A demodulation method and device of a DZT-OTFS system, equipment and medium

Abstract

The application discloses a demodulation method, device and equipment of a DZT-OTFS system and a medium, and relates to the technical field of wireless communication.The demodulation method comprises the following steps: transforming a receiving end OTFS signal to a DD domain to obtain a receiving end DD domain signal; extracting a receiving end pilot signal and a receiving end non-pilot signal from the receiving end DD domain signal; training a support vector regression model through the receiving end pilot signal and a sending end pilot signal to obtain optimal model parameters; inputting the receiving end non-pilot signal into the support vector regression model based on the optimal model parameters to obtain a predicted output of a sending end non-pilot signal; and demodulating the sending end non-pilot signal through QPSK or QAM to obtain a demodulation signal.The demodulation method reduces the signal processing complexity and decoding error rate of the receiving end of the DZT-OTFS system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, and in particular to a demodulation method and device for a DZT-OTFS system, an apparatus and a medium. BACKGROUND

[0002] With the rapid development of high-speed mobile scenarios such as Internet of Vehicles and low-orbit satellite communication, wireless channels are facing the dual interference of fractional Doppler spread and multipath delay spread. The traditional orthogonal frequency division multiplexing (OFDM) technology has been difficult to meet the communication reliability requirements due to its weak anti-Doppler ability. The orthogonal time frequency space (OTFS) technology emerges as the times require, and its core is to map information symbols to the delay-Doppler domain (DD domain), and to improve the communication robustness in high-speed mobile scenarios by directly processing the physical interference parameters (delay and Doppler) of the channel. In order to further simplify the signal transformation process of OTFS, discrete Zak transform (DZT) is introduced into the DZT-OTFS system, which directly realizes the conversion between the DD domain and the time domain through DZT / IDZT (inverse DZT), and theoretically can reduce the computational complexity. However, the demodulation link of the existing DZT-OTFS system still relies on the technical framework of traditional channel estimation and equalization and interpolation, and therefore still has many deficiencies, which are specifically manifested in:

[0003] (1) High complexity of channel estimation and detection algorithm: the existing method needs to estimate the channel matrix (such as channel estimation algorithm based on least squares and zero forcing (ZF)) through pilots first, and then offset the channel interference through an equalizer. However, in high-speed mobile scenarios, the channel parameters (Doppler shift and multipath attenuation) change rapidly with time, and the channel matrix needs to be frequently updated, resulting in a dramatic increase in computational load. At the same time, the equalizer (such as ZF equalization) has a significant noise amplification effect, and the demodulation performance decreases sharply in low signal-to-noise ratio scenarios, making it difficult to adapt to the real-time requirements of the DZT-OTFS system;

[0004] (2) Large interpolation error at non-pilot positions: the existing method needs to perform interpolation based on the channel estimation results of adjacent pilots when obtaining the transmitted signal at non-pilot positions. However, the channel interference at different positions in the DD domain has nonlinear differences (such as signal energy leakage caused by fractional Doppler), and linear interpolation cannot accurately fit this nonlinear relationship, which is prone to large interpolation errors, especially when the pilot interval is large or the channel time variation is severe, the error will further accumulate, resulting in an increase in the bit error rate and affecting the data transmission reliability;

[0005] (3) Pilot overhead and demodulation performance are difficult to balance: in order to reduce interpolation error, the existing method needs to increase the pilot density (such as shortening the pilot interval), but this will squeeze the transmission space of non-pilot data, resulting in the increase of pilot overhead; if the number of pilots is reduced, the estimation and interpolation accuracy will be reduced due to the lack of channel calibration points, forming a contradiction between calculation overhead and decoding performance, which cannot meet the actual application requirements of low overhead and high precision of the DZT-OTFS system.

[0006] Therefore, the present application is proposed. SUMMARY

[0007] The present application provides a demodulation method, device and equipment of a DZT-OTFS system and a medium to solve the above technical problems.

[0008] The present application is achieved by the following technical solutions:

[0009] In a first aspect, the present application provides a demodulation method of a DZT-OTFS system, comprising:

[0010] transforming the OTFS signal at the receiving end to the DD domain to obtain a DD domain signal at the receiving end;

[0011] extracting a pilot signal at the receiving end and a non-pilot signal at the receiving end from the DD domain signal at the receiving end;

[0012] training a support vector regression model through the pilot signal at the receiving end and the pilot signal at the transmitting end to fit the pilot signal at the receiving end and the pilot signal at the transmitting end through the support vector regression model to obtain optimal model parameters;

[0013] inputting the non-pilot signal at the receiving end into the support vector regression model based on the optimal model parameters to obtain a predicted output of the non-pilot signal at the transmitting end;

[0014] demodulating the non-pilot signal at the transmitting end through QPSK or QAM to obtain a demodulated signal.

