CN113660693B - Information transmission method applied to wireless communication system - Google Patents
Information transmission method applied to wireless communication system Download PDFInfo
- Publication number
- CN113660693B CN113660693B CN202110714427.8A CN202110714427A CN113660693B CN 113660693 B CN113660693 B CN 113660693B CN 202110714427 A CN202110714427 A CN 202110714427A CN 113660693 B CN113660693 B CN 113660693B
- Authority
- CN
- China
- Prior art keywords
- information
- convolution
- module
- network
- full
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 230000005540 biological transmission Effects 0.000 title claims abstract description 27
- 238000004891 communication Methods 0.000 title claims abstract description 24
- 230000006870 function Effects 0.000 claims abstract description 46
- 230000004913 activation Effects 0.000 claims abstract description 26
- 238000013528 artificial neural network Methods 0.000 claims abstract description 14
- 239000011159 matrix material Substances 0.000 claims description 70
- 210000002569 neuron Anatomy 0.000 claims description 21
- 238000007906 compression Methods 0.000 claims description 20
- 238000012549 training Methods 0.000 claims description 19
- 230000006835 compression Effects 0.000 claims description 18
- 238000013139 quantization Methods 0.000 claims description 18
- 238000004088 simulation Methods 0.000 claims description 8
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 230000001131 transforming effect Effects 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 239000013598 vector Substances 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000009826 distribution Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000011084 recovery Methods 0.000 description 5
- 238000001228 spectrum Methods 0.000 description 4
- 238000012795 verification Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000009365 direct transmission Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/02—Traffic management, e.g. flow control or congestion control
- H04W28/06—Optimizing the usage of the radio link, e.g. header compression, information sizing, discarding information
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
- H04L69/00—Network arrangements, protocols or services independent of the application payload and not provided for in the other groups of this subclass
- H04L69/04—Protocols for data compression, e.g. ROHC
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Computer Security & Cryptography (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses an information transmission method applied to a wireless communication system; the method comprises the following steps: the transmitting end compresses, quantizes and transmits the information to be transmitted through the encoder; the receiving end recovers the received compressed information through a decoder matched with the encoder; the encoder and the decoder are both neural networks; the decoder comprises a residual error network, wherein the residual error network comprises 5 information reconstruction modules and 1 addition module; each information reconstruction module comprises a first convolution sub-module, a full-connection sub-module and a second convolution sub-module; both convolution sub-modules adopt swish functions as activation functions; the full-connection sub-module comprises at least two full-connection layers which are adjacently connected; the adding module is used for adding the input of the residual error network and the output of the last information reconstruction module to reconstruct information. The invention can improve the information transmission performance in outdoor high-compression-rate scenes and reduce the mean square error of the reconstructed information and the original information.
Description
Technical Field
The invention belongs to the technical field of wireless communication, and particularly relates to an information transmission method applied to a wireless communication system.
Background
In a wireless communication system, transmission of a large amount of information requires consumption of a large amount of wireless spectrum resources. However, much information is information having data sparsity, and contains a large amount of 0 data. The direct transmission of such information is wasteful of precious radio spectrum resources.
In order to reduce occupation of wireless spectrum resources, a method for compressing, transmitting and reconstructing information with data sparsity by adopting a machine learning-based method has appeared in recent years; compared with the traditional compressed sensing method, the machine learning-based method has the advantages of low requirement on the sparsity of data, good robustness and low complexity.
However, the inventor finds that in the process of implementing the invention, in an outdoor high-compression-rate scene, the existing machine learning-based method shows poor performance in an information reconstruction link, and the average error (MSE) of the reconstructed information and the original information is quite different.
Disclosure of Invention
In order to solve the above problems in the prior art, the present invention provides an information transmission method applied to a wireless communication system. The technical problems to be solved by the invention are realized by the following technical scheme:
in a first aspect, the present invention provides an information transmission method applied to a wireless communication system, including:
the transmitting end compresses the information to be transmitted through the encoder to obtain compressed information; the information to be transmitted is an information matrix with data sparsity;
The transmitting end quantizes the compressed information;
the sending end sends the quantized compressed information so that the receiving end receives the quantized compressed information;
the receiving end recovers the received compressed information through a decoder matched with the encoder to obtain reconstructed information;
Wherein the encoder and the decoder are both neural networks; the decoder comprises a residual error network, wherein the residual error network comprises 5 information reconstruction modules and an addition module which are connected in sequence; each information reconstruction module comprises a first convolution sub-module, a full connection sub-module and a second convolution sub-module which are sequentially connected; the convolution units in the first convolution sub-module and the second convolution sub-module both adopt swish functions as activation functions; the full-connection sub-module comprises at least two full-connection layers which are adjacently connected; the adding module is used for adding the input of the residual error network and the output of the last information reconstruction module so as to reconstruct information.
Optionally, the wireless communication system includes: a multi-antenna FDD base station and a terminal device communicating with the FDD base station; the information to be transmitted is downlink channel information to be transmitted back to the FDD base station by the terminal equipment; the sending end is the terminal equipment, and the receiving end is the FDD base station;
before the terminal device compresses the downlink channel information through an encoder, the method further includes:
The terminal equipment estimates a downlink channel matrix of a space frequency domain according to signals sent by the FDD base station;
the terminal equipment performs two-dimensional discrete Fourier transform on the downlink channel matrix of the space frequency domain to obtain a downlink channel matrix of the time delay angle domain;
The terminal equipment extracts the real part of the front N p row elements of the downlink channel matrix of the time delay angle domain, and extracts the imaginary part of the front N p row elements of the downlink channel matrix of the time delay angle domain, so as to obtain a channel real part matrix and a channel imaginary part matrix; n p is a preset positive integer;
and the terminal equipment splices the real part matrix of the channel and the imaginary part matrix of the channel according to columns to obtain the downlink channel information.
Optionally, the encoder includes:
The first convolution unit is used for extracting two-dimensional first information features from the information to be transmitted;
A second convolution unit configured to extract a two-dimensional second information feature from the first information feature;
A first Reshape layer for transforming the second information feature into a one-dimensional second information feature;
a first fully-connected layer for compressing the one-dimensional second information feature into the compressed information of shorter length;
The compression rate of the encoder is determined according to the number of neurons of the first full-connection layer; the convolution kernel sizes of the first convolution unit and the second convolution unit are not lower than 5*5.
Optionally, the transmitting end quantizes the compressed information, including:
and the transmitting end performs mu-law quantization on the compressed information.
