A lasso-based vehicle millimeter wave radar interference signal elimination system and method, and a target detection system and method
By using sparse representation based on fusion lasso and generalized total variational techniques, interference signals from automotive millimeter-wave radar are suppressed in the time domain, solving the problems of high computational load and target signal loss, and achieving efficient interference suppression and target signal recovery.
Patent Information
- Application Number
- CN202310041003.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-01-12
- Publication Date
- 2026-02-24
- Estimated Expiration
- 2043-01-12
AI Technical Summary
Existing technologies suffer from high computational demands and performance limitations when suppressing interference signals from automotive millimeter-wave radar. They also tend to cause energy loss of the target signal and residual interference signals, making it impossible to effectively recover the target signal.
By employing sparse representation based on fusion lasso and generalized total variational technique, interference signals are extracted and eliminated in the time domain. Through sparse optimization model and non-convex norm processing, combined with target signal enhancement unit, white noise and zero Doppler interference are eliminated, and the target signal is recovered.
It can efficiently suppress interference signals in the time domain, save computation, improve real-time performance, avoid target signal loss, enhance target signal recovery, and improve detection accuracy.
Smart Images
Figure CN116047426B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of interference signal elimination of radar, and particularly relates to a vehicle millimeter wave radar interference signal elimination system and method based on fusion lasso, and a target detection system and method. BACKGROUND
[0002] With the development of unmanned driving technology, the automatic driving vehicle has an advanced driver-assistance system (ADAS), which realizes the lane change assistance, blind spot detection, adaptive cruise control and collision warning auxiliary functions, and has important significance for reducing driver errors and improving traffic safety.
[0003] The millimeter wave radar has all-weather characteristics and is of great significance for unmanned driving. The use of the same frequency spectrum by vehicle millimeter wave radars will cause mutual interference. According to the standard of the International Telecommunication Union, vehicle millimeter wave radars on the traffic road will share the same frequency spectrum, and a general automatic driving vehicle will be equipped with multiple millimeter wave radars around the vehicle. Therefore, with the popularization of automatic driving, the number of vehicle millimeter wave radars on the traffic road is increasing, and the mutual interference between millimeter wave radars will become more serious, the probability of mutual interference will increase, and the average power of mutual interference will increase, resulting in a decrease in the detection accuracy of the vehicle millimeter wave radar. In severe cases, the interference will produce false targets in the frequency spectrum of the vehicle millimeter wave radar, and ultimately lead to misjudgment of the vehicle millimeter wave radar.
[0004] The existing method based on morphological component analysis suppresses the interference signal in the frequency domain and the wavelet transform domain, and needs additional iterations, so the calculation amount is large. Moreover, the performance of the suppression of the interference signal in the frequency domain and the wavelet transform domain is limited by the resolution of the Fourier transform and the wavelet transform, so the performance is poor. In addition, the method based on morphological component analysis for suppressing interference in the frequency domain and the wavelet transform domain needs to decompose and reconstruct the signal, at this time, the energy distribution problem of the interference signal and the target signal is involved, and there is no effective morphological component analysis energy distribution algorithm at present, so the target signal energy loss and interference signal residue are easy to occur. In addition, this type of method has many parameters, and the interference signal suppression performance is quite different between different signals, so the fixed parameter setting cannot meet the interference signal elimination of all signals, and therefore the performance of this type of method is poor.
[0005] The existing algorithm for extracting interference in the time domain based on the characteristics of the interference signal directly cuts off the interference signal together with the target signal, and then restores the cut-off part by sparse or prediction method. However, the cut-off target signal contains important target detection information, and the restored part is only false information, so the correctness of this method is still questionable.
[0006] Existing methods based on traditional filters rely on prior knowledge of the interference signal. Current methods for obtaining prior knowledge have poor performance, resulting in residual interference signals. Summary of the Invention
[0007] To address the problems existing in the prior art, the present invention aims to provide a system and method for eliminating interference signals from automotive millimeter-wave radar and a target detection system and method based on fusion lasso, which can effectively suppress automotive millimeter-wave radar interference and false targets, and effectively recover the target signal.
