OTFS receiver signal detection method and system based on deep learning
By adopting deep learning technology and cyclic shift preprocessing in OTFS receivers, the challenges of channel estimation and signal detection in high mobility wireless communications are solved, and more stable communication and signal recovery are achieved.
Patent Information
- Application Number
- CN202411047749.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-08-01
AI Technical Summary
In the highly mobility wireless communication scenario, traditional time-frequency domain OFDM technology has severe inter-carrier interference in the dual-select channel, making it difficult to achieve accurate channel estimation and signal detection, resulting in unstable communication.
Using the deep learning-based OTFS receiver signal detection method, the 2D DD domain signal is cyclically shifted preprocessed, and feature extraction and signal recovery is used by CNN or lightweight neural network to learn and recover directly from the received signal. The original signal reduces dependence on prior CSI.
It improves the robustness and flexibility of signal detection, reduces the amount of parameter calculation, and enhances communication stability and signal recovery capabilities in high mobility scenarios.
Smart Images

Figure CN118971999A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of wireless communication technology, and relates to vehicle network communication, OTFS modulation theory, and receiver design based on deep learning, and specifically to a OTFS receiver signal detection method and system based on deep learning. Background Art
[0002] As the Internet of Everything gradually becomes a reality and communication networks are changing with each passing day, the development of the sixth generation (6G) network is considered to have entered its first year, and high-speed, reliable and stable communications are becoming more and more popular. 6G networks need to meet the needs of high mobility wireless communications (HMWC), which includes high-speed mobile communication scenarios such as vehicles, high-speed trains and satellites. Unlike quasi-static channels that only have frequency selectivity, there are complex scatterers between the transmitter and the receiver in modern communication paths, and there is also a Doppler effect caused by relative motion between the transmitter, the receiver and the scatterers. Therefore, the channel in the HMWC scenario has dual selective fading with both time selectivity and frequency selectivity. At the same time, due to high-speed movement, this fading is often large-scale. This large-scale dual selective fading brings huge challenges to communications in HMWC scenarios. How to obtain accurate channel estimation information, how to optimize and adjust the existing network deployment, and how to develop advanced signal processing technology to ensure reliable and stable communication have become urgent problems to be solved.
[0003] In 4G and 5G cellular networks, Orthogonal Frequency Division Multiplexing (OFDM) has become one of the commonly used multi-carrier modulation technologies. However, under the above channel characteristics, the traditional OFDM technology in the time-frequency (TF) domain has serious inter-carrier interference (ICI) in the dual-selective channel. The high Doppler shift caused by high-speed mobility and high frequency bands will seriously destroy the orthogonality between OFDM subcarriers. At the same time, as a multi-carrier modulation technology, Orthogonal Time-Frequency-Space (OTFS) modulation, as an emerging two-dimensional modulation scheme, maps data symbols to the delay Doppler (DD) domain and has become a promising solution for HMWC. OTFS converts the TF domain channel that experiences frequency and time fading into the DD domain channel. Since the signal modulation is in the DD domain, the channel parameters remain approximately constant. The channel response in the DD domain is the two-dimensional convolution of the symbol and the channel response, which enhances the stability and correlation of the symbol information. This conversion helps to overcome the inherent limitations of the TF domain processing framework of OFDM.
[0004] OTFS channel equalization and signal detection play a vital role in communication systems. It eliminates the influence of channel parameters on the transmitted signal, thereby restoring the original transmitted signal at the receiver. The core idea is to use the channel state information obtained from channel estimation to reversely compensate for the distortion caused by the wireless channel, and finally identify and recover the desired signal from the interference source, which is crucial for reliable and stable communication. However, due to the complexity of two-dimensional signal processing, there are considerable challenges in executing physical layer algorithms such as signal detection in the DD domain. For example, the OTFS-based channel estimation algorithm requires operations such as matched filtering and windowing to improve the accuracy of two-dimensional signal processing.
[0005] In recent years, following the success in computer vision, graphic representation, and natural language processing, the application of deep learning (DL) technology in the physical layer of wireless communications has attracted increasing attention. Deep learning is characterized by its ability to automatically learn and extract complex patterns and features from data, more accurately represent complex nonlinear relationships, and lead to a paradigm shift in data processing. Wireless channel conditions and interference patterns are often highly dynamic and complex, which poses substantial challenges to accurate modeling through traditional techniques. Deep learning can provide a sophisticated understanding of the multifaceted relationship between transmitted signals and received observations, thereby enabling more sophisticated and effective signal processing techniques.
