Method for determining the true velocity of a target object detected by radar measurement of an FMCW radar within a single measurement cycle using range migration

Abstract

Method for determining the true speed of a target object detected by radar measurement of an FMCW radar within a single measurement cycle, comprising the following steps: - Generating at least one reference signal for at least one specified reference speed, and - Generating an error measure (||e||) that quantifies the effect of the range migration deviation on the frequency response of the Doppler window, - Using the error measure (||e||) to determine which of the ambiguous intervals existing due to the sampling rate contains the true velocity of the target object, where, in the case of multiple reference speeds, the error measure (||e||) is used in such a way that at each boundary of a speed interval of each of the reference speeds an associated error measure (||e 1st ||; ||e 2ndll) the measured velocity of the target object is calculated relative to the reference velocities, and where a weighting of each error measure (||e 1st ||; ||e 2nd ||) per reference speed, and where the speed interval at which the error measure (||e 1st ||; ||e 2nd ||) is the smallest when the interval is determined into which the true velocity of the target object falls.

Description

[0001] The present invention relates to the field of radar-based speed measurement in the automotive sector. It solves the problem of velocity ambiguity within a measurement cycle.

[0002] Modern vehicles (cars, vans, trucks, motorcycles, etc.) are equipped with a multitude of sensors that provide the driver with information and control individual vehicle functions semi- or fully automatically. Radar technology is a key sensor principle for environmental perception in autonomous and semi-autonomous vehicles. Radar sensors in vehicles typically comprise multiple transmitting and receiving elements that form virtual channels of the radar sensor (Rx / Tx antenna pairs). In each receiving channel, the signal is down-converted to the baseband, filtered, and then digitized. Preprocessing the baseband signals for each receiving channel enables the detection and localization of a target object within the radar sensor's field of view.

[0003] Most radar sensors currently used in vehicles operate as multi-pulse radar sensors (also known as chirp sequence radar sensors), emitting several frequency-modulated pulses at short intervals. Preprocessing can include, in particular, a Fourier transform or optimal filtering of the baseband data for a single radar pulse (fast-time processing), a further Fourier transform for multiple pulses (so-called slow-time processing), beamforming, signal power detection, for example based on constant-force-alarm-rate methods, and peak determination.

[0004] One challenge for multi-pulse radar sensors with a constant pulse repetition time (PRT) is that they are limited to measuring relative velocity (or relative radial velocity) within the so-called principal interval. Only within this principal interval, between -vm / 2 and +vm / 2, can a maximum unambiguous velocity be measured. This means that velocities at a distance vm are mapped to the same value within the principal interval. Current radar sensors used in the automotive sector typically have vm values ​​in the range of 20 to 60 m / s, which can lead to ambiguous velocity measurements. This means that measured velocities can deviate from the actual velocity by several times vm.

[0005] One approach to solving this problem is to use different pulse repetition times, either between pulses or from one coherent processing interval (CPI) to the next. This can increase the maximum unique velocity. A disadvantage of using different pulse repetition times between successive intervals is that it increases latency and necessitates linking detections across multiple intervals. This is particularly error-prone in the case of multiple and / or rapidly moving targets, as is common in automotive applications. Approaches that vary the pulse repetition time between pulses within the same interval are generally unsuitable for processing with a fast Fourier transform (FFT) and can exhibit sidelobes in the velocity spectrum.

[0006] A disadvantage of previously known methods is the comparatively complex processing and the associated need for computing power. Furthermore, the existing approaches usually result in a loss of resolution.

[0007] WO 2020 / 201 337 A1 therefore proposes a device for determining the relative velocity of a target object by means of model fitting. EP 4 258 010 A1 discloses a vehicle radar with the capability to mitigate range migration. In Xu, L., Lien, J., Li, J.: Doppler-Range Processing for Enhanced High-Speed ​​Moving Target Detection Using LFMCW Automotive Radar. In: IEEE Transactions on Aerospace and Electronic Systems, Vol. 58, No. 1, pp. 568-580, 2022. - ISSN 0018-9251, an algorithm for reducing range / Doppler migration and velocity ambiguity is shown. However, since there is still room for improvement in these solutions, the invention is based on the objective of providing an improved method for determining the relative velocity of a target object.

