CN114745230A - OTFS signal receiving and recovering method based on deep neural network structure - Google Patents
OTFS signal receiving and recovering method based on deep neural network structure Download PDFInfo
- Publication number
- CN114745230A CN114745230A CN202210233419.6A CN202210233419A CN114745230A CN 114745230 A CN114745230 A CN 114745230A CN 202210233419 A CN202210233419 A CN 202210233419A CN 114745230 A CN114745230 A CN 114745230A
- Authority
- CN
- China
- Prior art keywords
- data
- training
- otfs
- channel
- network
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 90
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims abstract description 33
- 238000007781 pre-processing Methods 0.000 claims abstract description 17
- 238000012360 testing method Methods 0.000 claims description 51
- 238000011084 recovery Methods 0.000 claims description 19
- 238000010606 normalization Methods 0.000 claims description 15
- 238000013507 mapping Methods 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 10
- 125000004122 cyclic group Chemical group 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 5
- 238000012805 post-processing Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 230000003287 optical effect Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000002093 peripheral effect Effects 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 abstract description 22
- 238000007476 Maximum Likelihood Methods 0.000 abstract description 3
- 230000006854 communication Effects 0.000 description 17
- 238000004891 communication Methods 0.000 description 15
- 238000013135 deep learning Methods 0.000 description 8
- 230000005540 biological transmission Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 238000005562 fading Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000009825 accumulation Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000010295 mobile communication Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 239000000969 carrier Substances 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000007493 shaping process Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000010363 phase shift Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
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
- H04L25/0254—Channel estimation channel estimation algorithms using neural network algorithms
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/26—Systems using multi-frequency codes
- H04L27/2601—Multicarrier modulation systems
- H04L27/2602—Signal structure
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/26—Systems using multi-frequency codes
- H04L27/2601—Multicarrier modulation systems
- H04L27/2647—Arrangements specific to the receiver only
- H04L27/2655—Synchronisation arrangements
- H04L27/2689—Link with other circuits, i.e. special connections between synchronisation arrangements and other circuits for achieving synchronisation
- H04L27/2695—Link with other circuits, i.e. special connections between synchronisation arrangements and other circuits for achieving synchronisation with channel estimation, e.g. determination of delay spread, derivative or peak tracking
-
- 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)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Power Engineering (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention discloses an OTFS signal receiving and recovering method based on a deep neural network structure, which comprises the following steps: constructing an OTFS (over the air) transceiving system and constructing a bit information tag for transmitting a data set and corresponding data; acquiring an electromagnetic signal after passing through a delay Doppler channel, converting the electromagnetic signal into I Q data, and preprocessing the received I Q data into a network input format; constructing a multi-classification neural network based on a residual block (ResNet); training the multi-class neural network based on residual block (ResNet); and preprocessing the collected I Q data according to a signal acquisition mode, sending the preprocessed data to a trained multi-classification neural network based on a residual block (ResNet), and outputting an inference result. Compared with the common detection demodulation method, the performance of the invention is obviously improved and is close to the performance of the maximum likelihood detection algorithm.
Description
Technical Field
The invention belongs to the technical field of communication signal receiving detection and signal processing, and particularly relates to an OTFS signal receiving recovery method based on a deep neural network structure.
Background
The development of mobile communication technology is accompanied with the development rule of one generation every decade, the Moore's rule shows that the chip processing performance is rapidly improved, and with the large-scale deployment of 5G network, the world emergence of the internet of things of all things interconnection and the rapid development of emerging communication technology, wireless communication becomes an important mode for communication and transmission of information of the contemporary human society.
An ofdm (orthogonal Frequency Division multiplexing) orthogonal Frequency Division multiplexing technology is a core technology of fourth generation mobile communication, as a multi-carrier modulation technology, and multiple carriers are orthogonal to each other, so that the utilization rate of Frequency spectrum resources can be improved, the total bandwidth is divided into a plurality of narrow-band sub-carriers, and Frequency selective fading is effectively resisted. However, in a communication scenario of high-speed mobility, transmission performance may be significantly degraded due to time-varying characteristics of a channel, in order to solve a problem that an existing OFDM system is susceptible to frequency offset in a high-speed mobility environment and performance is deteriorated, r.hadani et al proposes a new modulation mode otfs (orthogonal time frequency space modulation), and in a delay-doppler channel, compared with the OFDM system, the OFDM system exhibits significant advantages. The time delay Doppler domain provides a selectable time-varying channel mobile terminal and reflector geometric model, each information symbol is expanded to a two-dimensional orthogonal basis function by utilizing the OTFS modulator in the representation mode, the function spans the whole time-frequency domain required by a transmission frame, and the basis function is specially designed to resist the dynamic characteristic of a time-varying multipath channel; hadani verified a general framework based on an ideal pulse shaping waveform OTFS in a study, comparing a turbo equalized encoding OTFS system with an encoding OFDM system, a remarkable effect was obtained, and meanwhile, a scholart also demonstrated that OTFS showed superior performance to OFDM when a millimeter wave channel generated high frequency dispersion.
The OTFS system is realized on the basis of the traditional OFDM system structure, and the OTFS preprocessing module is added at the sending end of the OFDM system, and the post-processing module is added at the receiving end of the OFDM system, so that the time-frequency domain signal transmission is converted into the time delay Doppler domain transmission. In a wireless communication system, a transmitting end generates bit information from an information source, the bit information is radiated out through an antenna after being subjected to information source coding, channel coding, modulation and pulse forming, signals reach a signal receiver after being transmitted through an air wireless channel, a receiving end carries radio frequency signals to baseband signals, and the information is recovered by adopting the processes of carrier synchronization, symbol synchronization, channel estimation, equalization, demodulation, channel decoding, information source decoding and the like. The essence of communication signal receiving is that baseband electromagnetic signals are restored into coded information, the coded information is consistent with a physical model of a mapping relation between a training sample and a label in deep learning, a network model is obtained through deep learning training, information transmission is guaranteed to be restored, system reasoning time delay can be reduced, the network model trained through samples acquired in a real communication environment has strong robustness, the information restoration capability of the communication system in a complex environment can be improved based on the network model trained through deep learning, meanwhile, the electromagnetic signals acquired based on the real environment are more consistent with an actual communication process, and a real application scene is achieved.