[0006] In practical applications, OTFS signal detection involves reversing the distortion caused by the wireless channel to the transmitted signal and distinguishing the signal or waveform pattern in the interference background to restore the original transmitted signal. In order to achieve this goal, deep learning technology can directly obtain the mapping from the received signal to the equalization output, eliminating the requirement for explicit feature engineering, thereby simplifying the overall equalization process, and can solve the problem of slow convergence of traditional algorithms through two basic features: sparsely connected neurons and shared weights of the receptive fields within the layer, which can improve computational efficiency. Therefore, the research on OTFS signal detection methods based on deep learning is of great benefit to today's OTFS receiver research. Summary of the invention
[0007] Purpose of the invention: The purpose of the present invention is to provide an OTFS receiver signal detection method and system based on deep learning, which utilizes deep learning and OTFS technology to achieve reliable and stable communication in vehicle network communication scenarios under high mobility, and improve the robustness and flexibility of signal detection.
[0008] Technical solution: To achieve the above-mentioned invention object, the present invention adopts the following technical solution:
[0009] A deep learning-based OTFS receiver signal detection method comprises the following steps:
[0010] The complex information in the two-dimensional DD domain signal is split into real and imaginary parts, and then concatenated in the third dimension to form a data format of (M, N, 2); where M is the number of subcarriers and N is the number of time slots;
[0011] The cyclic shift (SP) method is used to cut the two-dimensional DD domain signal of the input data into 9 segments according to the maximum delay value and the maximum Doppler frequency shift value. According to the characteristics of the two-dimensional circular convolution, the outer 8 segments are copied and the original frame perimeter is filled according to the law of DD domain information migration.
[0012] After filling, the two-dimensional DD domain signal is used as the input of the neural network in the OTFS signal receiver based on deep learning;
[0013] After obtaining the output of the neural network, the expanded part is removed and restored to the original size to obtain the equalized symbol of the OTFS modulation symbol; feature extraction and signal recovery are performed on the signal constellation diagram.
[0014] Furthermore, in the SP method, the extension length of the SP is the extension length of the maximum delay taps at both ends of the delay axis and the extension length of the maximum Doppler frequency taps at both ends of the Doppler axis.
[0015] Furthermore, in the SP method, the 9 segments are cut and filled as follows: the length of the segments at both ends of the delay axis is the maximum delay tap, and the length of the segments at both ends of the Doppler axis is the maximum Doppler frequency tap; the segments in the delay axis and Doppler axis directions are filled in a cyclic shift manner respectively.
[0016] Furthermore, the neural network in the deep learning-based OTFS signal receiver adopts a CNN network, and uses the two-dimensional convolution kernel of CNN to extract and fit the features of the signal in the two-dimensional DD domain; the data input into the two-dimensional convolution kernel module is preprocessed using the SP method.
[0017] Furthermore, the neural network in the deep learning-based OTFS signal receiver adopts a lightweight neural network MobileNet or MobileViT, and the SP method is used to preprocess the input two-dimensional DD domain signal data before the two-dimensional convolution kernel in the network performs data input and preprocessing.
[0018] Furthermore, the training data set of the deep learning neural network is a set of received symbols at the receiving end and modulation symbols at the transmitting end.
[0019] The present invention also provides an OTFS receiver signal detection system based on deep learning, comprising:
[0020] The preprocessing module is used to split the two-dimensional DD domain signal into real and imaginary parts, and splice them in the third dimension to form a (M, N, 2) data format; and use the SP method to cut the two-dimensional DD domain signal of the input data into 9 segments according to the maximum delay value and the maximum Doppler frequency shift value, and copy the outer 8 segments according to the characteristics of the two-dimensional circular convolution, and fill the original frame perimeter according to the law of DD domain information migration;
[0021] The deep learning module is used to fill the two-dimensional DD domain signal and use it as the input of the neural network in the OTFS signal receiver based on deep learning; after obtaining the output of the neural network, the expanded part is removed and restored to the original size to obtain the equalized symbol of the OTFS modulation symbol; feature extraction and signal recovery are performed on the signal constellation diagram.
[0022] The present invention also provides an in-vehicle network OTFS communication system based on deep learning, which uses OTFS technology to modulate, send and receive communication signals, and at the receiving end, uses an OTFS receiver signal detection method based on deep learning to recover and detect the signal.
[0023] The present invention also provides a computer system, comprising a memory, a processor, and a computer program / instruction stored in the memory and executable on the processor, wherein when the computer program / instruction is executed by the processor, the steps of the OTFS receiver signal detection method based on deep learning are implemented.
[0024] The present invention also provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the OTFS receiver signal detection method based on deep learning.
[0025] Beneficial effects: Compared with the prior art, the present invention has the following advantages: First, the OTFS signal detection receiver based on deep learning in the present invention does not require prior CSI information, and can directly learn and restore the original signal from a large amount of data. Second, the present invention proposes a cyclic shift SP method for data preprocessing of two-dimensional DD domain signals, which improves the performance of the deep learning network model to a certain extent and solves the limitations of the two-dimensional convolution kernel in processing edge information. Third, the present invention can use a lightweight neural network to optimize the network deployment of the receiver, which reduces the amount of parameter calculation while ensuring reliability and increases flexibility. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 It is a system communication scenario diagram of an embodiment of the present invention.
