An FMCW Bio-Radar Indoor Mutual Interference Suppression Method

Abstract

The present invention belongs to the technical field of radar mutual interference suppression, and particularly relates to a method for suppressing indoor mutual interference of an FMCW bio-radar, which comprises the following steps: S1: constructing an adaptive signal subspace projection mutual interference suppression framework based on deep reinforcement learning (DRL), including a frequency modulation interference feature recognition module, a Hankel-SVD module, a signal subspace projection module, a reward calculation module, a regulation factor diagonal matrix construction module, and a deep deterministic policy gradient algorithm (DDPG) agent; S2: training a frequency modulation interference feature recognition regression neural network mapping model with a simulation data set as the frequency modulation interference feature recognition module. The present invention constructs an adaptive signal subspace projection interference suppression framework based on DRL technology. When facing the mutual interference suppression problem of non-convex optimization problems, it adapts to the dynamic changes of the problem through the rewards feedback from the environment, and solves the technical problem that the optimization process of the previous signal separation method faces NP-hard problems.

Description

Technical Field

[0001] The present invention relates to the technical field of radar mutual interference suppression, and specifically provides a method for suppressing indoor mutual interference of FMCW bio-radar. Background Art

[0002] Compared with sensors such as wearable devices and cameras, millimeter-wave bio-radars have the advantages of non-contact, less privacy infringement, all-weather and full-light sensing capabilities, and providing fine-grained sensing information. At present, they have been widely used in related fields such as smart homes and smart healthcare. With the further development of millimeter-wave bio-radar technology, the number of millimeter-wave radar devices appearing in daily life will gradually increase in the future, providing all-weather support for life. When multiple FMCW bio-radar devices work in the same indoor environment, limited available spectrum resources and the instantaneous superposition of different radar signals will inevitably cause interference, thereby weakening the extraction and recognition of target feature signals by the target perception radar. When the frequency modulation slopes of the interfering radar and the target perception radar are the same, due to the high overlap of their frequency modulation modes, false target reflections unrelated to the actual target are likely to appear at the receiving end of the perception radar, resulting in false targets in the detection results. When their frequency modulation slopes are different, the transmitted signal of the interfering radar will cover the true target echo in the frequency spectrum, causing serious deviation in the time-frequency characteristics of the target signal. Mutual interference not only reduces the accuracy and real-time performance of activity monitoring, but also makes it impossible to timely and accurately identify and warn of real fall events, thus posing a potential threat to the safety and health of the monitored person. Therefore, we propose a method for suppressing indoor mutual interference of FMCW bio-radar to solve the above problems. Summary of the Invention

[0003] (I) Technical Problems to be Solved

[0004] Aiming at the deficiencies of the prior art, the present invention provides a method for suppressing indoor mutual interference of FMCW bio-radar, which solves the problems raised in the above background art.

[0005] (II) Technical Solutions

[0006] The present invention specifically adopts the following technical solutions to achieve the above objectives:

[0007] A method for suppressing indoor mutual interference of FMCW bio-radar includes the following steps:

[0008] S1: Construct an adaptive signal subspace projection mutual interference suppression framework based on deep reinforcement learning (DRL),

[0009] including a frequency modulation interference feature recognition module, a Hankel-SVD module, a signal subspace projection module, a reward calculation module, a regulation factor diagonal matrix construction module, and a deep deterministic policy gradient algorithm (DDPG) agent;

[0010] S2: Train a regression neural network mapping model for FM interference feature recognition using a simulation data set as the FM interference feature recognition module;

[0011] S3: The regulation factor diagonal matrix construction module constructs the DRL continuous action space data into a diagonal matrix and inputs it into the signal subspace projection module;

[0012] S4: Based on the characteristic that the useful signal is continuous in the time domain and sparse in the frequency domain, construct a DRL reward calculation module with the goal of maximizing the useful signal component in the original signal;

[0013] S5: Pre-train the DRL framework with the simulation data set, and then fine-tune it with the measured data to obtain an inter-interference suppression DRL model, and use the obtained DRL model to suppress the inter-interference signal in the original signal.

