-drive OFDM (orthogonal frequency division multiplexing) receiving method of data model
Technical Field
The invention relates to a data model dual-drive OFDM receiving method, belonging to the technical field of wireless communication.
Background
In recent years, deep learning is used as a basic technology in artificial intelligence, and has achieved great success in subjects such as computer vision and natural language processing, and the performance of the traditional machine learning method is surpassed in scenes such as image classification, facial recognition, voice recognition, machine translation, style conversion and the like, so that application of unmanned driving, intelligent disease diagnosis, intelligent home furnishing, personalized recommendation and the like becomes possible. Deep learning is a branch of the field of machine learning, and is a supervised learning method, and a group of optimal neural network parameters are obtained by minimizing a loss function between a predicted value and a true value of a deep neural network, so that the deep neural network can perform accurate prediction.
Deep learning has been explored in the application of the physical layer of wireless communication, and includes that the whole communication system is completely replaced by a deep neural network from end to end, or only partial modules of the communication system, such as an encoder, a decoder, a detector and the like, are replaced by the deep neural network. However, the currently adopted method does not apply the knowledge in the wireless communication field to the neural network design, so that the function of the neural network is completely a black box, the training of the neural network completely depends on a large amount of data driving, the parameter number of the neural network is also larger, and the training speed is slow.
Disclosure of Invention
The technical problem to be solved by the invention is to overcome the defects of the prior art, provide a data model dual-drive OFDM receiving method, and solve the problem that the prior art does not add communication knowledge and completely depends on data drive when a deep learning network is applied to a communication receiver.
The invention specifically adopts the following technical scheme to solve the technical problems:
a data model dual-drive OFDM receiving method comprises the following steps:
step 1, training a deep neural network in a channel estimation module and a signal detection module in a receiver respectively;
step 2, for the channel estimation module, inputting the received frequency domain pilot frequency and local frequency domain pilot frequency, and obtaining a least square channel estimation result by adopting a least square estimation method
Figure BDA0001723794620000011
And taking the real part and the imaginary part of the least square channel estimation, taking the real part and the imaginary part out, connecting the real part and the imaginary part in series, inputting the part into the trained deep neural network for improvement, and outputting the result to obtain a channel estimation result
Figure BDA0001723794620000012
Step 3, inputting the received frequency domain pilot frequency and the channel estimation result obtained by the channel estimation module to the signal detection module
Figure BDA0001723794620000021
Obtaining zero-forcing equalization result by adopting zero-forcing equalization method
Figure BDA0001723794620000022
And zero-forcing equalization results
Figure BDA0001723794620000023
Real and imaginary parts of the series and channel estimation results
Figure BDA0001723794620000024
Receiving signals, inputting the signals into a trained deep neural network for improvement and outputting the signals;
and 4, carrying out hard decision on the output of the deep neural network obtained in the step 3 and a set threshold, obtaining a decision result, and obtaining the estimation of the transmitted bit stream according to the decision result.
Further, as a preferred technical solution of the present invention, the loss function adopted in the deep neural network training in step 1 is a square error loss, and the adopted optimizer is an adaptive momentum estimation optimizer.
Further, as a preferred technical solution of the present invention, the loss function adopted in step 1 is a squared error loss, which specifically includes:
in the channel estimation module, the squared error loss employed is:
Figure BDA0001723794620000025
in the signal detection module, the square error loss employed is:
Figure BDA0001723794620000026
where N is the number of subcarriers, B is the number of bits to be estimated, H is the actual frequency domain channel,
Figure BDA0001723794620000027
is the channel estimation result, b is the actual transmitted bit stream,
Figure BDA0001723794620000028
is an estimate of the transmitted bit stream.
Further, as a preferred technical solution of the present invention, the step 2 obtains a least square channel estimation result
Figure BDA0001723794620000029
The formula is adopted:
Figure BDA00017237946200000210
where Y 'is the received frequency domain pilot and X' is the local frequency domain pilot.
By adopting the technical scheme, the invention can produce the following technical effects:
the invention reasonably integrates communication knowledge into the neural network design, obviously improves the performance under linear and nonlinear conditions, greatly shortens the network training time, and is suitable for a system adopting OFDM. In addition, when the environment is complicated, stable performance can be obtained by connecting the conventional modules in series.