[0027] Figure 2It is a model diagram of the OTFS communication system of an embodiment of the present invention.
[0028] Figure 3 2 is a format diagram of input signal data in an embodiment of the present invention.
[0029] Figure 4 This is a model diagram of an OTFS signal detection system based on deep learning in an embodiment of the present invention.
[0030] Figure 5 Schematic diagram of the cyclic shift SP method in an embodiment of the present invention.
[0031] Figure 6 1 is a signal constellation diagram at different stages in an embodiment of the present invention.
[0032] Figure 7 is a graph of average bit error rate of DD domain signals in an embodiment of the present invention.
[0033] Figure 8 4 is a diagram comparing BER performance under different modulation orders in an embodiment of the present invention.
[0034] Fig. 9 4 is a diagram comparing BER performance under imperfect CSI in an embodiment of the present invention.
[0035] Fig.10 3 is a diagram comparing BER performance without coding in an embodiment of the present invention.
[0036] Fig.11 It is a BER performance comparison diagram under RS coding in an embodiment of the present invention.
[0037] Fig.12 3 is a diagram comparing BER performance under BCH coding in an embodiment of the present invention. DETAILED DESCRIPTION
[0038] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given. It should be understood that the specific examples described herein are only used to explain the present invention, but the protection scope of the present invention is not limited to the following embodiments.
[0039] The present invention takes into account that in a high-mobility in-vehicle network communication scenario, vehicles all move at a certain speed, there are scatterers between the communication paths, and the communication channel has double selective fading. Due to the presence of scatterers and the high-speed movement of vehicles, the communication path between the vehicle and the road infrastructure will change in real time, and the parameters of the channel are not fixed in each time period. The traditional TF domain symbol modulation technology can no longer meet the requirements of stable and reliable communication under mobile conditions. Therefore, OTFS modulation of the modulated signal in the DD domain can be applied to high-speed mobile in-vehicle network communication scenarios, and a signal receiver system based on deep learning is used instead of the traditional signal detection system for channel equalization and signal detection.
[0040] Therefore, an embodiment of the present invention discloses a signal detection method for an OTFS receiver based on deep learning. The method first splits the complex information of the two-dimensional DD domain signal into a real part and an imaginary part, and splices them in the third dimension; then, the SP method is used to cut the two-dimensional DD domain signal of the input data into 9 fragments according to the maximum delay value and the maximum Doppler frequency shift value, and according to the characteristics of the two-dimensional cyclic circular convolution, the outer 8 fragments are copied, and the original frame circumference is filled according to the law of DD domain information migration; after the two-dimensional DD domain signal is filled, it is used as the input of the neural network in the OTFS signal receiver based on deep learning; after obtaining the output of the neural network, the expanded part is removed and restored to the original size to obtain the equalized symbol of the OTFS modulation symbol; feature extraction and signal recovery are performed on the signal constellation diagram.
[0041] Furthermore, the method takes into account the high-mobility vehicle network communication scenario, where the movement of vehicles and the presence of scatterers cause large-scale changes and fading in the communication path, and the channel is dynamic and changing. Different from the traditional OFDM modulation technology that modulates symbols in the TF domain, the OTFS technology that modulates signals in the DD domain is used in the high-mobility vehicle network communication scenario. The channel response in the DD domain is a two-dimensional convolution of the symbol and the channel response, which enhances the stability and relevance of the symbol information. This conversion helps to overcome the inherent limitations of the OFDM TF domain processing framework.
[0042] The embodiment of the present invention uses a deep learning method to extract relevant features in the DD domain signal without the need for a priori channel state information (CSI), wherein the goal of the signal detection receiver based on deep learning is to restore the received symbols of the receiving end to equalized symbols that are similar to the modulation symbols of the transmitting end, and perform feature extraction and signal recovery on the signal constellation diagram. During training and fitting, the two-dimensional DD domain signal is split into real and imaginary parts, and then spliced as a new input.
[0043] In the specific implementation, the two-dimensional convolution kernel of CNN can be used to extract and fit the features of the signal in the two-dimensional DD domain. At the same time, the lightweight neural network MobileNet is used to optimize the network deployment, and MobileViT is used to combine the spatial sensing bias of CNN and the correlation features between local pixels with the self-attention mechanism of Transformer, so as to extract both local and global information with a small number of parameters.
[0044] When the two-dimensional convolution kernel is performing convolution operation to capture feature information, the influence of delay and Doppler will cause the information at the edge of the DD domain signal to spread to the other side, and the two-dimensional convolution kernel with limited receiving domain cannot capture this change. After the two-dimensional DD domain signal is filled, it is used as new input data. At this time, after using the SP method, the two-dimensional convolution kernel can effectively capture the diffusion relationship of the edge information. After training, the data is restored to its original size by removing the peripheral extended fragments. The use of the SP method can improve the problem of limited receiving domain of the two-dimensional convolution kernel, thereby completing more reliable signal detection.