[0014] Further, the simulation data set in S2 is: within the same sensing space, based on the given rectangular room parameters and the installation coordinates of the sensing radar and the interference radar, and the sensing radar and the interference radar simultaneously sense the same target, through the FMCW bi-radar mutual interference signal model in the indoor environment, the original data of the sensing radar being interfered is obtained.

[0015] Further, the FM interference feature recognition module in S2 is: construct a data set to train a regression neural network based on the Monte Carlo simulation method, and this data set uses the signal-to-interference-plus-noise ratio , the proportion of interference components , the FM slope of the interference signal emitted by the interference radar as labels, and the simulated original IF signal data of N sampling points as features.

[0016] Further, the regulation factor diagonal matrix construction module in S3 is used to construct the DRL continuous action space data into a diagonal matrix. The DRL continuous action space data is , the regulation factor is negative to represent the weakening of the corresponding subspace, and positive to represent the retention or enhancement of the corresponding subspace. The constructed regulation factor diagonal matrix is .

[0017] Further, the signal subspace projection module in S3 is used to utilize the regulation factor diagonal matrix and the eigenvalues and eigenvectors input into this module to separate the low-rank approximation matrix of the obtained useful signal as , the low-rank approximation matrix of other signals is , where represents the identity matrix.

[0018] Further, the Hankel-SVD module in S1 is used to process the discrete original data input to the system Promoted to a Hankel matrix After that, the eigenvalue matrix of the discrete original data is obtained through SVD decomposition And the eigenvectors And , and the eigenvalues and eigenvectors are input into the signal subspace projection module.

[0019] Furthermore, the specific content of using the obtained DRL model to suppress the mutual interference signal in the original signal in S5 is: the original signal, that is, the IF signal of the interfered FMCW bio-radar, is input into the mutual interference suppression DRL model after passing through a low-pass filter with a cut-off frequency matching the effective sensing distance of the radar, and a high-fidelity target useful signal is separated from the interfered signal.

[0020] (III) Beneficial effects

[0021] Compared with the prior art, the present invention provides an FMCW bio-radar indoor mutual interference suppression method, which has the following beneficial effects:

[0022] In the present invention, an adaptive signal subspace projection interference suppression framework is constructed based on DRL technology. When facing the mutual interference suppression problem of non-convex optimization problems, it adapts to the dynamic changes of the problem through the rewards feedback from the environment, and solves the technical problem that the optimization process of the previous signal separation method faces NP-hard. The pre-trained frequency modulation interference feature recognition regression neural network mapping model is used as the frequency modulation interference feature recognition module in the above interference suppression framework, which improves the DRL training efficiency while reducing the DRL observation space dimension and avoiding the risk of state space explosion. An action space for mutual interference suppression suitable for DRL is constructed, and the DRL action space data is constructed into an adaptive signal subspace projection matrix, and the target useful signal is adaptively projected from the non-orthogonal signal subspace, and the signal projection adjustment factor is adaptively adjusted in various complex indoor environments, so as to accurately separate the time-frequency features of the useful signal. Description of the drawings

[0023] Figure 1 It is a schematic flow chart of the method of the present invention;

[0024] Figure 2 It is a schematic structural diagram of the adaptive signal subspace projection mutual interference suppression framework based on deep reinforcement learning of the present invention. Specific implementation manners

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] Embodiment

[0027] As Figure 1-2 shown, a method for suppressing indoor mutual interference of an FMCW bio-radar proposed in an embodiment of the present invention includes the following steps:

[0028] S1: Construct an adaptive signal subspace projection mutual interference suppression framework based on deep reinforcement learning (DRL),

[0029] including a frequency modulation interference feature recognition module, a Hankel-SVD module, a signal subspace projection module, a reward calculation module, a regulation factor diagonal matrix construction module, and a deep deterministic policy gradient algorithm (DDPG) agent, as Figure 2 shown;

[0030] wherein the Hankel-SVD module is used to elevate the discrete original data of the input system to a Hankel matrix . Due to the particularity of the construction of the Hankel matrix, the continuity of the useful signal in the time domain can be highlighted in the matrix, and the eigenvalues and eigenvectors , obtained by SVD decomposition can effectively reflect this continuity.

