A method for estimating RV two-dimensional spectral floor noise based on moving average

Abstract

This invention relates to a method for estimating the noise floor of a two-dimensional R-V spectrum based on moving averages. This method is used to process radar echo signals to obtain the noise floor. The radar echo signal is a three-dimensional spectrum, with its three dimensions corresponding to the range index R, the Doppler velocity index V, and the signal amplitude Am under the (R,V) index. The estimation method involves traversing all (R,V) values ​​of the radar echo signal and calculating the noise floor using a two-dimensional moving average method during the traversal. This invention uses a cross-shaped two-dimensional moving average method to estimate the noise floor, increasing the noise floor calculation sample from one-dimensional to two-dimensional. The calculated noise floor is more stable and closer to the true noise floor of the R-V spectrum.

Description

Technical Field

[0001] This invention relates to the field of automotive radar target detection technology, and specifically to a method for estimating RV two-dimensional spectral noise floor based on moving average. Background Technology

[0002] In automotive radar target detection, the RV spectrum (i.e., range-Doppler velocity two-dimensional FFT output spectrum) is often used to detect targets. Different points on the RV spectrum correspond to different ranges and velocities. When a strongly reflective target with a range of r and a velocity of v exists within the radar's detection range, the power at the corresponding position on its RV spectrum will increase. By detecting and identifying the index positions of these peak targets with power higher than the ambient power in the R and V coordinates, the radar ranging and velocity measurement results can be determined. A crucial step in correctly detecting these peak targets with power higher than the ambient power is calculating the ambient power (i.e., noise floor) corresponding to that peak target.

[0003] like Figure 1 As shown, existing methods for calculating the noise floor of the RV two-dimensional spectrum based on moving averages often employ the cross-moving average method, which involves performing one-dimensional moving averages on the distance and velocity dimensions near the peak points of the RV two-dimensional spectrum. Since this method simply combines the two-dimensional moving average noise floor calculation algorithm into a 1+1 approach, some shortcomings of the original one-dimensional moving average noise floor calculation remain. For example, when there are continuous peaks in the distance / velocity dimension, the calculated noise floor will be too high; when the environment changes rapidly, the calculated noise floor will fluctuate wildly. These scenarios are ubiquitous in automotive radar applications, such as parking lots, congested roads, and areas with guardrails and bushes. Therefore, the noise floor estimation obtained from the RV two-dimensional spectrum in current automotive radar systems has poor accuracy and stability. Summary of the Invention

[0004] To address the problems existing in the prior art, the purpose of this invention is to provide an estimation method for RV two-dimensional spectral noise based on moving average, so as to obtain accurate and stable noise floor.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A two-dimensional spectral noise floor estimation method based on moving average is used to process radar echo signals to obtain the noise floor. The radar echo signal is a three-dimensional spectrum, whose three dimensions correspond to the range dimension index R, the Doppler velocity dimension index V, and the signal amplitude Am under the (R,V) index, respectively. The estimation method is as follows: traverse all (R,V) of the radar echo signal, and calculate the noise floor by using the two-dimensional moving average method during the traversal process.

[0007] Specifically, for a given D that needs to be traversed, its current index is (Rd, Vd), denoted as D(Rd, Vd); when traversing to D(Rd, Vd), the following processing is performed:

[0008] Skip the guard interval Px of the distance dimension, take the D(Rd, Vd) distance dimension noise floor sample point set (Rd-Px-N: Rd-Px, Vd-m: Vd+m), (Rd+Px: Rd+Px+N, Vd-m: Vd+m), calculate their average values, and then obtain the reference noise floor NFr(Rd, Vd) of the distance dimension through the decision maker; where N is the length of the distance dimension sampling window, and m is half the width of the distance dimension sampling window;

[0009] Skip the guard interval Pv of the velocity dimension, take the velocity dimension noise floor sample point set D(Rd, Vd) (Rd-n: Rd+n, Vd-Pv-M: Vd-Pv), (Rd-n: Rd+n, Vd+Pv: Vd+Pv+M), calculate their respective average values ​​AVGv1 and AVGv2, and then obtain the velocity dimension reference noise floor NFv(Rd, Vd) through the decision maker; where M is the velocity dimension sampling window length and n is half of the velocity dimension sampling window width;

[0010] The noise floor obtained after traversing all (R, V) is the RV two-dimensional spectral noise floor.