Therefore, the invention provides a new method for designing the receiver of the OFDM system by combining deep learning and communication knowledge, the BER performance of the method is improved in linear and nonlinear situations compared with the BER performance of the traditional linear receiving method and the BER performance of the method in the full-connection method, the parameters are less than those of the full-connection method, the training speed is high, channel information can be obtained, and the scheme of offline training and online deployment is also suitable for implementation and application.
Drawings
Fig. 1 is a block diagram of a data model dual drive OFDM receiver according to the present invention.
Fig. 2 is a block diagram of a channel estimation module of the present invention.
FIG. 3 is a block diagram of a deep neural network in a signal detection module under the nonlinear condition of the present invention.
Detailed Description
The following describes embodiments of the present invention with reference to the drawings.
As shown in fig. 1, which is a schematic diagram of an OFDM receiver with dual driving of data models, the present invention provides an OFDM receiving method with dual driving of data models, which can be used for the receiver, but is not limited to the receiver structure, and the method specifically includes the following steps:
step 1, training a deep neural network in a channel estimation module and a signal detection module in a receiver respectively. The deep neural network driven by the data model in a dual mode is applied to a wireless communication receiver, the network is trained by a supervised learning method under the on-line condition, the trained network is used for forward prediction in the actual application, the specific structure of the receiver is divided into two sub-modules of channel estimation and signal detection, the deep neural network in the channel estimation module is trained firstly during the training, then the deep neural network in the signal detection module is trained, and a channel estimation result output by the channel estimation module is obtained firstly during the application and then is input into the signal detection module.
The method comprises the following steps that a loss function adopted for training of the deep neural network is square error loss, an adopted optimizer is an adaptive momentum estimation Adam optimizer, a small-batch training method is adopted, a dynamically adjusted learning rate is adopted, and in a channel estimation module, the square error loss is as follows:
Figure BDA0001723794620000031
in the signal detection module, the squared error loss is:
Figure BDA0001723794620000032
where N is the number of subcarriers, B is the number of bits to be estimated, H is the actual frequency domain channel,
Figure BDA0001723794620000033
is the channel estimation result, b is the actual transmitted bit stream,
Figure BDA0001723794620000034
is an estimate of the transmitted bit stream.
Step 2, for the channel estimation module, inputting the received frequency domain pilot frequency and local frequency domain pilot frequency, initializing by adopting a least square estimation method, and for the kth subcarrier, obtaining a least square channel estimation result by estimation
Figure BDA0001723794620000041
Comprises the following steps:
Figure BDA0001723794620000042
where Y 'is the received frequency domain pilot and X' is the local frequency domain pilot.
Then the least square channel estimation result is obtained
Figure BDA0001723794620000043
The real part and the imaginary part are taken out and connected in series, and are sent into a fully-connected deep neural network for improvement, and a channel estimation result is output
Figure BDA0001723794620000044
In this step, the deep neural network is a one-layer fully-connected network, no activation function is adopted, multiplicative parameter W of the network is initialized by adopting the real value of the weight matrix of the linear minimum mean square error channel estimation, additive parameter n is initialized to be all zero, all parameters are set to be trainable, and the calculation of the network implementation is realized at the moment
Figure BDA0001723794620000045
Because of the minimum mean square error channel estimation
Figure BDA0001723794620000046
And only real-valued calculation can be carried out in the deep learning network, so when the input of the network is
Figure BDA0001723794620000047
The real part and the imaginary part of the network are connected in series, multiplicative parameters W of the network are initialized into a weight matrix WLMMSEReal value matrix of
Figure BDA0001723794620000048
Wherein Re {. is a real part, and Im {. is an imaginary part; when the additive parameter n is initialized to be all zero, the initial value of the network output is
Figure BDA0001723794620000049
The real part and the imaginary part are connected in series, and then the network parameters are trained, so that the channel estimation result output by the network can be obtained
Figure BDA00017237946200000410
Accuracy exceeding
Figure BDA00017237946200000411
Step 3, inputting the received frequency domain pilot frequency and the channel estimation result obtained by the channel estimation module to the signal detection module
Figure BDA00017237946200000412
Initializing by adopting a zero-forcing equalization method, and obtaining a zero-forcing equalization result for the kth subcarrier
Figure BDA00017237946200000413
Comprises the following steps:
Figure BDA00017237946200000414
where Y is the received frequency domain data,
Figure BDA00017237946200000415
is the output of the channel estimation module.
And zero-forcing equalization results
Figure BDA00017237946200000416
Real and imaginary parts of the series and channel estimation results