In the existing research, a scholars in the aspect of channel estimation proposes a time-varying rayleigh fading estimation method based on deep learning, combines a cyclic neural network structure with a sliding window idea, and utilizes a neural network to establish, train and test a sliding bidirectional gating recursive unit channel estimator to complete time-varying rayleigh fading channel estimation. In the aspect of channel equalization, a scholars proposes an equalizer based on a deep neural network and a convolutional neural network, and has better error code performance compared with a traditional minimum mean square error equalizer, and in the aspect of signal demodulation, a scholars proposes a demodulator based on the convolutional neural network to demodulate a bipolar extended binary phase shift monitoring super-Nyquist speed signal, so that the problem of serious intersymbol interference is solved.
Disclosure of Invention
In view of the above, the present invention is directed to a method for receiving and recovering an OTFS signal based on a deep neural network structure.
In order to achieve the purpose, the technical scheme of the invention is realized as follows:
the embodiment of the invention provides an OTFS signal receiving and recovering method based on a deep neural network structure, which comprises the following steps:
constructing an OTFS (optical transport platform) receiving and transmitting system and constructing a bit information tag for transmitting a data set and corresponding data;
acquiring an electromagnetic signal passing through a delay Doppler channel, converting the electromagnetic signal into IQ data, and preprocessing the received IQ data into a network input format;
constructing a multi-classification neural network based on a residual block (ResNet);
training the multi-classification neural network based on residual block (ResNet);
sending the acquired IQ data into a trained multi-classification neural network based on a residual block (ResNet) after preprocessing according to a signal acquisition mode, and outputting an inference result.
In the above scheme, the constructing an OTFS transceiver system and the constructing a bit information tag for transmitting a data set and corresponding data specifically includes: the method comprises the steps of constructing an OTFS (over the air) receiving and transmitting system, generating a sample set and a label set for network model training, setting a baseband mapping modulation mode corresponding to the number of subcarriers and the number of symbols, randomly generating bit stream information with a fixed frame length by a transmitting end, carrying out source coding and channel coding, then carrying out modulation mapping to form corresponding symbols, setting a pilot sequence and generating a cyclic prefix sequence, then distributing data to corresponding subcarriers to complete OTFS modulation, radiating the data to an air channel by a radio peripheral device, matching the length of each frame of data of the transmitting end with the number of symbols which can be transmitted by one-time OTFS modulation, using each frame of bit data as a data label, and using the corresponding symbol which is modulated by the OTFS and then passes through a Doppler delay channel as a sample.
In the above scheme, the number M of subcarriers and the number N of frequency multiplexes on each subcarrier are set to 2 according to the OTFS modulation principle, and a corresponding baseband modulation scheme is selected, so that the length of each frame of data before encoding is: l is M × N × log2(k) The method comprises the steps that 8 bits of information are randomly generated by an information source, mapped into corresponding symbols through signal source coding and channel coding and QPSK modulation, set a pilot frequency sequence and generate a cyclic prefix sequence, then data are distributed to corresponding subcarriers to complete OTFS modulation, sent data are defined as s (t), a Doppler path parameter is set to be P2, and delay and Doppler translation amount channel matrixes on corresponding paths are set to be channel matrixes
Wherein τ ispIndicating the delay of the p-th path, vpRepresenting the Doppler frequency shift of the p-th path, the output of the transmitted signal passing through a delay Doppler channel is convolution of the transmitted information and the channel, the channel noise is Gaussian white noise w (t) with the mean value of zero, and the received signal is: r (t) ═ jj (τ, v) s (t- τ) ej2πv(t-τ)dτdv+w(t)。
In the above scheme, the acquiring the electromagnetic signal after passing through the delay doppler channel and converting the electromagnetic signal into IQ data, and preprocessing the received IQ data into a network input format specifically includes: a receiving end receives an analog electromagnetic signal of an air channel, digital orthogonal transformation is carried out after A/D sampling, IQ data are obtained through a signal extraction algorithm, and a Doppler delay factor and a signal-to-noise ratio factor are spliced into a network input sample.
In the above scheme, the received data after passing through the channel is s (n), then
Where n represents the number of received symbols, rI(n) represents taking the real part of the nth symbol, rQ(n) taking the imaginary part of the nth symbol, calculating the corresponding Doppler delay channel factor according to the Doppler frequency shift law
Recording the signal-to-noise ratio factor of the current signal as G (SNR), and if the sending end sends F frame data, constructing a sample set according to the following data splicing mode as follows:
wherein S (m) represents the m-th frame data, HkAnd representing the channel factor of the kth path, wherein the network input dimension is a matrix of LxZ, L represents the number of data symbols of each frame, Z represents the dimensions of a real part and an imaginary part of received data, the channel factor and a signal-to-noise ratio factor under multipath, the data is divided into a training set and a test set for the next-stage network model training, and the corresponding transmitted data is divided into a training label and a test label.
In the above scheme, the building of the multi-class neural network based on the residual block (ResNet) specifically includes: according to the mapping relationship between the bit information and the symbols, taking k-order QAM modulation as an example, the bit information carried by each symbol is: t is log2(k) Expanding the first dimension of an input sample matrix L multiplied by Z into tL multiplied by Z by using an deconvolution layer at an input layer of the network, namely expanding the transverse dimension of data by t times, and then expanding the first dimension into a two-dimensional convolution layer network, a batch normalization layer, a nonlinear activation layer, a maximum pooling layer and one or more residual blocks; wherein one channel in direct connection mode is one or more layer combinations of convolution layer, batch normalization layer and nonlinear activation layer, and the convolution layer and the batch normalization layerFor output, the other channel which is directly connected is output by a convolution layer and a batch normalization layer, the data of the two paths are added through a summation layer, and the classification result is output by a nonlinear activation layer, a global pooling layer, a full-link layer and a sigmoid activation function in the last layers of the network structure.
In the above scheme, training the multi-class neural network based on the residual block (ResNet) specifically includes: dividing a model training set and a test set and training a network to generate J frame data, storing IQ data after passing through a Doppler delay channel at a receiving end, taking I (I is more than 0 and less than J) group data of the IQ data as training set data, taking the residual data as test set data, taking the former I group of the transmitted data as a training label, taking the residual data as a corresponding test set label, and training the multi-classification neural network based on a residual error block (ResNet).
