SA RADAR SENSOR FOR MOTOR VEHICLES

Abstract

Radar sensor for motor vehicles. The radar sensor has a high-frequency part, which is configured to transmit sequences of modulated radar pulses and to receive the corresponding radar echoes, and an electronic evaluation part, which is configured to take distance and angle measurements using a synthetic aperture, and includes a scanning module, an FFT module for performing fast Fourier transforms to calculating a two-dimensional distance/velocity radar image, a transform module configured to transform the raw data, while simultaneously correcting migration effects, into a format that can be processed by the FFT module, by applying a transform function defined by a number N of coefficients, and a coefficient module for preparing the coefficients for the transform module. The coefficient module including a memory, in which there is stored an initial set of coefficients comprising fewer than N coefficients, and a recursion module, for recursive calculation of the remaining coefficients.

Claims

1. A radar sensor for a motor vehicle, comprising: a high-frequency part configured to transmit sequences of modulated radar pulses and to receive corresponding radar echoes; and an electronic evaluation part configured to take distance and angle measurements by a principle of synthetic aperture, the electronic evaluation part including: a scanning module configured to scan amplitudes of the received radar echoes as raw data in a two-dimensional data space, in which one dimension represents a change in the amplitudes over time within a radar pulse, and the other dimension represents the change in amplitudes from pulse to pulse, an FFT module with specialized hardware configured to perform fast Fourier transforms for calculating a two-dimensional distance/velocity radar image, a transform module configured to transform the raw data, while simultaneously correcting migration effects, into a format that can be processed by the FFT module, by applying a transform function defined by a number N of coefficients, and a coefficient module configured to prepare the coefficients for the transform module, wherein the coefficient module includes: a memory in which is stored an initial set of coefficients including fewer than N coefficients, and a recursion module configured for recursive calculation of the remaining of the N coefficients.

2. The radar sensor as recited in claim 1, wherein the coefficients are functions of two integer indices n and k, and wherein the memory contains at least one initial set of coefficients which correspond to a fixed value of the index n and all values of the index k.

3. The radar sensor as recited in claim 2, wherein a value range of the index n is divided into a plurality of blocks, the memory contains an initial set of coefficients for each of these blocks, and the recursion module is configured to carry out a separate recursion for each block.

4. The radar sensor as recited in claim 2, wherein the recursion module is configured to calculate, for the entire value range of the index n or for each block, two recursive sequences, which start at a value of n within the value range or block and progress in opposing directions to edges of the value range or block.

5. The radar sensor as recited in claim 2, wherein the recursion module is configured to perform recursions with different increments of the index n.

6. The radar sensor as recited in claim 1, wherein the transform module is configured to perform a convolution of two time-dependent functions in that the functions are transformed, with the aid of specialized hardware for performing fast Fourier transforms, into frequency domain and there multiplied by one another, and a product is inverse transformed, once again using specialized hardware for fast Fourier transforms, back into time domain.

7. The radar sensor as recited in claim 1, wherein the radar sensor is a rapid-chirp FMCW radar.

8. The radar sensor as recited in claim 6, wherein the transform function is a chirp z-transform function.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0024] FIG. 1 shows a block diagram of an SA radar sensor according to the present invention.

[0025] FIG. 2 shows a time diagram of a pulse sequence transmitted by the radar sensor.

[0026] FIG. 3 shows a flow chart for a chirp z-transform.

[0027] FIG. 4 shows a recursion scheme for calculating the coefficients of the chirp z-transform.

[0028] FIG. 5 shows an example of a modified recursion scheme.

DETAILED DESCRIPTION OF EXAMPLE EMBODIMENTS

[0029] The radar sensor shown in FIG. 1 features a high-frequency part 10, which is configured for example as a rapid-chirp FMCW radar and, in each measurement cycle, transmits a sequence of frequency-modulated radar pulses or ramps by way of an array of antennas 12. The radar signals that are reflected from objects 14 are received by antennas 12 again and, as is conventional in FMCW radar, are mixed with a proportion of the signal that is transmitted at the respective point in time, such that a beat signal of significantly reduced frequency (beat frequency) is obtained.

