CN114598575A - Deep learning channel estimation method based on self-attention mechanism - Google Patents

Deep learning channel estimation method based on self-attention mechanism Download PDF

Info

Publication number
CN114598575A
CN114598575A CN202210239196.4A CN202210239196A CN114598575A CN 114598575 A CN114598575 A CN 114598575A CN 202210239196 A CN202210239196 A CN 202210239196A CN 114598575 A CN114598575 A CN 114598575A
Authority
CN
China
Prior art keywords
matrix
channel
self
attention mechanism
channel estimation
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.)
Granted
Application number
CN202210239196.4A
Other languages
Chinese (zh)
Other versions
CN114598575B (en
Inventor
赵嗣强
邱玲
许逸丰
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.)
University of Science and Technology of China USTC
Original Assignee
University of Science and Technology of China USTC
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 University of Science and Technology of China USTC filed Critical University of Science and Technology of China USTC
Priority to CN202210239196.4A priority Critical patent/CN114598575B/en
Publication of CN114598575A publication Critical patent/CN114598575A/en
Application granted granted Critical
Publication of CN114598575B publication Critical patent/CN114598575B/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
    • 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
    • H04L25/0242—Channel estimation channel estimation algorithms using matrix methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Signal Processing (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Power Engineering (AREA)
  • Radio Transmission System (AREA)

Abstract

The invention discloses a deep learning channel estimation method based on a self-attention mechanism under a multi-input multi-output orthogonal frequency division multiplexing system, which is characterized in that the correlation between time-frequency domain channel impulse responses is utilized, channel characteristic information can be effectively extracted by adopting a self-attention mechanism module, so that a globally-dependent characteristic mapping is constructed, and the correlation between the time-frequency domain channel impulse responses can be deeply learned by utilizing the deep learning method. Compared with the channel estimation result of the existing MIMO OFDM system, the channel estimation result has ideal improvement on the accuracy.