In the above scheme, training the multi-class neural network based on the residual block (ResNet) specifically includes: setting the number of network training rounds Epochs, setting the size of Batch _ size of each Batch of training data of the network, setting an initial Learning rate Learning _ rate, wherein the self-defined Learning attenuation is half of the attenuation after every 5 Epochs, selecting I group data from J group data as training set data, testing the performance of a network model in the training process by using the rest as a test data set, using bit information of L length of each frame of sending data as a label, outputting the network as L two classifiers, optimizing parameters of the network model by calculating binary cross entropy of network input and output, training the network model by using the training set data to obtain a network initial model, testing the network model by using the test set data, judging whether the testing accuracy of the initial model reaches the expected accuracy, stopping training if the testing accuracy reaches the expected accuracy, finishing the storage of the corresponding model by a network model training task, and continuing to train the network model by using an adaptive gradient algorithm (Adam) optimizer according to the self-defined Learning rate attenuation mode if the testing accuracy does not reach the expected accuracy, and continuing to reduce the Learning rate And the model is used for updating the weight bias and other learnable parameter optimization network models in the network, and the training is continued until the test precision of the model reaches an expected value.
In the above scheme, the method further includes verifying a multi-class neural network based on a residual block (ResNet), specifically: loading the multi-classification neural network based on the residual block (ResNet) obtained by training into a verification code, obtaining IQ data obtained by a receiving end after OTFS modulation according to the method, storing corresponding label data, splicing the received IQ data into a network input dimensional format according to the method, preprocessing, inputting the preprocessed IQ data into a model, postprocessing the output result of the model into 0, 1-bit data, counting error conditions between the output of the model and information sent by a sending end, calculating the error rate of the frame data, and completing OTFS signal receiving and recovery if the expected performance requirements are met.
Compared with the prior art, the method and the device have the advantages that the deep neural network is utilized to recover the transmitted bit information from the received IQ data, the traditional signal recovery processing such as symbol synchronization, channel equalization, demodulation and decoding of the receiving end is replaced, and the factors such as channel fading and noise are overcome; the recovery of bit stream information is realized by adopting L classifiers based on binary cross entropy, the deconvolution operation is carried out on a network input layer, the method can flexibly adapt to baseband modulation modes of different orders, and the adaptability of the method is enhanced; experiments show that compared with a common detection demodulation method, the performance of the method is obviously improved and is close to that of a maximum likelihood detection algorithm.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:
fig. 1 is a schematic diagram of an OTFS signal receiving method according to an embodiment of the present invention;
FIG. 2 is a flowchart of an algorithm design according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating an example of a deep neural network-based network according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of signal detection and extraction according to an embodiment of the present invention;
FIG. 5 is a graph of accuracy and loss variation during the network model training process of the present invention;
FIG. 6 is a graph showing the comparison of the performance of the present invention under different M, N parameter settings.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The same or similar reference numerals in the drawings of the present embodiment correspond to the same or similar components; in the description of the present invention, it is to be understood that the terms "upper", "lower", "left", "right", "inner", "outer", etc. indicate orientations or positional relationships based on those shown in the drawings, and are only for convenience of description and simplicity of description, but do not indicate or imply that the referred devices or elements must have a specific orientation, be constructed in a specific orientation, and be operated, and thus, the terms describing the positional relationships in the drawings are only for illustrative purposes and are not to be construed as limitations of the present patent, and specific meanings of the terms may be understood by those skilled in the art according to specific situations.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, article, or apparatus that comprises a list of elements does not include only those elements but also other elements not expressly listed or inherent to such process, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, article, or apparatus that comprises the element.
The embodiment of the invention provides an OTFS signal receiving and recovering method based on a deep neural network structure, which comprises the following steps:
step 1: constructing an OTFS (over the air) transceiving system and constructing a bit information tag for transmitting a data set and corresponding data;
specifically, an OTFS (over the air) transceiving system is constructed, a sample set and a label set used for network model training are generated, a baseband mapping modulation mode corresponding to the number of subcarriers and the number of symbols is set, a sending end randomly generates bit stream information with a fixed frame length, the bit stream information is subjected to source coding and channel coding and then is modulated and mapped into corresponding symbols, a pilot sequence is set, a cyclic prefix sequence is generated, then data are distributed to corresponding subcarriers to complete OTFS modulation and are radiated to an air channel by a radio peripheral device, the length of each frame of data of the sending end is matched with the number of symbols which can be transmitted by one-time OTFS modulation, each frame of sent bit data serves as a data label, and the corresponding symbols which are subjected to Doppler delay channel after OTFS modulation serve as samples.
Setting the subcarrier number M equal to 2 and the frequency multiplexing number N equal to 2 on each subcarrier according to the OTFS modulation principle, and selecting a corresponding baseband modulation mode, wherein the length of each frame of data before coding is as follows: l is M × N × log2(k) The method comprises the steps that 8 bits of information are randomly generated by an information source, mapped into corresponding symbols through signal source coding and channel coding and QPSK modulation, set a pilot frequency sequence and generate a cyclic prefix sequence, then data are distributed to corresponding subcarriers to complete OTFS modulation, sent data are defined as s (t), a Doppler path parameter is set to be P2, and delay and Doppler translation amount channel matrixes on corresponding paths are set to be channel matrixes
Wherein tau ispRepresenting the delay of the p-th path, vpRepresenting the Doppler frequency shift of the p-th path, the output of the transmitted signal passing through a delay Doppler channel is convolution of the transmitted information and the channel, the channel noise is Gaussian white noise w (t) with the mean value of zero, and the received signal is: r (t) ═ jj (τ, v) s (t- τ) ej2πv(t-τ)dτdv+w(t)。
Step 2: acquiring an electromagnetic signal passing through a delay Doppler channel, converting the electromagnetic signal into IQ data, and preprocessing the received IQ data into a network input format;
specifically, the receiving end receives the analog electromagnetic signal of the air channel, performs digital orthogonal transformation after A/D sampling, obtains IQ data through a signal extraction algorithm, and splices a Doppler delay factor and a signal-to-noise factor into a network input sample.
