An OTFS pre-equalization design method for communication and perception integration

Abstract

The present invention provides an OTFS pre-equalization design method for communication-perception integration, belonging to the field of communication-perception integration technology. The method comprises the following steps: constructing an OTFS channel dataset; inputting the training dataset into a channel parameter prediction network for offline training until the channel parameter prediction network converges; reconstructing a predicted delay-Doppler domain communication channel matrix using the predicted channel parameters; inverting the channel reconstruction result, and inputting the obtained inverse matrix along with communication modulation symbols into a precoding design network based on a deep neural network for offline training until the precoding design network converges and the precoding design network parameters are fixed; and inputting a test dataset into the trained channel parameter prediction network to obtain a precoding design result. The present invention can reduce the processing complexity of the receiving end while balancing communication and perception performance requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of communication perception integration, and in particular to an OTFS pre-equalization design method for communication perception integration. Background Art

[0002] Orthogonal Time-Frequency-Space (OTFS) is a communication waveform that can mitigate Doppler shift caused by high-speed mobility. Compared to existing communication waveforms, it exhibits high reliability in high-speed mobility scenarios. OTFS also exhibits excellent communication and perception capabilities, which has sparked extensive research in recent years. To simultaneously meet specific communication and perception requirements, existing work often applies precoding to OTFS waveforms. Wu et al. employed precoding design to minimize the bit error rate of an OTFS-based integrated communication and perception system while ensuring perception accuracy. Keskin et al. proposed a multi-target detection and delay-Doppler-angle estimation algorithm based on a generalized likelihood ratio test for OTFS parameter perception. Based on this, they achieved a trade-off between perceived signal-to-noise ratio and communication rate by jointly optimizing the integrated waveform. However, these studies did not consider the channel obsolescence caused by high device mobility. To address this issue, channel prediction is used to assist communication. Liu et al. proposed a convolutional long short-term memory recurrent neural network (LSTM) that utilizes convolutional neural network modules for spatial feature extraction and long short-term memory modules to capture temporal dependencies. It is possible to predict the communication target angle using historical perception angle information and design beamforming accordingly. However, most of the aforementioned studies that consider channel or parameter prediction only use perception-assisted communication without considering perception performance optimization. In order to simultaneously ensure perception performance, Liu et al. proposed a total rate maximization problem based on the Cramer-Rao lower bound that includes perception constraints. However, this work cannot be applied to two-dimensional waveform optimization such as OTFS. In addition, these works do not consider the actual equalization of communication symbols. The receiver with limited processing power still needs to design a specific equalization algorithm for symbol detection, which may introduce computational processing overhead, which is inconsistent with the common computational processing limitations of the device. Summary of the Invention

[0003] The main purpose of the present invention is to propose an OTFS pre-equalization design method for communication and perception integration, which can reduce the processing complexity of the receiving end while meeting the perception performance requirements.

[0004] The present invention is achieved through the following technical solutions:

[0005] An OTFS pre-equalization design method for integrated communication and perception is provided. The OTFS symbol dimension is M×N, where M represents the number of elements in the delay dimension and N represents the number of elements in the Doppler shift dimension. The method includes the following steps:

[0006] Step S1: construct an OTFS channel data set, where each sample in the channel data set contains the delay, Doppler shift, and channel attenuation parameters of consecutive ζ+1 time slots, and divide the channel data set into a training set and a test set;

[0007] Step S2: input the training set into a channel parameter prediction network based on the neural extension analysis network, and use the first loss function of minimizing the mean square error between the predicted channel parameters and the actual channel parameters to perform offline training on the channel parameter prediction network until the channel parameter prediction network converges, and then fix the channel parameter prediction network parameters;

[0008] Step S3: Using the predicted channel parameters, and according to the relationship between the channel parameters and the OTFS channel, reconstruct the predicted delay-Doppler domain communication channel matrix.

[0009] Step S4: Invert the channel reconstruction result of step S3, and input the obtained inverse matrix together with the communication modulation symbol into the precoding design network based on the deep neural network to obtain the predicted precoding matrix. The precoding design network is offline trained with minimizing the objective function weighted communication mean square error, the perceptual Cramer-Laurent bound, and the regularization term as the second loss function until the precoding design network converges, and the precoding design network parameters are fixed;

[0010] Step S5: input the test set into the trained channel parameter prediction network, invert the output result of the channel parameter prediction network, and input it into the trained precoding design network together with the communication modulation symbol to output the precoding design result.