[0031] S2: Train a frequency modulation interference feature recognition regression neural network mapping model with a simulation data set as the frequency modulation interference feature recognition module;

[0032] The simulation data set is: within the same sensing space, based on the given rectangular room parameters and the installation coordinates of the sensing radar and the interference radar, and the sensing radar and the interference radar simultaneously sense the same target. Through the FMCW bio-radar mutual interference signal model in the indoor environment, the original data of the sensing radar being interfered is obtained.

[0033] The mutual interference signals of two FMCW bio-radars in the indoor environment include: a direct path interference signal, a target scattering interference signal, and a target useful echo signal generated in the indoor environment constructed by simulation. The mathematical formula of the above mutual interference signal model is as follows:

[0034] The transmitted signal of the sensing radar can be expressed as:

[0035]

[0036] where is the initial phase of the signal, represents the total time , where is the fast time, represents the slow time, represents the carrier frequency, represents the frequency modulation slope.

[0037] The instantaneous distance between the target and the radar is:

[0038]

[0039] where , represents the initial distance, represents the target moving speed.

[0040] The useful signal in the perceived radar target echo can be approximately expressed as:

[0041]

[0042] The direct path interference, that is, the signal from the interfering radar transmitting antenna and arriving at the perceived radar receiving antenna, causes the time delay of the signal obtained by the perceived radar to be:

[0043]

[0044] The time delay of the signal obtained by the perceived radar caused by the target scattering interference, that is, the signal from the interfering radar transmitting antenna scattered by the target and arriving at the perceived radar receiving antenna, is:

[0045]

[0046] The time delay jointly generated by the direct path interference and the scattering interference can be expressed as:

[0047]

[0048] Define the interference occurrence time as , and the interference duration as . Therefore, the interference signal received by the perceived radar is expressed as:

[0049]

[0050] is the initial phase of the signal transmitted by the intrusion radar; the carrier frequencies of the echo signal and the interference signal are the same. The FMCW radar system receives both of them simultaneously. The IF signal obtained after mixing the interference signal with the conjugate signal of the transmitted signal can be approximately expressed as:

[0051]

[0052] Among them represents the inherent background noise of the radar. The simulated original IF signal output by the mutual interference signal model is data for each fast time dimension of

[0053] The FM interference feature recognition module is: constructing a dataset based on the Monte Carlo simulation method to train a regression neural network. This dataset uses the signal-to-interference noise ratio , the proportion of interference components , the FM slope of the interference signal as labels, and the simulated original IF signal data of N sampling points as features.

[0054] S3: The regulation factor diagonal matrix construction module constructs the DRL continuous action space data into a diagonal matrix and inputs it into the signal subspace projection module;

[0055] Among them, the regulation factor diagonal matrix construction module is used to construct the DRL continuous action space data into a diagonal matrix. The DRL continuous action space data is , the regulation factor is a negative representation for weakening the corresponding subspace, and a positive representation for retaining or enhancing the corresponding subspace. The constructed regulation factor diagonal matrix is .

[0056] The signal subspace projection module is used to utilize the regulation factor diagonal matrix and the eigenvalues and eigenvectors input into this module. The low-rank approximation matrix of the separated useful signal is , and the low-rank approximation matrix of other signals is , where represents the identity matrix.

[0057] S4: Based on the characteristics of the useful signal being continuous in time domain and sparse in frequency domain, construct a DRL reward calculation module with the goal of maximizing the useful signal component in the original signal;

[0058] Through the objective function and the constraint calculate the reward obtained by the agent in the current round, and then input the reward value into the DDPG agent.