[0008] This task is solved by the features of the independent claim. Advantageous embodiments are the subject of dependent claims.

[0009] A method is proposed for determining the true velocity of a target object detected by radar measurement of an FMCW radar within a single measurement cycle, comprising the steps of: generating at least one reference signal for at least one specified reference velocity, generating an error measure that quantifies the effect of the deviation of the range migration on the frequency response of the Doppler window, and using the error measure to determine into which of the ambiguous intervals existing due to the sampling rate the true velocity of the target object falls.

[0010] In one embodiment, if a single reference velocity is provided, the error measure is used in such a way that at each boundary of a velocity interval, which is a multiple of the maximum detectable velocity of the radar measurement, a signal adjacent to the peak value is generated and the associated error measure is calculated, wherein the velocity interval in which the error measure is smallest is determined to be the interval in which the true velocity of the target object falls.

[0011] In an embodiment according to the invention, if several reference velocities are provided, the error measure is used in such a way that at each boundary of a velocity interval of each of the reference velocities an associated error measure of the measured velocity of the target object relative to the reference velocities is calculated, and wherein a weighting of each error measure is carried out for each reference velocity, and wherein the velocity interval in which the error measure is smallest is determined as the interval in which the true velocity of the target object falls.

[0012] In one implementation, the error measure serves as a measure of the true speed of the target object.

[0013] In one implementation, the sign of the velocity of the detected target object is determined based on the true absolute velocity by splitting the Fast Fourier Transform function into two sub-functions, where each sub-function corresponds to half a Range Gate.

[0014] In one implementation, the Fast Fourier Transform function is split into two subfunctions, and when one range gate after the other is processed, if a target object is detected, the Doppler bin is stored at a first range gate and the sign of the result of the difference of the two subfunctions is checked in a subsequent range gate, where a positive value indicates that the target object is moving towards the radar sensor and a negative value indicates that the target object is moving away from the radar sensor.

[0015] In one implementation, double peak values ​​of the measured radar signal are detected by evaluating an energy ratio of the two peak values ​​within a range gate after the complete FFT, which is divided into four sub-functions, and if the energy values ​​of the two peak values ​​are close to each other, and if both energy ratios are greater than a predefined threshold, double peak values ​​are detected.

[0016] In one version, the specified reference speed is zero.

[0017] In one implementation, sampling signals from radar measurements are aligned with sampling signals of the radar reference signal.

[0018] Further features and advantages of the invention will become apparent from the following description of exemplary embodiments of the invention, with reference to the figures in the drawing, which shows details of the invention, and from the claims. The individual features can be implemented individually or in any combination in a variant of the invention.

[0019] Preferred embodiments of the invention are explained in more detail below with reference to the accompanying drawing. Fig. Figure 1 shows the signal spectrum after a Doppler FFT of a stationary and a moving target object. Fig. Figure 2 shows an error curve over increasing speeds. Fig. Figure 3 shows a determination of the interval in which the true speed lies, according to one embodiment of the present invention. Fig. Figure 4 shows the signal spectrum after a Doppler FFT of a stationary and a very fast moving target object. Fig. Figure 5 shows a determination of the sign of the relative velocity according to one embodiment of the present invention.

[0020] In the following figure descriptions, identical elements or functions are marked with the same reference symbols.

[0021] For driver assistance systems, which include ADAS (Advanced Driver Assistance Systems) and AD (Automated Driving) systems, a radar sensor can provide distance and speed measurements of nearby objects. Among the many modulation schemes, FMCW (Frequency Modulation Continuous Wave) is the most widely used waveform in the automotive sector. With FMCW modulation, a sequence of so-called chirps is transmitted. The signal reflected after striking an object (hereinafter referred to as the target object) is captured by the radar sensor and converted into a baseband signal. An analog-to-digital converter then samples each chirp to obtain the corresponding sample values. This results in a two-dimensional data block x[n]. t ,m tOne dimension runs along the samples in the same chirp and is called the fast-time dimension. The other dimension, called the slow-time dimension, runs across the chirps, but along the same location of the analog-to-digital converter samples.