[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0009] A pulse-type interference signal cancellation system for automotive millimeter-wave radar based on fusion lasso, the system comprising an echo signal modeling unit, an interference signal extraction and cancellation unit, and a target signal enhancement unit;
[0010] The echo signal modeling unit models the radar echo signal y:
[0011] y = (s t +s i )+n
[0012] Where, s t It is the target signal, s i This represents the interference signal, where n represents the noise in the radar receiver.
[0013] The interference signal extraction and cancellation unit extracts and cancels pulse-type interference from the radar signal time domain to obtain the target signal;
[0014] First, based on the fused lasso sparse representation of the pulse signal, a sparse optimization model is established to extract the interference signal from the echo signal y:
[0015]
[0016] in, λ0 and ||·||1 represent the L2 norm and L1 norm, respectively, and λ0, λ1>0 are regularization parameters. These regularization parameters affect the convergence rate of the model. The two norms ||s i ||1 and||D0s i ||1's joint representation of the fusion lasso, ||D0s i ||1 represents the total variational norm, and the matrix This represents a first-order difference operator;
[0017] Then, the underestimation problem of the interference signal is solved by using a non-convex norm: the non-convex norm is defined as Φ(x; a), where a>0, and the parameter a determines the non-convexity of the penalty function. The model for extracting and eliminating the interference signal is expressed as:
[0018]
[0019] Target signal enhancement unit, used to enhance the target signal s t Enhancement processing is performed to eliminate white noise and zero Doppler.
[0020] The processing of the target signal enhancement unit is as follows:
[0021] The target signal enhancement model based on the generalized total variation is as follows:
[0022]
[0023] Among them, y p It is the echo signal after interference extraction and cancellation, λ2>0 is the regularization parameter, and the matrix is... It is a second-order difference operator. The expression follows the rule v = Sr, where v is the property of y. p The second derivative, operator S, transforms v into the sum of overlapping signals, and matrix r is a sparse representation of signal v, B n (r)=[r 2n ,r 2n+1 ] T and Parameters ρ1 and ρ2 are respectively Δ slope and Δ value The weight of Δ slope and Δ value The characteristics of the obtained target signal are determined, where Δ slope =v n +v n+1 Δ value =(v n -v n+1 The higher ρ1 / ρ2 induces higher discreteness of the target signal.
[0024] A vehicle millimeter-wave radar target detection system based on fusion lasso includes, as described above, a vehicle millimeter-wave radar pulse interference signal cancellation system and a target detection unit.
[0025] The target detection unit is used to perform fast time-dimensional target distance detection and slow time-dimensional coherent accumulation target velocity detection on the signal processed by the interference signal cancellation system, so as to obtain target distance and velocity information.
[0026] A method for canceling pulse interference signals from automotive millimeter-wave radar based on fusion lasso, the method being implemented using the system described above, includes the following steps:
[0027] Step 1: The data collected by the radar is input into the interference suppression system;
[0028] Step 2: The interference suppression system extracts and eliminates pulse-type interference from the radar signal in the time domain through the interference signal extraction and cancellation unit to obtain the target signal;
[0029] First, the radar echo signal y is modeled as:
[0030] y = (s t +s i )+n
[0031] Where, s t It is the target signal, s i This represents the interference signal, where n represents the noise in the radar receiver.
[0032] Then, based on the fused lasso sparse representation of the pulse signal, the following sparse optimization model is established to extract the interference signal from the echo signal:
[0033]
[0034] in, λ0 and ||·||1 represent the L2 norm and L1 norm, respectively, and λ0, λ1>0 are regularization parameters. These regularization parameters affect the convergence rate of the model. The two norms ||s i ||1 and||D0s i ||1's joint representation of the fusion lasso, ||D0s i ||1 represents the total variational norm, and the matrix This represents a first-order difference operator;
[0035] The underestimation problem of radar signals is solved by using a non-convex norm: the non-convex norm is defined as Φ(x; a), where a>0, and the parameter a determines the non-convexity of the penalty function. The model for extracting and eliminating interference signals is expressed as follows:
[0036]
[0037] Step 3: Enhance the target signal using the target signal enhancement unit, while eliminating white noise and zero Doppler interference.