[0011] During the traversal, each slide can only change one smallest unit, and two dimensions cannot be changed simultaneously.

[0012] During the traversal, the system slides row by row or column by column. The average value after each slide is the average value before the slide minus the average signal amplitude of the sample block that was slid out, plus the average signal amplitude of the sample block that was slid in. The reference noise floor after the slide is calculated in this way.

[0013] During the traversal, the index is slid along the distance dimension, becoming (Rd+1, Vd). The average distance dimension value AVGr1' corresponding to this index (Rd+1, Vd) is: AVGr1+(Am(Rd-Px+1, Vd-m: Vd+m)-Am(Rd-Px-N, Vd-m: Vd+m)) / (N*2m), and the average distance dimension value AVGr2' is: AVGr2+(Am(Rd+Px+N+1, Vd-m: Vd+m)-Am(Rd+Px, Vd-m: Vd+m)) / (N*2m). Based on the average distance dimension values ​​AVGr1' and AVGr2', the reference noise floor NFr(Rd+1, Vd) of the distance dimension is calculated.

[0014] The velocity dimension average value AVGv1' corresponding to the index (Rd+1, Vd) is: AVGv1+(Am(Rd+n+1, Vd-Pv-M: Vd-Pv)-Am(Rd-n, Vd-Pv-M: Vd-Pv)) / (M*2n), and the velocity dimension average value AVGv2' is: AVGv2+(Am(Rd+n+1, Vd+Pv: Vd+Pv+M)-Am(Rd-n, Vd+Pv: Vd+Pv+M)) / (M*2n); the reference noise floor NFv(Rd+1, Vd) of the velocity dimension is calculated based on the velocity dimension average values ​​AVGv1' and AVGv2'.

[0015] During the traversal, the index is slid along the velocity dimension, becoming (Rd, Vd+1). The average distance dimension value AVGr1' corresponding to this index (Rd, Vd+1) is: AVGr1+(Am(Rd-Px-N:Pd-Px, Vd+m+1)-Am(Rd-Px-N:Pd-Px, Vd-m)) / (N*2m), and the average distance dimension value AVGr2' is: AVGr2+(Am(Rd+Px:Rd+Px+N, Vd+m+1)-Am(Rd+Px:Rd+Px+N, Vd-m)) / (N*2m). The reference noise floor NFr(Rd, Vd+1) for the distance dimension is calculated based on the average distance dimension values ​​AVGr1' and AVGr2.

[0016] The velocity dimension average value AVGv1' corresponding to the index (Rd, Vd+1) is: AVGv1+(Am(Rd-n:Rd+n,Vd-Pv+1)-Am(Rd-n:Rd+n,Vd-Pv-M)) / (M*2n), and the velocity dimension average value AVGv2' is: AVGv2+(Am(Rd-n:Rd+n,Vd+Pv+M+1)-Am(Rd-n:Rd+n,Vd+Pv)) / (M*2n); the velocity dimension reference noise floor NFv(Rd, Vd+1) is calculated based on the velocity dimension average values ​​AVGv1' and AVGv2'.

[0017] After adopting the above scheme, the present invention estimates the noise floor based on the cross-shaped two-dimensional moving average method, which improves the noise floor calculation sample from one-dimensional to two-dimensional, making the calculated noise floor more stable and closer to the true noise floor of the RV spectrum.

