Radar system with monitoring of the frequency position of a sequence of similar transmission signals

Abstract

Method for a radar system for detecting the environment of a motor vehicle and implementing a function for driver assistance and / or autonomous driving maneuvers, comprising the following steps - Generating a frequency modulation using a controllable or adjustable oscillator, - Generating a sequence of K0 (K0>1) transmitted signals modulated in the transmission frequency, each having the same target frequency profile, possibly apart from a variation in the frequency position, i.e., in particular a variation in the initial and thus equivalently the center frequency, - Transmission of signals using transmitters, - Receiving transmitted signals reflected from objects using receiving devices, - Analyzing the frequency position of the transmitted signals and - Evaluation of the received signals, in particular for the detection of objects, using signal processing means, wherein the actual frequency response across the K0 transmitted signals, i.e., in particular the actual response of the initial, middle, or mid-frequency of the transmitted signals, or its deviation from the target response caused, for example, by frequency instability or frequency drift, is determined absolutely or relatively, i.e., up to an indeterminate constant component, wherein - a time-discrete signal is used for each transmission signal, which contains information about the frequency profile of the transmission signal and which is preferably generated by sampling an analog signal or by reading a free-running counter at predetermined times, - these time-discrete signals via the K0 transmission signals are unnormalized with respect to the position of their phase and / or their initial value, and - during an evaluation of these time-discrete signals, explicit normalization takes place or implicitly an influence of the position of their phase and / or their initial value is eliminated, and depending on an actual curve determined in this way and / or a deviation of the actual curve from the target curve of the frequency position determined in this way and / or a quality measure derived therefrom, a correction is made in the control of the oscillator and / or a correction in the evaluation of the received signals and / or an adjustment of the function for driver assistance and / or autonomous driving maneuvers up to and including its deactivation, whereby - the frequency of the emitted signals is linearly modulated, - the signals received by reflection of the transmitted signals from objects are converted into the low-frequency range by mixing with a signal whose frequency corresponds to the current transmission frequency or differs from it by a constant factor, - the low-frequency received signals are sampled NA times in an equidistant grid, - a first spectral analysis, particularly in the form of a discrete Fourier transform, is performed on each of these NA sample values ​​in order to enable, in particular, distance measurement of the objects and separation for the simultaneous detection of multiple objects, and - at each frequency reference point of the first spectral analysis, a second spectral analysis is carried out, in particular using the discrete Fourier transform, which extends over all or a subset of the values ​​occurring at the respective frequency reference point during the K0 transmission signals, in order to be able to realize, in particular, a relative velocity measurement of the objects and a separation for the simultaneous detection of several objects, and in the procedure (a) - from the determined actual frequency position across the K0 transmitted signals, the expected spectrum of the second spectral analysis of a target at a distance r with a relative velocity of zero is determined, - the resulting spectrum or its magnitude can be used directly to derive a measure of quality and / or - a quality factor is derived from the deviation of the spectrum from the spectrum expected at the target frequency position of a target at distance r with a relative velocity of zero, whereby the same window function is used for both spectral analyses and the resulting spectra are referenced to the same level for comparison and / or (b) a deviation from the actual to the target frequency response is corrected directly or after regression by multiplication with a complex unit vector between the first and second spectral analyses, wherein the phase of the complex unit vector is proportional to this deviation and proportional to the frequency of the respective frequency reference point of the first spectral analysis.

Description

[0001] The invention relates to a radar system for use in driver assistance systems in motor vehicles. According to the invention, the radar system comprises means and methods for analyzing and monitoring the frequency profile across a sequence of identical transmitted signals. If the frequency profile deviates too significantly from its target profile across the sequence of transmitted signals, either corrective measures are applied or the driver assistance system is partially or completely deactivated. State of the art

[0002] Motor vehicles are increasingly equipped with driver assistance systems that use sensors to perceive the surroundings and derive automatic vehicle reactions from the detected traffic situation and / or instruct the driver, particularly by issuing warnings. A distinction is made between comfort and safety functions.

[0003] In current development, FSRA (Full Speed ​​Range Adaptive Cruise Control) plays the most important role as a comfort feature. The vehicle regulates its own speed to the driver's desired speed, provided the traffic situation allows it; otherwise, the vehicle's speed is automatically adjusted to the traffic situation.

[0004] Safety features now come in a wide variety of forms. One group consists of functions designed to reduce braking or stopping distance in emergency situations; the spectrum of corresponding driver assistance functions ranges from automatic brake prefill to reduce braking latency (prefill), to enhanced brake assist (BAS+), and even autonomous emergency braking. Another group comprises lane change functions: these warn the driver or intervene in the steering if the driver intends to make a dangerous lane change, i.e., if a vehicle in the adjacent lane is either in the blind spot (referred to as BSD - "Blind Spot Detection") or is approaching rapidly from behind (LCA - "Lane Change Assist").

[0005] In the foreseeable future, however, the driver will no longer just be assisted, but the driver's task will increasingly be performed autonomously by the vehicle itself, i.e., the driver will be increasingly replaced; this is called autonomous driving.

[0006] Radar sensors are used for systems of the type described above, often in combination with sensors from other technologies, such as camera sensors. Radar sensors have the advantage of operating reliably even in poor weather conditions and, in addition to measuring the distance to objects, can also directly measure their radial relative velocity via the Doppler effect. Transmission frequencies of 24 GHz, 77 GHz, and 79 GHz are used.

[0007] As the functionality and intervention of such systems increase, the reliability requirements for the sensors are constantly rising. The radar image must correspond to reality, meaning that object sizes, especially distance, relative speed, and angle, must be correct, no objects should be missed, and no so-called ghost objects, which do not exist in reality, should be reported.

[0008] A key element of radar sensors is the modulation of the transmission frequency, enabling the measurement of distance and relative speed. The most common type of modulation is frequency modulation, specifically a linear change in frequency, often using a sequence of similar linearly modulated transmission signals. Incorrect frequency positions (e.g., center frequency) of the individual transmission signals (e.g., due to failure or malfunction of individual circuit components, or fundamental frequency instability or drift) can lead to the errors described above, such as incorrectly measured object sizes, undetected objects, and ghost objects. This could cause the driver assistance functions implemented with the radar system to malfunction; for example, in an emergency braking assistant, it could result in a malfunctioning emergency braking system.Ghost objects could trigger an unauthorized emergency braking maneuver, potentially leading to a rear-end collision with a following vehicle, with serious consequences including fatalities.

[0009] German patent application DE 10 2016 214 808 A1 describes an arrangement and a method for analyzing the frequency response over a sequence of identical transmission signals. This method is based on a period counter that is initialized at the beginning of each transmission signal (i.e., not free-running) and whose values ​​are read out at the end of each transmission signal. If the frequency response across the transmission signals deviates too significantly from the target response, either corrective measures are applied or the driver assistance system is partially or completely deactivated. The proposed approach must take into account that counters are difficult or expensive to implement at high frequencies (especially the precise timing of the readings) and have high power consumption (partly due to the relatively long counter length). To reduce the counter's frequency, either frequency division or mixing can be used.However, the upstream division has the disadvantage that the accuracy of the frequency analysis decreases with increasing division factor; and the upstream mixing requires the generation of another signal in the high-frequency range and thus significant effort.

[0010] Furthermore, DE 10 2015 103 149 A1 discloses a radar device for transmitting a signal in a frequency band, comprising a control device and an oscillator, in which an input of the oscillator is connected to the control device via a converter, the oscillator can be controlled by the control device to generate the signal, and the signal generated by the oscillator can be accessed at an output of the oscillator. At least one transmitting antenna is provided for transmitting the signal present at the output of the oscillator, the transmitting antenna being connected to the output of the oscillator.Furthermore, at least one receiving channel is provided for receiving a received signal, for processing the received signal and for forwarding the processed received signal to the control device, wherein the receiving channel has at least one receiving antenna and a mixer for mixing the received signal with the signal present at the output of the oscillator, wherein the mixer is connected to the output of the oscillator.

[0011] DE 10 2014 214 498 A1 discloses a radar system for environmental detection for a motor vehicle, in which negative effects of a received signal, which arises from strong reflection from the immediate vehicle-side environment of the radar system or from reflections and / or couplings within the radar system, are avoided or at least reduced.

[0012] Furthermore, F. Müller et al. describe a method for evaluating the linearity of an RF ramp in "A Method for the Analysis of Ramp-Inherent Linearity Distortions in Automotive Radar Applications" (German Microwave Conference GeMiC 2015; pp. 217-220), which is based on the analysis of the baseband signal. This simplifies the analysis, as the baseband frequency is only a few MHz and depends on chirp modulation parameters and a target distance. The instantaneous frequency is measured, so that deviations from the ideal slope can be detected immediately. Problem, solution and advantages of the invention

[0013] The object of the invention is to propose arrangements and methods for analyzing the frequency position of a sequence of similar transmitted signals in a radar system, which differ from the approach presented in DE 10 2016 214 808 A1 and may be, or are, more advantageous due to technological constraints and implementation aspects. In particular, methods are to be proposed which can either operate with large frequency divider factors or do not require counter reinitialization.

[0014] This task is generally solved using a method or radar system according to claims 1-11.

[0015] The advantages of the invention arise from the fact that it enables monitoring of the frequency position of the radar system in order to either apply corrective measures or to partially or completely deactivate the driver assistance system if the frequency position deviates too much from the target position, thus preventing malfunctions of the driver assistance system.

[0016] According to the invention, a method for a radar system for detecting the environment of a motor vehicle and implementing a function for driver assistance and / or autonomous driving maneuvers comprises the following steps: generating a frequency modulation by means of a controllable or adjustable oscillator; generating a sequence of K0 (K0>1) transmitted signals modulated in the transmission frequency, each having the same target frequency profile, optionally apart from a variation in the frequency position, i.e., in particular a variation in the initial and thus equivalently the center frequency; emitting transmitted signals by means of a transmitter; receiving transmitted signals reflected from objects by means of a receiver; analyzing the frequency position of the transmitted signals and evaluating the received signals, in particular for the detection of objects, by means of signal processing means, wherein the actual frequency profile resulting across the K0 transmitted signals isIn particular, the actual behavior of the initial, mid, or average frequency of the transmitted signals, or its deviation from the target behavior caused, for example, by frequency instability or frequency drift, is determined absolutely or relatively, i.e., down to an indeterminate constant component, wherein a time-discrete signal is used for each transmitted signal, which contains information about the frequency behavior of the transmitted signal and which is preferably generated at predetermined times by sampling an analog signal or by reading a free-running counter, these time-discrete signals are unnormalized with respect to the position of their phase and / or their initial value via the K0 transmitted signals, and during an evaluation of these time-discrete signals, an explicit normalization takes place or an implicit influence of the position of their phase and / or their initial value is eliminated.and depending on the actual curve determined in this way and / or a deviation of the actual curve from the target curve of the frequency position determined in this way and / or a quality measure derived therefrom, a correction in the control of the oscillator and / or a correction in the evaluation of the received signals and / or an adjustment of the function for driver assistance and / or autonomous driving maneuvers up to and including its deactivation.

[0017] Preferably, the method for a radar system can further be designed as follows: a signal is used for analysis which is frequency-reduced compared to the transmitted signal by frequency division by a factor of T>1 and / or by mixing; these low-frequency signals resulting from the transmitted signals are sampled, optionally after filtering; a complex value is determined from these sampled signals resulting from the transmitted signals, optionally after further filtering, in one or more time intervals per transmitted signal, in particular by correcting samples of the respective interval at least approximately by the expected phase shift resulting from the target frequency profile of the transmitted signals there, and accumulating these phase-corrected samples per interval.wherein the time interval(s) over the K0 transmitted signals preferably have at least approximately the same position, and the actual frequency response over the K0 transmitted signals or its deviation from the target response is characterized using the complex values ​​thus determined, in particular based on the phase of these values, which is why they are also referred to as phase characteristics in the following.

[0018] In the method, it is further preferably possible for a radar system in which the sampled signals are real-valued and are converted at least approximately into their analytical signals, i.e., the complex-valued signals with only the positive or negative frequency components, in the time intervals under consideration, wherein first-degree Hilbert filters with the zero at approximately the negative or positive of the respective center frequency of these frequency-modulated signal sections are preferably used.

[0019] In an advantageous embodiment, the method for a radar system can realize a scalar product between the samples and a preferably pre-calculated parameter vector by means of Hilbert filtering and the accumulation of the phase-corrected sample values ​​over the respective time period.

[0020] Preferably, in the method for a radar system, the phase difference of pairs of phase characteristics can be calculated, wherein the two values ​​of each pair belong to time periods with similar positions over the transmitted signals and there is at least approximately the same time interval and the same frequency profile between the two time periods, and from these phase differences or their change over the transmitted signals the profile of the frequency position is determined absolutely or relatively, i.e., up to an indeterminate constant component.

[0021] In the method for a radar system, the pairs of phase characteristics can preferably be formed from the same transmitted signal, wherein preferably one value is formed from a time period in the front area of ​​the transmitted signals and the second value from a time period in the rear area of ​​the transmitted signals.

[0022] In a further embodiment of the invention, in the method for a radar system, the pairs of phase characteristics can each be formed from different transmitted signals, for example from successive transmitted signals, wherein both values ​​belong to time periods of similar position over the transmitted signals.

[0023] According to the invention, the frequency of the radiated transmission signals is linearly modulated, the signals received by reflection of the transmission signals from objects are converted into the low-frequency range by mixing with a signal whose frequency corresponds to the current transmission frequency or differs from it by a constant factor, and the low-frequency received signals are distributed in an equidistant grid N A be scanned - times, via these N AA first spectral analysis, particularly in the form of a discrete Fourier transform, is performed on each sampled value to enable, in particular, distance measurement of the objects and separation for the simultaneous detection of multiple objects. At each frequency reference point of the first spectral analysis, a second spectral analysis is performed, particularly using the discrete Fourier transform, which extends over all or a subset of the values ​​occurring at the respective frequency reference point during the K0 transmission signals, in order to enable, in particular, a relative velocity measurement of the objects and separation for the simultaneous detection of multiple objects.

[0024] Preferably, in the method for a radar system, the expected spectrum of the second spectral analysis of a target at a distance r with a relative velocity of zero can be determined from the determined actual frequency position across the K0 transmitted signals, the resulting spectrum or its magnitude can be used directly to derive a quality measure, and / or a quality measure can be derived from the deviation of the spectrum from the spectrum expected at the target frequency position of a target at a distance r with a relative velocity of zero, wherein the same window function is used for both spectral analyses and the resulting spectra are referenced to the same level for comparison.

[0025] In the method for a radar system, it is further preferably possible to derive from a quality measure whether a detection could have been caused by a deviation of the actual frequency position across the K0 transmitted signals from the target frequency position from another detected object, and this detection can then be completely discarded or marked as a potential false detection.

[0026] In an advantageous embodiment of the method for a radar system, a regression can be performed on the difference between the target and actual frequency response, and the parameters of the compensation function and / or the deviation, in particular the standard deviation between the actual response and the compensation function, can be used as a measure of quality.

[0027] Preferably, in the method for a radar system, a deviation from the actual to the target frequency profile can be corrected directly or after performing a regression by multiplication with a complex unit vector between the first and second spectral analysis, wherein the phase of the complex unit vector is proportional to this deviation and proportional to the frequency of the respective frequency reference point of the first spectral analysis.

[0028] Furthermore, preferably in the method for a radar system, the center frequency of the transmitted signals or their deviation from the target center frequency can be determined and used for calculating the relative velocity and / or the angular position of objects.

[0029] According to the invention, a radar system is configured to carry out a method according to a previous preferred embodiment. Brief description of the drawings In Fig. Figure 1 shows an exemplary embodiment of a radar system. Fig. Figure 2 shows the frequency of the transmit and receive signals, which consists of so-called frequency ramps, as well as the respective antenna combinations used, consisting of transmit and receive antennas. Fig. Figure 3 shows a sampled signal in the presence of two objects before the first DFT (left) and after the first DFT (right). In Fig. 4 represents the complex spectral value rotating over the frequency ramps in distance gate 4, in which exactly one object is located. Fig. Figure 5 schematically shows the two-dimensional complex-valued spectrum e(j,l,m) after the second DFT for an antenna combination m. Fig. Figure 6 shows the different path lengths between the individual antennas and a distant object stationary relative to the sensor at an azimuth angle α. Az < 0. Fig. Figure 7a shows an antenna arrangement with one transmitting and eight receiving antennas, which corresponds to the antenna arrangement under consideration according to Fig. 1 is equivalent to 2 transmitting and 4 receiving antennas; in Fig. Figure 7b shows the different path lengths between the individual antennas and a distant object stationary relative to the sensor for this equivalent arrangement. Fig. Figure 8a shows the complex spectral value rotating over the antenna combinations in the distance-relative velocity gate (9,0) for the above antenna arrangements, in which exactly one object (stationary relative to the sensor) is located; in Fig. Figure 8b shows the magnitude of the corresponding spectrum according to the third DFT. Fig. Figure 9 schematically shows the data arrangement before the three-dimensional DFT (left) and the three-dimensional complex-valued spectrum w(j,l,n) afterwards (right). Fig. Figure 10 shows a section of the oscillator signal divided down by a factor of T=2048 (solid line), the sinusoidal signal resulting after filtering (dashed line) and its values ​​resulting after sampling (shown with dots). Fig. Figure 11 shows the frequency position error Δf E (k,0) for an actual curve that is curved compared to the target curve of the frequency position and also has a periodic disturbance. Fig. Figure 12a shows, in logarithmic representation (i.e., in dB), the magnitude of the velocity spectrum v(99,I,0) (solid line) for the actual frequency position and the magnitude of the velocity spectrum that results from the target frequency profile for a target at the same distance r=99m and with the same amplitude (dashed curve); in Fig. Figure 12b shows the relative difference between these two velocity spectra. Examples of implementation

[0030] The exemplary design of a radar system, which is located in Fig. Figure 1 is a rough representation. The radar system has two transmitting antennas, TX0 and TX1, for emitting signals and four receiving antennas, RX0-RX3, for receiving signals reflected from objects. The antennas are implemented as patch antennas on a planar circuit board, 1.1, which is oriented horizontally and vertically within the vehicle as shown in the figure. All antennas (transmitting and receiving) have the same beam pattern in elevation and azimuth. The four receiving antennas (and thus their phase centers, i.e., radiation centers) are each spaced at the same lateral, i.e., horizontal, distance of d = λ / 2 = 6.2 mm from each other, where λ = c / 24.15 GHz = 12.4 mm is the mean wavelength of the emitted signals. The horizontal distance between the two transmitting antennas is four times greater, i.e., 4d = 2λ.

[0031] Multiplexers 1.3 and 1.4 allow the selection of one of the two transmitting antennas and one of the four receiving antennas.

[0032] The transmission signals emitted by the respective selected transmitting antenna are derived from the high-frequency oscillator 1.2 in the 24 GHz range, which is controlled by a control voltage v steuerThe oscillator's frequency can be varied. The control voltage is generated in the control circuitry 1.9, which includes, for example, a phase-locked loop or a digital-to-analog converter. These components are controlled so that the oscillator's frequency response corresponds at least approximately to the desired frequency modulation. To analyze the oscillator frequency, it is reduced in circuit block 1.11 (by division and / or mixing), and the resulting signal is digitized in circuit block 1.12. Two different approaches to digitization are discussed below: an analog-to-digital converter and a counter. The evaluation of this digital information about the oscillator frequency takes place in the digital signal processing unit 1.10.

[0033] The signals received by the selected receiving antenna are also mixed down to the low-frequency range in the real-valued mixer 1.5 along with the signal from the oscillator 1.2. The received signals then pass through a bandpass filter 1.6 with the transfer function shown, an amplifier 1.7, and an analog-to-digital converter 1.8; subsequently, they are further processed in the digital signal processing unit 1.10.

[0034] In order to measure the distance to objects, - as in Fig. Figure 2 shows the frequency of the high-frequency oscillator, and thus the transmitted signals, changing linearly very rapidly (by 187.5 MHz in 8 µs, with a center frequency of 24.15 GHz); this is called a frequency ramp. The frequency ramps are repeated periodically (every 10 µs); there are a total of 2048 frequency ramps, all with the same target frequency profile. The 8 combinations of the 2 transmit and 4 receive antennas are periodically repeated via the frequency ramps in the sequence TX0 / RX0, TX0 / RX1, TX0 / RX2, TX0 / RX3, TX1 / RX0, TX1 / RX1, TX1 / RX2, and TX1 / RX3, with the next combination being selected before each frequency ramp. Fig. 2 is k the sweep variable over the 2048 / 8 = 256 frequency ramps for each antenna combination and m = 4.m TX +m RX the running variable across the 8 antenna combinations TXm TX / RXm RX .

[0035] The received signal from a single point-like object, after mixing and therefore also at the A / D converter, is a sinusoidal oscillation for each frequency ramp and each of the 8 antenna combinations; this can be visualized using... Fig. 2. Explain as follows: If the object has a radial relative velocity of zero to the radar system, the frequency difference Δf between the transmitted and received signals is constant and proportional to the signal propagation time Δt, and thus proportional to the radial distance r = c·Δt / 2, where c is the speed of light and the factor 1 / 2 accounts for the fact that the propagation time Δt refers to the round trip of the wave; the frequency difference Δf, according to the above interpretation, is Δf = 2r / c·187.5MHz / 8µs = r·156.250kHz / m. Since the received signal is mixed with the oscillator frequency, and thus the transmit frequency, a sinusoidal oscillation with frequency Δf results after the mixer.This frequency lies in the MHz range and, for a non-zero radial relative velocity, is further shifted by the Doppler frequency, which, however, lies only in the kHz range and is therefore approximately negligible compared to the frequency component due to the object distance. If there are multiple objects, the received signal is a superposition of several sinusoidal oscillations of different frequencies.

[0036] During each frequency ramp, the received signal is sampled 256 times at the A / D converter at intervals of 25 ns (i.e., at 40 MHz) (see Fig. 2), where the sampling always begins at the same time relative to the start of the ramp. As from Fig. As can be seen in section 2, signal sampling only makes sense in the time range where received signals from objects in the distance range of interest arrive - after ramp start, at least the propagation delay corresponding to the maximum distance of interest must be waited for (at a maximum distance of interest of 99m, this corresponds to 0.66µs); it should be noted that here and in the following, distance always refers to the radial distance.

[0037] Then, a discrete Fourier transform (DFT) in the form of a fast Fourier transform (FFT) is generated from the 256 samples of each frequency ramp. This allows objects at different distances, which result in different frequencies, to be separated (see Fig. 3; left: signal s(i,k,m) before DFT in the presence of two objects, right: magnitude |e(j,k,m)| of the DFT result; where k is the sweep variable over the frequency ramps per antenna combination and m is the sweep variable over the 8 antenna combinations TXm TX / RXm RX Each of the discrete frequency reference points j of the DFT corresponds to a distance r and can therefore, analogously to pulse radars, also be called a distance gate; in the above configuration, the distance gates have a spacing and thus a width Δr of exactly one meter (resulting from Δr·156.250kHz / m = 1 / (6.4µs)). Power peaks occur in the DFT at the distance gates where objects are located. Since the sampled received signals are real-valued (then no additional information in the upper half of the DFT, as it is symmetrical) and the upper transition region of the analog bandpass filter 1.6 according to Fig. Since filter 1 has a frequency bandwidth of 8.75 MHz (corresponding to the range of 56 frequency support points), only 100 of the 256 discrete frequency support points can be processed further (it should be noted that arbitrarily narrow filter transition ranges are not feasible). Filter 1.6 attenuates low frequencies and thus the received signals from nearby objects to prevent overloading amplifier 1.7 and A / D converter 1.8 (the signals received at the antennas become stronger with decreasing distance to the object).

[0038] Over the 256 frequency ramps (k = 0, 1, ..., 255), complex spectral values ​​e(j, k, m) are generated in each of the 8 antenna combinations m (m = 0, 1, ..., 7) for each distance gate j (i.e., each of the 100 considered frequency reference points). If there is exactly one object at the distance corresponding to a distance gate, the complex spectral value in this distance gate j rotates across the 256 frequency ramps of each of the 8 antenna combinations m = 0, 1, ..., 7 with the Doppler frequency, since the distance (in the millimeter range or below) and thus the phase of the associated oscillation changes uniformly from frequency ramp to frequency ramp (see Fig. 4; the phase change of 45° per frequency ramp shown there corresponds to a decrease in the object's distance of λ / (8·2) = 0.78 mm, where the wavelength λ = c / 24.15 GHz = 12.4 mm and the factor 2 in the denominator takes into account the back-and-forth propagation of the waves, from which the radial relative velocity v is calculated. rel= 0.78 mm / 80 µs = 35 km / h (a positive sign of the radial relative velocity is defined as distance). Several objects with different radial relative velocities at the same distance gate are separated by calculating a second DFT for each antenna combination and each distance gate using the complex spectral values ​​generated in the 256 frequency ramps. Each discrete frequency point I of this second DFT corresponds to a set of Doppler frequencies (because of the sampling of the Doppler frequency, it can only be determined to an unknown integer multiple of its sampling frequency) and thus to a set of radial relative velocities v. relof objects, such that the discrete frequency support points of the second DFT can be called relative velocity ports; for the radial relative velocity, the term "radial" is omitted from this point onward for the sake of linguistic simplicity. The uniqueness range of the relative velocity is given by 2·v rel,EB ·80µs=12.4mm to v rel,EB = 280 km / h. Therefore, the relative velocities v are assigned to the relative velocity gate I. rel =(l / 256+p).280km / h assigned, where p is an integer.

[0039] The second DFT not only serves to determine the relative velocity, but also increases the detection sensitivity through its integration - by about 10 log for 256 frequency ramps. 10 (256) = 24dB.

[0040] According to this second DFT for the relative velocities, a two-dimensional complex-valued spectrum v(j,l,m) is obtained for each antenna combination m, where the individual cells can be referred to as distance-relative velocity gates and power peaks occur at the respective associated distance-relative velocity gate due to objects (see Fig. 5).

[0041] Finally, the information from the eight antenna combinations is fused. The waves originating from the two transmitting antennas and reflected from a single point-like object arrive at the four receiving antennas depending on the azimuth angle α. Az with different phase relationships to each other, since the distances between the object and the transmitting and receiving antennas are slightly different. This will now be explained in more detail, whereby the object under consideration is initially assumed to be at rest relative to the sensor, i.e., it has a relative velocity of zero. In Fig. Figure 6 shows, in vertical projection, the phase centers of the antennas and the beam paths to a distant object stationary relative to the sensor at azimuth angle α. Az < 0 (positive α) Az (meaning right of the solder surface to the board plane) and elevation angle α EI = 0 (in the horizontal perpendicular plane to the circuit board plane); the object is so far away that the beam paths can be assumed to be parallel, i.e., the object is located in the far field of the antenna arrangement. The path length r(m) for the antenna combination m = 4·m TX +m RX from the transmitting antenna TXm TX to the object and back to the receiving antenna RXm RX This results in r(m)=2⋅rRP+sin(−αAz)⋅(a+mTX⋅4d+a+d / 2+mRX⋅d)=2⋅rRP+sin(−αAz)⋅(2a+d / 2+m⋅d), where r RPLet be the path length from a reference point RP on the antenna board to the object, and be the horizontal distance between the reference point and the transmitting antenna TX0. From this relationship, it can be seen that the distance changes linearly with the number m of the antenna combination. The quantity (2a+d / 2+md) represents the horizontal distance of the so-called relative phase center of the antenna combination m to the reference point RP and is the sum of the horizontal distances of the associated transmitting and receiving antennas to the reference point (the relative phase center of a combination of a transmitting and a receiving antenna is defined here as the sum of the two vectors from a reference point to the phase centers of the transmitting and receiving antennas).

[0042] The phase difference φ(m)-φ(0) between the received waves for the antenna combination m=0,1,...,7 and the antenna combination m=0 results from the different path lengths r(m) as follows: φ(m)−φ(0)=−2π / λ⋅[r(m)−r(0)]=−2π / λ⋅[2⋅rRP+sin(−αAz)⋅(2a+d / 2+m⋅d)−2 ⋅rRP−sin(−αAz)⋅(2a+d / 2+0⋅d)]=−2π / λ⋅sin(−αAz)⋅d⋅m=2π / λ⋅sin(αAz)⋅d⋅m and thus also changes linearly with the number m of the antenna combination. The amplitude of the signals received by the different antenna combinations is constant, since all antennas have the same beamforming characteristic and the distance of the antennas to the distant object differs only negligibly for level analysis.

[0043] As is immediately apparent, the following results for the in Fig. 7a Antenna arrangement shown with vertical projection according to Fig. 7b exactly the same relationships for the path length r(m) and the phase difference φ(m)-φ(0) as for the arrangement considered so far according to Fig. 1; the arrangement according to Fig. 7a has only one transmitting antenna TX0 and 8 equidistant receiving antennas RX0-RX7, where the antenna combination m = m RXnow from the transmitting antenna and the receiving antenna RXm RX is formed. Due to identical individual antennas and identical phase relationships between the antenna combinations, both antenna arrangements are equivalent with respect to angle measurement capability. The arrangement presented here according to Fig. However, option 1 has the advantage that it requires almost only half the horizontal extent compared to the conventional arrangement according to Fig. 7a, which allows the sensor size to be significantly reduced.

[0044] The phase differences φ(m)-φ(0), which increase or decrease linearly across the 8 antenna combinations m and depend on the azimuth angle, remain constant until after the second DFT, apart from any constant and therefore compensable phase shifts (e.g., due to different line lengths). Thus, if there is only one object in a distance-relative velocity gate (j,l), the complex spectral value v(j,l,m) at that location rotates across the 8 antenna combinations m = 0,1,...,7 with a constant rotational speed that depends on the azimuth angle (see example). Fig. 8a). Therefore, digital beamforming for the azimuth direction can be performed in each distance-relative velocity gate. This involves summing the complex values ​​of the eight antenna combinations, each of which is multiplied by a set of complex factors with linearly changing phase. Depending on the linear phase change of the respective factor set, radiation lobes with different beam directions result. The beamwidth of these radiation lobes is significantly smaller than that of the individual antennas. The summation described above is implemented using a 16-point DFT, where the eight values ​​of the eight antenna combinations are supplemented by eight zeros. The discrete frequency values ​​n = 0,1,...,15 of this DFT correspond to different phase differences Δφ = φ(m)-φ(m-1) = 2π·mods(n,16) / 16 between adjacent antenna combinations (mods(n,16) denotes the symmetric modulo, i.e. the mapping to the range -8...+8) and thus to different azimuth angles α. Az = arcsin(Δφ·λ / (2πd)) = arcsin(mods(n,16)·λ / (16d)) and can therefore be called angle gates. In Fig. 8b is the absolute value of the w(j,l,n) course of the spectrum of the third DFT for the conditions according to Fig. Figure 8a shows a point-like object at the azimuth angle α Az = 14.5° (referring to the depicted phase difference between adjacent antenna combinations of 45°, which corresponds to π / 4, n = 2 and for d = λ / 2 the azimuth angle α Az = arcsin(π / 4) = 14.5°). The third DFT not only serves to determine the azimuth angle, but its integration also increases the detection sensitivity – by approximately 10·log for 8 antenna combinations. 10 (8) = 9dB.

[0045] Previously, determining the azimuth angle assumed the object had a relative velocity of zero. If this is not the case, the phase between the received signals from the two transmitting antennas, activated 40 µs apart, changes proportionally to the relative velocity (assumed to be constant in the following), because the distance changes slightly during this period. Since each third DFT corresponds to a distance-relative velocity gate and thus to a specific relative velocity, the linear phase shift generated by the relative velocity across the eight antenna combinations can be compensated either before or after the third DFT. Compensation before the DFT requires shifting the phase of the complex input values; compensation after the DFT requires shifting the discrete frequency values ​​n corresponding to the output values.Due to the ambiguities for the relative velocity explained above, this compensation leads to different azimuth angles depending on the hypothesis used for the ambiguous relative velocity.

[0046] After this third DFT for the azimuth angles (including the compensation of the linear phase change generated by the relative velocity via the antenna combinations), a three-dimensional complex-valued spectrum w(j,l,n) is obtained, where the individual cells can be referred to as distance-relative velocity-angle gates and power peaks occur at the respective distance-relative velocity-angle gate due to objects (see Fig. 9; left: data arrangement before three-dimensional DFT, right: afterwards).

[0047] By determining the power peaks, one can detect objects and their dimensions: distance, relative velocity (apart from possible ambiguities, e.g.) and azimuth angle (each ambiguity hypothesis of the relative velocity corresponds to a value, see Fig. 9) determine. Since power peaks caused by the DFT windowing also exhibit levels in neighboring cells, the object dimensions can be determined much more accurately than the gate widths by interpolation based on these levels. It should be noted that the window functions of the three DFTs are chosen so that, on the one hand, the power peaks do not become too wide (for sufficient object separation), but on the other hand, the sidelobes of the window spectra do not become too high (in order to detect weakly reflecting objects in the presence of strongly reflecting objects). From the height of the power peaks, the object's reflection cross-section can be estimated as a fourth object dimension, indicating how strongly the object reflects the radar waves. Due to the noise present in every system (e.g.,Even without received object reflections, a certain power level is obtained according to the three-dimensional DFT (due to thermal noise); this noise level, which varies to some extent due to statistical effects, represents the lower physical limit of the detection capability. The detection threshold, above which objects are formed from power peaks, is set approximately 12 dB above the mean noise.

[0048] Up to this point, primarily point-like objects (i.e., neither extended in width nor length) with constant radial relative velocity and no lateral motion have been considered. In these cases, the power peaks after the three-dimensional Fourier transform are "sharp"; their shape corresponds to the three-dimensional discrete Fourier transform of the window functions shifted to the position of the three object dimensions: velocity, distance, and angle. With respect to each of these dimensions, the shape of the power peaks is the one-dimensional discrete Fourier transform of the respective window function shifted to the respective object size. Objects for which the above conditions do not apply exhibit "blurred" power peaks after the three-dimensional Fourier transform.

[0049] The described detection of objects and the determination of their corresponding dimensions constitute a measurement cycle and provide a snapshot of the environment; this is repeated cyclically approximately every 40 ms. To assess the environmental situation, the snapshots are tracked, filtered, and evaluated across successive cycles; the reasons for this include, in particular: • Some quantities cannot be determined directly in one cycle, but only from the change over successive cycles (e.g. longitudinal acceleration and lateral velocity), • The movement of objects can be validated over several cycles, resulting in a more robust and reliable description of the environment; for example, the change in (radial) distance over successive cycles must match the measured (radial) relative velocity, which provides redundancy and thus additional safety in the description of the environment. • Reduction of measurement noise through time-based filtering over multiple cycles.

[0050] Tracking and filtering object detections across successive cycles is also known as tracking. For each object, values ​​for the next cycle are predicted based on the tracked object dimensions of the current cycle. These predictions are then compared with the objects and their dimensions detected as a snapshot in the next cycle to match them appropriately. The predicted and measured object dimensions belonging to the same object are then fused, resulting in the current tracked object dimensions, which thus represent values ​​filtered across successive cycles. If certain object dimensions cannot be uniquely determined in a cycle, the different hypotheses must be considered during tracking.The environmental situation for the respective driver assistance function is analyzed and interpreted from the tracked objects and the associated tracked object dimensions in order to derive the corresponding actions.

[0051] It was explained above that for point-like objects with constant radial relative velocity and no lateral motion, sharp power peaks result after the three-dimensional Fourier transform. The shape of these peaks corresponds to the three-dimensional discrete Fourier transform of the window functions, shifted to the position of the three object dimensions: velocity, distance, and angle. With respect to each of these dimensions, the shape of the power peaks is the one-dimensional discrete Fourier transform of the respective window function, shifted to the respective object size. However, this only applies to an ideal circuit, especially ideal antennas and ideal frequency modulation. In reality, frequency modulation will never be perfect, e.g.,This is caused by physical noise effects such as thermal noise or thermal or electrical transients, or by the finite accuracy in digital circuits and digital-to-analog conversions due to quantization (e.g., the finite resolution of digital-to-analog converters for directly generating the oscillator control voltage or for the control signal of a phase-locked loop, i.e., a so-called PLL). In addition to such inherent errors caused by non-ideal circuits, significantly larger errors in frequency modulation can occur due to the failure or malfunction of individual circuit components.

[0052] The following discussion focuses solely on frequency modulation errors that result in a frequency deviation from the target value in the transmitted signals, where the frequency of a transmitted signal is characterized in particular by its center frequency. Thus, the focus is not on the linearity error of the frequency modulation (i.e., a deviation of the actual transmit frequency response within the transmitted signals from a linear regression with the target slope), but rather on whether and how the frequency ramps are offset from one another. Such an offset can be caused by slow frequency changes, for example, by low-frequency disturbances in the supply voltage.

[0053] Such a faulty frequency distribution (e.g., a varying center frequency with a constant target center frequency) leads to a blurring or fragmentation of power peaks in the relative velocity dimension, even for point-like objects. This can result in inaccurately measured relative velocity, obscuration of smaller objects by larger ones, and the generation of ghost objects. Consequently, the driver assistance function implemented with the radar system could malfunction; for example, in the case of an emergency braking assistant, ghost objects could trigger an unwarranted emergency braking maneuver, potentially leading to a rear-end collision with a following vehicle with serious consequences, including fatalities.

[0054] Therefore, it is important that the quality of the frequency response is constantly monitored and that any errors that occur are either corrected or the driver assistance function is deactivated if necessary.

[0055] To implement frequency monitoring, we will first consider the case where the oscillator signal is in circuit block 1.11 of the Fig. 1. The frequency is reduced by dividing by a factor of T=2048 and then in circuit block 1.12 by a real-valued analog-to-digital conversion with a sampling rate of f. A =40MHz is digitized.

[0056] After division by a factor of T=2048, a signal is obtained whose frequency lies in the range (24.15GHz±187.5 / 2) / 2048 = 11.8MHz±45.8kHz - with each frequency ramp, the frequency of the divided signal changes linearly from the lower to the upper value, i.e. by 91.6kHz, provided the oscillator frequency follows its intended curve. Fig. 1 with constant linear slope and constant frequency position (i.e., constant initial and thus center frequency); if the start time of the frequency ramp is defined as t=0, then the following applies to the divided frequency during each of the K=256 frequency ramps (k=0,1,...,K-1) from each of the 8 antenna combinations (m=0,1,...,7): fT(t,k,m)=fTM+bT⋅(t−4μs)=fTS+bT⋅t for 0≤t≤8μs with fTM=11.8MHz,fTS=11.75MHz and bT=91.6kHz / 8μs. Typically, the output signal of a divider has a rectangular shape; in Fig. Figure 10 shows a section of the signal (solid line). To achieve a sinusoidal waveform, this signal is filtered with a low-pass filter that suppresses all harmonics of the rectangular signal (the harmonics are at odd multiples of the respective fundamental frequency, i.e., around 3.11.8 MHz = 35.4 MHz, 5.11.8 MHz = 59 MHz, ...). For this purpose, a typical anti-aliasing low-pass filter with a cutoff frequency of half the sampling frequency, i.e., 20 MHz, can be used. The resulting sinusoidal signal is also shown in Fig. Figure 10 is shown (dashed curve). The phase profile φ T (t,k,m) of this sinusoidal signal s T (t,k,m) is obtained by integrating the divided frequency f T (t,k,m) such that: sT(t,k,m)=As⋅cos(φT(t,k,m)) with φT(t,k,m)=2π⋅[fTS⋅t+bT / 2⋅t2]+φ0(k,m), where φ0(k,m) represents the phase at the respective ramp start (i.e., at t=0) and generally varies from frequency ramp to frequency ramp; A s is the amplitude of the signal.

[0057] After sampling the sinusoidal signal s T (t,k,m) in 25ns grid (sampling rate f) A =40MHz) also result in the following in Fig. 10 plotted values; for the sampled signal s TA (n,k,m) holds true: sTA(n,k,m)=AS⋅cos(2π⋅[fTS / fA⋅n+bT / 2 / fA2⋅n2]+φ0(k,m)).

[0058] These total K0=2048 signals (one per frequency ramp) are further processed in digitized form in the digital signal processing unit 1.10 in order to determine the quality of the frequency response, which is explained below.

[0059] The above relationship for the sampled downsampled signal s TA(n,k,m) represents the ideal case, namely that the actual frequency response corresponds to the linear target response with a constant frequency position and that the signal is free of superimposed noise, which cannot be assumed in reality. Besides a systematic linearity error, not considered here, an error in the frequency position, and thus in the initial or center frequency, can occur, which varies across the frequency ramps (e.g., due to thermal transient effects or disturbances on the power supply); during a frequency ramp, this error Δf E (k,m) is somewhat simplified and considered constant. In the sampled, divided signal, this means an error smaller by the division factor T=2048. ΔfET(k,m)=ΔfE(k,m) / T, which is constant during a frequency ramp, but changes from ramp to ramp. In addition, the actual signal s TA(n,k,m) also includes a noise component r(n,k,m), which arises, for example, from phase noise of the oscillator and quantization effects during A / D conversion. The overall result is the real signal s TA (n,k,m) then to: sTA(n,k,m)=AS⋅cos(2π⋅[fTS / fA⋅n+bT / 2 / fA2⋅n2+ΔfET(k,m) / fA⋅n]+φ0(k,m))+r(n,k,m), where 0≤n≤8µs·f A , i.e. 0≤n≤320.

[0060] To reduce the frequency error Δf ET To determine (k,m), one can determine the phase response of the real signal s TA (n,k,m) compare with the phase response expected for the target frequency response. For this purpose, the real-valued signals s are used. TA (n,k,m) first into its corresponding complex-valued signal, i.e., into its analytical signal s TAC to convert (n,k,m): sTAC(n,k,m)=As⋅exp(i˜⋅(2π⋅[fTS / fA⋅n+bT / 2 / fA2⋅n2+ΔfET(k,m) / fA⋅n]+φ0(k,m)))+rC(n,k,m), where r C(n,k,m) the analytical signal is from the noise r(n,k,m) and has a much smaller amplitude than the useful part of the signal with amplitude A s ; denotes the imaginary unit. An analytical signal is generated by complex-valued filtering with a so-called ideal Hilbert filter, which suppresses all negative frequencies and passes all positive frequencies with a constant transfer factor of 1. In real filters, an infinitely narrow transition band between blocking negative frequencies and passing positive frequencies cannot be achieved. This is the case with the present signal s. TAC (n,k,m) but also not necessary, since it only applies in a narrow range around ±f TM = ±11.8MHz has relevant spectral components; therefore, filtering with a complex-valued first-degree filter with a zero at the frequency -f is sufficient. TM= -11.8 MHz. After this Hilbert filtering, the approximate analytical signal s is obtained. TAC (n,k,m) according to the formula above, where the complex-valued noise r C (n,k,m) also includes components at negative frequencies, which are not disruptive for further processing.

[0061] From the phases of the complex values ​​determined by measurement and processing s TAC (n,k,m) and the known parameters f TS , b T and f A Can one now determine the phase progression? φTEmess(n,k,m)=2π⋅ΔfET(k,m) / fA⋅n+φ0(k,m)+φr(n,k,m) determine, where φ r (n,k,m) is the unknown phase noise component. If one now calculates the phase difference between two time points (at indices n1 and n2), this is described by φTEmess(n2,k,m)−φTEmess(n1,k,m)=2π⋅ΔfET(k,m) / fA⋅(n2−n1)+φr(n2,k,m)−φr(n1,k,m); It should be emphasized that the start-up phase φ0(k,m), which generally varies across the ramps, is implicitly eliminated by the difference calculation. The frequency error Δf determined by measurement Emess (k,m) is thus obtained by evaluating the relationship ΔfEmess(k,m)=(φTEmess(n2,k,m)−φTEmess(n1,k,m))⋅T⋅fA / (2π⋅(n2−n1)), this measured frequency error Δf Emess (k,m) differs from the actual frequency error Δf E (k,m) to account for the measurement error ΔfEmessE(k,m)=(φr(n2,k,m)−φr(n1,k,m))⋅T⋅fA / (2π⋅(n2−n1))

[0062] To minimize this measurement error, widely separated time points n1 and n2 should be chosen, e.g. one at the end of the ramp (n2=320) and the other at the beginning of the sampling period of the received signals (n1=65).

[0063] It should also be noted that phases can only be determined to integer multiples of 2π, which is why the determination of Δf Emess(k,m) must be calculated using modulo calculus with respect to 2π and the uniquely determinable range of the frequency error is "only" T·f A / (n2-n1), which in the present example is over 321MHz and thus far above the expected errors.

[0064] Strictly speaking, slow frequency drift not only results in a change in the center frequency, but also, even within a frequency ramp, in a very small distortion compared to the linear target frequency response, which is generally uncritical from a system perspective. The frequency error Δf determined by measurement Emess (k,m) then represents the mean frequency error during a frequency ramp (apart from the measurement error Δf). EmessE (k,m)).

[0065] In reality, small systematic linearity errors can occur in frequency modulation (i.e., a deviation of the actual transmit frequency response within the transmit signals from a linear regression with the intended slope); one cause of this can be electrical transient effects. If, as in the example above, the same time points n1 and n2 are used to determine the phase difference for each frequency ramp, then such systematic linearity errors have no influence on the determination of the frequency error Δf. E (k,m); if different time points were used, this would no longer be the case, because the linearity error would lead to small changes in the measured mean frequency error (the mean frequency error then depends on the chosen time interval [n1,n2]). Therefore, it is preferable to use the same time points n1 and n2 for each frequency ramp.

[0066] In the above approach, the phase noise φ can be used to compensate for the noise. r(n,k,m) conditional measurement errors Δf EmessE (k,m) are significantly too large. An improvement can be achieved by taking not only one of the phase error measurements φ at the beginning and end of the selected time interval [n1,n2]. TEmess Instead of using (n,k,m), several are used to better average the phase noise. However, directly averaging the phase error measurements is unfavorable because, firstly, they are only unique over multiples of 2π (meaning phase jumps can occur), and secondly, from a signal-theoretic perspective, this would be averaging in a non-linear dimension. Therefore, the phase error measurements φ are averaged. TEmess (n,k,m) better indirectly via corresponding complex-valued unit pointers exp(i˜⋅φTEmess(n,k,m)); One accumulates such pointers and then determines the phase of the sum pointer. If the useful component 2π·Δf changes during the accumulation period... ET (k,m) / f A ·n of the phase error φTEmess (n,k,m) only a little (which is generally the case with the considered design) and is the phase noise φ rIf the values ​​(n, k, m) are uncorrelated over time n (which is usually a good approximation), then the influence of phase noise is reduced by a factor of N with respect to power and a factor of √N with respect to amplitude when accumulated over N values. If 64 values ​​are used in both the front and back time periods, the noise is effectively reduced by a factor of 64 with respect to power and a factor of 8 with respect to amplitude. However, it should also be noted that the distance between the midpoints of the two time periods decreases if only the interval [n1, n2] = [65, 320] is used: when using only one value in each time period, the distance was n2 - n1 = 320 - 65 = 255; Using the front time interval n1,...,n1+63 and the rear time interval n2-63,...,n2, the distance is reduced by 63 to 192. According to the above relationship for the measurement error Δf EmessE(k,m) this results in a loss of a factor of 255 / 192 = 1.33 with respect to amplitude and 1.76 with respect to power. Averaging therefore results in an overall reduction of noise by a factor of 36 with respect to power and a factor of 6 with respect to amplitude.

[0067] The phase error measurements φ are needed for averaging. TEmess (n,k,m) cannot be explicitly determined, since only the phase indicator exp(i˜⋅φTEmess(n,k,m)) interested and the complex-valued signal s TAC (n,k,m) includes this phase vector (see corresponding relationships above); therefore, for averaging the complex-valued signal s TAC (n,k,m) exit. For this, one must go to s TAC (n,k,m) then the phase component dependent on the target frequency curve 2π⋅[fTS / fA⋅n+bT / fA2⋅n2] eliminate what is obtained by multiplication with the phase pointer exp(−i˜⋅2π⋅[fTS / fA⋅n+bT / fA2⋅n2]) can be achieved (i.e., one essentially performs frequency normalization): sTACN(n,k,m)=As⋅exp(i˜⋅(2π⋅ΔfET(k,m) / fA⋅n+φ0(k,m)))+rCN(n,k,m); It should be noted that the statistical properties differ from the noise r. CN (n,k,m) do not change because they were multiplied by values ​​with amplitude 1. These signals are then accumulated in both time intervals, yielding the so-called phase characteristic P1(k,m) in the first time interval and P2(k,m) in the second. Then, for each frequency ramp, the phase characteristic P2(k,m) is multiplied by the complex conjugate of the phase characteristic P1(k,m), and the phase of this value is determined, thus giving the difference phase Δφ. TEmess (k,m) of the two phase parameters yields: ΔφTEmess(k,m)=2π⋅ΔfET(k,m) / fA⋅(n2−n1−(N−1))+Δφr(k,m); where N is the number of values ​​used for averaging per time period and Δφ r(k,m) represents the unknown phase noise component, which is now reduced by averaging over the N values ​​per time interval. It should be emphasized that the start phase φ0(k,m), which generally varies across the ramps, is implicitly eliminated again.

[0068] The frequency error Δf determined by measurement Emess (k,m) is thus obtained by evaluating the relationship ΔfEmess(k,m)=ΔφTEmess(k,m)⋅T⋅fA / (2π⋅(n2−n1−(N−1))), this measured frequency error Δf Emess (k,m) differs from the actual frequency error Δf E (k,m) to account for the measurement error ΔfEmessE(k,m)=Δφr(k,m)⋅T⋅fA / (2π⋅(n2−n1−(N−1))).

[0069] In the above calculation of the phase characteristics P1(k,m) and P2(k,m), the signal s TAC(n,k,m) were frequency-normalized by multiplication with a corresponding phase vector (i.e., the desired frequency response was removed), and then these values ​​were accumulated for the respective time intervals. Multiplication and accumulation can be formulated or determined together in one step as a scalar product; one vector consists of the signal s. TAC (n,k,m) in the considered time interval, the other consists of corresponding phase vectors. The signal s TAC (n,k,m) is obtained from the real-valued signal s by a complex-valued Hilbert filter of the first degree. TA (n,k,m) were created, i.e., by weighted addition of the two values ​​s TA (n,k,m) and s TA (n-1,k,m). This filtering and the scalar product determined from the filtered values ​​can also be viewed as a single scalar product. Thus, the phase characteristics can be directly derived from the real-valued signal s via a scalar product. TA(n,k,m) determine; one vector consists of the signal s TA (n,k,m) in the considered time interval plus the preceding value (the considered time interval is thus effectively extended by one time point), the other vector is essentially formed via the sum of phase vectors (the filter coefficients are also phase vectors), whereby it is sufficient to determine this parameter vector once a priori, since it is the same for all frequency ramps.

[0070] Above, it was assumed that the first-degree Hilbert filter always has its zero at the negative of the target band center, i.e., at -f. TM= -11.8 MHz. In principle, the zero could also be changed via the frequency ramp so that it always lies at the negative of the respective target frequency, and the parameter vectors of the scalar products calculated accordingly. The parameter vectors can also be approximated so that they always refer to the center frequency of the respective time interval; for this purpose, this center frequency is used for the Hilbert filter and also for the phase vectors of the frequency normalization (the quadratic component in the phase vectors of the frequency normalization is omitted).

[0071] So far, we have considered the case where there are two time periods per frequency ramp, one at the beginning and one at the end; the frequency error of each ramp was determined by comparing the phase characteristics in both periods. However, it is also possible to work with only one time period per frequency ramp (e.g., at the end of each ramp); in this case, the phase characteristics are compared across two consecutive ramps, i.e., the phase characteristic of the second ramp with that of the first, the phase characteristic of the third ramp with that of the second, and so on. To determine the absolute frequency error in this way, one would need to know the ideal frequency response (i.e., without disturbances) between the ramps precisely so that it could be taken into account during the frequency normalization from ramp to ramp (the exact frequency response is needed for the precise phase progression between two periods).However, the exact frequency response, particularly during the frequency return, is generally unknown, as it depends, for example, on the transient response of the PLL. Nevertheless, the frequency response between frequency ramps will always be the same; therefore, a relative determination of the frequency position, i.e., determining its change across the ramps, is possible. Only a constant component remains undefined, which is not very critical for monitoring, since only very large constant frequency position errors would be critical (which generally do not occur) and these can also be determined by other methods. Frequency normalization is then performed for each frequency ramp with identical phase vectors; that is, the unknown but constant phase shift from ramp to ramp that occurs in the undisturbed case is disregarded.In the undisturbed case, the phase characteristic will change by a constant value from ramp to ramp (precisely by this uncompensated constant phase progression); the phase difference Δφ is determined analogously to above via the phase characteristics. TEmess The phase difference (k,m) between two successive time intervals, i.e., two ramps, is therefore constant in the undisturbed case. However, if there is a change in the frequency, the measured phase differences Δφ change. TEmess (k,m); normalizing the phase differences to the first value Δφ TEmess (0,1), i.e., the phase difference between the first and second ramps, then the measured frequency error Δf is obtained. Emess (k,m) normalized to its first value Δf Emess (0, 1): ΔfEmessR(k,m)=ΔfEmess(k,m)−ΔfEmess(0,1)=(ΔφTEmess(k,m)−ΔφTEmess(0,1))⋅T / (2π⋅ΔtC), where Δt CThe interval between two frequency ramps is 10 µs. This determines the change in phase across the frequency ramps, here relative to the frequency at the beginning of the ramp sequence. The relative error Δf EmessR (k,m) of the frequency position represents the average value between the respective two time intervals (which lie at the end of two consecutive ramps) and thus slightly leads the average error in the sampling period of the received signal of the frequency ramp k,m. This could be corrected by interpolation with the subsequent value, but this is negligible for slow changes in the frequency position.

[0072] The advantage of this approach is that the time interval between the two considered time periods is longer, and thus the influence of the phase noise Δφ generated by the A / D converter, e.g. by quantization effects, is reduced. r (k,m) becomes smaller (the measurement error is Δφ r (k,m)·T / (2π-Δt C) and thus inversely proportional to the time interval between the midpoints of the two time periods). To further reduce the measurement error, one can use the relative error Δf. EmessR (k,m) filter or extend the period between the two used time intervals over all 8 antenna combinations (thus only one frequency position error is determined per index k, which is sufficient in the case of slow frequency drift e.g. due to thermal transient effects).

[0073] It should be emphasized that if two time periods from different frequency ramps are used to determine the relative frequency position, the frequency response between the two time periods must be identical (apart from the disturbances being measured). This means the target value for frequency generation (e.g., the drive signal of a PLL) must be the same. Therefore, if variable parameters are used in the frequency response, these cannot be varied arbitrarily. When varying the spacing of frequency ramps, one possible approach is to use a different spacing only for every second frequency ramp, while the spacing for the other ramps remains fixed. The frequency position error is then determined only from pairs with a constant spacing; pairs with varying spacing are not used, but rather the error is interpolated from the two adjacent values.

[0074] Regarding the example described at the beginning, using two time periods, each from a ramp, the following should be mentioned: if the exact phase progression between the two time periods is not taken into account during phase normalization, only a relative change in the frequency position is possible; however, the phase vectors for frequency normalization can then have any starting values ​​in both time periods, they only need to be the same across all frequency ramps (e.g., always starting phase zero).

[0075] It was previously assumed that all transmitted signals had the same target frequency. However, there are also applications where the frequency of the frequency ramps—that is, their initial and, consequently, their center frequency—is deliberately varied across the transmitted signals. This could, in principle, be considered in frequency normalization; however, one can also simply determine the deviation between the actual frequency (measured absolutely or relatively) and the target frequency by subtraction.

[0076] The following describes the further processing of a frequency position error Δf determined as above. E (k,m) explains (it should be noted that between measured frequency error Δf Emess (k,m) and actual frequency error Δf E (k,m) is no longer distinguished, as a sufficiently accurate determination is assumed).

[0077] The preferred approach is to determine the frequency position error Δf. E(k,m) to compensate. A change in frequency will change the phase of the received signal, since a different number of wave trains fit into the path to the object and back. For example, if the frequency increases by 1 MHz and the propagation time is 1 µs (object at a distance of 150 m), then exactly one more wave train fits, so the phase changes by 2π; the effect is proportional to the distance r of the object and the change in frequency Δf. E (k,m). This gives the general phase shift Δφ(k,m) of the received signal as follows: Δφ(k,m)=2π⋅ΔfE(k,m) / 1MHz⋅r / 150m.

[0078] This phase shift persists after the first DFT (for distance), meaning the complex value of the power peak at the distance gate corresponding to distance r exhibits this phase shift. To compensate, one can simply multiply by the complex unit vector with phase -Δφ(k,m). Since the distance gates containing objects are unknown after the first DFT, one multiplies all distance gates j (with a length of 1 m) by their corresponding phase correction vector. PK(j,k,m)=exp(−i˜⋅2π⋅ΔfE(k,m) / 1MHz⋅j / 150). If one uses the absolute frequency position error Δf instead of the E (k,m) only relatively determined, then only the change in frequency position is compensated, which is generally sufficient. If the change in frequency position were not compensated, this would lead to a distortion in the course of the second DFT (for relative velocity); this will be explained later.

[0079] A constant error in the frequency position, i.e., a constant deviation of the center frequency of the frequency ramps from their target value, changes the mean wavelength and thus affects the calculated relative velocity and angular position of objects (see derivations above). Using the actual center frequency instead of the target center frequency can avoid errors; however, this is only necessary with large deviations, as only then do the resulting errors become significant (the effects are much stronger when the center frequency changes across the frequency ramps).

[0080] It should be noted that to calculate the phase correction vector P K (j,k,m) can also use a filtered or smoothed frequency position error (e.g. smoothing of the measured frequency position error Δf) E (k,m) by a regression curve).

[0081] Are the frequency position errors Δf E(k,m) not compensated, it must be assessed whether its influence on the detection quality is still within an acceptable range, i.e., whether there are no unacceptable functional limitations; otherwise, the affected functions for driver assistance and / or autonomous driving maneuvers must be restricted or deactivated.

[0082] As an example, an actual frequency response is considered, which is curved compared to the constant target response and also exhibits a periodic disturbance - in Fig. 11 is the corresponding frequency position error Δf E(k,0) is shown. To assess such deviations, one can perform a regression (e.g., a first- or second-degree polynomial regression) and use the parameters of the adjustment function (e.g., the polynomial) and / or the deviation, especially the standard deviation, between the actual curve and the adjustment function as a measure of goodness of fit. However, these values ​​only provide a rough indication of the deterioration in detection quality (for the case considered here, where the frequency position error is not compensated).

[0083] For a more precise assessment, more in-depth analyses are necessary. An uncompensated frequency position error Δf E(k,m) primarily affects the second DFT (for relative velocity), which is calculated in each distance gate j=0,...,99 and each antenna combination m=0,1,...,7 for the input values ​​e(j,k,m) extending over the frequency ramps k=0,1,...,K-1. For a target in distance gate j and with relative velocity v rel applies: e(j,k,m)=w(k)⋅exp(i˜⋅2π⋅(k⋅vrel / 280km / h+ΔfE(k,m) / 1Mhz⋅j / 150)) where w(k) represents the window function used for the second DFT and the amplitude is assumed to be 1 and the starting frequency to be zero, as this is irrelevant for further consideration; the first phase component describes the linear phase change due to the relative velocity v rel , the second phase component represents the influence of the uncompensated frequency position error Δf E (k,m) represents and results from the above derivation of the phase correction values.

[0084] For a target with vanishing relative velocity (i.e., v) rel =0) the input signal of the second DFT then results in e(j,k,m)=w(k)⋅exp(i˜⋅2π⋅ΔfE(k,m) / 1MHz⋅j / 150); without frequency position error Δf E (k,m) would correspond to the window function w(k).

[0085] The above relationship shows that the frequency position error has a greater effect the higher the distance gate index j is, i.e., the greater the distance. Therefore, the maximum distance of interest, r=99m, i.e., j=99, is now considered: e(99,k,m)=w(k)⋅exp(i˜⋅2π⋅ΔfE(k,m) / 1.485MHz).

[0086] For the frequency position error Δf E (k,0) after Fig. 11 is in Fig. 12a shows the velocity spectrum v(99,I,0) resulting from the DFT in terms of magnitude (solid line; logarithmic representation, i.e., in dB), where the index I represents the relative velocity gates. For comparison, it shows Fig. 12a also the velocity spectrum, which is the target frequency profile for a target with the same amplitude, the same distance and the same v rel= 0 results (dashed curve; is the DFT of the window function w(k)). Due to the curvature in the actual frequency response, in addition to a small shift, the power peak belonging to the target broadens around I=0. This can, for example, lead to smaller targets in the vicinity of a larger one becoming undetectable (provided they are at the same distance and approximately the same angle, as they can otherwise be distinguished by these parameters). The "spreading" of the power into a broader power peak also leads to a reduction in level, so that the detection sensitivity decreases, which can result in smaller objects at greater distances generally no longer being detected.The periodic disturbance of the frequency position generates the additional smaller power peaks at I=11 and I=243; this is particularly critical because it can cause a real object at the same distance to generate ghost objects with a different relative velocity, which can result in strong braking (if the ghost object appears to have a slower velocity).

[0087] To assess the quality of the frequency response, the velocity spectra v(99,I,m) calculated for the actual frequency can, for example, be checked against a limit curve in terms of their magnitude; alternatively, the magnitude of the difference between the velocity spectra for the actual and target frequencies can also be checked against a limit curve. Fig. Figure 12b shows the difference in velocity spectra for the example above, normalized to the maximum of the velocity spectrum at the target frequency and plotted in dB. Due to the normalization to the maximum of the velocity spectrum at the target frequency, this is also referred to as the relative difference. Checking against a cutoff curve represents a binary quality measure (i.e., with the two outcomes "good" or "bad"); alternatively, an analogous quality measure can be defined, e.g., the maximum relative difference between the velocity spectra at the actual and target frequencies.

[0088] In the example according to Fig. 11 and Fig. 12. Due to a periodic error in the frequency distribution, additional smaller power peaks arise around the actual object, which can lead to ghost objects at the same distance as the real object. If an analysis of the actual frequency distribution reveals the amplitude of such interference lines, it is possible to check for each detection whether it was caused by an erroneous frequency distribution from another detection at the same distance, and then, if necessary, to discard this detection entirely or mark it as a potential false detection.

[0089] In the preceding considerations, the frequency position was monitored during the actual transmission signals (i.e., for the transmission signals whose corresponding received signals are evaluated for environmental sensing). To save the additional A / D converter for digitizing the divided oscillator signal, the A / D converter used for sampling the received signals could also be used for this purpose. However, in that case, frequency monitoring could not occur in parallel with environmental sensing; a further sequence of transmission signals with the same frequency response would have to be introduced solely for frequency monitoring. Frequency monitoring and environmental sensing would then occur at different frequency ramps, arranged either in two sequentially successive blocks or, preferably, nested within each other.The transmission power could also be switched off on the ramps used to monitor the frequency position (to save power and provided this does not affect the frequency position error).

[0090] So far, the case has been considered where an A / D converter is used to digitize the frequency-reduced oscillator signal; now, instead, circuit block 1.12 will be used. Fig. 1. A counter is used. The oscillator signal in circuit block 1.11 is then divided down by a factor of T=4, so that the center frequency at f TThe frequency is 6.04 GHz. The counter increments its value by 1 on each rising edge of the divided rectangular signal; it thus counts the number of periods of the divided signal. The counter is not reinitialized at the start of each frequency ramp, but simply continues counting indefinitely, even between frequency ramps – it can therefore be described as a free-running counter that does not require intermediate initializations.

[0091] For each of the 8 µs duration K=256 frequency ramps (k=0,1,...,K-1) from each of the 8 antenna combinations (m=0,1,...,7), the counter is read at two identical times, e.g., at t1=1.6 µs and t2=8 µs after the ramp start (and thus at the beginning and end of the sampling period of the received signals); the two read counter values ​​are denoted z1(k,m) and z2(k,m). Then, the difference Δz(k,m) between the two counter values ​​is calculated (which is why the absolute counter value is irrelevant, and therefore no re-initialization is required at the start of the ramp): Δz(k,m)=z2(k,m)−z1(k,m)

[0092] The counter difference indicates the number of periods in the time interval [t1,t2] for the signal divided by a factor of T=4; the oscillator signal then has a factor of T=4 more periods in this time interval. The average frequency can be calculated by dividing the number of periods by the length of the time interval: fM(k,m)=4⋅Δz(k,m) / (t2−t1).

[0093] The frequency error results from the deviation of the measured center frequency from its target value. The maximum measurement error of the counter is one period relative to the divided signal; in relation to the oscillator's frequency, this translates to a maximum error of 4 / (t2-t1) = 625 kHz. To reduce the measurement error, one can again compare the counter values ​​at two points in time that lie at different frequency ramps (the measurement error is inversely proportional to the interval between the two readout times). In general, the approaches discussed in detail above for digitization with an A / D converter can be applied analogously here.

[0094] It should also be mentioned that real counters are limited in their length, i.e., their maximum count, and therefore overflow can occur. When they reach their maximum count, they reset to 0 with the next edge to be counted. This corresponds to a modular calculation. If the counter evaluations, i.e., the difference calculation, are also performed using modular arithmetic, overflow does not distort the result, provided the number of periods to be counted between the two time points t1 and t2 does not exceed the counter length (i.e., the maximum count). For a time difference t2-t1 = 6.4 µs, a 16-bit counter is sufficient. If the counter is shorter, the center frequency can no longer be measured unambiguously.However, since only the deviation from a known target value and / or the change across the frequency ramps needs to be measured, uniqueness in the MHz range is sufficient; with a uniqueness range of 160MHz, an 8-bit counter is sufficient.

[0095] As can be seen from the derivations above, the accuracy of the measurement decreases with increasing division ratio T. On the other hand, the less the frequency is divided, the faster the counter must be – however, a fast counter is complex to implement in terms of circuitry and requires a lot of power. Frequency mixing can circumvent this problem, as it does not affect the measurement accuracy; however, generating a second signal in the 24 GHz range is complex. Therefore, a combination of division and mixing can also be implemented. For example, the oscillator signal can first be divided by a factor of 4 to the range of approximately 6.04 GHz and then downmixed at a fixed frequency of 5.8 GHz, so that the counter only has to operate in the range of about 200 MHz.

[0096] Another approach to reducing the division factor T is a counter that counts both the positive and negative edges of the divided signal.

[0097] Finally, the advantages and disadvantages of the two methods presented above for digitizing the frequency-reduced oscillator signal will be briefly discussed. Compared to the counter approach, analog-to-digital conversion (ADC) has the advantage of being able to work with larger frequency divider factors, since it can, in principle, measure frequencies and phases more accurately (given identical input signal frequencies). However, an ADC is generally more complex to implement than a counter (given identical input signal frequencies), and evaluating ADC values ​​is more complex than evaluating counter values.

[0098] It should be noted that the considerations and explanations according to the invention, illustrated by the examples above, can be transferred to general dimensioning and parameter design, i.e., they can also be applied to other numerical values. For example, the approaches according to the invention can also be applied to a radar in the 77 GHz range.

Claims

[1] Method for a radar system for detecting the surroundings of a motor vehicle and realizing a function for driver assistance and / or autonomous driving maneuvers with the steps - Generating a frequency modulation using a controllable or adjustable oscillator, - generating a sequence of K0 (K0>1) transmission signals modulated in the transmission frequency, each having the same nominal frequency characteristic, if necessary apart from a variation of the frequency position, i.e. in particular a variation of the initial and thus equivalently the center frequency, - transmission of transmission signals by means of transmission means, - Receiving transmission signals reflected from objects using receiving means, - Analyze the frequency position of the transmission signals and - Evaluation of the received signals, in particular for the detection of objects, by means of signal processing means, whereby the actual course of the frequency position resulting across the K0 transmission signals, i.e. in particular the actual course of the initial, the middle or the middle frequency of the transmission signals, or its deviation from the target course caused, for example, by frequency instability or frequency drift, is determined absolutely or relatively, i.e. up to an indeterminate constant component, whereby - a time-discrete signal is used for each transmission signal, which contains information about the frequency response of the transmission signal and which is preferably generated by sampling an analogue signal or by reading a free-running counter at predetermined times, - these time-discrete signals are unnormalized over the K0 transmit signals with respect to their phase and / or initial value, and - during an evaluation of these time-discrete signals, a normalisation is explicitly carried out or an influence of the position of their phase and / or their initial value is implicitly eliminated, and depending on an actual curve determined in this way and / or a deviation of the actual curve from the target curve of the frequency position and / or a quality measure derived therefrom, a correction in the control of the oscillator and / or a correction in the evaluation of the received signals and / or an adaptation of the function for driver assistance and / or for autonomous driving maneuvers up to and including its deactivation is carried out, whereby - the frequency of the transmitted signals is linearly modulated, - the signals received by reflection of the transmission signals from objects are converted into the low-frequency range by mixing them with a signal whose frequency corresponds to the current transmission frequency or differs from it by a constant factor, - the low-frequency received signals are sampled NA times in an equidistant grid, - a first spectral analysis is formed from these NA sample values, in particular in the form of a discrete Fourier transformation, in order to be able to carry out a distance measurement of the objects and a separation for the simultaneous detection of several objects, and - at frequency support points of the first spectral analysis, a second spectral analysis is carried out, in particular using the discrete Fourier transformation, which extends over all or a subset of the values ​​occurring at the respective frequency support point during the K0 transmission signals, in order to be able to realize in particular a relative speed measurement of the objects and a separation for the simultaneous detection of several objects, and in the method (a) - from the determined actual frequency position across the K0 transmission signals, the expected spectrum of the second spectral analysis of a target at a distance r with the relative speed zero is determined, - the resulting spectrum or its magnitude can be used directly to derive a quality measure and / or - a quality measure is derived from the deviation of the spectrum from the spectrum expected at the target frequency position of a target at a distance r with a relative speed of zero, whereby the same window function is used for both spectral analyses and the resulting spectra are related to the same level for comparison and / or (b) a deviation from the actual to the desired frequency position is corrected directly or after performing a regression by multiplication with a complex unit vector between the first and second spectral analysis, whereby the phase of the complex unit vector is proportional to this deviation and proportional to the frequency of the respective frequency support point of the first spectral analysis. [2] Method for a radar system according to claim 1, characterized by , that - a signal is used for analysis which is reduced in frequency compared to the transmitted signal by frequency division by a factor of T>1 and / or by mixing, - these low-frequency signals resulting from the transmission signals are sampled, if necessary after filtering, - a complex value is determined from these sampled signals resulting from the transmission signals, if necessary after further filtering, in one or more time periods per transmission signal, in particular by - that sample values ​​of the respective section are corrected at least approximately by the expected phase progression resulting from the target frequency curve of the transmission signals there, and - these phase-corrected samples are accumulated per section, - wherein the time period(s) over which K0 transmission signals preferably have at least approximately the same position, - and the actual course of the frequency position over the K0 transmission signals or its deviation from the target course is characterized with the help of the complex values ​​determined in this way, in particular based on the phase position of these values, which is why they are also called phase characteristics in the following. [3] Method for a radar system according to claim 2, in which the sampled signals are real-valued and are converted at least approximately into their analytical signals, i.e. the complex-valued signals with only the positive or negative frequency components, in the time periods under consideration, wherein first-degree Hilbert filters with the zero point at approximately the negative or positive of the respective center frequency of these frequency-modulated signal sections are preferably used for this purpose. [4] Method for a radar system according to claim 3, in which the Hilbert filtering and the accumulation of the phase-corrected sample values ​​over the respective time period are realized by a scalar product between the sample values ​​and a preferably pre-calculated parameter vector. [5] A method for a radar system according to any one of claims 2 to 4, wherein - for pairs of phase characteristic values, their phase difference is formed, whereby the two values ​​of each pair belong to time periods of the same position over the transmitted signals and at least approximately the same time interval and the same frequency curve lies between the two time periods, - and from these phase differences or their change over the transmission signals, the course of the frequency position is determined absolutely or relatively, i.e. up to an indeterminate constant component. [6] Method for a radar system according to claim 5, in which the pairs of phase characteristics are each formed from the same transmission signal, wherein preferably one value is formed from a time period in the front region of the transmission signals and the second value is formed from a time period in the rear region of the transmission signals. [7] Method for a radar system according to claim 5, in which the pairs of phase characteristics are each formed from different transmission signals, for example from successive transmission signals, both values ​​belonging to time periods of similar position over the transmission signals. [8] Method for a radar system according to one of the preceding claims, in which it is derived from a quality measure whether a detection could have resulted from a deviation of the actual frequency position over the K0 transmission signals from the target frequency position from another detected object, and this detection is then optionally completely rejected or marked as a potential false detection. [9] Method for a radar system according to one of the above claims, in which a regression is carried out on the difference between the desired and actual course of the frequency position and the parameters of the compensation function and / or the deviation, in particular the standard deviation between the actual course and the compensation function are used as a quality measure. [10] Method for a radar system according to one of the above claims, in which the center frequency of the transmission signals or their deviation from the desired center frequency is determined and used for calculating the relative speed and / or the angular position of objects. [11] Radar system arranged to carry out a method according to any one of the preceding claims.

Images