[0038] Step 3 is described in detail below:
[0039] The target signal enhancement model based on the generalized total variation is as follows:
[0040]
[0041] Among them, y p It is the echo signal after interference extraction and cancellation, λ2>0 is the regularization parameter, and the matrix is... It is a second-order difference operator. The expression follows the rule v = Sr, where v is the property of y. p The second derivative, operator S, transforms v into the sum of overlapping signals, and matrix r is a sparse representation of signal v, B n (r)=[r 2n ,r 2n+1 ] T and Parameters ρ1 and ρ2 are respectively Δ slope and Δ value The weight of Δ slope and Δ value The characteristics of the obtained target signal are determined, where Δ slope =v n +v n+1 Δ value =v n -v n+1 / 2, a higher ρ1 / ρ2 induces higher discreteness of the target signal.
[0042] A target detection method for automotive millimeter-wave radar based on fusion lasso, comprising the pulse interference signal cancellation method for automotive millimeter-wave radar as described above; the target detection method further includes...
[0043] The enhanced signal is subjected to fast time dimension target distance detection and slow time dimension coherent accumulation target velocity detection to obtain target distance and velocity information.
[0044] By adopting the above scheme, this invention suppresses radar interference in the time domain, solving the problem of large additional computational load caused by frequency domain and wavelet transform analysis methods. Therefore, this technology suppresses interference in the time domain, saving a significant amount of computation and improving the real-time performance of automotive millimeter-wave radar interference signal suppression. This technology is not a traditional method based on morphological component analysis, so there is no need to consider the energy distribution problem between the target signal and the interference signal. Suppressing radar interference in the time domain also avoids the interference suppression performance loss caused by the resolution of Fourier transform and wavelet transform analysis methods, thus achieving better performance. Moreover, this invention suppresses radar interference in the time domain without directly cutting off the part containing the interference signal, but extracts the interference signal through sparse optimization, so it does not cause loss of the target signal.
[0045] Furthermore, the present invention enhances the target signal through a target signal enhancement unit, while removing zero Doppler interference and white noise, thus effectively recovering the target signal. Attached Figure Description
[0046] Figure 1 This is a flowchart of the method of the present invention;
[0047] Figure 2 This is a comparative diagram showing the interference signal before and after elimination in an embodiment of the present invention. Detailed Implementation
[0048] This invention provides a pulse interference signal cancellation system for automotive millimeter-wave radar based on fusion lasso, which includes an echo signal modeling unit, an interference signal extraction and cancellation unit, and a target signal enhancement unit.
[0049] The received radar echo signal is input into this system for interference cancellation. Specifically, the echo signal modeling unit models the radar echo signal y:
[0050] y = (s t +s i )+n
[0051] Among them, s t It is the target signal, s i This represents the interference signal, where n represents the noise in the radar receiver.
[0052] The interference signal extraction and cancellation unit extracts and cancels pulse-type interference from the radar signal time domain to obtain the target signal.
[0053] First, based on the fused lasso sparse representation of the pulse signal, a sparse optimization model is established to extract the interference signal from the echo signal y:
[0054]
[0055] in, λ0 and ||·||1 represent the L2 norm and L1 norm, respectively, and λ0, λ1>0 are regularization parameters. These regularization parameters affect the convergence rate of the model. The two norms ||s i ||1 and||D0s i ||1's joint representation of the fusion lasso, ||D0s i ||1 represents the total variational norm, and the matrix This represents a first-order difference operator.
[0056] Then, the underestimation problem of the interference signal is solved by using a non-convex norm: the non-convex norm is defined as Φ(x; a), where a>0. The parameter a determines the non-convexity of the penalty function. A higher parameter a yields a penalty function with higher non-convexity. The model for extracting and eliminating the interference signal is expressed as:
[0057]
[0058] Target signal enhancement unit, used to enhance the target signal s t Enhancement processing is performed to eliminate white noise and zero Doppler.