Optionally, the decoder includes:
the second full-connection layer is used for decompressing the compressed information received by the decoder into a third information characteristic with equal length as the one-dimensional second information characteristic;
a second Reshape layer for converting the third information feature into a two-dimensional third information feature;
The residual error network is used for recovering fourth information features required by reconstruction information from the two-dimensional third information features;
a third convolution unit configured to extract the reconstructed information from the fourth information feature;
The first convolution submodule and the second convolution submodule in the residual network both comprise convolution units with convolution kernel sizes not lower than 5*5.
Optionally, the first convolution sub-module comprises a fourth convolution unit and a fifth convolution unit which are sequentially connected; the convolution kernel size of the fourth convolution unit is 7*7, and the convolution kernel size of the fifth convolution unit is 5*5;
The full-connection sub-module comprises a third Reshape layer, a third full-connection layer, a fourth full-connection layer and a fourth Reshape layer which are sequentially connected; the number of the neurons of the fourth full-connection layer is 4 times that of the neurons of the fourth full-connection layer;
The second convolution sub-module comprises a sixth convolution unit with a convolution kernel size 3*3;
Wherein the fourth convolution unit, the fifth convolution unit, and the sixth convolution unit each employ swish functions as activation functions.
Optionally, the encoder and the decoder are trained in the following manner;
Acquiring sample information;
inputting the sample information into a coding network, so that the coding network outputs compressed sample information;
transmitting the compressed sample information to an input of a decoding network through an analog wireless channel, so that the decoding network outputs reconstructed sample information;
calculating a mean square error according to the reconstructed sample information and the originally acquired sample information;
If the mean square error is not smaller than a threshold value, adjusting network parameters of the coding network and the decoding network based on an Adam optimization algorithm, and obtaining next sample information to continue training;
and if the mean square error is smaller than a threshold value, ending training, taking the current coding network as the coder, and taking the current decoding network as the decoder.
In a second aspect, the present invention provides an information transmission apparatus applied to a wireless communication system, comprising: transmitting means and receiving means; wherein,
The transmitting device includes:
The encoder is used for compressing the information to be transmitted to obtain compressed information;
a quantizer for performing μ -law quantization on the compressed information;
A transmitter for transmitting the quantized compressed information;
the receiving apparatus includes:
a receiver for receiving the quantized compressed information;
the decoder is used for recovering the compressed information received by the receiver to obtain reconstructed information; the decoder is matched with the encoder;
Wherein the encoder and the decoder are both neural networks; the decoder comprises a residual error network, wherein the residual error network comprises 5 information reconstruction modules and an addition module which are connected in sequence; each information reconstruction module comprises a first convolution sub-module, a full connection sub-module and a second convolution sub-module which are sequentially connected; the convolution units in the first convolution sub-module and the second convolution sub-module both adopt swish functions as activation functions; the adding module is used for adding the input of the residual error network and the output of the last information reconstruction module so as to reconstruct information.
Optionally, the wireless communication system includes: a multi-antenna FDD base station and a terminal device communicating with the FDD base station; the transmitting device is a device applied to the terminal equipment, and the receiving device is a device applied to the FDD base station; the information to be transmitted is downlink channel information to be transmitted back to the FDD base station by the terminal equipment;
the terminal device further includes: a downlink channel information acquisition module;
the downlink channel information acquisition module is configured to:
Estimating a downlink channel matrix of a space frequency domain according to signals sent by the FDD base station;
Performing two-dimensional discrete Fourier transform on the downlink channel matrix of the space frequency domain to obtain a downlink channel matrix of the time delay angle domain;
Extracting real parts of front N p row elements of a downlink channel matrix of the time delay angle domain, and extracting imaginary parts of front N p row elements of the downlink channel matrix of the time delay angle domain to obtain a channel real part matrix and a channel imaginary part matrix;
and splicing the real part matrix of the channel and the imaginary part matrix of the channel according to columns to obtain the downlink channel information.
Optionally, the quantizer is specifically configured to: and carrying out mu-law quantization on the compressed information.
Optionally, the encoder and the decoder are trained in the following manner;
Acquiring sample information;
inputting the sample information into a coding network, so that the coding network outputs compressed sample information;
transmitting the compressed sample information to an input of a decoding network through an analog wireless channel, so that the decoding network outputs reconstructed sample information;
calculating a mean square error according to the reconstructed sample information and the originally acquired sample information;
If the mean square error is not smaller than a threshold value, adjusting network parameters of the coding network and the decoding network based on an Adam optimization algorithm, and obtaining next sample information to continue training;
and if the mean square error is smaller than a threshold value, ending training, taking the current coding network as the coder, and taking the current decoding network as the decoder.
The invention has the beneficial effects that:
In the information transmission method applied to the wireless communication system, the receiving end adopts a neural network, namely a decoder, to recover and reconstruct the information; the decoder comprises a residual error network, wherein the residual error network comprises 5 information reconstruction modules and an addition module which are connected in sequence; each information reconstruction module comprises a first convolution sub-module, a full connection sub-module and a second convolution sub-module which are sequentially connected; therefore, the invention adopts a neural network with deeper depth to reduce the residual error between the reconstructed information and the original information; under the condition of high compression rate, even if the receiving end loses much information due to compression, the neural network can effectively recover and reconstruct the information. In order to enable the residual error network to have stronger information recovery performance, a full-connection sub-module is additionally arranged in each information reconstruction module of the residual error network, and the full-connection sub-module comprises at least two full-connection layers which are adjacently connected; because each neuron between every two adjacent full-connection layers is connected, more comprehensive useful information can be obtained from the characteristics input into the neurons, and the full-connection sub-module is additionally arranged, so that the residual error network has stronger information recovery performance. In order to reduce negative effects caused by deepening of network depth, the convolution unit in each convolution sub-module uses swish functions as activation functions. Through simulation verification, the invention has higher performance in an outdoor high-compression-rate scene, and can control the equipartition error of the reconstructed information and the original information within an acceptable range.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
Fig. 1 is a flowchart of an information transmission method applied to a wireless communication system according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of the structure of a residual network in a decoder used in an embodiment of the present invention;
FIG. 3 is a schematic diagram of an encoder used in an embodiment of the present invention;
FIG. 4 is a graph of statistics of compressed information output by an encoder in an embodiment of the present invention;
FIG. 5 is a schematic diagram of a decoder used in an embodiment of the present invention;
FIG. 6 is a schematic diagram of a training process for an encoder and decoder used in an embodiment of the present invention;
FIG. 7 is a schematic diagram of a pair of encoders and decoders used in performing simulation verification in an embodiment of the present invention;
fig. 8 is a schematic structural diagram of an information transmission device applied to a wireless communication system according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but embodiments of the present invention are not limited thereto.
