An OTFS timing synchronization method and device based on orthogonal feature fusion assistance

Abstract

The application discloses an OTFS timing synchronization method and device based on orthogonal feature fusion assistance, and the OTFS timing synchronization device comprises an initial signal blind detection unit which is used for converting an OTFS signal received by a receiver into an initial timing observation signal vector and sending the initial timing observation signal vector to a mode-driven feature extraction unit and a data-driven feature extraction unit respectively; the mode-driven feature extraction unit and the data-driven feature extraction unit are respectively used for extracting mode-driven orthogonal features and data-driven orthogonal features of the initial timing observation signal vector; an orthogonal feature fusion unit is used for performing feature fusion on the orthogonal features, and sending timing synchronization orthogonal fusion features obtained after the fusion to an intelligent decision unit; and the intelligent decision unit is used for obtaining a timing synchronization estimation value according to a timing offset tentative value, a timing offset maximum tentative value and the timing synchronization orthogonal fusion features. The application improves the timing synchronization accuracy of the OTFS system under the influence of Doppler and multipath and can be applied to an OFDM system.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning OTFS timing synchronization, and in particular to an OTFS timing synchronization method and device based on orthogonal feature fusion assistance. Background Art

[0002] As a key technology and research focus for future 6G wireless networks and mobile communications, the performance of OTFS (Orthogonal Time-Frequency-Space) timing synchronization will directly impact subsequent signal processing decisions. In modern wireless and mobile communication systems, traditional signal processing methods typically require complex and precise mathematical modeling to improve the accuracy of timing synchronization. However, the increasing complexity of the electromagnetic environment and the increasing difficulty in establishing channel models have limited traditional signal processing methods based on mathematical modeling. Furthermore, more stringent requirements are being placed on achieving higher-precision OTFS timing synchronization in future wireless communication scenarios. Existing maximum likelihood-based timing synchronization methods estimate parameters that influence timing synchronization to improve the system's anti-interference performance, thereby enhancing performance. However, maximum likelihood-based methods face a series of problems and challenges, such as complex modeling and high time costs, and urgently need improvement. Summary of the Invention

[0003] In view of this, the present invention provides an OTFS timing synchronization method and device based on orthogonal feature fusion assistance.

[0004] The present invention discloses an OTFS timing synchronization device based on orthogonal feature fusion assistance, which comprises an OTFS timing synchronization processing module and an orthogonal feature extraction and fusion module; the OTFS timing synchronization processing module comprises an initial signal blind detection unit and an intelligent decision unit; the orthogonal feature extraction and fusion module comprises a pattern-driven feature extraction unit, a data-driven feature extraction unit and an orthogonal feature fusion unit; the initial signal blind detection unit is used to convert an OTFS signal received by a receiver into an initial timing observation signal vector and send the initial timing observation signal vector to the pattern-driven feature extraction unit and the data-driven feature extraction unit respectively; the pattern-driven feature extraction unit and the data-driven feature extraction unit are used to extract the pattern-driven orthogonal feature and the data-driven orthogonal feature of the initial timing observation signal vector respectively, and send the extracted features to the orthogonal feature fusion unit; the orthogonal feature fusion unit is used to perform feature fusion on the orthogonal features and send the timing synchronization orthogonal fusion features obtained after fusion to the intelligent decision unit; the intelligent decision unit is used to obtain a timing synchronization estimation value according to a timing offset trial value, a timing offset maximum trial value and the timing synchronization orthogonal fusion features.

[0005] Furthermore, the initial signal blind detection unit is specifically used to:

[0006] y=[y(0),y(1),…,y(n),…,y(N-1)]T

[0007] y(nT) = y(t - argmaxΓ i (m)T)

[0008] Γ i (m) = f τ [y(t)]

[0009] where y(t) is the OTFS signal received by the receiver; Γ i (m) is the initial signal blind detection feature, n represents the sample index of the discrete signal; t represents the observation time; m represents the timing offset trial value in the detection feature; f τ [·] represents a signal detection algorithm with discretization processing, including cross-correlation, autocorrelation, matched filtering, and energy detection; argmax is a function that returns the index of the maximum value; T represents the sampling interval, which is determined by the specifications of the test equipment; N represents the minimum observation length to obtain a complete signal, which is set by the constraints of the transceiver.