As shown in fig. 4, which is a schematic diagram of signal detection extraction in the embodiment of the present invention, energy-based incoherent detection is adopted to directly perform modulo on a time-domain signal sampling value, and then square the time-domain signal sampling value, without knowing any prior knowledge of the detected signal, and without limitation on the type of the signal. Energy detection is to accumulate energy in a certain frequency band range, when the energy accumulation reaches a certain threshold, a signal is present, and when the energy accumulation is lower than the certain threshold, only noise is present. And (4) in the time domain signal, calculating an energy accumulation value in a sliding window and comparing the energy accumulation value with a threshold value to find out a signal starting end point, and intercepting an effective signal part as a subsequent processing signal.
The received data after passing through the channel is S (n), then
Where n denotes the number of received symbols, rI(n) represents taking the real part of the nth symbol, rQ(n) the imaginary part of the nth symbol is taken, and the corresponding Doppler delay channel factor is calculated according to the Doppler frequency shift law
Recording the signal-to-noise ratio factor of the current signal as G (SNR), and if the sending end sends F frame data, constructing a sample set according to the following data splicing mode as follows:
wherein S (m) represents the m-th frame data, HkAnd representing the channel factor of the kth path, wherein the network input dimension is a matrix of L multiplied by Z, L represents the number of data symbols of each frame, and Z represents the dimensions of a real part, an imaginary part, a channel factor and a signal-to-noise ratio factor of the received data under multipath. And dividing the data into a training set and a test set for the next-stage network model training, and dividing the corresponding sending data into a training label and a test label.
And step 3: constructing a multi-classification neural network based on a residual block (ResNet);
in particular, according toThe mapping relationship between bit information and symbols, taking k-order QAM modulation as an example, the bit information carried by each symbol is: t is log2(k) Expanding the first dimension of an input sample matrix L multiplied by Z into tL multiplied by Z by using an deconvolution layer at an input layer of the network, namely expanding the transverse dimension of data by t times, and then expanding the first dimension into a two-dimensional convolution layer network, a batch normalization layer, a nonlinear activation layer, a maximum pooling layer and one or more residual blocks; one channel of the direct connection mode is one or more layer combinations which are formed by combining a convolution layer, a batch normalization layer and a nonlinear activation layer, the convolution layer and the batch normalization layer are used as output, the other channel of the direct connection mode is output by the convolution layer and the batch normalization layer, data of two paths are added through a summation layer, and classification results are output by the nonlinear activation layer, a global pooling layer, a full connection layer and a sigmoid activation function in the last layers of the network structure.
And 4, step 4: training the multi-classification neural network based on residual block (ResNet);
specifically, a model training set and a test set are divided and a network is trained to generate J frame data, IQ data after passing through a Doppler delay channel are stored at a receiving end, I (0 < I < J) group data of the IQ data are used as training set data, the rest data are used as test set data, meanwhile, the former I group of transmitted data are used as training labels, the rest data are used as corresponding test set labels, and the multi-classification neural network based on a residual error block (ResNet) is trained.
Training the multi-classification neural network based on the residual block (ResNet), specifically comprising the following steps: setting the number of network training rounds Epochs, setting the size of Batch _ size of each Batch of training data of the network, setting an initial Learning rate Learning _ rate, wherein the self-defined Learning attenuation is half of the attenuation after every 5 Epochs, selecting I group data from J group data as training set data, testing the performance of a network model in the training process by using the rest as a test data set, using bit information of L length of each frame of sending data as a label, outputting the network as L two classifiers, optimizing parameters of the network model by calculating binary cross entropy of network input and output, training the network model by using the training set data to obtain a network initial model, testing the network model by using the test set data, judging whether the testing accuracy of the initial model reaches the expected accuracy, stopping training if the testing accuracy reaches the expected accuracy, finishing the storage of the corresponding model by a network model training task, and continuing to train the network model by using an adaptive gradient algorithm (Adam) optimizer according to the self-defined Learning rate attenuation mode if the testing accuracy does not reach the expected accuracy, and continuing to reduce the Learning rate And the model is used for updating the weight bias and other learnable parameter optimization network models in the network, and the training is continued until the test precision of the model reaches an expected value.
And 5: sending the acquired IQ data into a trained multi-classification neural network based on a residual block (ResNet) after preprocessing according to a signal acquisition mode, and outputting an inference result.
Further, the method further comprises verifying a multi-class neural network based on a residual block (ResNet), specifically: loading the trained multi-classification neural network based on the residual block (ResNet) into a verification code, acquiring IQ data obtained by a receiving end after OTFS modulation according to the method, storing corresponding label data, splicing the received IQ data into a network input dimension format according to the method, inputting the IQ data to a model after preprocessing, performing post-processing on a model output result into 0, 1 bit data, counting error conditions between the model output and information sent by a sending end to calculate the error rate of the frame data, and completing OTFS signal receiving and recovery if the expected performance requirements are met.
Example 1
With the large-scale deployment of 5G networks, the world emergence of internet of things with all things interconnection and the rapid development of emerging communication technologies, wireless communication becomes an important mode for communication and transmission of information of the contemporary human society, artificial intelligence is once again brought up and combined with the communication field to be widely researched in signal detection, channel estimation and demodulation recovery, and the method has important significance for improving the performance of mobile communication.