In order to improve information transmission performance in an outdoor high-compression-rate scene and reduce the mean square error of reconstructed information and original information, the embodiment of the invention provides an information transmission method applied to a wireless communication system. Referring to fig. 1, the method comprises the steps of:
S1: the transmitting end compresses the information to be transmitted through the encoder to obtain compressed information.
The encoder is a neural network and is used for extracting information characteristics from information to be transmitted, so that the information is compressed; in the embodiment of the invention, the information to be transmitted is an information matrix with data sparsity. For example, the information to be transmitted may be downlink channel information to be transmitted back to the FDD base station by the terminal device in a large-scale MIMO (Multiple-Input Multiple-Output) FDD (frequency division duplex) system; or the information to be transmitted may be a digital image.
S2: the transmitting end quantizes the compressed information.
The method for quantizing the compressed information by the transmitting end can be determined according to the data distribution condition of the compressed information; for example, when the data distribution of the compressed information has a uniform distribution trend, uniform quantization may be employed, and when the data distribution of the compressed information has a nonlinear distribution trend, non-uniform quantization may be employed.
S3: the transmitting end transmits the quantized compressed information so that the receiving end receives the quantized compressed information.
It can be understood that the transmitting end transmits the quantized compressed information to the receiving end through a wireless channel; that is, a wireless transmitting device is integrated in the transmitting end, and a wireless receiving device is integrated in the receiving end.
In addition, the transmitting end can further perform source coding and modulation on the quantized compressed information and then transmit the quantized compressed information into the space. Correspondingly, after receiving the wireless signal, the receiving end demodulates the wireless signal and decodes the information source, thereby completing the receiving of the quantized compressed information.
S4: the receiving end recovers the received compressed information through a decoder matched with the encoder to obtain reconstructed information.
It will be appreciated that the decoder that matches the encoder is also a neural network that includes a residual network; referring to fig. 2, the residual network includes 5 information reconstruction modules 101 and an addition module 102 connected in sequence; each information reconstruction module 101 comprises a first convolution sub-module, a full connection sub-module and a second convolution sub-module which are sequentially connected; the convolution units in the first convolution sub-module and the second convolution sub-module both adopt swish functions as activation functions; the full-connection sub-module comprises at least two full-connection layers which are adjacently connected; the summing block 102 is arranged to sum the input of the residual network and the output of the last information reconstruction block 101 to reconstruct the information.
When the Compression Rate (CR) of the encoder is low, such as when CR is equal to 4, the compression process does not lose much useful information, so it is sufficient to capture features and reconstruct information using a general convolutional neural network. However, when CR is high, such as when CR is equal to 64, the compression process loses much useful information, and reconstruction of the information is difficult to achieve using a general convolutional neural network. In order to solve the problem, the embodiment of the invention adopts a deeper neural network to reduce the residual error between the reconstructed information and the original information; therefore, under the condition of high compression rate, even if a receiving end loses much information due to compression, the neural network can effectively recover and reconstruct the information. In order to make the residual error network have stronger information recovery performance, a full-connection sub-module is additionally arranged in each information reconstruction module 101 of the residual error network, and the full-connection sub-module comprises at least two full-connection layers which are adjacently connected; because each neuron between every two adjacent full-connection layers is connected, more comprehensive useful information can be obtained from the characteristics input into the neurons, and the full-connection sub-module is additionally arranged, so that the residual error network has stronger information recovery performance. Therefore, the embodiment of the invention can have higher performance in an outdoor high-compression-rate scene, and the mean square error of the reconstructed information and the original information can be controlled in a smaller range. In addition, in order to reduce negative effects caused by deepening of the network depth, the embodiment of the invention uses swish functions as activation functions for convolution units in each convolution sub-module.
In one embodiment, a wireless communication system to which the method provided by the embodiment of the present invention can be applied includes: a multi-antenna FDD base station and a terminal device communicating with the FDD base station, where the wireless communication system specifically refers to a system applying a massive MIMO technology; correspondingly, the information to be transmitted is downlink channel information to be transmitted back to the FDD base station by the terminal equipment; at this time, the transmitting end is a terminal device, and the receiving end is an FDD base station.
Before the terminal device compresses the downlink channel information through the encoder, the terminal device also needs to obtain the information to be transmitted, specifically:
(1) The terminal equipment estimates a downlink channel matrix of the space frequency domain according to the signal sent by the FDD base station.
Here, the signal transmitted from the FDD base station mainly includes a precoding vector and a transmission vector on each subcarrier transmitted through the wireless channel; these two vectors are pre-agreed known quantities between the FDD base station and the terminal equipment; if y n is used to represent the signal received on the nth subcarrier of the terminal device, N e N c,Nc is the total number of subcarriers, the signal y n can be expressed as:
Wherein, Representing a precoding vector on an nth subcarrier, wherein N t is the number of antennas of the FDD base station; x n denotes a transmission vector on the nth subcarrier; z n denotes additive noise and interference on the nth subcarrier of the terminal device; The channel frequency domain response vector representing the nth subcarrier, the channel frequency domain response vector of N c subcarriers forms a downlink channel matrix/> (. Cndot.) H represents the conjugate transpose, symbol/>Representing the meaning of the collection.
(2) And performing two-dimensional discrete Fourier transform on the downlink channel matrix of the space frequency domain to obtain the downlink channel matrix of the time delay angle domain.
Specifically, the following two-dimensional discrete fourier transform is performed on the downlink channel matrix:
Wherein, And/>Is a DFT (Discrete Fourier Transform ) matrix pre-calculated according to N c and N t, and H represents a downlink channel matrix of the delay angle domain.
(3) Extracting real parts of front N p rows of elements in a downlink channel matrix of a time delay angle domain, and extracting imaginary parts of front N p rows of elements in the downlink channel matrix of the time delay angle domain to obtain a channel real part matrix and a channel imaginary part matrix; n p is a preset positive integer, and N p can be preset by manual experience according to an actual system.
Since the downlink channel matrix of massive MIMO is sparse in the delay angle domain and the multipath delay is limited, most of the elements in the downlink channel matrix of the delay angle domain are approximately 0, and only the first N p rows have non-zero values; therefore, the first N p rows in the total N c rows of the downlink channel matrix of the time delay angle domain can be extracted; in addition, to reduce the complexity of the encoder and decoder, the real and imaginary parts of the N p rows of elements may be separated, and the channel real matrix and the channel imaginary matrix are each of size N p×Nt.
(4) And splicing the real part matrix and the imaginary part matrix of the channel according to the columns to obtain downlink channel information.