[0030] An analog-to-digital converter stage 16 forms an interface between high-frequency part 10 and an evaluation part 18. There, the digitalized complex amplitudes of the beat signal are scanned at regular time intervals and stored as a time signal. Data is stored in a two-dimensional data space—that is to say the amplitudes A(n, k) are stored as functions of a “fast” index n and a “slow” index k.

[0031] In FIG. 2, frequency-modulated pulses 22 which are transmitted by the high-frequency part, and which are also designated as ramps or chirps, for two successive measurement cycles are illustrated schematically in a frequency/time diagram. Frequency f is entered on the vertical axis, and time t on the horizontal axis. Within each chirp, frequency f increases linearly. The center frequency of the pulses is designated fc, and the bandwidth is designated B. Fast index n counts successive scan time points 24 within an individual pulse 22, while slow index k counts the successive pulses within each measurement cycle. The number of scan time points 24 within each pulse is designated N.sub.fast, and the number of pulses 22 within each measurement cycle is designated N.sub.slow In practice, however, the numbers in N.sub.fast and N.sub.slow are significantly greater than in the diagram, which has been simplified. Typical values are for example N.sub.fast=256 and N.sub.slow=512.

[0032] Moreover, evaluation part 18 (FIG. 1) features an FFT module 26 for fast Fourier transforms (FFTs). The hardware of this module is configured to perform discrete Fourier transforms particularly fast and efficiently. As is conventional in rapid-chirp FMCW radar, a one- or two-dimensional Fourier transform is performed in the FFT module, in which one dimension is the sequence of scan time points counted by index n.

[0033] In conventional rapid-chirp radar (without SA evaluation), amplitudes A(n, k) captured in the scan module are transmitted directly to the FFT module. The two-dimensional spectrum generated by the FFT module then represents a distance/velocity radar image 28, in which each object 14 is apparent in the radar echo as a peak 14′ of which the location in the two-dimensional distance/velocity space indicates the distance d of the object and its relative velocity v. Because the frequency ramps of pulses 22 are very steep, the Doppler shift within a pulse that results from the relative movement of the objects is negligible, so the location of the peaks in the first dimension gives a good approximation to object distance d. Relative velocity v of the object results from the change in phases of the signals from pulse to pulse, measured in each case at the same scan time point, and is thus obtained by the Fourier transform in the second dimension.

[0034] In the case of the SA radar described here, however, objects 14 are not, or at least not primarily, vehicles that are moving forward and of which the distance and relative velocity is to be measured, but rather primarily stationary objects at the edge of the highway, of which the precise location (distance and angle) is to be measured. In FIG. 1, a vector Vf indicates the movement of the vehicle itself in which the radar sensor is installed, and thus also the movement of the radar sensor itself. The sight line from the radar sensor to object 14 forms an angle PHI with vector Vf (that is, the direction of travel). Because the radar sensor is itself moving, a radial velocity Vr=−Vf cos(PHI) is measured for object 14 even though the latter is stationary. By solving this equation for PHI, it is thus possible to convert the measured radial velocity Vr of object 14 into the directional angle PHI at which the object is seen, and consequently distance/velocity radar image 28 can also be interpreted as distance/angle radar image 30, it being possible to determine whether PHI is positive or negative (object to the right or left side of the vehicle) using the phase differences between the signals received by different antennas 12.

[0035] Because of the apparent change in location of object 14 over the period of a measurement cycle, however, migration effects occur, and these result in distortion of distance/velocity radar image 28. In order to correct this distortion, there is inserted between scan module 20 and FFT module 26 a transform module 32, which performs a transform that corrects the migration effects at the amplitudes A(n, k) captured in scan module 20 by a particular algorithm, for example the Keystone algorithm. Thus, the input data received by FFT module 26 does not directly comprise the amplitudes A(n, k), but transformed (migration-free) amplitudes T(n, k). Moreover, in transform module 32 the Fourier transform that is performed is already in the dimension corresponding to the sequence of pulses that are counted by index k.