[0015] The present application proposes a demodulation method based on a support vector regression (SVR) model to solve the problems of high complexity of current OTFS channel estimation and detection algorithms, large interpolation error of non-pilot data and imbalance between pilot overhead and performance. The SVR model is trained through pilot data in the DD data at the receiving end, and then the SVR model is used to directly predict the data symbols at the transmitting end according to the non-pilot data at the receiving end, thereby reducing the complexity of signal processing at the receiving end of the DZT-OTFS system and the decoding error rate.

[0016] The method of the application does not need to perform channel matrix estimation and equalization calculation of the traditional method, but directly fits the mapping relationship between the DD domain received signal and the sending end signal through the SVR model, and the model training only needs to use a small number of pilot samples, compared with the traditional channel estimation+ZF equalization method, the calculation amount of the receiving end is significantly reduced, and it is suitable for high real-time requirement scenes such as vehicle networking;

[0017] Unlike the limitation of relying on the channel characteristics of adjacent pilots in the traditional linear interpolation, the method of the application fits the nonlinear channel interference law in the DD domain through the SVR model, and the model can learn the interference differences of different delay-Doppler positions, such as signal offset caused by fractional Doppler and multipath fading, and directly predicts the sending end data according to the non-pilot position and the received signal without interpolation calculation; in the high-speed mobile scene, it has a lower bit error rate.

[0018] Further, the OTFS signal at the receiving end is transformed into the DD domain to obtain DD domain data, including:

[0019] The OTFS signal at the receiving end is subjected to cyclic prefix removal and matched filtering to obtain a time domain signal;

[0020] The time domain signal is sampled to obtain a time domain discrete signal;

[0021] The time domain discrete signal is subjected to DZT transformation to obtain a receiving end DD domain signal.

[0022] Further, a support vector regression model is trained through the receiving end pilot signal and the sending end pilot signal, including:

[0023] The receiving end pilot signal and the pilot position are used as training samples, and the sending end pilot signal is used as a sample label; wherein the training sample is represented as , is the th receiving end pilot signal, represents the pilot position of the th receiving end pilot signal, , respectively represent the delay domain index and the Doppler domain index, and the sample label is the sending end pilot signal at the pilot position . ;

[0024] All receiving end pilot signals in a frame are extracted to form a training sample set , and a support vector regression model is trained through the training sample set.

[0025] Further, the support vector regression model is trained through the training sample set, including:

[0026] An objective function is established to minimize model complexity and prediction error, expressed as:

[0027] ;

[0028] wherein, represents a norm of model parameters , , is a weight vector of the support vector regression model; represents a penalty factor; , is a non-negative slack variable, corresponds to an error of the support vector regression model prediction value being less than the pilot signal from the sending end, corresponds to an error of the support vector regression model prediction value being greater than the pilot signal from the sending end; is a total number of pilot points;

[0029] A constraint condition for establishing the objective function is:

[0030] ;

[0031] wherein, represents a mapping function of the position index, is a bias term of the support vector regression model, represents a preset tolerance error, ;

[0032] Solving the objective function under the constraint condition, the optimal model parameters and are obtained.

[0033] Further, the mapping function is expressed by a Gaussian kernel, expressed as:

[0034] ;

[0035] wherein, represents a Gaussian kernel function, represents a bandwidth of the Gaussian kernel, represents a pilot position index, , represents a natural exponential function.

[0036] The prediction output of the support vector regression model is expressed as:

[0037] ;

[0038] wherein, represents a predicted non-pilot signal at a non-pilot position , , These represent the time-delay index and the Doppler index, respectively. and For Lagrange multipliers, This represents the Gaussian kernel function.

[0039] Furthermore, the DZT-OTFS system inserts pilot signals at the 1st, 4th, 7th, 10th, 13th, and 16th column positions in the Doppler domain.

[0040] A second aspect of the present invention provides a demodulation apparatus for a DZT-OTFS system, comprising:

[0041] The conversion module is used to convert the received OTFS signal to the DD domain to obtain the received DD domain signal;

[0042] The extraction module is used to extract the receiving pilot signal and the receiving non-pilot signal from the receiving DD domain signal;

[0043] The training module is used to train the support vector regression model using the receiving pilot signal as training samples, so as to fit the receiving pilot signal and the transmitting pilot signal through the support vector regression model and obtain the optimal model parameters of the support vector regression model.

[0044] The prediction module is used to input the non-pilot signal from the receiving end into the support vector regression model based on the optimal model parameters to obtain the predicted output non-pilot signal from the transmitting end.