[0022] Often, a two-dimensional FFT is applied to the two-dimensional data block x[n]. t ,m t ] applied, so that the frequency response S[nf,mf] can be generated. If a reflection center is present in the field of view, a high response, i.e., a high-energy value at a frequency location in the two-dimensional spectrum, is achieved, so that the distance (range) and velocity of the target object from the two-dimensional location of the high response can be obtained.

[0023] To achieve a specific, unambiguous measurement of the maximum frequency, the Nyquist sampling principle dictates that the sampling rate should be at least twice as high as the highest frequency component being examined in the measured signal. However, this is difficult to achieve for many automotive applications, particularly with regard to speed. Consequently, an aliasing effect is often present, meaning that a signal originating from a high speed will fold back into a specific interval. The simple FFT-based peak detection method cannot determine the true speed; it merely outputs the folded speed value within this specific interval.

[0024] A universal approach to solving the speed ambiguity problem is to set two or more different pulse repetition frequencies in successive radar cycles (measurement cycles), keeping the pulse repetition time / pulse repetition frequency constant within each cycle. This results in distinct, unambiguous speed limits in these measurement cycles. The so-called Chinese Reminder Theorem is then applied to determine the true speed from the ambiguous speed estimates resulting from the different pulse repetition frequency configurations.

[0025] If the target object has a radial velocity component, the distance will shift from its original position with each chirp as soon as a new radar cycle (hereinafter also referred to as a measurement cycle) begins. This effect is called range migration. At low velocities, e.g., |v| < 20 m / s, or if the total chirp time in a measurement cycle is very short, the influence of range migration is usually negligible. However, if the total chirp time in a measurement cycle is long enough to provide sufficient velocity resolution, and the target object has a velocity that falls within the ambiguous intervals, the effect of range migration will produce artifacts strong enough to be measured.

[0026] The signal processing normally used in automotive radars, based on a simple FFT (Fast Fourier Transform), ignores the effect of range migration. In the present invention, however, this effect is used to determine the true speed (i.e., the actual relative speed of the target object to the radar) of a target object detected by the radar. That is, the true speed is determined from an ambiguous speed measurement within a measurement cycle by utilizing the effect of range migration.

[0027] The following steps are performed. First, a signal model is created. Here, a reference signal is generated for a specified reference velocity of the target object. Specifically, this is achieved by modifying the Doppler window (weighting the signals in the Doppler dimension), as described in detail below.

[0028] In a further step, an error measure is generated to quantify the effect of the range migration deviation on the frequency response of the Doppler window.

[0029] In a further step, this error measure is used to determine into which of the ambiguous intervals existing due to the sampling rate the true speed of the target object falls.

[0030] As previously described, a signal model is created in a first step. To simplify the description, only one Doppler dimension is considered. However, the method is also applicable to radar systems with multiple antenna channels, such as MIMO radars, by extracting the distance Doppler map at a specific angular position.

[0031] The signal model of a point-like target object subject to the effect of range migration (ignoring signal amplitude for simplification), after weights and before FFT, is: x[nt,mt]=wR[nt]⋅ej2π(fR+fRM[mt])Δtnt⋅wD[mt]⋅ej2πfDTpmt where the superposition frequency f R , which corresponds exclusively to the distance (English = range), is calculated as: fR=2cBeN⋅ΔtR

[0032] The shift in the superposition frequency f caused by range migration RM [mt] on the mt-en chirp is estimated to be about: fRM[mt]=2cBeN⋅Δtv⋅Tp⋅mt

[0033] Then a DFT (discrete Fourier transform) is performed along a so-called fast-time dimension nt of the time signal x[n]. t ,m t ] applied, which leads to: S[nf,mt]=wD[mt]⋅ej2πfDTpmt⋅WR[nf−(fR+fRM[mt])]

[0034] The term W R[•] also has a dependency on the chirp index mt under the influence of range migration. By rearranging the W R and W D to term W DR You will receive: WDR[nf,mt]=WD[mt]⋅WR[nf−(fR+fRM[mt])]

[0035] It can be seen that the effect of range migration can be considered a modification of the original Doppler window. More precisely, the main lobe of the frequency response of the Doppler window is flattened under the effect of range migration, as shown in Fig. 1 (dotted curve) shown.