Description

Deep learning channel estimation method based on self-attention mechanism
Technical Field
The invention belongs to the technical field of wireless communication, and particularly relates to a self-attention mechanism-based deep learning channel estimation method suitable for a multi-input multi-output orthogonal frequency division multiplexing system.
Background
The Emerging Telecommunications technology exchange ("Orthogonal frequency division multiplexing with sub-carrier power modulation for multiplexing the spectral frequencies of 6G and channels" in transformations on interference Telecommunications Technologies,2020,31(4): e3921.) states that Orthogonal frequency division multiplexing is widely used for 5G wireless communications due to its robustness to inter-carrier interference and inter-symbol interference and will continue to be used as a key technology in 6G communications. Mimo systems are capable of increasing channel capacity without increasing bandwidth, thereby accommodating a greater number of users within the available bandwidth, and are therefore often used in conjunction with orthogonal frequency division multiplexing techniques to achieve better data transmission rates, higher spectral utilization, and better resistance to multipath fading. The performance of the mimo ofdm system in a wireless fading environment depends on an accurate channel impulse response to a great extent, and it is very important to obtain an accurate channel impulse response through channel estimation. The institute of electrical and electronics engineers communication news ("Deep learning-based channel estimation." in IEEE Communications Letters,2019,23(4):652 and 655) proposes that the time-frequency response of a fast fading communication channel is regarded as a two-dimensional image, and a channel estimation method based on a Deep learning algorithm is adopted to estimate the channel impulse response. However, since the correlation between the time-frequency domain channel impulse responses is not effectively utilized, the accuracy of the channel impulse response obtained by the channel estimation method is not high.
Disclosure of Invention
The invention provides a channel estimation method based on deep learning of a self-attention mechanism under a multi-input multi-output orthogonal frequency division multiplexing system, so that the correlation between time-frequency domain channel impulse responses is effectively utilized, and the accuracy of an estimation result is further improved.
The invention relates to a channel estimation method based on deep learning of a self-attention mechanism under a multi-input multi-output orthogonal frequency division multiplexing system, which is characterized by comprising the following steps:
number N for one transmitting antennatThe number of receiving antennas is NrIn the mimo ofdm system, wherein the n-th transmission endtA certain ofdm symbol on the root transmit antenna is represented as:
Figure BDA0003543619310000011
wherein L is the number of subcarriers [. ]]TRepresents transposition, for the n-thrRoot jointAnd the receiving antenna, wherein the symbol on the k-th subcarrier is represented as:
Figure BDA0003543619310000012
wherein,
Figure BDA0003543619310000021
denotes the n-thtRoot transmitting antenna and nrThe channel impulse response between the receiving antennas corresponding to the k-th sub-carrier,
Figure BDA0003543619310000022
representing additive white gaussian noise on the receive antenna; for one of the pilot symbols, all NrThe symbol on the k-th subcarrier received by the root receive antenna is represented as:
Yk＝HkXk+Zk,(k＝1,...,L)
wherein
Figure BDA0003543619310000023
Figure BDA0003543619310000024
Respectively representing a receiving matrix at the receiving end, a transmitting matrix at the transmitting end, a noise matrix,
Figure BDA0003543619310000025
representing the frequency-domain channel impulse response matrix, H, on the k-th sub-carrierkExpressed as:
Figure BDA0003543619310000026
to HkThe channel estimation is carried out to obtain the channel impulse response estimation result of
Figure BDA0003543619310000027
Using a channel estimation method based on the least squares criterion, mostThe cost function shown in the following formula is reduced:
Figure BDA0003543619310000028
relating the cost function of the above formula to
Figure BDA0003543619310000029
Has a partial derivative of 0, i.e.
Figure BDA00035436193100000210
Obtaining the channel impulse response estimation result under the channel estimation method based on the least square criterion,
Figure BDA00035436193100000211
wherein, (. cndot.)-1An inverse matrix representing the matrix;
the specific process steps of the channel estimation method based on the deep learning of the self-attention mechanism are as follows:
step S1: estimating channel state information at pilot frequency symbol position by least square channel estimation algorithm and estimated channel matrix
Figure BDA00035436193100000212
Step S2: for the channel matrix
Figure BDA00035436193100000213
Performing up-sampling linear interpolation to obtain frequency domain channel response at data symbol, combining with frequency domain channel response at pilot frequency to obtain complete frequency domain channel response matrix, and remolding the complete frequency domain channel response matrix into L × Nsym×(Nt×NrX 2), where L represents the number of subcarriers, NsymThe number of sub-carriers is represented,Ntrepresenting the number of transmitting antennas, NrRepresents the number of receive antennas;
step S3: extracting features of the low-resolution matrix through two convolutional neural networks to obtain a feature map of the low-resolution matrix; obtaining a weighted self-attention mechanism feature mapping through a self-attention mechanism module according to the feature map, multiplying the self-attention mechanism feature mapping by a learning coefficient, and adding the learning coefficient to the original feature map to obtain a relationship feature which is interdependent between any two positions in the feature map; the relationship characteristic is respectively passed through a convolution network and a self-attention mechanism module twice to obtain a relationship dependent characteristic which is more comprehensive in learning;
