Novel coherent laser radar system for environment detection

Abstract

The invention relates to a coherent-working lidar system for environmental detection, which transmits a phase-modulated signal, in particular comprising irregular switching over discrete phase values, preferably over only two phase values differing by approximately 180 DEG, the frequency at which the phase is modulated varies at least in a segment-by-segment continuous manner, preferably with an at least approximately linear course.

Description

Technical Field

[0001] The invention relates to a coherently operating laser radar system, which is particularly suitable for environmental detection in motor vehicle applications. The novel coherent laser radar system operates with phase modulation and, according to the invention, additionally has frequency modulation, wherein the frequency varies continuously at least in sections, preferably with an at least approximately linear course / curve. The laser radar system advantageously utilizes continuous frequency changes, changes the radiation direction (beam direction) in a first spatial direction by means of at least one waveguide and / or, when using a real-valued mixer, correctly determines the received frequency and thus uniquely determines the relative speed of the object. For the change of the radiation direction in a second spatial direction orthogonal to the first spatial direction, different devices can be used, in particular devices with controllable optical material properties. The novel laser radar system can be realized with a high semiconductor integration density. Background Art

[0002] More and more motor vehicles are equipped with driver assistance systems. These systems use sensor systems to detect the environment and, based on the traffic situation thus detected, derive automatic vehicle reactions and / or provide instructions to the driver, particularly warnings. These system functions are categorized as comfort functions and safety functions.

[0003] However, current developments are going even further. The driver is not only assisted, but their tasks are increasingly performed autonomously by the vehicle, meaning they are gradually being replaced; this is known as automated driving.

[0004] Especially for autonomous driving, sensors are required that provide highly accurate and easily machine-evaluable environmental information. Radar systems, due to their limited angular accuracy and resolution, are unable to meet today's high detection requirements, either alone or in combination with camera systems. Therefore, lidar systems are being used in parallel. They offer high angular resolution (horizontally and vertically) comparable to cameras, and additionally provide distance information and resolution per pixel. Currently, so-called time-of-flight lidar systems are predominantly used. These treat electromagnetic radiation as particles and, therefore, can only directly measure distance, not relative velocity. However, coherent lidar systems are gaining increasing attention. These treat electromagnetic radiation as waves (like radar systems) and can directly measure relative velocity of objects using the Doppler effect. Other advantages of coherent lidar systems include robustness against interfering radiation from other sources (such as other lidar systems or sunlight) and increased sensitivity at longer distances, thus enabling a longer detection range. Furthermore, coherent lidar systems are believed to have greater potential for semiconductor integration, potentially reducing manufacturing costs.

[0005] In coherent lidar systems, the transmitted electromagnetic wave is modulated so that at least one of its parameters—i.e., amplitude, frequency, or phase—varies over time; otherwise, distance measurement is impossible. The most commonly used modulation scheme for coherent lidar systems is linear frequency modulation (FMCW), which typically consists of two frequency ramps with opposite signs. However, this modulation suffers from ambiguity, especially in the case of multiple reflections in the same radiation direction, and generating highly linear frequency changes is expensive. Phase modulation (e.g., by pseudo-randomly switching discrete phase values ​​at a fixed transmission frequency) suffers from or lessens these drawbacks, but digital evaluation of the received signal is more complex, and the solutions proposed in the prior art have drawbacks (particularly with respect to sensitivity and detection range). Currently, no known modulation form can achieve continuous frequency variation within a single pixel data acquisition and across pixels, which is advantageous for changing the radiation direction using waveguides. Changing the radiation direction in both spatial directions is still often achieved mechanically, which is costly, large, and suffers from robustness drawbacks. Real-valued mixers are often used (due to their significantly reduced implementation complexity compared to complex-valued mixers); however, determining the sign of the received frequency is often problematic or impossible, leading to two assumptions about the relative velocity and, possibly, the distance of the object. Currently, the potential for semiconductor integration in coherent lidar systems is largely untapped. Summary of the Invention

[0006] The object of the present invention is to provide a novel coherent lidar system which has high performance while allowing a low-cost implementation and a small installation space.

[0007] This object is achieved primarily by the lidar system described in claim 1; suitable embodiments of the invention are claimed in the dependent claims. A key concept is the use of phase modulation superimposed with frequency modulation, which is particularly advantageous for changing the radiation direction using at least one waveguide and allows the use of real-valued mixers.

[0008] The advantages of the present invention are particularly derived from the fact that a high-performance lidar system can be realized through high semiconductor integration, with low cost and small structural space.

[0009] The coherently operating lidar system for environmental detection according to the present invention emits a phase-modulated signal, which is distinguished by the fact that the frequency underlying the phase modulation varies at least continuously in sections, preferably with an at least approximately linear course. The phase-modulated signal includes an irregular switching of discrete phase values, preferably with only two phase values ​​that differ by approximately 180°.

[0010] Furthermore, the laser radar system receives a signal reflected from an object. This reflected signal is delayed relative to the transmitted signal due to a distance-dependent propagation time. Its frequency is shifted by both the Doppler effect, which is related to relative velocity, and the distance-dependent propagation time caused by linear frequency changes. This signal is converted to a low-frequency signal through frequency mixing and digitized into a received sequence. According to the present invention, a digital signal processing device performs two-dimensional correlation filtering on the signal reflected from the object to determine the a priori unknown time and frequency shifts in the signal reflected from the object. The distance to the object is determined from the determined time shift, and the contribution due to the time shift is subtracted from the determined frequency shift to determine the radial relative velocity of the object. This correlation filtering constitutes optimal filtering to ensure that the laser radar system achieves the most accurate measurements possible (i.e., high sensitivity and a large range).

[0011] Preferably, at least a portion of the two-dimensional correlation filtering is implemented by a fixed-wired / hard-wired (i.e. hardware-implemented and unchangeable) digital circuit implemented as a pipeline (i.e. a structure having multiple calculation stages separated by intermediate memories), wherein for each clock cycle / cycle of the digital circuit, multiple output values ​​or all output values ​​are determined in one of the two dimensions and multiple output values ​​or all output values ​​are determined over a series of clock cycles in the other dimension.

[0012] Advantageously, the fixed-wiring digital circuit may have a pre-stage in which the signal sequence or the complex conjugate of the signal sequence is multiplied by a phase modulation sequence or the complex conjugate of the phase modulation sequence at a relatively offset position, the multiplied sequence is then decimated if necessary and / or zero-padded if necessary, and then a multi-stage fast Fourier transform (FFT) is performed, wherein each single calculation operation is implemented in a dedicated circuit, in the first stage the offset between the signal sequence and the modulation sequence changes with each clock cycle, and the Fourier transform result is generated at the output of the subsequent stage, wherein the result respectively relates to the output data generated by the pre-stage a plurality of cycles ago.

[0013] Advantageously, the fixed-wiring digital circuit can be used for multiple, preferably all, pixels of the lidar system due to its high transmission rate; these pixels are generated in particular by optical radiation scanning and / or parallel reception paths, each pixel corresponding to a data acquisition and evaluation, in particular the generation of a detection result.

[0014] Furthermore, a binary phase modulation can be used, ie a phase modulation consisting of only two phase values ​​differing by approximately 180°, whereby the multiplication with the modulation sequence value is performed by means of a switchable inverter.

[0015] According to an advantageous design of the present invention,

[0016] The multiplication of the signal required for the fast Fourier transformation by the rotation factor is performed using a small number of additions and / or subtractions of the shifted signal values, preferably using at most one addition and / or subtraction to perform a real-valued multiplication.

[0017] Preferably, the bit length used is varied within the pipeline stages of the fixed-wire digital circuitry, and is preferably only set so that the quantisation noise produced by the digital circuitry does not significantly increase the magnitude of the system noise produced by the analogue parts of the receiver.

[0018] Advantageously, the Fast Fourier Transform can be implemented in a frequency-domain decimated structure, thereby avoiding reordering the input data in the form of long lines, where the longest lines and non-trivial multiplications are placed in the early stages with lower bit lengths.

[0019] Appropriately,

[0020] Truncation and / or pure bit inversion can be used for quantization, the resulting average error effect being compensated by adding correction values ​​in one stage of the digital circuit.

[0021] Furthermore, the hardwired digital circuit can be expanded with one or more additional stages for evaluating the results of the two-dimensional correlation, in particular for modulus formation or power formation and subsequent summation and / or maximum value search.

[0022] Advantageously, coupling and reflection components contained in the digital reception signal from within the laser radar system or its immediate surroundings, in particular from the cover, can be substantially eliminated by adding or subtracting a compensation sequence.

[0023] Advantageously, the digital received signal is compensated for coupling and reflection components from within the lidar system or its immediate surroundings, in particular from the housing, by using at least one of the following methods: The product of the received sequence and the modulation sequence is calculated at identical or similar points in time, and one or more average values ​​are determined, wherein the multiple average values ​​are formed by observing multiple segments of the phase modulation sequence and / or different phase value combinations. To form the average value of all values ​​of the product of the received sequence and the modulation sequence, a fixed-wire digital circuit output value is used, which corresponds to the zero frequency point of the Fourier transform in the first stage and the zero offset between the signal sequence and the modulation sequence, preferably while pausing the clock cycles of the first stage during the calculation of this value. The average value is formed over multiple detection cycles and / or over different, preferably adjacent, pixels.

[0024] According to an advantageous design of the present invention, the coupling and reflection components contained in the digital received signal from within the lidar system or its immediate surroundings, in particular the cover, are compensated by the following sequence: the sequence is formed in a fixed-wiring digital circuit by correctly assigning symbols to one or more average values ​​determined by the product of the received sequence and the modulation sequence.

[0025] The method according to the invention is used to determine and implement correction values ​​by which coupling effects and reflection effects within the lidar system or in its immediate surroundings, in particular in the cover, can be compensated. The method according to the invention can also be used without superimposed frequency modulation.

[0026] According to an advantageous embodiment of the present invention, the received sequence or its complex conjugate is multiplied by a phase modulation sequence or its complex conjugate at relatively offset positions, wherein the respective offset corresponds to the propagation time. The resulting product sequence is multiplied by a complex twiddling factor with an at least approximately linearly varying phase, which at least partially compensates the frequency contribution of the linear frequency modulation for the respective propagation time, wherein a frequency shift is additionally achieved if necessary. The sampling rate of the sequence is then decimated if necessary. A discrete Fourier transform is then performed, preferably via a fast Fourier transform. This operation is performed for all propagation times of interest, wherein the sequence of complex twiddling factors varies with the respective time shift.

[0027] For frequency modulations that are not exactly linear, linearity errors can be taken into account in the following way: Essentially, in a two-dimensional correlation filter, the received sequence or its complex conjugate is multiplied by a phase modulation sequence or its complex conjugate at relatively offset positions, where the respective offset corresponds to the propagation time. The resulting product sequence is multiplied by a sequence of complex twiddling factors, where the twiddling factors at least approximately correct the non-constant frequency contribution of the nonlinear frequency modulation for the respective propagation time. A discrete Fourier transform is then performed, preferably implemented as a fast Fourier transform. This operation is repeated for all propagation times of interest, with the sequence of complex twiddling factors used to correct for linearity errors in the frequency modulation varying with the respective time shift.

[0028] The multiplication by the complex twiddle factor for correcting frequency modulation linearity errors and / or propagation-time-dependent frequency shifts is preferably also implemented in a fixed-wire digital circuit, with the two twiddle factors optionally being summed / combined. Advantageously, this multiplication is implemented using a single programmable structure with low precision in the low bit range and thus low complexity.

[0029] In another advantageous embodiment of the present invention, the fixed-wire digital circuit has a pre-stage in which the Fourier transform of the signal sequence or its complex conjugate is multiplied by the Fourier transform of the relatively offset modulation sequence or its complex conjugate, and the sequence resulting from the multiplication is subsequently decimated and / or extended with zero padding if necessary, and then subjected to a multi-stage fast Fourier transform, each individual calculation operation being implemented in a dedicated circuit, in the first stage, the offset between the two Fourier transforms is changed with each clock cycle in the first stage, and an inverse Fourier transform result is generated at the output of the subsequent stage, wherein the result respectively relates to the output data generated by the pre-stage a number of cycles ago.

[0030] According to an advantageous design of the present invention,

[0031] Since, in particular for objects at greater distances, only one relative velocity assumption is possible or at least more reasonable based on the known, propagation-time-dependent frequency shift components, the use of real-valued mixers allows for a sign-correct determination of the received frequency and thus a unique determination of the relative velocity of the object.

[0032] According to another advantageous design of the present invention,

[0033] When using a real-valued mixer, symbol-correct determination of the received frequency, and thus unique determination of the relative velocity of the object, is achieved by varying the slope of the frequency change, acquiring and evaluating object data for different slopes, and by utilizing the fact that the two frequency shift effects, relative velocity and propagation time, have different relationships to each other.

[0034] Advantageously, the sign of the frequency change slope is changed, rather than the modulus, whereby the two frequency shift effects of relative speed and propagation time are superimposed with different signs, thereby enabling the sign of the received frequency to be determined.

[0035] According to an advantageous embodiment of the invention, the frequency change is used to change the radiation direction, in particular at least in a stepwise continuous manner, in order to be able to detect data from a plurality of pixels in different directions.