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
As shown in fig. 1, the present invention provides a method for receiving an OTFS signal based on a deep neural network structure, including the following steps:
r(t)＝∫∫h(τ,v)s(t-τ)ej2πv(t-τ)dτdv+w(t)
and 2, acquiring the electromagnetic signals, preprocessing the electromagnetic signals, acquiring IQ data of the receiving end, preprocessing the IQ data and constructing a data set for training a network model. The received data after passing through the delay Doppler channel is S (n), then
Where n denotes the number of received symbols, rI(n) denotes the real part of the nth symbol, rQAnd (N) represents the imaginary part of the nth symbol, and when M is 2 and N is 2, the number of received symbols N is 4. Calculating the corresponding Doppler delay channel factor according to the Doppler frequency shift law
Recording the signal-to-noise ratio factor of the current signal as g (snr), and if the sending end sends F100000 frame data, constructing a sample set according to the following data splicing method as follows:
wherein S (m) represents the m-th frame data (0 < m.ltoreq.F), HkThe channel factor representing the kth path (0 < k ≦ 2), where M is 2 and N is 2, the network input dimension is L × Z ═ 4,7]L represents the number of data symbols per frame, and Z represents the dimensions of the real part, imaginary part, channel factor and signal-to-noise ratio factor of the received data under multipath, so that the dimension of data is a three-dimensional array of F × L × Z100000 × 4 × 7. Dividing 80000 groups in the data into training set residual 20000 groups as a test set for next stage network model training, and dividing 80000 groups in corresponding sending data into training label residual 20000 groups as test labels;
step 3, constructing a deep neural network structure based on the residual block, constructing a network structure for training, and according to the mapping relationship between the bit information and the symbols, taking QPSK modulation as an example in this example, then the bit information carried by each symbol is: t is log2(k) 2, expanding an input sample matrix L multiplied by Z and 4 multiplied by 7 to 2L multiplied by Z and 8 multiplied by 7 in a first dimension of an input sample matrix L multiplied by Z and 8 multiplied by 7 in an input layer of the network, and then expanding a two-dimensional convolution layer network, a batch normalization layer, a nonlinear activation layer, a maximum pooling layer and one or more residual blocks, wherein the problem of deep network model degradation is solved in the residual block by a short-circuit direct connection mode, wherein the main path of the residual block adopts a mode of convolution layer and batch normalizationThe method comprises the following steps that one or more layers are combined, wherein the layers are combined with nonlinear active layers, a convolution layer and a batch normalization layer are used as output, a sub-path of a residual block is used as the output of the convolution layer and the batch normalization layer, feature vectors of the two paths are added and transmitted downwards through summation of adding layers, and a classification result is output by a sigmoid active function through the nonlinear active layer, a global pooling layer, a full-connection layer and finally 2L-8 two classifiers in the last layer of a network structure;
and 4, dividing the training set and the test set and training the network, generating data of J100000 frames according to the mode in the step (1), storing IQ data after passing through a Doppler delay channel at a receiving end, taking data of I80000 (I is more than 0 and less than J) of the IQ data as training set data, taking the rest 20000 group data as test set data, taking the former I80000 group of the transmitted data as a training label, and taking the rest 20000 data as a corresponding test set label. Training the residual error network described in (3), specifically, setting the number of network training rounds of Epochs to 50, setting the size of each Batch of training data of the network to Batch _ size to 64, setting the initial Learning rate of the training network to 0.003, self-defining the Learning attenuation to be half of the original attenuation after every 5 Epochs, selecting 80000 groups of data from 100000 groups of data as training set data, testing the performance of the network model in the training process by using the rest as a test data set, using the 8-bit-length bit information of each frame of the transmitted data as a label, using the network output to be 8 classifiers activated by 8 sigmoids, optimizing the parameters of the network model by calculating the binary cross entropy of the input and output of the network, training the network model by using the training set data to obtain the network initial model, testing the network model by using the test set data, and judging whether the test accuracy of the initial model reaches the expected accuracy, if so, stopping training, finishing the training task of the network model and storing the corresponding model, otherwise, adopting an adaptive gradient algorithm (Adam) optimizer to reduce the learning rate according to a self-defined learning rate attenuation mode and continue training the network model, updating the weight bias and other learnable parameters in the network to optimize the network model, so that the network model optimally learns along the optimal solution direction, and continuing training until the model testing precision reaches an expected value;
and 5, model deployment and performance verification, loading the network model obtained by training in the method (4) into a verification code, obtaining IQ data obtained by a receiving end after OTFS modulation according to the method (1), storing corresponding label data, splicing the received IQ data into a network input dimension format according to the method (1), preprocessing the IQ data to be used as the input of the model, carrying out post-processing on the output result of the model, setting the value of the output result of the model to be more than or equal to 0.5 to 1, setting the value of the output result of the model to be less than 0.5 to 0, splicing the IQ data to be used as the finally output bit stream information of the receiving network again, counting the error condition between the output result of the final model and the information sent by the sending end to calculate the error bit rate of the frame data, completing the receiving and recovering of the OTFS signal if the expected performance requirement is met, and returning to the step 4 to retrain the model if the error performance meets the requirement. And recovering the model which is finally trained and stored as OTFS receiving end information from baseband IQ data to a model of bit stream information.
The invention constructs a sample based on electromagnetic signals by collecting electromagnetic signal data of different bit stream information sent by an OTFS (optical transport system) modulation system, the sent bit information is used as a data set of a label, a network model based on a deep residual error structure is trained, and the trained network model is utilized to recover the bit stream information of a sending end so as to achieve the purpose of recovering the sent bit information. The network model trained based on deep learning can improve the information recovery capability of the communication system in a complex environment, and meanwhile, electromagnetic signals acquired in a real environment better conform to the actual communication process, so that the network model has a practical application scene in the complex environment.
The above description is only one embodiment of the present invention, but the scope of the present invention is not limited thereto, and any modification or replacement within the technical scope of the present invention disclosed by the claims should fall within the scope of the present invention.
Simulation conditions are as follows: as shown in fig. 1, in the OTFS modulation, a transmitting end converts symbol information (such as QAM) after modulation mapping to a Delay-Doppler Domain (DD) grid, converts a signal to a time-frequency domain through an ISFFT operation, completes the expansion of the signal in time and frequency, modulates a multicarrier symbol into a time-domain signal through heisenberg transform (heisenberg transform), and transmits the time-domain signal to a wireless channel after shaping filtering. At a receiving end, the sampled signals form the size of a sample input by a network model after being preprocessed, and the trained model is utilized to demodulate and decode the time domain IQ data to recover bit information.
Simulation content: the invention carries out receiving recovery on OTFS signals, mainly maps IQ data to a process of bit information at a signal receiving end by using a mode of training a deep neural network model, and a specific algorithm implementation flow is shown in figure 2.