Figure BDA00017237946200000417
The received signal is input into the trained deep neural network for improvement and output. Wherein the received signal may be received user data.
For the signal detection module, two different deep neural networks are adopted based on whether the system is linear or nonlinear:
when the OFDM system is a linear system, a fully-connected deep neural network is adopted; the input of the network is the result of zero-forcing equalization, and the network adopts a fully-connected network with two layers.
When the OFDM system is a nonlinear system, for example, a system with intersymbol interference due to the removal of a cyclic prefix or a system with truncated noise due to the truncation for reducing the peak-to-average energy ratio, the signal detection improvement network adopts a bidirectional long-time and short-time memory BilSTM cyclic neural network and a one-layer fully-connected neural network. The input of the network is the zero-forcing equalization result, the channel estimation result output by the channel estimation module and the received signal, the network adopts a three-layer BiLSTM recurrent neural network with OFDM carrier number in time stages, and then the output of the BiLSTM recurrent neural network is sent to a layer of fully-connected network.
And when the OFDM system is a nonlinear system, the number of the neurons in the last layer of the deep neural network of the signal detection module is equal to the estimated bit number B, is determined by the number of bits contained in partial continuous subcarriers, and can change with different modulation orders of the adopted constellation.
And 4, carrying out hard decision on the output of the deep neural network obtained in the step 3 and a set threshold, obtaining a decision result, and obtaining the estimation of the transmitted bit stream according to the decision result.
In this embodiment, a hard decision may be made on the output of the deep neural network in the signal detection module with 0.5 as a threshold, and a decision that the output is greater than 0.5 is bit 1 and a decision that the output is less than 0.5 is bit 0, so as to obtain an estimate of the transmitted bit stream.
When the channel state is deteriorated or changed too fast, the traditional communication method is adopted, namely the least square channel estimation is calculated by the receiving pilot frequency and the local pilot frequency, then the result of the least square channel estimation and the receiving frequency domain data are sent to a zero-forcing detection module, then the constellation demapping is carried out on the result of the zero-forcing detection, and the sending bit stream estimation with general accuracy and stability is obtained
Figure BDA0001723794620000051
Based on the above method, the present invention provides a specific embodiment.
As shown in fig. 1, a data model dual-drive neural network is applied to a wireless communication receiver, functions of three modules of channel estimation, signal equalization and QAM demodulation of an original OFDM system are replaced, specifically, the channel estimation and signal detection sub-modules are divided into two sub-modules, a network is trained by a supervised learning method on line, the trained network is used for forward prediction in actual application, the receiver is specifically divided into two sub-modules of channel estimation and signal detection, a deep neural network in a channel estimation module is trained firstly and then a deep neural network in a signal detection module is trained in training, and a channel estimation result output by the channel estimation module is obtained firstly and then input into the signal detection module in application.
As shown in FIG. 2, for the channel estimation module, the received frequency domain pilot and the local frequency domain pilot are input, and the least square estimation result is firstly adopted for initialization, and for the k-th subcarrier (k is more than or equal to 1 and less than or equal to 64), the least square channel estimation is carried out
Figure BDA0001723794620000052
Comprises the following steps:
Figure BDA0001723794620000053
y 'is received frequency domain pilot frequency, X' is local frequency domain pilot frequency, then the real part and imaginary part of the least square estimation result are taken out and connected in series to form a vector with length of 128, the vector is sent to a fully-connected deep neural network for improvement, and the channel estimation result with length of 128 is output
Figure BDA0001723794620000061
The deep neural network is a one-layer fully-connected network, an activation function is not adopted, multiplicative parameters of the layer network are initialized by adopting real values of a weight matrix of linear minimum mean square error channel estimation, initial values of additive parameters are set to be all zero, and all the parameters are set to be trainable. Wherein the minimum mean square error channel estimate
Figure BDA0001723794620000062
Weight matrix WLMMSEReal value matrix of
Figure BDA0001723794620000063
Comprises the following steps:
Figure BDA0001723794620000064
the dimension is 128 × 128. The deep neural network training process comprises the steps of setting a label value as a real frequency domain channel H, and setting a loss function as a square error loss
Figure BDA0001723794620000065
The optimizer is an Adam optimizer, small-batch gradient descent is adopted during training, 50 batches are adopted in each round, the size of each batch is 1000 samples, the training is carried out for 2000 rounds, the learning rate of the first 1000 rounds is 0.001, and the learning rate of the last 1000 rounds is 0.0001.