[0036] As previously described, an error measure is generated in a second step. This error measure is a metric for the artifacts of the range migration.

[0037] For example, a = [a o a1 a2 a3 a4] T set as the signal values ​​in the vicinity of the peak value of the original frequency response of w o[mt]. In this case, the true (real) signal (radar measurement signal) Y = [y0 y1 y2 y3 y 4] T a different amplitude (and therefore a different scaling factor k) and additional noise n i exhibit: y i = k · a i + n i

[0038] The noise can be further decomposed into a general noise level n and a frequency-specific variation term e i : y i = k · a i + n + e i .

[0039] Thus, the signals adjacent to the peak value y = k · a + n · 1 + e can be obtained, where 1=

[11111] T and e=[e0e1e2e3e4]T.

[0040] Is it called A=[a01a11a21a31a41] The signals adjacent to the peak value can be rearranged to y = A · kn+e.

[0041] The value of k n = [kn] Tcan be determined using least squares, e.g. k^n=(ATA)−1ATy

[0042] However, the direct use of the least squares method can lead to a k n lead, which fits well into an arbitrary y.

[0043] It is therefore proposed to impose some restrictions on k n to add to the optimization problem. For example, such a constraint can be added in the form of a loss function: L(kn)=‖y−A⋅kn‖2+qk(k−1)2+qnn2 where q k and q n Predefined weighting factors are used to restrict the values ​​of k and n. It should be noted that the restriction of k close to 1 is based on the normalized versions of the vectors y = y / ||y|| and a = a / ||a|| included in the calculation. The restriction term n 2 This refers to the expected noise level compared to the signal of the target object. Therefore, the factor q cann be set according to the expected, measured signal-to-noise ratio.

[0044] The minimizer of L(k n ) is similar to the unconstrained least-squares solution: k n = (A T A) + -1 (A T y) + , where the new matrices (A T A) + and (A T y) + subject to the restriction of k n Calculated as: (ATA)+=ATA+(qk00qn) and (ATy)+=ATy+(qk0)

[0045] The residual error should still be obtained without the constraint term as: ‖e‖=‖y−A⋅k^n‖

[0046] The error sources of ||e|| can be categorized into two classes: the deviation of the main lobe from model A and other signal noise. In detail, the influence of the main lobe shape deviation will dominate for target objects with a good signal-to-noise ratio (SNR). This is assumed to result primarily from the flattening effect of range migration. Numerical calculations also show that the flattening effect of range migration becomes more pronounced as the relative velocity v increases. Thus, ||e|| can be used as an error measure to reflect the underlying true velocity of the target object. Fig. Figure 2 shows an example of the error measure where the error increases monotonically with increasing speed.

[0047] In a further step, the true speed is determined. Here, the speed value estimated using spectral analysis is expressed as v.f denoted. The true velocity v will be a value that represents a displacement of several maximum unique velocities v. max is, that is, v = v f + h · v max , where h is the integer describing the true interval in which the true velocity v lies. The error measure of the range migration artifacts (as defined above) can be used to set limits and to determine the value of h. Two cases can be distinguished for this purpose:

[0048] In the first case, a single reference velocity and several limit values ​​are specified. In this case, matrix A is generated for an ideal response of the Doppler window, i.e., at a target object velocity of v=0. Then, at each limit, v i = h · v max a neighborhood to be observed y i generated at a defined SNR and the error ||e i|| calculated. The errors at each limit can be used as a limit to determine the velocity interval into which the true measurement of y lies. v is supposed to fall.

