MIMO radar waveform optimization method, system and device facing signal-to-interference-and-noise ratio

Abstract

This invention discloses a method, system, and apparatus for optimizing MIMO radar waveforms based on signal-to-interference-plus-noise ratio (SINR), relating to the field of radar signal processing technology. The method includes: initializing MIMO radar system parameters; constructing a deep convolutional neural network model for optimizing initial waveform phase data; inputting the initialized waveform phase data into the deep convolutional neural network, processing it through grouped convolution and identity mapping, and outputting the optimized waveform phase; generating a transmitted signal based on the optimized waveform phase and establishing a received signal model; calculating the SINR value corresponding to the current transmitted waveform; comparing the calculated SINR value with a threshold value; if it is greater than the threshold, continuing iterative optimization; otherwise, outputting the current waveform as the optimal waveform. This method utilizes the nonlinear modeling capability and powerful data processing capability driven by neural network data to solve the problem of high computational cost in traditional robust optimization.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, and specifically to a method, system, and apparatus for optimizing MIMO radar waveforms based on signal-to-interference-plus-noise ratio. Background Technology

[0002] MIMO radar transmits different waveforms simultaneously through multiple transmitting antennas and receives echo signals simultaneously through multiple receiving antennas, achieving superior performance compared to traditional phased array radars by utilizing waveform diversity and spatial diversity. In recent years, MIMO radar has become a research hotspot in radar signal processing, and waveform design is a crucial aspect. The core value of waveform optimization lies in its decisive impact on key system performance: First, optimizing the directional power focusing of the waveform can significantly improve the target direction signal-to-interference-plus-noise ratio (SINR), enhancing the detection capability of weak targets in complex electromagnetic environments, which is vital in military scenarios such as air defense, anti-missile defense, and anti-submarine warfare. Second, optimizing the autocorrelation and cross-correlation characteristics of the waveform can reduce sidelobe levels and improve range and angular resolution, providing a guarantee for high-resolution imaging and multi-target detection. Third, in emerging scenarios such as radar-communication integration, waveform design must simultaneously consider detection accuracy and communication rate, becoming a key breakthrough for achieving "integrated perception." Therefore, the goal of waveform optimization is to design a set of optimized transmitted waveforms to maximize the system's performance in target detection, parameter estimation, and anti-interference under hardware and energy constraints.

[0003] The modern electromagnetic environment is becoming increasingly complex. In the military field, there are strong countermeasures such as active suppression and deception interference. In civilian scenarios, there is the impact of multi-system interference caused by spectrum sharing and urban multipath clutter. The effectiveness of traditional radar's "passive defense" method of relying on receiver anti-interference is limited, and the decrease in signal-to-interference-plus-noise ratio of the echo directly leads to the deterioration of detection performance. MIMO radar can maximize SINR by actively designing the pattern of the transmitted waveform, focusing energy in the target direction and forming a null in the interference direction. This is the core path to improve the system's anti-interference capability. At present, the research methods of MIMO radar waveform optimization with the goal of maximizing SINR cover multiple technical paths, including convex optimization and relaxation methods. These methods can obtain the global optimal lower bound, but their performance is limited in constant modulus constraint scenarios. Alternating direction multiplier methods and other alternating optimization and decomposition methods solve complex problems by partitioning them into blocks. They have the advantages of convergence stability and parallelization, and are the mainstream choice in engineering implementation. Manifold optimization methods use Riemannian manifolds constructed by constant modulus constraints to search for optimal solutions without additional relaxation processing. They are more suitable for large-scale MIMO systems, but they also have the problem of high computational complexity.

[0004] In summary, current waveform optimization research suffers from multiple constraints such as constant mode coupling, which leads to non-convexity of the problem and forms an NP-HARD problem. Traditional convex optimization is difficult to solve directly, and the contradiction between computational complexity and real-time performance is prominent. The computational load increases dramatically with the number of antenna channels, and the optimization performance is highly dependent on accurate channel information. Mismatch of target parameters or dynamic changes in the environment will significantly degrade SINR, making it difficult to meet the waveform update requirements of cognitive radar. Summary of the Invention

[0005] To address the shortcomings of existing technologies where computational complexity increases dramatically with the number of antenna channels, making it difficult to meet the needs of cognitive radar waveform updates, this invention proposes a MIMO radar waveform optimization method, system, and device oriented towards signal-to-interference-plus-noise ratio (SINR). By introducing a convolutional neural network into the MIMO radar waveform optimization process aimed at maximizing SINR, the problem of existing technologies is solved.