[0059] S5: Pre-train the DRL framework with the simulation dataset, and then fine-tune it with the measured data to obtain the mutual interference suppression DRL model, and use the obtained DRL model to suppress the mutual interference signal in the original signal;

[0060] The specific content of using the obtained DRL model to suppress the mutual interference signals in the original signal is as follows: The original signal, that is, the IF signal of the interfered FMCW bio-radar, is input into the mutual interference suppression DRL model after passing through a low-pass filter with a cut-off frequency matched to the effective sensing distance of the radar, and a high-fidelity target useful signal is separated from the interfered signal.

[0061] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for suppressing indoor mutual interference of FMCW bio-radar, characterized by: The steps include: S1: Construct a DRL-based adaptive signal subspace projection mutual interference suppression framework, It includes FM interference feature recognition module, Hankel-SVD module, signal subspace projection module, reward calculation module, control factor diagonal matrix construction module, and deep deterministic policy gradient algorithm (DDPG) agent; S2: Use the simulation data set to train a FM interference feature recognition regression neural network mapping model as a FM interference feature recognition module; S3: The control factor diagonal matrix construction module constructs the DRL continuous action space data into a diagonal matrix and inputs it into the signal subspace projection module; S4: Based on the characteristics of useful signals being continuous in time domain and sparse in frequency domain, a DRL reward calculation module is constructed with the goal of maximizing the useful signal component in the original signal; S5: Use the simulation data set to pre-train the DRL framework, and then use the measured data to fine-tune it to obtain the mutual interference suppression DRL model, and use the obtained DRL model to suppress the mutual interference signal in the original signal.

2. The method for suppressing indoor mutual interference of FMCW bio-radar according to claim 1, characterized in that: The simulation data set in S2 is: in the same perception space, based on the given rectangular room parameters and the installation coordinates of the perception radar and the interference radar, and the perception radar and the interference radar perceive the same target at the same time, and obtain the original data of the perception radar being interfered through the FMCW bio-radar mutual interference signal model in the indoor environment.

3. The method for suppressing indoor mutual interference of FMCW bio-radar according to claim 1, characterized in that: The FM interference feature recognition module in S2 is: based on the Monte Carlo simulation method, a data set is constructed to train a regression neural network. The data set is based on the signal interference noise ratio , the proportion of interference components , the frequency modulation slope of the jamming signal emitted by the jamming radar The label is the label, and the analog original IF signal data of N sampling points is the feature.

4. The method for suppressing indoor mutual interference of FMCW bio-radar according to claim 1, characterized in that: The control factor diagonal matrix construction module in S3 is used to construct the DRL continuous action space data into a diagonal matrix. The DRL continuous action space data is , the negative regulation factor represents the weakening of the corresponding subspace, and the positive regulation factor represents the retention or enhancement of the corresponding subspace. The constructed diagonal matrix of regulation factors is .

5. The method for suppressing indoor mutual interference of FMCW bio-radar according to claim 4, characterized in that: The signal subspace projection module in S3 is used to separate the low-rank approximate matrix of the useful signal by using the diagonal matrix of the control factor and the eigenvalues ​​and eigenvectors input into the module: , the low-rank approximation matrix of other signals is ,in Represents the identity matrix.

6. The method for suppressing indoor mutual interference of FMCW bio-radar according to claim 1, characterized in that: The Hankel-SVD module in S1 is used to upgrade the discrete original data of the input system to a Hankel matrix, obtain the eigenvalues ​​and eigenvectors of the discrete original data through SVD decomposition, and input the eigenvalues ​​and eigenvectors into the signal subspace projection module.

7. The method for suppressing indoor mutual interference of FMCW bio-radar according to claim 1, characterized in that: The specific content of using the obtained DRL model to suppress the mutual interference signal in the original signal in S5 is: the original signal, that is, the IF signal of the interfered FMCW bioradar, is input into the mutual interference suppression DRL model after passing through a low-pass filter with a cutoff frequency matching the effective perception distance of the radar, and a high-fidelity target useful signal is separated from the interfered signal.

Images