[0049] As in Fig. As indicated in point 2, this applies in the case where the observation vector y is used. v calculated errors ||e v || lies within two reference errors, i.e., ||e i || < ||e v || < ||e i+1 ||, the true velocity v also lies between the two corresponding velocities v i and v i+1, which were used to fix the errors ||e i || and all i+1 to obtain ll. That means that the following holds: v i < v < v i+1 This resolves the ambiguity regarding speed.

[0050] In the second case, several reference speeds are specified, as in Fig. 3 shown. All interval boundaries, i.e. v max , 2 v max , 3 v maxetc. are chosen as reference velocities. By evaluating the error measure of the observation vector y v (Neighborhood of the peak value as a result of the Doppler FFT for a signal observed at velocity v) with all reference velocities, the two best reference velocities (here v) can be determined. max and 2 v max ) can be used to calculate a weighted total position v w to calculate. The hypothesis h opt with the smallest distance between v opt = v f + h opt · v max and v w is considered to be the one through y v in Fig. 3. Displayed true speed considered.

[0051] For example, a model matrix A h for every speed at a limit h · v max set up. During the application, the observation vector y is v Calculated for each reference speed using: ||e h || = ||y v- A h · k̂ n , h || Is it provided that the two smallest errors ||e 1st || and ||e 2nd || corresponding to the interval index h1 and h2, a weighted subinterval h can be w to be calculated: hw=w1st⋅h1st+w2nd⋅h2nd where the weights w 1st (for h 1st ) and w 2nd (for h 2nd ) with the corresponding error ||e 1st || and ||e 2nd || will be evaluated via: wh=1‖eh‖+b⋅1S

[0052] The parameter b is a predefined constant that avoids division by zero and the influence of the error ||e h ||regulated. S is a normalization factor that ensures that w 1st + w 2nd = 1, i.e.: S=1‖e1st‖+b+1‖e2nd‖+b

[0053] The weighted sum h w results in a subinterval value that lies between h 1st and h 2ndlies and is closer to the one with the smallest error.

[0054] Subsequently, the true interval h can be determined. opt can be obtained using the following equation: hopt=argminh|(vf+h⋅vmax)−hw⋅vmax|

[0055] In another iteration, the sign of the velocity is determined by splitting the Doppler FFT into several sub-functions FFT1, FFT2 (in Fig. (5 to be seen) is divided. In automotive applications, both positive and negative velocities are often of interest. The procedure described above can determine the best hypothesis h. opt Determine among all positive intervals, or among all negative intervals. However, it cannot be the best hypothesis h. opt Determine under positive and negative intervals at the same time, since the applied distance window W R [n f] is usually an even function (i.e., symmetric about W). R[0] due to the zero-group delay condition). Furthermore, at the range gate RG (distance gate) nf=fR+fRM[M2], at which the highest spectral Doppler response is expected, the response of the range window: WR[nf−(fR+fRM[mt])]=WR[fRM[M2]−fRM[mt]]

[0056] At a specific chirp index mt, the frequency shift f RM+ and f RM- Due to range migration of a positive and a negative velocity, the following property arises: fRM+[M2]−fRM+[mt]=−(fRM−[M2]−fRM−[mt])

[0057] Due to the symmetry of W R [n f ]: WR[fRM+[M2]−fRM+[mt]]=WR[fRM−[M2]−fRM−[mt]]

[0058] Consequently, the following applies at the Range Gate nf=fR+fRM[M2] the following ratio: WR[nf−(fR+fRM+[mt])]=WR[nf−(fR+fRM−[mt])]

[0059] In summary, a target object moving away from the sensor generates the same signal as a target object moving towards the sensor at the same absolute velocity. To determine the sign of the velocity, two methods are proposed below, both based on knowledge of the true absolute velocity as determined in the method described above.

