A CP-OFDM SAR imaging method combined with interpolation network

Abstract

This invention belongs to the field of radar imaging technology, specifically to a CP-OFDM SAR imaging method combined with an interpolation network. By analyzing the limitations of currently popular CP-OFDM SAR imaging algorithms in specific applications and combining them with the interpolation network GAIN used in deep learning, this method can accurately estimate numerical results that are incalculable in the imaging algorithm even when the dataset is incomplete and the training sample size is limited. Furthermore, the invention does not involve complex computational operations and is relatively simple to implement.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar imaging, and in particular relates to a CP-OFDM SAR imaging method combined with an interpolation network. Background Art

[0002] With the boom in research on the integration of radar and communications, more and more studies are beginning to apply communication signals to radar. One of the research hotspots is the combination of Orthogonal Frequency Division Multiplexing (OFDM) signals and Synthetic Aperture Radar (SAR) imaging. Currently, there are many studies on OFDM SAR imaging algorithms, among which the best imaging effect is OFDM SAR imaging with a cyclic prefix (CP). This algorithm compares the range resolution units in the imaging scene to subchannels in communications. By adding a sufficiently long CP to eliminate interference between range units, it can achieve a zero sidelobe effect in range-dimensional pulse compression, realizing relatively ideal SAR imaging.

[0003] Currently, the concept of CP-OFDM SAR imaging has been gradually applied to fields such as multi-input multi-output (MIMO) SAR imaging and spotlight SAR imaging. However, because this imaging algorithm requires the OFDM signal's frequency domain vector, i.e., the subcarrier coefficients, to have a constant modulus, its application in practical applications is limited. For example, when the OFDM signal's frequency domain vector is randomly composed of "1s" and "0s" or when a DC component is included in the signal's frequency domain vector, the original CP-OFDM SAR imaging algorithm will produce an infinite result in the range compression step, preventing the subsequent imaging steps from proceeding and ultimately failing to obtain an imaging result. Therefore, appropriate improvements to the CP-OFDM SAR imaging algorithm to address the aforementioned situations where "0s" appear in the OFDM signal's frequency domain vector can not only retain the algorithm's advantages but also further expand its application scope. Summary of the Invention

[0004] The present invention addresses the problem of range compression failure caused by the presence of zeros in the OFDM signal's frequency domain vector. This paper proposes a CP-OFDM SAR imaging algorithm incorporating an interpolation network (GAIN). This method is primarily used for the range compression step. To elaborate on the present invention and its specific implementation steps, we first briefly introduce the existing CP-OFDM SAR algorithm.

[0005] Considering a single static side-viewing SAR imaging scenario, assume that the number of subcarriers of the OFDM signal is N and the bandwidth is B. Assume that the frequency domain vector of the signal is S = [S0, S1, ..., S N-1 ] T , the duration of the signal without CP is T, and the length of CP is T GI Assuming that the imaging scene can be divided into M range units, the received signal from the mth range unit can be written as:

[0006]

[0007] Where t and η are fast time variables and slow time variables respectively. a (η) represents the azimuthal envelope, g m Represents the Radar Cross Section (RCS) coefficient, R m (η) represents the instantaneous slant distance, ω(t,η) represents the noise, and by simplifying the above formula, we can get:

[0008]

[0009] Among them, [s0,s1,...,s N+M-2 ] T is the transmission signal sequence, [d0,d1,...,d M-1 ] T is the weighted RCS coefficient. After simplification and Fast Fourier Transform (FFT) operation, formula (2) can be written as:

[0010] U k =D k S k '+W k ,k=0,1,...,N-1 (3)

[0011] Among them, S k '=S k exp{j2pk(M-1) / N}, let D=[D0,D1,...,D N-1 ] T Formula (4) gives D k estimated value.