[0036] Advantageously, a change in the radiation direction by changing the frequency can be achieved by using a waveguide with a plurality of coupling points for transmission and reception, which are preferably located in an equidistant grid on a line.

[0037] Preferably, the change in radiation direction caused by the frequency change is used for a first spatial direction, and another device is provided for changing the radiation direction in a second spatial direction, which is at least essentially perpendicular to the first spatial direction. Preferably, one of the two spatial directions is a horizontal direction and the other is a vertical direction.

[0038] In an advantageous embodiment of the present invention, the change in radiation direction is achieved by material transmission or material reflection, and at least one optical material property is changed by a control parameter, in particular by applying a voltage or current, wherein the control parameter and the resulting change in the optical material property can be different locally.

[0039] For changing the radiation direction in a second spatial direction, the body has a constant, in particular triangular, cross-section in the waveguide orientation direction and, if necessary, can be combined with a lens which also has a constant cross-section in this dimension for beam focusing in the second spatial direction, the electrical control parameters for changing the optical properties being applied with reference to the orientation direction of the body with constant cross-section.

[0040] Furthermore, for changing the radiation direction in the second spatial direction, a planar transmissive or reflective liquid crystal element having a one-dimensional lattice structure and controlled by voltage can be used.

[0041] Advantageously, for changing the radiation direction in the second spatial direction, a transmissive or reflective liquid crystal array can be used, which has only a small number of elements, preferably only one rod-shaped element, in the waveguide orientation direction. Preferably, the liquid crystal array is also used for radiation bunching alone for the second spatial direction or for radiation bunching of an auxiliary lens, preferably with the aid of the liquid crystal array to compensate for manufacturing tolerances of the relevant optical components and their relative arrangement.

[0042] Suitably, a plurality of transceiver paths operating in parallel, each having a waveguide, may be provided, wherein the waveguide is used to change the radiation direction in the first spatial direction by changing the frequency, preferably being powered by a common laser source through subsequent phase modulation, and the plurality of transceiver paths open up different radiation directions in parallel in at least one of the two spatial directions.

[0043] According to an advantageous design of the present invention, different radiation directions in the second spatial direction can be generated by multiple parallel-operating transceiver paths, preferably by waveguides having the same radiation characteristics and arranged in parallel, or by electrically controlling parameters to change the optical properties of the transmissive or reflective material. Through multiple parallel-operating transceiver paths, for corresponding control of the transmissive or reflective material, the radiation direction either addresses a group of directly adjacent pixels or addresses a group of pixels that are at least partially spaced farther apart, wherein the control of the transmissive or reflective material is gradually changed to cover the entire target radiation direction range, and if necessary, the angular spacing of the pixels increases outward on non-equidistantly arranged side-by-side waveguides, in particular spaced waveguide groups, the spacing within the group increases outward, and preferably the optical radiation width in the second spatial direction increases outward.

[0044] The waveguides, which are preferably arranged in parallel and adjacent to one another, can differ from one another in order to open up different radiation direction regions in the first spatial direction by the same frequency change, so that only part of the radiation direction region in the first spatial direction is covered by a single waveguide.

[0045] Advantageously, the change in the radiation direction in the second spatial direction can be achieved continuously by electronic or mechanical means, thereby achieving a slow scan in the first spatial direction by frequency change, in particular in the following way: during a complete scan in the second spatial direction, the scan is only advanced by approximately one pixel in the first spatial direction by the continuous frequency change.

[0046] In another advantageous design of the present invention, a plurality of transceiver paths operating in parallel and preferably powered by a common laser source are provided, each of the transceiver paths having a switch matrix for performing sequential switching within the corresponding waveguide group, wherein waveguides of preferably the same type are respectively used to change the radiation direction in the first spatial direction by changing the frequency, and all waveguides are arranged side by side so that they completely open up the second spatial direction. If necessary, the angular spacing of the pixels is increased outward by the non-equidistant arrangement of the side-by-side waveguides, and preferably the width of the light radiation in the second spatial direction is increased.

[0047] In another lidar system according to the present invention, a transceiver path is provided including a switch matrix for sequentially switching between a plurality of waveguides, which are preferably parallel and adjacent to each other, wherein the waveguides are different so as to open up different radiation direction areas in the first spatial direction through corresponding frequency changes of the same type, so that a single waveguide only covers part of the radiation direction area in the first spatial direction and reduces the frequency tuning range required by the laser source, and the radiation direction change in the second spatial direction is achieved in a continuous manner by means of electronic or mechanical methods, thereby achieving slow scanning in the first spatial direction through frequency changes, especially during a complete scan in the second spatial direction, the scan is only advanced by approximately one pixel in the first spatial direction through continuous frequency changes.

[0048] Advantageously, the change in radiation direction that can be achieved by changing the frequency has different speeds, characterized in particular in that the radiation direction changes more slowly in the central region than in the outer regions.

[0049] Suitably, the frequency change for achieving the change in radiation direction may be generated by one or more frequency-tunable laser sources, whereby, in the case of using a plurality of laser sources, each laser source covers only a part of the frequency range required for the change in radiation direction.

[0050] According to the present invention, the lidar system also includes a transceiver unit and a scanner for a second spatial direction. The transceiver unit has one or more waveguides, which are used to perform frequency scanning in the first spatial direction. The scanner can also be used in combination with other modulation forms, especially when the frequency changes step by step, and can even be used in incoherent lidar systems.

[0051] In another advantageous embodiment of the present invention, when using a real-valued mixer, the correct determination of the received frequency and thus the unique determination of the relative velocity of the object is achieved in the following manner: the frequency change for changing the radiation direction in the first spatial direction has different, in particular alternating, signs in different radiation directions in the second spatial direction, and these different signs cause the two frequency shift effects of the relative velocity and propagation time to be superimposed with different signs on different, in particular adjacent, radiation direction planes when detecting the same object, so that the sign of the received frequency can be determined.

[0052] In lidar systems, deviations in the radiation direction may occur, in particular due to misalignment (deviation from the ideal position or functional state), inaccurately known frequencies and / or inaccurately known correspondences between optical properties and control parameters. These deviations can advantageously be determined based on the measured radial relative velocity to the stationary object so that they can subsequently be taken into account and / or corrected.

[0053] Suitably, a road surface may be used as the stationary object, the angle of which is preferably determined in the vertical direction based on the measured distance and the sensor installation height.

[0054] The method for determining the deviation of the radiation direction according to the invention can also be used for other modulation forms, since it is essentially based only on the Doppler measurement capability inherent in coherent lidar systems.

[0055] Preferably, the device for changing the radiation direction is used to compensate for misalignment and / or to adaptively adjust the detection range, in particular as a function of the traffic situation.

[0056] In another advantageous embodiment of the present invention, the position of the road surface, in particular the position of the road surface at a greater distance, is determined in the following manner: in order to better distinguish system noise, at the same vertical angle, for multiple pixels belonging to different, in particular adjacent horizontal angles and / or for different, in particular adjacent distances within a pixel, the correlation power or the complex product between the conjugate complex numbers of the first correlation and the second correlation is accumulated; for the product of the two correlations, its received signal originates at least partially from the same reflection point on the road surface, in particular by alternating distribution of the received sequence values.

[0057] The method according to the invention for determining the position of a road surface at a greater distance can also be used for other modulation forms, such as pure linear frequency modulation (usually consisting of two frequency ramps of opposite sign). The correlation values ​​used in this case are based on correspondingly different correlation calculations, which in the case of pure linear frequency modulation are performed in the form of a fast Fourier transform of each frequency ramp.

[0058] Advantageously, when using a real-valued mixer, sign-correct determination of the receiving frequency, and thereby unique determination of the relative velocity of the object, is achieved in the following manner: among the discrete phase values ​​used in an irregular sequence for phase modulation, there are at least two phase values ​​that are neither in phase nor in phase opposition; the correlation between the frequency-shifted receiving sequence and the complex modulation sequence or its complex conjugate is calculated for different receiving frequencies, and the sign of the receiving frequency is identified using the characteristic that the correlation has different levels at positive and negative receiving frequencies.

[0059] Advantageously, a value set of at least three phase values ​​may be used, which are nominally evenly distributed over one period (one circumferential period), but which may deviate significantly from the nominal position.

[0060] Preferably, the phase value is achieved by switching between lines of slightly different lengths; or by a combination of switching between lines of slightly different lengths and a switchable inverter.

[0061] The method according to the invention for determining the sign of a received frequency in a real mixer by means of a phase modulation comprising at least two phase values ​​which are neither in phase nor in opposition can also be used without superimposed frequency modulation.

[0062] In a further advantageous embodiment of the invention, the device for changing the radiation direction is monitored, in particular to ensure eye safety, in such a way that the received signal is checked in the corresponding spatial direction with respect to changes in object reflection or with respect to changes in the composition of internal reflections and couplings as well as cover reflections.

[0063] According to the present invention, the laser radar system particularly includes the following three main components, and the electronic components of the three main components are preferably on the same or at most two circuit boards: a photonic chip, which particularly includes a frequency-tunable laser source, a phase modulation unit and a waveguide, the phase modulation unit having a switchable inverter and parallel transmit and receive paths, a digital chip, which particularly includes an analog-to-digital converter, a fixed-wired computing logic device, one or more microcontrollers and / or digital signal processors, and a device for generating control signals, the fixed-wired computing logic device is used to determine two-dimensional correlation and has a subsequent evaluation device, and the third is a scanner for a spatial direction, which is realized by the following material: the material has electrically controllable optical properties, in particular a liquid crystal element or a liquid crystal array, or the scanner is realized by a switch matrix for each parallel transmit and receive path, or is realized mechanically. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A coherent lidar system using binary phase modulation is shown.

[0065] Figure 2 The pseudo-random binary phase modulation process is shown.

[0066] Figure 3 The real part of the low-frequency analog received signal for an object is shown.

[0067] Figure 4 Shows the two-dimensional correlation of the received signals of two objects.

[0068] Figure 5 The linear variation of the transmission frequency during phase modulation is shown.

[0069] Figure 6 The received frequency is shown with a time shift relative to the transmitted frequency in the absence of a Doppler shift (ie, the relative velocity is zero).

[0070] Figure 7 The two-dimensional correlation of the received signals of two objects is shown, wherein a linear frequency change is superimposed on a phase modulation.

[0071] FIG8 shows a cost-optimized fixed-wiring digital circuit for implementing two-dimensional correlation filtering; wherein: Figure 8a Shows an overview, Figures 8b to 8d Three modules are shown in detail.

[0072] Figure 9 A complex-valued multiplier implementation of an optimized cost for an exemplary twiddle factor is shown.

[0073] FIG10 shows a cost-optimized fixed-wire digital circuit for implementing two-dimensional correlation filtering using frequency shifting and decimation prior to the Fast Fourier Transform (FFT); wherein Figure 10a Shows an overview, Figures 10b to 10d Three modules are shown in detail.

[0074] Figure 11 Implementation of the rotation factor for frequency shifting is shown.

[0075] Figure 12 The course of the quadratic error in the transmission frequency relative to an ideal linear variation is shown.

[0076] Figure 13 The received frequency range is shown as a function of the object distance, with the range where a real-valued mixer would have ambiguity due to unknown signs being shaded.

[0077] Figure 14 The relationship between the receiving frequency range and the object distance is shown when the modulation bandwidth is inverted.

[0078] Figure 15 The relationship between the receiving frequency range and the object distance is shown when the modulation bandwidth is inverted and doubled.

[0079] Figure 16 A waveguide with equidistant coupling points is shown for focusing and scanning by frequency variation in a first spatial direction.

[0080] Figure 17 The linear progression of the beam angle over time and the corresponding transmission frequency for rotating the beam angle are shown.

[0081] Figure 18 The figure shows the change of radiation angle over time and the corresponding transmission frequency at different scanning speeds.

[0082] Figure 19 The arrangement of the waveguide and the lens for focusing in the second spatial direction is shown from two different angles.

[0083] Figure 20 Show Figure 19 The setup in

[15] also adds a prism-shaped device made of an electrically controlled dielectric constant material for scanning in a second spatial direction.

[0084] Figure 21 The arrangement of the waveguide and the transmissive one-dimensional liquid crystal array for focusing and scanning in the second spatial direction is shown from three different angles.

[0085] Figure 22 The setup of the waveguide and reflective one-dimensional liquid crystal array for focusing and scanning in the second spatial direction is shown from three different angles.

[0086] Figure 23 A two-dimensional liquid crystal array is shown.

[0087] Figure 24 An arrangement with 32 parallel and equidistant waveguides is shown.

[0088] Figure 25 Shown is a two-dimensional pixel field for 32 identical and equidistant waveguides with a smaller pitch.

[0089] Figure 26 Shown is a two-dimensional pixel field for 32 identical and equidistant waveguides with a pitch 32 times that of the former.

[0090] Figure 27 Shown is a two-dimensional pixel field for 32 identical waveguides with different pitches.

[0091] Figure 28 A two-dimensional pixel field of 32 waveguides of different equidistant spacing with a small pitch is shown, and is continuously stepped and scanned in the second spatial direction.

[0092] Figure 29 A two-dimensional pixel field of 32 differently equidistant waveguides with closely spaced intervals is shown, and is continuously scanned in a second spatial direction.

[0093] Figure 30 The derivation of the radial component of the relative velocity of a stationary object with an elevation angle of 0° is shown.

