G-LMP moving target detection method based on local maximum potential

Abstract

The invention belongs to the field of centralized MIMO (Multiple Input Multiple Output) radar signal detection, and relates to a self-adaptive detector algorithm for a weak moving target, combined with an optimized quantizer based on a PSO-SA (Particle Swarm Optimization-Sub-Association) algorithm and based on G-LMP (Gradient-Local Maximum Potential) detection of a local maximum potential algorithm under a low-bit quantization condition. The method specifically comprises the following steps: S1, setting a radar system model, setting a centralized MIMO radar, and performing multi-input and multi-output; s2, transmitting and receiving at the same time by each antenna to obtain a radar echo signal, and performing low-bit quantization processing on the received signal; s3, adopting a similar GLRT thought, and designing a G-LMP detection algorithm based on a local maximum potential; and S4, designing a quantizer based on a PSO-SA algorithm obtained by combining a particle swarm algorithm and a simulated annealing algorithm, and carrying out multiple iterations to obtain an appropriate quantization threshold value so as to improve the detection performance. Under the condition that a large amount of echo data is obtained by a centralized MIMO radar, a low-bit sampling method is introduced, the occupied bandwidth is reduced, the detector with better performance is designed, and compared with a single traditional detector, the detection performance is effectively improved while the operation cost is reduced.

Description

Technical Field

[0001] This invention belongs to the field of centralized MIMO radar target detection. Specifically, it is an adaptive detector technology for weakly moving targets under low-bit quantization conditions, which combines an optimized quantizer based on the PSO-SA algorithm and a G-LMP test based on the local maximum potential algorithm. Background Technology

[0002] In ground-based scenarios (such as intelligent traffic monitoring, border security reconnaissance, and low-altitude target early warning), weak-signal moving target detection is one of the core application requirements of radar systems. In these scenarios, targets are often affected by the superposition of ground clutter, environmental noise, and equipment noise, resulting in weak signal strength and dynamic characteristics, which significantly increases the difficulty of detection. Centralized MIMO radar, with its advantages of maximizing the utilization of spatial resources and waveform diversity technology, can achieve efficient scanning across the entire airspace and exhibits superior target detection and parameter estimation potential in low signal-to-noise ratio environments, making it the preferred technical solution for weak-signal moving target detection.

[0003] However, the multi-antenna architecture of centralized MIMO radar brings new technical challenges: On the one hand, the deployment of a large number of antenna elements causes the amount of data after echo signal sampling to grow exponentially, placing stringent demands on hardware storage capacity and computing power. In practical applications, hardware resources are often constrained by size, power consumption, and cost, making it difficult to handle high-load data processing tasks. On the other hand, there is a core contradiction in the quantization process of converting analog echo signals to digital signals—theoretically, the closer the quantization bits are to infinite quantization, the higher the signal restoration accuracy. However, this directly leads to a surge in system hardware costs and a decrease in operating efficiency, contradicting the actual needs of limited hardware resources. While low-bit quantization technology can effectively reduce the consumption of hardware resources and alleviate resource pressure, it is prone to further weakening the detectability of weak signal targets due to the loss of signal restoration accuracy.

[0004] In addition, existing radar target detection algorithms still have performance bottlenecks: most detection methods model the problem as a binary hypothesis test and rely on the accurate estimation of the noise covariance matrix to achieve target discrimination. However, in practical applications, there is often a problem of insufficient training samples, which leads to low accuracy of covariance matrix estimation. At the same time, traditional algorithms are highly dependent on training data and are difficult to adapt to the dynamic changes in noise characteristics in ground scene, which further restricts the detection performance of weak signal moving targets.

[0005] In summary, in centralized MIMO radar systems with limited hardware resources, balancing resource consumption during quantization with signal reconstruction accuracy, while overcoming the performance limitations of existing detection algorithms under conditions of low signal-to-noise ratio and insufficient samples, to achieve efficient detection of weak-signal moving targets, has become a pressing technical challenge. Based on this, a solution combining multi-bit quantization technology with the G-LMP (Local Maximum Potential Test) detection algorithm has emerged, providing a new technical path to resolve the core contradiction between hardware constraints and improved detection performance. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a G-LMP detector based on local maximum potential in low-bit scenarios, thus solving the problems mentioned in the background section.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention specifically adopts the following technical solution:

[0010] A method for improving detection performance involves multi-bit quantization of centralized MIMO radar echo data, followed by the design of a G-LMP detector based on local maximum potential using a GLRT-like approach, and finally, an optimized quantizer. The method includes the following steps:

[0011] S1, Construct a radar signal system model. Considering that the MIMO radar system uses a uniform linear array, we have... One transmitting antenna, There are 1 receiving antenna, and the target to be detected is a weak moving target. Each antenna independently receives the unknown parameters of the target. Observations were conducted. The radar wavelength was... The antenna spacing is , The angle between the far-field target and the perpendicular direction of the linear array is The transmit-receive matrix of a radar signal can be defined as:

[0012]

[0013] In the formula, The transmitted signal uses linear frequency modulation (LFM). The signals from each transmitting antenna are:

[0014]

[0015] Where j represents the imaginary part, and l represents the sequence. ;

[0016] The received signal from the radar can be represented as:

[0017]

[0018]

[0019] in, Representing the Each receiving antenna in Time series The received signal; target reflectivity For unknown deterministic parameters; The additive complex white Gaussian noise matrix satisfies ; Let be the pulse repetition interval. The formula can be simplified to the following form:

[0020]

[0021] in, The real and imaginary parts of the echo are respectively and ; and They are respectively The real and imaginary parts; ; and They are respectively The real and imaginary parts;

[0022] S2 performs low-bit processing on the echo signal to reduce bandwidth usage.

[0023] By quantizing the echo signal, a quantized signal can be obtained:

[0024]

[0025] in, This represents q-bit quantization of the signal; and Represents the real and imaginary parts of the quantized binary code.

[0026] The echo signal is transmitted to the fusion center (FC) via the receiving channel; that is, the signal transmitted from the k-th receiving channel to the FC is... The information received by the FC is represented as ;

[0027]

[0028] in,

[0029]

[0030]

[0031] This refers to the bit error rate. Let be the Hamming distance between the i-th quantized value and the r-th quantized value.

[0032] Assuming the noise is complex Gaussian white noise, the probability distribution under this condition is:

[0033]

[0034]

[0035] In the formula, .

[0036] Based on the above system model, a binary hypothesis model for object detection is constructed:

[0037]

[0038] Assuming the null hypothesis, only noise exists, and the signal reflection coefficient is 0; The assumption is that noise and signal coexist, and there is a received signal; this is the alternative hypothesis.

[0039] S3, construct the G-LMP test statistic.

[0040] In the classic local maximum potential algorithm, To avoid redundant parameters that might arise under alternative assumptions, a "GLRT-like" approach is adopted to address the parameters. The search yielded a cluster of [elements / data / data]. Then, calculate the statistic for a cluster and maximize that cluster statistic to obtain the final detection statistic. Local maximum potential can be used as the optimal criterion for one-sided asymptotic testing problems. The formula for the quantified detection statistic is:

[0041]

[0042] in, express Assuming The likelihood function value, This represents the threshold of the G-LMP detector, considering... The independence of the real and imaginary parts in the processing is addressed here by rewriting the G-LMP test statistic as a measure of the independence of the real and imaginary parts. The form of differentiating the real and imaginary parts is as follows:

[0043]

[0044] Through derivation, the explicit expression for Fisher information is:

[0045]

[0046] After deriving the formulas for the real and imaginary parts, the simplified form of the G-LMP test statistic can be obtained as follows:

[0047]

[0048] S4. An optimization quantizer is designed using the PSO-SA algorithm, which combines the Particle Swarm Optimization (PSO) algorithm and the Simulated Annealing (SA) algorithm. The core logic of the PSO-SA hybrid algorithm is to complement the advantages of both: PSO's swarm cooperative search ensures convergence efficiency, while SA's probabilistic acceptance of different solutions enhances global optimization capabilities and avoids PSO's tendency to get trapped in local optima. The goal of the hybrid algorithm is to use PSO's swarm search to quickly approach the optimal solution region, and SA's probabilistic acceptance mechanism to break the "clustering trap" of local optima, while retaining PSO's information-sharing characteristics.

[0049] The asymptotic performance expression for G-LMP is as follows:

[0050]

[0051] For the corresponding progressive PDF, This represents a chi-square distribution with 2 degrees of freedom. A non-central chi-square distribution with 2 degrees of freedom. This represents the non-centrality parameter.