[0045] When applying neural networks to extract features from 2D DD domain signals, the 2D convolution kernel has a limited receptive field and can only aggregate local information, which makes it impossible for neural networks to effectively process edge symbols, and the information of these edge symbols may be located on the other side of the entire frame. Applying the SP method to preprocess the input data, the information that constitutes the edge symbols is moved to their adjacent positions, and this information can be well aggregated by the neural network model.
[0046] Furthermore, for the lightweight neural networks MobileNet and MobileViT, which also have a large number of two-dimensional convolution kernel modules, the SP method is used to preprocess the input two-dimensional DD domain signal data before the two-dimensional convolution kernel performs data input and preprocessing.
[0047] Based on the same inventive concept, the present invention also discloses an OTFS receiver signal detection system based on deep learning, including: a preprocessing module, used to split the two-dimensional DD domain signal into a real part and an imaginary part, and splice them in the third dimension; and using the SP method to cut the two-dimensional DD domain signal of the input data into 9 fragments according to the maximum delay value and the maximum Doppler frequency shift value, and according to the characteristics of the two-dimensional cyclic circular convolution, copy the outer 8 fragments, and fill the original frame perimeter according to the law of DD domain information migration; a deep learning module, used to use the two-dimensional DD domain signal as the input of the neural network in the OTFS signal receiver based on deep learning after filling; and after obtaining the output of the neural network, remove the expanded part and restore it to the original size to obtain the equalized symbol of the OTFS modulation symbol; perform feature extraction and signal recovery on the signal constellation diagram.
[0048] The embodiment of the present invention also discloses an in-vehicle network OTFS communication system based on deep learning, which considers an in-vehicle network communication scenario in which a vehicle is traveling at a certain speed, wherein there are scatterers between the communication paths, and the communication channel has double selective fading. Due to the presence of scatterers and the high-speed movement of vehicles, the communication path between the vehicle and the road infrastructure will change in real time, and the parameters of the channel are not fixed in each time period. The system uses OTFS technology to modulate, send and receive communication signals, and uses an OTFS signal detection receiver based on deep learning at the receiving end to recover and detect the signals. The following is a detailed introduction to the embodiments of the present invention in conjunction with specific scenarios.
[0049] like Figure 1 As shown in the system communication scenario diagram, the embodiment of the present invention takes into account the high-mobility vehicle network communication scenario. The downlink communication path between the transmitter Tx 1 and the receiving vehicle Rx can be equivalent to two, one of which is a direct path, and the other is also used as a propagation path after passing through the scatterer 1, and finally superimposed at Rx. At the same time, since the vehicle is in continuous motion, these two communication paths are also constantly changing, and have frequency selective fading and time selective fading. Similarly, the uplink communication path between the transmitter Tx 2 and the road communication facilities is also constantly changing due to the movement of the vehicle. At the same time, the communication path between Tx 2 and Rx is also affected by the vehicle scatterer 2, so that the communication information finally superimposed at the receiving end interferes with each other, and each undergoes large-scale dual selective fading. Under this condition, the OTFS technology of the DD domain modulated signal that can sparse the delay and Doppler changes can be applied.
[0050] like Figure 2 As shown in the OTFS communication system model diagram, the communication process in the inner box is the general process of the traditional OFDM technology, while the process in the outer box is realized by adding the pre-processing ISFFT and post-processing SFFT on the basis of OFDM. Among them, OFDM modulates symbols in the time frequency TF domain, while OTFS modulates symbols in the delay Doppler DD domain.
[0051] OTFS communication system model:
[0052] At the transmitting end, the transmitted data bit stream u first passes through the encoder for channel coding, which increases redundancy while improving anti-interference and error correction capabilities. Then, the symbol is modulated into a two-dimensional signal, such as quadrature amplitude modulation (QAM), by a mapper. At this time, the transmitted symbol in the DD domain is represented as That is, the signal is sent in a two-dimensional DD domain with M subcarriers and N time slots. Through two-dimensional ISFFT, the DD domain symbol To TF domain symbol It can be expressed as:
[0053]
[0054] After adding the send window function, the above formula can be expressed as:
[0055]
[0056] Where W tx [m,n] represents the send window function.
[0057] Next, we will have the emission waveform g tx The Heisenberg transform of (t) is applied to X TF , and obtain the discrete time domain signal x(t). Specifically, the Heisenberg transform corresponds to the inverse fast Fourier transform (IFFT) implemented in OFDM modulation. Therefore, the time domain signal x(t) can be expressed as:
[0058]
[0059] where g tx (t-nT) represents the received pulse waveform, T is the time slot duration and Δf is the subcarrier spacing.