[0045] The demodulation module is used to demodulate the transmitting non-pilot signal using QPSK or QAM to obtain a demodulated signal.

[0046] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the demodulation method of the DZT-OTFS system according to any one of the first aspects of the present invention.

[0047] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the demodulation method of the DZT-OTFS system according to any one of the first aspects of the present invention.

[0048] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0049] The SVR model is trained through pilot data in the receiving end DD data, and then the SVR model is used to directly predict the sending end data symbol according to the non-pilot data of the receiving end, so that the signal processing complexity and error rate of the receiving end of the DZT-OTFS system are reduced, and the problems of high complexity of the current OTFS channel estimation and detection algorithm and large interpolation error of the non-pilot data are solved.

[0050] The SVR model is trained through single pilot small sample data, the proportion of pilot symbols is significantly reduced, and the dynamic real-time decoding capability is improved.

[0051] Compared with the traditional channel estimation and equalization demodulation method, the present application opens up a new paradigm for directly completing the demodulation of the DZT-OTFS communication system by using a small amount of pilot data. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows, and it should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor. In the drawings:

[0053] Figure 1 It is a DZT-OTFS system demodulation block diagram based on support vector regression model;

[0054] Figure 2 It is a flow chart of the demodulation method of the DZT-OTFS system based on support vector regression;

[0055] Figure 3 It is a signal structure diagram of the receiving end DD domain in a frame;

[0056] Figure 4 It is a comparison diagram of the simulation results of the error rate of SVR detection and ZF detection under AWGN channel and QPSK modulation;

[0057] Figure 5 It is a simulation comparison diagram of the error rate of SVR detection and ZF detection varying with signal-to-noise ratio under different modulation orders. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solutions and advantages of the present application more clear and obvious, the present application will be further described in detail below in combination with embodiments and drawings, and the exemplary embodiments of the present application and their descriptions are only used to explain the present application, and not as a limitation on the present application.

[0059] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification and in the claims and the above description of the drawings are intended to cover both the exclusive and the non-exclusive inclusion of the steps or elements being claimed, such that the process, method, system, product, or apparatus that comprises or has the steps or elements does not have to necessarily be limited to or otherwise inherently include other steps or elements.

[0060] The terminology used in the various embodiments of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the various embodiments of the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of the present application belong. The terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0061] The concept of the professional terms involved in the present application is explained as follows.

[0062] OTFS: Orthogonal Time Frequency Space Modulation, is a new modulation technology, OTFS can effectively combat the influence of high Doppler spread channel by mapping information symbols to delay-Doppler (DD) domain.

[0063] DD domain: Delay-Doppler domain, is a set of two-dimensional data, which can be regarded as a two-dimensional grid, the horizontal coordinate is the Doppler domain index (range 1~16, corresponding to the Doppler shift of-8kHz~+8kHz, each index corresponds to a frequency shift of 1kHz), the vertical coordinate is the time delay domain index (range 1~256, corresponding to the propagation delay of 0~128μs, each index corresponds to a delay of 0.5μs), each grid point represents a "signal symbol of a certain delay + a certain Doppler shift".

[0064] DZT: Discrete Zak Transform, which acts as a signal translation between time domain and DD domain, can directly realize the conversion between DD domain and time domain, and the computational complexity is reduced by about 30%, which can ensure real-time demodulation.

[0065] DZT-OTFS: Discrete Zak Transform-based Orthogonal Time Frequency Space modulation technique, traditional OTFS adopts two-step transformation: first convert DD domain to time-frequency domain, then convert time-frequency domain to time domain; DZT-OTFS directly transforms DD domain to time domain through discrete Zak transform, which significantly reduces the computational complexity and improves the performance. The core advantage of DZT-OTFS system lies in its robustness to fractional delay and Doppler spread, which is suitable for high-speed mobile scenarios such as vehicle networking and high-speed rail.

[0066] SVR: Support Vector Regression, an extension of Support Vector Machine (SVM), is used to solve regression problems. Unlike traditional regression methods, SVR creates a "margin band" on both sides of the objective function, only calculating the loss for samples outside the margin band, thereby improving the robustness of the model. The core of SVR is to optimize the model by maximizing the width of the margin band and minimizing the total loss. Specifically, SVR creates a margin band with a width of ε on both sides of the objective function, and does not calculate the loss for samples within the margin band. Only samples outside the margin band will affect the model.

[0067] OTFS modulation technology is attracting attention due to its robustness in high mobility scenarios, but its demodulation performance is highly dependent on the accuracy of channel estimation. Current mainstream channel estimation methods can be divided into the following three categories.

[0068] (1) Pilot-based estimation method.