[0036] The transform that corrects the migration effects is defined by a set of coefficients c(n, k), which for their part are a function of indices n and k.

[0037] In the case of the Keystone algorithm, for example:

c(n,k)=exp(i(πk.sup.2/N.sub.slow)(1+nB/(fc N.sub.fast)))  (1)

In principle, these coefficients c(n, k) only need to be calculated once for each index pair n, k in order then to be stored in a memory of evaluation part 18. However, memory space is then required for N.sub.fast×N.sub.slow complex coefficients (131,072 in the example shown). If the radar sensor has various operating modes which differ for example in respect of the center frequency fc (for example in order to avoid interference with the radar sensors of other vehicles), then the required memory space proliferates accordingly.

[0038] On the other hand, if each individual coefficient is calculated by the above-indicated formula as needed, then during each measurement cycle a high number of very complex calculations has to be performed, with the result that a computer with a high processing capacity is required.

[0039] However, from the above-indicated formula (1) it is possible to derive the following recursive formula:

[00001]$\begin{array}{cc}\begin{array}{c}c\left(n,k\right)=\exp \left(i\left(\Pi /N_{\mathrm{slow}}\right){\mathrm{Bk}}^2/\left(N_{\mathrm{fast}}\mathrm{fc}\right)\right)c\left(n-1,k\right)\\ =D\left(k\right)c\left(n-1,k\right)\end{array}& \left(2\right)\end{array}$

The constant D(k) only needs to be calculated once for each k and can then be stored. Moreover, if an initial set of coefficients

c(0,k)=exp(iπk.sup.2/N.sub.slow)

is stored, then all the coefficients can be calculated recursively, only a simple multiplication needing to be performed for each increment of n and each value of the index k. The required memory space is significantly reduced, since memory space is now only needed for the N.sub.slow initial values c(0, k) and the constants D(k).

[0040] In this way, a favorable compromise is achieved between required memory and processing capacity, with the result that the overall costs for hardware can be significantly reduced.

[0041] As FIG. 1 shows, evaluation part 18 contains a coefficient module 34 that provides the coefficients c(n, k) for transform module 32. The coefficient module contains a memory 36, in which there is stored, for each of the N.sub.slow values of the index k, an initial value c(0, k) and the associated factor D(k), and a recursion module 38 for recursive calculation of the coefficients c(n, k) for n>0. Once recursion module 38 has calculated a set of coefficients c(n−1, k) and transmitted it to transform module 32, this set is also stored in a register of the recursion module and then used to calculate the next set of coefficients c(n, k).

[0042] Operation of transform module 32 is illustrated in more detail in FIG. 3. The complex amplitudes A(n, k) are functions of time, to be more precise of fast time (index n) and slow time (index k). Mathematically, the chirp z-transform is a convolution of this time-dependent function with a further time-dependent function, which may be expressed as a phase factor exp(iPSI) with a time-dependent phase PSI. Rather than performing this convolution directly by numerical integration, a mathematically equivalent operation is performed which has in its kernel a Fourier transform of both functions into the frequency domain, multiplication of the two frequency-dependent functions, and then inverse transform into the time domain. For this, first the complex amplitudes A(n, k) for each index pair are multiplied by the phase factor exp(iPSI), which is a function of these indices n and k. Then, in an FFT stage 40, the product undergoes a complex Fourier transform cFFT.sub.slow in the “slow” dimension (index k). In parallel with this, in a further FFT stage 42 the phase factor exp(iPSI) also undergoes the same Fourier transform. The functions that are transformed into the frequency domain are multiplied by one another and then, in a further FFT stage 44 with inverse Fourier transform cIFFT.sub.slow, undergo inverse transform back into the time domain. The result is then multiplied once again, for each index pair n, k, by the phase factor exp(iPSI). The product of this provides the transformed amplitudes T(n, k) which are then transmitted, as input data, to FFT module 26 for the Fourier transform in the dimension specified by the index n.