[0060] In addition, for a linear time-varying (LTV) channel containing p resolvable paths, the channel response in the DD domain can be expressed as:
[0061]
[0062] where h p , τ p and v p represent the complex fading coefficient, time delay and Doppler frequency associated with the p resolvable paths, respectively.
[0063] Therefore, the time domain received signal y(t) after the LTV channel can be expressed as:
[0064] y(t)=∫∫H DD (τ,ν)x(t-τ)e j2πv(t-τ) dτdv+ω(t)
[0065] Where ω(t) represents the mean of 0 and the variance of σ 2 The time domain received signal y(t) is first converted to the TF domain by the Wigner transform. Similar to the Heisenberg transform, the Wigner transform corresponds to the fast Fourier transform (FFT) implemented in OFDM modulation. Then the matching window g is added rx (t), then g rx The cross ambiguity function Y(t,f) between (t) and y(t) can be expressed as:
[0066]
[0067] Then, the received window function W rx [m,n] are added and sampled at intervals of t = nT, f = mΔf, and the matched filter output in the TF domain is obtained as follows:
[0068] Y TF [m,n]=W rx [m,n]Y(t,f)| t=nT,f=mΔf
[0069] At this point, the input-output relationship of OTFS can be derived as follows:
[0070]
[0071] Among them, H m,n [m′,n′] is g tx (t), H DD (τ,ν) and g rx (t) is the combined effect. Indicates g tx (t) and g rx The cross-mixing function of (t) is written as:
[0072]
[0073] Therefore, H m,n [m′,n′] can be expressed by the formula:
[0074]
[0075] (mm′)Δf-ν)e j2π(v+m′Δf)((n-n′)T-τ) e j2πvn′T dτdv
[0076] Applying the Wigner transform, that is, FFT, to the receiving end, the received symbol in the TF domain can be expressed as:
[0077]
[0078] Afterwards, Y TF Use two-dimensional SFFT to convert to DD domain, the expression is:
[0079]
[0080] Finally, the data symbol Y on the DD field at the receiving end DD After the demapper, it is restored to a one-dimensional sequence signal, and then decoded by the decoder to obtain the final output bit stream
[0081] like Figure 3 As shown in the format of the input signal data, in traditional deep learning neural network training, the two-dimensional convolution kernel usually processes three-dimensional signals, where the size of the third dimension is usually 2 or 3, representing black and white images and color images in image processing, respectively. However, unlike traditional image data, the signal in the DD domain The data in is composed entirely of complex numbers, that is, in the form of a+bj. DD The number of channels is 1, and the format is usually only in complex form (M, N). However, a two-dimensional convolution kernel usually cannot process two-dimensional data with complex elements and 1 channel. In order to meet the input data format requirements of the two-dimensional convolution kernel and avoid the situation where complex information cannot be processed, the complex information of the received signal can be split into two parts, the real part and the imaginary part, and then spliced in the same dimension as channel information.
[0082] Signal detection method based on deep learning:
[0083] Y DD After being split into the real part and the imaginary part, they are concatenated in the third dimension and finally the (M, N, 2) input format is input into the deep learning-based OTFS signal detection receiver, so that the two-dimensional convolution kernel can process the DD domain signal similarly to the black and white image.
[0084] Therefore, let r be the two-dimensional received symbol, Re(r) be the real part of the received symbol, and Im(r) be the imaginary part of the received symbol. Then the input data x of the neural network is d It can be expressed as:
[0085] x d =[Re(r),Im(r)]
[0086] Among them are
[0087] After that, the training data set of the deep learning neural network is the set of received symbols at the receiving end and the modulation symbols at the transmitting end. In other words, the received symbol x d As the input signal of the deep learning neural network, the modulation symbol y d As a label for the deep learning neural network to judge whether the learning and fitting are correct or not. Therefore, the data set Ω can be expressed as:
[0088]
[0089] where x d With y d The superscripts represent the corresponding data indexes, N B Represents the total number of samples in the dataset.
[0090] After the training data set Ω is input into the OTFS signal detection receiver based on deep learning, the two-dimensional convolution kernel in the convolution layer will extract the corresponding data features from the input data. The purpose of training is to optimize the network parameters based on the training data set generated by the traditional OTFS transmitter, so that the trained model can obtain excellent bit error rate (BER) performance and improve the generalization ability of the network. Among them, the goal of the OTFS signal detection receiver based on deep learning is to learn the functional relationship with the training weight θ Therefore, the output data of the neural network It can be expressed as:
[0091]
[0092] The reliability of wireless communication systems is mainly reflected in the BER. Therefore, the goal of the intelligent receiver for OTFS communication is to minimize the BER between the input bit stream and the output bit stream. Therefore, the purpose of the OTFS signal detection receiver based on deep learning is to minimize y d and The error between them is reduced to minimize the final BER. It can be expressed as:
[0093]
[0094] Where L represents the mean square error (MSE) loss function, and the MSE loss calculation method is also given. D Represents the data length, and the goal of the neural network is to minimize L.