[0059] Specifically, the target signal enhancement model is established based on the generalized total variation (GTV) as follows:
[0060]
[0061] Among them, y p It is the echo signal after interference extraction and cancellation, λ2>0 is the regularization parameter, and the matrix is... It is a second-order difference operator. The expression follows the rule v = Sr, where v is the property of y. p The second derivative, operator S, transforms v into the sum of overlapping signals, and matrix r is a sparse representation of signal v, B n (r)=[r 2n ,r 2n+1 ] T and Parameters ρ1 and ρ2 are respectively Δ slope and Δ value The weight of Δ slope and Δ value The characteristics of the obtained target signal are determined, where Δ slope =v n +v n+1 Δ value =(v n -v n+1 The higher ρ1 / ρ2 induces higher discreteness of the target signal.
[0062] By adding a target detection unit to the aforementioned automotive millimeter-wave radar pulse interference signal cancellation system, an automotive millimeter-wave radar target detection system based on fusion lasso can be formed.
[0063] The target detection unit performs fast time-dimensional target range detection and slow time-dimensional coherent accumulation target velocity detection on the signal processed by the interference cancellation system. Fast time-dimensional target range detection involves performing a Fast Fourier Transform on the enhanced target signal to obtain the target's range information in a one-dimensional spectrum. Subsequently, in the slow time dimension, multiple enhanced target signals are coherently accumulated to obtain the target's range and velocity information in a two-dimensional range-Doppler spectrum.
[0064] like Figure 1 As shown, based on the same inventive concept, this invention also provides a method for canceling pulse interference signals of automotive millimeter-wave radar based on fusion lasso, specifically including the following steps:
[0065] Step 1: The data collected by the radar is input into the interference suppression system;
[0066] Step 2: The interference suppression system extracts and eliminates pulse-type interference from the radar signal time domain through the interference signal extraction and elimination unit to obtain the target signal.
[0067] Specifically, the steps include the following:
[0068] The radar echo signal y is modeled as:
[0069] y = (s t +s i )+n
[0070] Among them, s t It is the target signal, s i This represents the interference signal, where n represents the noise in the radar receiver.
[0071] Based on the fused lasso sparse representation of the pulse signal, the following sparse optimization model is established to extract the interference signal from the echo signal:
[0072]
[0073] in, λ0 and ||·||1 represent the L2 norm and L1 norm, respectively, and λ0, λ1>0 are regularization parameters. These regularization parameters affect the convergence rate of the model. The two norms ||s i ||1 and||D0s i ||1's joint representation of the fusion lasso, ||D0s i ||1 represents the total variational norm, and the matrix This represents a first-order difference operator;
[0074] The underestimation problem of radar signals is solved by using a non-convex norm: the non-convex norm is defined as Φ(x; a), where a>0. The parameter a determines the non-convexity of the penalty function. A higher parameter a yields a penalty function with higher non-convexity. The model for extracting and eliminating interference signals is expressed as follows:
[0075]
[0076] Step 3: Enhance the target signal using the target signal enhancement unit, while eliminating white noise and zero Doppler interference. This includes the following steps:
[0077] The target signal enhancement model is established based on the generalized total variation (GTV) as follows:
[0078]
[0079] Among them, y p It is the echo signal after interference extraction and cancellation, λ2>0 is the regularization parameter, and the matrix is... It is a second-order difference operator. The expression follows the rule v = Sr, where v is the property of y. p The second derivative, operator S, transforms v into the sum of overlapping signals, and matrix r is a sparse representation of signal v, B n (r)=[r 2n ,r 2n+1 ] T and Parameters ρ1 and ρ2 are respectively Δ slope and Δ value The weight of Δ slope and Δ value The characteristics of the obtained target signal are determined, where Δ slope =v n +v n+1 Δ value =(v n -v n+1 The higher ρ1 / ρ2 induces higher discreteness of the target signal.
[0080] The signal enhancement unit described above is an embodiment of the present invention, and can also be implemented based on signal denoising and signal enhancement techniques other than total variation.