[0010] Further, the mode-driven feature extraction unit obtains the mode-driven orthogonal feature through the following formula:

[0011] λ m = f m (y)

[0012] wherein f m (·) represents a two-dimensional feature extraction algorithm process, including 2D-FFT, SFFT, MUSIC, and sub-set combination; k represents the vertical axis index of two-dimensional variable domain processing, l represents the horizontal axis index after two-dimensional variable domain processing, K represents the maximum sampling point number of the vertical axis, L represents the maximum sampling point number of the horizontal axis, 2D-FFT is two-dimensional Fourier transform, for two-dimensional features, the horizontal axis is time and the vertical axis is frequency, SFFT is symplectic transform, for two-dimensional features, the horizontal axis is Doppler and the vertical axis is delay, MUSIC represents multiple signal classification algorithm, both the horizontal axis and the vertical axis are trial parameters for classification; subscript m indicates that λ is obtained from the modeled mode driving;

[0013] The data-driven feature extraction unit obtains the data-driven orthogonal feature through the following formula:

[0014] λ d = f Θ (y, Θ)

[0015] wherein f Θ(·) represents the AI ​​reasoning process, including 2D-CNN, GNN, DNN, and RNN; Θ is the neural network parameter, which is obtained by training with real data; p represents the vertical axis index of the two-dimensional feature output by the neural network, q represents the horizontal axis index of the two-dimensional feature output by the neural network, and P and Q represent the height and width of the two-dimensional feature output by the neural network, respectively; the subscript d indicates that λ is data-driven;

[0016] The orthogonal feature fusion unit obtains the timing synchronous orthogonal fusion feature through the following formula:

[0017] Γ=f c (λ m ,λ d )

[0018] Among them, Γ=[γ(1),γ(2),…,γ(N nn )] T ;γ is the element in the fusion feature Γ, f c (·) is the orthogonal feature fusion method, including maximum ratio merging, superposition, multiplication and splicing; N nn Represents the length of the fused orthogonal feature vector, which is set based on engineering experience.

[0019] Furthermore, the two-dimensional feature extraction algorithm can be a and b are indices in the inner product summation.

[0020] Furthermore, the intelligent decision-making unit obtains the timing synchronization estimation value through the following formula

[0021]

[0022] O=[O(0),…,O(τ trial ),…,O(τ max )] T

[0023] O(τ trial )=||yG(Γ,Θ τ ,τ trial )||2

[0024] Among them, τ max is the maximum trial value of the timing offset, y represents the signal received by the receiver OTFS, G(·) represents the intelligent decision model, which consists of an ELM network of SLFN; Θ τ represents the ELM network model parameters; τ trial represents the timing offset trial value, τ trial is the discretization representation of the continuous timing offset, which is determined by the resolution; O represents the OTFS timing metric, G(,Θ τ ,) represents theτ Parametric description of the ELM network.

[0025] The present invention also discloses an OTFS timing synchronization method based on orthogonal feature fusion assistance, which is applicable to any of the above-mentioned OTFS timing synchronization devices based on orthogonal feature fusion assistance, and includes:

[0026] Step 1: Based on the OTFS signal y(t) received by the receiver, obtain the N-dimensional initial timing observation signal vector y = [y(0), y(1), ..., y(N-1)] T ;

[0027] Step 2: Obtain N based on the initial timing observation signal vector y nn dimensional timing synchronization orthogonal fusion feature Γ=[γ(0),γ(1),…,γ(N nn -1)] T ;

[0028] Step 3: Obtain the timing synchronization estimate value based on the timing synchronization orthogonal fusion feature Γ

[0029] Furthermore, the step 1 includes:

[0030] Assuming that autocorrelation is used as the initial signal target detection, the signal detection of the sampled signal is expressed as:

[0031]

[0032] According to Γ i (m), y(nT)=y(t-argmaxΓ) is calculated by the following formula i (m)T) to display:

[0033]

[0034] The initial timing observation signal vector is y = [y(0), y(1), ..., y(N-1)] T .