[0006] A MIMO radar waveform optimization method oriented towards signal-to-interference-plus-noise ratio includes the following steps: Acquire the initial waveform phase data of the MIMO radar; Initial waveform phase data is input into a pre-trained waveform optimization model based on a deep convolutional neural network. Phase feature maps are extracted by convolution operations on the initial waveform phase data. Max pooling is then performed on the phase feature maps along the waveform length dimension to output pooled features. The number of channels in the pooled features is divided into multiple groups, and each group of channels is convolved independently. Intermediate features are generated by concatenating the output features of each group along the channel dimension. Convolution operations are then performed on the intermediate features to fuse the output features across channels, resulting in a fused feature with the same number of channels as the pooled features. The fused feature is then element-wise added to the pooled features to obtain an enhanced feature after residual concatenation. Optimized waveform phase data that satisfies constant modulus constraints is obtained based on the enhanced features. The transmitted signal is generated based on the optimized waveform phase, and a received signal model is established based on the received signal model after the transmitted signal is reflected by the target and received by the radar's receiving antenna. Based on the received signal model, the signal-to-interference-plus-noise ratio (SIR) of the transmitted signal is calculated. The waveform corresponding to the transmitted signal with an SIR value less than a preset threshold is taken as the optimal waveform of the MIMO radar.

[0007] Furthermore, the initial waveform phase data includes the number of transmitting antennas N. t Number of receiving antennas N r Waveform length M, signal-to-interference-plus-noise ratio threshold and the receive steering vector of the antenna array calculated based on the target angle. and launch guidance vector .

[0008] Furthermore, the received signal model is expressed as: ; in, For K interference sources, the transmitted signal The sum of the resulting interference signals; The mean is 0 and the variance is Additive white Gaussian noise; The target reflection coefficient is; the transmitted signal is The m-th transmitted waveform vector of a MIMO radar system based on Nt transmit antennas is: The optimized m-th transmitted waveform of the n-th antenna is represented as follows: , , .

[0009] Furthermore, the calculation of the signal-to-interference-plus-noise ratio (SIR) corresponding to the transmitted signal based on the received signal model specifically includes the following steps: Regarding the received signal power, according to the received signal model The average power after M samplings is expressed as follows: ; in, For the transmitted waveform The covariance matrix; Similarly, the transmitted signal is obtained. The echo energy power generated by K interference sources is: ; in, For the first k The emission coefficient of the interference; After M samplings by Nr receiving antennas, the total noise power is: ; Based on the total noise power, the radar transmission waveform after optimization and iteration is obtained. The corresponding signal-to-interference-plus-noise ratio of the echo is: .

[0010] Furthermore, the number of channels N of the pooling feature is... t Divided into G groups, each group contains N t / G channels; perform independent convolution on each group of channel data, and concatenate the output features of each group of convolutions along the channel dimension to generate N channel data. t Intermediate features of ×G.

[0011] Furthermore, before using the waveform corresponding to the transmitted signal with a signal-to-interference-plus-noise ratio (SIR) less than a preset threshold as the optimal waveform for the MIMO radar, an iterative comparison is performed. Specifically, this includes: comparing the currently optimized SIR with the preset threshold; if the SIR is greater than the threshold, using the currently optimized waveform phase data as new input waveform phase data and inputting it into the waveform optimization model for further optimization; if the SIR is less than or equal to the threshold, terminating the iteration and outputting the waveform corresponding to the currently transmitted signal as the optimal waveform.

[0012] Furthermore, the process of obtaining optimized waveform phase data that satisfies constant modulus constraints based on the enhancement features is specifically achieved by sequentially inputting the enhancement features into a fully connected layer and an activation function layer. The activation function layer uses an activation function that satisfies constant modulus constraints to map the output of the fully connected layer to generate optimized waveform phase data.