[0081] Adding step 4 to the above method will allow you to complete the target signal detection.
[0082] Step 4: Perform fast time dimension target distance detection and slow time dimension coherent accumulation target velocity detection on the enhanced signal to obtain target distance and velocity information.
[0083] To fully explain the content of this invention and the effects it achieves, a specific embodiment will be described in detail below.
[0084] This embodiment uses raw data from an interference experiment collected by two automotive millimeter-wave radars operating at 77 GHz. The main radar's scan period is 29.56 μs, and its frequency modulation parameters are 29.31 × 10⁻⁶. 12 The interference source has a scan period of 72.3 μs and a frequency modulation parameter of 9.99 × 10 Hz / s. 12 Hz / s. To generate a stronger interference signal, the interference source was placed in front of the main radar to simulate a scenario of vehicles encountering each other on a real traffic road.
[0085] Comparison of automotive millimeter-wave radar interference suppression before and after the present invention Figure 2 As shown. Figure 2 (a) The interference signal from automotive millimeter-wave radar exists in the time domain as pulses, with the smaller amplitude signal representing the target signal. Due to the presence of the interference signal, Figure 2 (b) The target in the spectrum is submerged in interference. Figure 2 (c) The target signal after interference suppression of the radar signal using this technique. At this point, the target signal is enhanced, and interference signals, zero-Doppler interference, and white noise are eliminated. Therefore, in Figure 2 (d) The target spikes visible in the spectrum have been recovered. Before interference cancellation, the target spikes were... Figure 2 The coherent accumulation result in (e) is submerged, indicating strong interference and zero Doppler interference. This technique is used to suppress the interference, and the target peak is visible in... Figure 2 (f) is restored, and zero Doppler interference and white noise are eliminated.
[0086] In summary, this invention suppresses radar interference in the time domain, solving the problem of excessive computational burden caused by frequency domain and wavelet transform analysis methods. Therefore, this technique suppresses interference in the time domain, saving significant computational resources and improving the real-time performance of automotive millimeter-wave radar interference signal suppression. This technique is not based on traditional morphological component analysis methods, thus eliminating the need to consider the energy distribution between the target signal and the interference signal. Suppressing radar interference in the time domain also avoids the performance loss caused by the resolution of Fourier transform and wavelet transform analysis methods, resulting in better performance. Furthermore, this invention suppresses radar interference in the time domain without directly cutting off the portion containing the interference signal; instead, it extracts the interference signal through sparse optimization, thus avoiding the loss of the target signal.
[0087] Furthermore, the present invention enhances the target signal through a target signal enhancement unit, while removing zero Doppler interference and white noise, thus effectively recovering the target signal.
[0088] The above description is merely an embodiment of the present invention and does not constitute any limitation on the technical scope of the present invention. Therefore, any minor modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the scope of the technical solution of the present invention.
Claims
1. A pulse interference signal cancellation system for automotive millimeter-wave radar based on fusion lasso, characterized in that: The system includes an echo signal modeling unit, an interference signal extraction and cancellation unit, and a target signal enhancement unit; The echo signal modeling unit models the radar echo signal y: y=(s t +s i )+n Where, s t It is the target signal, s i This represents the interference signal, where n represents the noise in the radar receiver. The interference signal extraction and cancellation unit extracts and cancels pulse-type interference from the radar signal time domain to obtain the target signal; First, based on the fused lasso sparse representation of the pulse signal, a sparse optimization model is established to extract the interference signal from the echo signal y: in, λ0 and ||·||1 represent the L2 norm and L1 norm, respectively, and λ0, λ1>0 are regularization parameters. These regularization parameters affect the convergence rate of the model. The two norms ||s i ||1 and||D0s i ||1's joint representation of the fusion lasso, ||D0s i ||1 represents the total variational norm, and the matrix This represents a first-order difference operator; Then, the underestimation problem of the interference signal is solved by using a non-convex norm: the non-convex norm is defined as Φ(x; a), where a>0, and the parameter a determines the non-convexity of the penalty function. The model for extracting and eliminating the interference signal is expressed as: Target signal enhancement unit, used to enhance the target signal s t Enhancement processing is performed to eliminate white noise and zero Doppler.