[0011] Furthermore, in step S1, the OTFS channel parameters are extracted by channel measurement or ray tracing method, and the extraction results of the channel parameters of consecutive ζ+1 time slots are placed in a sample of the data set. This is repeated N' times to complete the extraction of N' samples of the data set, wherein the channel parameters include time delay, Doppler frequency shift and channel attenuation parameters.

[0012] Furthermore, step S2 includes the following steps:

[0013] Step S21: Input the channel number parameters of the past z time slots in the training set data into the channel parameter prediction network as z time steps, and output the predicted channel parameter value of the z+1th time slot;

[0014] Step S22: Substitute the predicted channel parameter value of the z+1th time slot into the first loss function, and perform offline training on the channel parameter prediction network with the goal of minimizing the first loss function;

[0015] Step S23: Repeat steps S21 to S22 until the channel parameter prediction network converges, and fix the channel parameter prediction network parameters.

[0016] Furthermore, the step S4 specifically includes:

[0017] Step S41: Invert the channel reconstruction result, separate the real and imaginary parts of the obtained inverse matrix, and reconnect them in dimension 1 to form a tensor of size (2M×N,M×N), and flatten the tensor into a one-dimensional tensor t z ;

[0018] Step S42: Separate the real and imaginary parts of the communication modulation symbol and reconnect them in dimension 1 to form a one-dimensional tensor t with a size of (2M×N) s ;

[0019] Step S43: tensor t z With tensor t s Connect to form a one-dimensional tensor t of size (2M×N×M×N+2M×N) c ;

[0020] Step S44: tensor t c Input to the fully connected layer using the tanh activation function;

[0021] Step S45: Perform a residual connection on the inverse matrix obtained in step S41 and the result of step S44, and reshape the output into a matrix of size (2M×N, M×N) through the reshape layer.

[0022] Step S46: Matrix The first MN row Corresponding to the real part of the precoding matrix, the last MN rows Corresponding to the imaginary part of the precoding matrix, the real part and the imaginary part of the output are connected to obtain the complex precoding matrix

[0023] Step S47: Power normalize the precoding matrix obtained in step S46 to obtain a predicted precoding matrix. Among them, P max represents the maximum transmit power, |·| F Indicates the F norm;

[0024] Step S48: The predicted precoding matrix Substitute into the second loss function, iteratively train the network parameters until convergence, and fix the precoding design network parameters.

[0025] Furthermore, the OTFS symbol dimension is M×n, where M represents the number of elements in the time delay dimension, and N represents the number of elements in the Doppler shift dimension.

[0026] Furthermore, the delay-Doppler domain communication channel matrix Expressed as Where P represents the number of propagation paths, p represents the p-th path, represents the Kronecker product, h p represents the channel attenuation of path p, Π represents the channel delay matrix, l p =τ p MΔf represents the delay tap of path p, τ p represents the delay of path p, M represents the number of elements in the delay dimension, Δf represents the subcarrier spacing, Δ represents the channel Doppler shift matrix, k p =ν p NT represents the Doppler shift tap of path p, ν p Represents the Doppler frequency shift, N represents the number of elements in the Doppler shift dimension, and T represents the time slot length.

[0027] Furthermore, the second loss function is expressed as Among them, ρ C , ρ S , ρ L They are communication mean square error weight, perception Cramer bound weight, regularization term weight, MSE C is the communication mean square error, CRLB S is the perceptual Cramer bound, and Φ represents the regularization term.

[0028] Furthermore, the communication modulation symbol is a random QPSK modulation sequence.

[0029] From the above description of the present invention, it can be seen that compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. The present invention first constructs an OTFS channel data set and divides the data set into a training set and a test set, then inputs the training set into a channel parameter prediction network, and uses the predicted channel parameters to reconstruct a predicted delay-Doppler domain communication channel matrix, then inverts the channel reconstruction result, and inputs the obtained inverse matrix into the precoding design network together with the communication modulation symbol to obtain a predicted precoding matrix, then inputs the test set into the trained channel parameter prediction network, and inverts the output result of the channel parameter prediction network, and then inputs it into the trained precoding design network together with the communication modulation symbol, and outputs the precoding design result to complete the OTFS equalization design. This process effectively overcomes the impact of the deterioration of communication and perception performance caused by outdated channel state information. Communication users can directly demodulate signals without equalization, thereby significantly reducing the processing complexity of the receiving end, and thus balancing the performance of communication and perception functions in the integrated communication and perception system. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] Figure 1 Flowchart of the present invention.