Specifically, the real part matrix and the imaginary part matrix of the channel are spliced according to the columns to obtain a matrix of N p×Nt multiplied by 2, and the matrix is used as downlink channel information.
It can be understood that, in an FDD system, for an uplink transmission link, an FDD base station can obtain accurate information through a pilot signal transmitted from a terminal device; for the downlink transmission, the terminal device needs to transmit back downlink channel information to the FDD base station via the uplink. With the development of communication technology, the massive MIMO technology becomes a key technology of a future mobile communication system due to higher channel capacity and lower inter-user interference; the amount of downlink channel information of an FDD system to which the MIMO technology is applied is proportional to the number of antennas of an FDD base station, and the greater the number of antennas of the FDD base station, the greater the amount of downlink channel information, and a large amount of spectrum resources need to be consumed. Therefore, the method provided by the embodiment of the invention is suitable for compressing the downlink channel information.
The encoder and decoder used in the embodiments of the present invention are respectively exemplified below.
The encoder is first illustrated. Referring to fig. 3, the encoder may include: a first convolution unit 20, a second convolution unit 30, a first Reshape layer 40, and a first full connection layer 50.
Wherein,
A first convolution unit 20 for extracting a two-dimensional first information feature from the information to be transmitted.
A second convolution unit 30 for extracting a two-dimensional second information feature from the first information feature.
A first Reshape layer 40 for transforming the second information feature into a one-dimensional second information feature.
A first fully connected layer 50 for compressing the one-dimensional second information feature into compressed information of a shorter length.
Wherein the compression ratio of the encoder is determined according to the number of neurons of the first full connection layer 50; if the first fully connected layer 50 has M neurons, the Compression Rate (CR) of the encoder can be calculated as:
And, the convolution kernel sizes of the first convolution unit 20 and the second convolution unit 30 are not less than 5*5. The purpose of this is to capture more non-zero features using a larger convolution kernel size, thereby extracting more useful information; preferably, the convolution kernel sizes of the first convolution unit 20 and the second convolution unit 30 may be 7*7. In addition, the convolution kernel sizes of the first convolution unit 20 and the second convolution unit 30 may be equal or unequal; when the convolution kernels of the first convolution unit 20 and the second convolution unit 30 are not equal in size, the convolution kernel of the first convolution unit 20 may be larger than the convolution kernel of the second convolution unit 30.
Wherein the first convolution unit 20 and the second convolution unit 30 each comprise: convolutional layer, BN (Batch Normalization ) layer, and active layer. Wherein the convolution layer, the BN layer and the activation layer are sequentially connected; the activation layer may use a ReLU (RECTIFIED LINEAR Unit, linear activation Unit) function as the activation function, or swish function as the activation function in concert with the encoder. swish activation functions are expressed as follows:
where x represents the input of an activation function, e is a natural base, and sigmoid (·) is another common activation function.
Based on the coding network shown above, statistics is performed on the cumulative probability of the compressed information in different values, the statistical result is shown in fig. 4, it can be seen that the compressed information is in nonlinear distribution, and the value close to zero in the compressed information is the vast majority. Therefore, embodiments of the present invention preferably employ μ -law quantization methods to quantize the compression.
The decoder is illustrated below. Referring to fig. 5, the decoder may include: a second full connection layer 60, a second Reshape layer 70, a residual network 10 and a third convolution unit 80.
The second full connection layer 60 is configured to decompress the compressed information received by the decoder into a third information feature equal in length to the one-dimensional second information feature.
It will be appreciated that the number of neurons of the second fully connected layer 60 is equal to the length of the one-dimensional second information feature, so that a third information feature equal to the length of the one-dimensional second information feature can be output; the third information feature is also a one-dimensional feature.
A second Reshape layer 70 for converting the one-dimensional third information feature to a two-dimensional third information feature.
A residual network 10 for recovering a fourth information feature required for reconstructing the information from the two-dimensional third information feature.
Specifically, the residual network 10 first reconstructs the information by 5 information reconstruction modules 101; the input of the residual network 10, i.e. the input of the first information reconstruction module 101, is then added to the output of the last information reconstruction module 101 by an addition module 102, resulting in a fourth information feature required for reconstructing the information.
The first convolution sub-module and the second convolution sub-module of each information reconstruction module 101 in the residual network 10 each include a convolution unit with a convolution kernel size not less than 5*5. Preferably, the convolution kernel size of the convolution unit in the first convolution sub-module is 7*7, and the convolution kernel size of the convolution unit in the second convolution sub-module is 5*5; in this way, a larger convolution kernel is employed to reduce the residual of the reconstructed information from the original information.
A third convolution unit 80 for extracting reconstructed information from the fourth information feature. The third convolution unit 80 also includes: a convolutional layer, a BN layer, and an active layer.
In an alternative implementation, the first convolution sub-module in the residual network 10 includes a fourth convolution unit and a fifth convolution unit that are sequentially connected; the convolution kernel size of the fourth convolution unit is 7*7, and the convolution kernel size of the fifth convolution unit is 5*5. The fully-connected sub-module in the residual error network 10 comprises a third Reshape layer, a third fully-connected layer, a fourth fully-connected layer and a fourth Reshape layer which are sequentially connected; the number of the neurons of the fourth full-connection layer is 4 times that of the neurons of the fourth full-connection layer; the second convolution sub-module in the residual network 10 comprises a sixth convolution unit having a convolution kernel size 3*3.
Wherein the fourth convolution unit, the fifth convolution unit, the sixth convolution unit and the third convolution unit 80 following the residual network 10 each use swish functions as activation functions. Similarly, the fourth convolution unit, the fifth convolution unit, and the sixth convolution unit each include: a convolutional layer, a BN layer, and an active layer that uses swish functions as the active functions. Here, the embodiment of the present invention adopts swish functions as the activation functions of the convolutional layers, mainly because the decoding network used in the embodiment of the present invention has a larger depth, and when the common ReLU function is used as the activation function, the gradient disappears, and after the swish function is tried to be used, the effect is found to be effectively reduced, so the embodiment of the present invention adopts swish functions as the activation functions of the convolutional layers.
It can be seen that each information reconstruction module 101 includes two adjacent full connection layers of the third full connection layer and the fourth full connection layer, so that more comprehensive useful information can be obtained from the features input into the information reconstruction module 101, and therefore, adding the 5*2 full connection layers in the residual error network 10 can enable the residual error network 10 to have stronger information recovery performance.
In addition, all the full connection layers in the embodiment of the invention adopt a sigmoid function as an activation function.