2. The automotive millimeter-wave radar pulse interference signal cancellation system based on fusion lasso as described in claim 1, characterized in that: The processing of the target signal enhancement unit is as follows: The target signal enhancement model based on the generalized total variation is as follows: Among them, y p It is the echo signal after interference extraction and cancellation, λ2>0 is the regularization parameter, and the matrix is... It is a second-order difference operator. The expression follows the rule v = Sr, where v is the property of y. p The second derivative, operator S, transforms v into the sum of overlapping signals, and matrix r is a sparse representation of signal v, B n (r)=[r 2n ,r 2n+1 ] T and Parameters ρ1 and ρ2 are respectively Δ slope and Δ value The weight of Δ slope and Δ value The characteristics of the obtained target signal are determined, where Δ slope =v n +v n+1 Δ value =(v n -v n+1 The higher ρ1 / ρ2 induces higher discreteness of the target signal.
3. A vehicle millimeter-wave radar target detection system based on fusion lasso, characterized in that: The system includes the automotive millimeter-wave radar pulse interference signal cancellation system as described in claim 1 or 2, and a target detection unit. The target detection unit is used to perform fast time-dimensional target distance detection and slow time-dimensional coherent accumulation target velocity detection on the signal processed by the interference signal cancellation system, so as to obtain target distance and velocity information.
4. A method for canceling pulse interference signals of automotive millimeter-wave radar based on fusion lasso, characterized in that: The method is implemented using the system described in claim 1, and includes the following steps: Step 1: The data collected by the radar is input into the interference suppression system; Step 2: The interference suppression system extracts and eliminates pulse-type interference from the radar signal in the time domain through the interference signal extraction and cancellation unit to obtain the target signal; First, the radar echo signal y is modeled as: y=(s t +s i )+n Where, s t It is the target signal, s i This represents the interference signal, where n represents the noise in the radar receiver. Then, based on the fused lasso sparse representation of the pulse signal, the following sparse optimization model is established to extract the interference signal from the echo signal: in, λ0 and ||·||1 represent the L2 norm and L1 norm, respectively, and λ0, λ1>0 are regularization parameters. These regularization parameters affect the convergence rate of the model. The two norms ||s i ||1 and||D0s i ||1's joint representation of the fusion lasso, ||D0s i ||1 represents the total variational norm, and the matrix This represents a first-order difference operator; The underestimation problem of radar signals is solved by using a non-convex norm: the non-convex norm is defined as Φ(x; a), where a>0, and the parameter a determines the non-convexity of the penalty function. The model for extracting and eliminating interference signals is expressed as follows: Step 3: Enhance the target signal using the target signal enhancement unit, while eliminating white noise and zero Doppler interference.
5. The method for canceling pulse interference signals of automotive millimeter-wave radar based on fusion lasso according to claim 4, characterized in that: Step 3 is described in detail below: The target signal enhancement model based on the generalized total variation is as follows: Among them, y p It is the echo signal after interference extraction and cancellation, λ2>0 is the regularization parameter, and the matrix is... It is a second-order difference operator. The expression follows the rule v = Sr, where v is the property of y. p The second derivative, operator S, transforms v into the sum of overlapping signals, and matrix r is a sparse representation of signal v, B n (r)=[r 2n ,r 2n+1 ] T and Parameters ρ1 and ρ2 are respectively Δ slope and Δ value The weight of Δ slope and Δ value The characteristics of the obtained target signal are determined, where Δ slope =v n +v n+1 Δ value =(v n -v n+1 The higher ρ1 / ρ2 induces higher discreteness of the target signal.
6. A vehicle millimeter-wave radar target detection method based on fusion lasso, characterized in that: The target detection method includes the vehicle millimeter-wave radar pulse interference signal cancellation method as described in claim 4 or 5, and the target detection method further includes... The enhanced signal is subjected to fast time dimension target distance detection and slow time dimension coherent accumulation target velocity detection to obtain target distance and velocity information.