For the signal detection module, the received frequency domain data and the channel estimation of the channel estimation module are inputAs a result, the zero-forcing equalization result is used for initialization, and for the k-th sub-carrier, the zero-forcing equalization result
Figure BDA0001723794620000066
Comprises the following steps:
Figure BDA0001723794620000067
where Y is the received frequency domain data,
Figure BDA0001723794620000068
is the output of the channel estimation module, and then the real part and the imaginary part of the zero-forcing equalization result are taken out and connected in series, or the result of the same channel estimation is taken out
Figure BDA0001723794620000069
The received signals are sent to a deep neural network for improvement, the output of the deep neural network in a signal detection module is hard judged by 0.5 bit threshold, more than 0.5 is judged as bit 1, less than 0.5 is judged as bit 0, and the estimation of the transmitted bit stream is output
Figure BDA00017237946200000610
Based on whether the OFDM system is linear or nonlinear, the deep neural network adopts two different signal detection modules. For a linear system, the deep neural network adopts a fully-connected deep neural network; for systems with nonlinear effects, such as systems with intersymbol interference due to the removal of cyclic prefixes, or systems with truncated noise due to the truncation for reducing the peak-to-average energy ratio, the deep neural network adopts a BilSTM cyclic neural network and a layer of fully-connected neural network.
For a fully connected network under a linear system, the input is
Figure BDA00017237946200000611
The real part and the imaginary part of the two-layer network are connected in series to form a vector with the length of 128, and the output is bit stream recovery with the length of 48 and neural of the two-layer networkThe number of elements is 120 and 48 respectively, the adopted activation functions are ReLU and Sigmoid respectively, multiplicative parameters are initialized by a He initialization method, and additive parameters are all initialized to zero.
For the BilSTM recurrent neural network under the nonlinear system, as shown in FIG. 3, three layers of 64 time-phased BilSTM recurrent neural networks are adopted, the number of states of each layer is respectively 20,10 and 6, and the output of the last layer of BilSTM is sent to a layer of full-connection network with the number of neurons being 48 and the activation function being Sigmoid. The inputs for the kth time period of the first layer of BilSTM are:
Figure BDA0001723794620000071
wherein Re {. is a real part, and Im {. is an imaginary part.
The process of training the signal detection improvement network under the above linear or nonlinear system is that the label value is set as the actual transmission bit stream b, and the loss function is the square error loss
Figure BDA0001723794620000072
The optimizer is an Adam optimizer, small batch gradient descent is adopted during training, 50 batches are adopted in each round, and the size of each batch is 1000 samples. A total of 5000 rounds of training, with a dynamically adjusted learning rate, the initial learning rate was set to 0.001, decreasing to one fifth of the current per 2000 rounds.
When the channel state is deteriorated or changed too fast, the traditional communication method is adopted, namely the least square channel estimation is calculated by the receiving pilot frequency and the local pilot frequency, then the result of the least square channel estimation and the receiving frequency domain data are sent to a zero-forcing detection module, then the constellation demapping is carried out on the result of the zero-forcing detection, and the sending bit stream estimation with general accuracy and stability is obtained
Figure BDA0001723794620000073
By specific parameters of experimental examples, in an OFDM system with 64 subcarriers, the data frame format is one pilot OFDM symbol and one data OFDM symbol, and the pilot and data occupy 64 subcarriers. The constellation modulation mode of the pilot frequency is QPSK, and the constellation modulation mode of the data adopts 64QAM of the LTE standard. At a transmitting end, data bits are 64 multiplied by 6 to 384 bits, and are converted into time domain sending signals through 64QAM constellation modulation, pilot frequency adding framing and IFFT; after passing through a multipath channel, the receiving end performs FFT to obtain frequency domain received pilot and data, and the frequency domain received pilot and data are sent to the OFDM system receiver based on the combination of communication knowledge and deep learning of the present invention to obtain the recovery of 8 × 6 to 48 bits on 8 continuous subcarriers, such as the 1 st to 8 th subcarriers or the 9 th to 16 th subcarriers. Training 64/8 is 8 independent signal detection modules, and hard decision is made on the output of the deep neural network, so that all 384-bit recovery can be obtained.
Therefore, the invention provides a new method for designing the receiver of the OFDM system by combining deep learning and communication knowledge, the BER performance of the method is improved in linear and nonlinear situations compared with the BER performance of the traditional linear receiving method and the BER performance of the method in the full-connection method, the parameters are less than those of the full-connection method, the training speed is high, channel information can be obtained, and the scheme of offline training and online deployment is also suitable for implementation and application.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.