The following describes the training mode of the encoder and decoder. In order to avoid labeling training samples and reduce network errors, the embodiment of the invention adopts an end-to-end training mode to synchronously train the encoder and the decoder. Referring to fig. 6, a specific training process may include:
s601: sample information is acquired.
S602: sample information is input to the encoding network such that the encoding network outputs compressed sample information.
S603: the compressed sample information is transmitted over the analog wireless channel to an input of the decoding network to cause the decoding network to output reconstructed sample information.
The simulated wireless channel can be simulated by matlab software, and in the related art, the simulation of wireless channels such as Rayleigh channel and multipath channel is performed based on matlab, and the simulation can be selected and used according to the actual wireless channel of the system, which is not described in detail in the embodiment of the invention.
S604: a Mean Square Error (MSE) is calculated from the reconstructed sample information and the originally acquired sample information. The calculation formula is as follows:
Where H i represents the i-th sample information originally acquired, Representing sample information reconstructed for the ith sample information, the term 2 is a euclidean norm, N represents the total number of sample information that have currently been trained.
S605: and if the calculated mean square error is not smaller than the threshold value, adjusting network parameters of the coding network and the decoding network based on an Adam optimization algorithm, and obtaining next sample information to continue training.
Here, after the next sample information is acquired, the process returns to step S602 to continue execution.
S606: and if the calculated mean square error is smaller than the threshold value, ending training, taking the current coding network as an encoder, and taking the current decoding network as a decoder.
The wireless transmission process is innovatively incorporated into the training process of the coding network and the decoding network, so that the two neural networks can learn the transmission characteristics of the wireless channel, and the coding network and the decoding network after training are more accurate.
The simulation verification situation of the information transmission method provided by the embodiment of the invention is described below. The simulation configuration includes: the FDD system works at 300MHz, the base station has N c =1024 subcarriers and N t =32 uniform linear array antennas, N p is set to be 32, so the size of downlink channel information is 32×32×2, and a wireless channel is simulated by adopting a COST 2100MIMO channel model; the network structure of the encoder and decoder is shown in fig. 7. In fig. 7, "Conv" represents the meaning of convolution, and "Refine" represents the information reconstruction module 101; "FC" means the meaning of a fully attached layer; the grey filled rectangular blocks represent features that are input to the layers, and the numbers next to them represent the size of the features.
In the encoder shown in fig. 7, from left to right, the 1 st block is the first convolution unit 20, the 2 nd block is the second convolution unit 30, the 3 rd block is the first Reshape th layer 40, and the 4 th block is the first full connection layer 50.
In the decoder shown in fig. 7, from right to left, the 1 st module is the second full connection layer 60, the 2 nd module is the second Reshape th layer 70, the 3 rd, 4 th, 5 th, 6 th and 7 th modules are the 5 information reconstruction modules 101, the 8 th module is the addition module 102, and the 9 th module is the third convolution unit 80, respectively; in the partial enlarged view of the information reconstruction module 101 represented by the 4 th module, the 1 st module is a fourth convolution unit in the first convolution sub-module, and the 2 nd module is a fifth convolution unit in the first convolution sub-module; the 3 rd module is a third Reshape th layer in the full-connection sub-module, the 4 th module is a third full-connection layer in the full-connection sub-module, the 5 th module is a fourth full-connection layer in the full-connection sub-module, and the 6 th module is a fourth Reshape th layer in the full-connection sub-module; the 7 th module is a sixth convolution unit of the second convolution sub-module.
First, without considering quantization and noise, the performance of the information transmission method provided by the embodiment of the present invention is shown in table 1:
TABLE 1
| CR | NMSE(dB) | ρ |
| 8 | -17.88 | 0.97 |
| 16 | -14.7 | 0.96 |
| 32 | -14.42 | 0.96 |
| 64 | -11.34 | 0.94 |
The NMSE (normalized mean square error) is calculated as follows:
e {.cndot. } represents the desire to find, the meaning of the remaining parameters is described above.
Ρ is a parameter for measuring the effect of the beamforming vector, and the larger ρ is, the more consistent the correlation between each h recovered by the decoder and the correlation between the true h are; ρ is defined as follows:
as can be seen from table 1, in the embodiment of the present invention, the NMSE of the embodiment of the present invention is always kept below-10 dB under different high compression rates, and ρ is also higher than the existing machine learning based method; existing machine learning based methods can only ensure that p is close to 0.9 at compression ratios below 16, whereas p will deteriorate to 0.7 or even 0.6 at compression ratios above 16. The ρ of the embodiment of the present invention at high compression ratio is also maintained above 0.9.
In consideration of noise interference, the performance of the information transmission method provided by the embodiment of the present invention is shown in table 2:
TABLE 2
Wherein S1 represents the signal-to-noise ratio set when the coding network and the decoding network are trained, S2 represents the signal-to-noise ratio set when the coding network and the decoding network are trained, and the values in table 2 are NMSE. It can be seen from table 2 that the robustness to noise is also relatively strong in the embodiments of the present invention.
In the case where quantization is considered, that is, in the presence of quantization noise, the performance of the information transmission method provided in the embodiment of the present invention is shown in table 3:
TABLE 3 Table 3
| CR | NMSE(UQ) | ρ(UQ) | NMSE(μ) | ρ(μ) |
| 8 | -12.88 | 0.95 | -13.5 | 0.96 |
| 16 | -11.23 | 0.94 | -11.42 | 0.94 |
| 32 | -9.82 | 0.93 | -9.87 | 0.93 |
| 64 | -8.26 | 0.91 | -8.05 | 0.91 |
In table 3, UQ represents uniform quantization, and μ represents μ -law quantization. It can be seen from table 3 that the performance of the inventive examples is slightly degraded in the presence of quantization noise, but still falls within the higher performance category. Furthermore, as can be seen from table 3, the performance of the inventive examples under μ -law quantization is slightly higher than that under uniform quantization.
Based on the same inventive concept, an embodiment of the present invention further provides an information transmission apparatus applied to a wireless communication system, as shown in fig. 8, including: a transmitting means 81 and a receiving means 82.
Wherein the transmitting device 81 comprises:
an encoder 811 for compressing information to be transmitted to obtain compressed information;
a quantizer 812 for μ -law quantization of the compressed information;
a transmitter 813 for transmitting the quantized compressed information;
The receiving device 82 includes:
A receiver 821 for receiving the quantized compressed information;
a decoder 822, configured to recover the compressed information received by the receiver to obtain reconstructed information; the decoder is matched with the encoder;
Wherein, the encoder and the decoder are both neural networks; the decoder comprises a residual error network 10, wherein the residual error network 10 comprises 5 information reconstruction modules 101 and an addition module 102 which are connected in sequence; each information reconstruction module 101 comprises a first convolution sub-module, a full connection sub-module and a second convolution sub-module which are sequentially connected; the convolution units in the first convolution sub-module and the second convolution sub-module both adopt swish functions as activation functions; the summing block 102 is arranged to sum the input of the residual network 10 and the output of the last information reconstruction block 101 to reconstruct the information.