[0052] During the initialization phase, PSO initialization and SA initialization are performed separately, and the parameters from both are fused. For the PSO algorithm, a particle swarm of size N is generated, where each particle contains: position... ,speed Initialize the individual optimal position of each particle. and the global optimal position Simultaneously, parameters such as inertia weight and learning factor are set. For the SA algorithm, the initial temperature is set. Cooling coefficient Termination temperature Combining both aspects, a unified objective function is used. Measure the solution.

[0053] During the iterative search phase, the following operations are performed on each particle in each iteration to balance "group cooperation" and "probability of escaping": The candidate positions and velocities of the particles are updated according to the PSO rules:

[0054]

[0055]

[0056] The acceptance criterion of SA is used to determine whether to accept a candidate solution. If the new solution is better, it is accepted directly and the particle position is updated at the same time. If the new solution is worse, it is not discarded directly, but accepted according to probability.

[0057] Update the individual optimal and global optimal values. For the newly received position... ,like Then update . Traversing all particles If it exists Then update .

[0058] After each round, the temperature is reduced according to the cooling coefficient:

[0059]

[0060] The process terminates when the maximum number of iterations is reached, the temperature drops below the termination temperature, or the globally optimal fitness shows no significant improvement for K consecutive rounds.

[0061] (III) Beneficial Effects

[0062] Compared with existing technologies, this invention provides a novel adaptive target detection algorithm that, under centralized MIMO detection of weakly moving targets, performs low-bit quantization processing on the echo signal, combines it with an optimized quantizer, and then uses the G-LMP algorithm for detection. This algorithm offers the following advantages:

[0063] This method quantizes the received signals from MIMO radar with different levels of precision, reducing bandwidth usage, improving computational efficiency, and significantly compressing data volume. Subsequently, a target detection process is performed for each signal with different quantization precisions: an optimized quantizer algorithm combining SA and PSO is used to construct an appropriate quantization threshold strategy; a "GLRT-like" algorithm, namely G-LMP, is constructed to obtain a parameter cluster regarding redundant parameters, which is then maximized to finally calculate the detection cluster. Compared to the maximum likelihood estimation (MLE) method used by conventional detectors, this method, utilizing the class concept, further improves the accuracy of parameter estimation and overall detection performance, demonstrating good application potential in the field of weak signal radar target detection. Attached Figure Description

[0064] Figure 1 This is a flowchart of a G-LMP moving target detection method based on local maximum potential according to the present invention.

[0065] Figure 2 This is a comparison chart showing the effects of using the target detection method of this invention; Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Example

[0068] like Figure 1 As shown, an embodiment of the present invention proposes a G-LMP moving target detection algorithm based on local maximum potential, which includes the following steps:

[0069] S1, Construct a radar signal system model. Considering that the MIMO radar system uses a uniform linear array, we have... One transmitting antenna, There are 1 receiving antenna, and the target to be detected is a weak moving target. Each antenna independently receives the unknown parameters of the target. Observations were conducted. The radar wavelength was... The antenna spacing is , The angle between the far-field target and the perpendicular direction of the linear array is The transmit-receive matrix of a radar signal can be defined as:

[0070]

[0071] In the formula, The transmitted signal uses linear frequency modulation (LFM). The signals from each transmitting antenna are:

[0072]

[0073] Where j represents the imaginary part, and l represents the sequence. ;

[0074] The received signal from the radar can be represented as:

[0075]

[0076]

[0077] in, Representing the Each receiving antenna in Time series The received signal; target reflectivity For unknown deterministic parameters; The additive complex white Gaussian noise matrix satisfies ; Let be the pulse repetition interval. The formula can be simplified to the following form:

[0078]

[0079] in, The real and imaginary parts of the echo are respectively and ; and They are respectively The real and imaginary parts; ; and They are respectively The real and imaginary parts;

[0080] S2 performs low-bit processing on the echo signal to reduce bandwidth usage.

[0081] By quantizing the echo signal, a quantized signal can be obtained:

[0082]

[0083] in, This represents q-bit quantization of the signal; and Represents the real and imaginary parts of the quantized binary code.

[0084] The echo signal is transmitted to the fusion center (FC) via the receiving channel; that is, the signal transmitted from the k-th receiving channel to the FC is... The information received by the FC is represented as ;

[0085]

[0086] in,

[0087]

[0088]

[0089] This refers to the bit error rate. Let be the Hamming distance between the i-th quantized value and the r-th quantized value.

[0090] Assuming the noise is complex Gaussian white noise, the probability distribution under this condition is:

[0091]

[0092]

[0093] In the formula, .