[0018] In addition, the present invention combines a specific sliding sequence, and the data calculation after the sliding is based on the data of the previous sliding. That is, the average value after the sliding is the average value before the sliding minus the average signal amplitude of the sample area block that slid out plus the average signal amplitude of the sample area block that slid in, thereby reducing the calculation time of the average value of the two-dimensional surface sampling data during the sliding process, realizing fast sliding and improving the noise floor estimation speed. Attached Figure Description

[0019] Figure 1 This is a block diagram illustrating the principle of the existing method;

[0020] Figure 2 This is a block diagram illustrating the principle of the method of the present invention;

[0021] Figure 3 This is the first RV two-dimensional spectrum sliding mode of the present invention;

[0022] Figure 4 This is the second RV two-dimensional spectrum sliding mode of the present invention. Detailed Implementation

[0023] This invention discloses a method for estimating the RV two-dimensional spectral noise floor based on moving average, which is used to process radar echo signals to obtain the noise floor. The radar echo signal is a three-dimensional spectrum, with its three dimensions corresponding to the range index R, the Doppler velocity index V, and the signal amplitude Am under the (R,V) index. The estimation method involves traversing all (R,V) values ​​of the radar echo signal and calculating the noise floor using a two-dimensional moving average method during the traversal. Each (R,V) value corresponds to an Am value.

[0024] Specifically, such as Figure 2 As shown, the two-dimensional moving average method is as follows:

[0025] For a given D that needs to be traversed, its current index is (Rd, Vd), denoted as D(Rd, Vd); when traversing to D(Rd, Vd), the following processing is performed:

[0026] Skip the guard interval Px of the distance dimension, take the D(Rd, Vd) distance dimension noise floor sample point set (Rd-Px-N: Rd-Px, Vd-m: Vd+m), (Rd+Px: Rd+Px+N, Vd-m: Vd+m), calculate their average values, and then obtain the reference noise floor NFr(Rd, Vd) of the distance dimension through the decision maker; where N is the length of the distance dimension sampling window, and m is half the width of the distance dimension sampling window;

[0027] Skip the guard interval Pv of the velocity dimension, take the velocity dimension noise floor sample point set D(Rd, Vd) (Rd-n: Rd+n, Vd-Pv-M: Vd-Pv), (Rd-n: Rd+n, Vd+Pv: Vd+Pv+M), calculate their respective average values ​​AVGv1 and AVGv2, and then obtain the velocity dimension reference noise floor NFv(Rd, Vd) through the decision maker; where M is the velocity dimension sampling window length and n is half of the velocity dimension sampling window width;

[0028] The two-dimensional spectral noise floor calculation is now complete. The final CFAR decision threshold can then be obtained by weighting the Wr and Wv coefficients, and compared with the original data to obtain the CFAR output. This part of the process is completely identical to that in existing technologies, so it will not be described in detail here.

[0029] The noise floor obtained after traversing all (R, V) is the RV two-dimensional spectral noise floor.

[0030] During the traversal, slide row by row or column by column. Each slide can only change one smallest unit and cannot change two dimensions at the same time.

[0031] like Figure 3 As shown, during the traversal, the sliding is performed according to the distance dimension. The index after sliding becomes (Rd+1, Vd). Then, the average distance dimension value AVGr1' corresponding to the index (Rd+1, Vd) is: AVGr1+(Am(Rd-Px+1, Vd-m: Vd+m)-Am(Rd-Px-N, Vd-m: Vd+m)) / (N*2m), and the average distance dimension value AVGr2' is: AVGr2+(Am(Rd+Px+N+1, Vd-m: Vd+m)-Am(Rd+Px, Vd-m: Vd+m)) / (N*2m). The reference noise floor NFr(Rd+1, Vd) of the distance dimension is calculated based on the average distance dimension values ​​AVGr1' and AVGr2'.