Optionally, the wireless communication system includes: a multi-antenna FDD base station and a terminal device communicating with the FDD base station; the transmitting device 81 is a device applied to a terminal device, and the receiving device 82 is a device applied to an FDD base station; the information to be transmitted is downlink channel information to be transmitted back to the FDD base station by the terminal equipment;
the terminal device further includes: a downlink channel information acquisition module;
a downlink channel information acquisition module, configured to:
estimating a downlink channel matrix of a space frequency domain according to signals sent by an FDD base station;
Performing two-dimensional discrete Fourier transform on the downlink channel matrix of the space frequency domain to obtain a downlink channel matrix of the time delay angle domain;
Extracting real parts of front N p row elements of a downlink channel matrix of a time delay angle domain, and extracting imaginary parts of front N p row elements of the downlink channel matrix of the time delay angle domain to obtain a channel real part matrix and a channel imaginary part matrix;
and splicing the real part matrix and the imaginary part matrix of the channel according to the columns to obtain downlink channel information.
Optionally, the encoder 811 includes:
The first convolution unit is used for extracting two-dimensional first information features from information to be transmitted;
A second convolution unit 30 for extracting a two-dimensional second information feature from the first information feature;
a first Reshape layer 40 for transforming the second information feature into a one-dimensional second information feature;
A first fully-connected layer 50 for compressing the one-dimensional second information feature into compressed information of a shorter length;
Wherein the compression ratio of the encoder is determined according to the number of neurons of the first full connection layer 50; the convolution kernel sizes of the first convolution unit 20 and the second convolution unit 30 are not less than 5*5.
Optionally, quantizer 812 is specifically configured to: mu-law quantization is performed on the compressed information.
Optionally, the decoder 822 includes:
A second full connection layer 60 for decompressing the compressed information received by the decoder into a third information feature of equal length as the one-dimensional second information feature;
A second Reshape layer 70 to convert the third information feature to a two-dimensional third information feature;
A residual network 10 for recovering a fourth information feature required for reconstructing the information from the two-dimensional third information feature;
a third convolution unit 80 for extracting reconstructed information from the fourth information feature;
the first convolution submodule and the second convolution submodule in the residual network 10 each include a convolution unit with a convolution kernel size not lower than 5*5.
Optionally, the first convolution sub-module comprises a fourth convolution unit and a fifth convolution unit which are sequentially connected; the convolution kernel size of the fourth convolution unit is 7*7, and the convolution kernel size of the fifth convolution unit is 5*5;
The full-connection sub-module comprises a third Reshape layer, a third full-connection layer, a fourth full-connection layer and a fourth Reshape layer which are sequentially connected; the number of the neurons of the fourth full-connection layer is 4 times that of the neurons of the fourth full-connection layer;
A second convolution sub-module comprising a sixth convolution unit having a convolution kernel size 3*3;
wherein the fourth convolution unit, the fifth convolution unit and the sixth convolution unit all use swish functions as activation functions.
Optionally, the encoder 811 and decoder 821 are trained in the following manner;
Acquiring sample information;
inputting the sample information into the encoding network so that the encoding network outputs compressed sample information;
transmitting the compressed sample information to an input of a decoding network through an analog wireless channel, so that the decoding network outputs reconstructed sample information;
calculating a mean square error according to the reconstructed sample information and the originally acquired sample information;
if the mean square error is not smaller than the threshold value, adjusting network parameters of the coding network and the decoding network based on an Adam optimization algorithm, and obtaining next sample information to continue training;
if the mean square error is less than the threshold, the training is ended, the current coding network is taken as the encoder 811, and the current decoding network is taken as the decoder 821.
For the device embodiments, the description is relatively simple as it is substantially similar to the method embodiments, and reference is made to the description of the method embodiments for relevant points.
It should be noted that the terms "first," "second," and "second" are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implying a number of technical features being indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Further, one skilled in the art can engage and combine the different embodiments or examples described in this specification.
Although the invention is described herein in connection with various embodiments, other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims. In the description, the word "comprising" does not exclude other elements or steps, and the "a" or "an" does not exclude a plurality. Some measures are described in mutually different embodiments, but this does not mean that these measures cannot be combined to produce a good effect.
The foregoing is a further detailed description of the invention in connection with the preferred embodiments, and it is not intended that the invention be limited to the specific embodiments described. It will be apparent to those skilled in the art that several simple deductions or substitutions may be made without departing from the spirit of the invention, and these should be considered to be within the scope of the invention.