Claims (8)

1. A data model dual-drive OFDM receiving method is characterized by comprising the following steps:
step 1, training a deep neural network in a channel estimation module and a signal detection module in a receiver respectively;
step 2, for the channel estimation module, inputting the received frequency domain pilot frequency and local frequency domain pilot frequency, and obtaining a least square channel estimation result by adopting a least square estimation method
Figure FDA0001723794610000011
And taking the real part and the imaginary part of the least square channel estimation, taking the real part and the imaginary part out, connecting the real part and the imaginary part in series, inputting the part into the trained deep neural network for improvement, and outputting the result to obtain a channel estimation result
Figure FDA0001723794610000012
Step 3, inputting the received frequency domain pilot frequency and the channel estimation result obtained by the channel estimation module to the signal detection module
Figure FDA0001723794610000013
Obtaining zero-forcing equalization result by adopting zero-forcing equalization method
Figure FDA0001723794610000014
And zero-forcing equalization results
Figure FDA0001723794610000015
Real and imaginary parts of the series and channel estimation results
Figure FDA0001723794610000016
Receiving signals, inputting the signals into a trained deep neural network for improvement and outputting the signals;
and 4, carrying out hard decision on the output of the deep neural network obtained in the step 3 and a set threshold, obtaining a decision result, and obtaining the estimation of the transmitted bit stream according to the decision result.
2. The data model dual-drive OFDM receiving method as claimed in claim 1, wherein the loss function used for deep neural network training in step 1 is a squared error loss, and the optimizer used is an adaptive momentum estimation optimizer.
3. The data model dual-drive OFDM receiving method according to claim 2, wherein the loss function used in step 1 is a squared error loss, and specifically includes:
in the channel estimation module, the squared error loss employed is:
Figure FDA0001723794610000017
in the signal detection module, the square error loss employed is:
Figure FDA0001723794610000018
where N is the number of subcarriers, B is the number of bits to be estimated, H is the actual frequency domain channel,
Figure FDA0001723794610000019
is the channel estimation result, b is the actual transmitted bit stream,
Figure FDA00017237946100000110
is an estimate of the transmitted bit stream.
4. The data model dual-drive OFDM receiving method according to claim 1, wherein the least square channel estimation result obtained in step 2
Figure FDA00017237946100000111
The formula is adopted:
Figure FDA00017237946100000112
where Y 'is the received frequency domain pilot and X' is the local frequency domain pilot.
5. The data model dual-drive OFDM receiving method according to claim 1, wherein the deep neural network in the channel estimation module in step 2 is a one-layer fully-connected network, wherein,
Figure FDA0001723794610000021
and, minimum mean square error channel estimation
Figure FDA0001723794610000022
Multiplicative parameter W of network is initialized to weight matrix WLMMSEReal value matrix of
Figure FDA0001723794610000023
Wherein Re {. is a real part, and Im {. is an imaginary part; when the additive parameter n is initialized to be all zero, the initial value of the network output is
Figure FDA0001723794610000024
The real and imaginary parts of (a) are connected in series.
6. The data model dual-drive OFDM receiving method of claim 1, wherein in step 3, a zero-forcing equalization result is obtained
Figure FDA0001723794610000025
The formula is adopted:
Figure FDA0001723794610000026
where Y is the received frequency domain data,
Figure FDA0001723794610000027
is the channel estimation result.
7. The data model dual-drive OFDM receiving method according to claim 1, wherein the deep neural network in step 3 specifically comprises:
when the OFDM system is a linear system, a fully-connected deep neural network is adopted;
when the OFDM system is a nonlinear system, a BilSTM recurrent neural network and a layer of fully-connected deep neural network are adopted.
8. The data model dual-drive OFDM receiving method as claimed in claim 7, wherein in step 3, when the OFDM system is a nonlinear system, the number of neurons of a layer-depth fully-connected deep neural network is equal to the number of bits to be estimated.
CN201810743534.1A 2018-07-09 2018-07-09 Dual-drive OFDM (orthogonal frequency division multiplexing) receiving method of data model Active CN109067688B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810743534.1A CN109067688B (en) 2018-07-09 2018-07-09 Dual-drive OFDM (orthogonal frequency division multiplexing) receiving method of data model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810743534.1A CN109067688B (en) 2018-07-09 2018-07-09 Dual-drive OFDM (orthogonal frequency division multiplexing) receiving method of data model