Fig. 6 shows error bit curves for different numbers of subcarriers and different numbers of symbols set in the OTFS modulation scheme, where fig. 6(a) is an error bit curve for comparing single-tap least mean square OFDM, message passing detection algorithm, maximum ratio detection algorithm and the receiving algorithm of the present invention when M is 2 and N is 2, fig. 6(b) is an error bit comparison curve for different algorithms when M is 4 and N is 2, fig. 6(c) is an error bit comparison curve for different algorithms when M is 4 and N is 4, fig. 6(d) is an error bit comparison curve for different algorithms when M is 2 and N is 2 baseband symbols are mapped to 16QAM, and fig. 6(a) sets the number of subcarriers to 2 in the OTFS modulation scheme and sets the number of symbols to 2, and 4-QAM modulation scheme is used to detect error rate performance under different receiving conditions. Wherein, the bit error rate of MRC detector under 4-QAM modulation, the detection iteration is set to 10 times, the iteration time of message passing MP algorithm is also set to 10 times, and the performance of MMSE equalizer with OFDM single tap. Compared with the OFDM modulation, the performance of the OTFS modulation and demodulation mode under a delay Doppler channel has obvious advantages through simulation results, wherein the performance of detection based on an MP algorithm is more excellent than that of an MRC detection algorithm when the signal-to-noise ratio is low, but the MRC detection algorithm advantages are shown when the signal-to-noise ratio is increased.
Fig. 6(b) is a performance curve of different algorithms when the number of subcarriers is 4 and the number of symbols is 2, and it can be known that the performance of the conventional detection algorithm is improved as the number of subcarriers M increases compared with fig. a, and meanwhile, a network model based on deep learning training still has better performance. Fig. 6(c) shows that the number of subcarriers is 4 and the number of symbols is also 4, and compared with fig. a, the performance of the conventional MP detection algorithm is improved with the increase of the number of symbols of the subcarriers, but the receiving method provided by the present invention still has certain performance advantages, in fig. 6(d), the baseband symbol mapping is set to 16QAM, and the number of bits represented by a single symbol is increased compared with 4QAM, and the performance of the conventional MP detection algorithm and the receiving method provided by the present invention is reduced, but the method provided by the present invention still has advantages over the MP detection algorithm, and meanwhile, as the number of subcarriers and the number of symbols are increased, more difficulty is brought to the training of a network model, but the OTFS signal reception recovery task can still be completed by adjusting the network structure, so when the data samples are abundant enough, the receiving method provided by the present invention has better performance.
The invention relates to an OTFS signal receiving method based on a deep neural network structure, which constructs a deep learning receiving model, recovers transmitted bit information from received IQ data by utilizing the deep neural network, replaces signal recovery processing such as symbol synchronization, channel equalization, demodulation, decoding and the like of a traditional receiving end, and overcomes factors such as channel fading, noise and the like. Different from a common classification network, the invention adopts L classifiers based on binary cross entropy to realize the recovery of bit stream information, carries out deconvolution operation on a network input layer, can flexibly adapt to baseband modulation modes of different orders and enhances the adaptability of the invention. Experiments show that the OTFS signal receiving method based on deep learning has obviously improved performance compared with a common detection demodulation method, is close to the performance of a maximum likelihood detection algorithm, and has practical application value in the aspect that the robustness of a network model can be enhanced based on sample data acquired in a real environment.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention.
Claims (9)
1. An OTFS signal receiving and recovering method based on a deep neural network structure is characterized by comprising the following steps:
constructing an OTFS (optical transport platform) receiving and transmitting system and constructing a bit information tag for transmitting a data set and corresponding data;
acquiring an electromagnetic signal passing through a delay Doppler channel, converting the electromagnetic signal into IQ data, and preprocessing the received IQ data into a network input format;
constructing a multi-classification neural network based on a residual block (ResNet);
training the multi-classification neural network based on residual block (ResNet);
sending the acquired IQ data into a trained multi-classification neural network based on a residual block (ResNet) after preprocessing according to a signal acquisition mode, and outputting an inference result.
2. The OTFS signal reception recovery method based on the deep neural network structure according to claim 1, wherein the OTFS transceiver system is constructed, and a bit information tag for transmitting a data set and corresponding data is constructed, specifically: the method comprises the steps of constructing an OTFS (over the air) receiving and transmitting system, generating a sample set and a label set for network model training, setting a baseband mapping modulation mode corresponding to the number of subcarriers and the number of symbols, randomly generating bit stream information with a fixed frame length by a transmitting end, carrying out source coding and channel coding, then carrying out modulation mapping to form corresponding symbols, setting a pilot sequence and generating a cyclic prefix sequence, then distributing data to corresponding subcarriers to complete OTFS modulation, radiating the data to an air channel by a radio peripheral device, matching the length of each frame of data of the transmitting end with the number of symbols which can be transmitted by one-time OTFS modulation, using each frame of bit data as a data label, and using the corresponding symbol which is modulated by the OTFS and then passes through a Doppler delay channel as a sample.
3. The OTFS signal receiving and recovering method according to claim 2, wherein the number M of subcarriers M-2 and the number N of frequency multiplexes on each subcarrier N-2 are set according to an OTFS modulation principle, and a corresponding baseband modulation scheme is selected, so that a length of each frame of data before encoding is:
L＝M×N×log2(k) the method comprises the steps that 8 bits of information are randomly generated by an information source, mapped into corresponding symbols through signal source coding and channel coding and QPSK modulation, set a pilot frequency sequence and generate a cyclic prefix sequence, then data are distributed to corresponding subcarriers to complete OTFS modulation, sent data are defined as s (t), a Doppler path parameter is set to be P2, and delay and Doppler translation amount channel matrixes on corresponding paths are set to be channel matrixes
Wherein tau ispRepresenting the delay of the p-th path, vpThe doppler frequency shift of the p-th path is represented, the output of the transmitted signal passing through the delay doppler channel is convolution of the transmitted information and the channel, the channel noise is white gaussian noise w (t) with the mean value of zero, and the received signal is:
r(t)＝∫∫h(τ,v)s(t-τ)ej2πv(t-τ)dτdv+w(t)。
4. the OTFS signal reception recovery method according to any one of claims 1 to 3, wherein the obtaining of the electromagnetic signal after passing through the delay-doppler channel and converting the electromagnetic signal into IQ data, and preprocessing the received IQ data into a network input format, specifically: a receiving end receives an analog electromagnetic signal of an air channel, digital orthogonal transformation is carried out after A/D sampling, IQ data are obtained through a signal extraction algorithm, and a Doppler delay factor and a signal-to-noise ratio factor are spliced into a network input sample.