Claims (5)
1. An information transmission method applied to a wireless communication system, comprising:
the transmitting end compresses the information to be transmitted through the encoder to obtain compressed information; the information to be transmitted is an information matrix with data sparsity;
The transmitting end quantizes the compressed information;
the sending end sends the quantized compressed information so that the receiving end receives the quantized compressed information;
the receiving end recovers the received compressed information through a decoder matched with the encoder to obtain reconstructed information;
Wherein the encoder and the decoder are both neural networks; the decoder comprises a residual error network, wherein the residual error network comprises 5 information reconstruction modules and an addition module which are connected in sequence; each information reconstruction module comprises a first convolution sub-module, a full connection sub-module and a second convolution sub-module which are sequentially connected; the convolution units in the first convolution sub-module and the second convolution sub-module both adopt swish functions as activation functions; the full-connection sub-module comprises at least two full-connection layers which are adjacently connected; the adding module is used for adding the input of the residual error network and the output of the last information reconstruction module so as to reconstruct information;
wherein the encoder and the decoder are trained and obtained in the following manner:
Acquiring sample information;
inputting the sample information into a coding network, so that the coding network outputs compressed sample information;
transmitting the compressed sample information to an input of a decoding network through an analog wireless channel, so that the decoding network outputs reconstructed sample information; the simulated wireless channel is realized by using matlab software in a simulation way;
calculating a mean square error according to the reconstructed sample information and the originally acquired sample information;
If the mean square error is not smaller than a threshold value, adjusting network parameters of the coding network and the decoding network based on an Adam optimization algorithm, and obtaining next sample information to continue training;
if the mean square error is smaller than a threshold value, training is finished, a current coding network is used as the coder, and a current decoding network is used as the decoder;
the encoder includes:
The first convolution unit is used for extracting two-dimensional first information features from the information to be transmitted;
A second convolution unit configured to extract a two-dimensional second information feature from the first information feature;
A first Reshape layer for transforming the second information feature into a one-dimensional second information feature;
a first fully-connected layer for compressing the one-dimensional second information feature into the compressed information of shorter length;
The compression rate of the encoder is determined according to the number of neurons of the first full-connection layer; the convolution kernel sizes of the first convolution unit and the second convolution unit are not lower than 5*5; the compression ratio is equal to 64;
The decoder includes:
the second full-connection layer is used for decompressing the compressed information received by the decoder into a third information characteristic with equal length as the one-dimensional second information characteristic;
a second Reshape layer for converting the third information feature into a two-dimensional third information feature;
The residual error network is used for recovering fourth information features required by reconstruction information from the two-dimensional third information features;
a third convolution unit configured to extract the reconstructed information from the fourth information feature;
The first convolution submodule and the second convolution submodule in the residual network both comprise convolution units with convolution kernel sizes not lower than 5*5;
The first convolution sub-module comprises a fourth convolution unit and a fifth convolution unit which are sequentially connected; the convolution kernel size of the fourth convolution unit is 7*7, and the convolution kernel size of the fifth convolution unit is 5*5;
The full-connection sub-module comprises a third Reshape layer, a third full-connection layer, a fourth full-connection layer and a fourth Reshape layer which are sequentially connected; the number of the neurons of the fourth full-connection layer is 4 times that of the neurons of the fourth full-connection layer;
The second convolution sub-module comprises a sixth convolution unit with a convolution kernel size 3*3;
Wherein the fourth convolution unit, the fifth convolution unit, and the sixth convolution unit each employ swish functions as activation functions.
2. The method of claim 1, wherein the wireless communication system comprises: a multi-antenna FDD base station and a terminal device communicating with the FDD base station; the information to be transmitted is downlink channel information to be transmitted back to the FDD base station by the terminal equipment; the sending end is the terminal equipment, and the receiving end is the FDD base station;
before the terminal device compresses the downlink channel information through an encoder, the method further includes:
The terminal equipment estimates a downlink channel matrix of a space frequency domain according to signals sent by the FDD base station;
the terminal equipment performs two-dimensional discrete Fourier transform on the downlink channel matrix of the space frequency domain to obtain a downlink channel matrix of the time delay angle domain;
The terminal equipment extracts the real part of the front N p rows of elements in the downlink channel matrix of the time delay angle domain, and extracts the imaginary part of the front N p rows of elements in the downlink channel matrix of the time delay angle domain, so as to obtain a channel real part matrix and a channel imaginary part matrix; n p is a preset positive integer;
and the terminal equipment splices the real part matrix of the channel and the imaginary part matrix of the channel according to columns to obtain the downlink channel information.
3. The method of claim 1, wherein the transmitting end quantizes the compressed information, comprising:
and the transmitting end performs mu-law quantization on the compressed information.
4. An information transmission apparatus applied to a wireless communication system, comprising: transmitting means and receiving means; wherein,
The transmitting device includes:
The encoder is used for compressing the information to be transmitted to obtain compressed information;
a quantizer for performing μ -law quantization on the compressed information;
A transmitter for transmitting the quantized compressed information;
the receiving apparatus includes:
a receiver for receiving the quantized compressed information;
the decoder is used for recovering the compressed information received by the receiver to obtain reconstructed information; the decoder is matched with the encoder;
Wherein the encoder and the decoder are both neural networks; the decoder comprises a residual error network, wherein the residual error network comprises 5 information reconstruction modules and an addition module which are connected in sequence; each information reconstruction module comprises a first convolution sub-module, a full connection sub-module and a second convolution sub-module which are sequentially connected; the convolution units in the first convolution sub-module and the second convolution sub-module both adopt swish functions as activation functions; the adding module is used for adding the input of the residual error network and the output of the last information reconstruction module so as to reconstruct information;
Wherein the encoder and the decoder are trained in the following manner;
Acquiring sample information;
inputting the sample information into a coding network, so that the coding network outputs compressed sample information;
transmitting the compressed sample information to an input of a decoding network through an analog wireless channel, so that the decoding network outputs reconstructed sample information; the simulated wireless channel is realized by using matlab software in a simulation way;
calculating a mean square error according to the reconstructed sample information and the originally acquired sample information;
If the mean square error is not smaller than a threshold value, adjusting network parameters of the coding network and the decoding network based on an Adam optimization algorithm, and obtaining next sample information to continue training;
if the mean square error is smaller than a threshold value, training is finished, a current coding network is used as the coder, and a current decoding network is used as the decoder;
the encoder includes:
The first convolution unit is used for extracting two-dimensional first information features from the information to be transmitted;
A second convolution unit configured to extract a two-dimensional second information feature from the first information feature;
A first Reshape layer for transforming the second information feature into a one-dimensional second information feature;
a first fully-connected layer for compressing the one-dimensional second information feature into the compressed information of shorter length;
The compression rate of the encoder is determined according to the number of neurons of the first full-connection layer; the convolution kernel sizes of the first convolution unit and the second convolution unit are not lower than 5*5; the compression ratio is equal to 64;
The decoder includes:
the second full-connection layer is used for decompressing the compressed information received by the decoder into a third information characteristic with equal length as the one-dimensional second information characteristic;
a second Reshape layer for converting the third information feature into a two-dimensional third information feature;
The residual error network is used for recovering fourth information features required by reconstruction information from the two-dimensional third information features;
a third convolution unit configured to extract the reconstructed information from the fourth information feature;
The first convolution submodule and the second convolution submodule in the residual network both comprise convolution units with convolution kernel sizes not lower than 5*5;
The first convolution sub-module comprises a fourth convolution unit and a fifth convolution unit which are sequentially connected; the convolution kernel size of the fourth convolution unit is 7*7, and the convolution kernel size of the fifth convolution unit is 5*5;
The full-connection sub-module comprises a third Reshape layer, a third full-connection layer, a fourth full-connection layer and a fourth Reshape layer which are sequentially connected; the number of the neurons of the fourth full-connection layer is 4 times that of the neurons of the fourth full-connection layer;
The second convolution sub-module comprises a sixth convolution unit with a convolution kernel size 3*3;
Wherein the fourth convolution unit, the fifth convolution unit, and the sixth convolution unit each employ swish functions as activation functions.