[0032] The velocity dimension average value AVGv1' corresponding to the index (Rd+1, Vd) is: AVGv1+(Am(Rd+n+1, Vd-Pv-M: Vd-Pv)-Am(Rd-n, Vd-Pv-M: Vd-Pv)) / (M*2n), and the velocity dimension average value AVGv2' is: AVGv2+(Am(Rd+n+1, Vd+Pv: Vd+Pv+M)-Am(Rd-n, Vd+Pv: Vd+Pv+M)) / (M*2n); the reference noise floor NFv(Rd+1, Vd) of the velocity dimension is calculated based on the velocity dimension average values ​​AVGv1' and AVGv2'.

[0033] like Figure 4As shown, during the traversal, the sliding is performed according to the velocity dimension, and the index after sliding becomes (Rd, Vd+1). Then, the average distance dimension AVGr1' corresponding to this index (Rd, Vd+1) is: AVGr1+(Am(Rd-Px-N:Pd-Px, Vd+m+1)-Am(Rd-Px-N:Pd-Px, Vd-m)) / (N*2m), and the average distance dimension AVGr2' is: AVGr2+(Am(Rd+Px:Rd+Px+N, Vd+m+1)-Am(Rd+Px:Rd+Px+N, Vd-m)) / (N*2m). The reference noise floor NFr(Rd, Vd+1) of the distance dimension is calculated based on the average distance dimension AVGr1' and AVGr2.

[0034] The velocity dimension average value AVGv1' corresponding to the index (Rd, Vd+1) is: AVGv1+(Am(Rd-n:Rd+n,Vd-Pv+1)-Am(Rd-n:Rd+n,Vd-Pv-M)) / (M*2n), and the velocity dimension average value AVGv2' is: AVGv2+(Am(Rd-n:Rd+n,Vd+Pv+M+1)-Am(Rd-n:Rd+n,Vd+Pv)) / (M*2n); the velocity dimension reference noise floor NFv(Rd, Vd+1) is calculated based on the velocity dimension average values ​​AVGv1' and AVGv2'.

[0035] The above-described sliding sequence and the fast calculation method for the sliding mean of a two-dimensional surface are also applicable to the reverse operation (i.e., from...). Figure 3 , Figure 4 (Rn, Vm) begins to slide in the opposite direction), which will not be listed here.

[0036] In summary, this invention estimates the noise floor based on the cross-shaped two-dimensional moving average method, which improves the noise floor calculation sample from one dimension to two dimensions, making the calculated noise floor more stable and closer to the true noise floor of the RV spectrum.

[0037] In addition, the present invention combines a specific sliding sequence, and the data calculation after the sliding is based on the data of the previous sliding. That is, the average value after the sliding is the average value before the sliding minus the average signal amplitude of the sample area block that slid out plus the average signal amplitude of the sample area block that slid in, thereby reducing the calculation time of the average value of the two-dimensional surface sampling data during the sliding process and improving the noise floor estimation speed.