step S4: the relation dependency characteristics are processed by a convolution network, and a matrix which is consistent with the final channel matrix dimension needing to be estimated is output
Figure BDA0003543619310000032
The invention relates to a deep learning channel estimation method based on a self-attention mechanism under a multi-input multi-output orthogonal frequency division multiplexing system; the channel estimation method effectively utilizes the correlation between the time-frequency domain channel impulse responses, which is not considered in the existing channel estimation work; the invention provides a deep learning channel estimation method based on a self-attention mechanism, which can effectively construct the interdependent relation of any two positions by outputting the channel characteristics to a super-resolution neural network of the self-attention mechanism; because certain correlation exists among elements of a channel matrix generated by the adopted channel model, the self-attention mechanism module is adopted to more effectively utilize the correlation among the time-frequency domain channel impulse responses, and compared with the channel estimation result of the existing multi-input multi-output orthogonal frequency division multiplexing system, the channel estimation result has ideal improvement on the accuracy.
Description of the drawings:
FIG. 1 is a flow chart of a neural network implementation of a deep learning channel estimation method based on a self-attention mechanism according to the present invention;
FIG. 2 is a schematic diagram of a self-attention mechanism module used in the present invention.
Fig. 3 is a graph comparing the Mean Square Error (MSE) performance of the channel estimation of the present invention method with the existing mimo ofdm system under different signal-to-noise ratio (SNR) settings;
fig. 4 is a graph comparing Bit Error Rate (BER) performance of the channel estimation of the present invention method with that of the existing mimo ofdm system under different signal-to-noise ratio (SNR) settings.
Detailed Description
The following describes and explains the self-attention mechanism-based deep learning channel estimation method in the mimo-ofdm system in further detail by way of embodiments with reference to the accompanying drawings.
Example 1:
to facilitate understanding of the specific implementation of the method, the following describes in detail how the present invention utilizes a self-attention mechanism-based deep learning method for channel estimation. Number N for one transmitting antennatThe number of receiving antennas is NrIn the mimo ofdm system, wherein the n-th transmission endtA certain ofdm symbol on the root transmit antenna is represented as:
Figure BDA0003543619310000031
wherein L is the number of subcarriers [. ]]TRepresents transposition, for the n-thrThe symbol on the k-th subcarrier above the receive antenna is represented as:
Figure BDA0003543619310000041
wherein,
Figure BDA0003543619310000042
denotes the n-thtRoot transmitting antenna and nrThe channel impulse response between the receiving antennas corresponding to the k-th sub-carrier,
Figure BDA0003543619310000043
representing additive white gaussian noise on the receive antenna; for one of the pilot symbols, all NrThe symbol on the k-th subcarrier received by the root receive antenna is represented as:
Yk＝HkXk+Zk,(k＝1,...,L)
wherein
Figure BDA0003543619310000044
Figure BDA0003543619310000045
Respectively representing a receiving matrix at the receiving end, a transmitting matrix at the transmitting end, a noise matrix,
Figure BDA0003543619310000046
representing the frequency-domain channel impulse response matrix, H, on the k-th sub-carrierkExpressed as:
Figure BDA0003543619310000047
to HkThe channel estimation is carried out to obtain the channel impulse response estimation result of
Figure BDA0003543619310000048
Using a channel estimation method based on a least squares criterion, a cost function shown in the following formula is minimized:
Figure BDA0003543619310000049
make the cost function of the above equation about
Figure BDA00035436193100000410
Has a partial derivative of 0, i.e.
Figure BDA00035436193100000411
Obtaining the channel impulse response estimation result under the channel estimation method based on the least square criterion,
Figure BDA00035436193100000412
wherein, (.)-1An inverse matrix representing the matrix;
the specific process steps of the channel estimation method based on the deep learning of the self-attention mechanism are as follows:
step S1: estimating channel state information at pilot frequency symbol position by least square channel estimation algorithm and estimated channel matrix
Figure BDA00035436193100000413
Step S2: for the channel matrix
Figure BDA00035436193100000414
Performing up-sampling linear interpolation to obtain frequency domain channel response at the data symbol, combining with the frequency domain channel response at the pilot frequency to obtain a complete frequency domain channel response matrix, and remolding the complete frequency domain channel response matrix into a matrix with a size of LxNsym×(Nt×NrX 2) where L is the number of subcarriers, NsymIs the number of subcarriers, NtNumber of transmitting antennas, NrIs the number of receive antennas;
step S3: extracting features of the low-resolution matrix through two convolutional neural networks to obtain a feature map of the low-resolution matrix; obtaining a weighted self-attention mechanism feature mapping through a self-attention mechanism module according to the feature map, multiplying the self-attention mechanism feature mapping by a learning coefficient, and adding the learning coefficient to the original feature map to obtain a relationship feature which is interdependent between any two positions in the feature map; the relationship characteristic is respectively passed through a convolution network and a self-attention mechanism module twice to obtain a relationship dependency characteristic which is more comprehensively learned;