[0060] In a first procedure, after determining the true absolute velocity v opt = v f + h opt · v max at the Range Gate nf=fR+fRM[M2] The highest frequency response was determined, and two inverted integrations were performed: SD=∑mtM−1S[G(mt),mt] e−j2πfDTpmt

[0061] The range gate index G(mt) at a chirp is not constant, but depends on the chirp index. A first integration S D,1 starts at the Range Gate G1(0)=nf−ΔGRM[M2], where the shift of the range gate per chirp is due to the speed ΔGRM[mt]=2Becvopt⋅Tp⋅mt The value is then used. Sampled values ​​from the following range gates are used at the mt-en chirp: G1(mt)=G1(0)+ΔGRM[mt]

[0062] The second integration S D,2 starts at the Range Gate G2(0)=nf+ΔGRM[M2] and calls up the sample values G2(mt)=G2(0)+ΔGRM[mt] away.

[0063] The integrations mentioned above can be effectively performed with two Doppler FFTs at f D are applied, with both along the assumed respective range migration traces across the chirps, unlike the Classical Doppler FFT, which is performed along the chirps in a range gate.

[0064] The sign of the velocity can be obtained by: |S D,1 | - |S D,2 |

[0065] Another option is to use the result obtained by classical Doppler FFT ||S D,0 || at f D to evaluate, but along the chirps at the Range Gate nf=fR+fRM[M2]. The sign is preserved by the sign of |S (D,1) |-|S( D,0 )| or |S( D,0 )|-|S( D,2 )|.

[0066] The first method cannot be used if a radar system processes data only range gate by range gate, especially if the data from one range gate is no longer available after processing. In this case, data from multiple range gates, required for the first method, is unavailable. Therefore, in a second method, the Doppler FFT is performed on the chirps in the same range gate. However, the full FFT S[n] is not used. f , m f ] in two shorter FFTs S 2,0 and S 2,1 to be carried out as follows: S[nf,mf]=∑mt=0M2−1S[nf,mf] e−j2πmfMmt+∑mt=M2M−1S[nf,mf] e−j2πmfMmt =S2,0[nf,mf]+S2,1[nf,mf]

[0067] When a target object is detected in a range gate (designated RG1), its Doppler bin is measured. D (Doppler gate) stored. In a subsequent range gate (designated RG2), the sign of |S is stored. 2,0 [RG2, m D ]| - |S 2,1 [RG2, m D ]| checked. A positive value indicates that the target object is moving towards the radar sensor.

[0068] The selection of RG2 depends on the frequency response of the spacing window. Advantageously, RG2 should be chosen to be RG1 + offset, where the offset corresponds to the distance between the peak value and the first trough in the frequency response.

[0069] In another iteration, double peak values ​​are determined by decomposing the Doppler FFT into several sub-functions. Based on the above methods, knowledge of the correct Doppler bin of the target object is expected, which is typically achieved via a processing chain involving peak value acquisition, CFAR, etc. However, if the true velocity is very high, e.g., above 60 m / s, double peak values ​​appear around the true Doppler bin, as shown in Fig. 4 can be seen. Methods such as peak detection cannot be used to determine the correct Doppler location of the target object.

[0070] Double peak values ​​only appear when the range migration effect is significant, i.e., for target objects with high relative velocities to the radar, e.g., >60 m / s. In this case, the main lobe of the frequency response of the distance window will be located at the range gate, e.g., at nf=fR+fRM[M2], The peak value is located in the middle chirp samples of the range gate. The preceding and trailing chirps primarily sample the side lobes of the frequency response of the distance window. Consequently, the Doppler FFT result on the samples of the middle chirps will show a significantly higher response than the Doppler FFT result on the samples of the preceding and trailing chirps.

[0071] To take advantage of this fact, the full Doppler FFT is split into four short FFTs using a range gate: S[nf,mf]=∑k=03∑mt=kM4(k+1)M4−1S[nf,mf] e−j2πmfMmt =∑k=03S4,k[nf,mf] where S4,k[nf,mf]=e−jπ2mf⋅k∑m't=0M4−1S[nf,m't+kM4] e−j2πmfMm't

[0072] Unlike decimation in time FFT, simple partitioning cannot fully exploit the symmetry of the twiddle factors. However, the twiddle factors do e−jπ2mf⋅k phases of a variety of π2, so that complex multiplication is merely a reversal of the sign, a setting of output values ​​to zero, or an exchange of real and imaginary components, and does not introduce any further complex multiplication.