[0013] The present invention also includes a MIMO radar waveform optimization system for signal-to-interference-plus-noise ratio, comprising: The acquisition module is used to acquire the initial waveform phase data of the MIMO radar; The optimization module is used to input initial waveform phase data into a pre-trained waveform optimization model based on a deep convolutional neural network. It extracts phase feature maps by performing convolution operations on the initial waveform phase data; performs max pooling along the waveform length dimension on the phase feature maps to output pooled features; divides the pooled features into multiple groups based on their channel count, and performs independent convolution on each group of channel data; concatenates the output features of each group of convolutions along the channel dimension to generate intermediate features; performs convolution operations on the intermediate features to fuse the output features of each group across channels, obtaining a fused feature with the same number of channels as the pooled features; adds the fused feature to the pooled features element-wise to obtain an enhanced feature after residual connection; and obtains optimized waveform phase data that satisfies constant modulus constraints based on the enhanced features. The output module is used to generate a transmitted signal based on the optimized waveform phase and to establish a received signal model of the transmitted signal after being reflected by the target and received by the radar's receiving antenna. Based on the received signal model, the signal-to-interference-plus-noise ratio (SIR) of the transmitted signal is calculated. The waveform of the transmitted signal with an SIR value less than a preset threshold is taken as the optimal waveform of the MIMO radar.

[0014] The present invention also includes a computer device for optimizing MIMO radar waveforms based on signal-to-interference-plus-noise ratio (SINR), comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the MIMO radar waveform optimization method based on SINR.

[0015] The present invention also includes a readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, are used to perform the steps of the MIMO radar waveform optimization method for signal-to-interference-plus-noise ratio.

[0016] This invention provides a MIMO radar waveform optimization method oriented towards signal-to-interference-plus-noise ratio (SINR), which has the following beneficial effects: This invention introduces convolutional neural networks into MIMO radar waveform optimization processing aimed at maximizing SINR. Grouped convolution divides the input channel into several subgroups for parallel convolution, significantly reducing computational complexity and the number of parameters. This adapts to the limited computing power and storage resources of radar platforms, enabling lightweight model deployment. Simultaneously, grouped convolution enhances channel feature independence, aligning with the characteristic of different transmit antenna waveform parameters corresponding to different channels in MIMO radar, avoiding feature redundancy between channels, and thus accurately mapping the differentiated relationship between each antenna waveform and SINR, improving the targeting of waveform optimization. Identity mapping achieves lossless gradient backpropagation through direct connections, solving the gradient vanishing problem in deep networks and ensuring the stability of constant modulus constraints and the complex nonlinear relationship between SINR. It also effectively integrates waveform phase data features extracted by shallow networks with waveform optimization strategies learned by deep networks under interference environments, improving SINR optimization accuracy. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the MIMO radar waveform optimization method for signal-to-interference-plus-noise ratio in an embodiment of the present invention. Figure 2 This is a schematic diagram of a deep neural network structure used for waveform optimization in an embodiment of the present invention; Figure 3 This is a schematic diagram comparing the power direction of the transmitted waveforms before and after optimization in an embodiment of the present invention after the completion of the received beamforming process. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] This invention proposes a waveform optimization method for MIMO radar focusing on signal-to-interference-plus-noise ratio (SINR). This method introduces a deep convolutional neural network (CNN) into the waveform optimization field, leveraging the data-driven nonlinear modeling capabilities and powerful data processing capabilities of the neural network. It eliminates the need for relaxation of non-convex constraints such as constant modulus, and by embedding a constraint penalty term in a newly constructed loss function, it can directly learn the nonlinear mapping between the waveform and SINR, avoiding the performance loss and high computational complexity of existing methods. Furthermore, relying on an "offline training + online inference" model, the trained neural network can quickly output the optimal waveform adapted to the current interference and system state, overcoming the bottleneck of high computational cost in iterative algorithms for large-scale MIMO systems and solving the problem of high computational cost in traditional robust optimization. This method effectively overcomes the bottlenecks of traditional optimization algorithms in non-convex constraints, real-time performance, and multi-objective compatibility, powerfully promoting the application of theoretical achievements in waveform optimization for SINR problems to engineering practice.

[0020] like Figure 1 As shown, the method specifically includes the following steps: S1. Determine the number N of the MIMO radar's transmitting antennas. t Receiving antenna N r Waveform length M and threshold .

[0021] S2. Based on the target's direction of arrival angle Calculate the receive steering vector of the antenna array. and launch guidance vector .

[0022] S3. Generate randomized initial waveform phase data of size N. t ×M, and then feed the waveform phase data (referring to: randomized initial waveform phase data) into the deep neural network for SINR-oriented MIMO radar waveform optimization designed in this invention. The deep neural network structure for waveform optimization in this invention is as follows: Figure 2 As shown.