step S4: the relation dependency characteristics are processed by a convolution network, and a matrix which is consistent with the final channel matrix dimension needing to be estimated is output
Figure BDA0003543619310000051
Fig. 1 shows a flow chart of a neural network implementation of the deep learning channel estimation method based on the self-attention mechanism designed by the present invention.
In this embodiment, the step S1 is as follows: estimating channel state information at pilot frequency symbol position by least square channel estimation algorithm and estimated channel matrix
Figure BDA0003543619310000052
The method specifically comprises the following steps:
by transmitting pilot symbols X with known pilot positionskUsing the received signal YkUsing least squares
Figure BDA0003543619310000053
Estimating the channel state information of the symbol at all pilot frequencies to obtain an estimated channel matrix
Figure BDA0003543619310000054
I.e., channel matrix a0 in fig. 1.
In an embodiment, the step S2 is as follows: for the channel matrix
Figure BDA0003543619310000055
Performing up-sampling linear interpolation to obtain frequency domain channel response at data symbol, combining with frequency domain channel response at pilot frequency to obtain complete frequency domain channel response matrix, and remolding the complete frequency domain channel response matrix into L × Nsym×(Nt×NrX 2), where L is the number of subcarriers, NsymIs the number of subcarriers, NtNumber of transmitting antennas, NrSpecifically, the method includes, for receiving the number of antennas:
to pair
Figure BDA0003543619310000056
Performing upsampling linear interpolation, i.e. the first step upsampling operation a1 in fig. 1, where the method replaces upsampling by linear interpolation, and assuming that the interval between two pilots is L, the frequency domain channel response estimate at the middle position between the mth pilot and the (m + 1) th pilot can be expressed as:
Figure BDA0003543619310000057
after the frequency domain channel response at the data symbol is obtained, in order to facilitate the operation of the convolution network, the real part and the imaginary part of the matrix are respectively used as the channel dimensions of the channel matrix, so that the matrix is reshaped into a matrix with the size of L multiplied by Nsym×(Nt×NrX 2) low resolution matrix.
In an embodiment, the step S3 is as follows: extracting features of the low-resolution matrix through two convolutional neural networks to obtain a feature map of the low-resolution matrix; obtaining a weighted self-attention mechanism feature mapping through a self-attention mechanism module according to the feature map, multiplying the self-attention mechanism feature mapping by a learning coefficient, and adding the learning coefficient to the original feature map to obtain a relationship feature which is interdependent between any two positions in the feature map; and obtaining a relationship dependence characteristic which is more comprehensively learned by respectively passing the relationship characteristic through a convolution network and a self-attention mechanism module twice, wherein the relationship characteristic specifically comprises the following steps:
the low resolution matrix is first passed through two convolutional neural networks a2 and A3 as shown in fig. 1 to obtain a signature x:
Figure BDA0003543619310000061
wherein Inter (-) represents an interpolation function, i.e. an upsampling module in the network, W1，W2Respectively representing the weights of the convolutional neural networks, and inputting the obtained characteristic diagram intoFrom the attention feature module (i.e., a4 shown in fig. 1). Fig. 2 is a schematic structural diagram of a self-attention mechanism module used in the present invention, in which a feature diagram is B0 shown in fig. 2, and then three feature matrices query, key, and value are obtained by respectively passing through 3 convolution networks with convolution kernel size of 1 × 1 and 3 feature mapping functions f, g, and h:
query(x)＝f(x)＝Wfx+bf，
key(x)＝g(x)＝Wgx+bg，
value(x)＝h(x)＝Whx+bh，
where W, b represent the weight and bias of the convolutional network, respectively. The three feature matrices query, key and value described above correspond to B1, B2 and B3, respectively, shown in FIG. 2. Then, by utilizing a method for recombining matrix dimensions, the query matrix, the key matrix and the value matrix are respectively recombined into a size of
Figure BDA0003543619310000062
The matrix Query, Key, Value, i.e. B4, B5, B6 shown in fig. 2, wherein
Figure BDA0003543619310000063
After that, the matrix multiplication operation is carried out on the Query and the Key, and a size of the matrix multiplication operation is obtained through a softmax module
Figure BDA0003543619310000064
The interdependent Attention mapping matrix Attention _ map, i.e., B7 shown in fig. 2:
Figure BDA0003543619310000065
Attention_mapj,ithe degree of dependency on the ith area considered when synthesizing the jth area is shown, namely, the interdependence relation between any two positions in the characteristics is realized,
Figure BDA0003543619310000066
because the obtained Attention _ map is a weight matrix which is added according to rows and is 1, the Attention _ map is firstly transposed, then multiplied by the Value matrix to obtain the weighted summation of the Attention _ map at each position, and finally, the weighted summation is obtained through a convolution network with the convolution kernel size of 1 × 1, so that the self-Attention feature mapping o (i.e. B8 shown in fig. 2) is obtained, then, the self-Attention feature mapping is multiplied by a coefficient to be added to the original feature mapping, and the final output is:
y＝γO+x,
where y is B9 shown in fig. 2, γ is a learnable scalar quantity initialized to 0, and since it can be easier to learn in locally adjacent network features, the introduced γ starts learning from the local part first, and then gradually learns to give higher weight to the non-local features, which is also consistent with the intuition of the human body: firstly, learning a simple task, then gradually increasing the complexity of the task, and continuously superposing weighted self-attention feature mapping on the original feature mapping along with the continuous deep learning, thereby finally obtaining a globally-dependent feature mapping.
In one embodiment, the step S4 is as follows: the relation dependency characteristics are processed by a convolution network, and a matrix which is consistent with the final channel matrix dimension needing to be estimated is output
Figure BDA0003543619310000071
The method specifically comprises the following steps:
by setting the number of convolution kernels, the finally output channel dimension is converted to be consistent with the estimated channel matrix dimension through convolution (namely, A5 shown in FIG. 1), and finally, the feedback of the neural network is carried out by solving the following optimization problem, so that a better estimation result is obtained:
Figure BDA0003543619310000072
where Θ represents the training parameters of the entire network.
After the above steps, the recovered estimated channel (i.e., a6 shown in fig. 1) is finally output.
The invention relates to a deep learning channel estimation method based on a self-attention mechanism under a multi-input multi-output orthogonal frequency division multiplexing system; the channel estimation method effectively utilizes the correlation between channel time-frequency responses, which is not considered in the existing channel estimation work; the invention provides a deep learning channel estimation method based on a self-attention mechanism, which can effectively construct the interdependent relation of any two positions by outputting the channel characteristics to a super-resolution neural network of the self-attention mechanism; because certain correlation exists among elements of a channel matrix generated by the adopted channel model, the correlation among channel time-frequency responses is more effectively utilized by adopting the self-attention mechanism module, and the accuracy of a channel estimation result is ideally improved compared with the channel estimation result of the existing multi-input multi-output orthogonal frequency division multiplexing system.
The deep learning channel estimation method based on the self-attention mechanism in the mimo ofdm system of the present invention is compared with the existing channel estimation method in the system by using simulation. The indexes compared for measuring the accuracy of channel estimation are Mean Square Error (MSE) and Bit Error Rate (BER).
The simulation of the deep learning channel estimation method based on the self-attention mechanism in the multi-input multi-output orthogonal frequency division multiplexing system of the embodiment is specifically set as follows:
for simulation of different signal-to-noise ratios, the number of transmitting antennas is 2, the number of receiving antennas is 2, the number of subcarriers is 64, a guard interval is a cyclic prefix, the ratio of the cyclic prefix is 1/4, a modulation mode is quadrature phase shift keying, noise is additive white gaussian noise, transmitting power is normalized, and the signal-to-noise ratio is expressed in a logarithmic function mode.
Fig. 3 is a comparison result of mean square error between the method of the present invention and the existing estimation method under different snr, wherein the top solid line marked by C3 indicates that the existing estimation method is based on the least square method, the solid line marked by C2 indicates that the existing estimation method is based on the super-resolution deep learning method, and the bottom solid line marked by C1 indicates the method of the present invention. As can be seen from fig. 3, the mean square error of the mimo ofdm system using the method of the present invention is smaller than that using the least square method and the super-resolution deep learning method. At a high signal-to-noise ratio, the method has a gain of about 1.1dB relative to a super-resolution deep learning-based method, and at a low signal-to-noise ratio, the method has a gain of about 0.8 dB.
Fig. 4 is a comparison result of bit error rates of the method of the present invention and the existing estimation method under different signal-to-noise ratios, in which the top solid line marked by D3 indicates that the existing estimation method is based on the least square method, the middle solid line marked by D2 indicates that the existing estimation method is based on the super-resolution deep learning method, and the bottom solid line marked by D1 indicates the method of the present invention. As can be seen from fig. 4, the bit error rate of the mimo ofdm system using the method of the present invention is smaller than that using the least square method and the super-resolution deep learning method. And with the increase of the signal-to-noise ratio, the performance gap between the performance of the method and the performance of the existing estimation method is gradually widened.
Through the above implementation example, it is proved that, compared with the existing channel estimation method, the deep learning channel estimation based on the self-attention mechanism of the present invention in the mimo-ofdm system has more accurate channel estimation result due to the effective utilization of the correlation between channels, and has ideal performance in both the mean square error and the bit error rate.