[0035] Furthermore, the step 2 includes:

[0036] The mode-driven orthogonal characteristics are obtained by the following formula:

[0037] λ m =f m (y)

[0038] in, f m (·) represents the two-dimensional feature extraction algorithm process, including 2D-FFT, SFFT, MUSIC and diversity merging;

[0039] The data-driven orthogonal feature is obtained by the following formula:

[0040] λ d =f Θ (y,Θ)

[0041] Wherein, f Θ (·) represents an AI inference process, including 2D-CNN, GNN, DNN, RNN; Θ is a neural network parameter, which is obtained by training real data;

[0042] The timing synchronization orthogonal fusion feature is obtained by the following formula:

[0043] Γ=f c (λ m ,λ d )

[0044] Wherein, Γ=[γ(1),γ(2),…,γ(N nn )] T ;f c (·) is an orthogonal feature fusion method, including maximum ratio combination, superposition, multiplication and splicing; N nn represents the length of the fused orthogonal feature vector, which is set by engineering experience.

[0045] Further, the SFFT is used to extract the mode-driven orthogonal feature,

[0046]

[0047] Wherein, KL=N and K=L;

[0048] The extracted mode-driven orthogonal feature λ m represents:

[0049]

[0050] Due to the adoption of the above technical solutions, the present application has the following advantages:

[0051] 1. The present application improves the timing synchronization accuracy of the OTFS system under the influence of Doppler and multipath and can be applied to the OFDM system, which specifically comprises: firstly, obtaining an initial timing observation signal vector by using an initial signal blind detection unit; then, according to an orthogonal feature extraction and fusion module, extracting the mode and data-driven feature of the observation signal vector and fusing them to enhance the input feature of the intelligent decision network, and further optimizing the OTFS timing synchronization accuracy; finally, using an ELM network as an intelligent decision network to perform timing classification and output a timing synchronization offset estimation value.

[0052] 2. Targeting high-precision, high-reliability future wireless networks and mobile communications, this invention utilizes the characteristics of OTFS signals to design a pattern- and data-driven feature extraction and fusion mechanism. This enhances the accuracy of AI-powered timing synchronization in OTFS systems, resulting in a highly reliable, high-precision, and low-mathematical-modeling timing synchronization method and an easily deployable device. This improves synchronization performance in real-world scenarios and offers numerous solutions for timing synchronization that meets low-latency processing requirements in these scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments described in the embodiments of the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0054] Figure 1 Schematic diagram of a flow chart of an OTFS timing synchronization method based on orthogonal feature fusion assistance according to an embodiment of the present invention;

[0055] Figure 2 This is a design diagram of an initial signal blind detection unit according to an embodiment of the present invention;

[0056] Figure 3 This is a design diagram of the orthogonal feature extraction and fusion module according to an embodiment of the present invention;

[0057] Figure 4 This is a design diagram of an intelligent decision-making unit according to an embodiment of the present invention;

[0058] Figure 5 Schematic diagram of an OTFS timing synchronization device based on orthogonal feature fusion assistance according to an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described with reference to the accompanying drawings and embodiments. The embodiments described are only a part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by those skilled in the art should fall within the scope of protection of the embodiments of the present invention.

[0060] See also Figures 2 to 5The present invention provides an embodiment of an OTFS timing synchronization device based on orthogonal feature fusion assistance, which includes an OTFS timing synchronization processing module and an orthogonal feature extraction and fusion module; the OTFS timing synchronization processing module includes an initial signal blind detection unit and an intelligent decision unit; the orthogonal feature extraction and fusion module includes a pattern-driven feature extraction unit, a data-driven feature extraction unit and an orthogonal feature fusion unit; the initial signal blind detection unit is used to convert the OTFS signal received by the receiver into an initial timing observation signal vector and send it to the pattern-driven feature extraction unit and the data-driven feature extraction unit respectively; the pattern-driven feature extraction unit and the data-driven feature extraction unit are used to extract the pattern-driven orthogonal features and the data-driven orthogonal features of the initial timing observation signal vector respectively, and send them to the orthogonal feature fusion unit; the orthogonal feature fusion unit is used to perform feature fusion on the orthogonal features, and send the timing synchronization orthogonal fusion features obtained after fusion to the intelligent decision unit; the intelligent decision unit is used to obtain a timing synchronization estimation value according to a timing offset trial value, a timing offset maximum trial value and the timing synchronization orthogonal fusion feature.