[0033] Figure 2 This is a diagram of the simulation results of the perceptual Cramer-Rao bound of the present invention.

[0034] Figure 3 This is a simulation result diagram of the communication symbol mean square error of the present invention. DETAILED DESCRIPTION

[0035] The present invention is further described below through specific embodiments.

[0036] like Figure 1 As shown in FIG, an OTFS pre-equalization design method for integrated communication and perception is based on a communication and perception integrated system using an OTFS waveform. The system includes a base station and T = 3 communication users, which are also perception targets. The OTFS symbol dimension used in this system is M×N (i.e., 4×4), where M represents the number of elements in the delay dimension and N represents the number of elements in the Doppler shift dimension.

[0037] The method comprises the following steps:

[0038] Step S1: construct an OTFS channel data set, where each sample in the channel data set contains the delay, Doppler shift, and channel attenuation parameters of consecutive ζ+1 time slots, and divide the channel data set into a training set and a test set;

[0039] Specifically, the OTFS channel parameters are extracted through channel measurement or ray tracing methods. The channel parameter extraction results of consecutive ζ+1=9 time slots are placed in a sample of the data set. This is repeated N'=640 times to complete the extraction of N'=640 samples of the data set. The data set is then divided into training and test sets in proportion. The channel parameters include time delay, Doppler frequency shift, and channel attenuation parameters.

[0040] Step S2: inputting the training set into a channel parameter prediction network based on a neural extended analysis network, using minimization of the mean square error between the predicted channel parameters and the actual channel parameters as a first loss function, and performing offline training on the channel parameter prediction network until the channel parameter prediction network converges, and fixing the channel parameter prediction network parameters. In this embodiment, the channel parameter prediction network is the neural extended analysis network;

[0041] The specific steps include:

[0042] Step S21: Input the channel number parameters of the past ζ=8 time slots in the training set data into the channel parameter prediction network as ζ=8 time steps, and output the predicted channel parameter prediction value of the ζ+1=9th time slot;

[0043] Step S22: Substitute the predicted channel parameter value of the ζ+1=9th time slot into the first loss function, and perform offline training on the channel parameter prediction network with the goal of minimizing the first loss function;

[0044] Step S23: Repeat steps S21 to S22 until the channel parameter prediction network converges, and fix the channel parameter prediction network parameters.

[0045] Step S3: Using the predicted channel parameters, and according to the relationship between the channel parameters and the OTFS channel, reconstruct the predicted delay-Doppler domain communication channel matrix. The matrix is ​​represented as Where P=3 represents the number of propagation paths, p represents the pth path, represents the Kronecker product, h p represents the channel attenuation of path p, which is a random complex number, Π represents the channel delay matrix, l p =τ p MΔf represents the delay tap of path p, τ p represents the delay of path p, which is a random positive number, M represents the number of elements in the delay dimension, Δf = 6000 Hz represents the subcarrier spacing, Δ represents the channel Doppler shift matrix, k p =ν p NT represents the Doppler shift tap of path p, ν prepresents the Doppler shift of path p, which is a random real number, N represents the number of elements in the Doppler shift dimension, and T = 0.0025s represents the time slot length;

[0046] Step S4, inverting the channel reconstruction result of step S3, and inputting the obtained inverse matrix together with the communication modulation symbol into the precoding design network based on the deep neural network to obtain the predicted precoding matrix, minimizing the objective function weighted communication mean square error, the perceptual Cramer-Laurent bound and the regularization term as the second loss function, and performing offline training on the precoding design network until the precoding design network converges, and fixing the precoding design network parameters; wherein, in this embodiment, the communication modulation symbol is a random QPSK modulation sequence; the precoding design network is a deep neural network including four fully connected layers with residual connections between input and output;

[0047] The second loss function is expressed as:

[0048]

[0049] Among them, ρ C =5,ρ S (The value is 10 -11 , 10 -3 , 10 -2 , 10 -1 ,1),ρ L =0.01 are the communication mean square error weight, the perception Cramer bound weight, the regularization term weight, and the MSE weight. C is the communication mean square error, CRLB S is the perceptual Cramer-Laur bound, Φ represents the regularization term;

[0050] The specific steps include:

[0051] Step S41: Invert the channel reconstruction result, separate the real and imaginary parts of the obtained inverse matrix, and reconnect them in dimension 1 to form a tensor of size (2M×N,M×N), and flatten the tensor into a one-dimensional tensor t z ;

[0052] Step S42: Separate the real and imaginary parts of the communication modulation symbol and reconnect them in dimension 1 to form a one-dimensional tensor t with a size of (2M×N) s ;

[0053] Step S43: tensor t z With tensor t s Connect to form a one-dimensional tensor t of size (2M×N×M×N+2M×N) c ;

[0054] Step S44: tensor t cInput to the fully connected layer using the tanh activation function;

[0055] Step S45: Perform a residual connection on the inverse matrix obtained in step S41 and the result of step S44, and reshape the output into a matrix of size (2M×N, M×N) through the reshape layer.