Claims (1)

1. A deep learning channel estimation method based on a self-attention mechanism is characterized in that:
number N for one transmitting antennatThe number of receiving antennas is NrIn the mimo ofdm system, wherein the n-th transmission endtA certain ofdm symbol on the root transmit antenna is represented as:
Figure FDA0003543619300000011
wherein L is the number of subcarriers [. ]]TRepresents transposition, for the n-thrRoot receiving antenna, k sub-carrier above itThe symbols above are represented as:
Figure FDA0003543619300000012
wherein,
Figure FDA0003543619300000013
denotes the n-thtRoot transmitting antenna and nrThe channel impulse response between the receiving antennas corresponding to the k-th sub-carrier,
Figure FDA0003543619300000014
representing additive white gaussian noise on the receive antenna; for one of the pilot symbols, all NrThe symbol on the k-th subcarrier received by the root receive antenna is represented as:
Yk＝HkXk+Zk,(k＝1,...,L)
wherein
Figure FDA0003543619300000015
Figure FDA0003543619300000016
Respectively representing a receiving matrix at the receiving end, a transmitting matrix at the transmitting end, a noise matrix,
Figure FDA0003543619300000017
representing the frequency-domain channel impulse response matrix, H, on the k-th sub-carrierkExpressed as:
Figure FDA0003543619300000018
to HkPerforming channel estimation to obtain channel impulse response estimation result of
Figure FDA0003543619300000019
Using a channel estimation method based on a least squares criterion, a cost function shown in the following formula is minimized:
Figure FDA00035436193000000110
make the cost function of the above equation about
Figure FDA00035436193000000111
Has a partial derivative of 0, i.e.
Figure FDA00035436193000000112
Obtaining the channel impulse response estimation result under the channel estimation method based on the least square criterion,
Figure FDA00035436193000000113
wherein, (.)-1An inverse matrix representing the matrix;
the specific process steps of the channel estimation method based on the deep learning of the self-attention mechanism are as follows:
step S1: estimating channel state information at pilot frequency symbol position by least square channel estimation algorithm and estimated channel matrix
Figure FDA0003543619300000021
Step S2: for the channel matrix
Figure FDA0003543619300000022
Performing up-sampling linear interpolation to obtain frequency domain channel response at data symbol, combining with frequency domain channel response at pilot frequency to obtain complete frequency domain channel response matrix, and remolding the complete frequency domain channel response matrix into large matrixIs as small as LxNsym×(Nt×NrX 2), where L is the number of subcarriers, NsymIs the number of subcarriers, NtNumber of transmitting antennas, NrIs the number of receive antennas;
step S3: extracting features of the low-resolution matrix through two convolutional neural networks to obtain a feature map of the low-resolution matrix; obtaining a weighted self-attention mechanism feature mapping through a self-attention mechanism module according to the feature map, multiplying the self-attention mechanism feature mapping by a learning coefficient, and adding the learning coefficient to the original feature map to obtain a relationship feature which is interdependent between any two positions in the feature map; the relationship characteristic is respectively passed through a convolution network and a self-attention mechanism module twice to obtain a relationship dependent characteristic which is more comprehensive in learning;
step S4: the relation dependency characteristics are processed by a convolution network, and a matrix which is consistent with the final channel matrix dimension needing to be estimated is output
Figure FDA0003543619300000023
CN202210239196.4A 2022-03-11 2022-03-11 Deep learning channel estimation method based on self-attention mechanism Active CN114598575B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210239196.4A CN114598575B (en) 2022-03-11 2022-03-11 Deep learning channel estimation method based on self-attention mechanism