[0043] The procedure shown in FIG. 3 has the advantage that the Fourier transforms in FFT stages 40, 42 and 44 can be performed very fast and efficiently with the aid of specialized hardware. Where necessary, the hardware of FFT module 32 may also be used for this purpose. The other operations that are to be executed in transform module 32 are simple multiplications, which can be carried out substantially faster than numerical convolution operations.

[0044] The multiplications by phase factor exp(iPSI) are performed in transform module 32 such that, first, for a fixed value of index n (for example n=0), the products are calculated for all values of index k, and this is then continued for the next highest value of index n. Coefficient module 34 can then supply the coefficients c(n, k) that are needed for calculation of the phase factors in the order in which they are needed for multiplication by the phase factor. In FFT stage 42, too, in each case integration is carried out with a fixed n over index k. The recursive calculation of the coefficients in coefficient module 34 thus need only be carried out once for each measurement cycle.

[0045] In principle, the calculations in transform module 32 for the different values of n may be carried out in any order. For this reason, it is not mandatory to start the recursion in coefficient module 34 with n=0. For example, it would also be possible to start with a value of n in the vicinity of N.sub.fast/2, and then to continue with two recursive sequences to smaller values of n and larger values of n, as shown schematically in FIG. 4. This has the advantage on the one hand that each of the two recursive sequences is only half as long as the sequence from n=0 to n=N.sub.fast. Because rounding errors, which are unavoidable in the calculation, tend to accumulate as recursion progresses, the fact that the recursive sequences are shortened has the effect of a smaller error accumulation.

[0046] A further advantage of this procedure results from the fact that the transformed amplitudes T(n, k) in FFT module 26 are typically multiplied using a window function, which suppresses the amplitudes at the edges of the relevant time interval (where n=0 and n=N.sub.fast). This windowing serves to mitigate artifacts resulting from the fact that the transform can only be performed over a finite time interval. If, in addition, when calculating the coefficients the recursion progresses from the center to the edges of the time interval, this results in the advantage that the accumulated errors at the edges of the interval are also suppressed by the window function.

[0047] One way of further suppressing error accumulation is for the value range of the indices n to be divided into a plurality of blocks and for the recursion then to be performed in blocks, preferably once again progressing from the center to the edges, as a result of which the recursive sequences are further shortened.

[0048] FIG. 5 shows an example of a recursion scheme in which the value range for the index n (0-255) is divided into two blocks, which extend from 0-128 and from 129-255. In this case, two sets of initial coefficients c(96, k) and c(160, k) are stored in memory 36. The recursion then progresses from these initial values to the edges of the respective block. However, the initial values for n=96 and n=160 do not lie in the center of the respective block but are displaced toward the center of the entire value range. For this reason, the recursive sequences, which progress to n=128 and n=129 respectively, are shorter than the recursive sequences that progress to the outer edges n=0 and n=255. Thus, there is a smaller error accumulation in the center of the value range than at the outer edges, where the errors are additionally also suppressed by the windowing.

[0049] Moreover, the error may also be reduced in that, instead of the one set of constants D(k), a plurality of sets are stored, which are previously calculated exactly for different step sizes—that is to say different increments of the index n. For example, a set D_1(k) representing an increment of length 1 and an additional set D_10(k) representing increments of length 10 may be used, with the result that every tenth iteration can be calculated using D_10(k) and the iterations in between can be calculated using D_1(k). This also reduces the number of iterations by a factor of 10. The number of sets D_i(k) used, and their gradation, may in this case be selected as desired, and depends on the need for accuracy, the numerical representation and the available memory.