[0094] Based on the above system model, a binary hypothesis model for object detection is constructed:

[0095]

[0096] Assuming the null hypothesis, only noise exists, and the signal reflection coefficient is 0; The assumption is that noise and signal coexist, and there is a received signal; this is the alternative hypothesis.

[0097] S3, construct the G-LMP test statistic.

[0098] In the classic local maximum potential algorithm, To avoid redundant parameters that might arise under alternative assumptions, a "GLRT-like" approach is adopted to address the parameters. The search yielded a cluster of [elements / data / data]. Then, calculate the statistic for a cluster and maximize that cluster statistic to obtain the final detection statistic. Local maximum potential can be used as the optimal criterion for one-sided asymptotic testing problems. The formula for the quantified detection statistic is:

[0099]

[0100] in, express Assuming The likelihood function value, This represents the threshold of the G-LMP detector, considering... The independence of the real and imaginary parts in the processing is addressed here by rewriting the G-LMP test statistic as a measure of the independence of the real and imaginary parts. The form of differentiating the real and imaginary parts is as follows:

[0101]

[0102] Through derivation, the explicit expression for Fisher information is:

[0103]

[0104] After deriving the formulas for the real and imaginary parts, the simplified form of the G-LMP test statistic can be obtained as follows:

[0105]

[0106] S4. An optimization quantizer is designed using the PSO-SA algorithm, which combines the Particle Swarm Optimization (PSO) algorithm and the Simulated Annealing (SA) algorithm. The core logic of the PSO-SA hybrid algorithm is to complement the advantages of both: PSO's swarm cooperative search ensures convergence efficiency, while SA's probabilistic acceptance of different solutions enhances global optimization capabilities and avoids PSO's tendency to get trapped in local optima. The goal of the hybrid algorithm is to use PSO's swarm search to quickly approach the optimal solution region, and SA's probabilistic acceptance mechanism to break the "clustering trap" of local optima, while retaining PSO's information-sharing characteristics.

[0107] The asymptotic performance expression for G-LMP is as follows:

[0108]

[0109] For the corresponding progressive PDF, This represents a chi-square distribution with 2 degrees of freedom. A non-central chi-square distribution with 2 degrees of freedom. This represents the non-centrality parameter.

[0110] During the initialization phase, PSO initialization and SA initialization are performed separately, and the parameters from both are fused. For the PSO algorithm, a particle swarm of size N is generated, where each particle contains: position... ,speed Initialize the individual optimal position of each particle. and the global optimal position Simultaneously, parameters such as inertia weight and learning factor are set. For the SA algorithm, the initial temperature is set. Cooling coefficient Termination temperature Combining both aspects, a unified objective function is used. Measure the solution.

[0111] During the iterative search phase, the following operations are performed on each particle in each iteration to balance "group cooperation" and "probability of escaping": The candidate positions and velocities of the particles are updated according to the PSO rules:

[0112]

[0113]

[0114] The acceptance criterion of SA is used to determine whether to accept a candidate solution. If the new solution is better, it is accepted directly and the particle position is updated at the same time. If the new solution is worse, it is not discarded directly, but accepted according to probability.

[0115] Update the individual optimal and global optimal values. For the newly received position... ,like Then update . Traversing all particles If it exists Then update .

[0116] After each round, the temperature is reduced according to the cooling coefficient:

[0117]

[0118] The process terminates when the maximum number of iterations is reached, the temperature drops below the termination temperature, or the globally optimal fitness shows no significant improvement for K consecutive rounds.

[0119] This method relies on the parallel operation mechanism of multiple transmitters and receivers in a large-scale MIMO radar system. The transmission and reception of these signals inevitably generate massive amounts of data, leading to a significant increase in data volume. This method is primarily designed to address application scenarios with heavy data loads and difficulties in capturing weak signals. It effectively compresses the data volume by implementing low-bit quantization sampling processing of the received signal with varying precision. Simultaneously, to ensure that detection performance is not affected by quantization, a G-LMP detection strategy based on local maximum potential is adopted.

[0120] Compared to traditional schemes that rely solely on local maximum potential detectors in low-bit quantization scenarios, this method introduces a GLRT-like approach to effectively calculate the final target detection rate. This not only significantly improves the detection probability and overall detection performance but also ensures the stability of the constant false alarm rate (CFAR) attribute. Furthermore, the quantizer structure is optimized using a PSO-SA algorithm combining particle swarm optimization and simulated annealing, enabling iterative optimization of the detection threshold. Moreover, the detection performance gradually improves with increasing quantization bits. In practical applications, this method can provide strong support for the accurate detection of weak moving targets in large-scale MIMO radar, effectively improving overall detection performance.