[0095] like Figure 4As shown in the model diagram of the OTFS signal detection system based on deep learning, the transmitter generates the input data 01 bit stream, and the input bit stream is channel encoded by the encoder, and the constellation mapping is performed by the mapper. Taking 4-QAM modulation as an example, after the input bit stream passes through the encoder and the mapper, it is distributed in four areas on the coordinate axis as shown in the figure, that is, the mapping of the four constellations - the constellation diagram. At this time, the signal has become a two-dimensional DD domain signal. After that, this signal is modulated by OTFS (ISFFT+IFFT), and then passes through the LTV channel with noise and interference, and the signal and the channel coefficient of double selective fading are two-dimensionally convolved. The signal passing through the channel is OTFS demodulated at the receiving end, and at this time it has entered the designed OTFS signal detection receiver based on deep learning. It can be seen that the demodulated two-dimensional DD domain signal shows a chaotic form on the constellation diagram after being interfered by the double selective fading of the channel. This received symbol is used as the input data of the neural network for deep learning, and the output equalized symbol has basically restored the distribution style of the four constellations similar to the modulation symbol of the transmitter. Then the equalized symbols are passed through the decoder and demapper to obtain the output bit stream, and finally the output bit stream and the input bit stream are subjected to BER calculation to evaluate the signal detection performance. In the entire OTFS communication process, the part after the channel until the output bit stream is replaced by the OTFS signal detection receiver system based on deep learning designed by the present invention.
[0096] like Figure 5 As shown in the schematic diagram of the circular shift SP method, since the neural network extracts data features through filters, the receptive field of the neural network is limited and can only aggregate local information. When the neural network is applied to process the OTFS frame at the receiving end, this feature causes the neural network model to be unable to effectively process edge symbols, and the information of these edge symbols may be located on the other side of the entire frame. For example, as shown in the figure, due to the Doppler frequency shift, the information of area 6 will spread to area 4, and this change is migrated from one edge to another relative edge, and the size of the convolution kernel is often much smaller than the size of the DD domain signal, resulting in it being insufficient to capture this change relationship. For this reason, the present invention designs an SP method to solve the conflict between the limited receptive field of CNN and the two-dimensional DD domain signal characteristics of OTFS. First, the received OTFS frame is cut into 9 fragments, and then according to the characteristics of the two-dimensional circular convolution, the outer 8 fragments are copied, and the original frame perimeter is filled according to the law of DD domain information migration. After the filling is completed, it is used as a new input to enter the neural network for training. At this time, when the original convolution kernel performs feature extraction at the previous position, it can capture the information of area 4 and area 6 that were originally separated at both ends.
[0097] Circular shift SP method:
[0098] In the extension of SP, Q represents the maximum time delay tap, and P represents the maximum Doppler frequency tap. The extension length of SP is the extension length of the maximum time delay taps at both ends of the delay axis and the extension length of the maximum Doppler frequency taps at both ends of the Doppler axis.
[0099] Let E(·,(Q,P)) be the SP expansion function with the maximum delay tap Q and the maximum Doppler frequency tap P, and It is the reverse process of SP expansion, that is, after obtaining the output of the neural network, the SP expansion part is removed and the data format is restored to be the same as the label data for loss calculation. In this way, the input data of the neural network can be written as:
[0100] x sp =E(x d ; (Q,P))
[0101] where x sp Indicates the input data after SP expansion.
[0102] At this time, after the data is SP expanded before entering the neural network and SP is removed after passing through the neural network, its final output data can be rewritten as:
[0103]
[0104] Through the SP method, the information constituting the edge symbols is moved to their adjacent positions, and this information can be well aggregated by the neural network. In this way, when the neural network model uses the convolution window for feature extraction, the information lost at the edge will be captured, which can significantly improve the performance of channel equalization and signal detection. In addition, since there are a certain number of two-dimensional convolution kernels in MobileNet and MobileViT, the present invention also applies the SP method to lightweight neural networks to obtain certain performance improvements.
[0105] like Figure 6 The signal constellation diagrams at different stages are shown in Figure 1, which show the signal constellation diagrams at different stages under the conditions of 4-QAM modulation and uncoded OTFS signal when the SNR is 20dB. Three of the figures are Figure 6 Three constellations in Figure 1One to one correspondence. It can be seen that the modulated signal at the transmitter in (a) is equidistantly distributed in the four quadrants on the constellation diagram due to the 4-QAM modulation. At the same time, the received signal at the receiving end is interfered by the double selective fading of the channel and the influence of additive Gaussian white noise, so its constellation distribution on the constellation diagram of (b) becomes chaotic. After the channel is equalized by the OTFS signal detection receiver based on deep learning, it can be seen that the constellation diagram of the equalized signal obtained in (c) is very close to the original modulated signal at the transmitter. It can be seen that the OTFS signal detection method based on deep learning of the present invention is effective.