[0069] For OTFS systems, this method estimates channel parameters by embedding pilot symbols in the delay-Doppler domain. Classic schemes such as embedded pilot design place pilots in the center of the symbol, isolate data and pilot interference through guard symbols, and use threshold methods to select valid paths and estimate integer Doppler and delay parameters. Some literature proposes extracting pilot energy from local regions of the received signal, combining threshold detection to eliminate noise interference, and is suitable for low complexity scenarios, but has limited adaptability to fractional Doppler. Existing DZT-OTFS channel estimation methods mainly construct a cost function through pilot symbols, estimate channel parameters using path-by-path search, and the disadvantages are large pilot overhead and limited interpolation accuracy due to fractional delay spread effects. By setting an energy threshold to separate multipath components, combining linear interpolation to recover channel response, the disadvantages are sensitivity to fractional Doppler, large interpolation error, and inability to suppress pilot and data interference.

[0070] (2) Sparse recovery method based on compressed sensing.

[0071] The sparse property of OTFS channel provides a natural advantage for compressive sensing. Some researches transform channel estimation into an optimization problem by jointly modeling the channel in three-dimensional delay-Doppler-angle domain. The conditional expectation maximization (EM) algorithm is used to iteratively optimize the hyperparameters, combined with orthogonal search to filter out the effective paths, which significantly reduces the number of parameters to be estimated. Such methods perform well in sparse channel scenarios such as low-orbit satellites, but have weak adaptability to dynamic multipath environments.

[0072] (3) Data-driven deep learning methods.

[0073] In recent years, deep learning has been introduced to solve complex noise and fractional Doppler problems. For example, the deep residual network (DRDN) based method models channel estimation as a noise removal problem, learns noise residuals through three-dimensional convolution blocks, and improves denoising accuracy with an adaptive threshold module. Such methods perform well in nonlinear channel modeling, but rely on a large amount of training data and have high computational cost.

[0074] Interpolation methods are mainly used to extend the estimation results of the pilot area to the full delay-Doppler grid, and the common strategies are as follows.

[0075] (1) Two-dimensional linear interpolation: A local area is constructed around the pilot, and the channel response of adjacent grids is estimated by bilinear interpolation. This method is simple to calculate, but energy leakage caused by fractional Doppler will reduce the interpolation accuracy.

[0076] (2) Pilot structure based area expansion: In the embedded pilot scheme, the receiver extends the effective area by deconvolution operation, and restores the complete channel matrix combined with the pilot energy amplitude. Some documents further propose to isolate the interference by phase rotation to reconstruct the Doppler spread vector, but require accurate pre-compensation of the Doppler scale factor.

[0077] (3) Nonlinear interpolation techniques: To address the fractional Doppler effect, some researches introduce polynomial fitting or sparse representation for interpolation, but the complexity is high and relies on channel prior knowledge.

[0078] In view of the shortcomings of existing pilot schemes and interpolation methods, the present application proposes a demodulation method for DZT-OTFS system based on support vector regression (SVR). The basic idea is to use an embedded pilot structure and place a single pilot symbol at the center of the delay-Doppler domain, use a small amount of pilot data from the transmitter and receiver as the training set, and use the SVR algorithm to model the regression of the small amount of pilot data from the transmitter and receiver. Then use the data received by the receiver as the test set, and directly predict the demodulation scheme of the transmitter data using the trained model. Compared with traditional channel estimation and equalization methods, the method of the present application can directly complete the demodulation of the DZT-OTFS communication system using a small amount of pilot data. And solves the problem of the contradiction between the complexity and accuracy of channel estimation and detection of DZT-OTFS system.

[0079] Referring to Figure 1 As shown in the figure, the demodulation method of the DZT-OTFS system based on the support vector regression model mainly consists of five parts: (1) DZT-OTFS received signal processing; (2) extracting the signal at the embedded pilot of the DD domain at the receiving end; (3) training the SVR model based on the pilot data of the transmitting and receiving ends; (4) predicting the data signal at the non-pilot end of the transmitting end by using the trained SVR model; and (5) QPSK or QAM demodulation to recover the original binary data.

[0080] The flow of the five parts will be expanded in detail below, referring to Figure 2 As shown in the figure, the demodulation method comprises the following steps.

[0081] S1, transform the OTFS signal at the receiving end to the DD domain to obtain the DD domain signal at the receiving end.

[0082] The transformation at the receiving end is the inverse operation of the signal processing flow at the transmitting end. Let the transmitting signal be represented as a matrix in the DD domain , the transmitting end performs inverse discrete Zak transform (IDZT) on the DD domain signal to time domain signal , adds a cyclic prefix (CP), and then transmits through a multipath fading channel. The core goal of the receiving end is to reverse this process and finally map the received signal back to the DD domain, in preparation for subsequent SVR training and non-pilot data prediction.