[0023] Considering that the data dimension and computational complexity of MIMO radar to be optimized will increase with the number of transmitting antennas N tWith the increase in the number of waveforms M, the SINR (Signal Indicator Ratio) increases significantly. This invention, based on traditional convolutional neural networks, introduces grouped convolution and identity mapping mechanisms to construct a waveform optimization processing model suitable for large-scale data scenarios, aiming to maximize SINR. Grouped convolution divides the input channels into several subgroups of parallel convolutions, significantly reducing computational complexity and the number of parameters. This adapts to the limited computing power and storage resources of radar platforms, enabling lightweight model deployment. Simultaneously, grouped convolution enhances the independence of channel features, aligning with the characteristic of different transmit antenna waveform parameters corresponding to different channels in MIMO radar, avoiding feature redundancy between channels, and thus accurately corresponding to the differentiated mapping relationship between each antenna waveform and SINR, improving the targeting of waveform optimization. Identity mapping achieves lossless gradient backpropagation through direct connections, solving the gradient vanishing problem in deep networks, ensuring the stability of constant modulus constraints and the complex nonlinear relationship between SINR, and effectively integrating waveform phase data features extracted by shallow networks with waveform optimization strategies learned by deep networks under interference environments, improving SINR optimization accuracy.

[0024] S4, with a size of N t The waveform phase data of ×M is fed into a deep convolutional neural network for waveform optimization. First, the data undergoes a preprocessing module that includes convolution, normalization, and ReLU activation, resulting in N channels. t The preprocessing results are then used; the data is then subjected to maximization pooling to reduce the data size while strengthening the correlation between the parameters and the SINR optimization objective; the resulting N is then... t The data from each channel is divided into G groups, with each group containing N channels. t / G) are processed by grouped convolution, and the convolution output is N. t Data from 1 channel; G groups of data are grouped and convolved to obtain N data. t The batch of data consists of ×G channels. This batch of data is then merged and convolutionally processed to restore the data to N channels. t The channel is then used; the result of the previous maximum pooling is then identity-connected with the merged convolution result, which solves the gradient vanishing problem in deep networks and ensures the accuracy of waveform optimization; finally, the output waveform is passed through a fully connected layer and an activation function for signal-to-interference-plus-noise ratio optimization, and the output waveform size remains the same. Based on the optimized phase distribution and the constant modulus constraint requirements of the waveform, the activation function designed in this invention is as follows:

[0025] .

[0026] S5. Calculate the transmitted and received signals based on the output waveform of the neural network. Here, the result after this optimization is... n The first antenna m The transmitted waveform is represented as follows: , , It is possible to obtain N based on t The MIMO radar system with the first transmitting antenna at the transmitting end... m The transmitted waveform vectors are: Then the transmitted signal can be obtained as The transmitted signal is reflected by the target and detected by the radar's N. r The received signal, as received by each receiving antenna, can be represented as:

[0027] ; in, For K interference sources, the transmitted signal The sum of the resulting interference signals, The mean is 0 and the variance is Additive white Gaussian noise, The target reflectance coefficient.

[0028] S6. Calculate the signal-to-interference-plus-noise ratio (SINR) based on the output waveform of the neural network. For the received signal power, it can be determined according to the signal model. The average power after M samplings is expressed as follows:

[0029] ; in, For the transmitted waveform The covariance matrix, Indicates conjugate. This indicates the conjugate transpose. This represents the mathematical expectation. Similarly, the transmitted signal can be obtained. The echo energy power generated by K interference sources is ( For the first k (Emission coefficient of each interference)

[0030] .

[0031] N r After M samplings by the receiving antennas, the total noise power obtained by the system is: Based on the above results, the radar transmission waveform after optimization iteration via neural network can be obtained. The corresponding signal-to-interference-plus-noise ratio of the echo is:

[0032] .

[0033] S7. Compare the SINR of the waveform after this optimization iteration with the threshold value. The waveform data is compared. When the waveform exceeds the threshold, it is saved and returned to the neural network for further optimization. When the waveform is less than or equal to the threshold, the optimization iteration ends and the current waveform data is used as the final optimized waveform result. This waveform can achieve the optimal SINR performance under certain parameter conditions when used in MIMO radar.

[0034] S8. Verify the effectiveness of this method. The antenna array is configured with N transmitting antennas. t =16, Number of receiving antennas N r =8, waveform length 256, assuming target azimuth 10 degrees, there are two strong interference signals in the -60 degree and 40 degree directions respectively. After optimization using this method, SINR increased from -22.76dB to 8.80dB, demonstrating that this method can effectively suppress interference and improve the system's anti-interference capability. Figure 3 To optimize the power pointing of the transmitted waveforms before and after receiving beamforming, a comparison was made.