[0106] like Figure 7 As shown in the figure, in order to better demonstrate the limitations of the SP method, the concept of average bit error rate, that is, average BER, is introduced. Specifically, the two-dimensional DD domain signal is represented as a two-dimensional matrix. At the same time, BER is not calculated based on the total number of erroneous bits in a frame, but the BER of each element in the two-dimensional matrix in the total frame of a test set is calculated separately. For example, the DD domain signal is represented as a matrix in the form of [M,N], and the element [1,1] is taken as the research object. The average BER of this element is the number of erroneous bits at position [1,1] in all test set frames, divided by the total test set frame length. As Figure 7 As shown in the figure, the performance of the average BER of the uncoded OTFS signal of 4-QAM with or without SP in the DD domain after CNN-based signal detection at a signal-to-noise ratio of 20dB is demonstrated. When there is no SP filling in (a), due to the influence of Doppler frequency shift and the limited receptive field of the two-dimensional convolution kernel, it can be clearly seen that the average BER is higher on both sides of the DD domain signal, and the closer to the edge, the higher the average BER. On the contrary, (b) after SP expansion, the original feature of higher average BER at the edge disappears, and is replaced by a lower BER on the entire plane, and the scatter points of the average BER at the level are much less than when there is no SP expansion. It can be seen that the data preprocessing of the SP method is effective in solving the problem of limited information capture and loss of global information in the convolution window.
[0107] like Figure 8As shown in the BER performance comparison diagram under different modulation orders, it shows the BER performance comparison of OTFS signal detection using traditional signal detection algorithm and OTFS signal detection method based on deep learning under high-order QAM modulation without coding. It can be seen that with the increase of modulation order, the BER of traditional signal detection algorithms MRC and LMMSE has increased significantly and maintained at a higher BER value. On the contrary, the method of signal detection using CNN still maintains a relatively stable BER performance in high-order modulation, which proves the stability of the OTFS receiver signal detection method based on deep learning in high-order modulation of the present invention. And for each order of modulation, the BER performance of the SP method is always better than that of the non-SP method, which also proves its advantage.
[0108] like Fig. 9 As shown in the BER performance comparison under imperfect CSI, the BER performance comparison between the traditional signal detection method and the CNN signal detection method is shown when the uncoded OTFS signal under 4-QAM modulation introduces imperfect channel state information (CSI). Since the traditional signal detection algorithm relies on accurate CSI, when the channel estimation error parameter ε appears, the BER performance of MRC and LMMSE deteriorates sharply, and is very sensitive to the accuracy of CSI. On the contrary, since the deep learning-based OTFS signal detection method of the present invention does not require prior CSI information, reliable signal recovery can be performed even if a channel estimation error occurs.
[0109] like Fig.10 The BER performance comparison chart under no coding shows the BER performance comparison of OTFS signals under no coding and 4-QAM modulation under traditional signal detection algorithms and signal detection methods of different network models. It can be seen that under low signal-to-noise ratio (SNR≤12dB), the OTFS receiver signal detection method based on deep learning shows better BER performance than the traditional signal detection methods MRC and LMMSE when having perfect CSI. At the same time, compared with the signal detection method without SP extension, the signal detection method with SP extension also shows better BER performance. When compared horizontally in the network model, MobileViT showed the best BER performance, followed by CNN, and MobileNetV1 and MobileNetV2 were second, but the overall performance was relatively close, showing the advantages of lightweight neural networks, which provided help in optimizing network deployment and ensuring flexibility.
[0110] like Fig.11The BER performance comparison under RS coding is shown in the figure, which shows the BER performance comparison of OTFS signal detection with RS coding under 4-QAM modulation. When the vehicle network EVA channel coding shows a certain stability, the signal detection receiver based on deep learning of the present invention achieves better BER performance than the traditional OTFS receiver while reflecting the effectiveness of the neural network. In particular, when the SNR is 12dB, MobileViT, MobileNetV1 with SP extension, and CNN all achieve a BER of 0 for signal detection. Under the same signal-to-noise ratio, the traditional signal detection algorithms LMMSE and MRC only reach 10 -3 At the same time, in other signal-to-noise ratios, the signal detection algorithm based on deep learning of the present invention still shows better BER performance than the traditional signal detection algorithm, which reflects its effectiveness.
[0111] like Fig.12 As shown in the BER performance comparison diagram under BCH coding, the BER performance comparison of the OTFS signal detection method with BCH coding under 4-QAM modulation is shown. When the SNR is 12dB, the BER of MobileViT and CNN also reaches an order of magnitude performance. At the same time, under this condition, the OTFS receiver signal detection method based on deep learning of the present invention still has better signal detection performance than the traditional signal detection algorithm. Similarly, the use of the SP method also enables the OTFS signal detection method based on deep learning to achieve a certain improvement in effect.