[0012]

[0013] For the existing CP-OFDM SAR imaging algorithm, its main step is to calculate the d in formula (2) m Estimation is performed to achieve distance compression, and d m The estimate is usually When the digital frequency domain vector of the OFDM signal is composed of "0" and "1" randomly or the DC component is considered in the frequency domain vector, the corresponding S k = 0, which will lead to the corresponding is infinite, so the subsequent imaging processing steps cannot be completed and it is difficult to achieve the final imaging. Through the simulation of some imaging examples and the study of the corresponding discrete distribution of D, we found that the D of different distance units at the same azimuth position is k (k=0,1,...,N-1) are interconnected, and the changing rules of D in different directions are similar. Therefore, the above rules can be used to calculate S k =0 calculated Re-evaluation is performed so that the subsequent imaging steps can be performed according to the original algorithm. The Generative Adversarial Imputation Nets (GAIN) network in deep learning can estimate the value of the random missing type, and its training process does not require a complete data set. Therefore, the present invention combines GAIN with the CP-OFDM SAR imaging algorithm to solve the problem of S k =0 when imaging problem.

[0014] The technical solution of the present invention is: a CP-OFDM SAR imaging algorithm combined with an interpolation network (GAIN), comprising the following steps:

[0015] Step 1: Process the SAR original echo signal u(t,η) according to formulas (1)-(4) to obtain the distance units of the entire imaging scene. ( These values ​​are stored as a two-dimensional matrix according to the corresponding azimuth and distance positions.

[0016] Step 2: Before using GAIN for valuation, Perform preprocessing. Set the median value of the matrix to be infinite Assign 0, and then input the matrix into the GAIN network for training, using right Make an estimate. Let the output be In order to improve the GAIN network To improve the accuracy of the estimation, we can increase the number of radar flights and obtain multiple flight data in the same scene. Input into GAIN to make the estimated value closer to the true value.

[0017] Step 3: By performing IFFT operation, the estimated value of the weighted RCS coefficient in the mth distance unit can be obtained. At this point, the distance compression step is also completed accordingly.

[0018] Step 4: The value obtained in the previous step is subjected to azimuth compression processing according to the traditional matched filtering method, thereby obtaining the final SAR imaging result.

[0019] The present invention considers an application scenario where the CP-OFDM SAR algorithm is limited (i.e., when there is a "0" in the frequency domain vector of the OFDM signal), and proposes applying GAIN estimation to the range compression step to ensure the smooth progress of subsequent processing steps.

[0020] The beneficial effects of the present invention are that the CP-OFDM SAR algorithm is no longer restricted by the condition that "0" exists in the signal frequency domain vector, the application scope of the algorithm is expanded while the imaging accuracy is guaranteed, and the algorithm is simple to implement. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flow chart of the present invention;

[0022] Figure 2 Generated for the echo data of a radar flight Input the imaging result after GAIN processing;

[0023] Figure 3 Generated from echo data from five radar flights Input the imaging result after GAIN processing;

[0024] Figure 4 This is the one-dimensional distance comparison result when different training data are input into GAIN;

[0025] Figure 5 The one-dimensional azimuth comparison results when different training data are input into GAIN;

[0026] Figure 6 The final imaging result obtained after processing with the traditional sinc interpolation algorithm. DETAILED DESCRIPTION

[0027] The present invention is described in detail below with reference to the accompanying drawings and simulation examples to demonstrate the practicability of the present invention.

[0028] As attached Figure 1As shown, through a CP-OFDM SAR imaging algorithm combined with an interpolation network (GAIN) of the present invention, it is possible to use the CP-OFDM SAR algorithm to achieve accurate imaging in the range dimension even when the OFDM signal frequency domain vector contains "0". The specific implementation steps are as follows:

[0029] Step 1: Preprocess the original echo u(t,η) in the range direction, mainly by removing CP and other simplification operations, and the following formula can be obtained:

[0030]

[0031] Among them, [s0,s1,...,s N+M-2 ] T is the transmission signal sequence, [d0,d1,...,d M-1 ] T is the weighted RCS coefficient. Then after simplification and FFT operation, we can get:

[0032] U k =D k S k '+W k ,k=0,1,...,N-1

[0033] Transform the above formula and let S k '=S k exp{j2pk(M-1) / N}, then we can get D k Estimated value of

[0034]

[0035] The distance units of the entire imaging scene obtained by the above formula are ( Also included) Save as a two-dimensional matrix according to the corresponding azimuth position and distance position

[0036] Step 2: To facilitate the processing of GAIN, first set the median of the matrix to be infinite Assign it to 0, then input it into the GAIN network for training, using right Make an estimate. Let the value estimated by GAIN be The output matrix is

[0037] Step 3: By performing the IFFT operation in the distance direction, the estimated value of the weighted RCS coefficient in the mth distance unit can be obtained. At this point, the distance compression step is also completed accordingly.