[0035] This invention introduces convolutional neural networks into MIMO radar waveform optimization aimed at maximizing SINR, achieving multiple core effects: precise adaptation to non-convex constraints such as constant modulus, avoiding the performance loss of relaxation processing in traditional methods, and significantly improving SINR and weak target detection capabilities in complex electromagnetic environments; reducing modeling complexity and solution difficulty, meeting the needs of dynamic radar scenarios, and breaking through the computational complexity bottleneck of waveform optimization processing in large-scale MIMO systems; and, depending on specific requirements, it can also integrate constraints such as power and spectrum to achieve multi-objective collaborative optimization. However, because this invention uses convolutional neural networks from deep learning, it also suffers from problems such as poor interpretability of the solution model and limited generalization ability.

[0036] Based on the same inventive concept, this invention also proposes a MIMO radar waveform optimization system oriented towards signal-to-interference-plus-noise ratio, comprising: The acquisition module is used to acquire the initial waveform phase data of the MIMO radar.

[0037] The optimization module is used to input the initial waveform phase data into a waveform optimization model based on a deep convolutional neural network. It extracts phase feature maps by performing convolution operations on the initial waveform phase data; it then performs max pooling along the waveform length dimension to output pooled features; it divides the pooled features into multiple groups, performs independent convolution on each group of channels, and concatenates the outputs of each group along the channel dimension to generate intermediate features; it performs convolution operations on the intermediate features to fuse the features from each group, outputting fused features; it adds the fused features element-wise to the pooled features to obtain enhanced features after residual concatenation; and finally, it obtains optimized waveform phase data that satisfies constant modulus constraints based on the enhanced features.

[0038] The output module is used to generate a transmitted signal based on the optimized waveform phase and to establish a received signal model of the transmitted signal after being reflected by the target and received by the radar's receiving antenna. Based on the received signal model, the signal-to-interference-plus-noise ratio (SIR) corresponding to the current transmitted waveform is calculated. The current waveform with an SIR value less than a preset threshold is taken as the optimal waveform for the MIMO radar.

[0039] The present invention also proposes a computer device for optimizing MIMO radar waveforms based on signal-to-interference-plus-noise ratio (SINR), comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the SINR-based MIMO radar waveform optimization method.

[0040] The present invention also proposes a readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, are used to perform steps of a MIMO radar waveform optimization method oriented towards signal-to-interference-plus-noise ratio.

[0041] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A MIMO radar waveform optimization method oriented towards signal-to-interference-plus-noise ratio, characterized in that, Includes the following steps: Acquire the initial waveform phase data of the MIMO radar; Initial waveform phase data is input into a pre-trained waveform optimization model based on a deep convolutional neural network. Phase feature maps are extracted by convolution operations on the initial waveform phase data. Max pooling is then performed on the phase feature maps along the waveform length dimension to output pooled features. The number of channels in the pooled features is divided into multiple groups, and each group of channels is convolved independently. Intermediate features are generated by concatenating the output features of each group along the channel dimension. Convolution operations are then performed on the intermediate features to fuse the output features across channels, resulting in a fused feature with the same number of channels as the pooled features. Finally, the fused feature is element-wise added to the pooled features to obtain the enhanced feature after residual connection. The optimized waveform phase data that satisfies the constant modulus constraint is obtained based on the enhanced features; The transmitted signal is generated based on the optimized waveform phase, and a received signal model is established based on the received signal model after the transmitted signal is reflected by the target and received by the radar's receiving antenna. Based on the received signal model, the signal-to-interference-plus-noise ratio (SIR) of the transmitted signal is calculated. The waveform corresponding to the transmitted signal with a signal-to-interference-plus-noise ratio (SINR) less than a preset threshold is taken as the optimal waveform for the MIMO radar.

2. The MIMO radar waveform optimization method for signal-to-interference-plus-noise ratio as described in claim 1, characterized in that, The initial waveform phase data includes the number of transmitting antennas N. t Number of receiving antennas N r Waveform length M, signal-to-interference-plus-noise ratio threshold and the receive steering vector of the antenna array calculated based on the target angle. and launch guidance vector .

