Distributed signal non-coherent fusion detection method based on variable indexes

Abstract

The invention relates to a distributed signal non-coherent fusion detection method based on variable indexes, and belongs to the field of radar signal processing. The method comprises the following steps: aiming at each resolution unit grid of a unified distance-azimuth-pitching three-dimensional observation space, extracting to-be-detected unit data and reference unit data of each observation channel of a distributed MIMO radar, and constructing AMF CFAR detection statistics of each observation channel of the resolution unit grid as local statistics; setting a probability that a uniform background is misjudged as a non-uniform background by using a non-target condition local statistic CFAR characteristic, and obtaining a screening effective local statistic uniformity judgment threshold through Monte-Carlo simulation by adopting a variable index-based uniformity judgment method; according to the false alarm probability of the detection threshold, iterative optimization and Monte-Carlo simulation are combined, and the detection threshold required for detection of global statistics formed after non-coherent fusion of different numbers of effective local statistics is solved; and traversing all resolution unit grids by using the obtained uniformity discrimination threshold and detection threshold to complete observation space detection.

Description

Technical Field

[0001] The invention belongs to the field of radar signal processing, and in particular relates to a distributed signal detection technology. Background Art

[0002] Distributed MIMO radar uses the spatial scattering diversity of the target RCS to obtain spatial diversity gain, thereby improving the target detection capability. However, whether the statistical characteristics of each local statistic are consistent has a great impact on the detection performance of the distributed MIMO radar. Generally speaking, the target characteristics and their interactions with the radar and the environment usually lead to inconsistencies in the statistical characteristics of each local statistic. The most important of these is the inconsistency of the signal-to-noise ratio contained in each local statistic, which complicates the problem of non-coherent fusion between local statistics. The main reasons for this inconsistency include: (1) The distances of the same target relative to different receiving stations are different, resulting in differences in the target echo energy; (2) Due to different observation angles, the instantaneous values ​​of the target RCS obtained by different receiving stations are different; (3) The clutter characteristics obtained by different receiving stations and the relative characteristics of clutter and target are different. Therefore, after the clutter suppression of each station itself, the remaining target energy is also different, resulting in a large difference in signal-to-noise ratio; (4) The measurement error makes it necessary to expand the number of "range-azimuth-elevation" search grids of each receiving channel during the fusion process, which will inevitably lead to the absence of targets in some local statistics involved in the fusion.

[0003] If the signal-to-noise ratio of a local statistic is not significantly lower than that of other local statistics, then the local statistic is called a valid local statistic. Since not all local statistics involved in non-coherent fusion detection are valid local statistics, a direct idea is to select valid local statistics from these local statistics and perform non-coherent fusion processing. There are two problems to be solved in this idea: (1) Problems with screening methods. From the perspective of scope of application, data-based adaptive screening methods are obviously superior to non-adaptive screening methods such as direct non-coherent summation, taking the maximum value, and taking the sum of the first several larger values.

[0004] (2) After non-coherent fusion, the problem of determining the detection threshold. Since the number of valid local statistics obtained in each screening is different, the detection statistics after fusion are statistically different, and the corresponding detection threshold needs to be determined according to the number of local statistics fused.

[0005] In order to solve the above problems, the present invention provides a distributed signal non-coherent fusion detection method based on variable index. Summary of the invention

[0006] The purpose of the present invention is to provide a distributed signal non-coherent fusion detection method based on a variable index to solve the problems of inconsistent signal-to-noise ratios of local statistics and poor detection effect of direct non-coherent fusion faced by distributed MIMO radar non-coherent fusion detection. The technical problems to be solved include: (1) The problem of screening effective local statistics based on variable index; (2) Under the premise that each local statistic has the CFAR characteristic, the problem of obtaining the detection threshold is solved.

[0007] The variable index-based distributed signal non-coherent fusion detection method of the present invention is characterized by comprising the following technical measures: Step 1: For each resolution unit grid in the unified "range-azimuth-elevation" three-dimensional observation space, extract the detection unit data and reference unit data in each observation channel of the corresponding distributed MIMO radar, and then construct the AMFCFAR detection statistics of each observation channel corresponding to the resolution unit grid as the local statistics based on the distributed signal parameters and the complex Gaussian clutter background; Step 2: Using the CFAR characteristics of local statistics under target-free conditions and the probability of misjudging a uniform background as a non-uniform background, a uniformity determination method based on a variable index is adopted to obtain a uniformity determination threshold for screening effective local statistics through Monte-Carlo simulation; Step 3: According to the false alarm probability corresponding to the detection threshold, a method combining iterative optimization and Monte-Carlo simulation is used to solve the detection threshold required for detection of the global statistics formed by non-coherent fusion of different numbers of effective local statistics; Step 4: Using the uniformity discrimination threshold obtained in step 2 and the detection threshold obtained in step 3, traverse all the resolution unit grids in step 1 to complete the detection of the entire observation space.