Publications (2)

Publication Number Publication Date
CN109067688A CN109067688A (en) 2018-12-21
CN109067688B true CN109067688B (en) 2021-09-07

Family

ID=64819570

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810743534.1A Active CN109067688B (en) 2018-07-09 2018-07-09 Dual-drive OFDM (orthogonal frequency division multiplexing) receiving method of data model

Country Status (1)

Country Link
CN (1) CN109067688B (en)

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109889462B (en) * 2019-01-21 2021-05-28 东南大学 Neural network assisted iterative receiving method suitable for high-speed visible light communication
CN109672639B (en) * 2019-01-22 2021-07-16 哈尔滨工业大学 Signal demodulation method based on machine learning
CN109831399A (en) * 2019-01-22 2019-05-31 东南大学 A kind of OFDM receiver of artificial intelligence auxiliary
CN109714086B (en) * 2019-01-23 2021-09-14 上海大学 Optimized MIMO detection method based on deep learning
CN109921882B (en) * 2019-02-20 2021-08-27 深圳市宝链人工智能科技有限公司 Deep learning-based MIMO decoding method, device and storage medium
CN111628946B (en) * 2019-02-28 2021-10-26 华为技术有限公司 Channel estimation method and receiving equipment
CN109831400A (en) * 2019-03-06 2019-05-31 深兰科技（上海）有限公司 A kind of received method and apparatus of OFDM signal
CN110266620B (en) * 2019-07-08 2020-06-30 电子科技大学 Convolutional neural network-based 3D MIMO-OFDM system channel estimation method
CN110661734B (en) * 2019-09-20 2020-07-28 西安交通大学 Channel estimation method, device and readable storage medium based on deep neural network
CN110719239B (en) * 2019-09-29 2022-03-11 东南大学 Data model dual-drive combined MIMO channel estimation and signal detection method
CN111245754B (en) * 2020-01-03 2021-07-13 西安交通大学 Self-adaptive space domain equalization method based on deep learning
CN111669344B (en) * 2020-06-01 2022-01-07 西北工业大学 Time-varying OFDM system signal detection method based on deep learning
CN111786923A (en) * 2020-06-05 2020-10-16 中国电子科技集团公司第七研究所 Channel estimation method for time-frequency double-channel selection of orthogonal frequency division multiplexing system
CN112383496A (en) * 2020-11-09 2021-02-19 西安科技大学 Mine communication method, system, computer equipment and medium based on depth receiver
CN112637093B (en) * 2020-12-09 2022-05-17 齐鲁工业大学 Signal detection method based on model-driven deep learning
CN112866150A (en) * 2021-01-13 2021-05-28 上海复旦通讯股份有限公司 Wireless channel feature extraction and dimension reduction method and system
CN113037668B (en) * 2021-05-20 2023-03-10 武汉科技大学 Millimeter wave point-to-point communication channel equalization method
CN113676431B (en) * 2021-07-08 2023-01-10 东南大学 Model-driven MIMO-OFDM receiving method without cyclic prefix
CN113708891A (en) * 2021-08-12 2021-11-26 电子科技大学 Communication system adopting deep learning CS decoder
CN113839743A (en) * 2021-09-18 2021-12-24 电子科技大学 RLL code decoder for wireless communication receiving end
CN113872911B (en) * 2021-10-15 2023-10-24 齐鲁工业大学 Model-driven method and system for suppressing peak-to-average ratio of orthogonal frequency division multiplexing system
CN114301742B (en) * 2021-12-23 2024-01-09 北京邮电大学 Channel estimation method and device
CN115250217B (en) * 2022-07-22 2023-05-19 西华大学 OFDM system channel estimation method and device assisted by differential coding and neural network
CN116455706A (en) * 2023-04-12 2023-07-18 华南理工大学 Channel estimation and equalization design method and system