3. The MIMO radar waveform optimization method for signal-to-interference-plus-noise ratio as described in claim 2, characterized in that, The received signal model is represented as follows: ； in, For K interference sources, the transmitted signal The sum of the resulting interference signals; The mean is 0 and the variance is Additive white Gaussian noise; The target reflection coefficient is; the transmitted signal is Based on N t The vector of the m-th transmitted waveform of a MIMO radar system with multiple transmitting antennas at the transmitter is: The optimized m-th transmitted waveform of the n-th antenna is represented as follows: , , .

4. The MIMO radar waveform optimization method oriented towards signal-to-interference-plus-noise ratio according to claim 3, characterized in that, The calculation of the signal-to-interference-plus-noise ratio (SINR) of the transmitted signal based on the received signal model specifically includes the following steps: Regarding the received signal power, according to the received signal model The average power after M samplings is expressed as follows: ； in, For the transmitted waveform The covariance matrix; Similarly, the transmitted signal is obtained. The echo energy power generated by K interference sources is: ； in, For the first k The emission coefficient of the interference; N r After M samplings by the receiving antennas, the total noise power obtained is: ; Based on the total noise power, the radar transmission waveform after optimization and iteration is obtained. The corresponding signal-to-interference-plus-noise ratio of the echo is: 。 5. The MIMO radar waveform optimization method for signal-to-interference-plus-noise ratio as described in claim 1, characterized in that, The number of channels N of the pooling feature t Divided into G groups, each group contains N t / G channels; perform independent convolution on each group of channel data, and concatenate the output features of each group of convolutions along the channel dimension to generate N channel data. t Intermediate features of ×G.

6. The MIMO radar waveform optimization method for signal-to-interference-plus-noise ratio as described in claim 1, characterized in that, Before using the waveform corresponding to the transmitted signal with a signal-to-interference-plus-noise ratio (SIR) less than a preset threshold as the optimal waveform for the MIMO radar, the process includes iterative comparison. Specifically, this involves comparing the currently optimized SIR with the preset threshold. If the SIR is greater than the threshold, the currently optimized waveform phase data is used as new input waveform phase data and input into the waveform optimization model for further optimization. If the SIR is less than or equal to the threshold, the iteration is terminated, and the waveform corresponding to the currently transmitted signal is output as the optimal waveform.

7. The MIMO radar waveform optimization method for signal-to-interference-plus-noise ratio as described in claim 1, characterized in that, The optimized waveform phase data that satisfies the constant modulus constraint is obtained based on the enhancement features by sequentially inputting the enhancement features into a fully connected layer and an activation function layer. The activation function layer uses an activation function that satisfies the constant modulus constraint to map the output of the fully connected layer to generate the optimized waveform phase data.

8. A MIMO radar waveform optimization system oriented towards signal-to-interference-plus-noise ratio, characterized in that, include: The acquisition module is used to acquire the initial waveform phase data of the MIMO radar; The optimization module is used to input initial waveform phase data into a pre-trained waveform optimization model based on a deep convolutional neural network. It extracts phase feature maps by performing convolution operations on the initial waveform phase data; it then performs max pooling along the waveform length dimension to output pooled features; it divides the pooled features into multiple groups based on the number of channels, and performs independent convolution on each group of channels; it concatenates the output features of each group of convolutions along the channel dimension to generate intermediate features; it performs convolution operations on the intermediate features to fuse the output features of each group across channels, obtaining a fused feature with the same number of channels as the pooled features; and it adds the fused feature to the pooled features element-wise to obtain the enhanced feature after residual connection. The optimized waveform phase data that satisfies the constant modulus constraint is obtained based on the enhanced features; The output module is used to generate a transmitted signal based on the optimized waveform phase and to establish a received signal model of the transmitted signal after being reflected by the target and received by the radar's receiving antenna; based on the received signal model, the signal-to-interference-plus-noise ratio (SIR) of the transmitted signal is calculated. The waveform corresponding to the transmitted signal with a signal-to-interference-plus-noise ratio (SINR) less than a preset threshold is taken as the optimal waveform for the MIMO radar.

9. A computer device for optimizing MIMO radar waveforms based on signal-to-interference-plus-noise ratio, characterized in that, include: A memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the MIMO radar waveform optimization method for signal-to-interference-plus-noise ratio as described in any one of claims 1-7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program, which includes program instructions that, when executed by a processor, are used to perform the steps of the MIMO radar waveform optimization method for signal-to-interference-plus-noise ratio as described in any one of claims 1-7.

Images