[0008] Compared with non-adaptive screening methods such as direct non-coherent summation, taking the maximum value, taking the sum of the first several larger values, the distributed signal non-coherent fusion detection method based on variable index described in the present invention has the following beneficial effects: (1) This method solves the problem of screening effective local statistics by using a uniformity determination method based on a variable index.

[0009] (2) This method combines iterative optimization with Monte-Carlo simulation to solve the problem of finding the detection threshold corresponding to the sum of different numbers of effective local statistics. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1It is a flow chart of a distributed signal non-coherent fusion detection method based on variable index. DETAILED DESCRIPTION

[0011] The present invention will be described in further detail below in conjunction with the accompanying drawings. Figure 1 The specific implementation of the present invention is divided into the following steps: (1) For each resolution unit grid in the unified “range-azimuth-elevation” three-dimensional observation space, the data of the unit to be detected and its reference unit data in each observation channel of the corresponding distributed MIMO radar are extracted. Then, based on the distributed signal parameters and the complex Gaussian clutter background, the AMF CFAR detection statistics of each observation channel corresponding to the resolution unit grid are constructed as local statistics. (2) Using the CFAR characteristics of local statistics under target-free conditions and the probability of misjudging a uniform background as a non-uniform background, a uniformity judgment method based on a variable index is adopted. Through Monte-Carlo simulation, the uniformity judgment threshold for screening effective local statistics is obtained. (3) Based on the false alarm probability corresponding to the detection threshold, a method combining iterative optimization and Monte-Carlo simulation is used to solve the detection threshold required for detection using the global statistics formed by incoherent fusion of different numbers of effective local statistics. (4) Using the uniformity discrimination threshold obtained in step (2) and the detection threshold obtained in step (3), traverse all the resolution unit grids in step (1) to complete the detection of the entire observation space.

[0012] The following is a detailed description of the above steps: (1) For each resolution unit grid in the unified “range-azimuth-elevation” three-dimensional observation space, extract the to-be-detected unit data and its reference unit data in each observation channel of the corresponding distributed MIMO radar, and then construct the AMF CFAR detection statistics of each observation channel corresponding to the resolution unit grid as the local statistics based on the distributed signal parameters and the complex Gaussian clutter background. The specific operations of this step are as follows: ① For distributed MIMO radar, a unified “range-azimuth-elevation” three-dimensional observation space is established, and the space is divided into grids according to the distance resolution unit size. The grid numbers are recorded as , the total is recorded as , ; ② Distributed MIMO radar includes observation channel, corresponding to the resolution unit grid, The data of the unit to be detected obtained in the observation channel is recorded as , the reference unit data is recorded as , , ; The reference unit data does not contain the target and the mean is , the covariance is Complex Gaussian distribution; without target, The mean is , the covariance is Complex Gaussian distribution; under target conditions, The mean is , the covariance is The complex Gaussian distribution of ③ Construct the corresponding The AMF CFAR detection statistic of each observation channel of the resolution unit grid is called the local statistic and is denoted as , , , in, Represents the sample covariance matrix, which is the covariance matrix estimates; ④ Traverse all resolution unit grids and obtain the local statistics corresponding to each resolution unit grid. , , .

[0013] (2) Using the CFAR characteristics of local statistics under target-free conditions and the probability of misjudging a uniform background as a non-uniform background, a uniformity judgment method based on a variable index is adopted to obtain the uniformity judgment threshold for screening effective local statistics through Monte-Carlo simulation. The specific operations of this step are as follows: ① The probability of misjudging a uniform background as a non-uniform background is set to 0.1; the number of Monte-Carlo simulations is set to ; Pick one , denoted as ; ② No. , , in the next Monte-Carlo simulation, refer to the data format and calculation method of ② and ③ in step (1) to generate the simulation The local statistics of the units to be detected without targets, Recorded as , ; Then sort them from large to small, record them as, , For the sorted sequence , The variable index is calculated as follows , , ,and Always equal to 1; ③ Repeat step ② times, get Variable index sequence , ; Extract the first Elements of samples, arrange them from large to small, and take the The item is used as the uniformity judgment threshold corresponding to the element, denoted as ; Use this operation to traverse all elements in the variable index sequence and get The uniformity judgment threshold is denoted as ; The CFAR characteristic ensures the uniformity judgment threshold Universality across all resolution unit grids.