[0083] S2, extract the receiving end pilot signal and the receiving end non-pilot signal from the DD domain signal at the receiving end.

[0084] As shown in Figure 3 is the DD domain signal at the receiving end within a frame, the time delay domain is 256, the Doppler domain is 16, and the pilot symbols embedded in the DD domain can be obtained by extracting the pilot symbols, and the pilot symbol and the pilot index position can be obtained, which are represented as follows:

[0085]

[0086] In the formula, represents the receiving end pilot signal at the pilot position , represents the position label of the pilot signal, i.e. is the th pilot signal, , represent the time delay domain index and the Doppler domain index of the th pilot signal, respectively. is the pilot position set, is the pilot value at the pilot position at the receiving end.

[0087] Non-pilot position is non-pilot signal.

[0088] S3, training the support vector regression model by the receiving pilot signal and the sending pilot signal.

[0089] The training aims to fit the receiving pilot signal and the sending pilot signal by the support vector regression model to obtain the optimal model parameters, that is, to complete channel modeling by using a small amount of pilot data.

[0090] During the training, each receiving pilot signal is input as the fitting data of the support vector regression model, and the sending pilot signal corresponding to the pilot position is taken as the fitting target. Only a small amount of pilot data is needed to form the training set, so that the support vector regression model can be constructed, and the sending data at the non-pilot position can be predicted by using the support vector regression model, thereby solving the contradiction between the detection complexity and the precision of the DZT-OTFS system channel estimation.

[0091] S4, inputting the receiving non-pilot signal into the support vector regression model based on the optimal model parameters to obtain the sending non-pilot signal predicted by the model.

[0092] Taking the DD domain signal in a frame at the receiving end as an example, the pilot signal and the non-pilot signal in the frame are extracted according to the pilot position, the SVR is trained by using the pilot signal, the non-pilot signal is input into the trained SVR, and the sending non-pilot signal in the frame is predicted. The processing of the DD domain signal in the next frame is the same, so that dynamic real-time decoding is realized.

[0093] For a slowly varying channel, multi-frame joint training and prediction can be used, and for a fast varying channel, single-frame processing or short multi-frame (2-3 frames) joint training can be used. High-performance devices such as base stations and edge nodes can use multi-frame parallel processing to improve the throughput, and low-performance devices such as vehicle terminals and portable terminals mainly use single-frame processing to avoid excessive delay.

[0094] S5, demodulating the sending non-pilot signal by QPSK or QAM to obtain a demodulated signal.

[0095] The binary data at the sending end is recovered by symbol demapping by QPSK or QAM, and thus the demodulation process of the DD domain signal at the receiving end of the DZT-OTFS system is completed. Whether QPSK or QAM is used can be selected according to the actual channel condition.

[0096] QPSK (Quadrature Phase Shift Keying) distinguishes symbols by phase change, and each symbol corresponds to 2-bit binary number (for example, “00” corresponds to 1+ j at a phase of 0°, “01” corresponds to -1+ j, "10" corresponds to -1 of phase 180° j , "11" corresponds to 1-j of phase 270°, the advantage is that the phase change is stable, and the noise influence is small, so the anti-interference ability is strong, and it is suitable for the channel time-varying scene of Internet of Vehicles and the like.

[0097] QAM (Quadrature Amplitude Modulation) combines phase change and amplitude change to distinguish symbols, and the number of symbols is more (such as 16QAM has 16 symbols, and 64QAM has 64 symbols), and each symbol can correspond to more binary numbers (16QAM corresponds to 4 bits, and 64QAM corresponds to 6 bits). The advantage is high spectral efficiency, and more data can be transmitted in the same time, but the anti-interference ability is weaker than QPSK, and it is suitable for stable channel quality (such as static terminal communication) scenes.

[0098] Specifically, based on the inverse process of the signal processing of the transmitting end, the OTFS signal at the receiving end is transformed into the DD domain to obtain DD domain data, including:

[0099] S1-1, the CP and the matching filter processing of the OTFS signal at the receiving end are performed to obtain a time domain signal.

[0100] The filter coefficients of the matching filter are completely matched with the signal waveform (such as the time domain pulse shape of OTFS) of the transmitting end, which can maximize the extraction of the energy of the useful signal, and at the same time, suppress the noise, adjacent channel interference and the like introduced by the channel, and output a pure time domain signal with higher signal-to-noise ratio.

[0101] S1-2, the time domain signal is sampled to obtain a time domain discrete signal.