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210239196.4A CN114598575B (en) 2022-03-11 2022-03-11 Deep learning channel estimation method based on self-attention mechanism

Publications (2)

Publication Number Publication Date
CN114598575A true CN114598575A (en) 2022-06-07
CN114598575B CN114598575B (en) 2024-01-09

Family

ID=81817558

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210239196.4A Active CN114598575B (en) 2022-03-11 2022-03-11 Deep learning channel estimation method based on self-attention mechanism

Country Status (1)

Country Link
CN (1) CN114598575B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115664898A (en) * 2022-10-24 2023-01-31 四川农业大学 OFDM system channel estimation method and system based on complex convolution neural network
CN118282811A (en) * 2024-06-04 2024-07-02 中科南京移动通信与计算创新研究院 MIMO channel estimation method based on AI

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100290548A1 (en) * 2007-12-25 2010-11-18 Panasonic Corporation Wireless communication apparatus, wireless communication system and wireless communication method
CN113160375A (en) * 2021-05-26 2021-07-23 郑健青 Three-dimensional reconstruction and camera pose estimation method based on multi-task learning algorithm
US11115095B1 (en) * 2020-03-05 2021-09-07 Cornell University Finite-alphabet beamforming for multi-antenna wideband systems
US20220052883A1 (en) * 2020-08-11 2022-02-17 University Of Manitoba Method for massive unsourced random access
CN114066052A (en) * 2021-11-15 2022-02-18 国网江苏省电力有限公司营销服务中心 Adjustable load power multi-step prediction method based on improved TCN correction accumulated error

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100290548A1 (en) * 2007-12-25 2010-11-18 Panasonic Corporation Wireless communication apparatus, wireless communication system and wireless communication method
US11115095B1 (en) * 2020-03-05 2021-09-07 Cornell University Finite-alphabet beamforming for multi-antenna wideband systems
US20220052883A1 (en) * 2020-08-11 2022-02-17 University Of Manitoba Method for massive unsourced random access
CN113160375A (en) * 2021-05-26 2021-07-23 郑健青 Three-dimensional reconstruction and camera pose estimation method based on multi-task learning algorithm
CN114066052A (en) * 2021-11-15 2022-02-18 国网江苏省电力有限公司营销服务中心 Adjustable load power multi-step prediction method based on improved TCN correction accumulated error

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ZHAOSIQIANG等: "Deep Learning-Based channel estimation with SRGAN in OFDM Systems", 《2021IEEE WIRELESS COMMUNICATION AND NETWORKING CONFERENCE(WCNC)》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115664898A (en) * 2022-10-24 2023-01-31 四川农业大学 OFDM system channel estimation method and system based on complex convolution neural network
CN115664898B (en) * 2022-10-24 2023-09-08 四川农业大学 OFDM system channel estimation method and system based on complex convolution neural network
CN118282811A (en) * 2024-06-04 2024-07-02 中科南京移动通信与计算创新研究院 MIMO channel estimation method based on AI

Also Published As

Publication number Publication date
CN114598575B (en) 2024-01-09

Similar Documents

Publication Publication Date Title
Li et al. Cross domain iterative detection for orthogonal time frequency space modulation
CN111404849B (en) OFDM channel estimation and signal detection method based on deep learning
Muquet et al. Subspace-based blind and semi-blind channel estimation for OFDM systems
CN114598575B (en) Deep learning channel estimation method based on self-attention mechanism
CN108881076B (en) MIMO-FBMC/OQAM system channel estimation method based on compressed sensing
JP2008529409A (en) Delay-limited channel estimation for multi-carrier systems
CN106506415B (en) A kind of method of multi-user MIMO-OFDM system channel estimation
CN106888178B (en) A kind of OFDM/OQAM system channel estimation method
US20120033683A1 (en) Method and Receiver For Jointly Decoding Received Communication Signals Using Maximum Likelihood Detection
Zhang et al. Label-assisted transmission for short packet communications: A machine learning approach
Vilas Boas et al. Artificial intelligence for channel estimation in multicarrier systems for B5G/6G communications: a survey
Ponnaluru et al. RETRACTED ARTICLE: Deep learning for estimating the channel in orthogonal frequency division multiplexing systems
Ru et al. Model-driven channel estimation for OFDM systems based on image super-resolution network
CN101132381B (en) Pilot frequency data transmission channel estimation method for MIMO-OFDM system
CN115699690A (en) Generalized orthogonal linear frequency modulated waveform
Tabatabaee et al. Novel transceiver beamforming schemes for a MIMO-GFDM system
CN109617851B (en) Channel estimation method and device based on DFT smooth filtering
Bhoyar et al. Leaky least mean square (LLMS) algorithm for channel estimation in BPSK-QPSK-PSK MIMO-OFDM system
CN111030741B (en) Precoding algorithm for interference suppression of multi-carrier system under fast time-varying scene
Hussein et al. Least Square Estimation‐Based Different Fast Fading Channel Models in MIMO‐OFDM Systems
Zhang et al. Efficient estimation of fast fading OFDM channels
Kahlon et al. Channel estimation techniques in MIMO-OFDM systems–review article
Mei et al. LMMSE channel estimation for OFDM systems with channel correlation function selection
Gao et al. An Attention-Aided Deep Neural Network Design for Channel Estimation in Massive MIMO Systems
Zheng et al. Deep learning-aided receiver against nonlinear distortion of HPA in OFDM systems

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