[0056] Step S46: Matrix The first MN row Corresponding to the real part of the precoding matrix, the last MN rows Corresponding to the imaginary part of the precoding matrix, the real part and the imaginary part of the output are connected to obtain the complex precoding matrix

[0057] Step S47: Power normalize the precoding matrix obtained in step S46 to obtain a predicted precoding matrix. Among them, P max represents the maximum transmit power, |·| F Indicates the F norm;

[0058] Step S48: The predicted precoding matrix Substitute into the second loss function, iteratively train the network parameters until convergence, and fix the precoding design network parameters.

[0059] Step S5: input the test set into the trained channel parameter prediction network, invert the output result of the channel parameter prediction network, and input it into the trained precoding design network together with the communication modulation symbol to output the precoding design result.

[0060] Figure 2 In the figure, the horizontal axis is the perception weight coefficient, which takes values ​​of 10 -11 , 10 -3 , 10 -2 , 10 -1 ,1,The vertical axis is the perceptual Cramer-Rao bound. The simulation experiment compares and analyzes four cases: 1) zero-forcing precoding (ZF precoding); 2) precoding designed using real channel state information (real CSI); 3) precoding designed using predicted channel state information (predicted CSI), that is, this method; 4) precoding designed using outdated channel state information (outdated CSI);

[0061] Depend on Figure 2 It can be seen that the present invention can approach the Cramer-Rao bound performance of the precoding designed using real channel state information. Compared with the precoding designed using outdated channel state information, it has a lower Cramer-Rao bound, which means higher perception accuracy. At the same time, as the perception weight coefficient increases, the perception Cramer-Rao bound further decreases, and the perception accuracy is further improved.

[0062] Figure 3 In the figure, the horizontal axis is the perception weight coefficient, which takes values ​​of 10 -11 , 10 -3 , 10 -2 , 10 -1 ,1, the vertical axis is the mean square error of the communication symbols. The simulation experiment compares and analyzes four cases: 1) zero-forcing precoding (ZF precoding); 2) precoding designed using real channel state information (real CSI); 3) precoding designed using predicted channel state information (predicted CSI), that is, this method; 4) precoding designed using outdated channel state information (outdated CSI);

[0063] Depend on Figure 3 It can be seen that the present invention can approach the communication symbol mean square error performance of the precoding designed using real channel state information, and has a lower communication symbol mean square error than the precoding designed using outdated channel state information. At the same time, as the perceptual weight coefficient decreases, the communication symbol mean square error is further reduced.

[0064] from Figure 2 and Figure 3 It can be seen that the present invention can achieve a compromise between communication and perception functions under different communication and perception requirements.

[0065] In the present invention, the terms "first", "second", "third", etc. are only used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. In the description, the directions or positional relationships indicated by "upper", "lower", "left", "right", "front", and "back" are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention, and are not intended to indicate or imply that the device referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the scope of protection of the present invention. For those skilled in the art, the specific meanings of the above terms in this application can be understood according to the specific circumstances.

[0066] In this application, unless otherwise specified, "plurality" refers to two or more. "And / or" describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. The character " / " generally indicates that the related objects are in an "or" relationship.

[0067] The above is only a specific implementation of the present invention, but the design concept of the present invention is not limited to this. Any non-substantial changes to the present invention using this concept shall be deemed as an infringement of the protection scope of the present invention.