[0121] Finally, it should be noted that the above content is only a preferred embodiment of the invention and is not intended to limit the scope of the invention. Even though the invention has been described in detail based on the foregoing embodiments, those skilled in the art still have the right to adjust the technical solutions described in the embodiments or make equivalent substitutions for some technical features. As long as it is based on the core spirit and basic principles of the invention, any form of modification, equivalent substitution, and optimization is included within the protection scope of the invention.

Claims

1. A target detection algorithm combining multi-bit quantization and G-LMP detection algorithm (i.e. G-LMP algorithm based on local maximum potential test) for weak signal moving target detection under the condition of limited hardware resources centralized MIMO radar in the ground object scene, characterized in that, Specifically comprising the following steps: S1: In the complex terrain scene, under the condition of limited hardware resources, the weak moving target detection of ultra-long distance is carried out by using centralized MIMO radar, a signal detection model is established combined with the characteristics of the scene, and it is converted into an asymptotic one-sided hypothesis test problem. S2: In view of the processing burden caused by the increase of the number of antennas and a large amount of data, the radar received echo signal is subjected to multi-bit quantization processing, and then fused and detected. S3: Since the generalized likelihood ratio detection needs to iteratively obtain the maximum likelihood estimate value, the maximum likelihood value of the above-sought parameters is a non-closed solution, so a novel G-LMP detector based on local maximum potential is used for detection, which can avoid solving the maximum likelihood value of unknown parameters, significantly reduce the calculation amount, and improve the detection probability. S4: In order to improve the detection performance, in combination with multi-bit quantization, a corresponding quantizer needs to be designed. The traditional quantizer is a uniform quantizer, and in this paper, an optimized quantizer is used, and the simulated annealing algorithm SA and the particle swarm algorithm PSO are combined to optimize the quantization threshold. Reduce the computational complexity and improve the algorithm performance.

2. The target detection algorithm combining multi-bit quantization and G-LMP detection algorithm (i.e. G-LMP algorithm based on local maximum potential test) under the condition of weak signal moving target detection of hardware resource limited centralized MIMO radar in ground object scene according to claim 1, characterized in that: The formula of the radar echo signal of the weak target signal is: where the target reflectivity is an unknown deterministic parameter; is an additive complex Gaussian white noise matrix satisfying ; , is the pulse repetition interval.

3. The target detection algorithm combining multi-bit quantization and G-LMP detection algorithm (i.e. G-LMP algorithm based on local maximum potential test) under the condition of weak signal moving target detection in the hardware resource limited centralized MIMO radar in the ground object scene according to claim 1, characterized in that: By sampling the echo signal to q bits, the obtained signal can be expressed as: where, ; , . represents the signal received by the kth receiving antenna at time t in sequence l, and represent the real and imaginary parts of the echo signal, respectively; denotes that the kth receiving channel is quantized by a q-bit quantizer, represents the quantized echo, and represent the real and imaginary parts of the quantized echo, respectively. It is necessary to determine whether the reflecting target exists, and a binary hypothesis testing problem is constructed: wherein, represents additive complex white Gaussian noise. represents the null hypothesis that only noise is present. represents the alternative hypothesis that both target and noise are present.

4. The target detection algorithm combining multi-bit quantization and G-LMP detection algorithm (i.e. G-LMP algorithm based on local maximum potential test) under the condition of weak signal moving target detection in the hardware resource limited centralized MIMO radar in the ground object scene according to claim 1, characterized in that: The "GLRT-like" approach is taken, and the test statistic is expressed as Excess parameters in the middle The parameter cluster is searched for and maximized, and the test statistic expression 5. The target detection algorithm combining multi-bit quantization and G-LMP detection algorithm (i.e. G-LMP algorithm based on local maximum potential test) under the condition of weak signal moving target detection in the hardware resource limited centralized MIMO radar in the ground object scene according to claim 1, characterized in that: Based on the simulated annealing algorithm SA and particle swarm optimization algorithm PSO algorithm combination optimization quantizer design. With PSO algorithm guarantee convergence efficiency, with SA algorithm enhance global optimization ability, avoid PSO fall into the defect of local optimal threshold Is a controllable variable, is one of the key factors affecting the detection probability, always have , .

Images