[0061] In this embodiment, the initial signal blind detection unit is specifically used to:

[0062] The initial signal blind detection unit is specifically used for:

[0063] y=[y(0),y(1),…,y(n),…,y(N-1)] T

[0064] y(nT)=y(t-argmaxΓ i (m)T)

[0065] Γ i (m) = f τ [y(t)]

[0066] Where y(t) is the OTFS signal received by the receiver; Γ i (m) is the initial signal blind detection feature, n is the sample index of the discrete signal; t is the observation time; m is the timing offset trial value in the detection feature; f τ [·] represents a signal detection algorithm with discretization processing, including cross-correlation, autocorrelation, matched filtering, and energy detection. argmax is a function that returns the maximum value index. T is the sampling interval, which is determined by the test equipment specifications. N is the minimum observation length for obtaining a complete signal, which is set by the constraints of both the transceiver and the transmitter.

[0067] In this embodiment, the mode-driven feature extraction unit obtains the mode-driven orthogonal feature by the following formula:

[0068] λ m =f m (y)

[0069] wherein, f m (·) represents a two-dimensional feature extraction algorithm process, including 2D-FFT, SFFT, MUSIC and sub-set combination; k represents the vertical axis index of two-dimensional variable domain processing, l represents the horizontal axis index after two-dimensional variable domain processing, K represents the maximum sampling point number of the vertical axis, L represents the maximum sampling point number of the horizontal axis, 2D-FFT is two-dimensional Fourier transform, for two-dimensional features, the horizontal axis is time and the vertical axis is frequency, SFFT is symplectic transform, for two-dimensional features, the horizontal axis is Doppler and the vertical axis is delay, MUSIC represents multiple signal classification algorithm, both the horizontal axis and the vertical axis are trial parameters for classification; subscript m represents that λ is obtained by modeling mode driving;

[0070] The data-driven feature extraction unit obtains data-driven orthogonal features through the following formula:

[0071] λ d =f Θ (y,Θ)

[0072] wherein, f Θ (·) represents an AI inference process, including 2D-CNN, GNN, DNN, RNN; Θ is a neural network parameter, which is obtained by training real data; p represents the vertical coordinate index of the two-dimensional features constituted by the neural network output, q represents the horizontal coordinate index of the two-dimensional features constituted by the neural network output, P and Q represent the height and width of the two-dimensional features constituted by the neural network output respectively; subscript d represents that λ is obtained by data driving;

[0073] The orthogonal feature fusion unit obtains timing synchronization orthogonal fusion features through the following formula:

[0074] Γ=f c (λ m ,λ d )

[0075] wherein, Γ=[γ(1),γ(2),…,γ(N nn )] T ; γ is an element in the fusion feature Γ, f c (·) is an orthogonal feature fusion method, including maximum ratio combination, superposition, multiplication and splicing; N nn represents the length of the fused orthogonal feature vector, which is set by engineering experience.

[0076] In this embodiment, the two-dimensional feature extraction algorithm can be a and b are indexes in inner product summation.

[0077] In this embodiment, the intelligent decision unit obtains timing synchronization estimation value through the following formula

[0078]

[0079] O=[O(0),…,O(τ trial ),…,O(τ max )] T

[0080] O(τ trial )=||yG(Γ,Θ τ ,τ trial )||2

[0081] Among them, τ max is the maximum trial value of the timing offset, y represents the signal received by the receiver OTFS, G(·) represents the intelligent decision model, which consists of an ELM network of SLFN; Θ τ represents the ELM network model parameters; τ trial represents the timing offset trial value, τ trial is the discretization representation of the continuous timing offset, which is determined by the resolution. For example, if the true offset is 12.4ns and the resolution is 1ns, the estimated offset is 12ns, with an ambiguity of 0.4ns. O represents the OTFS timing metric, G(,Θ τ ,) represents the τ Parametric description of the ELM network.

[0082] See also Figure 1 The present invention further provides an embodiment of an OTFS timing synchronization method based on orthogonal feature fusion assistance, which is applicable to the OTFS timing synchronization device based on orthogonal feature fusion assistance described in the above embodiment, and includes:

[0083] Step 1: Based on the OTFS signal y(t) received by the receiver, obtain the N-dimensional initial timing observation signal vector y = [y(0), y(1), ..., y(N-1)] T ;

[0084] Step 2: Obtain N based on the initial timing observation signal vector y nn dimensional timing synchronization orthogonal fusion feature Γ=[γ(0),γ(1),…,γ(N nn -1)] T ;

[0085] Step 3: Obtain the timing synchronization estimate value based on the timing synchronization orthogonal fusion feature Γ

[0086] Step 1 includes:

[0087] like Figure 2In the initial signal blind detection process, it is assumed that the timing synchronization training sequence of the OTFS system is a Zadoff-Chu sequence, the sampling frequency is 32 MHz, and the symbol rate is 4 MHz. The OTFS received signal is sampled and sent to the initial signal blind detection unit to obtain the initial timing observation signal vector, and the vector length N = 256.