[0038] 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 method for estimating the noise floor of a R-V two-dimensional spectrum based on a moving average, for processing a radar return signal to obtain a noise floor, the radar return signal being a three-dimensional spectrum with three dimensions corresponding to a range dimension index R, a Doppler velocity dimension index V, and a signal amplitude Am at (R, V) index, characterized in that: The estimation method is: traversing all (R, V) of the radar echo signal, and calculating the noise floor by the two-dimensional sliding average method during the traversal; Specifically, for a certain D that needs to be traversed, the current index is (Rd, Vd), denoted as D(Rd, Vd); when D(Rd, Vd) is traversed, the following processing is performed: Skip the protection interval Px of the distance dimension, take the distance dimension noise floor sample point set (Rd-Px-N:Rd-Px, Vd-m:Vd+m) and (Rd+Px:Rd+Px+N, Vd-m:Vd+m) of D(Rd, Vd), calculate the average values respectively, and then obtain the reference noise floor NFr(Rd, Vd) of the distance dimension through the judge; wherein, N is the length of the distance dimension sampling window, and m is half of the width of the distance dimension sampling window; Skip the protection interval Pv of the velocity dimension, take the velocity dimension noise floor sample point set (Rd-n:Rd+n, Vd-Pv-M:Vd-Pv) and (Rd-n:Rd+n, Vd+Pv:Vd+Pv+M) of D(Rd, Vd), calculate the average values AVGv1 and AVGv2 respectively, and then obtain the reference noise floor NFv(Rd, Vd) of the velocity dimension through the judge; wherein, M is the length of the velocity dimension sampling window, and n is half of the width of the velocity dimension sampling window; After traversing all (R, V), all the noise floors obtained are the R-V two-dimensional spectrum noise floor; During the traversal, slide row by row or column by column, and the average value after each sliding is the average value before sliding minus the average value of the signal amplitude of the sample area block that slides out and plus the average value of the signal amplitude of the sample area block that slides in, and the reference noise floor after sliding is calculated based on this; The traversal process includes the following order operations: During the traversal, slide according to the distance dimension, and the index after sliding becomes (Rd+1, Vd), then the distance dimension average value AVGr1' corresponding to the index (Rd+1, Vd) is: AVGr1+(Am(Rd-Px+1, Vd-m:Vd+m)-Am(Rd-Px-N, Vd-m:Vd+m)) / (N*2m), and the distance dimension average value AVGr2' is: AVGr2+(Am(Rd+Px+N+1, Vd-m:Vd+m)-Am(Rd+Px, Vd-m:Vd+m)) / (N*2m); the reference noise floor NFr(Rd+1, Vd) of the distance dimension is calculated based on the distance dimension average values AVGr1' and AVGr2'; The velocity dimension average value AVGv1' corresponding to the index (Rd+1, Vd) is: AVGv1+(Am(Rd+n+1, Vd-Pv-M:Vd-Pv)-Am(Rd-n, Vd-Pv-M:Vd-Pv)) / (M*2n), and the velocity dimension average value AVGv2' is: AVGv2+(Am(Rd+n+1, Vd+Pv:Vd+Pv+M)-Am(Rd-n, Vd+Pv:Vd+Pv+M)) / (M*2n); the reference noise floor NFv(Rd+1, Vd) of the velocity dimension is calculated based on the velocity dimension average values AVGv1' and AVGv2'; Or, In the traversal process, the index becomes (Rd, Vd+1) after sliding in the speed dimension, and the distance dimension average value AVGr1' corresponding to the index (Rd, Vd+1) is: AVGr1+(Am(Rd-Px-N:Pd-Px, Vd+m+1)-Am(Rd-Px-N:Pd-Px, Vd-m)) / (N*2m), and the distance dimension average value AVGr2' is: AVGr2+(Am(Rd+Px:Rd+Px+N, Vd+m+1)-Am(Rd+Px:Rd+Px+N, Vd-m)) / (N*2m); the reference noise NFr(Rd, Vd+1) in the distance dimension is calculated according to the distance dimension average values AVGr1' and AVGr2; The speed dimension average value AVGv1' corresponding to the index (Rd, Vd+1) is: AVGv1+(Am(Rd-n:Rd+n, Vd-Pv+1)-Am(Rd-n:Rd+n, Vd-Pv-M)) / (M*2n), and the speed dimension average value AVGv2' is: AVGv2+(Am(Rd-n:Rd+n, Vd+Pv+M+1)-Am(Rd-n:Rd+n, Vd+Pv)) / (M*2n); the reference noise NFv(Rd, Vd+1) in the speed dimension is calculated according to the speed dimension average values AVGv1' and AVGv2'.

2. The method of claim 1, wherein the method is based on a sliding average of the R-V two-dimensional spectrum. In the traversal process, only one minimum unit can be changed each time, and two dimensions cannot be changed at the same time.

3. The method of claim 1, wherein the method is based on a sliding average of the R-V two-dimensional spectrum. The traversal process includes reverse direction operation, and the sliding order of the reverse direction operation and the calculation method of the speed dimension average value and the distance dimension average value in the sliding process are the same as those of the forward direction operation.

Images