5. The OTFS signal reception recovery method based on the deep neural network structure, according to claim 4, characterized by passing throughThe received data after the channel is S (n), then
Where n represents the number of received symbols, rI(n) denotes the real part of the nth symbol, rQ(n) taking the imaginary part of the nth symbol, calculating the corresponding Doppler delay channel factor according to the Doppler frequency shift law
Recording the signal-to-noise ratio factor of the current signal as G (SNR), and if the sending end sends F frame data, constructing a sample set according to the following data splicing mode as follows:
wherein S (m) represents the m-th frame data, HkAnd representing the channel factor of the kth path, wherein the network input dimension is a matrix of LxZ, L represents the number of data symbols of each frame, Z represents the dimensions of a real part and an imaginary part of received data, the channel factor and a signal-to-noise ratio factor under multipath, the data is divided into a training set and a test set for the next-stage network model training, and the corresponding transmitted data is divided into a training label and a test label.
6. The OTFS signal reception recovery method based on the deep neural network structure, according to claim 5, wherein the building of the multi-classification neural network based on the residual block (ResNet) is specifically as follows: according to the mapping relationship between the bit information and the symbols, taking k-order QAM modulation as an example, the bit information carried by each symbol is: t is log2(k) Expanding the first dimension of an input sample matrix L multiplied by Z into tL multiplied by Z by using an deconvolution layer at an input layer of the network, namely expanding the transverse dimension of data by t times, and then expanding the first dimension into a two-dimensional convolution layer network, a batch normalization layer, a nonlinear activation layer, a maximum pooling layer and one or more residual blocks; wherein one channel of the direct connection mode is a convolution layer, a batch normalization layer and a non-linearThe active layer is a combined layer or a plurality of combined layers, the convolution layer and the batch normalization layer are used as output, the other channel which is directly connected is used as the output of the convolution layer and the batch normalization layer, the data of the two paths are added through a summation layer, and the classification result is output by the nonlinear active layer, the global pooling layer, the full-connection layer and the sigmoid active function in the last layers of the network structure.
7. The OTFS signal reception recovery method based on the deep neural network structure of claim 6, wherein the multi-classification neural network based on the residual block (ResNet) is trained, specifically: dividing a model training set and a test set and training a network to generate J frame data, storing IQ data after passing through a Doppler delay channel at a receiving end, taking I (I is more than 0 and less than J) group data of the IQ data as training set data, taking the residual data as test set data, taking the former I group of the transmitted data as a training label, taking the residual data as a corresponding test set label, and training the multi-classification neural network based on a residual error block (ResNet).
8. The OTFS signal reception recovery method based on the deep neural network structure of claim 7, wherein training the multi-class neural network based on the residual block (ResNet) specifically comprises: setting the number of network training rounds Epochs, setting the size of Batch _ size of each Batch of training data of the network, setting an initial Learning rate Learning _ rate, wherein the self-defined Learning attenuation is half of the attenuation after every 5 Epochs, selecting I group data from J group data as training set data, testing the performance of a network model in the training process by using the rest as a test data set, using bit information of L length of each frame of sending data as a label, outputting the network as L two classifiers, optimizing parameters of the network model by calculating binary cross entropy of network input and output, training the network model by using the training set data to obtain a network initial model, testing the network model by using the test set data, judging whether the testing accuracy of the initial model reaches the expected accuracy, stopping training if the testing accuracy reaches the expected accuracy, finishing the storage of the corresponding model by a network model training task, and continuing to train the network model by using an adaptive gradient algorithm (Adam) optimizer according to the self-defined Learning rate attenuation mode if the testing accuracy does not reach the expected accuracy, and continuing to reduce the Learning rate And the model is used for updating the weight bias and the like in the network to learn the parameter optimization network model, and continuing training until the test precision of the model reaches an expected value.
9. The OTFS signal reception recovery method based on the deep neural network structure of claim 8, wherein the method further comprises verifying a multi-class neural network based on a residual block (ResNet), specifically: loading the trained multi-classification neural network based on the residual block (ResNet) into a verification code, acquiring IQ data obtained by a receiving end after OTFS modulation according to the method, storing corresponding label data, splicing the received IQ data into a network input dimension format according to the method, inputting the IQ data to a model after preprocessing, performing post-processing on a model output result into 0, 1 bit data, counting error conditions between the model output and information sent by a sending end to calculate the error rate of the frame data, and completing OTFS signal receiving and recovery if the expected performance requirements are met.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210233419.6A CN114745230B (en) | 2022-03-10 | 2022-03-10 | OTFS signal receiving and recovering method based on deep neural network structure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210233419.6A CN114745230B (en) | 2022-03-10 | 2022-03-10 | OTFS signal receiving and recovering method based on deep neural network structure |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114745230A true CN114745230A (en) | 2022-07-12 |
| CN114745230B CN114745230B (en) | 2024-01-16 |
Family
ID=82275341
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210233419.6A Active CN114745230B (en) | 2022-03-10 | 2022-03-10 | OTFS signal receiving and recovering method based on deep neural network structure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114745230B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115296962A (en) * | 2022-08-02 | 2022-11-04 | 北京信息科技大学 | OTFS system signal detection method based on Viterbi network |
| CN115514596A (en) * | 2022-08-16 | 2022-12-23 | 西安科技大学 | Convolution neural network-based OTFS communication receiver signal processing method and device |