5. The apparatus of claim 4, wherein the wireless communication system comprises: a multi-antenna FDD base station and a terminal device communicating with the FDD base station; the transmitting device is a device applied to the terminal equipment, and the receiving device is a device applied to the FDD base station; the information to be transmitted is downlink channel information to be transmitted back to the FDD base station by the terminal equipment;
the terminal device further includes: a downlink channel information acquisition module;
the downlink channel information acquisition module is configured to:
Estimating a downlink channel matrix of a space frequency domain according to signals sent by the FDD base station;
Performing two-dimensional discrete Fourier transform on the downlink channel matrix of the space frequency domain to obtain a downlink channel matrix of the time delay angle domain;
Extracting real parts of front N p row elements of a downlink channel matrix of the time delay angle domain, and extracting imaginary parts of front N p row elements of the downlink channel matrix of the time delay angle domain to obtain a channel real part matrix and a channel imaginary part matrix;
and splicing the real part matrix of the channel and the imaginary part matrix of the channel according to columns to obtain the downlink channel information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110714427.8A CN113660693B (en) | 2021-06-25 | 2021-06-25 | Information transmission method applied to wireless communication system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110714427.8A CN113660693B (en) | 2021-06-25 | 2021-06-25 | Information transmission method applied to wireless communication system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113660693A CN113660693A (en) | 2021-11-16 |
| CN113660693B true CN113660693B (en) | 2024-06-11 |
Family
ID=78489003
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110714427.8A Active CN113660693B (en) | 2021-06-25 | 2021-06-25 | Information transmission method applied to wireless communication system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113660693B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108390706A (en) * | 2018-01-30 | 2018-08-10 | 东南大学 | A kind of extensive mimo channel state information feedback method based on deep learning |
| CN108847876A (en) * | 2018-07-26 | 2018-11-20 | 东南大学 | A kind of extensive MIMO time varying channel states Information Compression feedback and method for reconstructing |
| CN110311718A (en) * | 2019-07-05 | 2019-10-08 | 东南大学 | Quantization and inverse quantization method in a kind of extensive mimo channel status information feedback |
| CN111555781A (en) * | 2020-04-27 | 2020-08-18 | 天津大学 | Large-scale MIMO channel state information compression and reconstruction method based on deep learning attention mechanism |
| CN112737985A (en) * | 2020-12-25 | 2021-04-30 | 东南大学 | Large-scale MIMO channel joint estimation and feedback method based on deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI687063B (en) * | 2019-01-04 | 2020-03-01 | 財團法人工業技術研究院 | A communication system and codec method based on deep learning and channel state information |
-
2021
- 2021-06-25 CN CN202110714427.8A patent/CN113660693B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108390706A (en) * | 2018-01-30 | 2018-08-10 | 东南大学 | A kind of extensive mimo channel state information feedback method based on deep learning |
| CN108847876A (en) * | 2018-07-26 | 2018-11-20 | 东南大学 | A kind of extensive MIMO time varying channel states Information Compression feedback and method for reconstructing |
| CN110311718A (en) * | 2019-07-05 | 2019-10-08 | 东南大学 | Quantization and inverse quantization method in a kind of extensive mimo channel status information feedback |
| CN111555781A (en) * | 2020-04-27 | 2020-08-18 | 天津大学 | Large-scale MIMO channel state information compression and reconstruction method based on deep learning attention mechanism |
| CN112737985A (en) * | 2020-12-25 | 2021-04-30 | 东南大学 | Large-scale MIMO channel joint estimation and feedback method based on deep learning |
Non-Patent Citations (4)
| Title |
|---|
| Bit-Level Optimized Neural Network for Multi-Antenna Channel Quantization;Chao Lu等;《IEEE Wireless Communications Letters》;第09卷(第01期);87-90 * |
| Convolutional Neural Network-Based Multiple-Rate Compressive Sensing for Massive MIMO CSI Feedback: Design, Simulation, and Analysis;Jiajia Guo等;《IEEE Transactions on Wireless Communications》;第19卷(第04期);2827 -2840 * |
| 基于深度学习的大规模MIMO系统信道状态信息反馈研究;2020-12-16;《北京邮电大学博士论文》;全文 * |
| 大规模MIMO系统中基于CNN的延迟CSI反馈改进算法;王月;段红光;郑兴林;;电讯技术(第07期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113660693A (en) | 2021-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112737985B (en) | Large-scale MIMO channel joint estimation and feedback method based on deep learning | |
| CN110311718B (en) | Quantization and inverse quantization method in massive MIMO channel state information feedback | |
| CN108390706B (en) | Large-scale MIMO channel state information feedback method based on deep learning | |
| CN110350958B (en) | CSI multi-time rate compression feedback method of large-scale MIMO based on neural network | |
| CN109672464B (en) | FCFNN-based large-scale MIMO channel state information feedback method | |
| Liu et al. | A Markovian model-driven deep learning framework for massive MIMO CSI feedback | |
| CN108847876B (en) | Large-scale MIMO time-varying channel state information compression feedback and reconstruction method | |
| CN111464220B (en) | Channel state information reconstruction method based on deep learning | |
| Liu et al. | An efficient deep learning framework for low rate massive MIMO CSI reporting | |
| Yang et al. | Deep convolutional compression for massive MIMO CSI feedback | |
| Cai et al. | Attention model for massive MIMO CSI compression feedback and recovery | |
| CN113098804B (en) | Channel state information feedback method based on deep learning and entropy coding | |
| CN113691288B (en) | Joint pilot frequency, feedback and multi-user hybrid coding method based on deep learning | |
| CN113872652B (en) | CSI feedback method based on 3D MIMO time-varying system | |
| CN113726375B (en) | Channel information compression feedback reconstruction method based on deep learning | |
| CN116248156A (en) | Deep learning-based large-scale MIMO channel state information feedback and reconstruction method | |
| CN113660020A (en) | Wireless communication channel information transmission method, system and decoder | |
| CN115865145A (en) | Large-scale MIMO channel state information feedback method based on Transformer | |
| CN114884549A (en) | Large-scale MIMO channel state information feedback method based on deep learning | |
| Ravula et al. | Deep autoencoder-based massive MIMO CSI feedback with quantization and entropy coding | |
| CN113660693B (en) | Information transmission method applied to wireless communication system | |
| CN112929304A (en) | Wireless channel acquisition method assisted by environment knowledge base | |
| CN112054826A (en) | Single-user low-complexity hybrid precoding method based on intermediate channel | |
| CN106789776B (en) | High-precision channel estimation method of MIMO-OFDM system | |
| Chakma et al. | Deep decoder CsiNet for FDD massive MIMO system |
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 |