Claims

1. An OTFS pre-equalization design method for integrated communication and perception, wherein the OTFS symbol dimension is M×N, where M represents the number of elements in the delay dimension and N represents the number of elements in the Doppler shift dimension. The method is characterized by: The steps include: Step S1: construct an OTFS channel data set, where each sample in the channel data set contains the delay, Doppler shift, and channel attenuation parameters of consecutive ζ+1 time slots, and divide the channel data set into a training set and a test set; Step S2: input the training set into a channel parameter prediction network based on the neural extension analysis network, and use the first loss function of minimizing the mean square error between the predicted channel parameters and the actual channel parameters to perform offline training on the channel parameter prediction network until the channel parameter prediction network converges, and then fix the channel parameter prediction network parameters; Step S3: Using the predicted channel parameters, and according to the relationship between the channel parameters and the OTFS channel, reconstruct the predicted delay-Doppler domain communication channel matrix. Step S4: Invert the channel reconstruction result of step S3, and input the obtained inverse matrix together with the communication modulation symbol into the precoding design network based on the deep neural network to obtain the predicted precoding matrix. The precoding design network is offline trained with minimizing the objective function weighted communication mean square error, the perceptual Cramer-Laurent bound, and the regularization term as the second loss function until the precoding design network converges, and the precoding design network parameters are fixed; Step S5: input the test set into the trained channel parameter prediction network, invert the output result of the channel parameter prediction network, and input it into the trained precoding design network together with the communication modulation symbol to output the precoding design result.

2. The OTFS pre-equalization design method for integrated communication and perception according to claim 1, characterized in that: In step S1, the OTFS channel parameters are extracted by channel measurement or ray tracing, and the channel parameter extraction results of consecutive ζ+1 time slots are placed in a sample of the data set. This is repeated N' times to complete the extraction of N' samples of the data set. The channel parameters include time delay, Doppler shift, and channel attenuation parameters.

3. The OTFS pre-equalization design method for integrated communication and perception according to claim 2, characterized in that: The step S2 comprises the following steps: Step S21: Input the channel number parameters of the past z time slots in the training set data into the channel parameter prediction network as z time steps, and output the predicted channel parameter value of the z+1th time slot; Step S22: Substitute the predicted channel parameter value of the z+1th time slot into the first loss function, and perform offline training on the channel parameter prediction network with the goal of minimizing the first loss function; Step S23: Repeat steps S21 to S22 until the channel parameter prediction network converges, and fix the channel parameter prediction network parameters.

4. The OTFS pre-equalization design method for integrated communication and perception according to claim 3, characterized in that: The step S4 specifically includes: Step S41: Invert the channel reconstruction result, separate the real and imaginary parts of the obtained inverse matrix, and reconnect them in dimension 1 to form a tensor of size (2M×N,M×N), and flatten the tensor into a one-dimensional tensor t z ; Step S42: Separate the real and imaginary parts of the communication modulation symbol and reconnect them in dimension 1 to form a one-dimensional tensor t with a size of (2M×N) s ; Step S43: tensor t z With tensor t s Connect to form a one-dimensional tensor t of size (2M×N×M×N+2M×N c ; Step S44: tensor t c Input to the fully connected layer using the tanh activation function; Step S45: Perform a residual connection on the inverse matrix obtained in step S41 and the result of step S44, and reshape the output into a matrix of size (2M×N, M×N) through the reshape layer. Step S46: Matrix The first MN row Corresponding to the real part of the precoding matrix, the last MN rows Corresponding to the imaginary part of the precoding matrix, the real part and the imaginary part of the output are connected to obtain the complex precoding matrix Step S47: Power normalize the precoding matrix obtained in step S46 to obtain a predicted precoding matrix. Among them, P max represents the maximum transmit power, |·| F Indicates the F norm; Step S48: The predicted precoding matrix Substitute into the second loss function, iteratively train the network parameters until convergence, and fix the precoding design network parameters.

5. The OTFS pre-equalization design method for integrated communication and perception according to any one of claims 1 to 4, characterized in that: The delay-Doppler domain communication channel matrix Expressed as Where P represents the number of propagation paths, p represents the p-th path, represents the Kronecker product, h p represents the channel attenuation of path p, Π represents the channel delay matrix, l p =τ p MΔf represents the delay tap of path p, τ p represents the delay of path p, M represents the number of elements in the delay dimension, Δf represents the subcarrier spacing, Δ represents the channel Doppler shift matrix, k p =ν p NT represents the Doppler shift tap of path p, ν p Represents the Doppler frequency shift, N represents the number of elements in the Doppler shift dimension, and T represents the time slot length.

6. The OTFS pre-equalization design method for integrated communication and perception according to any one of claims 1 to 4, characterized in that: The second loss function is expressed as Among them, ρ C , ρ S , ρ L They are communication mean square error weight, perception Cramer bound weight, regularization term weight, MSE C is the communication mean square error, CRLB S is the perceptual Cramer bound, and Φ represents the regularization term.

7. The OTFS pre-equalization design method for integrated communication and perception according to any one of claims 1 to 4, characterized in that: The communication modulation symbol is a random QPSK modulation sequence.

Images