[0014] (3) According to the false alarm probability corresponding to the detection threshold, a method combining iterative optimization and Monte-Carlo simulation is used to solve the detection threshold required for detection of the global statistics formed by the incoherent fusion of different numbers of effective local statistics. The specific operations of this step are as follows: ① Set the false alarm probability corresponding to the detection threshold to , setting the initial false alarm probability required for iterative optimization operation to ; The number of Monte-Carlo simulations is set to ; Pick any one , denoted as ; ② No. , , in the next Monte-Carlo simulation, refer to the data format and calculation method of ② and ③ in step (1) to generate the simulation The local statistics of the units to be detected without targets are recorded as , , and sort them from large to small, recorded as, ; Then for the sorted sequence , according to the first item, the sum of the first to the second item, ..., the sum of the first to the second item Sum the items, ..., 1st to 2nd The sum of the terms forms a matrix of length The new sequence is called dimensional cumulative sequence, Recorded as ,in , ; ③ Repeat step ② times, get Dimensional cumulative sequence , ;extract The first dimension in the cumulative sequence Elements of samples, arrange them from large to small, and take the The item is used as the detection threshold corresponding to the element, denoted as ; Use this operation to traverse dimensional accumulation of all elements in the sequence, we get The detection threshold is denoted as ; ④ Refer to the data format and calculation method of ② and ③ in step (1) again, Simulation generates the detection unit without target conditions indivual dimensional local statistics sequence, denoted as , ; ⑤ Extraction step ④ indivual dimensional local statistics sequence, sort it from large to small, and record it as, ; Then for the sorted sequence ,according to from To 1, calculate the variable index in sequence , , , and is always equal to 1; then use the uniformity judgment threshold obtained in step (2) ,according to from To 1, the following uniformity judgments are performed in sequence: , until the judgment is first established, at which time is the number of valid local statistics, denoted by ; Finally, construct the following judgment formula: , If the above equation holds, then the counter Add 1, otherwise the counter No increase; ⑥ Traverse step ④ indivual Dimensional local statistics sequence, execute the processing of step ⑤, and get the total count , and then get the false alarm probability of the current iteration output ; ⑦ If , then the iteration ends and the output is generated in step ③ As the second threshold, the CFAR property of the local statistics ensures that the second threshold is universal across all resolution cell grids. Otherwise, if , then reduce the , execute steps ② to ⑦ again; if , then increase the , execute steps ② to ⑦ again.

[0015] (4) Using the uniformity discrimination threshold obtained in step (2) and the detection threshold obtained in step (3), traverse all the resolution unit grids in step (1) to complete the detection of the entire observation space; the specific operations of this step are as follows: ① For distributed MIMO radar The local statistics of the resolution unit grid are obtained through step (1) , ; Then sort them from large to small, record them as, ; ② For the sorted sequence ,according to from To 1, calculate the variable index in sequence , , , and is always equal to 1; then use the uniformity judgment threshold obtained in step (2) ,according to from To 1, the following uniformity judgments are performed in sequence: , until the judgment is first established, at which time is the number of valid local statistics, denoted by ; Then the global statistics after non-coherent fusion of effective local statistics are constructed as follows, recorded as , and the detection threshold output from step (3) For comparison, , If the above formula is established, then determine There is a target in the resolution unit grid, otherwise there is no target; ③ Traverse all resolution unit grids , execute steps ① and ② to complete the detection of the entire observation space.

Claims

1. A distributed signal non-coherent fusion detection method based on variable index, characterized in that: The following steps are involved: S1. For each resolution unit grid in the unified "range-azimuth-elevation" three-dimensional observation space, the to-be-detected unit data and the reference unit data in each observation channel of the corresponding distributed MIMO radar are taken out, and then the AMF CFAR detection statistics of each observation channel corresponding to the resolution unit grid are constructed as local statistics according to the distributed signal parameters and the complex Gaussian clutter background; S2. Using the CFAR characteristics of local statistics under target-free conditions and the probability of misjudging a uniform background as a non-uniform background, a uniformity determination method based on a variable index is adopted to obtain a uniformity determination threshold for screening effective local statistics through Monte-Carlo simulation. S3, according to the false alarm probability corresponding to the detection threshold, a method combining iterative optimization and Monte-Carlo simulation is used to solve the detection threshold required for detection of the global statistics formed by non-coherent fusion of different numbers of effective local statistics; S4. Using the uniformity determination threshold obtained in step S2 and the detection threshold obtained in step S3, all the resolution unit grids in step S1 are traversed to complete the detection of the entire observation space.