[0094] Figure 31 The derivation of the radial component of the relative velocity of a stationary object at any position angle is shown, as well as the derivation of the angle seen from the road surface in the vertical direction.

[0095] Figure 32 The irradiated road surface area is shown.

[0096] Figure 33 An implementation of three equidistant phase values ​​is shown, with a switch between line segments of three different lengths.

[0097] Figure 34 An implementation of four equidistant phase values ​​is shown, wherein a switchable inverter and a switch are placed between two line segments of different lengths.

[0098] Figure 35 A modulation sequence according to the prior art is shown, which is composed of two subsequences.

[0099] Figure 36 A new modulation sequence with two subsequences periodically nested within each other is shown.

[0100] Figure 37 Taking two objects as an example, two fast Fourier transforms of the first modulation subsequence are shown.

[0101] Figure 38 Taking two objects as an example, two correlations of the second modulation subsequence are shown.

[0102] Figure 39 As an alternative, a novel modulation sequence is shown having two subsequences periodically nested within each other.

[0103] Figure 40 A novel modulation sequence according to the present invention is shown, which comprises two periodically nested subsequences and superimposed linear frequency modulation. DETAILED DESCRIPTION

[0104] Figure 1 A schematic diagram of a coherent lidar system 1.1 is shown. A coherent signal with a wavelength range of approximately λ = 1550 nanometers (nm) is generated using a laser source 1.2; the coherence length is at least a few microseconds, and according to the prior art, the frequency remains constant. The signal then enters a switchable inverter 1.3, which changes the sign of the signal, which is equivalent to a phase shift of 180°. The change of the signal sign only occurs in a fixed grid, for example 3.33 nanoseconds (ns). According to the prior art, the change of the signal sign is pseudo-random, i.e., at T m = 3.33 nanoseconds later, the frequency or probability of the signal sign changing is only 50%. Figure 2 The history of the modulation sequence b(n) is shown. The history consists of the values ​​+1 and -1 and is also called binary. m = 4096 cycles, i.e., every 13.6 microseconds (μs). The modulated signal passes through amplifier 1.4, circulator 1.5 (i.e., transceiver duplexer), and is transmitted through transceiver unit 1.6. It is partially reflected by an object 1.7; the delay depends on the distance r to the object and is therefore variable.

[0105] t0=2r / c,(1a)

[0106] Where c = 3·10 8 Meters per second (m / s) is the speed of light. The frequency shift depends on the radial relative velocity v and is therefore variable. It is generated by the Doppler effect.

[0107] f D =2v / λ(1b)

[0108] The signal is then received by the transceiver unit 1.6 and transmitted to the subsequent receiving path through the circulator 1.5. In the complex mixer 1.8 (also called IQ mixer), the modulated received signal is superimposed with the unmodulated laser signal and converted into a complex low-frequency signal with the help of the photodiode unit 1.9; as shown in formula (1b), the frequency f of this signal is e Corresponding to the Doppler frequency shift f D :

[0109] f e = f D = 2v / λ, (2)

[0110] Modulation of the signal delays the signal propagation time relative to the transmitted signal. Figure 3 The low-frequency analog received signal e is shown in the target case. a The signal is then converted to real part of (t) in the analog-to-digital conversion unit 1.10 at a sampling frequency f s =300MHz, that is, every T s = 3.33 nanoseconds for sampling and digitization, and the real part of the result is Figure 3 It is marked as a point; since the sampling time T s = 3.33 nanoseconds and modulation time T m =3.33 nanoseconds, so the number of sampling values ​​per modulation cycle is N s Also with the grid length N m Equal, that is, N s =N m =4096, so N=4096 is used uniformly below, without a subscript. The complex-valued sampled signal e(n), also referred to as the received sequence, can be described as follows:

[0111]

[0112] Where, it is assumed that the propagation time t0 is the modulation time T m = an integer multiple of 3.33 nanoseconds m0, so it is also the sampling time T s = integer multiples of 3.33 nanoseconds:

[0113] m0=t0 / T m =2r / c / T m (3b)

[0114] And, corresponding to the Doppler shift,

[0115] k0=N·T s ·f D =N·T s 2v / λ (3c)

[0116] Also an integer; a is the complex-valued amplitude of the received sequence, and "exp" represents the exponential function. Is an imaginary unit.

[0117] According to equation (3), the complex-valued received sequence e(n) refers to a single object with no longitudinal extension and an ideal receiver. In reality, there may be multiple objects and / or extended objects, and additional noise r is generated in the receiver. e (n), especially thermal noise; then the resulting received sequence is

[0118]

[0119] where, "sum i＝1、......、I " represents the sum function of I non-vertically extended single objects at sequence numbers i = 1,..., I.

[0120] The discrete propagation times m 0.i and discrete Doppler frequency shifts k 0,i of I objects can be determined from the received sequence e(n) in time periods n = 0, 1,..., N - 1. To obtain as accurate a determination result as possible, that is, to achieve as good a separation of the signal and noise as possible, so as to obtain the maximum sensitivity and range of the lidar system, the so-called optimal filtering method needs to be adopted, that is, through the two-dimensional space between the received sequence e(n) and the ideal amplitude-normalized received sequence of a single object for correlation filtering: where, m = 0,..., M - 1 and k = 0,..., N - 1; (5)

[0121] Here, M - 1 corresponds to the assumed or interested maximum object distance, and the Doppler frequency shift k is assumed to be able to take any value. Thus, the two-dimensional correlation E m，k results in

[0122]

[0123] where, "conj" represents the complex conjugate operation, and the modulation sequence b(n) remains unchanged during this process due to its real-valued nature. The correlation E m，k has a modulus peak (often called a power peak) at the object position (m, k) = (m 0，i , k 0，i ); Figure 4 shows the modulus of the two-dimensional correlation of two objects with the same received amplitude, and their positions are respectively at (m 0，1 , k 0，1 ) = (300, 3846) and (m 0，2 , k 0，2 ) = (101, 1000), which corresponds to their distances r1 = 150 meters (m) and r2 = 50.5 meters and radial relative velocities v1 = -7.1 m / s and v2 = 56.8 m / s (for the asymmetric range k = 0,..., N - 1 selected above for k, for k > N / 2, a negative Doppler frequency is assigned, and thus a negative relative velocity is assigned; a negative relative velocity indicates the relative away movement of the object). <00, 01020>

[0124] To determine the distance r i and the radial relative velocity v i, we need to determine the two-dimensional correlation E m，k The modulus peaks are calculated, where only the modulus peaks above the detection threshold are used to distinguish them from the system noise. According to equations (3b) and (3c), r i and v i The position of the modulus peak, i.e. the discrete propagation time m, can be 0，i and discrete frequency shift k 0，i The calculation is as follows:

[0125] r i =m 0，i ·c·T m / 2, (7a)

[0126] v i =k 0，i ·λ / (2N·T s ). (7b)

[0127] Therefore, the distances and relative velocities of multiple objects can be determined directly and unambiguously using a single modulation sequence. This is a significant advantage over the linear frequency modulation with two frequency slopes of opposite signs commonly used in coherent lidar systems, where ambiguity is unavoidable in the presence of multiple objects.

[0128] The calculation of this two-dimensional correlation and the downstream analytical evaluation are performed in the digital signal processing unit 1.11. This represents a high computational workload of the order N·M·N. However, the above equation (6) can also be viewed as the discrete Fourier transform of the product e(n)·b(nm), n=0, ..., N-1; this calculation must be performed for each m=0, ..., M-1; the discrete Fourier transform (DFT) is calculated by the fast Fourier transform:

[0129] E m，k =FFT k (e(n)·b(nm)), where m=0, ..., M-1, (8)

[0130] Here, k=0, ..., N-1 is the output dimension of the fast Fourier transform, that is, the discrete frequency, and the computational workload is reduced to the order of M·N·log2(N).

[0131] Phase modulation superimposed with linear frequency modulation

[0132] According to the prior art, it has been previously assumed that the frequency of the applied phase modulation sequence b(n) is constant, i.e. does not vary. Below we will consider the method according to the invention for continuously varying the frequency, wherein firstly an ideal linear variation is assumed; Figure 5 As shown, the frequency of the laser source 1.2, and therefore the transmission frequency fTX (t) for a duration of T pm =N·T m =13.7 microseconds modulation period with B = 800MHz (with slope B / T pm =58.6MHz / microsecond) increases linearly. Therefore, the general transmission frequency f TX (t) Satisfy:

[0133] f TX (t) = f TX0 +t·B / T pm , where T pm =N·T m (9)

[0134] In the absence of Doppler shift (i.e., initially assuming the relative velocity is zero), Figure 6 As shown, the frequency f of the received signal RX (t) is delayed accordingly due to the propagation time t0:

[0135] f RX (t) = f TX (t-t0)=f TX0 +(t-t0)·B / T pm , (10)

[0136] Therefore, it is related to the transmission frequency f TX (t) is shifted downward compared to the frequency shift caused by the propagation time f r for

[0137] f r =-t0·B / T pm (11a)

[0138] Substituting the propagation time t0 according to equation (1a):

[0139] f r =-2r / c·B / T pm ; (11b)

[0140] For the example of an object distance r = 150 m, a propagation time t0 = 1 μs and the above modulation values, the frequency shift caused by the propagation time is f r =73.2MHz. According to equation (11b), the frequency shift f caused by the propagation time is r Proportional to the object distance r.

[0141] The total frequency shift and the received signal frequency after mixing f e The Doppler shift f according to equation (1b) D and the propagation time-induced component f according to equation (11b) r composition:

[0142] f e =f D +f r =2v / λ-2r / c·B / T pm (12)

[0143] This is a particular difference from the case of constant transmission frequency initially considered in the prior art. At this reception frequency, the phase modulation sequence delayed by the propagation time is applied unchanged, so that for the sampled and digitized reception sequence e(n), equation (3a) still applies, whereas for the discrete reception frequency k0, instead of equation (3c) the following applies:

[0144] k0=N·T s ·(f D +f r )=N·T s ·2·v / λ-N·T s ·2·r / c·B / T pm , (13a)

[0145] And using the discrete propagation time m0 and T according to Equation (3b) pm =N·T m :

[0146] k0=N·T s ·(f D -f r )=N·T s ·2·v / λ-m0·T s B, (13b)

[0147] The discrete receiving frequency k0 is also initially assumed to be an integer. Therefore, the optimization filtering method is to normalize the receiving sequence two-dimensional space by the receiving sequence e(n) and the possible ideal amplitude of a single object. The method of filtering the correlation between the two is still effective, which is characterized by the two-dimensional correlation E m，k Formula (6) and (8); Formula (8) applies here:

[0148] E m,k =FFT k (e(n)·b(nm)), where m=0, ..., M-1, ((8))

[0149] That is, a fast Fourier transform is performed by corresponding multiplication between the received sequence e(n) and the corresponding shifted modulation sequence b(nm), thereby achieving a computationally low computational cost. Figure 7 Shows the two-dimensional correlation E of the above two object examples m，kThe distance between the two objects is r1 = 150 m and r2 = 50.5 m, the radial relative velocity is v1 = -7.1 m / s and v2 = 56.8 m / s; the discrete time displacement is m 0，1 =300 and m 0，2 =101 remains unchanged, while the discrete frequency shift is reduced by the component -m caused by the propagation time according to equation (13b) 0，1 ·T s B=-800 and m 0，2 ·T s B = -264 becomes k 0，1 =3846-800=3046 and k 0，2 =1000-264=736.

[0150] For the peak position of the modulus value of the correlation (m 0，i , k 0，i ) Determine the object distance r i , formula (7a) still applies:

[0151] r i =m 0，i ·c·T m / 2, ((7a))

[0152] For the radial relative velocity v i To determine the discrete frequency shift k caused by the propagation time, it is necessary to consider 0，i The components of - Using formula (13b) we can get:

[0153] v i =λ / (2N·T s )·(k 0，i +m 0，i ·T s ·B); (14)

[0154] Different from the original formula (7b), we now need to start from the discrete frequency shift k 0，i Subtract the contribution due to propagation time from 0，i ·T s B, the contribution value and the discrete propagation time m 0，i Proportional.

[0155] The related system approach and the advantages brought by combining phase modulation and varying frequency will be explained in detail in the following sections.

[0156] Implementing the calculation logic of two-dimensional correlation

[0157] First, we discuss how to implement the two-dimensional correlation E in the digital signal processing unit 1.11 m，kCalculation of. Previously, we considered the received sequence e(n) in the time period n=0.1, ....., N-1, and the corresponding correlation E m，k Refers to a single detection direction, i.e., horizontal and vertical directions based on one pixel. In fact, in each detection cycle, which was originally assumed to last 100 milliseconds, about 160,000 detection directions, i.e., pixels, are covered; this is generally achieved by a combination of parallel transceivers, i.e., parallel detection of pixels, and scanning, i.e., sequential detection of pixels. Parallel transceivers mean Figure 1 There are a plurality of, for example, 32, elements 1.4 to 1.10 (except, if necessary, the common focus rotation device and the radiation deflection device in the transceiver unit 1.6).