[0088] Assuming that autocorrelation is used as the initial signal target detection, the signal detection of the sampled signal is expressed as:

[0089]

[0090] According to Γ i (m), y(nT)=y(t-argmaxΓ) is calculated by i (m)T) for example demonstration:

[0091]

[0092] At this point, the initial timing observation signal vector of the initial signal blind detection unit is y = [y(0), y(1), ..., y(255)] T .

[0093] Step 2 includes:

[0094] The mode-driven orthogonal characteristics are obtained by the following formula:

[0095] λ m =f m (y)

[0096] in, f m (·) represents the two-dimensional feature extraction algorithm process, including 2D-FFT, SFFT, MUSIC and diversity merging;

[0097] The data-driven orthogonal features are obtained by the following formula:

[0098] λ d =f Θ (y,Θ)

[0099] in, f Θ (·) represents the AI ​​reasoning process, including 2D-CNN, GNN, DNN, and RNN; Θ is the neural network parameter, which is obtained by training with real data;

[0100] The timing synchronization orthogonal fusion feature is obtained by the following formula:

[0101] Γ=f c (λ m ,λ d )

[0102] Among them, Γ=[γ(1),γ(2),…,γ(N nn )] T ;f c (·) is the orthogonal feature fusion method, including maximum ratio merging, superposition, multiplication and splicing; N nn Represents the length of the fused orthogonal feature vector, which is set based on engineering experience.

[0103] like Figure 3 In the medium mode driven orthogonal feature extraction, it is assumed that the training sequence for timing synchronization of the OTFS system is the Zadoff-Chu sequence;

[0104] According to the above initial timing observation signal vector y=[y(0),y(1),…,y(255)] T , assuming that the mode-driven orthogonal feature extraction adopts SFFT, the extracted mode-driven orthogonal feature λ m It can represent:

[0105]

[0106] Where KL=N=256 and K=L=16;

[0107] So far, the mode-driven orthogonal characteristics in Example 2 are:

[0108]

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. An OTFS timing synchronization device based on orthogonal feature fusion assistance, characterized in that: It includes an OTFS timing synchronization processing module and an orthogonal feature extraction and fusion module; the OTFS timing synchronization processing module includes an initial signal blind detection unit and an intelligent decision-making unit; the orthogonal feature extraction and fusion module includes a pattern-driven feature extraction unit, a data-driven feature extraction unit and an orthogonal feature fusion unit; The initial signal blind detection unit is used to convert the OTFS signal received by the receiver into an initial timing observation signal vector and send it to the pattern-driven feature extraction unit and the data-driven feature extraction unit respectively; The pattern-driven feature extraction unit and the data-driven feature extraction unit are used to extract the pattern-driven orthogonal features and the data-driven orthogonal features of the initial timing observation signal vector respectively, and send them to the orthogonal feature fusion unit; The orthogonal feature fusion unit is used to fuse the orthogonal features and send the timing synchronization orthogonal fusion features obtained after fusion to the intelligent decision unit; the intelligent decision unit is used to obtain the timing synchronization estimation value based on the timing offset trial value, the timing offset maximum trial value and the timing synchronization orthogonal fusion features; The mode-driven feature extraction unit obtains the mode-driven orthogonal feature by the following formula: l m =f m (y) in, f m (·) represents the two-dimensional feature extraction algorithm; k represents the vertical axis index after two-dimensional domain transformation, l represents the horizontal axis index after two-dimensional domain transformation, K represents the maximum number of sampling points on the vertical axis, and L represents the maximum number of sampling points on the horizontal axis; the subscript m indicates that λ is driven by the modeling mode; The data-driven feature extraction unit obtains the data-driven orthogonal features through the following formula: l d =f Θ (y,Θ) in, f Θ (·) represents the AI ​​reasoning process, Θ is the neural network parameter, p represents the vertical coordinate index of the two-dimensional feature output by the neural network, q represents the horizontal coordinate index of the two-dimensional feature output by the neural network, P and Q represent the height and width of the two-dimensional feature output by the neural network, respectively; the subscript d indicates that λ is driven by data; The orthogonal feature fusion unit obtains the timing synchronous orthogonal fusion feature through the following formula: C=f c (l m ,l d ) Among them, Γ=[γ(1),γ(2),…,γ(N nn )] T ;γ is the element in the fusion feature Γ, f c (·) is the orthogonal feature fusion method; N nn Represents the length of the fused orthogonal feature vector.