[0073] In summary, the following steps can be taken to handle Doppler peak values, as described in Fig. 5 shown: 1. In a range gate, if two peak values ​​are very close to each other (e.g., |m f,2 - m f,1| ≤ 5) along the Doppler dimension and after the complete Doppler FFT are recorded, and if their energy values ​​are very similar (e.g. within an absolute difference of ≤2dB), these are considered double peak values ​​and processed further. 2. The following energy ratio is determined for the double peak values: ΔSnf,mf2=|S4,1[nf,mf]|2+|S4,2[nf,mf]|2|S4,0[nf,mf]|2+|S4,3[nf,mf]|2 3. If both ΔSnf,mf,12 as well as ΔSnf,mf,22 greater than a limit T ΔS2 are the peak values ​​at m f,1 and m f,2 as double peak values ​​due to distance migration. The true Doppler locus is midway between m f,1 and m f,2 to find.

[0074] The choice of the limit T ΔS2 should consider both the SNR level and the shape of the Doppler window.

[0075] In a further implementation, sampling signals from radar measurements are aligned with sampling signals of the radar reference signal. Thus, the peaks of the main lobes coincide. This enables higher measurement accuracy. The effectiveness of the matching formula k̂ n = (A T A) + -1 (A T y) + suggests that the data a = [a a a1 a2 a3 a4] T to calculate model A and the data in signal y = [y0 y1 y2 y3 y4] Tare aligned with each other, meaning, for example, that both a2 and y2 coincide with the peak value of the frequency response window. However, during actual application, the theoretical peak value of the real observed data can lie at any point between two adjacent sampling points. By curve-fitting the data vector y and re-sampleting the fitted curve, the data vector y and the model vector a can be aligned.

[0076] For example, B-Spline can be used to perform curve fitting: y(x)=∑ic[i]⋅B(x−i) where B(x) is the spline basis function.

[0077] Assume that the original sampling position y0 in the neighborhood vector y corresponds to x s , then the optimal scanning position x opt , in order to be aligned with the model vector a, satisfy: -0.5 ≤ x opt - x s ≤ 0.5

[0078] For the model vector where the midpoint a2 in a represents the peak value in the response, the re-sampled data vector y should also have a peak value at y(x). opt + 2). Additionally, due to the symmetry of the frequency response window, the optimal re-sampling point should be the main lobe: xopt=argmaxx(p0|y(x+2)−0.5(y(x+1)+y(x+3))|−p1|y(x+1)−y(x+3)|− p2|y(x)−y(x+4)| where the first term p 0| The expression y(x + 2) - 0.5(y(x + 1) + y(x + 3))| expresses that the true peak (or the trough in the case of Doppler peaks) has the greatest difference to its neighboring sampling points. The second and third terms p1|y(x + 1) - y(x + 3)1 + p2|y(x) - y(x + 4)| restrict the symmetry form of the main lobe.

[0079] p0, p1, p2 are three predetermined weights that control the contribution of the peak difference and the symmetry constraint to the optimization. More precisely, if p1=p2=0, the optimization degrades to the case of searching for the highest (or lowest in the case of double peaks) value on the fitted curve as the location for (x opt + 2).

[0080] This method is used in the field of semi-autonomous to autonomous driving. For this purpose, a vehicle is equipped with appropriate sensors and actuators. It also has at least one control unit on which the described methods are implemented as a computer program.