[0158] Scanning can be achieved, for example, by continuous mechanical movement (e.g., of a mirror element) or electronically, by continuous variation or by sequential switching (possible approaches to electronic scanning will be described later). In continuous scanning, pixels may also partially overlap, i.e., the last N values ​​of the received sequence e(n) of one pixel are also used as the first values ​​of the next pixel. If we now assume that M=500 (corresponding to the maximum distance of 249.5 meters in the above explanation), the fast Fourier transform of length 4096 in equation (8) must be calculated 800 million times per second. A microprocessor or DSP (digital signal processor) cannot perform such a large number of fast Fourier transform calculations; the clock frequency of such processors is generally in the range of 1 GHz, i.e., almost the entire fast Fourier transform of length 4096 must be calculated per clock frequency. However, modern processors, even when using parallel vector computing units, can usually only perform a maximum of 100 multiplication and addition operations per clock frequency, which is several orders of magnitude lower than the number of operations required for a fast Fourier transform of length 4096. Therefore, such a large number of fast Fourier transforms can only be implemented through dedicated computational logic implemented in hardware. Because the fast Fourier transform algorithm consists of many sub-units called butterfly operations, hardware-implemented FFT dedicated computational logic usually implements multiple butterfly operation units. These butterfly operation units contain programmable multipliers because the twiddle factors to be multiplied vary with the butterfly operation sequence. However, the implementation of programmable multipliers is very complex.

[0159] Since the 800 million Fast Fourier Transforms to be calculated per second roughly corresponds to the 1 GHz clock frequency achievable by this type of computational logic, it is possible to implement each Fast Fourier Transform butterfly, along with each adder and multiplier (for the corresponding twiddle factor) contained therein, directly in the computational logic, eliminating the need for programmable multipliers. As shown in module 8.4 of Figure 8 , a Fast Fourier Transform of length 4096 consists of log2(4096)=12 sequential stages, each with 2048 butterfly units. Each butterfly unit determines two output values ​​from two complex-valued input values ​​by performing complex-valued additions and subtractions and a complex-valued multiplication (in Figure 8 , the example of a butterfly unit in the first Fast Fourier Transform stage is highlighted with a bold line). Since a large number of sequential computational operations cannot be completed within a single clock cycle, registers must be inserted for intermediate storage. In Figure 8 , such intermediate storage is inserted between the 12 Fast Fourier Transform stages, using module z. -1 In practice, the number of intermediate memories can be greater, because, for example, the input feed lines of the first stage are very long, and additional intermediate memories may be required if necessary (the intermediate memories receive the values ​​at their inputs in every clock cycle). Therefore, the calculation is performed in a so-called pipeline - the fast Fourier transform calculation spans multiple clock cycles, and the calculation circuit contains multiple fast Fourier transform data; in each clock cycle, the fast Fourier transform input data is fed into the calculation circuit implemented as a pipeline, and the result of the fast Fourier transform, that is, the output data, appears at the output of the calculation circuit after multiple clock cycles, so that a new fast Fourier transform result is obtained in each clock cycle.

[0160] The main work of realizing this operation logic is done by the multiplier. Among them, the product between the complex value signal and the rotation factor is

[0161]

[0162] where p=1, ..., log2(N)-1 and q=1, ..., N / 2 p -1,

[0163] That is, a unit pointer / unit vector (modulus = 1); four real-valued multipliers are usually required. Each of these real-valued multipliers is generally implemented by adding a large number of shifted values. However, a high precision factor is not required here; for example, an error of up to 1 / 32 can be tolerated, even if quantized values ​​are used.

[0164]

[0165] Here, "round" represents a rounding function. By this rounding function, the correlation E m，kThe noise generated at the output is lower than the required dynamic range and is generally lower than the receiver noise. Moreover, the signal loss caused by this noise is negligible. Therefore, only multipliers with factors of ±1 / 16, ±2 / 16, ..., ±15 / 16 need to be implemented. For example, a multiplier with a factor of 7 / 16 is used; since

[0166] 7 / 16=8 / 16–1 / 16=1 / 2-1 / 16=2 -1 -2 -4

[0167] This is achieved by subtracting the input value shifted right by four bits from the input value shifted right by one bit—assuming a binary representation is used here; the above 7 / 16 representation is called the Canonical Signed Digital Code (CSD). With the exception of the factors ±11 / 16 and ±13 / 16, all of the above factors can be implemented with at most one addition or subtraction according to equation (16); to avoid using two additions or subtractions for these factors ±11 / 16 and ±13 / 16, they are approximated as ±10 / 16 and ±14 / 16, which still results in acceptable quantization noise.

[0168] Figure 9 The complex-valued multiplier for the twiddle factor is shown in

[0169]

[0170] From a 9-bit binary input value After multiplication, the output value of the same length is obtained Example values ​​are input. The real and imaginary parts of the output values ​​are each implemented using two real-valued multipliers and an adder. The two partial results of the real and imaginary parts are then added together. Since the negative values ​​of the real and imaginary parts of the input values ​​are also required, a negation operation is required. This is done bit by bit, meaning the extra 1, the so-called least significant bit (LSB), is omitted. The resulting error can be compensated by adding a correction value to the input value of the fast Fourier transform. To do this, the effect of the missing 1 during the negation is determined at the output of the fast Fourier transform and transferred to the input via an inverse discrete Fourier transform (DFT). When right-shifting the binary value (to implement the multiplier), the trailing part is simply omitted, meaning no rounding is performed. The resulting error is a mean error, which can also be compensated by adding a correction value to the input value of the fast Fourier transform. When right-shifting, the bit length of the value remains unchanged; therefore, in the double-compensated representation considered here, the highest bit of the input value is simply expanded, i.e., copied. The right shift itself is simple to implement with the appropriate wiring, requiring no additional effort. Since the modulo value of the twiddle factor being multiplied is always 1, the input and output values ​​of the multiplier have the same value range; no additional bits are required to expand upward.

[0171] In the butterfly unit, complex-valued addition and subtraction are performed on two complex-valued values. Therefore, the modulus of the result may be twice that of the input value, so the value range must be expanded by one bit. Thus, the bit length increases by 12 over the 12 stages of the fast Fourier transform. However, the noise component of the value due to receiver noise also increases with the addition and subtraction, with an average increase of √2. Therefore, the noise amplitude doubles after every two stages. Therefore, the lowest-valued bit, the least significant bit (LSB), can be omitted at every other stage (i.e., scaled by a factor of 0.5). The resulting quantization noise is smaller than the effect of receiver noise because the value range at the fast Fourier transform input is selected so that the receiver noise already has multiple least significant bits (LSBs). The effect of simply omitting the least significant bit (LSB) (i.e., not rounding), namely the resulting average error, can be compensated by adding a correction value back to the fast Fourier transform input value. In the circuit of Figure 8, this scaling is omitted in the final stage because there are no other computational steps within the FFT that can benefit from a reduction in bit length; therefore, the bit length is increased from 8 bits at the input to 15 bits at the output by the FFT.

[0172] According to FIG8 , a fast Fourier transform is implemented in a structure with frequency domain decimation (decimation-in-frequency-FFT) so that the longest wiring in the structure and the non-trivial multiplication operation are placed in the early stage with lower bit length (the last two stages have no multiplication operation; the factor (Only the corresponding wiring is shown.) Furthermore, this structure avoids the need to reorder the input data in the form of long lines; in this case, the output data does not need to be reordered into its natural order for further processing. Therefore, this structure requires less effort to implement than the alternative fast Fourier transform structure with time-domain decimation (decimation-in-time FFT).

[0173] According to formula (8), to determine the correlation E m，k , the Fast Fourier Transform needs to be applied to the product of the received sequence e(n) and the shifted modulation sequence b(nm). Due to the periodic nature of the modulation sequence b(n) (its period is N), the product of the unshifted modulation sequence b(n) and the periodically shifted received sequence e(mod N (n+m)), where “mod N ” represents the modular function modulo N, and then the Fast Fourier Transform is applied to it:

[0174] E m，k =FFT k (e(mod N (n+m))·b(n)), where m=0, ..., M-1; (17)

[0175] The correlation values ​​differ from those in equation (8) in phase, but are identical in modulus, and only this modulus is relevant for further evaluation, so for simplicity the same notation is used here (this relationship arises from the time shift of the Fourier transform). N The product of (n+m)) and the modulation sequence b(n) is formed in module 8.2 of Figure 8 and is implemented by a switchable inverter. For values ​​of 1 in b(n), the input value remains unchanged, and for values ​​of -1, it is simply inverted bit by bit. This inversion actually requires the addition of a least significant bit (LSB), and the effect of omitting this addition is compensated by adding a correction value to the input value of the fast Fourier transform in module 8.3. It should be noted that a switchable inverter is only required if the modulation sequence is likely to change. If the modulation sequence does not change, a hard inversion can be implemented, but of course only when the inversion is in progress. The periodic offset of the received sequence is determined by register chain z -1 Implementation where the received sequence is initially loaded with values.

[0176] Prior to this, in module 8.1, a correction value c1(n) is also added to the received sequence to compensate for the effects of coupling and reflections inside the lidar system or in its immediate vicinity, in particular in the cover; this will be discussed in more detail later.

[0177] As mentioned above, to simplify the quantization calculations, the inversion is implemented using pure truncation and pure bit-by-bit inversion; the effects of the resulting mean value error are compensated by adding a correction value c2(n) in module 8.3 before the fast Fourier transform. This correction stage can also be implemented after the fast Fourier transform instead of before it.

[0178] After the fast Fourier transform, that is, when constructing the correlation E m，k Afterwards, the result is further processed. First, a modulus value is generated for each of the N=4096 complex values ​​in module 8.5. Since high precision is not required here, the complex values ​​can be The modulus |i| uses the following approximation:

[0179] |i|=max(u+v / 8, uu / 8+v / 2), where u=max(|i Re |,|i Im |) and v=min(|i Re |,|i Im |), (18)

[0180] Here, "max" and "min" represent a maximum value function and a minimum value function, respectively; the calculation can be performed with fewer logical operations.

[0181] The resulting modulo values ​​of N = 4096 values ​​are used for both the summation operation in module 8.6 and the maximum value calculation in module 8.7. Both modules are cascaded; in each of the 12 stages, the sum or maximum value of a pair of values ​​is calculated. The registers required between the stages are not shown.

[0182] To distinguish between peaks in the magnitude of the correlation generated by the object and peaks in the noise, a summation calculation is performed to estimate the noise level. Since the magnitude of the correlation generated by the object is very small, this means that most of the values ​​represent noise. Therefore, dividing the summation result by 4096 (i.e., shifting it right by 12 bits) provides a good estimate of the noise level.

[0183] In the frequency shift dimension (which contains not only the Doppler component caused by the relative velocity but also the component caused by the propagation time), the maximum value of the modulus of N = 4096 FFT output values ​​and their associated label k are determined for each time shift m (corresponding to the distance). If this maximum value is at least three times the estimated noise, it is considered to be generated by the object; according to the corresponding time shift m = m 0，i and the corresponding frequency shift mark k=k 0，i , the distance and relative speed of each corresponding object i can be determined using equations (7a) and (14), and its reflectivity can be determined from the correlation level. As shown in module 8.7, if only the absolute maximum is determined, only the object with the highest reflectance intensity in each corresponding pixel at a given distance can be determined. If we want to cover the very rare case where two objects with different relative speeds exist within a distance (approximately half a meter), the cascaded structure of the maximum search, for example, eight modules of equal length, can also output the corresponding maximum values ​​of multiple value blocks. If the input data of the maximum search are appropriately configured, multiple modules can also be used to interpolate the peak value of the modulus in the fast Fourier transform to more accurately determine the frequency shift; because this modulus peak usually appears in two adjacent fast Fourier transform values ​​(because it is not at the integer index k0 as previously considered), if the input data is configured accordingly and placed in different modules of the maximum search, two values ​​can be obtained.

[0184] It should also be noted that no window function is used for the FFT, i.e. the FFT input values ​​are not multiplied by a bell-shaped curve; this is only necessary or meaningful when two objects with similar relative velocities but significantly different reflection intensities appear at the same distance from a pixel and need to be separated. In particular, if no window function is used at the FFT input and the frequency shift index k0 is not an integer, i.e. the modulus peak is split between two adjacent FFT values, then the sensitivity of the FFT output is reduced (i.e. the ability to detect objects with weak reflectivity and at a distance is reduced). This effect can be reduced by choosing a longer FFT length than its input signal, i.e. adding zeros to the input signal, which is called zero padding.

[0185] Regarding the marker determination in the maximum search, it should also be noted that, thanks to the cascaded implementation, the marker can be constructed bit by bit, starting with the least significant bit (LSB). At the output of each comparison between two values, in addition to the current maximum value, there is also a marker value whose bit length corresponds to the stage number. The resulting marker is related to the linear numbering at the maximum search input. Since the numbering at the fast Fourier transform output is mixed with the frequency-shifted marker k, a scaling / mapping operation is required later.

[0186] For the output dimension of the fast Fourier transform, i.e. the discrete frequencies, the asymmetric region k = 0, ..., N-1 has usually been considered; but the actual frequency shift k can assume two signs, generally limited by a previous low-pass filtering, for example as part of an analog-to-digital converter, where, for simplicity, it is assumed here that the limit range is k = -N / 2, ..., +N / 2, whereby the upper half of k = 0, ..., N-1 can be mapped to a negative value by subtracting N.

[0187] In the design discussed here (the sampling time T s = 3.33 nanoseconds and the linear frequency modulation bandwidth B = 800MHz), the relative velocity range at the fast Fourier transform output is about ±419 km / h when the target distance is zero (according to formula (14), where k 0，i =±N / 2), and when the maximum object distance is 249.5 meters, the relative speed range is about -147...+690 km / h (according to formula (14), where k 0，i = ±N / 2 and m 0，i = M-1 = 499); the possible range of relative velocities or the functionally interesting range is thus completely covered, with the exception of negative velocities at long distances, which even significantly exceed the coverage range - it will be shown later how a uniform over-coverage of the relative velocity range, i.e., for all distances, can be achieved in order to shorten the fast Fourier transform length by means of subsequent decimation. In general, decimation can be used before the fast Fourier transform if the possible frequency shift range is smaller than the frequency range of the fast Fourier transform, i.e., for example, if the sampling time and the modulation time are short and / or the length N of the modulation sequence is greater than the range considered for the adjacent positions; in the simplest case, such decimation is performed by calculating the partial sum of a sequence of products of the shifted received sequence and the modulation sequence.

[0188] As mentioned above, the received signal is low-pass filtered after the mixer—this can be done in a dedicated filter or as part of the analog-to-digital converter (especially if it is a delta-sigma converter). To achieve optimal sensitivity (i.e., optimal signal-to-noise ratio), an optimal filter should be used relative to the modulation format: if the modulated signal on the transmit side is rectangular and its shape is retained in the received signal (i.e., after mixing), the impulse response of the low-pass filter will also have this rectangular course. After low-pass filtering, the received signal elements will have a triangular course with double the length—but this course is only accurately obtained when the frequency shift is zero. For other frequency shifts, the larger the frequency shift, the less effective the filtering effect (from the perspective of maximum sensitivity) will be using this low-pass filter; this can be achieved by shortening the sampling time T. s (i.e. increasing the sampling frequency) can reduce the relative range of frequency shift and thus reduce the signal-to-noise ratio loss in low-pass filtering, where the modulation time T mIt can also be chosen to be higher than the sampling time in order to be combined with the decimation before the Fast Fourier Transform without significantly increasing the required computational effort. m，k Generally, a modulus peak will appear between two adjacent discrete distances m, so that the value can be interpolated to determine the distance more accurately.

[0189] At the output of the fixed-wire digital circuit shown in FIG8 , information is received every clock cycle, i.e., approximately every nanosecond, regarding the presence and relative velocity of an object at each respective pixel and at each respective distance under consideration. The received sequence e(n) for each pixel is loaded into a register once and stored, then calculated with a circular shift over M = 500 clock cycles (corresponding to 500 different offsets m and, therefore, different distances). After M = 500 clock cycles, the received sequence for the next pixel is loaded. All, for example, 160,000 pixels are evaluated sequentially within a detection cycle lasting 100 milliseconds.

[0190] The logic of the digital circuit shown in Figure 8 is primarily composed of additions—requiring approximately 300,000 adders with an average length of 12 bits. As the semiconductor technology structure size of digital circuits continues to shrink, such large-scale fixed wiring can be achieved from both a cost and power consumption perspective. Compared to frequency modulation in existing technologies, phase modulation shifts more of the implementation cost of coherent lidar systems to the digital domain (because the analog part becomes simpler, as described later, there is no need for high-precision linear and complex-valued receivers that change frequency). Due to the rapid advancement of semiconductor technology used in digital circuits, the cost is more optimized, meaning the solution is more economical and reasonable.

[0191] Now let us consider an alternative structure of Figure 8. According to Equation (6), the two-dimensional correlation E m，k It can also be viewed as a one-dimensional time correlation between the received sequence e(n) and the sequence b(n) exp(j2π·n / N·k), which is calculated for each k as follows:

[0192] Where k = 0, ..., N-1, (19)

[0193] Among them, "CC m " represents the cyclic correlation between two sequences of length N, where m = 0, ..., N-1 is the dimension of the correlation output; since N > M in the considered design, the cyclic correlation will handle more discrete distances m than required.

[0194] The cyclic correlation in the time domain is equivalent to the product of discrete Fourier transforms in the frequency domain:

[0195] Where k = 0, ..., N-1, (20)

[0196] Among them, IFFT m represents the inverse fast Fourier transform, m = 0, ..., N-1 is its output dimension (here it is assumed that the inverse discrete Fourier transform (DFT) is implemented by a fast Fourier transform). According to the frequency shift theorem of the Fourier transform, the factor exp(-j2π·n / N·k) applied to the modulation sequence b(n) in the time domain represents an offset in the frequency domain, that is, the Fourier transform:

[0197] E m，k =IFFT m (E(l)·B(l+k))

[0198] Where E(l) = FFT l (e(n)), B(l) = FFT l (b(n)) and k=0, ..., N-1, (21)

[0199] Due to the frequency shift theorem of Fourier transform and the cyclic characteristics of discrete Fourier transform, the correlation modulus can be further converted as follows:

[0200] |E m，k |=|IFFT m (E(lk)·B(l))|

[0201] =|IFFT m (E(mod N (lk))·B(l))|

[0202] Where E(l) = FFT l (e(n)), B(l) = FFT l (b(n)) and k=0, ..., N-1. (22)

[0203] This relationship can be converted into a structure similar to that shown in Figure 8. The input value of the structure is the fast Fourier transform of the received sequence that needs to be calculated in advance. It is multiplied by the fast Fourier transform of the modulation sequence b(n) determined in advance in the form of a cyclic shift k. Then an inverse fast Fourier transform (IFFT) is performed, which differs from the fast Fourier transform only in the sign of the rotation factor; the output dimension of the inverse fast Fourier transform (IFFT) is the distance dimension m (in the structure shown in Figure 8, it is the frequency shift dimension k), that is, for a discrete frequency shift k, the two-dimensional correlation E m，kIn the distance dimension m=0, ..., N-1. Then the modulus calculation is performed again, followed by the summation and maximum calculation of the distance dimension. Although there may be multiple reflections at different distances in a pixel, the probability that these reflections have the same frequency shift (composed of the absolute value of the relative speed and the distance) is very low, so it is sufficient to determine the absolute maximum here; for example, consider the case of fog and the possibility of a stationary object in a pixel: although all reflections have the same relative speed, they have different frequency shifts due to different distances. By marking m=m 0，i and k = k 0，i The maximum value above the detection threshold of can be used to determine the distance and relative speed of the corresponding object i again with the help of equations (7a) and (14).

[0204] If the same modulation sequence b(n) is always used, the product E(mod N The multiplier for (lk))·B(l) can be implemented in fixed-wiring form—through approximation and CSD (Canonical Sign Digital) representation—with only low implementation overhead. If the modulation sequence changes, a programmable multiplier may be required, which significantly increases the implementation effort.

[0205] As mentioned above, assuming N > M (N ≈ 8M), this often results in processing more discrete distances m = 0, ..., N-1 than necessary. This can be avoided by decimating the values ​​before the IFFT. If the decimation factor is 8, the N = 4096 values ​​are reduced to just 512 values, which are then fed into the IFFT. In the simplest case, decimation is achieved by summing the corresponding eight adjacent values. This reduces the IFFT's dimension from the original 4096 to 512, significantly reducing the workload. With a length of 512, the IFFT still covers the entire distance range of length M = 500.

[0206] However, it should be noted that such a structure requires N = 4096 clock cycles per pixel (the Fast Fourier Transform of the input signal must calculate this many shifts); this is eight times the structure shown in FIG8 , which, however, reaches the maximum achievable value at a clock frequency of approximately 1 GHz. Therefore, this alternative structure of equation (22) must be constructed many times, which greatly offsets the advantage of a shorter inverse Fast Fourier Transform (IFFT) length. This alternative structure only makes sense when the product of the sampling length N (according to the design of the modulation length considered above) and the number of pixels is small, because the structure only needs to be constructed a few times, preferably once.

[0207] Optimize calculation logic using rotation factors and decimation

[0208] As mentioned above, the total frequency shift of the received signal after mixing, that is, the frequency f e It consists of the components caused by the Doppler shift and the propagation time due to the linear frequency modulation:

[0209] f e =f D +f r =2v / λ-2r / c·B / T pm , ((12))

[0210] For the discrete frequency k0 of the received sequence e(n), the following equation applies:

[0211] k0=N·T s ·(f D -f r )=N·T s ·2·v / λ-m0·T s B. ((13b))

[0212] The component caused by the propagation time (in the above formula, the distance r or the second component of the discrete propagation time m0) causes the estimation of the relevant frequency range to depend on the distance and increase accordingly. This means that compared with pure phase modulation, that is, without additional linear frequency modulation, the calculation structure shown in Figure 8 actually requires a longer fast Fourier transform and therefore requires more computational effort. This problem can be solved by eliminating the frequency component caused by the propagation time -m0·T before the fast Fourier transform. s B is solved; now we briefly derive it, and for this purpose, we convert the two-dimensional correlation E into m，k The conversion is as follows:

[0213]

[0214] The output dimension of the Fast Fourier Transform is now discrete frequencies k , so according to equations (13a) and (24), the peak position of the modulus of the fast Fourier transform is k 0 corresponds only to the relative velocity of the respective objects:

[0215] k 0=N·T s ·2·v / λ. (25)

[0216] The product of the received sequence e(n) and the shifted modulation sequence b(nm) is multiplied by the rotation factor before the fast Fourier transform

[0217]

[0218] This multiplication is performed in module 10.8 of the calculation structure shown in Figure 10, that is, before the start of the fast Fourier transform, together with the operation of adding the correction value c2(n) in module 10.3 to compensate for the mean error caused by truncation and pure bitwise inversion in the FFT; the modules in Figure 10 that are basically the same as the calculation structure shown in Figure 8 (mainly differing only in dimension) are marked with the same sub-numbers .1-.7.

[0219] When the fast Fourier transform dimension N=4096 remains unchanged, the discrete frequency k Its coverage is k = -N / 2, ..., N / 2, the range of relative speed v is about ±419 kilometers per hour (km / h) (derived from v = λ / (2N·T s )· k ,in, k = ±N / 2), which is significantly higher than the relevant speed range, such as v 最小 (v min ) = -80 km / h to v 最大 (v max ) = +280 km / h - For objects moving towards each other (positive sign of v), the absolute value of the relative velocity has a higher functional relevance than for objects moving away (negative sign). k The corresponding asymmetric range can be shifted by the corresponding shift frequency

[0220] k offset =N·T s ·2·(v 最大 +v 最小 ) / 2 / λ=N·T s ·(v 最大 +v 最小 ) / λ (27)

[0221] Frequency is

[0222]

[0223] Using the symmetric range for the conversion, the time-corrected and centered frequency can be obtained using equations (23) and (27):

[0224]

[0225] Using this frequency for the Fast Fourier Transform k In this case, the two-dimensional correlation E m，k Similar to the derivation of its representation according to formula (24):

[0226]

[0227] Therefore, before the fast Fourier transform, the product sequence e(n)·b(nm) must be combined with the modified rotation factor

[0228]

[0229] Multiply.

[0230] For example, v 最小 = -80 km / h to v 最大 = Relative speed range of +280 km / h, the frequency at the output of the fast Fourier transform k The symmetry range is k =-881, ..., +881. This range is smaller than half the length of the fast Fourier transform N / 2 = 2048, so a 2-fold decimation can be performed before the fast Fourier transform. In the simplest case, as shown in module 10.9 of Figure 10, this can be done by adding two consecutive values ​​out of a total of N values. Subsequently, the N / 2 sum values ​​constitute the input value of the fast Fourier transform, and module 10.3 is located before it to add the correction value c2(n). Since the length is halved N / 2 = 2048, the fast Fourier transform is reduced by one level, that is, there are only 11 levels (see the fast Fourier transform module 10.4 in the calculation structure shown in Figure 10). The frequency at the output of the fast Fourier transform Only extends to the range The subsequent modules 10.5-10.7 for calculating the modulus, sum and maximum also have only half the dimension N / 2=2048.

[0231] As mentioned above, the calculation structure needs to cover about 160,000 pixels; based on the previous considerations, each pixel needs to calculate M = 500 offsets m between the received sequence and the modulation sequence, which requires M = 500 clock cycles. When the maximum achievable clock cycle of the calculation logic is about 1GHz, only a single detection cycle of only about 100 milliseconds can be achieved (certain consumption such as data loading and system reconfiguration must also be considered here). However, in practice, a cycle time of 50 milliseconds is often required, which requires the implementation of a double calculation structure. To avoid this, the modulation time T m Effectively doubled, so that only half the offset needs to be calculated for each pixel; this can also be done by m =3.33 nanoseconds, this effective doubling is achieved by defining every two adjacent values ​​of the modulation sequence b(n) to be equal:

[0232] b(n+1)=b(n) where n=0, 2, 4, ..., N-2, (32)

[0233] Therefore, the sampling time T s and modulation time Tm = are still equal, so the received sequence e(n) and the modulated sequence b(n) are still correlated with the same discrete time n. The only difference between the above considerations and the formula is that for the offset m, only one of every two values ​​needs to be considered, that is, only even values ​​m = 0, 2, 4, ..., M-2 are considered. The two-dimensional correlation E in the calculation structure m，k The representation of (where it is the received sequence that is shifted, not the modulation sequence) is similar to equation (17) using equation (30) above, and taking into account the decimation:

[0234]

[0235] Where s(n)=e(mod N (n+m))·b(n)+e(mod N (n+1+m))·b(n+1),

[0236] n=0,2,4,......,N-2,m=0,2,4,......,M-2 and

[0237] According to formula (33), the periodically shifted received sequence e(mod N The multiplication of (n+m)) with the modulation sequence b(n) is described in module 10.2; according to module 10.1, the addition of the upstream correction value c1(n) remains unchanged from the initial structure in FIG. 8.

[0238] Since the structure shown in FIG10 only requires M / 2=250 times per pixel, which is half of the initial structure shown in FIG8 , a half-cycle time of 50 milliseconds can be achieved; in addition, since the data dimension of the main module is halved, the implementation workload is only about half of the original.

[0239] The simplest way to implement decimation is to add two adjacent values. The first-order low-pass filter (i.e., length 2) thus implemented has a large transition area and a slow slope; this leads to two consequences: on the one hand, in the relevant frequency range Significant changes in the mid-level (in and On the other hand, due to the interference of high frequency noise (in up to 2.15 dB when decimating), a loss of sensitivity occurs after decimation. By means of a higher-order low-pass filter, the transition region can be designed more clearly; for a third-order low-pass filter with four coefficients [0.5, 1, 1, -0.5], the level difference is only 0.91 dB and the maximum sensitivity loss is 1.88 dB. This third-order low-pass filter requires three additions, but still no multiplications, since the coefficients with a modulus of 0.5 can be implemented by shifting one position to the right, i.e. by pure wiring (the negative sign of the coefficient -0.5 can be simplified again by using bitwise inversion). Sharper edges can be achieved using higher-order low-pass filters and coefficients that are not in the quadratic grid (i.e. they themselves require one or more additions). It should also be noted that in principle frequencies above ±881 can also be considered and determined. (Relative speeds outside the corresponding range -80 km / h, ..., +280 km / h can therefore also be calculated); however, the level loss and sensitivity loss increase, and There will be ambiguity in the above (because the frequency will be mapped to a range below ±1024).

[0240] Now, the rotation factor d described using equation (33) is explained. n，m The multiplication operation is performed using a unit rotation vector with a phase range covering 0…2π full angle. The phase of the rotation factor does not need to be arbitrarily accurate, but can be selected from a finite set of phases that are closest to d n，m The simplest method is to only implement four phases: 0, π / 2, π, and 3π / 2, that is, the rotation factor is 1. -1 and This does not require a multiplication operation; however, the resulting effects are not negligible, namely the effective sensitivity loss and the generation of noise or pseudo-modulus peaks in the FFT output. Therefore, the method with 8 phase values ​​0, π / 4, ..., 7π / 4 will be considered below, that is, the rotation factor achieved is

[0241]

[0242] Therefore, in addition to the simple rotation factors ±1 and In addition, there is a rotation factor The factor 1 / √2=0.707 can be approximated by 1-2 -2 =0.75, i.e., it is efficiently implemented by addition (except for the inversion and right shift by 2 bits, which can be easily implemented without consuming too much work).

[0243] Due to the rotation factor d n，m will change with the change of the offset m, that is, change in the clock cycle of the computing logic, so a programmable structure is needed to implement them, such as Figure 11 n (there are N = 4096 such structures in total) as shown in Module 11.2. By using the rotation factor d n，m Multiplying, will have real part i Re (n, m) and the imaginary part i Im The complex input value (n, m) is converted to a value with real part o Re (n, m) and the imaginary part o Im The complex output value of (n, m); depends on which of the 8 rotation factors is used,

[0244]

[0245] The 6 switches (switching between 0 and the corresponding input values) and the 4 switchable bit inverters are controlled by the logic 11.21. In this logic, the required 10 binary switching signals are determined by the corresponding discrete phase values ​​p at the input. n，m =0, ..., 7, which is calculated from 3 bits. The corresponding values ​​p n，m , that is, according to formula (34):

[0246] p n，m =mod8(round(8·n / N·[m·T s ·B- k offset ])), where m = 0, 2, 4, ..., M-2, (36)

[0247] In module 11.1, the linear part (increasing gradually in the clock cycles of the calculation structure) with respect to the offset m is realized by recursively adding the value in register R2, and the constant part (- k offset ) is achieved by initially loading the integrator register RI from register R1. The two values ​​in registers R1 and R2 are round(-2 14 ·n / N· k offset ) and round(2 14 ·2·n / N·T s B) Compared with the value in formula (36), the value is increased by a factor of 2 11 Scaling, that is, the calculation is actually performed with 11 decimal places, in particular when gradually integrating the value of register 11.12, to achieve a sufficient accuracy (this value is integrated M / 2 = 250 times, that is, the initial rounding error of the register value will be magnified by a factor of 250; it should be noted that since only even values ​​of m are used, it is also necessary to round (2 14 ·2·n / N·T s=B) with a factor of 2. Since the integrator is implemented using 14-bit two's complement arithmetic, overflows above the effective value of 8 are ignored, essentially performing the modulo operation in (36). By using the top 3 bits, i.e., the 3 most significant bits (MSBs), quantization to the range p n，m =0, ..., 7; in fact, this is not rounding according to formula (36), but truncation is implemented, which is different from the rounding of the average 0.5 - because this average error is calculated over all N rotation factors (n = 0, ..., N-1), and therefore only represents a constant phase shift, which is not important from a functional point of view (a constant phase shift of the fast Fourier transform input data will only result in a constant phase shift of the fast Fourier transform output data, so its modulus, that is, its only relevant quantity, will not change).

[0248] according to Figure 11 As shown, the realization and rotation factor d n，m The multiplication operation requires only a small amount of effort. Alternatively, with a little more effort, a 3-bit programmable factor (i.e. for the values ​​-1, -0.75, ..., 0.5, 0.75) can be calculated using a conventional complex-valued multiplier (composed of 4 real-valued multipliers); since the value +1 cannot be realized, it may be more useful to scale the rotation factor by a factor of 0.875, for example. Similar to Figure 11 The factors used by the four real-valued multipliers can be obtained from the corresponding values ​​p by means of a logic n，m , wherein the value can also be longer than 3 bits (because more than 8 different twiddle factors can be realized using such a multiplier).

[0249] Compensation for nonlinear frequency history

[0250] Previously, the sending frequency f TX (t) Assuming an ideal linear process:

[0251] f TX (t) = f TX0 +t·B / T pm , where T pm =N·T m . ((9))

[0252] In reality, the modulation is not perfect; Figure 12 As shown, it usually has a quadratic error.

[0253] f TX，Q (t) = Q·(tT pm / 2) 2 (37a)

[0254] And the relevant sending frequency is obtained:

[0255] f TX (t) = f TX0 +t·B / T pm +Q·(tT pm / 2) 2 (37b)

[0256] In the absence of Doppler shift, the receiving frequency f RX (t) relative to the transmission frequency f TX The offset of (t-t0) corresponds to the propagation time t0; the frequency shift f caused by the propagation time between the transmitting and receiving frequencies r (t), we get:

[0257] f r (t) = f RX (t)-f TX (t) = f TX (t-t0)-f TX (t)=

[0258] =f TX0 +(t-t0)·B / T pm +Q·(t-t0-T pm / 2) 2 -[f TX0 +t·B / T pm +Q·(tT pm / 2) 2 ]

[0259] =-t0·B / T pm -Q·2(tT pm / 2)·t0+Q·t0 2 ; (38)

[0260] The next two terms describe the error

[0261] f r，Q (t) = f TX，Q (t-t0)-f TX，Q (t) (39a)

[0262] The frequency shift associated with the propagation time (according to equation (11a), the ideal linear frequency shift is not included in this frequency shift) is:

[0263] f r，Q (t) = -Q·2·(tT pm / 2)·t0+Q·t0 2 (39b)

[0264] After mixing and digitization, the error is directly transferred to the received sequence e(n), that is, the error k transferred to the discrete receiving frequency0，Q , and similarly to formula (13a), we can get:

[0265] k 0，Q =N·T s ·f r，Q (t) = N·T s ·(-Q·2·t·t0+Q·T pm ·t0+Q·t0 2 ),

[0266] Where t = n·T s (where n=0, ..., N-1), t0=m0·T m (See formula (3b)), where T pm =N·T m :

[0267] k 0，Q (n) = Q·N 2 ·T m 2 ·T s ·m0+Q·N·T m 2 ·T s m0 2 -Q·2·N·T m ·T s 2 ·m0·n

[0268] Where n=0, ..., N-1. (40)

[0269] Over the N values ​​of the received sequence e(n), the first two terms are constant, i.e., they represent a constant frequency error and are expressed in the two-dimensional correlation E m，k In the two-dimensional correlation E, only the modulus peak position in the frequency dimension k is changed; in contrast, the last term varies linearly with discrete time n, i.e., the frequency changes in the received sequence, which will m，k This results in a broadening of the modulus peak in the frequency dimension k—a width of approximately Q·2·N 2 ·T s 2 ·T m m0. The displacement and broadening of the modulus peak increase with increasing discrete propagation time m0 (i.e., with object distance). While the effects of displacement can be accounted for by simply converting the frequency k0 to relative velocity, broadening leads to reduced sensitivity (reduced modulus peak height), reduced accuracy in determining the modulus peak position and relative velocity, and poor separation of two objects at the same distance and with similar relative velocities.

[0270] The following explains how to avoid the broadening of the modulus peak. According to the frequency error f in equation (39) r，Q (t), the phase error can be obtained as follows

[0271]

[0272] The integration constant is omitted in this derivation because the constant phase component is independent of the function. Expressed in discrete time n, where t = n·T s , t0=m0·T m and T pm =N·T m , from which we can conclude:

[0273]

[0274] Where n=0, ..., N-1. (42)

[0275] For objects with discrete propagation time m0, this phase error can be corrected as follows: m，k When the product sequence e(n)·b(nm) is multiplied by the rotation factor, the negative value of the phase error is added to each offset m.

[0276]

[0277] Where n=0, ..., N-1 and m=0, ..., M-1 (43)

[0278] (Note that in the method shown in Figure 10, only a subset of even values ​​of m is used.) These rotation factors can be implemented according to equations (27) or (31) with the above rotation factor d n，m In order to compensate for the propagation time-dependent frequency shift and the possibly asymmetric Doppler frequency range, a rotation factor with the sum of the phases of the two respective rotation factors needs to be implemented. According to equation (43), d n，m，Q The first two phase components in are linear in the offset m, so in achieving something like Figure 11 When the twiddle factors are used, they can be implemented as an addition to register R2, which is the input to the integrator:

[0279] R2=round(2 14 ·2·[n / N·T s ·B+Q·T m ·T s 2 ·m·n 2 -Q·N·T m 2 ·T s·m·n]); (44)

[0280] The additional part is the two subsequent parts, which are proportional to the magnitude Q of the secondary frequency error. The subsequent part proportional to the square of the offset m in Equation (44) is usually negligible because M << N and it is significantly smaller than the other parts; however, if it is to be implemented, a second-order integrator is required - thus, there is also a third register, which is integrated into an upstream integrator, and the output of this integrator enters as another input into Figure 11 the integrator shown.

[0281] In principle, the correction of the frequency error can also be carried out after the fast Fourier transform instead of before the fast Fourier transform; multiplying by the rotation factor d n，m，Q before the fast Fourier transform, n = 0,..., N - 1, is equivalent to performing a convolution on the spectrum of this sequence of rotation factors (i.e., the discrete Fourier transform (DFT) or the fast Fourier transform). In the case of the secondary frequency error under consideration, this spectrum has a wide modulus peak near zero (the width of the modulus peak increases with the offset m); therefore, the convolution can be restricted to the summation of a few values and weighted with complex-valued factors - nonetheless, this implementation is much more complex than the method of correcting before the fast Fourier transform introduced above, also because these weighting factors depend on the offset m.

[0282] Previously, the secondary error of the transmission frequency described by Equation (37) was considered. The above considerations can also be applied to the general error f TX，Q (t) of the transmission frequency, that is, the general linear process deviation. To correct the transmission frequency error, when using Equation (41a) and t = n·T s and t0 = m0·T m the rotation factors are multiplied by the general form of the product sequence e(n)·b(n - m) as follows:[[]]

[0283]

[0284] These rotation factors can be recombined with the rotation factor d n，m in Equation (31) to compensate for the frequency shift related to the propagation time and the asymmetric Doppler frequency range. For an implementation similar to Figure 11 the discrete phase values p n，m range from 0, 1,..., 7 (i.e., with a length of 3 bits), which is an extension of Equation (36) and involves an ideal linear frequency modulation:

[0285]

[0286] For generating these discrete phase values Figure 11The module 11.1 shown can be replaced by a register in which the corresponding pre-calculated phase value is written. In addition to using a register for storing only 3-bit length phase values, a long register can also be used, in which all m discrete phase values, i.e., M / 2=250 3-bit length values ​​in the example considered above, are written, wherein the register is run for m clock cycles, i.e., pushed in a loop; the advantage of this is that the writing process does not have to be performed when calculating a pixel, and when multiple pixels have the same transmission frequency error, it only needs to be written once (and then pushed in a loop). This method of pre-calculating register values ​​can also be used with Figure 11 The method shown is combined with the use of an integrator circuit so that the registers only need to store the portion that cannot be realized by an integrator circuit. This may result in one register being used to generate multiple phase values ​​p. n，m , i.e. for different n and / or m, thus reducing the number of required registers and the amount of pre-calculation work (also for microcontrollers).

[0287] To compensate for a nonlinear frequency history, the error—that is, the deviation from a linear history—must be known. As mentioned above, uncompensated frequency history errors lead, in particular, to a broadening of the modulus peak in the frequency dimension k. Assuming a quadratic error, the width of the modulus peak has a fixed relationship to the error magnitude Q (see above). A nonlinear frequency history across pixels also leads to angular errors (which will be discussed in detail later, including how these angular errors can be determined). Knowing these angular errors allows inferences about the errors in the frequency history—not only across pixels, but also within pixels, since the error curve is typically quadratic, at least in certain regions.

[0288] Compensation for internal coupling and cover reflections

[0289] As mentioned above, in the fixed-wired computational logic modules 8.1 and 10.1 shown in Figures 8 and 10, correction values ​​c1(n) are added to the received sequence to compensate for coupling and reflection effects within the lidar system or its immediate surroundings, particularly in the housing. It's worth noting that radar sensors and coherent lidar sensors are based on the same coherent operating concept, but operate in a different electromagnetic frequency range, offering advantages in compensating for these effects. We will now further explain how these correction values ​​are determined in a coherent lidar system using phase modulation and how they can be easily implemented.

[0290] On the one hand, it is assumed that the discrete distance m = 0 is exactly at the position where the distance is zero, and on the other hand, it is assumed that the distance of these couplings and reflections is also small enough to be negligible (so the receiving frequency f e= 0; the Doppler component is also zero), in this case the level P generated by these couplings and reflections in the received sequence can be determined by the average value of the product of the received sequence e(n) and the unshifted modulation sequence b(n) (the signal portion of the received sequence originating from the actual object is on the one hand of lower level, and on the other hand, since it corresponds to the shifted modulation sequence and usually has a received frequency f e ≠0 and is therefore largely averaged). The correction value c1(n) is obtained by multiplying the level P thus determined by the value of the unshifted modulation sequence b(n) by ±1. The determination of this average value (by summing N=4096 values ​​e(n)·b(n) and right-shifting the result by 12 bits) can be performed outside or within the hard-wired calculation logic. If it is implemented within the calculation logic, an additional module can be implemented, or the two-dimensional correlation E at the discrete distance m=0 and discrete frequency k=0 can be used directly in the existing calculation logic. m，k The value of , which corresponds to the sum of e(n)·b(n). If the correlation result is used, it is necessary to consider that the calculation logic is implemented in a pipelined manner, so the calculation lasts for several clock cycles; therefore, if the level value calculated by the two-dimensional correlation is to be used in the same pixel, the offset of the input sequence in module 8.2 or 10.2 must be paused, that is, the clock cycle of this module must be paused. As an alternative, the level value of the previous cycle or the previously processed adjacent pixels can also be used (as long as the value P between the pixels does not change significantly). For the level value P determined in this way, the fixed-wire calculation logic assigns the correction value c1(n) to the value ±P; if the same modulation sequence b(n) is always used, fixed wiring of each corresponding sign can be achieved (the negative sign is preferably only achieved by bit-by-bit inversion), and if b(n) changes, a switchable inverter is required.

[0291] Due to the aforementioned smearing (loss) of the received pulse waveform, especially reflections from the cover, which can have a significant propagation time, signal components may still be visible even at a discrete distance m = 1. Therefore, the level in the received sequence e(n) depends not only on the respective value of b(n) but also on the previous value b(n-1). Therefore, two average values ​​must be determined: one for b(n) and b(n-1) at time n, when they have the same sign, and another for b(n) and b(n-1) when they have different signs. In the hard-wired logic, these two level values ​​are then sign-correctly assigned to the correction value c1(n), depending on whether b(n) and b(n-1) have the same or different signs at the respective n. In principle, these two level values ​​can also be determined by the sums of e(n)·b(n) and e(n)·b(n-1) for all n. These sums can be determined by explicit calculation or obtained from a two-dimensional correlation. If the tailing and / or offset of the received pulses is so severe that the value of the received sequence e(n) is affected by three adjacent values ​​of b(n), the above method should be extended accordingly.

[0292] Due to the linear frequency changes superimposed by phase modulation, the received sequence, especially the one reflected from the cover, may have a relatively low frequency. Even if only a small fraction of a cycle passes through the N = 4096 values ​​of a pixel, the compensation mechanism described above, which assumes a zero received frequency and a constant phase, is no longer fully effective. To reduce frequency effects, the level values ​​can be determined segment by segment, for example, in four segments of length N / 4 = 1024. Alternatively, if the received frequency is known to be relatively low, this frequency can be eliminated by multiplying by a corresponding rotation factor before determining the level value, and then reapplying these rotation factors when determining the correction value c1(n); however, this is relatively complex. The level values ​​can also be re-determined from a two-dimensional correlation; in this case, the unknown frequency can also be determined by interpolation.

[0293] If the effects of coupling and reflection are at least approximately constant over time and / or within a small pixel area, averaging can be performed over the detection period and / or pixel area. However, the effects are often not constant across all pixels, e.g., the phase position of the cover reflection may vary due to different radiation directions.

[0294] Determining the sign of the received frequency using linear frequency modulation when using a real-valued mixer

[0295] The following sections explain the advantages and systematic approach of combining phase modulation and varying frequency.

[0296] exist Figure 1As shown in the previously considered lidar system, the mixer is set as a complex-valued mixer. Compared with a real-valued mixer (with only one real-valued output), the workload of the receiving path is significantly increased (almost doubled). When using a real-valued mixer, only the modulus of the received frequency can be determined, but not its sign, because the correlation E m，k There are two modulus peaks (m0, +k0) and (m0, -k0) in the prior art. In the prior art, if the transmission frequency is constant, only the absolute value of the radial relative velocity can be determined, but its sign cannot be determined. According to the prior art, in order to determine the sign, the sign can be determined by tracking - that is, tracking over multiple acquisition cycles - and / or by rationality verification - for example, whether the absolute value of the measured radial relative velocity corresponds to a stationary object - but this also has some disadvantages. These disadvantages can be reduced by superimposing linear frequency modulation on phase modulation. According to formula (12):

[0297] f e =f D +f r =2v / λ-2r / c·B / T pm , ((12))

[0298] After mixing and digitization, the received frequency f e The Doppler shift f generated by the relative velocity v D and the frequency shift f caused by linear frequency modulation r Composition, where f r Proportional to the object distance r determined by means of phase modulation. Proportional to the relevant relative speed range v 最小 ,……,v 最大 Corresponding receiving frequency range f e Therefore, the shift occurs with the object distance r; Figure 13 Shown is v 最小 = -80 km / h and v 最大 = +280 km / h, and the linear frequency modulation design considered above, that is, B = 800 MHz and T pm = 13.7 microseconds. Consider the target distance r = 200 meters as an example; the relevant frequency range extends to f e =-106MHz, ..., 22.3MHz. Since only the modulus of the received frequency can be determined in a real-valued mixer, but its sign cannot be determined, the f e=-22.3, ..., 22.3 MHz, which corresponds to a relative speed range of v = 156, ..., 280 km / h, so there will be ambiguity - for example, it is impossible to distinguish between relative speeds v = 156 km / h and v = 280 km / h, nor can it be distinguished between v = 186 km / h and v = 250 km / h; for all measured absolute values ​​|f e For receive frequencies > 22.3 MHz, only negative signs can be used, so the relative speed can be determined unambiguously. Figure 13 For the maximum object distance of 249.5 m, almost the entire relevant frequency range f e =-126.2, ..., 2.9 MHz are all negative values, so no ambiguity occurs except for the relative speed range v = 265, ..., 280 km / h.

[0299] To reduce ambiguity in the received frequency domain, a negative modulation bandwidth can be selected, ie the same absolute value as B=-800 MHz mentioned above, which means that the transmitted frequency decreases linearly with time. Figure 14 The corresponding relationship for the receiving frequency is shown; there is no longer any ambiguity in the object distance for r>73.4 meters. However, the absolute value of the maximum receiving frequency is now higher than the positive modulation bandwidth (198.1 MHz instead of 126.2 MHz), so a higher sampling frequency is required, at least around f s =450MHz (instead of f s =300MHz). The range of ambiguity can be further reduced by increasing the absolute value of the modulation bandwidth B; for twice the absolute value of the modulation bandwidth and a negative sign, i.e. B = -1600MHz, according to Figure 15 As shown, ambiguity only occurs when the object distance r is less than 36.7 meters; however, in this case, the sampling frequency needs to be further increased. It is also worth mentioning that if the frequency shift caused by the propagation time is eliminated by multiplying the corresponding rotation factor before the fast Fourier transform as described above, and then decimating accordingly, then a higher sampling frequency (and more samples per pixel, with the data acquisition time remaining unchanged) can keep the fast Fourier transform length used to calculate the two-dimensional correlation unchanged.

[0300] The propagation-time-dependent frequency shift effect of linear frequency modulation reduces the ambiguity in the relative velocity determination of real-valued mixers when determining the distance to close objects. Since tracks are typically set during tracking, the ambiguity is resolved by assigning the detection results generated by a single detection cycle to a track.

[0301] If the lowest possible sampling frequency is used, only a small modulation bandwidth can be used, which can lead to ambiguity in the relative velocity determination over the entire range. This can be resolved by varying the modulation bandwidth during the detection cycle, particularly by alternating its sign. For the receive frequency, with a modulation bandwidth of +B, we have:

[0302] f e，1 =f D +f r =2v / λ-2r / c·B / T pm , (47a)

[0303] In the case of inverting the modulation bandwidth B:

[0304] f e，2 =f D -f r =2v / λ+2r / c·B / T pm , (47b)

[0305] Among them, the frequency shift related to the propagation time

[0306] f r =-2r / c·B / T pm ((11b))

[0307] is known, since the object distance r is determined by means of phase modulation. The Doppler shift f obtained by adding these two equations is D for

[0308] f D =(f e，1 +f e，2 ) / 2. (48)

[0309] However, when using a real-valued mixer, only the absolute value of the two received frequencies |f e，1 | and |f e，2 |. Now let's focus on the frequency shift f caused by the positive propagation time. r It follows how to unambiguously determine the relative velocity v from this. If the Doppler shift f D The modulus value is lower than f r (i.e. |f D | <f r ), then according to formula (47), f e，1 >0, that is, f e，1 =|f e，1 |, and f e，2 <0, that is, f e，2 =-|f e，2 |, for the difference in the modulus of the received frequency, use formula (47):

[0310] |fe，1 |-|f e，2 |=f e，1 -(-f e，2 )=f e，1 +f e，2 =2·f D Applicable to |f D | <f r ,

[0311] For |f D | <f r The absolute value of the difference:

[0312] ||f e，1 |-|f e，2 ||=2·|f D |<2·f r Applicable to |f D | <f r .

[0313] For the Doppler shift f D ≥f r , both receiving frequencies are non-negative, that is, f e，1 =|f e，1 | and f e，2 =|f e，2 |, so using formula (47) we get:

[0314] |f e，1 |-|f e，2 |=f e，1 -f e，2 =2·f r Applicable to f D ≥f r ,

[0315] For the Doppler frequency shift f D ≤-f r , both receiving frequencies are negative, that is, f e，1 =-|f e，1 | and f e，2 =-|f e，2 |, so:

[0316] |f e，1 |-|f e，2 |=-f e，1 -(-f e，2 )=-f e，1 +f e，2 =-2·f r Applicable to f D ≤-f r .