[0038]

[0039] Step 4: The value obtained in the previous step is compressed in azimuth according to the traditional matched filtering method to obtain the final SAR imaging result.

[0040] Simulation Example

[0041] According to the above method, Figure 1 The main simulation parameters of the process shown are set as follows: considering a single static side-view SAR imaging scenario, the pulse repetition frequency is 800 Hz, the bandwidth of the OFDM signal is 150 MHz, the number of subcarriers is 512, the carrier frequency is 9 GHz, and the duration of the OFDM signal is 3.42 us when CP is not considered. Assuming that the imaging scene can be divided into 96 range units, the corresponding CP length is 95 and the duration is 0.63 us.

[0042] Figure 2 The following is a diagram of the echo data (800 rows and 512 columns) generated from a radar flight. After inputting GAIN processing, the final surface target imaging result is obtained. Although the imaging effect is relatively general, the final imaging result can be obtained, which shows that the improved algorithm is feasible;

[0043] Figure 3 The echo data (3200 rows and 512 columns) generated by five radar flights are given. After inputting GAIN processing, the final surface target imaging result is obtained; Figure 3 and Figure 2 By comparison, it can be seen that with the increase of training data, the imaging quality is significantly improved, and Figure 3 The imaging effect is close to ideal. Combined with the actual situation that the cost of radar data acquisition is relatively high, it can be shown that the improved algorithm has certain practicality.

[0044] Figure 4 This is the one-dimensional distance comparison result when different training data are input into GAIN;

[0045] Figure 5 The one-dimensional azimuth comparison results when different training data are input into GAIN;

[0046] Figure 6 This is the final imaging result obtained after processing with the traditional sinc interpolation algorithm. Comparing the image obtained by GAIN estimation with the result obtained by sinc interpolation shows the superiority of GAIN estimation and proves the performance of this algorithm.

[0047] In summary, judging from the processing results, the method provided by the present invention has practical value and good imaging performance.

Claims

1. A CP-OFDM SAR imaging method combined with an interpolation network, wherein the interpolation network is a generative adversarial estimation network, characterized in that: The following steps are involved: S1. SAR original echo signal Processing, including removing CP and distance FFT, obtains the matrix corresponding to the weighted RCS coefficient estimate ; S2, the matrix The elements with infinite median values ​​are assigned 0, and the processed matrix is ​​sent to the interpolation network for training. right Make an estimate and define the value estimated by the interpolation network as , the output matrix is ; S3, Perform IFFT operation in the range direction to obtain the estimated value of the weighted RCS coefficient in the mth range unit : ， in, is the number of subcarriers of OFDM signal, To divide the imaging scene into distance units; S4, the value obtained Perform azimuth compression processing to obtain the final SAR imaging results.

2. The CP-OFDM SAR imaging method in combination with an interpolation network according to claim 1, characterized in that: The specific method of step S1 is: The imaging scene is defined as a single static side-view SAR imaging scene, and the number of subcarriers of the OFDM signal is , the bandwidth is , the frequency domain vector of the signal is , the duration of the signal without adding CP is , the CP length is ; Divide the imaging scene into distance units, the received signal from the mth distance unit is: ， in, and are fast time variables and slow time variables, respectively. represents the azimuthal envelope, represents the RCS coefficient, represents the instantaneous slant distance, represents noise; after simplification, we get: ， in, is the transmission signal sequence, is the weighted RCS coefficient; after simplification and Fourier transform operation, we get: ， in, ,make ,get The estimated value of is: 。

Images