[0102] The time domain signal after the matching filter processing is a continuous time signal, and the subsequent discrete Zak transform needs to input a discrete signal, so the time domain sampling operation needs to be performed, and the sampling frequency needs to meet the Nyquist sampling theorem to ensure that the sampling frequency is greater than or equal to twice the highest frequency of the signal, so as to avoid signal aliasing, and finally obtain a discrete time domain signal.

[0103] S1-3, the DZT transform (Discrete Zak Transform) of the time domain discrete signal is performed to obtain the DD domain signal at the receiving end.

[0104] The core of the transform is to decompose the signal in the time domain dimension into a signal matrix in the delay-Doppler two-dimensional domain through mathematical mapping, and the received signal model can be written as:

[0105]

[0106] In the formula, is the DD domain signal at the receiving end, is the discrete time domain signal, and are the sampling points in the time domain and the frequency domain respectively, , This is an index for the DD field.

[0107] Taking single-frame processing as an example, the method of training a support vector regression model by establishing a training sample set through the pilot signals of the receiving end and the pilot signals of the transmitting end within a frame includes the following steps.

[0108] S3-1, take a receiving pilot signal and its pilot position as a training sample, and take the transmitting pilot signal at the corresponding pilot position as the sample label of the sample.

[0109] The training samples can be represented as Sample labels can be represented as ,

[0110] S3-2, Extract all receiver pilot signals within a frame to form a training sample set. ;

[0111] S3-3 uses the training sample set from S3-2 to train the support vector regression model and obtain the optimal model parameters.

[0112] in, Indicates pilot position The pilot signal at the receiving end, , These represent the time delay domain index and the Doppler domain index, respectively, with the sample label being the pilot position. The pilot signal at the origin .

[0113] by Figure 3 Taking a frame of DD domain data as an example, pilot signals are inserted at columns 1, 4, 7, 10, 13, and 16 in the Doppler domain, while the remaining positions are non-pilot signals. A total of 6 × 256 sample data are extracted, and the non-pilot data to be predicted is (16-6) × 256. This pilot embedding method evenly covers the entire Doppler range from columns 1 to 16, without missing any columns. This allows the SVR to learn the channel characteristics of the entire Doppler domain through pilot samples, such as signal attenuation and noise distribution at different frequency shifts, avoiding the loss of channel characteristics due to the absence of pilot signals in certain areas.

[0114] The specific method for training a support vector regression model using a training sample set is as follows.

[0115] (1) Establish the objective function:

[0116]

[0117] in, Represents model parameters (Represents the weight vector of the support vector regression model) Norm; This is a penalty factor used to balance model complexity and prediction error. The larger the value, the heavier the penalty for error; , These are non-negative slack variables, used to allow prediction errors to exceed a tolerance range. The predicted value of the corresponding support vector regression model is smaller than the error of the transmitted pilot signal. The predicted value of the corresponding support vector regression model is greater than the error of the pilot signal at the transmitting end; This represents the total number of pilot points.

[0118] (2) Establish constraints:

[0119]

[0120] in, Indicates the use of pilot position indexing The mapping function that maps to a high-dimensional feature space. These are the model parameters (representing the bias terms of the support vector regression model). This indicates the preset tolerance error. That is, the deviation between the predicted value and the sample label is allowed to be within a certain range. No penalty is incurred when the area is within the specified range.

[0121] (3) Solve the objective function under the above constraints to obtain the optimal model parameters. and .

[0122] In the specific solution, Lagrange multipliers are introduced. , The constrained optimization problem described above is transformed into a dual problem. The optimal Lagrange multipliers are then solved using a numerical optimization algorithm (such as the Sequential Minimum Optimization (SMO)). Finally, the optimal model parameters are determined. and .

[0123] Furthermore, the mapping function is expressed using a Gaussian kernel function, i.e. This avoids explicit processing of high-dimensional feature spaces.

[0124] The Gaussian kernel is represented as:

[0125]

[0126] in, This represents the bandwidth of the Gaussian kernel. Indicates the pilot position index. , This represents the natural exponential function.

[0127] Furthermore, by using the trained SVR to predict the transmitted non-pilot signal, the predicted output is expressed as:

[0128]

[0129] in, Indicates the predicted non-pilot position The originating signal is not a pilot signal. , These represent the time-delay index and the Doppler index, respectively. and For Lagrange multipliers, This is the Gaussian kernel function.

[0130] This invention compares the demodulation performance of SVR detection (the demodulation method of this invention) with the traditional linear detection (Zero-Forcing Detection) algorithm by simulating an AWGN channel. Secondly, it simulates the demodulation performance of SVR detection and ZF detection in a low-Earth orbit satellite communication channel.