2. The variable index-based distributed signal non-coherent fusion detection method according to claim 1, characterized in that: The step S1 specifically includes: S11. For distributed MIMO radar, a unified "range-azimuth-elevation" three-dimensional observation space is established, and the space is divided into grids according to the distance resolution unit size. The grid numbers are recorded as , the total is recorded as , ; S12, Distributed MIMO Radar includes observation channel, corresponding to the resolution unit grid, The data of the unit to be detected obtained in the observation channel is recorded as , the reference unit data is recorded as , , ; The reference unit data does not contain the target and the mean is , the covariance is Complex Gaussian distribution; without target, The mean is , the covariance is Complex Gaussian distribution; under target conditions, The mean is , the covariance is The complex Gaussian distribution of S13, construct the corresponding The AMF CFAR detection statistic of each observation channel of the resolution unit grid is called the local statistic and is denoted as , , ， in, Represents the sample covariance matrix, which is the covariance matrix estimates; S14, traverse all resolution unit grids to obtain each local statistic corresponding to each resolution unit grid, , , .

3. The variable index-based distributed signal non-coherent fusion detection method according to claim 1, characterized in that: The step S2 specifically includes: S21, set the probability of misjudging the uniform background as non-uniform background to 0.1; the number of Monte-Carlo simulations is set to ; Pick any one , denoted as ; S22, , , in the next Monte-Carlo simulation, refer to the data format and calculation method of S12 and S13 in step S1, and the simulation generates The local statistics of the units to be detected without targets are recorded as , ; Then sort them from large to small, record them as, ， For the sorted sequence , calculate the variable index according to the following formula , ， ,and Always equal to 1; S23, repeat step S22 times, get Variable index sequence ， ； Extract the first Elements of samples, arrange them from large to small, and take the The item is used as the uniformity judgment threshold corresponding to the element, denoted as ; Use this operation to traverse all elements in the variable index sequence and get A uniformity judgment threshold, Recorded as ; The CFAR characteristic ensures the uniformity judgment threshold Universality across all resolution unit grids.

4. The variable index-based distributed signal non-coherent fusion detection method according to claim 1, characterized in that: The step S3 specifically includes: S31, set the false alarm probability corresponding to the detection threshold to , setting the initial false alarm probability required for iterative optimization operation to ; The number of Monte-Carlo simulations is set to ; Pick any one , denoted as ; S32, , , in the next Monte-Carlo simulation, refer to the data format and calculation method of S12 and S13 in step S1, and the simulation generates The local statistics of the units to be detected without targets are recorded as , , and sort them from large to small, recorded as, ； Then for the sorted sequence , according to the first item, the sum of the first to the second item, ..., the sum of the first to the second item Sum the items, ..., 1st to 2nd The sum of the terms forms a matrix of length The new sequence is called dimensional cumulative sequence, Recorded as ,in , ; S33, repeat step S32 times, get indivual Dimensional cumulative sequence , ;extract The first dimension in the cumulative sequence Elements of samples, arrange them from large to small, and take the The item is used as the detection threshold corresponding to the element, denoted as ; Use this operation to traverse dimensional accumulation of all elements in the sequence, we get The detection threshold is denoted as ; S34, again referring to the data form and calculation method of S12 and S13 in step S1, simulate and generate the target-free condition of the unit to be detected indivual dimensional local statistics sequence, denoted as, , ; S35, extract the first indivual dimensional local statistics sequence, sort it from large to small, and record it as, ; Then for the sorted sequence ,according to from To 1, calculate the variable index in sequence , ， ， and is always equal to 1; then use the uniformity judgment threshold obtained in step S2 ,according to from To 1, the following uniformity judgments are performed in sequence: , until the judgment is first established, at which time is the number of valid local statistics, denoted by ; Finally, construct the following judgment formula, ， If the above equation holds, then the counter Add 1, otherwise the counter No increase; S36, traverse step S34 indivual Dimensional local statistics sequence, execute the processing of step S35 to obtain the total count , and then get the false alarm probability of the current iteration output ; S37, if , then the iteration ends and the output is generated in step S33 As the second threshold, the CFAR property of the local statistics ensures that the second threshold is universal in all resolution unit grids; otherwise, if , then reduce the step S31 , execute steps S32 to S37 again; if , then increase the value in step S31 , execute steps S32 to S37 again.

5. The variable index-based distributed signal non-coherent fusion detection method according to claim 1, characterized in that: The step S4 specifically includes: S41, for distributed MIMO radar The resolution unit grid is obtained through step S1. Local Statistics , ; Then sort them from large to small, record them as, ; S42. For the sorted sequence ,according to from To 1, calculate the variable index in sequence , ， ,and Always equal to 1; Then use the uniformity judgment threshold obtained in step S2 ,according to from To 1, the following uniformity judgments are performed in sequence: , until the judgment is first established, at which time is the number of valid local statistics, denoted by ; Then the global statistics after non-coherent fusion of effective local statistics are constructed as follows, recorded as and the detection threshold output from step S3 For comparison, ， If the above formula is established, then determine There is a target in the resolution unit grid, otherwise there is no target; S43, traverse all resolution unit grids , execute step S41 and step S42 to complete the detection of the entire observation space.

Images