| CN115865199A (en) * | 2022-11-10 | 2023-03-28 | 北京理工大学 | Nonlinear compensation method for optical fiber communication based on residual error neural network |
| CN115987734A (en) * | 2022-12-02 | 2023-04-18 | 西安电子科技大学 | Low-complexity OTFS system symbol detection method based on deep neural network |
| CN116055273A (en) * | 2023-01-19 | 2023-05-02 | 浙江工业大学 | QPSK receiver cascaded by neural network and auxiliary model training method thereof |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3433969A1 (en) * | 2016-03-23 | 2019-01-30 | Cohere Technologies | Receiver-side processing of orthogonal time frequency space modulated signals |
| US20190238189A1 (en) * | 2016-09-30 | 2019-08-01 | Cohere Technologies, Inc. | Uplink user resource allocation for orthogonal time frequency space modulation |
| CN111669344A (en) * | 2020-06-01 | 2020-09-15 | 西北工业大学 | Time-varying OFDM system signal detection method based on deep learning |
| US20210279851A1 (en) * | 2020-03-09 | 2021-09-09 | Canon Kabushiki Kaisha | Manufacturing method of learning data, learning method, learning data manufacturing apparatus, learning apparatus, and memory medium |
| CN113708855A (en) * | 2021-09-29 | 2021-11-26 | 北京信息科技大学 | OTFS data drive receiving method, system and medium based on deep learning |
| CN113726697A (en) * | 2021-09-14 | 2021-11-30 | 电子科技大学 | OTFS symbol detection method based on confidence space dynamic decision |
| CN113852580A (en) * | 2021-09-14 | 2021-12-28 | 电子科技大学 | MIMO-OTFS symbol detection method based on multi-level separation |
| CN113965440A (en) * | 2021-10-20 | 2022-01-21 | 嘉兴学院 | Dual-channel hybrid deep neural network OFDM wireless communication blind receiving method |
| CN114024811A (en) * | 2021-09-18 | 2022-02-08 | 浙江大学 | OTFS waveform PAPR suppression method and device based on deep learning |
-
2022
- 2022-03-10 CN CN202210233419.6A patent/CN114745230B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3433969A1 (en) * | 2016-03-23 | 2019-01-30 | Cohere Technologies | Receiver-side processing of orthogonal time frequency space modulated signals |
| US20190238189A1 (en) * | 2016-09-30 | 2019-08-01 | Cohere Technologies, Inc. | Uplink user resource allocation for orthogonal time frequency space modulation |
| US20210279851A1 (en) * | 2020-03-09 | 2021-09-09 | Canon Kabushiki Kaisha | Manufacturing method of learning data, learning method, learning data manufacturing apparatus, learning apparatus, and memory medium |
| CN111669344A (en) * | 2020-06-01 | 2020-09-15 | 西北工业大学 | Time-varying OFDM system signal detection method based on deep learning |
| CN113726697A (en) * | 2021-09-14 | 2021-11-30 | 电子科技大学 | OTFS symbol detection method based on confidence space dynamic decision |
| CN113852580A (en) * | 2021-09-14 | 2021-12-28 | 电子科技大学 | MIMO-OTFS symbol detection method based on multi-level separation |
| CN114024811A (en) * | 2021-09-18 | 2022-02-08 | 浙江大学 | OTFS waveform PAPR suppression method and device based on deep learning |
| CN113708855A (en) * | 2021-09-29 | 2021-11-26 | 北京信息科技大学 | OTFS data drive receiving method, system and medium based on deep learning |
| CN113965440A (en) * | 2021-10-20 | 2022-01-21 | 嘉兴学院 | Dual-channel hybrid deep neural network OFDM wireless communication blind receiving method |
Non-Patent Citations (2)
| Title |
|---|
| SHILIAN ZHENG: "Spectrum sensing based on deep learning classification for cognitive radios", 《CHINA COMMUNICATIONS》 * |
| 涂岳: "基于深度学习的OTFS信号解调技术研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115296962A (en) * | 2022-08-02 | 2022-11-04 | 北京信息科技大学 | OTFS system signal detection method based on Viterbi network |
| CN115296962B (en) * | 2022-08-02 | 2023-06-09 | 北京信息科技大学 | OTFS system signal detection method based on Viterbi network |
| CN115514596A (en) * | 2022-08-16 | 2022-12-23 | 西安科技大学 | Convolution neural network-based OTFS communication receiver signal processing method and device |
| CN115514596B (en) * | 2022-08-16 | 2024-06-04 | 西安科技大学 | OTFS communication receiver signal processing method and device based on convolutional neural network |
| CN115865199A (en) * | 2022-11-10 | 2023-03-28 | 北京理工大学 | Nonlinear compensation method for optical fiber communication based on residual error neural network |
| CN115987734A (en) * | 2022-12-02 | 2023-04-18 | 西安电子科技大学 | Low-complexity OTFS system symbol detection method based on deep neural network |
| CN116055273A (en) * | 2023-01-19 | 2023-05-02 | 浙江工业大学 | QPSK receiver cascaded by neural network and auxiliary model training method thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114745230B (en) | 2024-01-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114745230B (en) | OTFS signal receiving and recovering method based on deep neural network structure | |
| CN106357311B (en) | Detection method of MIMO-OFDM system based on carrier index modulation | |
| KR100918717B1 (en) | Sequence estimating method and device in mimo ofdm communication system | |
| CN106453190A (en) | Multi-user sub-carrier index modulation orthogonal frequency-division multiplexing (SIM-OFDM) transmission method | |
| CN105591717A (en) | Low-complexity detection method for MIMO-OFDM system with subcarrier index modulation | |
| WO2006125343A1 (en) | A orthogonal frequency-time division multiplex transmitter, a receiver and methods | |
| Jiang et al. | AI-aided online adaptive OFDM receiver: Design and experimental results | |
| CN111368979B (en) | PAPR suppression method for MIMO-OFDM system | |
| CN113971430A (en) | Signal detection and model training method, device, equipment and storage medium | |
| CN111628833B (en) | MIMO antenna number estimation method based on convolutional neural network | |
| KR101043698B1 (en) | Apparatus and Method for detecting signal in spatial multiplexing system | |
| CN111669344A (en) | Time-varying OFDM system signal detection method based on deep learning | |
| CN114745248B (en) | DM-GSM signal detection method based on convolutional neural network | |
| Ye et al. | Bilinear convolutional auto-encoder based pilot-free end-to-end communication systems | |
| CN115118566A (en) | Improved transmission of information in wireless communications | |
| CN113726697A (en) | OTFS symbol detection method based on confidence space dynamic decision | |
| CN112636855A (en) | OFDM signal detection method | |
| Liu et al. | RecNet: Deep learning-based OFDM receiver with semi-blind channel estimation | |
| Li et al. | Model-driven deep learning scheme for adaptive transmission in MIMO-SCFDE system | |
| CN101764636A (en) | Space-time block coding DFT-S-OFDM transmission method and device | |
| Krishna Reddy et al. | Enhancing PAPR performance in MIMO-OFDM system using hybrid optimal MMSE-MLSE equalizers | |
| Tian et al. | Research on Nonbinary LDPC-OTFS Scheme in High Mobile Communication Scenarios | |
| CN114337876B (en) | AMP detection method based on NSGA2 algorithm training and system for implementing same | |
| Zhang et al. | Turbo receiver for polar-coded OFDM systems with unknown CSI | |
| Turhan et al. | Deep learning aided generalized frequency division multiplexing |
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 |