[0131] like Figure 4 The figure shows a comparison of the bit error rate (BER) of SVR detection and ZF detection under AWGN channel and QPSK modulation. By comparing the BER performance of SVR and ZF detection methods at different signal-to-noise ratios (SNR), the adaptability of the two methods in different channel environments is revealed. As the SNR increases from 0dB to 20dB, the BER of both methods decreases exponentially, with higher BER at low SNR (0-6dB). Signal quality degradation leads to significant bit errors; while at high SNR (14-20dB), the BER converges rapidly to The following demonstrates the effectiveness of the decoding algorithm of this invention in high-quality channels. Comparison shows that SVR has significant advantages at high SNR (>10dB), and the BER can be reduced to... Its steep descent curve reflects its high sensitivity to high signal-to-noise ratios, and its overall performance is superior to ZF detection.

[0132] Figure 5 The bit error rate (BER) of SVR detection and ZF detection under different modulation orders (QPSK and 16QAM) was compared with the signal-to-noise ratio (SNR). Specifically, due to the inherent characteristics of higher-order modulation, 16QAM modulation (such as 16-QAM) has a significantly higher BER than QPSK modulation (such as QPSK) at the same SNR, requiring a higher SNR (approximately 25 dB) to achieve the same result. The QPSK modulation combined with the SVR detection has the best robustness in a low SNR (<10 dB) scene, and is suitable for a complex channel environment; and the 16QAM modulation combined with the ZF algorithm can meet the requirements of high data rate and low bit error rate in a high SNR (>25 dB) scene, and is suitable for a high-quality channel.

[0133] The embodiment of the present application further provides a demodulation device of the DZT-OTFS system, which is used for executing the demodulation method of any of the above-mentioned embodiments of the present application, and the device comprises:

[0134] A conversion module is configured to convert the received OTFS signal into a DD domain to obtain a received DD domain signal.

[0135] An extraction module is configured to extract a received pilot signal and a received non-pilot signal from the received DD domain signal.

[0136] A training module is configured to train a support vector regression model by taking the received pilot signal as a training sample, to fit the received pilot signal and the transmitted pilot signal by the support vector regression model, and to obtain optimal model parameters of the support vector regression model.

[0137] A prediction module is configured to input the received non-pilot signal into the support vector regression model based on the optimal model parameters to obtain a predicted output of the transmitted non-pilot signal.

[0138] A demodulation module is configured to demodulate the transmitted non-pilot signal by QPSK or QAM to obtain a demodulated signal.

[0139] Further, the conversion module comprises a cyclic prefix removal module, a matched filter module, a sampling module and a DZT conversion module connected in sequence. The cyclic prefix removal module is configured to perform cyclic prefix removal processing on the received OTFS signal. The matched filter module is configured to perform matched filter processing on the received OTFS signal after the cyclic prefix removal processing to obtain a time domain signal. The sampling module is configured to sample the time domain signal to obtain a time domain discrete signal. The DZT conversion module is configured to perform DZT conversion on the time domain discrete signal to obtain the received DD domain signal.

[0140] Further, the extraction module comprises a pilot identification module, a pilot signal extraction module and a non-pilot signal extraction module.

[0141] Further, the training module comprises a regression model module, a target function module, a constraint condition module, a kernel function module and an optimization solution module. The regression model module, the target function module, the constraint condition module and the kernel function module are respectively used to create a support vector regression model, a target function, a constraint condition and a kernel function and store corresponding model data, and the optimization solution module is used to receive a training sample set and iteratively train model parameters of the support vector regression model according to the training sample set, the target function, the constraint condition and the kernel function.

[0142] The prediction module calls the optimal model parameters solved by the optimization solution module, and predicts the originating non-pilot signal based on the support vector regression model of the optimal model parameters.

[0143] Embodiments of the present application also provide an electronic device comprising a processor and a memory, the number of processors can be one or more. The memory as a kind of computer readable storage medium, can be used to store software programs, computer executable programs and modules. The processor executes various functions of the electronic device and data processing by running the software programs, instructions and modules stored in the memory, so as to realize the demodulation method of the DZT-OTFS system of any one of the above embodiments of the present application.

[0144] The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function; the data storage area can store data created according to the use of the terminal and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state memory device. In some examples, the memory can further include a memory remotely arranged with respect to the processor, which can be connected to the electronic device through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0145] Embodiments of the present application also provide a computer readable storage medium having a computer program stored thereon, when the computer program is executed by a processor, the demodulation method of the DZT-OTFS system of any one of the embodiments of the present application is realized.

[0146] The computer storage medium of the embodiments of the present application can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination thereof. More specific examples (non-exhaustive list) of the computer readable storage medium include: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or apparatus.