[0112] Those skilled in the art can understand that the technical solution of the present invention, in essence or in other words, the part that contributes to the prior art, can be embodied in the form of a software program product, which is stored in a storage medium and includes several instructions for enabling a computer system (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in the embodiment of the present invention. The storage medium includes: various media that can store computer programs, such as a USB flash drive, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.
[0113] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A signal detection method for OTFS receiver based on deep learning, It is characterized in that The steps include: The complex information in the two-dimensional delay Doppler DD domain signal is split into real and imaginary parts, and then concatenated in the third dimension to form a data format of (M, N, 2); where M is the number of subcarriers and N is the number of time slots; The cyclic shift SP method is used to cut the two-dimensional DD domain signal of the input data into 9 segments according to the maximum delay value and the maximum Doppler frequency shift value. According to the characteristics of the two-dimensional circular convolution, the outer 8 segments are copied and the original frame perimeter is filled according to the law of DD domain information migration. After filling, the two-dimensional DD domain signal is used as the input of the neural network in the OTFS signal receiver based on deep learning; After obtaining the output of the neural network, the expanded part is removed and restored to the original size to obtain the equalized symbol of the OTFS modulation symbol; feature extraction and signal recovery are performed on the signal constellation diagram.
2. According to the OTFS receiver signal detection method based on deep learning according to claim 1, It is characterized in that In the SP method, the extension length of the SP is the extension length of the maximum delay taps at both ends of the delay axis and the extension length of the maximum Doppler frequency taps at both ends of the Doppler axis.
3. According to the OTFS receiver signal detection method based on deep learning according to claim 1, It is characterized in that In the SP method, the 9 segments are cut and filled in the following manner: the length of the segments at both ends of the delay axis is the maximum delay tap, and the length of the segments at both ends of the Doppler axis is the maximum Doppler frequency tap; the segments in the delay axis and Doppler axis directions are filled in a cyclic shift manner respectively.
4. The OTFS receiver signal detection method based on deep learning according to claim 1, It is characterized in that The neural network in the OTFS signal receiver based on deep learning adopts a CNN network, and uses the two-dimensional convolution kernel of CNN to extract and fit the signal in the two-dimensional DD domain; the data input into the two-dimensional convolution kernel module is preprocessed using the SP method.
5. The OTFS receiver signal detection method based on deep learning according to claim 1, It is characterized in that The neural network in the deep learning-based OTFS signal receiver adopts a lightweight neural network MobileNet or MobileViT. Before the two-dimensional convolution kernel in the network performs data input and preprocessing, the SP method is used to preprocess the input two-dimensional DD domain signal data.
6. The OTFS receiver signal detection method based on deep learning according to claim 1, It is characterized in that The training data set of the deep learning neural network is a collection of received symbols at the receiving end and modulated symbols at the transmitting end.
7. A deep learning based OTFS receiver signal detection system, It is characterized in that include: A preprocessing module is used to split the two-dimensional delay Doppler DD domain signal into real and imaginary parts, and splice them in the third dimension to form a data format of (M, N, 2); where M is the number of subcarriers and N is the number of time slots; And adopt the cyclic shift SP method to cut the two-dimensional DD domain signal of the input data into 9 segments according to the maximum delay value and the maximum Doppler frequency shift value. According to the characteristics of the two-dimensional circular convolution, the outer 8 segments are copied and the original frame perimeter is filled according to the law of DD domain information migration. The deep learning module is used to fill the two-dimensional DD domain signal and use it as the input of the neural network in the OTFS signal receiver based on deep learning; after obtaining the output of the neural network, the expanded part is removed and restored to the original size to obtain the equalized symbol of the OTFS modulation symbol; feature extraction and signal recovery are performed on the signal constellation diagram.
8. A deep learning-based OTFS communication system for vehicle networks. It is characterized in that The system uses OTFS technology to modulate, send and receive communication signals, and at the receiving end uses an OTFS receiver signal detection method based on deep learning according to any one of claims 1-5 to recover and detect the signal.
9. A computer system comprising a memory, a processor and a computer program / instructions stored in the memory and executable on the processor, It is characterized in that When the computer program / instructions are executed by the processor, the steps of the OTFS receiver signal detection method based on deep learning according to any one of claims 1 to 6 are implemented.
10. A computer program product comprising a computer program / instructions, It is characterized in that When the computer program / instructions are executed by the processor, the steps of the OTFS receiver signal detection method based on deep learning according to any one of claims 1 to 6 are implemented.
Citation Information
Patent Citations
OTFS data drive receiving method and system based on deep learning, and medium
CN113708855A
OTFS communication receiver signal processing method and device based on convolutional neural network
CN115514596A
Discrete fourier transform spread orthogonal time frequency space modulation method
US20230016587A1
Cited By
OTFS channel estimation method based on SwinTrans-CNN model
CN119232530A
OTFS system signal detection method based on adaptive wavelet scattering convolutional neural network
CN120075013A