2. The OTFS timing synchronization device based on orthogonal feature fusion assistance according to claim 1 is characterized in that: The initial signal blind detection unit is specifically used for: y=[y(0),y(1),…,y(n),…,y(N-1)] T y(nT)=y(t-arg maxΓ i (m)T) Γ i (m)=f τ [y(t)] Where y(t) is the OTFS signal received by the receiver; Γ i (m) is the initial signal blind detection feature, n is the sample index of the discrete signal; t is the observation time; m is the timing offset trial value in the detection feature; f τ [·] denotes a signal detection algorithm with discretization; argmax is a function that returns the maximum value index; T is the sampling interval; and N is the minimum observation length required to obtain a complete signal.

3. The OTFS timing synchronization device based on orthogonal feature fusion assistance according to claim 1 is characterized in that: The two-dimensional feature extraction algorithm can be a and b are indices in the inner product summation.

4. The OTFS timing synchronization device based on orthogonal feature fusion assistance according to claim 1, characterized in that: The intelligent decision-making unit obtains the timing synchronization estimation value through the following formula O=[O(0),…,O(τ trial ),…,O(τ max )] T O(t trial )=||yG(Γ,Θ τ ,t trial )||2 Among them, τ max is the maximum trial value of the timing offset, y represents the signal received by the receiver OTFS, G(·) represents the intelligent decision model, which consists of an ELM network of SLFN; Θ τ represents the ELM network model parameters; τ trial represents the timing offset trial value, τ trial is the discretization representation of the continuous timing offset, which is determined by the resolution; O represents the OTFS timing metric, G(,Θ τ ,) represents the τ Parametric description of the ELM network.

5. An OTFS timing synchronization method based on orthogonal feature fusion assistance, applicable to the OTFS timing synchronization device based on orthogonal feature fusion assistance according to any one of claims 1 to 4, characterized in that: include: Step 1: Based on the OTFS signal y(t) received by the receiver, obtain the N-dimensional initial timing observation signal vector y = [y(0), y(1), ..., y(N-1)] T ; Step 2: Obtain N based on the initial timing observation signal vector y nn dimensional timing synchronization orthogonal fusion feature Γ=[γ(0),γ(1),…,γ(N nn -1)] T ; Step 3: Obtain the timing synchronization estimate value based on the timing synchronization orthogonal fusion feature Γ 6. The OTFS timing synchronization method based on orthogonal feature fusion assistance according to claim 5 is characterized in that: The step 1 comprises: Assuming that autocorrelation is used as the initial signal target detection, the signal detection of the sampled signal is expressed as: According to Γ i (m), y(nT)=y(t-argmaxΓ) is calculated by the following formula i (m)T) to display: The initial timing observation signal vector is y = [y(0), y(1), ..., y(N-1)] T .

7. The OTFS timing synchronization method based on orthogonal feature fusion assistance according to claim 5 is characterized in that: The step 2 includes: The mode-driven orthogonal characteristics are obtained by the following formula: l m =f m (y) in, f m (·) represents the two-dimensional feature extraction algorithm; The data-driven orthogonal features are obtained by the following formula: l d =f Θ (y,Θ) in, f Θ (·) represents the AI ​​reasoning process, Θ is the neural network parameter; The timing synchronization orthogonal fusion feature is obtained by the following formula: C=f c (l m ,l d ) Among them, Γ=[γ(1),γ(2),…,γ(N nn )] T ;f c (·) is the orthogonal feature fusion method, N nn Represents the length of the fused orthogonal feature vector.

8. The OTFS timing synchronization method based on orthogonal feature fusion assistance according to claim 7 is characterized in that: Adopt SFFT extraction mode to drive orthogonal features, Where KL = N and K = L; The extracted mode-driven orthogonal feature λ m express:

Images