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101340266A (en) * 2008-06-06 2009-01-07 山东大学 Dual-selection channel tracking method based on particle filter and neural network
CN104022978A (en) * 2014-06-18 2014-09-03 中国联合网络通信集团有限公司 Half-blindness channel estimating method and system

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101340266A (en) * 2008-06-06 2009-01-07 山东大学 Dual-selection channel tracking method based on particle filter and neural network
CN104022978A (en) * 2014-06-18 2014-09-03 中国联合网络通信集团有限公司 Half-blindness channel estimating method and system

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
An Introduction to Deep Learning for the Physical Layer;Tim O’Shea等;《IEEE Transactions on Cognitive Communications and Networking》;20171002;全文 *
Framewise Phoneme Classification with Bidirectional LSTM and Other Neural Network Architectures;Alex Graves等;《Neural Networks》;20050819;第602-610页 *
Neural Network Channel Estimation Based on Least Mean Error Algorithm in the OFDM Systems;Jun Sun等;《International Symposium on Neural Networks:Advances in Neural Networks》;20060601;全文 *
Power of Deep Learning for Channel Estimation and Signal Detection in OFDM Systems;HaoYe等;《IEEE WIRELESS COMMUNICATIONS LETTERS》;20170928;第7卷(第1期);第1节至第3节 *

Also Published As

Publication number Publication date
CN109067688A (en) 2018-12-21

Similar Documents

Publication Publication Date Title
CN109067688B (en) Dual-drive OFDM (orthogonal frequency division multiplexing) receiving method of data model
CN109617847B (en) OFDM receiving method without cyclic prefix based on model-driven deep learning
CN111404849B (en) OFDM channel estimation and signal detection method based on deep learning
CN109474352B (en) Underwater sound orthogonal frequency division multiplexing communication method based on deep learning
CN112637093B (en) Signal detection method based on model-driven deep learning
Lin et al. A novel OFDM autoencoder featuring CNN-based channel estimation for internet of vessels
CN112600772B (en) OFDM channel estimation and signal detection method based on data-driven neural network
CN109246038B (en) Dual-drive GFDM receiver and method for data model
CN109194595B (en) Neural network-based channel environment self-adaptive OFDM receiving method
CN111669344B (en) Time-varying OFDM system signal detection method based on deep learning
Zhang et al. Deep learning based on orthogonal approximate message passing for CP-free OFDM
CN111614587A (en) SC-FDE system signal detection method based on self-adaptive integrated deep learning model
Chi et al. Challenges and prospects of machine learning in visible light communication
CN111786923A (en) Channel estimation method for time-frequency double-channel selection of orthogonal frequency division multiplexing system
CN110311876A (en) The implementation method of underwater sound OFDM receiver based on deep neural network
CN113708855A (en) OTFS data drive receiving method, system and medium based on deep learning
CN115664898A (en) OFDM system channel estimation method and system based on complex convolution neural network
Zhao et al. Multi-task learning based underwater acoustic OFDM communications
CN104580058B (en) A kind of ofdm system inter-sub-carrier interference self elimination method
Cheng et al. Using back propagation neural network for channel estimation and compensation in OFDM systems
Vilas Boas et al. Artificial intelligence for channel estimation in multicarrier systems for B5G/6G communications: a survey
CN109391315B (en) Data model dual-drive MIMO receiver
Ye et al. Bilinear convolutional auto-encoder based pilot-free end-to-end communication systems
CN111313971A (en) Lightgbm equalization system and method for improving IMDD short-distance optical communication system
CN112636855A (en) OFDM signal detection method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant