MIMO radar system with non-coherent TX integration and transmitters of varying power levels

Abstract

The invention relates to a method for a radar system and a radar system for environmental detection with transmitters having several parallel transmitting antennas for emitting signals which contain one or more sequences of K individual signals, means for changing the phase of the transmitted individual signals by which a phase profile different for the transmitting antennas is realized over the K individual signals, receivers with several receiving antennas for receiving transmitted signals reflected from objects and signal processing means for processing the received signals, wherein not all transmitting antennas emit the same average transmit level, and a non-coherent integration (TX integration) is performed over certain signal components, wherein this non-coherent TX integration is performed multiple times assuming different radial relative velocities and, optionally, distances of an object.In non-coherent TX integration, signal components are weighted more heavily the higher the radiated transmit level of the respective assigned transmitting antenna is, and the result of the non-coherent TX integration is used to identify the phase progression resulting from the object's relative velocity and, if applicable, object distance, and / or a quantity derived therefrom.

Description

[0001] The invention relates to a radar method and radar system for use in driver assistance systems in motor vehicles. The radar system has several transmitting and receiving antennas operating in parallel, which is referred to as MIMO (Multiple-Input Multiple-Output), and has different transmission levels on the transmitting antennas, which, according to the invention, is used for object identification in the case of non-coherent TX integration. 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, in particular by issuing warnings. A distinction is made between comfort and safety functions.

[0003] In current vehicle development, FSRA (Full Speed ​​Range Adaptive Cruise Control) plays an 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 for reducing braking or stopping distance in emergency situations, up to and including 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] Nowadays, the driver is no longer just assisted, but the driver's task is increasingly performed autonomously by the vehicle, i.e., the driver is increasingly replaced; this is referred to as 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, among others, of operating reliably even in poor weather conditions and being able to directly measure not only the distance to objects but also their radial relative velocity via the Doppler effect. Currently, 77 GHz and 79 GHz are typically used as transmission frequencies.

[0007] The functions mentioned above require high detection quality, for which precise angle determination is essential. Therefore, MIMO radars are increasingly being used. These radars have multiple fully parallel transmitting and receiving antennas and utilize all possible combinations of transmitting and receiving antennas to achieve the best possible angle determination. For parallel operation of the transmitting antennas, their transmitted signals must be modulated differently so that the components they cause in the received signals can be separated. A commonly used approach employs a sequence of identical or similar transmitting signals, over which the phase is linearly changed, with the modulation rate—that is, the rate of change of the linear phase shift—varying across the transmitting antennas.To separate the signal components originating from the various transmitting antennas, the result of a discrete Fourier transform of the individual received signals is often used, since the different rates of linear phase change result in the components of the different transmit paths being located at different frequency points. A challenge here is correctly assigning the power peaks occurring in this spectrum to the individual transmitting antennas (also known as TX demodulation). WO 2018 / 137835 A1 proposes that, in addition to the phase modulation described, the transmitting antennas transmit at different levels, resulting in power peaks with varying amplitudes in the spectrum. The assignment of the power peaks to the transmitting antennas is then based on the relative magnitude of the power peaks.However, this method has limited robustness - especially in situations with multiple objects whose power peaks overlap at least partially, and with small and / or distant objects, i.e. objects that result in a weak received signal. Problem, solution and advantages of the invention

[0008] The object of the invention is to provide improved TX demodulation for automotive MIMO radar systems, which is also suitable for scenarios with multiple objects and objects with a low received signal.

[0009] This problem is fundamentally solved by a method for a radar system according to claim 1 and a corresponding radar system according to the dependent claim, wherein the radar system may also explicitly include the features described in the dependent claims. Advantageous embodiments of the invention are claimed in the dependent claims. The core idea is that integration is performed not only over the spectra of the received path signals, but also incoherent integration in the grids of the transmitted path signals expected in the spectra, wherein the weights used in the incoherent integration are based on the transmit level of the respective associated transmitting antenna.From the result, in which individual objects generate power peaks at several positions of an equidistant grid, the correct, i.e., actual frequency support points of objects are identified, in particular as the frequency support point where the maximum of the respective power peaks lies.

[0010] The advantages of the invention arise from the fact that high detection quality and sensitivity can be ensured, and thus the hardware requirements can be reduced if necessary.

[0011] The radar system to which the inventive method for environmental detection relates comprises: 1) transmitting means with several parallel transmitting antennas for emitting signals, which contain one or more sequences of K individual signals, the general form of which is preferably the same or similar, wherein their frequency can change, in particular, linearly; 2) means for changing the phase of the transmitted individual signals, by which a phase response different for the transmitting antennas is achieved over the K individual signals, wherein this difference in the phase response for different transmitting antennas over the K individual signals is at least approximately linear, optionally excluding phase jumps due to the phase uniqueness range of 2π; 3) receiving means with one or more receiving antennas for receiving signals reflected from objects.wherein the signals received by an object are composed of components originating from different transmitting antennas, these components exhibiting a phase response that differs linearly across the K individual signals, as well as a similar phase progression, which is generated in particular by the relative motion of the object and, in the case of a frequency shift across the K individual signals, also depends on the object's distance, and 4) signal processing means for processing the received signals. The method is characterized in that 1) not all transmitting antennas radiate the same average transmit level, 2) a non-coherent integration is performed over signal components whose phase responses across the K individual signals lie within the expected grid of the signal components of an object originating from different transmitting antennas and exhibiting linearly different phase responses, which is hereinafter referred to as non-coherent TX integration.wherein this non-coherent TX integration is performed multiple times assuming different radial relative velocities and, if applicable, distances of an object, 3) in the non-coherent TX integration, signal components are weighted more heavily the higher the radiated transmit level of the respective associated transmitting antenna is, and 4) the result of the multiple non-coherent TX integration is used to identify the phase shift resulting from the object's relative velocity and, if applicable, object distance, and / or a quantity derived therefrom.

[0012] Advantageously, the weighting of the signal components used in non-coherent TX integration correlates with the average transmit level of the respective assigned transmitting antenna, i.e., is proportional to the average transmit power or transmit amplitude of the respective assigned antenna, possibly minus a constant value, in order to advantageously achieve the most robust possible identification of the phase progression resulting from the object relative velocity and, if applicable, object distance, and / or a quantity derived therefrom, especially in the case that the signals contain contributions from multiple objects and / or significant noise.

[0013] Advantageously, prior to the non-coherent TX integration via the received signals to multiple receiving antennas or via signals derived from them, a coherent integration, in particular in the form of digital beamforming, hereinafter referred to as RX beamforming, can be carried out, and the result of the multiple non-coherent TX integration from different RX beamforming directions can be used to identify the phase progression resulting from the object relative velocity and, if applicable, object distance, and / or a quantity derived therefrom.

[0014] Furthermore, a non-coherent integration can be performed over the received signals to several receiving antennas or over signals derived from them, which is expediently carried out before the non-coherent TX integration, and the result of the entire non-coherent integration, performed multiple times under the assumption of different radial relative velocities and, if necessary, distances of an object, can be used to identify the phase progression resulting from the object's relative velocity and, if necessary, object distance, and / or a quantity derived therefrom.

[0015] Non-coherent TX integration can be implemented using power values. Preferably, it can be implemented using the magnitude values ​​of the signal components.

[0016] In an advantageous embodiment of the invention, the method can be characterized in that 1) the K individual signals preferably lie in an at least approximately equidistant grid, 2) a discrete Fourier transform of length L is performed on their K received signals or signals derived from them, optionally after padding with zeros, i.e., a so-called zero-padding, wherein the L frequency reference points of the discrete Fourier transform are hereinafter referred to as Doppler gates, 3) the phase response over the K individual signals includes a linear component, i.e., with a different slope, for each transmitting antenna, by which the signal components from the different transmitting antennas lie at different Doppler gates in the discrete Fourier transform.4) the non-coherent TX integration is performed via Doppler gates in the grid expected from the different slopes of the linear components, and 5) this non-coherent TX integration is performed for different positions of this grid, i.e., different, in particular all, Doppler gates as the first grid point and thus for different, in particular all possible, object relative velocities and, if applicable, object distances.

[0017] Furthermore, the linear components of the phase responses, when mapped to the phase uniqueness range of 0...2π, ​​can have a common period P, which preferably represents an integer divisor of the number L of Doppler gates, whereby the Doppler gates corresponding to the transmitting antennas represent a subset of an equidistant grid with a distance L / P.

[0018] Advantageously, the summation necessary for non-coherent TX integration for preferably all P grid layers in an equidistant grid with integer spacing L / P can be realized by performing a cyclic correlation between a vector with the P values ​​of the discrete Fourier transform in this grid and an occupancy vector of length P, wherein the values ​​of the occupancy vector for all indices, which for a transmitting antenna correspond to its linear phase change within one period P as multiples of 2π, assume non-zero values ​​and are otherwise 0, and wherein this cyclic correlation can preferably be carried out as a fast correlation, i.e. by multiplying two spectra in the frequency domain.

[0019] Advantageously, the non-zero values ​​of the occupancy vector correlate to the average transmission level of the respective assigned transmitting antenna, i.e., are in particular proportional to the average transmission power or transmission amplitude of the respective assigned transmitting antenna, possibly minus a constant value.

[0020] Furthermore, different transmission levels can be caused by hardware-related effects such as different lengths of the antenna leads via the transmitting antennas and / or can be generated specifically, in particular by a configurable transmission power generation, especially so that the period P can be chosen as small as possible, with the number of transmitting antennas representing the lower limit.

[0021] Advantageously, the results of the non-coherent TX integration of raster layers with spacings that are integer multiples of L / P are investigated and, in particular, compared to identify objects as well as their relative velocity and, if applicable, distance, preferably also using a detection threshold that depends in particular on the noise level, and, when using an RX beamforming, this is done in all its beam directions.

[0022] Advantageously, for object identification, the grid position used in the grid with a distance L / P is the one where the incoherent TX integration has the highest amount and therefore the highest performance.

[0023] Furthermore, for object identification, grid layers with a distance L / P can also be used where the power of the non-coherent TX integration is not maximum, but is above a threshold which may depend on the distance of this grid layer to the grid layer of the maximum and in particular is at least a minimum distance above the value generated there by an object which is located at the grid layer of the maximum and generates the power level there.

[0024] If the transmitting antennas are located in different planes when viewed vertically, the transmission levels of the respective transmitting antennas can differ in at least some planes, preferably in such a way that the diversity of the transmission levels is similar in as many or all planes as possible, thereby reducing the influence of superposition effects caused by reflections on a road surface.

[0025] Furthermore, there can be several similar groups, in particular pairs of transmitting antennas, in each group where the different transmission levels of their antennas are at least approximately the same.

[0026] Furthermore, the individual transmitted signals can be linearly frequency-modulated, with their center frequency optionally changing successively and preferably linearly, or represent OFDM signals or be generated with pseudorandom fast phase modulation, in particular characterized in that a signal separation into gates corresponding to different distances, so-called distance gates, is realized from the received values ​​resulting for each individual transmitted signal by means of a transformation, and the procedures described in the claims above are applied in the different distance gates. Brief description of the drawings Fig. Figure 1a shows an exemplary embodiment of a radar system; in Fig. Figure 1b shows in detail the arrangement of the 12 transmitting antennas and the 16 receiving antennas, and Fig. Figure 1c shows the two-dimensional equidistant antenna array with 16x8 channels synthesized from eight of the transmitting antennas and all 16 receiving antennas. In Fig. Figure 2 shows the frequency modulation consisting of a sequence of frequency ramps. Fig. Figure 3 shows, for a single object, the magnitude of the spectrum resulting after two-dimensional transformation for a transmit and a receive path in the object distance gate - in Fig. 3a without random phase modulation component, in Fig. 3b with random phase modulation component; inaccuracies occurring in real phase shifters are assumed, i.e. small phase and amplitude errors. Fig. Figure 4 shows the magnitude of the spectrum resulting after two-dimensional transformation for a receive path with all components of the 12 parallel transmit paths. Fig. Figure 5 shows the performance spectrum after non-coherent integration across all reception paths. In Fig. Figure 6 shows the performance spectrum after non-coherent integration over all receive paths and all transmit paths, while still assuming only a single object. Fig. Figure 7 shows the performance spectrum after non-coherent integration over all receive paths and all transmit paths in the case of two objects. Fig. Figure 8 shows, for the now always considered case of two objects, the magnitude of the spectrum which results from the spectrum after two-dimensional transformation by beam shaping via the receive paths (RX beam shaping). In Fig. Figure 9 shows the power spectrum resulting after RX beamforming and non-coherent integration over all transmit path components. Fig. Figure 10 shows the magnitude of the beamformer spectrum in the Doppler gate of the first object, which is generated by two-dimensional digital beamforming via antenna channels formed from all transmit and receive path signals. Fig. 1c results. Fig. Figure 11 shows the power of the maximum of the two-dimensional beamformer spectrum formed per Doppler port over all beam directions. Example of implementation

[0027] The exemplary design of a radar system is considered according to Fig. 1a, whose antenna 1.1 is an antenna arrangement according to Fig. 1b. The radar system has M TX = 12 transmitting antennas TX0-11 for broadcasting signals and M RX= 16 receiving antennas RX0-RX15 for receiving transmitted signals reflected from objects. All antennas (transmitting and receiving) have the same beam pattern in elevation and azimuth. Four transmitting antennas TX0-3 and four transmitting antennas TX4-7 each form vertical arrays; the vertical spacing of the antennas (and thus their phase centers, i.e., radiation centers) is 5λ, where λ = c / 76.5GHz = 3.92mm is the mean wavelength of the transmitted signals in the used frequency band of 76-77GHz and c = 3*10 8m / s is the speed of light. These two arrays, each with four transmitting antennas, are located at the left (TX0-3) and right (TX4-7) edges of the antenna array and have a horizontal spacing of 24λ. Additionally, there are two more transmitting antennas each on the left inner side (TX8-9) and right inner side (TX10-11), which are 3.5λ and 2.5λ, respectively, from the transmitting antennas at the respective edges of the antenna array and are 5.5λ apart vertically. The 16 receiving antennas are divided into two horizontal arrays (RX0-7 and RX8-15), each with eight equidistant antennas, located at the top and bottom of the antenna array and with a vertical spacing of 20λ; the horizontal spacing between the antennas is 3λ.All combinations of the eight transmitting antennas on the right and left (TX0-3 and TX4-7) and the 16 receiving antennas below and above (RX0-7 and RX8-15) synthesize a two-dimensional equidistant array with 16x8 = 128 antenna channels in a horizontal grid of 3λ and a vertical grid of 5λ (see . Fig. 1c), which is used to determine the azimuth and elevation angles of objects. Since these grids represent multiples of half the wavelength λ / 2, ambiguities arise in the angle range -90° to 90°. To resolve the azimuth ambiguities, the antenna channels from transmitting antennas TX1, TX8, TX10, and TX5, as well as all receiving antennas, are used; to resolve the elevation ambiguities, the antenna channels from transmitting antennas TX8-TX11 and all receiving antennas are used.

[0028] As in Fig. As shown in Figure 1a, the transmitted signals emitted by the transmitting antennas are derived from the high-frequency oscillator 1.2 in the 76-77 GHz range, which is controlled by a control voltage V. Steuer Its frequency can be changed. The control voltage is generated in the control elements 1.8, which include, for example, a phase-locked loop or a digital-to-analog converter, which are controlled so that the frequency response of the oscillator corresponds to the desired frequency modulation. The phase of the transmitted signals can be individually adjusted for the M using phase shifter 1.3. TX= 12 transmitting antennas can be set and varied; the phase shifters realize 64 at least approximately uniformly distributed phase values ​​over the phase uniqueness range 0...2π. These phase shifters modulate the transmitted signals from the different transmitting antennas differently, so that parallel transmission on all transmitting antennas, i.e., MIMO operation, is possible because, after demodulation in the received signals, the components originating from the different transmitting antennas can be separated.

[0029] The ones from the M RXThe signals received by the 16 receiving antennas are simultaneously down-converted into the low-frequency range in the real-valued mixers 1.4, along with the signal from the oscillator 1.2. The received signals then pass through the bandpass filters 1.5 with the transfer function shown, the amplifiers 1.6, and the analog-to-digital converters 1.7. Finally, they are further processed in the digital signal processing unit 1.9.

[0030] In order to measure the distance to objects, - as in Fig. 2 shown - the frequency f TX the high-frequency oscillator and thus the transmitted signals change very rapidly and linearly (in T ch = 51.2µs around B ch = 600MHz, where the center frequency f c = 76.5 GHz); this is referred to as a frequency ramp (often also called a "chirp"). The frequency ramps are in a fixed grid T DThe frequency ramps are repeated periodically every 70 µs; in total, there are K = 512 frequency ramps, all of which have the same frequency profile, i.e., the same frequency slope, the same frequency position (i.e., in particular, the same start and center frequencies), and the same duration. In recent years, this type of modulation has become increasingly widespread and established in radar systems for the detection of vehicles' surroundings. It allows for a long sensor range and high velocity resolution (due to the long data acquisition time) as well as high distance resolution (due to the use of a high modulation bandwidth).

[0031] During each frequency ramp k=0,...,K-1, the received signals from each of the M RX = 16 A / D converters l = 2048 times each at intervals of T s = sampled at 25ns (i.e., at 40MHz), with the sampling always starting at the same time relative to the start of the ramp (see Fig. 2); the one in the receive path m RXThe resulting digital samples with index i=0,..., l-1 are combined with s(i,k,m RX ) denotes. 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 travel time corresponding to the maximum distance of interest must be waited for (for a maximum distance of interest of 200m, this corresponds to 1.33µs); it should be noted that here and in the following, distance always refers to the radial distance, and relative velocity to its radial component.

[0032] As is known from the prior art (see, e.g., EP 2 629 113 B1) and can also be easily deduced, the transmission signal from a transmitting antenna m TX caused sampling signal ŝ(i,k,m) TX ,m RX) in the case of a single point-like object at a distance d, a sinusoidal oscillation with respect to index i is represented, which can be described to a very good approximation as follows: s^(i,k,mTX,mRX)=A⋅sin[2π⋅i / l_⋅j0+φv(k)+φPM,TX(k,mTX)+φα˜,TX(mRX)+φα˜,RX(mRX)] with j0=d / (meter)⋅Bch / 150MHz, The frequency of the oscillation is proportional to the object distance d (j0 is a normalized frequency). A radial relative movement of the object to the sensor causes a phase shift contribution φ that changes across the K = 512 frequency ramps. v (k) of the sinusoidal oscillation; for a motion with a constant radial velocity component v, the following results: φv(k)=2π⋅k / K⋅l0 with l0=2KTDvfc / c, That is, a linear phase shift across the frequency ramps k, where the rate of phase change is proportional to the radial relative velocity v of the object. The phase contribution φ PM , TX (k,m TX) in the above reference (1) describes the phase modulation (also referred to as TX modulation) realized with the phase shifters 1.3: φPM,TX(k,mTX)=φPM,lin(k,mTX)+φPM,r(k), which has two components: Firstly, the component that changes linearly over the frequency ramps k φPM,lin(k,mTX)=2π⋅k⋅p(mTX) / P with different rate of change, i.e. modulation speed p(m) across the transmission paths TX ) / P with integer common period P and normalized modulation speeds p(m) considered here as integers TX ) (normalized with respect to the slowest modulation speed 1 / P); this component serves to later separate the components originating from the different transmit paths in the received signals (which will be explained in detail later). Secondly, a random or pseudorandom component φ is generated via the frequency ramps k. PM,r(k), which is identical for all transmission paths and which is generated by a random selection for each frequency ramp k from the 64 phase values ​​of the phase shifters 1.3; this component serves to suppress unwanted effects, which will be derived and explained later. The two trailing phase components φ, independent of the frequency ramps k α̃,TX (m TX ) and φ α̃,RX (m RX ) represent the angles of azimuth α Az and elevation angle α El The phase angles of the object for the various transmit and receive paths are represented. It should also be noted that real-world phase values ​​lie in the range 0...2π, ​​since, due to the cyclic nature of phases, all phase values ​​can be mapped into this range, which mathematically represents a modulo functionality.

[0033] Finally, with regard to reference (1), it should be mentioned that the amplitude A of the received signal is assumed to be independent of the transmit and receive paths, i.e., all transmit and receive paths are assumed to be of equal strength; this assumption has no influence on the following considerations.

[0034] In the digital signal processing unit 1.9, the received signals s(i,k,m) are used. RX ) per frequency ramp k and receive path m RX After multiplication by a suitable window function w1(i), a first discrete Fourier transform (DFT) is performed over the time index i=0,..., l-1, since this corresponds to optimal filtering for the signal shape according to definition (1); the DFT is expediently implemented with a fast Fourier transform (FFT). With the relationship sin(x)=(exp(j^⋅x)−exp(−j^⋅x)) / (2j^), where "exp" denotes the exponential function and ĵ is the imaginary unit, the DFT is obtained, i.e., the spectrum Ŝ1(j,k,m). TX ,m RX ) of the sampling signal ŝ(i,k,m) caused by a transmit path TX ,m RX ) to: S^1(j,k,mTX,mRX)=A / (2j^). [exp(j(φv(k)+φPM,TX(k,mTX)+φα˜,TX(mTX)+φα˜,RX(mRX)))⋅W1(modJ(j−j0)) −exp(−j^(φv(k)+φPM,TX(k,mTX)+φα˜,TX(mTX)y+φα˜,RX(mRX)))⋅W1(modJ(j+j0))] where j=0,..., 1-1 is the scroll variable for the image domain, i.e., the frequency domain of the DFT, and represents the so-called distance gates (because the frequency of the received signal is proportional to the distance), W1(j) is the spectrum of the window function w1(i) used, and "mod J “represents the modulo function of the modulo J. The spectrum W1(j) of the window function has a rather sharp power peak at j = 0, which extends over approximately three frequency values ​​j. The spectrum Ŝ1(j,k,m) caused by a single object TX ,mRX According to equation (6), the signal exhibits two power peaks at frequencies j0 and l-j0 (assuming 0 ≤ j0 ≤ l, which is valid due to non-negative distances and the effect of the bandpass filters 1.5). If 0 ≤ j0 ≤ l / 2, then the power peak at l-j0, i.e., in the upper half of the spectrum, carries no additional information. This is generally true for the upper half of the spectrum because, due to the real-valued input signal, it is the complex conjugate of the lower half; therefore, only the lower half of the spectrum is considered for further processing, i.e., only the frequencies or distance gates j=0...l / 2.

[0035] Before a second DFT over dimension k can be performed, the random phase component φ must be determined. PM , r (k) of the phase modulation φ PM , TX (k,m TX) according to section (4a) (the remaining phase components represent a linear phase response over dimension k, which is a prerequisite for applying the DFT as optimal filtering). For the distance gate range of interest j0=0...l / 2 (i.e., objects located there), the first term in section (6) is relevant (since only the lower half of the spectrum j=0...l / 2 is considered), so that the random phase modulation component is compensated with exp(-ĵ·-φ). PM,r (k)) to multiply; then the result is: S^1,comp(j,k,mTX,mRX)=A / (2j^) [exp(j^(φv(k)+φPM,lin(k,mTX)+φα˜,TX(mTX)+φα˜,RX(mRX)))⋅W1(modJ(j−j0) )−exp(−j^(φv(k)+φPM,lin(k,mTX)+2φPM,r(k)+φα˜,RX(mRX)))⋅W1(modJ(j+j0)) ]. Per distance gate j and receive path m RX After multiplication by a window function w2(k), a second DFT is performed, this time over the frequency ramp index k (preferably again via an FFT); this yields the two-dimensional spectrum. S^2(j,l,mTX,MRX)=A / (2j^)⋅[exp(j^(φα˜,TX(mTX)+φα˜,RX(mRX)))⋅W12(modJ(j−j0),modK(l−l0−p( mTX)⋅K / P))−RPM(modK(l+l0+p(mTX)⋅K / P))⋅exp(−j^(φα,TX(mTX)+φα,RX(mRX)))⋅W1(modJ(j+j0))], where l=0,..., K-1 is the tracking variable for the image domain, i.e., the frequency domain of the second DFT, and represents the so-called Doppler gates (because frequency is proportional to relative velocity, apart from the additional phase modulation component p(m)). TX )·K / P), W 12 (j,l) the two-dimensional spectrum of the two-dimensional window function w1(i)·w2(k) used is (has power peak at j = 0 and l = 0) and R PM (l) the spectrum of the unit vector exhibiting a random phase profile exp(-ĵ2φ PM,r (k)) represents (2φ PM,r (k) randomly assumes values ​​from 32 different phase values ​​uniformly distributed over 2π - the modulo property of the phase is already taken into account); thus, R is also PM(l) itself is noise, which is Rayleigh-distributed. For an object in the distance gate range of interest j0=0...l / 2, the following results in the two-dimensional spectrum: Ŝ2(j,l,m) TX ,m RX ) according to reference (8) of the first term a power peak at distance gate j = j0 and Doppler gate l = mod K (l0+p(m TX )·K / P); the second term generates noise distributed over all Doppler gates k at j = l-j0 (and its immediate neighborhood), where this j lies outside the distance gate range of interest and consideration j=0...l / 2.

[0036] Now, an object above the distance gate range of interest is considered, i.e., in the range j0 = (l / 2 + 1)...(l - 1); due to the relatively low attenuation of the transition range of the bandpass filter 1.5, objects can be received particularly well above j0 = l / 2 – this is referred to as overreach. Then, in the two-dimensional spectrum, Ŝ2(j,l,m) resultsTX ,m RX According to equation (8), the first term generates a power peak at j = j0 above the considered distance gate range j = 0...1 / 2, while the second term now generates noise distributed across all Doppler gates k at distance gate j = l-j0 in the relevant and considered distance gate range j = 0...l / 2. However, this noise will not lead to detection even if it is significantly above the system noise, since detections are only made for power peaks in dimension k that are significantly above the overall noise level there. If it were the random phase modulation component φ PM,r (k) do not exist, so the two-dimensional spectrum Ŝ2(j,l,m) would be TX ,m RX ) read as follows (can be seen from references (7) and (8)): S^2(j,l,mTX,mRX)=A / (2j^). [exp(j^(φα˜,TX(mTX)+φα˜,RX(mRX)))⋅W12(modJ(j−j0),modK(l−l0−p(mTX)⋅K / P)) −exp(−j^(φα,TX(mTX)+φα˜,RX(mRX)))⋅W12(modJ(j+j0),modK(I+I0+p(mTX)⋅K / P))].

[0037] This would mean that an overrange, i.e. an object in the range j0=(l / 2+1)...(l-1), would now lead to a power peak at the distance gate j = l-j0 in the distance gate range of interest and consideration j=0...l / 2 due to the second term; thus, a detection would be erroneously formed which not only has an incorrect (too short) distance, but whose measured values ​​for relative velocity and angle are also incorrect (because of incorrect signs of the contributing quantities in the second term of equation (9)).

[0038] Due to the random phase modulation component φ PM , r (k) false detections due to overreach are thus prevented; the noise generated instead has an average power of approximately 26dB below the power peak that would occur without a random component (the 26dB results from the DFT integration gain of 10·log 10(K = 512) = 27dB minus approximately 1dB window loss). It should also be mentioned that the above considerations do not apply to a binary phase shifter, i.e., one with only the two states 0 and π; because then the phase component is 2φ. PM,r (k) in the second term of reference (7) is always 0 (phase 2π corresponds to phase 0) and therefore has no effect, so that the second term of the two-dimensional spectrum Ŝ2(j,l,m) TX ,m RX ) does not represent noise, but according to reference (9) also generates a power peak, which then leads to the undesired over-range effect. Therefore, at least three different phase values ​​are necessary to convert over-ranges into noise.

[0039] Problems caused by over-propagation arise from a convolution effect, i.e., the mapping of frequencies to other frequencies. If frequencies retain certain power components at other frequencies, this is referred to as convolution – such effects are collectively known as spectral convolution effects. This convolution effect occurs with very close objects; for example, consider an object at distance gate j0 = 0.5 (it should be noted that signals from such close objects can be received despite the high attenuation of the 1.5 bandpass filters, because the short distance results in a very strong signal at the receiving antennas). Without the random phase modulation component φ PM , r (k) would then be in the two-dimensional spectrum Ŝ2(j,l,m) TX ,m RX ) according to reference (9) at distance gate j = 0, proportions of both terms are effective (the power peak with the shape of the two-dimensional window spectrum W). 12(It has a certain width and typically extends across three distance gates); the first term provides the correct information, while the second term represents incorrect information. These erroneous components from the second term can either lead to distortions of the object's measured values ​​(if they overlap with the real components of the first term and then, for example, influence distance interpolation) or to non-real detections, i.e., ghost detections (if they do not overlap with the real components of the first term and thus form independent power spikes). Due to the random phase modulation component φ PM , r (k) represents the second term of the two-dimensional spectrum Ŝ2(j,l,m) TX ,m RX) according to reference (8) noise, so that no ghost detections occur; and because the noise is far below the real power peaks of the first term, the influence on the measured values ​​of the object (distance, relative velocity and angle) is also negligibly small.

[0040] Negative effects of spectral convolutions are thus caused by the random phase modulation component φ. PM,r (k) avoided.

[0041] A fundamental problem with phase modulation is that the means used, especially phase shifters, are never ideal and therefore always exhibit certain errors. The 64 phase values ​​of the phase shifters considered here, which are ideally uniformly distributed over the phase range 0...2π = 0...360°, should have a standard deviation of 5°; and additionally, the amplitude of the realized phasors should also have a standard deviation of 10%. The modulation period used should be P = 32, and for the considered transmission path m TX,0 Let the modulation speed be p(m). TX,0) = 8. Without a random modulation component, the four target phase values ​​0, 90°, 180°, and 270° are to be repeated periodically according to equation (4b), which are generated with the phase shifter indices 0, 16, 32, and 48; these should have actual phase values ​​deviating from the target values: -5°, 91°, 185°, and 263°, and amplitude values ​​1.02, 1.1, 0.85, and 1.11. For an object at a distance gate j0 < l / 2 and a Doppler gate l0 = 168, the resulting two-dimensional spectrum at the object distance gate j0 and a transmission path m is TX,0 and a reception path m RX,0 , also Ŝ2(j0,l,m TX,0, m RX,0 ), in Fig. Figure 3a shows the power output in dB; in addition to the regular power peak at l = l0+8K / P = 296, further power peaks occur at a rate of 8·K / P = 128 (harmonic frequencies with a period of P / 8 = 4), which are generated by the non-ideal phase shift values ​​and their periodic repetition and can lead to false detections, i.e., ghost detections. If the harmonics coincide with the correct power peaks of other transmitting antennas, they can distort the angle calculation result. It should also be mentioned that in Fig. 3a the noise that lies significantly below the power peaks comes from the system noise, which is not represented in the formulas above.

[0042] When using the random superimposed phase component φ PM,r (k) of the phase modulation φ PM,TX (k,m TX,0) according to reference (4a), four phase values ​​are no longer repeated periodically, but quasi-random phase values ​​are used (the linear phase component is no longer visible); thus, the phase modulation errors (phase and amplitude) are also no longer periodic, but quasi-random, so that in the spectrum Ŝ2(j0,l,m TX,0, m RX.0 ) after Fig. 3b No more harmonic power peaks occur – their energy is distributed in noise. The superimposed random phase modulation component φ PM,r (k) thus avoids ghost detections due to the inaccuracies that always occur in the phase modulation means and therefore also allows the use of rather poor phase shifters, which can lead to a reduction in costs.

[0043] In addition to the advantages already presented due to the random phase modulation component φ PM , r(k) this also causes emissions via the receiving antennas (due to their limited TX isolation), internal couplings between transmit and receive paths, and interference from other radar systems to be decorrelated and thus converted into noise - they therefore cannot significantly degrade the measurement quality of objects and cannot generate ghost detections.

[0044] So far, only the contribution from one transmission path has been recorded. TX considered. That of all M TX = 12 transmission paths generated a two-dimensional total spectrum S2(j,l,m) RX ) is the sum of the individual contributions represented in the formulas above (i.e., the sum over m). TX = 0,...,M TX -1). Because the random phase modulation component φ PM,r (k) constant over all M TXSince there are 12 transmission paths, its compensation and the second DFT only need to be calculated once for all transmission paths, not separately for each one (the latter would be necessary if the random phase modulation component differed between the transmission paths, which would result in significantly increased computational effort). The modulation used to distinguish the transmission paths, i.e., the different linear phase response φ between the transmission paths, PM,lin (k,m TX This also does not require multiple second DFTs, as it does not need to be compensated for before the DFT because, without compensation in the DFT, it simply leads to a corresponding shift in the power peaks. After this second DFT, performed jointly for all transmission paths (i.e., only once), the following results are then obtained: M TX= 12 power peaks at the positions corresponding to the transmit paths (i.e., their linear phase response); this allows the separation of the components caused by the different transmit paths, which is necessary for MIMO operation (i.e., using all combinations of transmit and receive antennas for angle formation when the transmit antennas are operating in parallel) – this will be explained in detail later. The advantages and effects of the random phase modulation component φ shown above PM,r (k) naturally remain within the overall spectrum of all M TX = 12 transmission paths are obtained (they apply to each individual transmission path and therefore also to their sum).

[0045] So far, only a single point-like object has been considered. The above considerations remain valid even in the case of multiple and / or extended objects (since this simply means a linear superposition of several individual signals).

[0046] So far, a real-valued mixer has been considered. With a complex-valued mixer (also called an IQ mixer), ideally there is only one power peak; however, in reality, IQ generation is not quite perfect, so there is also a smaller power peak at a negative frequency, which can lead to ghost detections or distortion of measured values. This is due to the random phase modulation component φ. PM , r (k) This power peak is also converted into noise, so that negative effects from it are avoided.

[0047] Now, the radar system is to be... Fig. Section 1 describes how, in MIMO operation (i.e., using all combinations of transmitting and receiving antennas for angle formation when the transmitting antennas are operating in parallel), the separation of the components caused by the different transmission paths can be achieved. After the second DFT (i.e., after a two-dimensional transformation) performed jointly (i.e., only once) for all transmission paths, the spectrum S2(j,l,m) is obtained for a single point-like object with a distance gate j0 < l / 2 and a Doppler gate l0. RX ) in the receive path m RX as the sum of the individual transmission paths m TX belonging sub-spectra Ŝ2(j,l,m TX ,m RX ) according to designation (8): S2(j,l,mRX)=A / (2j^). summTX[exp(j^(φα˜,TM(mTX)+φα˜,RX(mRX)))⋅W12(modJ(j−j0),modK(l−l0−p(mTX)⋅K / P))] with mTX=0,…,MTX−1 , where “sum mTX “the sum over all m TX = 0,...,M TX-1 means; the second term in reference (8), which generates noise distributed over all Doppler gates l at j = l-j0 (and its immediate neighborhood), is omitted here because this j lies outside the distance gate range of interest and consideration j=0...l / 2.

[0048] This spectrum S2(j,l,m RX ) indicates distance gate j0 of object M TX = 12 power peaks at the positions corresponding to the transmission paths (i.e., their linear phase response) and is in Fig. 4. Example values ​​in dB and scaled for a receive path m RX,0 depicted (where the scaling is such that the power peaks without noise would be exactly at 0dB); the object is located at the Doppler gate l0 = 168, which is for the M TX = 12 signal paths used normalized modulation speeds p(m) TX ) are p(mTX=0,…11)=[04681014172022252729] and the received signal exhibits system noise, which results in a noise floor in the spectrum that is significantly below the power peaks. For each of the M RX The power peaks have the same position and shape across all 16 receive paths. The phases of the complex values ​​at the line peaks vary across the associated transmit and receive paths – they depend on the azimuth and elevation angle of the object.

[0049] The positions resulting from section (10), i.e., double goals l0(mTX)=modK(l0+p(mTX)⋅K / P), mTX=0,…,MTX−1, The power peaks depend on the generally unknown and determinable Doppler gate l0 of the object. To determine this Doppler gate l0 of the object and assign the power peaks to the transmission paths m TX To determine this, according to the state of the art, a non-coherent integration is first performed over the M that is contained within it. RX = 16 resulting spectra S2(j,l,m) RX) carried out according to reference (10): PRX−NC(j,l)=summRx[|S2(j,l,mRX)|] =|A / 2|2⋅summRX[|summTX[exp(j^(φα˜,TX(mTX)+φα˜,RX(mRX))). W12(modJ(j−j0),modK(I−I0−p(mTX)⋅K / P))]|2] with mRX=0,…,MTX-1, Here, as is generally customary, the incoherent integration is formed as the sum of the powers. The power spectrum P resulting from the object removal gate j0 serves as an example. RX-NC (j0,l) in Fig. 5 in dB and displayed on a scale. As also shown by comparing the Fig. 4 and Fig. As can be seen in Figure 5, the non-coherent integration significantly reduces the variance within the noise floor; this allows for better detection of small and / or distant objects whose power peaks lie within or only slightly above the noise floor. For the identification of the object Doppler gate l0 and the correct assignment of the power peaks to the transmit paths m TXCan a pattern comparison between the expected raster p(m) TX )·K / P of the power peaks and the measured power spectrum P RX-NC (j0,l) occurring power peaks are performed; pattern matching is possible because the modulation speeds p(m TX ) are chosen such that they have a non-periodic staffing with respect to the modulation period P = 32: for each q=0,…,P−1:{modP(q+p(0,…,MTX−1))}≠{modP(p(0,…,MTX−1))}, where “{.}” denotes the set of the respective M TX Values ​​are designated.

[0050] One approach according to the invention is that for each possible Doppler gate l0 = 0,...,K-1, the values ​​of P are non-coherently determined. RX-NC (j,l) in the expected grid l = l0(m TX ) the peak power output is integrated; with P RX-NC (j,l) according to reference (13) and l0(m TX ) according to reference (12) it follows: PRX−NC,TX−NC(j,l)=summ_TX[PRX−NC(j,modK(l+p(m_TX)⋅K / P))] =|A / 2|2⋅summ_TX[summRX[|summTX[exp(j^(φα˜,TX(mTX)+φα˜,RX(mRX))). W12(modJ(j-j0),modK(l+p(m_TX)⋅K / P−l0−p(mTX)⋅K / P))|2]] with m_TX=0,…,MTX−1 and mTX=0,…,MTX−1; Here, the loop variable l is again used for the possible Doppler ports l0 = 0,...,K-1, and since summation over the transmitting antennas is performed twice in equation (15b), two different loop variables m must be used. TX and m TX can be used. The power spectrum P resulting from the object removal gate j0 is an example. RX-NC,TX-NC (j0,l) in Fig. The image is scaled to 6 (maximum), with the object still positioned at the Doppler gate l0 = 168. A total of 30 power peaks now occur, lying on an equidistant grid with a grid length K / P = 16 (the fact that two grid points are unoccupied will be explained later), and exhibiting different level levels. The maximum level occurs at the Doppler gate l0 = 168 of the object, as this is where the power peaks of all M converge. TX = 12 transmission paths are added together (the power P there) RX-NC,TX-NC (j0,l0) is therefore 12 times higher than the power peaks in P RX-NC (j0,l0(m TX ))); the other 29 peak performances of P RX-NC,TX-NC (j0,l) are lower because of the choice of modulation speeds p(m) TX ) according to reference (14) only over a portion of the peak power of P RX-NC(j0,l) is added together – there are a maximum of six power peaks, a minimum of four. Thus, the Doppler gate l0 of the object can now be identified as the position of the highest power peak; the object distance gate j0 results from the position of the power peak in dimension j. The complex channel values ​​Ŝ2(j0,l0,m) required for angle determination TX ,m RX ), so the M RX ·M TX = 192 combinations of all transmit and receive paths, can be represented by the two-dimensional spectra S2(j,l,m) RX ) at the point l = l0(m TX ) can be taken from reference (12): S^2(j0,l0,mTX,mRX)=S2(j0,l0(mTX),mRX)=S2(j0,modK(l0+p(mTX)⋅K / P),mRX)

[0051] The described method fails if the power peaks in the power spectrum P resulting from two-dimensional non-coherent integration RX-NC,TX-NC(j,l) not lie above the noise floor or if there are multiple objects at the same distance that are in the same grid l = G g with Gg={g+(0,…,P−1)⋅K / P} with g=0,…,K / P−1

[0052] Generate line peaks. For example, consider two point-like objects at the same distance gate j0, or at the Doppler gates l. 0,1 = 168 and l 0,2 = 168+9*16 = 312, at the azimuth angles α Az,1 = 4.78° and α Az,2 = 14.5° and the elevation angles α EI,1 = 4.3° and α El,2 = -4.3° considered, with the additional object 2 at Doppler gate l 0,2 = 312 should generate a received signal that is 6dB lower than the previously considered object 1 at l 0,1 = 168; the power spectrum P resulting from two-dimensional non-coherent integration RX-NC,TX-NC (j0,l) is in Fig. 7 scaled (with the same scaling as in Fig. 6) shown. The power peaks of both objects lie in the same grid. I=G8={8+(0,…,31)⋅16}, That is, they overlap and are therefore inseparable. The highest power peak occurs at the Doppler gate l. 0,1 = 168 of object 1 with a stronger received signal, so that this object is identifiable in the measurement. The other object 2 is not identifiable even when the second highest power peak is used – this occurs at Doppler gate l = 200, where no object is located at all.

[0053] To solve this problem, the following procedure according to the invention is used: The spectrum S2(j,l,m) resulting after distance and Doppler transformation RX In contrast to the above, the signal is now coherently integrated across the receive paths, for which digital steel forming is applied. Since the arrangement of the RX antennas according to Fig. 1b a two-dimensional array with M each RX,Az= 8 antennas in horizontal direction and M RX,El Since the array represents two antennas in a vertical direction, two-dimensional digital beamforming is applied (for azimuth and elevation); for this purpose, S2(j,l,m) RX ) the M RX = 16 receive paths m RX To arrange 0,...,15 as a two-dimensional array: S2,RX-BF(j,l,mRX,Az=0…7,mRX,EI=0)=S2(j,l,mRX=0…7) S2,RX−BF(j,l,mRX,Az=0…7,mRX,EI=0)=S2(j,l,mRX=0…7)

[0054] For two-dimensional beamforming (hereinafter also referred to as RX-Beamforming with the abbreviation RX-BF), a two-dimensional DFT preferably implemented via FFTs with 100% zeropadding (i.e., extension by M) is used. RX,Az = 8 or M RX,El (= 2 zeros) is used so that it has the dimension (N RX-BF,Az =16)×(N RX-BF,El =4), and the window functions w 3,RX-BF (m RX,Az ) for azimuth (Chebyshev window with 30dB sidelobe suppression) and w4,RX-BF (m RX,El ) for elevation (because of only two elevation channels, w 4,RX-BF (m RX,El ) constant) used: SRX−BF(j,l,nRX−BF,Az,nRX−BF,EI)= DFTNRX−BF,EI[w4,RX−BF(mRX,EI)⋅DFTNRX−BF,Az[w3,RX−BF(mRX,Az)⋅S2,RF−BF(j,l,mRX,Az,mRX,EI)]], where n RX-BF,Az = 0,...,N RX-BF,Az -1 the loop variable for the image domain, i.e. the frequency domain of the third discrete Fourier transform DFT NRX-BF,Az is and represents the so-called RX-BF azimuths (because frequency is proportional to the electrical azimuth angle) and n RX-BF,El = 0,...,N RX-BF,El -1 the loop variable for the image domain of the fourth discrete Fourier transform (DFT) NRX-BF,El represents the so-called RX-BF elevation stores; RX-BF azimuth and RX-BF elevation stores are also referred to as RX-BF azimuth and RX-BF elevation beams in the following.

[0055] With the four-dimensional spectrum W 1234,RX-BF (j,l,n RX-BF,Az , nRX-BF,El ) the four-dimensional window function w1(i)·w2(k)·w 3,RX-BF (m RX,Az )·w 4,RX-BF (m RX,El ), which is also formed with 100% zeropadding in the third and fourth dimensions and which has its maximum at indices 0, and with S2(j,l,m RX ) according to reference (10) the four-dimensional spectrum S is obtained RX-BF (j,l,n RX-BF,Az ,n RX-BF,El ) for a point-like object in the RX-BF azimuth beam n RX-BF,Az,0 and RX-BF elevation beam n RX-BF,El,0 to SRX−BF(j,l,nRX−BF,Az,nRX−BF,El)= A / (2j^)⋅summTX[exp(j^⋅φα˜,TX(mTX))⋅W1234,RX−BF(modJ(j−j0),modK(I−I0−p(mTX)⋅K / P), modNRX−BF,Az(nRX−BF,Az−nRX−BF,Az,0),modNRX−BF,El(nRX−BF,El−nRX−BF,El,0))] with mTX=0,…,MTX−1; For the sake of completeness, it should also be mentioned that, firstly, the RX-BF azimuth beam n RX-BF,Az,0 and the RX-BF elevation beam n RX-BF,El,0 as follows with the azimuth angle α Az and the elevation angle α Elrelated to the object: nRX−BF,Az,0=modNRX−BF,Az(sin(αAz)⋅3⋅NRX−BF,Az) nRX−BF,EI,0=modNRX−BF,EI(sin(αEI)⋅20⋅NRX−BF,EI), where the factors 3 and 20 come from the antenna spacings 3λ (in the horizontal direction) and 20λ (in the vertical direction), and secondly for the angle-related phase φ α̃,RX (m RX ) in reference (10) applies: φα˜,RX(mRX=0…7)=2π⋅mRX⋅nRX−BF,Az,0 / NRX−BF,Az φα˜,RX(mRX=8...15)=2π⋅((mRX−BF,Az,0−8)⋅nRX−BF,Az,0 / NRX−BF,Az+nRX−BF,EI,0 / NRX−BF,EI)

[0056] The spectrum S resulting after four-dimensional transformation RX-BF (j,l,n RX-BF,Az, n RX-BF,El ) of a single object, according to the above reference (19b), therefore, M TX = 12 power peaks, which occur at the distance gate j0, the RX-BF azimuth beam n RX-BF,Az,0 , the RX-BF elevation beam n RX-BF,El,0 and the 12 different Doppler gates mod K (l0+p(m TX)·K / P). For the case of two objects considered above, this is the case at their distance gate j0 and their identical RX-BF elevation beam n. RX-BF,El,0 = 2 resulting spectrum S RX-BF (j0,l,n RX-BF,Az ,n RX-BF,El,0 ) in Fig. 8 shown in magnitude in dB and scaled (to maximum) (although the two elevation angles α El,1 = 4.3° and α El,2 Since the objects differ by -4.3°, they fall into the same RX-BF elevation beam n due to the large vertical distance and the associated high angular ambiguity according to reference (20b). RX-BF,El,0 = 2). Since the two objects have different azimuth angles α Az,1 = 4.78° and α Az,2 = 14.5°, they are located at the different RX-BF azimuth beams n RX-BF,Az,0,1 = 4 and n RX-BF,Az,0,2 = 12 and are therefore separated in the spectrum; at each of the two RX-BF azimuth beams, M TX = 12 power peaks, specifically at the Doppler gates I=I0,1 / 2(mTX)=modK(l0,1 / 2+p(mTX)⋅K / P)with the object Doppler gates I0,1=168 and l0,2=312.

[0057] Analogous to reference (15), the RX-BF spectrum now shows S RX-BF (j,l,n RX-BF,Az, n RX-BF,El ) for each possible Doppler gate l0 = 0,...,K-1 incoherently over the values ​​in the expected grid l = l0(m TX ) = mod K (l0+p(m TX )·K / P) of the power peaks integrated: PRX−BF,TX−NC(j,l,nRX−BF,Az,nRX−BF,EI)=summ_TX[SRX−BF(j,modK(l+p(m_TX)⋅K / P),nRX−BF,Az,nRX−BF,EI)|2] with m_TX=0,...,MTX−1; Here, the loop variable l is again used for the possible Doppler gates l0 = 0,...,K-1. For a single object, this results with reference (19b): PRX−BF,TX−NC(j,l,nRX−BF,Az,nRX−BF,EI)=|A / 2|2⋅summ_TX[|summTX[exp(j^⋅φα˜,TX(mTX)). W1234,RX−BF(modJ(j−j0),modK(l+p(m_TX)⋅K / P−l0−p(mTX)⋅K / P), modNRX−BF,Az(nRZ−BF,Az−nRX−BF,Az,0),modNRX−BF,EI(nRX−BF,EI−nRX−BF,EI,0))]|2]with m_TX=0,...,MTX−1 and mTX=0,...,MTX−1.

[0058] For the above example of the two objects, this is the case at their distance gate j0 and their RX-BF elevation beam n. RX-BF,El,0 = 2 resulting performance spectrum P RX-BF,TX-NC (j0,l,n RX BF,Az ,n RX BF,El,0 ) in Fig. 9 scaled representation; the 30 power peaks occurring in the same grid l = G8 are now separated, since they have different RX-BF azimuth beams n RX-BF,Az,0,1 = 4 and n RX-BF,Az,0,2 = 12 occur. Therefore, the Doppler portal l can now be determined for both objects. 0,1 = 168 or l 0,2= 312 from the position of the highest power peak in the respective RX-BF azimuth beam; thus, both objects are detectable and – as will be shown later – their angles can be determined without mutual interference. The inventive combination of coherent integration over the receive paths and subsequent non-coherent integration over the transmit paths thus solves the problem that arises with completely non-coherent integration (i.e., over both receive and transmit paths).

[0059] The inventive approach also has advantages with weak signals, since coherent integration allows small signals to be distinguished from noise significantly better than non-coherent integration: With coherent integration, the distance between the signal and the average noise is increased (in the case of N signals, by a factor of N, i.e., in the case of M RX= 16 reception paths by a factor of 16, which corresponds to 10·log10(16) = 12dB); with non-coherent integration, the distance to the mean noise remains constant, only the variance within the noise floor is reduced (i.e., the noise assumes a more planar shape, as e.g., by comparing the Fig. 5 and Fig. 6 (Spectra with incoherent integration) with Fig. 4 (spectrum before non-coherent integration) can be seen, i.e., the statistically caused peaks in the noise floor decrease), which also leads to smaller signals being better distinguished from noise, but this effect is a few dB less than the gain in the signal-to-noise ratio resulting from coherent integration.

[0060] As mentioned above, object detection requires differentiating the power peaks they generate from noise or noise peaks; for this purpose, the noise level must be determined. Since the noise level can vary not only across the distance gates but also across the RX-BF beams (e.g., due to interference and phase noise effects), it is measured at each distance gate and at each of the N RX-BF,Az xN RX-BF,El = 64 RX-BF beams are individually determined. This is done using the respective K = 512 values ​​of the RX-BF power spectrum P resulting from non-coherent TX integration. RX-BF,TX-NC (j,l,n RX-BF,Az ,n RX-BF,El), i.e., using the Doppler dimension l, a so-called ordered statistic (OS) is performed, and the smallest value, e.g., the 40% smallest (i.e., the 205 smallest), is sought. The detection threshold for distinguishing between object power peaks and noise peaks is set approximately 4.5 times higher, or about 6.6 dB, above the noise level value determined in this way (then the probability that statistically induced noise peaks lie above the detection threshold and thus falsely lead to a detection is low enough).

[0061] To determine the noise level, P can be used instead of the RX-BF power spectrum. RX-BF,TX-NC (j,l,n RX-BF,Az ,n RX-BF,El ) after non-coherent integration over the transmission paths also the RX-BF power spectrum |S RX-BF (j,l,n RX-BF,Az ,n RX-BF,El )| 2before this non-coherent integration, which has two advantageous aspects: Firstly, the noise level can be determined in parallel with the non-coherent TX integration, provided the computer used has parallel processing units, resulting in reduced latency. Secondly, only 12 power peaks per object are generated in the power spectrum before non-coherent TX integration, whereas there are 30 power peaks per object in the power spectrum after non-coherent TX integration;In particular, when power peaks from multiple objects occur in an RX-BF beam, the noise level is overestimated by the ordered statistic. This effect is greater the more power peaks occur in the spectrum and thus per object. Therefore, using the RX-BF power spectrum after non-coherent TX integration can lead to a greater overestimation of the noise level than using the RX-BF power spectrum before non-coherent integration. This can lead to the failure to detect small objects (whose power peaks then lie below the noise threshold increased by the overestimation of the noise). Thus, using the power spectrum before non-coherent TX integration for noise level estimation results in more robust detection in the case of multiple objects at similar distances with a similar RX-BF beam.

[0062] If the noise level is estimated from the power spectrum before incoherent integration, it must be translated to the noise level after incoherent integration (since detection, i.e., checking for the noise threshold, takes place after incoherent integration); assuming an approximately constant and uncorrelated noise across the Doppler ports, the translation factor M TX = 12 (so in the case without the noise overestimation due to power peaks, the noise level after incoherent integration is 12 times higher than before).

[0063] In addition to the noise level, other criteria can be used to determine the detection threshold. In particular, it is useful to limit the dynamic range per RX-BF beam and per distance gate; that is, the detection threshold may be at most a certain value below the maximum value obtained in the respective RX-BF beam (via the K = 512 values) or in the respective distance gate (via the 512.64 values). This prevents, for example, sidelobes from window functions from causing false detections.

[0064] As described above and in Fig. 9 are shown in the RX-BF performance spectrum P RX-BF,TX-NC (j,l,n RX-BF,Az, n RX-BF,ElAfter non-coherent TX integration, a total of 30 power spikes per object are generated. The highest power spike is at the object's Doppler gate l0 and is referred to as the valid or correct power spike. The other 29 lower-level power spikes are referred to as invalid power spikes or TX modulation artifacts. To detect objects, the valid power spike must be identified, while the invalid power spikes, i.e., the TX modulation artifacts, are to be ignored, as they do not indicate the presence of an object at their Doppler gates. Thus, a simple approach is to use a grid G g According to reference (17), only the highest power peak is classified as valid per RX-BF beam (and of course per distance gate, which will not be mentioned again below), but this leads to the situation that in the case of two objects whose power peaks are in the same RX-BF beam and in the same grid Gg In such cases, only the stronger object (i.e., the one with the stronger received signal) is detected. The approaches described below can be used to reduce this problem.

[0065] In the case of a single object, the largest TX modulation artifacts lie approximately 3 dB below the maximum, i.e., correct, power peak (provided this is significantly above the noise level); therefore, at least the second-largest power peak can also be considered valid if its level is only slightly below the largest (whereby the permissible level difference can depend on the distance from the noise level) – however, it must be taken into account that if excessively large level differences are allowed (e.g., the 3 dB itself), even with power peaks far above the noise level, an incorrect identification of the Doppler gate of the smaller object can occur, since the objects in the complex RF-BF spectrum S RX-BFoverlap, and thus the power peaks there can mutually reinforce or weaken each other depending on the respective phase position.

[0066] As in Fig. As can be seen in 9, the respective grid G g The power peaks do not occupy all 32 positions; rather, the two neighbors (left and right neighbors) to the correct, i.e., largest, power peak are unoccupied. Therefore, if power peaks also occur at these two positions, they must originate from a second object. They can be considered valid if the second largest power peak occurs there, or if their level is even a few dB below the second largest power peak. The fact that the two neighbors to the correct power peak in grid G g Having the TX modulation artifacts unoccupied has the following two advantages in particular: - The detection of pedestrians in the presence of other strong reflections from the stationary environment (e.g., buildings, trees, traffic signs, parked vehicles, etc.) is important; due to the comparatively low speed of a pedestrian, their Doppler gate and the Doppler gate of infrastructure reflections can differ by a grid spacing K / P = 16, which, due to the comparatively low reflectivity of pedestrians, can lead to the detection of reflections in the same grid G ​​being incorrect. g The pedestrian's peak power output is significantly lower than that of the infrastructure - but because the correct peak power output of the pedestrian is now located at an unoccupied grid neighbor position of the significantly larger peak power output of the infrastructure, the pedestrian can be detected. - So that the power peaks of two objects with different relative velocities do not occur in the same grid G ​​over many radar cycles gIt is advantageous to choose period T. D the K frequency ramps (see Fig. 2) changed from radar cycle to radar cycle (i.e., from data acquisition to data acquisition) - i.e., instead of T D = 70µs will be used for the next K frequency ramps, e.g. T D = 72µs is used. This changes the ratio to the grating spacing of 16 of the gratings G. g The corresponding relative velocity difference is reduced accordingly, i.e., by 2.8% in this numerical example, so that with two objects of different relative velocity, it is impossible for them to be in exactly the same grid G ​​in two successive radar cycles. g lie (at least apart from rounding effects for the integer grids); the greater the relative velocity difference of the two objects, the further they shift due to a changed T Dtheir two grids relative to each other. However, if the relative velocity difference is only about one grid spacing, then a moderate change in the ramp repetition time T can D It is not sufficient to clearly separate the power peaks because power peaks typically extend over about three Doppler gates (due to their window function used in the FFT); however, since the two grid neighbors of the correct power peaks are each unoccupied, a lack of grid separation does not pose a significant problem.

[0067] So that in grid G g If the two adjacent positions to the correct power peak are not occupied, the normalized modulation speeds p(m) must be used. TX ) are chosen so that they differ by at least 2 (also taking into account the periodic wrap-around caused by the modulo property): for each(mTX,1=0,...,MTX−1)≠(mTX,2=0,...,MTX−1): modP(p(mTX,1)−p(mTX,2),P)∈{2,...P−2}; This condition is fulfilled for the example considered here according to reference (11). Instead of a completely omitted occupancy of the two adjacent positions, a significantly reduced occupancy compared to the other TX modulation artifact positions could also be used; for example, for exactly two modulation speeds p(m) TX ) are allowed to be directly next to each other, i.e., differing by only one.

[0068] For the most robust detection possible, the modulation speeds p(m) are used. TX) chosen such that, for a single point-like object, the difference between the largest and second-largest power peaks is as large as possible, possibly subject to further boundary conditions, e.g., the boundary condition considered here that the neighboring positions to the correct power peak are not occupied; for this purpose, preferably all possible combinations of modulation speeds are searched, cleverly avoiding the fact that the same set of modulation speeds (only with different order) is examined multiple times, and combinations that can be excluded a priori (here, those in which two p(m) TX (less than a distance of 2) are not considered (otherwise 12 would be considered). 32 (to search combinations).

[0069] It should also be mentioned that in incoherent integration via the M TX = 12 transmission paths in total M TX 2= 144 power peaks are used, i.e., the sum of the power peaks in the resulting power spectrum P RX-BF,TX-NC is 144 times the power of the 12 individual, assumed to be equally high power peaks before non-coherent TX integration, i.e. in S RX-BF . Since at the correct peak performance of P RX-BF,TX-NC M TX The 12 peak power levels are summed up, the other M are distributed TX ·(M TX -1) = 132 power peaks on the TX modulation artifact positions, so that a theoretical lower limit for the minimum number of summed power peaks in the largest TX modulation artifact can be determined - in the example under consideration it would be five (132 / 29 increased to the next integer); however, as in the example considered here, there can be no combination that realizes this theoretical minimum (here the best combinations have six power peaks in the two largest TX modulation artifacts).

[0070] For the non-coherent integration, power values ​​were used above, as is generally customary. Alternatively, and according to the invention, magnitudes (i.e., absolute values) can also be used; in particular, for very weak objects, this leads to a slightly reduced probability that the largest power peak is not the actual power peak, but rather a TX modulation artifact, resulting in a false identification. Whether calculating power or magnitudes requires less computational effort depends on the computing platform used.

[0071] The above describes how power peaks can be checked against a detection threshold (derived from the noise level and, if necessary, additional criteria, especially for limiting the dynamic range) and to ensure they are not TX modulation artifacts. This check is advantageously performed at each distance gate not only for power peaks but for every value, i.e., across all Doppler gates and RX-BF beams (in particular, it is not checked whether the value represents a local maximum in Doppler gate dimensions). The result is also needed in a later processing step and is stored as a bit field of length 64 for each Doppler gate (bits with the value one indicate that the corresponding RX-BF beam is valid in the respective Doppler gate, i.e., the value there is above the detection threshold and is not a TX modulation artifact).All Doppler gates in which at least one of the 64 RX-BF beams passes the detection threshold test and shows no TX modulation artifacts are then declared valid. If the local maximum were checked at this stage, objects that only become visible after full digital beamforming (i.e., using the transmit paths) could be missed; therefore, the local maximum test, i.e., the power peak test, is only performed after full digital beamforming.

[0072] The next section describes how the valid Doppler gates determined in this way are further processed. In the example above, with two objects located at the same distance gate j0 but different RX-BF beams, the Doppler gates l 0,1 = 168 as well as l 0,2 = 312 and because power peaks extend across three Doppler ports there, the quantity L results v the valid double goals to Lv={167,168,169,311,312,313}.

[0073] First, the quantity L will be determined. v The valid Doppler gates are extended to include their direct neighbors, i.e., the Doppler gates before and after them, provided these are not already valid themselves – this is necessary to later check for a local maximum in the Doppler gate dimension; this is called the set L. p the processed Doppler gates are referred to as: Lp={166,167,168,169,170,310,311,312,313,314}.

[0074] As in Fig. As shown in Figure 1c and executed at the beginning, the combinations of the eight transmitting antennas on the right and left (TX0-3 and TX4-7) and the 16 receiving antennas below and above (RX0-7 and RX8-15) synthesize a two-dimensional equidistant array with (M TXRX,Az =16)×(M TXRX,El =8) = 128 antenna channels in a horizontal grid of 3λ and a vertical grid of 5λ. The complex channel values ​​Ŝ2(j0,l) p ,m TX ,m RX) these antenna channels become the two-dimensional spectra S2(j,l,m) resulting after distance and Doppler transformation. RX ) to the Doppler gate to be processed in each case p corresponding places l = l p (m TX ) taken from reference (12): S^2(j0,lp,mTX,mRX)=S2(j0,lp(mTX),mRX)=S2(j0,modK(lp+p(mTX))⋅K / P,mRX) with mRX=0,...,15 and mTX=0,...,7.

[0075] These 128 values ​​are now subjected to two-dimensional digital beamforming (in contrast to the RX beamformer above, coherent integration is now performed not only on the signals from the receive paths, but also on the transmit paths); for this purpose, they must be arranged according to the two-dimensional antenna array: S^2,TXRX−BF(j0,lp,mTXRX,Az=0...7,mTXRX,EI=0...3)=S^2(j0,lp,mTX=0...3,mRX=0...7)T S^2,TXRX−BF(j0,lp,mTXRX,Az=8...15,mTXRX,EI=0...3)=S^2(j0,lp,mTX=4...7,mRX=0...7)T S^2,TXRX−BF(j0,lp,mTXRX,Az=0...7,mTXRX,EI=4...7)=S^2(j0,lp,mTX=0...3,mRX=8...15)T S^2,TXRX−BF(j0,lp,mTXRX,Az=8...15,mTXRX,EI=4...7)=S^2(j0,lp,mTX=4...7,mRX=8...15)T.

[0076] For two-dimensional beamforming (hereinafter also referred to as TXRX-Beamforming with the abbreviation TXRX-BF), a two-dimensional DFT preferably implemented via FFTs with 100% zeropadding (i.e., extension by M) is used. TXRX,Az = 16 or M TXRX,El (= 8 zeros) is used so that it has the dimension (N TXRX-BF,Az =32) × (N TXRX-BF,Az =16), and the window functions w 3,TXRX-BF (m RX,Az ) for azimuth (Chebyshev window with 30dB sidelobe suppression) and w 4,TXRX-BF (m RX,El ) used for elevation (also Chebyshev window with 30dB sidelobe suppression): STXRX−BF(j0,lp,nTXRX−BF,Az,nTXRX−BF,EI)=DFTNTXRX−BF,EI[w4,TXRX−BF(mTXRX,EI)⋅ DFTNTXRX−BF,Az[w3,TXRX−BF(mTXRX,Az)⋅S^2,TXRX−BF(j0,lp,mTXRX,Az,mTXRX,EI)]], where n TXRX-BF,Az = 0,...,N TXRX-BF,Az -1 the loop variable for the image domain, i.e. the frequency domain of the third discrete Fourier transform DFT NTXRX-BF,Az is and represents the so-called TXRX-BF azimuths (because frequency is proportional to the electrical azimuth angle) and n TXRX-BF,El = 0,...,N TXRX-BF,El -1 the loop variable for the image domain of the fourth discrete Fourier transform (DFT) NTXRX-BF,El represents the so-called TXRX-BF elevation stores; TXRX-BF azimuth and TXRX-BF elevation stores are also referred to as TXRX-BF azimuth and TXRX-BF elevation beams in the following.

[0077] With the spectrum W 34,TXRX-BF (jn RX-BF,Az, n RX-BF,El ) the two-dimensional window function w used for beam shaping 3,TXRX-BF (m RX,Az )·w 4,TXRX-BF (m RX,El), which is also formed with 100% zeropadding in the third and fourth dimensions and which has its maximum at indices 0, results in the TXRX beamformer spectrum S TXRX-BF (j0,l p ,n TXRX-BF,Az ,n TXRX-BF,El ) for a point-like object at distance gate j0, Doppler gate l0, TXRX-BF azimuth beam n TXRX-BF,Az,0 and TXRX-BF elevation beam n TXRX-BF,El,0 to STXRX−BF(j0,lp,nTXRX−BF,Az,nTXRX−BF,EI)=A / (2j^)⋅W12(0,modK(Ip−I0))⋅W34,TXRX−BF( modNTXRX−BF,Az(nTXRX−BF,Az−nRTXRX−BF,Az,0),modNTXRX−BF,EI(nTXRX−BF,EI−nTXRX−BF,EI,0); For the sake of completeness, it should also be mentioned that, firstly, the TXRX-BF azimuth beam n TXRX-BF,Az,0 and the TXRX-BF elevation beam n TXRX-BF,El,0 as follows with the azimuth angle α Az and the elevation angle α El related to the object: nTXRX−BF,Az,0=modNTXRX−BF,Az(sin(αAz)⋅3⋅NTXRX−BF,Az) nTXRX−BF,EI,0=modNTXRX−BF,EI(sin(αEI)⋅5⋅NTXRX−BF,EI), where the factors 3 and 5 come from the antenna spacings 3λ (in the horizontal direction) and 5λ (in the vertical direction).

[0078] We will now consider the example above again, with two objects at the same distance gate j0 and at the Doppler gates l. 0,1 = 168 or l 0,2 = 312, whereby the signals corresponding to the transmission paths appear in the same grid G8 (see references (17) and (12)); the azimuth angles α Az,1 = 4.78° and α Az,2 = 14.5° of the two objects lead to the TXRX-BF azimuth beams according to reference (27a). RTXRX-BF,Az,0,1 = 8 and n RTXRX-BF,Az,0,2 = 24, the elevation angles α El,1 = 4.3° and α El,2 = -4.3° according to reference (27b) on the TXRX-BF elevation beams n RTXRX-BF,El,0,1 = 6 and n RTXRX-BF,El,0,2 = 10. In Fig. 10 is the one located in the Doppler Gate l p = l 0,1 = 168 of object 1 resulting TXRX-BF spectrum S TXRX-BF (j0,l 0,1 ,n TXRX-BF,Az ,n TXRX-BF,El) shown in magnitude in dB and scaled; for the TXRX-BF beam of object 1, i.e., at n RTXRX-BF,Az,0,1 = 8 and n RTXRX-BF,El,0,1 As expected, a performance peak occurs at = 6, but also in the area of ​​the TXRX-BF azimuth beam n RTXRX-BF,Az,0,2 = 24 of object 2, a power peak occurs, particularly in the elevation direction, which is highly fragmented, even though object 2 is located at a different Doppler gate (at l). 0,2 = 312). This fragmented power peak arises because the Doppler gates resulting from reference (12) l 0,1 (m TX ) the transmit path signals of object 1 also partially correspond to transmit path signals of object 2 - the Doppler gates correspond to a total of four positions. 0,2 (m TX ) of object 2 the Doppler gates l 0,1 (m TX) of object 1; thus, some signals from object 2 are also included in the digital beamforming, however, they have an incorrect assignment to the transmit paths and therefore generally an incorrect calibration (the previous considerations and formulas have not addressed the fact that the antenna channels generally also have different phase angles and slightly different amplitudes due to hardware-related effects such as different antenna feed line lengths; these differences can be determined, for example, during sensor production in order to then compensate for them during signal processing using appropriate calibration factors). As a result, no meaningful beamforming is achieved for these signals from object 2, particularly in the elevation direction (elevation is dominated by the transmit paths), while a blurred power peak is still visible in azimuth (azimuth is dominated by the receive paths). Thus, in the TXRX-BF spectrum S TXRX-BF (j0,l0,1 ,n TXRX-BF,Az, n TXRX-BF,El Since these components of object 2 do not subsequently lead to detections (which would have an incorrect Doppler gate and incorrect TXRX-BF beam), they must be identified as TX modulation artifacts. For this purpose, it is checked for each value in the TXRX-BF spectrum whether the RX-BF power spectrum P resulting from incoherent TX integration is present at the corresponding position. RX-BF,TX-NC (j0,l 0,1 ,n RX-BF,Az, n Rx-BF,El ) was valid (for this purpose, the result of the test there for detection threshold and no TX modulation artifact was temporarily stored in bit fields); the necessary relationship between RX-BF beams and TXRX-BF beams results from references (20) and (27) to: nRX−BF,Az=nTXRX−BF,Az⋅NRX−BF,Az / NTXRX−BF,Az=nTXRX−BF,Az / 2 nRX−BF,EI=modNTXRX−BF,EI / NRX−BF,EI(nTXRX−BF,Az)=mod4(nTXRX−BF,Az).

[0079] To identify power peaks, local maxima of the TXRX-BF spectrum S arising in four-dimensional space are used. TXRX-BF (j,l,n TXRX-BF,Az ,n TXRX-BF,El ) determined, i.e., via TXRX-BF azimuth beams, via TXRX-BF elevation beams, via Doppler gates l (therefore the set L p the processed Doppler gates are also extended to include the neighbors of the valid Doppler gates) and via the distance gates (for this, the corresponding values ​​of the TXRX-BF spectrum in the two adjacent distance gates must also be calculated).

[0080] Additionally, such identified power peaks must be checked against a detection threshold, which is based in particular on the noise level. It could be argued that this is implicitly addressed in the above test for validity in the RX-BF power spectrum P resulting from non-coherent TX integration. RX-BF,TX-NCis included (where a detection threshold is also checked), but several TXRX-BF beams of the TXRX-BF spectrum considered here correspond to the RX-BF beams used there (since TXRX beamforming uses many more channels, especially in elevation, and therefore has a better resolution). Therefore, particularly with multiple TXRX-BF power peaks at the same RX-BF beam, it must be checked whether the smaller power peaks are also above the noise. For a noise-derived detection threshold, the noise determined during RX-BF beamforming (before or after non-coherent TX integration) must be translated to the TXRX beamformer in the respective distance gate and corresponding RX-BF beam (according to reference (28)). When using the RX-BF noise before non-coherent integration, it must be multiplied by the factor TNoise=[sum(w3,TXRX−BF(mTXRX,Az)2)⋅sum(w4,TXRX−BF(mTXRX,EI)2)] / [sum(w3,RX−BF(mRX,Az)2)⋅sum(w4,RX−BF(mRX,EI)2)] The values ​​are multiplied to obtain the noise in the TXRX beamformer (assuming approximately constant and uncorrelated noise across all receive paths). The detection threshold is then set approximately 25 times higher than this noise (the mean noise is 1.97 times higher than the OS40% noise calculated here, and the detection threshold is intended to be 11 dB, i.e., 12.6 times higher than the mean noise). Additionally, the detection threshold could also include a dynamic range limitation per TXRX beamformer and / or across the entire distance gate. To prevent detections from side lobes of the digital beamforming (from window functions used and imperfect sensors), so-called side lobe thresholds are also used, which do not allow detections that are too far (e.g. more than 10dB) below the maximum in the respective TXRX azimuth beam and the maximum in the respective TXRX elevation beam.

[0081] The following criteria exist for identifying detections in the TXRX-BF spectrum: exceeding the detection threshold, validity in the non-coherently integrated RX beamformer, and a local maximum in four-dimensional space; these criteria can, of course, be checked in any order.

[0082] Such detections in the TXRX-BF spectrum are – as already explained at the beginning – still ambiguous in azimuth and elevation angles, since the distances in the two-dimensional equidistant antenna array used are according to Fig. 1c lies above half a wavelength. To resolve the azimuth ambiguities, the antenna channels from transmitting antennas TX1, TX8, TX10, and TX5, and all receiving antennas, can be used; to resolve the elevation ambiguities, the antenna channels from transmitting antennas TX8-TX11 and all receiving antennas can be used, since two of these transmitting antennas (TX8 and TX10 for azimuth, and TX9 and TX11 for elevation) lie half a wavelength outside the respective equidistant grid. The complex channel values ​​Ŝ2(j,l,m) TX ,m RX ) the additional antenna channels required are, like the channel values ​​required for TXRX beamforming according to reference (24), the spectra S2(j,l,m RX ) at the points corresponding to the Doppler gate l0 to be processed, l = l0(m TX ) taken from reference (12): S^2(j0,l0,mTX,mRX)=S2(j0,l0,(mTX),mRX)=S2(j0,modK(l0+p(mTX)⋅K / P),mRX) with mRX=0,...,15 and mTX=8,...,11.

[0083] To resolve the azimuth ambiguities, digital beamforming is used for the four transmission paths m TX = 1, 8, 10, 5 and all receive paths - either at all azimuth angles or, advantageously, only at the six hypotheses resulting from the TXRX beamforming (due to the 3λ spacing, there are six hypotheses in -90°...+90°). For elevation, this can be done for the four transmit paths m TX = 8,9,10,11 and all reception paths are carried out, finding the correct one among 10 hypotheses (where large elevation angles can be excluded a priori due to the vertical extent of the individual antennas and the associated focusing).

[0084] Let us now consider again the example above with two objects at the same distance gate j0 and at the Doppler gates l. 0,1 = 168 or l 0,2 = 312, whereby the signals corresponding to the transmission paths appear in the same grid G8. As also shown by the Fig. As explained in section 10, object 2 is generated in the TXRX-BF spectrum of the Doppler gate l 0,1Object 1 emits components that do not form a sharp power peak. Depending on the position of the two objects, these components from Object 2 can extend into or beyond the power peak of Object 1, which can lead to errors, particularly when determining the angle of Object 1. These angular errors (i.e., their magnitude and sign) depend on the phase at which the components from both objects superimpose in the region of the TXRX-BF beam from Object 1 (the superposition in the TXRX-BF spectrum occurs in the complex domain). To prevent the same or at least similar angular errors from occurring in successive cycles (in which the angles of the objects do not change significantly) – otherwise, the angular averaging used in tracking (i.e., monitoring detections over multiple radar cycles) would be ineffective – the phase modulation must be varied from radar cycle to radar cycle.

[0085] A first approach to this is that in the phase modulation according to reference (4) the component φ changes linearly over the frequency ramps k PM,lin (k,m TX ) the start phase φ0(m TX ,c) via the transmission paths m TX = 0,...,M TX -1 is different and varies over the radar cycles c = 0,1,... (in the original reference (4b) all start phases were zero): φPM,lin(k,mTX)=φ0(mTX,C)+2π⋅k⋅p(mTX) / P; The start-up phase, for example, assumes random values ​​from the entire phase range 0...2π. During TX demodulation, i.e., the extraction of the complex antenna channel values ​​from the TXRX-BF spectrum at the corresponding Doppler gates, the respective start-up phase must be compensated by multiplication with a corresponding complex unit vector. For demodulation of object 1, the signals resulting from the associated Doppler gates are multiplied by the complex unit vector corresponding to the respective transmit path. However, for Doppler gates where transmit path signals from object 2 are also present, these originate from a different transmit path than those from object 1, so the start-up phases are incorrectly compensated for these transmit path signals from object 2.This means that the phases of these components originating from object 2 and incorporated into the TXRX beamforming of object 1 change over the radar cycles, so that their contributions in the spectrum change not only in phase but also in general form; thus, it is also impossible for object 2 to always have a maximum contribution in the area of ​​the TXRX-BF beam from object 1 over several cycles.

[0086] A second approach to changing the phase modulation via radar cycles can be achieved by varying the modulation velocities p(m). TX ) can be realized. Generally, different sets of such modulation speeds can be used for this purpose (i.e., those defined by the values ​​p(m). TXThe set formed changes, or only the mapping of the transmission paths is changed when using the same set. This changes the mapping of transmission paths from object 2 to the Doppler gates of object 1 (at least their mapping, and possibly also the Doppler gates themselves).

[0087] The same effect can be achieved by the third approach, which involves changing the period P of the linear modulation across the cycles.

[0088] When assigning the modulation speeds p(m) / P of the linear phase components to the transmission paths m TXIt is expedient to consider the following effect: Due to the reflective properties of the road surface, the transmitted signal reaches an object both directly and via reflections off the road surface; depending on the phase difference of the two components, this can lead to an amplifying effect, but also to attenuation up to total cancellation (at a 180° phase difference). The phase difference depends on the height of the transmitting antennas above the road surface (and of course also on the height of the object, which is not relevant here); since in the antenna arrangement considered here, according to Fig. 1b Since the transmitting antennas are essentially arranged in four planes (apart from the additional offset of antennas TX9 and TX11), the effective transmit power at the object, and thus also the power of the corresponding received signals, will differ in the presence of road reflections. If, for example, total cancellation occurs at transmitting antennas of the second and fourth planes (viewed from the bottom) (i.e., for antennas TX1, TX8, TX10, and TX5 in the second plane, and TX3 and TX7 in the fourth plane), and the other six transmitting antennas are assigned to a TX modulation artifact (there are two TX modulation artifacts, each assigned to six transmitting antennas), then the level of this TX modulation artifact would be the same as in the actual Doppler gate, so that unambiguous object identification would no longer be possible. Therefore, the assignment of the modulation velocities p(m) / P of the linear phase components to the transmit paths m should TXAdvantageously, the distribution of the transmitted paths used by the TX modulation artifacts across the different vertical transmit antenna planes should correspond as closely as possible, in percentage terms, to the distribution of the actual Doppler gate (where, in the approach presented here, all transmit paths are included). For TX modulation artifacts to which six transmit antennas are assigned, there should be two transmit antennas each from the second and third vertical planes, and one each from the first and fourth planes. In general, this cannot be achieved exactly for all TX modulation artifacts; by searching through all possible assignment combinations, the assignment that comes closest to this goal can be determined.

[0089] Another criterion for assigning the modulation speeds p(m) / P to the transmission paths m TXIt's possible that the strength of the transmission paths is taken into account (unlike the case of equally strong transmission paths considered here, real sensors exhibit slight differences due to hardware effects). Therefore, weaker transmission paths are more appropriately used for the TX modulation artifacts assigned the most transmission paths (six transmitting antennas in this example).

[0090] So far, only integer ratios K / P of the length K of the second FFT (across the frequency ramps) and period P of the linear component of the phase modulation have been considered (in particular, the example K / P = 512 / 32 = 16); the power peaks of the various transmit paths in the second DFT and the power peaks in the spectra used for object identification (i.e., the power peaks at the correct Doppler gate and at the TX modulation artifact positions) then assume positions from an equidistant grid with grid spacing K / P. Now, we will consider what effects would result with non-integer ratios K / P, for example, with P = 24 and an unchanged K = 512, i.e., K / P = 21.33.To identify the correct Doppler gates, one would then have to search this grid. Since only values ​​for integer Doppler gates are available, either the nearest integer Doppler gates of the grid would have to be used (leading to varying distances of 21 and 22), or corresponding intermediate values ​​would have to be calculated by interpolation. However, neither the use of a non-equidistant grid nor interpolation may be supported when using special hardware-implemented computational logics, which are frequently used in radar processing. Furthermore, non-integer K / P can lead to adverse performance: If the nearest integer Doppler gate of the grid is used each time, then the distance of the used values ​​to the actual position of the power peaks is not constant—for example, if the target is at the non-integer Doppler gate 10.In case 33, the values ​​used are sometimes located exactly at the position of the respective power peak, but sometimes also a third of a Doppler gate before or after it. This means that the levels used do not correspond exactly to the ratio of the power peaks, which can lead to less accurate identification of the object Doppler gates (with an integer K / P ratio, for example, the K / P = 16 considered above, all values ​​used are located a third of a Doppler gate before the respective position of the power peaks, so that the level ratio used corresponds to that of the power peaks). While interpolation can reduce this problem, errors can still occur because, for example, the shape of the power peaks assumed for interpolation differs from the actual shape (for example, because the object exhibits relative acceleration).If digital beamforming is also performed via the transmit paths, the required signals from the transmit paths of the second DFT are taken from the corresponding positions, as explained and used above; however, since these positions lie on a grid with a non-integer grid spacing, the values ​​used (determined from the nearest Doppler gate or by interpolation) do not match exactly, exhibiting not only amplitude errors but also phase errors, which can lead to large errors in the beamforming spectrum.

[0091] For the reasons stated above, it is not advisable to use non-integer K / P ratios. With the considered K = 512, however, this leads to the conclusion that only values ​​of P that are powers of two are possible. To reduce the number of TX modulation artifacts, a period P = 24 would also be desirable. For this purpose, zero-padding can be performed on the second DFT to a length L = 768, resulting in L = 768 Doppler gates and the relevant ratio L / P = 768 / 24 = 32, which is again an integer.

[0092] With a period P that is not a power of two, but phase shifters whose number of realizable values ​​in an equidistant grid is typically a power of two (64 phase values ​​in the example under consideration), many of the required phase values ​​2π·m / P, where m is an integer, cannot be realized exactly. One approach is to always use the nearest available value (instead of the phase 2π·5 / 24, the phase 2π·13 / 64 is used); however, the resulting quantization error then repeats itself every period P (possibly even with a smaller period), which ultimately leads to harmonics, i.e., further power peaks in the spectrum (analogous to Fig. 3a for non-ideal phase shifters), which can lead in particular to false detections and errors in angle formation. To avoid this (analogous to Fig. 3b), the quantization errors must be decorrelated, which can be achieved, for example, by a superimposed random phase component. The random component φ considered above can be used for this purpose. PM,r (k) is used, which is identical for all transmission paths and varies over the frequency ramps k = 0,...,K-1; however, in contrast to above, it is no longer defined by random selection from the 64 phase values ​​of the phase shifters (i.e., as a value from 2π·m / 64 with m = 0,...,63), but rather it takes on any arbitrary value from 0...2π. The phase shifter value used in each case is the one which corresponds to the phase value calculated according to (4) (composed of a random component φ). PM , r (k) and linear component φ PM,lin (k,m TX ) possibly with a random start phase) is closest.

[0093] In the processing sequence described above, the RX beamforming is performed after the two DFTs across the individual values ​​per frequency ramp and across the frequency ramps; the order can of course be changed arbitrarily - for example, the RX beamforming could also take place at the very beginning.

[0094] The inventive method of RX beamforming for identifying valid Doppler gates (and thus for distinguishing TX modulation artifacts) exploits the fact that, with equidistant RX antennas, targets with different angles can be separated by beamforming. This applies not only to the two-dimensional RX array considered here, but also to one-dimensional RX arrays. At least reduced separation (i.e., with a reduced dynamic range) is still possible even if there are moderate deviations from an equidistant arrangement in the RX antenna array – either because many antennas have small offsets from an equidistant grid or only a few antennas have larger offsets.

[0095] In simpler radar sensors with fewer antennas, the RX antennas are often arranged in a highly non-equidistant manner (for example, they represent a so-called sparse array). If RX beamforming is then performed, large sidelobes appear in the beamforming spectrum in addition to the actual power peak. This means that if components from two or more objects occur in an RX-BF spectrum at angles to different RX-BF beams, they are no longer clearly distinguishable, which can lead to incorrect identification of their Doppler gates. Therefore, to identify the valid Doppler gates (as distinct from the TX modulation artifacts), it may be advantageous not to perform and utilize RX beamforming with subsequent non-coherent integration over the received signals, but rather, according to the invention, to use coherent integration over the receive and transmit paths, i.e., the two-dimensional beamforming according to reference (26), whereby this is then applied in all Doppler gates l = 0,...,K-1 is to be determined (and not just in a subset L). p ).

[0096] For this TXRX beam shaping S TXRX-BF (j,l,n TXRX-BF,Az, n TXRX-BF,El ) the maximum power P is reached in each Doppler port l = 0,...,K-1 TXRX-BF,max (j,l) via the N TXRX-BF,Az × N TXRX-BF,El Beams formed: PTXRX−BF,max(j,l)=maxnTXRX−BF,Az & nTXRX−BF,EI(|STXRX−BF(j,l,nTXRX−BF,Az,nTXRX−BF,EI)|2), where “max n&m “the maximum over all combinations of the indices n and m means; for the above example of the two objects, the TXRX-BF power maximum P resulting at distance gate j0 TXRX-BF,max (j0,l) in dB and scaled (to maximum) in Fig. Figure 11 is shown. The identification of the valid Doppler gates is now shown in P TXRX-BF,max (j0,l) performed analogously to the procedure for the non-coherently integrated RX-BF, but here only in one vector and not in several vectors to different beams.

[0097] Even in the initially considered, incoherently integrated power spectrum P across receive and transmit paths RX-NC,TX-NC (j0,l) according to reference (15), the identification of the valid Doppler gates only took place in a vector, i.e., without ray formation and thus without angular separation. There, the signal-to-noise ratio did not improve through integration, but only the variance in the noise floor decreased. In comparison, the determination of P TXRX-BF,max (j0,l) used coherent integration to achieve a significantly improved signal-to-noise ratio (as shown by comparison of the Fig. 11 and Fig. 7 evident); this allows objects with a weak received signal (distant objects and / or objects with low reflectivity) to be better identified.

[0098] Another advantage of the TXRX-BF maximum power output P TXRX-BF,max(j0,l) is due to the fact that the level of the TX modulation artifacts is generally not the full sum of the signal levels of the transmit paths located there, but is lower, because the transmit path signals do not add up in phase in any TXRX-BF beam; this is also confirmed by comparing Fig. 11 and Fig. 7 clearly, since in Fig. 11 for P TXRX-BF,max The level difference of the TX modulation artifacts to the maximum is on average more than twice as low as in Fig. 7 for P RX-NC,TX-NC (j0,l) - when comparing, factor 2 must be taken into account, since for P TXRX-BF,max a signal sum is formed over the contributing transmit path signals, for P RX-NC,TX-NC (j0,l) a power sum, so that the reduced number of transmit path signals used in the TX modulation artifacts has a correspondingly different effect when power levels are compared. This advantage also leads to the fact that at the TXRX-BF power maximum P TXRX-BF,max(j0,l) the smaller object 2 at the Doppler gate l 0,2 = 312 is fundamentally identifiable, as it forms the second largest performance peak (see Fig. 11), whereas this is the case with the non-coherently integrated service spectrum P RX-NC,TX-NC (j0,l) after Fig. 7 is not the case (there the level of TX modulation artifacts is higher at l = 136 and l = 200).

[0099] To prevent the recurring case over radar cycles in which, during TXRX beam shaping, many or all contributing transmit path signals overlap at least approximately coherently in a beam due to a TX modulation artifact, it is also advantageous to apply the variation of the start phases over the transmit paths (see Ref. (31)).

[0100] Generally, even when using the TXRX-BF maximum power output, P TXRX-BF,max(j0,l) all previously presented, advantageous and inventive designs of the TX modulation, i.e. the phase profiles different over the transmit paths, are used.

[0101] It should be added that this naturally applies when using the non-coherently integrated performance spectrum P RX-NC,TX-NC (j0,l) allow all previously presented, advantageous and inventive designs of TX modulation to be applied.

[0102] Now, an advantageous implementation of the non-coherent TX integration according to references (15) and (22) will be discussed. There, the values ​​of P are determined for each Doppler port l0 = 0,...,K-1. RX-NC (j,l) or |S RX-BF (j,l,n RX-BF,Az, n RX-BF,El )| 2 in the expected grid l = l0(m TX ) the power peaks are integrated. For all P Doppler gates l0 in a grid G g According to reference (17), the values ​​to be summed also lie in this grid; the grid l = l0(m TX) these values ​​are determined according to reference (12) by the modulation rates p(m TX ) is defined and shifts successively with l0. Therefore, the calculation of the P sums for this grid can be performed as a cyclic correlation between the P values ​​in this grid and an allocation vector b(n) with n=0,...,(P-1), where b(n) = 1 for all n which correspond to a p(m TX) correspond, and otherwise b(n) = 0. The cyclic correlation can advantageously be performed as a fast correlation, i.e., by an FFT of the P values ​​of the grid, a multiplication of this FFT with the pre-calculated conjugate complex of the FFT of the occupancy vector b(n), and a subsequent inverse FFT; if a normal FFT (i.e., not an inverse FFT) is used for the inverse transformation, then the conjugate complex of the first FFT must be multiplied with the pre-calculated FFT of the occupancy vector b(n), and a scaling factor P must be taken into account if necessary.Since the values ​​used for summation are real-valued, summation can be achieved simultaneously for two different grids by fast correlation, by assigning the values ​​of one grid to the real part and those of the other grid to the imaginary part of a complex-valued vector (the FFT works with complex values); if a normal FFT (i.e., not an inverse FFT) is used for the inverse transformation, the complex conjugate of the result must then be calculated.The computing units used for radar processing are generally optimized for FFTs, so this approach according to the invention can lead to a significant reduction in the required computing time for non-coherent TX integration; this is particularly important for the approach with prior RX beam shaping, since the non-coherent TX integration has to be performed in each RX-BF beam, i.e. 64 times, and therefore represents the most computationally intensive step of the entire calculation.

[0103] If the signals corresponding to the different transmission paths are not of equal strength, they can be used with different weights in non-coherent TX integration, for example, to optimize the signal-to-noise ratio. Such different weights can be implemented in the occupancy vector b(n) for cyclic correlation by ensuring that the occupied elements no longer all assume the value 1, but rather their respective weights.

[0104] Different transmission levels across the transmitting antennas can occur due to hardware-related effects such as different lengths of the antenna leads or tolerances in the electronic components generating the transmission power; these generally unintentional and unknown differences are determined during sensor production or even during sensor operation in order to ascertain the calibration factors required for signal processing.

[0105] Different transmission levels can also be selectively generated (via appropriate configuration of the signal power generation) and used via the transmitting antennas; this will now be briefly explained: In the example considered so far, there is M TX= 12 transmitting antennas and a modulation period of P = 32 is used; to enable robust detection of the correct Doppler gate of the objects, the period P is significantly higher than the number of transmitting antennas. For even better environmental perception, future radar systems will have even more transmitting antennas, e.g., M TX = 32. Then a significantly larger modulation period P would also be necessary, which would increase the number of TX modulation artifacts and the probability that the power peaks of two objects in the same grid G g to come to rest, significantly increased; in addition, a large number of modulation artifacts also leads to an overestimation of the noise level, from which the detection threshold is derived, and thus to a reduced dynamic range.

[0106] In the approach considered so far, with at least approximately the same transmission level across the transmitting antennas, identifying the correct Doppler gate in the grid of power peaks requires unoccupied positions, thus creating an identifiable pattern. Alternatively, an identifiable pattern can also be generated by having the power peaks exhibit different levels (due to different transmission levels at the respective transmitting antennas). For example, at M TX = 32 transmitting antennas continue to have a modulation period P = 32, full transmit power for the first 12 transmitting antennas and modulation speeds p(m) TX) according to equation (11) and for the remaining 20 transmitting antennas, the other modulation speeds and half the transmit power are used. The non-coherent TX integration is then performed over all 32 positions in the grid, whereby the 20 positions corresponding to transmitting antennas with half the transmit power are weighted half as much as the 12 positions corresponding to transmitting antennas with full transmit power. The occupancy vector b(n) then has the values ​​1 and 0.5. The line values ​​after non-coherent TX integration remain relatively unchanged (and thus also the identification of the correct Doppler gate) if a constant value is subtracted from the weights – for example, in the example under consideration, the value 0.5, resulting in an occupancy vector with only 12 non-zero values.

[0107] When using two levels for the transmit levels of the transmitting antennas and in the above case that the modulation period P corresponds to the number of transmitting antennas (which represents the smallest possible value for P), the most robust identification of the correct Doppler gate is achieved when both levels occur equally often (in this sense, the above example is therefore not quite optimal).

[0108] To increase the robustness of identifying the correct Doppler gate, more than two different transmit levels can be used, and the modulation period P can be designed to be larger than the number of transmitting antennas (for example, P = 48 with 32 transmitting antennas, resulting in 16 zeros in the occupancy vector).

[0109] To make the identification of the correct Doppler gate as robust as possible against noise contained in the received signals, the weighting used in the example above is optimal: the weights are chosen proportionally to the transmit level of the respective transmitting antenna.

[0110] The average transmit level across all frequency ramps is relevant for the magnitude of the power peaks. A reduction in the average transmit level can be achieved not only by an equal level reduction across all frequency ramps, but also by omitting transmission on some frequency ramps. However, this results in multiple power peaks per transmitting antenna in the Doppler spectrum (two power peaks occur if transmission only occurs on every other frequency ramp), which in principle necessitates a longer modulation period. This approach is primarily interesting when only binary phase shifters are available (i.e., only phases 0° and 180° are possible) in order to synthesize further linear phase profiles (phase profile 0, 90°, 180°, 270°, 0°, ... with every other frequency ramp disabled, requires only phases 0° and 180°).

[0111] In the antenna arrangement according to Fig. 1b The transmitting antennas are located on different planes when viewed vertically. Due to superposition effects caused by reflections off a road surface, the signal components originating from different planes can vary in level (because the reflections cause an object to be seen from different directions, and the corresponding signal components overlap). Thus, different received levels result not only from the different transmit levels. Consequently, the level variation of the power peaks used for non-coherent TX integration does not correspond to the expected pattern, which can lead to errors in identifying the correct Doppler gate. To reduce or completely prevent the probability of such errors, the transmit levels of the respective transmitting antennas should differ in as many planes as possible, preferably such that the diversity of the transmit levels is similar in as many or all planes as possible.

[0112] Antenna arrays are often used in which several transmitting antennas are arranged in a grid at constant spacing. In such cases, particularly when dealing with multiple objects at the same distance and with the same relative speed, it can be advantageous for the distribution of the different transmission levels across these antennas to be irregular.

[0113] Some antenna arrays have several similar groups of transmitting antennas, especially antenna pairs (which are used differently in several signal processing steps, for example, firstly as a common, i.e., larger, antenna, and secondly for phase comparison of the signal components generated by the individual antennas). It can be advantageous to make the different transmit levels of the antennas in each group at least approximately the same.

[0114] A potential disadvantage is that deliberately different transmission levels of the transmitting antennas mean the sensor's full sensitivity potential is not utilized (because some transmitting antennas do not use their full transmission power). However, this is not critical, or only slightly so, for sensors with many transmitting antennas, as they already possess very high sensitivity, and a reduction in the overall transmission power may be necessary for regulatory reasons. Furthermore, reducing the transmission power also reduces the sensor's power consumption (less transmission power requires less electrical power to generate).

[0115] In the modulation considered so far according to Fig.2. All frequency ramps have the same frequency position, i.e., starting frequency (and thus center frequency). For improved distance resolution, as described in DE 10 2020 210 079 B3, the starting frequency can be changed linearly across the frequency ramps, preferably with the spacing of the frequency ramps also changing linearly. This has no effect on the approaches according to the invention and their advantages; they remain unchanged. The only difference is that the Doppler gate of an object is then not only determined by the relative velocity of the object, but also has a contribution from the object's distance, which increases linearly with this distance (due to a center frequency changing linearly across the frequency ramps, the phase position changes linearly, since the number of wave trains that fit into the beam path to the object and back changes, and this change is greater the longer the beam path is).

[0116] So far, a linear frequency ramp has been considered as the transmitted signal, which is repeated sequentially. However, other signal forms can be used instead of a frequency ramp, for example, a signal with pseudorandom binary phase modulation (i.e., the sign is changed very rapidly and pseudorandomly within the signal) or an OFDM signal (OFDM = Orthogonal Frequency Division Multiplexing). The sampled values ​​(after analog-to-digital conversion) cannot then be used directly as input values ​​for the first DFT described above; instead, the Fourier transform must first be calculated for each transmitted signal and receiving antenna and divided by the spectrum of the transmitted signal.

[0117] Finally, it should be noted that it is obvious to a person skilled in the art how the considerations and explanations according to the invention, illustrated by the above application examples, can be transferred to general dimensions and parameter designs, i.e., they can also be applied to other numerical values. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] WO 2018 / 137835 A1

[0007] EP 2 629 113 B1

[0032] DE 10 2020 210 079 B3

[0115]

Claims

[1] Method for a radar system for environmental sensing with - Transmitting devices with several parallel transmitting antennas for radiating transmission signals, which contain one or more sequences of K individual signals, the general form of which is preferably the same or similar, wherein their frequency position can change in particular linearly, - Means for changing the phase of the transmitted individual signals, by which a different phase response for the transmitting antennas is realized over the K individual signals, wherein this difference in the phase response for different transmitting antennas over the K individual signals is at least approximately linear, possibly excluding phase jumps due to the phase uniqueness range of 2π, - Receiving means with one or more receiving antennas for receiving transmitted signals reflected from objects, wherein the signals received from an object are composed of components originating from the various transmitting antennas, wherein these components include a phase response that differs linearly across the K individual signals as well as a similar phase progression, which is generated in particular by the relative movement of the object and, in the case of a frequency shift across the K individual signals, also depends on the object distance, - and signal processing equipment for processing the received signals, characterized by , that - not all transmitting antennas radiate the same average transmission level, - a non-coherent integration is performed over signal components whose phase responses lie within the expected grid of the signal components of an object originating from different transmitting antennas and exhibiting linearly different phase responses, which is referred to below as non-coherent TX integration, wherein this non-coherent TX integration is performed multiple times under the assumption of different radial relative velocities and, if necessary, distances of an object, - In non-coherent TX integration, signal components are weighted more heavily the higher the radiated transmit level of the respective assigned transmitting antenna is, - and the result of multiple non-coherent TX integration is used to identify the phase progression resulting from the object's relative velocity and, if applicable, object distance, and / or a quantity derived therefrom. [2] Method according to claim 1, wherein in the non-coherent TX integration the weighting of the signal components correlates with the mean transmit level of the respective assigned transmitting antenna, i.e., in particular is proportional to the mean transmit power or transmit amplitude of the respective assigned antenna, optionally less a constant value, in order to expediently achieve the most robust possible identification of the phase progression resulting from the object relative velocity and optionally object distance and / or a quantity derived therefrom, in particular in the case that contributions from several objects and / or significant noise are present in the signals. [3] Method according to claim 1 or 2, characterized by, that prior to the non-coherent TX integration via the received signals to multiple receiving antennas or via signals derived from them, a coherent integration, in particular in the form of digital beamforming, hereinafter referred to as RX beamforming, is carried out, and the result of the multiple non-coherent TX integration from different RX beamforming directions is used to identify the phase progression resulting from the object relative velocity and, if applicable, object distance, and / or a quantity derived therefrom. [4] Method according to claim 1 or 2, characterized by, that a non-coherent integration is performed over the received signals to several receiving antennas or over signals derived from them, which is expediently carried out before the non-coherent TX integration, and the result of the entire non-coherent integration, performed multiple times under the assumption of different radial relative velocities and, if necessary, distances of an object, is used to identify the phase progression resulting from the object's relative velocity and, if necessary, object distance, and / or a quantity derived therefrom. [5] Method according to any of the above claims, wherein the incoherent integration is performed over power values ​​or over magnitude values ​​of the signal components. [6] Method according to any one of the above claims, wherein - the K individual signals preferably lie in an equidistant grid, at least approximately, - a discrete Fourier transform of length L is performed on their K received signals or signals derived from them, possibly after extension with zeros, i.e. a so-called zero-padding, wherein the L frequency reference points of the discrete Fourier transform are referred to below as Doppler gates, - the phase response across the K individual signals contains a different linear component for each transmitting antenna, i.e., with a different slope, which results in the signal components from the different transmitting antennas lying at different Doppler gates in the discrete Fourier transform, - the non-coherent TX integration is performed via Doppler gates in the grid expected by the different slopes of the linear components, - and this non-coherent TX integration is performed for different positions of this grid, i.e., different, in particular all Doppler gates as the first grid point and thus for different, in particular all possible object relative velocities and, if necessary, object distances. [7] Method according to claim 6, wherein the linear components of the phase responses when mapped to the phase uniqueness range of 0...2π have a common period P, which preferably represents an integer divisor of the number L of Doppler gates, whereby the Doppler gates corresponding to the transmitting antennas represent a subset of an equidistant grating with a distance L / P. [8] Method according to claim 7, wherein the summation necessary for non-coherent TX integration for preferably all P grid layers in an equidistant grid with integer spacing L / P is realized by performing a cyclic correlation between a vector with the P values ​​of the discrete Fourier transform in this grid and an occupancy vector of length P, wherein the values ​​of the occupancy vector for all indices which correspond to the linear phase change of a transmitting antenna within one period P as multiples of 2π, assume non-zero values ​​and are otherwise 0, and wherein this cyclic correlation can preferably be performed as a fast correlation, i.e. by multiplying two spectra in the frequency domain. [9] Method according to claim 8, wherein the non-zero values ​​of the occupancy vector correlate to the mean transmit level of the respective associated transmitting antenna, i.e., in particular, are proportional to the mean transmit power or transmit amplitude of the respective associated transmitting antenna, optionally less a constant value. [10] Method according to one of the above claims, in which different transmission levels are caused by hardware-related effects such as different lengths of the antenna leads via the transmitting antennas and / or are specifically generated, in particular by a configurable transmission power generation, especially so that the period P can be chosen as small as possible, wherein the number of transmitting antennas represents the lower limit. [11] Method according to one of claims 6-10, wherein the results of the non-coherent TX integration of raster layers with spacings which are integer multiples of L / P are examined and in particular compared in order to identify objects as well as their relative velocity and, if applicable, distance, wherein preferably a detection threshold which depends in particular on the noise level is also used and, when using an RX beam shaping, this is done in all its beam directions. [12] Method according to claim 11, wherein for object identification the grid position is used which, from those in the grid at a distance L / P, the non-coherent TX integration assumes the highest amount and thus the highest performance. [13] Method according to one of claims 6-12, in which, for object identification, grid layers in the grid with distance L / P are also used where the power of the non-coherent TX integration is not maximum, but is above a threshold which may depend on the distance of this grid layer to the grid layer of the maximum and in particular is at least a minimum distance above the value generated there by an object which is located at the grid layer of the maximum and generates the power level there. [14] Method according to one of the above claims, wherein the transmitting antennas are located in different planes when viewed in the vertical direction and the transmit levels of the respective transmitting antennas differ in at least some planes, preferably such that the diversity of the transmit levels is similar in as many or as many planes as possible, thereby reducing the influence of superposition effects caused by reflections on a road surface. [15] Method according to one of the above claims, wherein there are several similar groups, in particular pairs of transmitting antennas, wherein in each group the different transmit levels of their antennas are at least approximately equal. [16] A method according to any of the above claims, wherein the individual transmit signals are linearly frequency-modulated, optionally with their center frequency changing successively and preferably linearly, or represent OFDM signals or are generated with pseudorandom fast phase modulation, in particular characterized by , that the received values ​​resulting from each individual transmitted signal are transformed to create signal separation into gates corresponding to different distances, so-called distance gates, and that the procedures described in the above claims are applied in the different distance gates. [17] Radar system for environmental sensing with - Transmitting devices with several parallel transmitting antennas for radiating transmission signals, which contain one or more sequences of K individual signals, the general form of which is preferably the same or similar, wherein their frequency position can change in particular linearly, - Means for changing the phase of the transmitted individual signals, by which a different phase response for the transmitting antennas is realized over the K individual signals, wherein this difference in the phase response for different transmitting antennas over the K individual signals is at least approximately linear, possibly excluding phase jumps due to the phase uniqueness range of 2π, - Receiving means with one or more receiving antennas for receiving transmitted signals reflected from objects, wherein the signals received from an object are composed of components originating from the various transmitting antennas, wherein these components include a phase response that differs linearly across the K individual signals as well as a similar phase progression, which is generated in particular by the relative movement of the object and, in the case of a frequency shift across the K individual signals, also depends on the object distance, - and signal processing equipment for processing the received signals, characterized by , that - not all transmitting antennas radiate the same average transmission level, - a non-coherent integration is performed over signal components whose phase responses lie within the expected grid of the signal components of an object originating from different transmitting antennas and exhibiting linearly different phase responses, which is referred to below as non-coherent TX integration, wherein this non-coherent TX integration is performed multiple times under the assumption of different radial relative velocities and, if necessary, distances of an object, - In non-coherent TX integration, signal components are weighted more heavily the higher the radiated transmit level of the respective assigned transmitting antenna is, - and the result of multiple non-coherent TX integration is used to identify the phase progression resulting from the object's relative velocity and, if applicable, object distance, and / or a quantity derived therefrom.

Images