[0081] Determining distances, angles and relative velocities to target objects, i.e. objects located within the field of view of the vehicle's radar sensor, is necessary to avoid, for example, rear-end collisions. It refers to: Be Effective bandwidth at each chirp c Speed ​​of light N, M Number of analog-to-digital converter samples in one chirp; number of chirps Δt Analog-to-digital converter sampling interval within a chirp R radial distance of the target object at the start of the measurement cycle RG Range Gate W R [n f] Frequency response of the distance window at n f -ten Range Gate W D [m f Frequency response of the distance window at m f -ten Doppler bin W R [n t ] nt-ter coefficient of the distance window in the time domain W D [m t ] m t -th coefficient of the Doppler window in the time domain w DR [n f , m t Impulse response of the Doppler window modified by range migration mt Chirp Index f c Carrier frequency fR Superposition frequency f RM [mt] Shift in the superposition frequency due to range migration at the mt-th chirp v true velocity of the target object v f ambiguous velocity estimate (converted from spectral analysis) vmax is a unique velocity interval - if the true velocity lies within this interval, then v = vf x[n t , m t ] (complex) time signal of the nt-th sample in the mt-th chirp S [n f , m f ] DFT sample value at the n f -ten Range Gate and the m f -ten Doppler bin T p Pulse repetition time h i Index of the ambiguous intervals of interest in velocity ||e|| Error measure y vObservation vector (neighborhood of the peak value as a result of the Doppler FFT for a signal observed at velocity v)

Claims

[1] Method for determining the true velocity of a target object detected by radar measurement of an FMCW radar within a single measurement cycle, comprising the steps: - Generating at least one reference signal for at least one specified reference speed, and - Generating an error measure (||e||) that quantifies the effect of the range migration deviation on the frequency response of the Doppler window, - Using the error measure (||e||) to determine which of the ambiguous intervals existing due to the sampling rate contains the true velocity of the target object, where, in the case of multiple reference speeds, the error measure (||e||) is used in such a way that at each boundary of a speed interval of each of the reference speeds an associated error measure (||e 1st ||; ||e 2ndll) the measured velocity of the target object is calculated relative to the reference velocities, and where a weighting of each error measure (||e 1st ||; ||e 2nd ||) per reference speed, and where the speed interval at which the error measure (||e 1st ||; ||e 2nd ||) is the smallest when the interval is determined into which the true velocity of the target object falls. [2] Method according to claim 1, wherein, in the case that a single reference speed is provided, the use of the error measure (||e||) is carried out such that at each boundary of a speed interval which is a multiple of the maximum detectable speed (v) i = h · v max ) of the radar measurement is a signal adjacent to the peak value (y) i ) generated and the associated error measure (all ill) is calculated, whereby the velocity interval in which the error measure (||e||) is smallest is determined as the interval into which the true velocity of the target object falls. [3] Method according to any of the preceding claims, wherein the error measure (||e||) serves as a measure of the true velocity of the target object. [4] Method according to one of the preceding claims, wherein, based on the true absolute velocity, the sign of the velocity of the detected target object is determined by splitting the Fast Fourier Transform function into two sub-functions, each sub-function corresponding to half a Range Gate. [5] A method according to any of the preceding claims, wherein the Fast Fourier Transform function is split into two sub-functions, and when one range gate is processed after the other, in the event that a target object is detected, the Doppler effect (m D) is stored at a first range gate (RG1) and the sign of the result of the difference of the two subfunctions |S 2,0 [RG2, m D ]| - |S 2,1 [RG2, m D ]| is checked in a subsequent Range Gate (RG2), where in the case of a positive value it is detected that the target object is moving towards the radar sensor and in the case of a negative value it is detected that the target object is moving away from the radar sensor. [6] A method according to any of the preceding claims, wherein double peak values ​​of the measured radar signal are detected by determining an energy ratio in the case that within a range gate two closely spaced peak values ​​are detected along a Doppler dimension after the complete FFT divided into four sub-functions, and the energy values ​​of the two peak values ​​are close to each other. (ΔSnf,mf2) the two peak values ​​are evaluated and in the event that both energy ratios (ΔSnf,mf2) Double peak values ​​are detected if the values ​​are greater than a specified threshold. [7] Method according to any of the preceding claims, wherein the specified reference velocity is zero. [8] Method according to one of the preceding claims, wherein sampling signals from radar measurements are aligned to sampling signals of the radar reference signal.

Images