[0147] The computer readable signal medium can include a data signal propagated in a baseband or as a part of a carrier wave, in which a computer readable program code is carried. Such a propagated data signal can take on many forms, including but not limited to electro-magnetic signal, optical signal or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can transmit, propagate or transport a program for use by or in connection with an instruction execution system, apparatus or device.

[0148] The embodiments of the present application also provide a computer program product, which, when running on a computer, causes the computer to execute the demodulation method of the DZT-OTFS system of any one of the above embodiments of the present application.

[0149] The above specific embodiments further illustrate the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A demodulation method of a DZT-OTFS system, characterized by, The method comprises the following steps: transforming the receiving end OTFS signal into the DD domain to obtain a receiving end DD domain signal; extracting a receiving end pilot signal and a receiving end non-pilot signal from the receiving end DD domain signal; training a support vector regression model through the receiving end pilot signal and a transmitting end pilot signal to fit the receiving end pilot signal and the transmitting end pilot signal through the support vector regression model to obtain optimal model parameters of the support vector regression model; inputting the receiving end non-pilot signal into the support vector regression model based on the optimal model parameters to obtain a predicted output of the transmitting end non-pilot signal; demodulating the transmitting end non-pilot signal through QPSK or QAM to obtain a demodulated signal; training the support vector regression model through the receiving end pilot signal and the transmitting end pilot signal, comprising: The receiving-end pilot signal and the pilot position are taken as training samples, and the sending-end pilot signal is taken as a sample label; wherein the training sample is represented as , is the first receiving-end pilot signal, is the pilot position of the first receiving-end pilot signal, is the pilot position of the first receiving-end pilot signal, , , respectively represent a time delay domain index and a Doppler domain index, and the sample label is a sending-end pilot signal at the pilot position . ​ Extract all the received pilot signals in a frame to form a training sample set ; establishing an objective function by minimizing model complexity and minimizing prediction error, expressed as: wherein, denotes a model parameter norm of is a weight vector of the support vector regression model; denotes a penalty factor; , is a non-negative slack variable, corresponds to an error of the support vector regression model prediction value being less than the originating pilot signal, corresponds to an error of the support vector regression model prediction value being greater than the originating pilot signal; is a total number of pilot points;​ the constraint condition for establishing the objective function is: wherein, a mapping function representing a position index, a bias term for a support vector regression model, represents a pre-set tolerance error, ; solving the objective function under the constraints to obtain optimal model parameters and ; the mapping function is expressed by a Gaussian kernel, expressed as: wherein denotes a Gaussian kernel function, denotes the bandwidth of the Gaussian kernel, denotes the pilot position index, i≠j, denotes the natural exponential function; the prediction output of the support vector regression model is expressed as: wherein, represents a predicted non-pilot position at which a non-pilot signal from a source is expected, , represent a delay domain index and a Doppler domain index, respectively, and is a Lagrange multiplier, represents a Gaussian kernel function.

2. The method of demodulation of a DZT-OTFS system according to claim 1, characterized in that, transforming the receiving end OTFS signal into the DD domain to obtain DD domain data, comprising: performing cyclic prefix removal and matched filtering processing on the receiving end OTFS signal to obtain a time domain signal; sampling the time domain signal to obtain a time domain discrete signal; performing DZT transformation on the time domain discrete signal to obtain the receiving end DD domain signal.

3. The demodulation method of a DZT-OTFS system according to any one of claims 1-2, characterized in that, The transmitting end of the DZT-OTFS system inserts a pilot signal at the 1st, 4th, 7th, 10th, 13th and 16th column positions in the Doppler domain.

4. An apparatus for performing a demodulation method of a DZT-OTFS system according to any one of claims 1 to 3, characterized in that, The method comprises the following steps: a conversion module is configured to transform a receiving end OTFS signal into the DD domain to obtain a receiving end DD domain signal; an extraction module is configured to extract a receiving end pilot signal and a receiving end non-pilot signal from the receiving end DD domain signal; a training module is configured to train a support vector regression model by taking the receiving end pilot signal as a training sample to fit the receiving end pilot signal and a transmitting end pilot signal through the support vector regression model and obtain optimal model parameters of the support vector regression model; a prediction module is configured to input the receiving end non-pilot signal into the support vector regression model based on the optimal model parameters to obtain a predicted output of the transmitting end non-pilot signal; a demodulation module is configured to demodulate the transmitting end non-pilot signal through QPSK or QAM to obtain a demodulated signal.

5. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the demodulation method of the DZT-OTFS system according to any one of claims 1-3.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the demodulation method of the DZT-OTFS system according to any one of claims 1-3.

Images