[0317] Therefore, the difference between the measured values ​​of the two received frequencies |fe，1 |-|f e，2 | clearly distinguishes and identifies three situations:

[0318] |f D | <f r (ie -f r <f D <f r ), f D ≥f r and f D ≤-f r :

[0319] ||f e，1 |-|f e，2 ||<2·f r →|f D | <f r ,

[0320] |f e，1 |-|f e，2 |=2·f r →f D ≥f r ,

[0321] |f e，1 |-|f e，2 |=-2·f r →f D ≤-f r (49)

[0322] For each of these three cases, the signs of the two received frequencies are well-defined (see the corresponding sections above), so the Doppler shift f can be determined unambiguously using equation (48) D :

[0323] ||f e，1 |-|f e，2 ||<2·f r →|f D | <f r →f e，1 =|f e，1 |,f e，2 =-|f e，2 |→f D =(|f e，1 |-|f e，2 |) / 2,

[0324] |f e，1 |-|f e，2 |=2·f r →f D ≥f r →f e，1 =|fe，1 |,f e，2 =|f e，2 |→f D =(|f e，1 |+|f e，2 |) / 2,

[0325] |f e，1 |-|f e，2 |=-2·f r →f D ≤-f r →f e，1 =-|f e，1 |,f e，2 =-|f e，2 |→f D =(-|f e，1 |-|f e，2 |) / 2. (50)

[0326] Previously, we considered the frequency shift f caused by the propagation time r The positive sign of is (i.e., the modulation bandwidth B<0 due to Eq. (11b)); similar considerations apply to the negative sign (i.e., the modulation bandwidth B>0):

[0327] ||f e，1 |-|f e，2 ||<2·|f r |→|f D |<|f r |→f e，1 =-|f e，1 |,f e，2 =-|f e，2 |→f D =(-|f e，1 |+|f e，2 |) / 2,

[0328] |f e，1 |-|f e，2 |=2·|f r |→f D ≤-|f r |→f e，1 =-|f e，1 |,f e，2 =-|f e，2 |→f D =(-|f e，1 |-|f e，2 |) / 2,

[0329] |f e，1 |-|f e，2 |=-2·|f r |→f D ≥|fr |→f e，1 =|f e，1 |,f e，2 =|f e，2 |→f D =(|f e，1 |+|f e，2 |) / 2. (51)

[0331] These two relationships can be expressed using the notation V for the modulation bandwidth B. B = ±1To summarize, then, for the radial relative velocity v = λ / 2·f D It can be clearly concluded that:

[0332] For ||f e，1 |-|f e，2 ||<2·|f r |:v=-V B ·λ / 4·(|f e，1 |-|f e，2 |),

[0333] For |f e，1 |-|f e，2 |≥2·|f r |:v=-V B ·λ / 4·(|f e，1 |+|f e，2 |),

[0334] For |f e，1 |-|f e，2 |≤-2·|f r |:v=+V B ·λ / 4·(|f e，1 |+|f e，2 |); (52)

[0335] In this case, calculation errors and slight changes in relative speed and / or distance between two detection cycles must also be taken into account. e，1 |-|f e，2 The value of | will slightly exceed the practical limit ±2·|f r |, so the comparison uses "≥" or "≤" instead of "=". Therefore, the relative speed v can be determined unambiguously by first calculating the difference |f between the two received frequencies e，1 |-|f e，2 |, and with ±2·|f r | compare, and then apply the corresponding formula to the three areas obtained to calculate v. Only when f r= 0, that is, when the distance is zero, the sign of the relative velocity v cannot be determined, because at this time the second and third cases in the above equation (52) cannot be distinguished and are both possible (this ambiguity can also be explained by the fact that the received frequency only consists of the Doppler shift when viewed from the absolute value). Due to the error and change between the two detection cycles, even if f r If the value of f is very small, the Doppler frequency shift sign cannot be determined, and thus the relative speed cannot be determined. However, r Very small values ​​correspond to the area directly in front of the sensor, where normally no objects are present, and even if they are present, the sign of their relative velocity is known from historical records and / or is approximately zero in most cases. Note that to derive Equation (52), the Doppler shift f averaged over two detection periods according to Equation (48) is effectively used. D Of course, the Doppler frequency shift f based on only one detection cycle can also be used D Calculations according to equation (47a) or (47b) are performed. In addition, the frequency shift caused by slightly different distances in the two detection cycles can also be considered; then, different values ​​of f are used in equations (47a) and (47b). r，1 and f r，2 , and add their difference as the smaller component in Equation (48).

[0336] Therefore, the relative velocity ambiguity can be resolved by comparing the received frequencies measured during the two detection cycles. As with conventional tracking, the detection results must be distributed across cycles. However, conventional methods for resolving ambiguity through tracking typically require tracking for significantly more cycles because, in the aforementioned case, the measured distance history—that is, the change in distance measured across cycles—is compared with the expected change in distance based on the corresponding velocity hypothesis. This may require multiple cycles if the received frequency modulus is small, resulting in a small difference between the two velocity hypotheses.

[0337] The method of resolving velocity ambiguity by changing the modulation bandwidth, for example by alternating symbols, can be applied to more than just two detection cycles. As an alternative, a pixel can be acquired with different values ​​or symbols of the modulation bandwidth B within one detection cycle; however, under given hardware conditions, this will halve the number of pixels or require doubling the number of parallel transmit and receive paths. To avoid this, closely spaced pixels with different values ​​or symbols of the modulation bandwidth B can be acquired, especially adjacent pixels, rather than identical pixels; since real objects are usually extended and therefore detected in multiple pixels, and the relative speed and distance are at least approximately the same, the relative speed v can be determined explicitly by applying equation (52) again, where the two received frequencies belong to pixels with two symbols of different modulation bandwidths B. For pixels that are close in position, especially adjacent pixels, the method of setting different B values ​​is: in a system with scanning, different B values ​​are selected for different (especially adjacent) scanning planes.

[0338] Continuous scanning by changing the frequency

[0339] The following describes one possible implementation of continuous scanning in a detection plane. This involves using a device whose radiation direction (for transmission and reception) depends on frequency. Examples of such devices include dispersive materials, lattice structures, or waveguides, with this article focusing on the latter (of course, the method described can also be applied similarly to other devices). Figure 16 A waveguide 16.2 of length 1 cm, implemented in a photonic semiconductor chip 16.1, is shown, with a lateral feed 16.3 of frequency f and equidistant coupling points 16.4 of respective spacing λ0 / 2, where λ0=1550 nm (outdoor wavelength up to f0=c / λ0=195 THz); the coupling points are at Figure 16 The dotted shape indicates that they can also have a certain extension size. The coupling point is used to couple out the wave when transmitting and to couple in the wave when receiving. The waveguide structure is periodic, that is, each phase has the same shape between two adjacent coupling points. Therefore, the phase difference between two adjacent coupling points is It depends on the frequency f used and is a constant, so the phase of the wave transmitted at the k-th coupling point (k=0, ..., K-1, where K=12900) is

[0340] Where, k=0, ..., K-1(53)

[0341] in," mod 2π " indicates that the modulus is 2π, i.e., a symmetric modulus function that maps to the symmetric region -π, ..., +π; the construction of the modulus takes into account that there are usually many wavelengths between the two coupling points, and is also to make the dependence of the phase difference on the frequency as strong as possible, which can be achieved by Figure 16The waveform structure shown in the figure can be realized (as shown in the figure, the waveform structure can extend inward or parallel to the chip surface). is an integer multiple of 2π (i.e., there are integer wavelengths between the coupling points), the radiation direction is perpendicular to the waveguide. situations such as Figure 16 As shown, the radiation direction is tilted, with an angle of γ1(f), which is defined as: the compensation phase difference of the wave path difference Δl between adjacent coupling points Since the phase difference corresponding to the path difference Δl is 2π·Δl / λ, where the free space wavelength λ = c / f, and Δl = sin(γ1(f))·λ0 / 2, where λ0 = c / f0, we can conclude that:

[0342]

[0343] thus

[0344]

[0345] Here, "asin" represents the inverse sine function. If, for example And λ = λ0, we get angle γ1 = 30°. The radiation angle applies to both sending and receiving - this also conforms to the reciprocity principle.

[0346] Continuous spatial scanning is achieved by continuous frequency variation, which can be varied at least approximately linearly even during the acquisition of a single pixel of data. Figure 5 As shown, the frequency change superimposed on the phase modulation is a direct result of the frequency change sweep. As mentioned above, the required sampling frequency f s As the modulation width B, i.e. the frequency change during one pixel, increases, this actually limits the sampling frequency, for the pixel duration T considered here. pm = 13.7 μs and a maximum range of slightly less than 250 m, its absolute value should be |B|≤2 GHz (for f s ≤800MHz). If more than 500 pixels are to be scanned (e.g. horizontally), the total frequency needs to be changed by about 1THz, which corresponds to about 0.5% of the average frequency f0 = c / λ0 = 195THz (at a free-space wavelength of λ0 = 1550nm). For a hypothetical scanning range of -20°, ..., +20°, the phase difference between two adjacent coupling points is The phase difference must be changed by about -0.34π, ..., +0.34π, or about 0.68π, which requires that the phase difference be It has a high frequency-dependent sensitivity, which can be achieved by using a high-dispersion waveguide in the used frequency domain in addition to using a waveform waveguide variation curve.

[0347] First, assume that the radiation angle γ1 changes linearly with time in the range of -20°, ..., +20°, that is,

[0348] γ1(t)=A1·t, (56)

[0349] And the phase difference Has a constant slope over the small frequency range considered:

[0350] Where L is an integer; (57)

[0351] Here, it is considered that at the intermediate frequency f0, the radiation direction should be zero (i.e. ). According to formula (54), we can get:

[0352] π·sin(A1·t)·f / f0=-F1·(f-f0)

[0353] This gives the required time-dependent frequency history f(t):

[0354] f(t)=f0 / (1+π·sin(A1·t) / (f0·F1)) (58)

[0355] For a scan duration of 4.5 ms (500 partially overlapping pixels at 9 ms intervals) and a predetermined frequency change of 1 THz (→ F1 = 2.13 msec (Ts)), Figure 17 The figure shows the time course of the radiation angle γ1(t) and the frequency f(t). Note that there is actually a slight frequency difference between the transmitting and receiving frequencies f(t) (frequency shift caused by the Doppler effect and propagation time). The resulting difference in the radiation angle is much smaller than the radiation width itself.

[0356] According to equation (56), the linear variation of the radiation angle γ1(t) over time means that, with equidistant pixel spacing in time, the angular resolution is constant over the entire detection range. Especially for forward-facing lidar sensors, i.e., those facing the direction of travel, different angular resolutions may be more beneficial: a higher angular resolution is required along the direction of travel (γ1 = 0), while a faster scanning speed can be achieved in the outward direction (γ1 = ±20°; γ1 here refers to the horizontal direction as an example). Figure 18An example of this is shown in . It is worth noting that with faster scanning, the area covered by the lidar radiation on the object also becomes larger, so the signal received during the acquisition duration of a pixel loses coherence (at least partially different points on the object are detected at different times); this reduced coherence slightly reduces the sensitivity and thus the range, but this is not a significant issue in the outer regions where the range requirements are lower. On the other hand, the Doppler accuracy is slightly reduced (because the modulus peak in the two-dimensional correlation of the Doppler dimension is slightly blurred), but this is also not a significant issue since the accuracy of the Doppler (i.e., relative velocity) itself is usually very high.

[0357] Even at a constant sweep speed, according to equation (56), the frequency history according to equation (58) is not completely linear in time (especially due to the included sine function). With a non-constant sweep speed, the nonlinearity of the frequency history in time becomes even more pronounced (see also Figure 18 ). Another reason why the frequency history appears nonlinear in time is that the waveguide has a highly dispersed characteristic (i.e., according to Eq. (57), the phase difference The linear relationship between the frequency and the frequency f no longer holds. Such nonlinearity of the time history of the frequency over a detection scan (i.e. 500 pixels in the above example) can lead to significant nonlinearity of the frequency variation even within a single pixel, i.e. the frequency error f r，Q (t), in particular according to the quadratic form of equation (39b). As mentioned above, such frequency errors can be obtained by rotating the frequency by the factor d n，m In the case of quadratic error, it is only necessary to adjust according to formula (44) Figure 11 Register R2 in the structure shown. By adjusting the value of this register, the modulation bandwidth B that varies with each pixel (due to the varying slope of the frequency history f(t)) can also be taken into account. This way, the output dimensions of the fast Fourier transform always correspond to the same relative speed.

[0358] In principle, a corresponding frequency change profile allows for nearly any scanning profile—a major advantage of scanning with frequency change. The detection range can also be varied within each cycle, particularly depending on the traffic situation; thus, a wider detection range is more valuable at lower vehicle speeds (e.g., in city traffic) than at higher speeds (e.g., on motorways). Problems with incorrect sensor alignment (e.g., a sensor offset 2° to the right instead of facing the direction of travel) can be easily resolved by adjusting the used frequency range accordingly, essentially shifting it slightly.

[0359] In principle, instead of the previously considered continuous scanning, step scanning can also be used, that is, the frequency varies from pixel to pixel, but the frequency remains constant within a pixel (in this case, constant frequency phase modulation in the prior art can also be applied, that is, without superimposing frequency modulation). However, this step scanning itself has some disadvantages (nevertheless, it should not be excluded as a possible implementation form), including:

[0360] The inability to overlap pixels results in a reduction in the data acquisition time per pixel for a given hardware and cycle time. Furthermore, data acquisition time can be further reduced because switching frequencies requires a certain amount of time (especially before the frequency reaches the new value) and then must wait for the maximum propagation time of the received signal (approximately 1.67 microseconds at a maximum distance of 249.5 meters). In the previous design, the considered pixel interval was approximately 9 microseconds, resulting in a data acquisition time of only 6 microseconds, more than halved compared to continuous scanning with a pixel duration of 13.7 microseconds. The main drawback of this significant reduction in data acquisition time per pixel is a reduction in sensitivity (approximately 3.5 dB under the above conditions), resulting in a reduction in range (approximately 19%); the reduction in resolution and accuracy for determining relative velocity is less significant. The penalty is even greater if the pixel pitch is reduced by a factor of, for example, 2, to reduce the number of parallel transmit and receive paths by half (using 16 paths instead of 32).

[0361] - In the case of real-valued mixers, the sign of the relative velocity cannot be determined - The above-mentioned method for resolving the ambiguity by the unknown sign of the received frequency is based on superimposed linear frequency modulation.

[0362] - Objects with very narrow scan directions may be missed if, due to very fast scanning (e.g. outward scanning), the ray width is smaller than the ray movement during the data acquisition pause between two pixels.

[0363] -The implementation of frequency continuous scanning may be simpler and better than the implementation of hierarchical scanning.

[0364] - Frequency-continuous scanning slightly reduces the speckle effect (i.e., statistical fluctuations in the received level from diffuse reflectors) because the speckle effect is frequency-dependent; in addition, spatial scanning also slightly reduces the speckle effect because the illuminated area changes slightly during the data acquisition of one pixel.

[0365] Frequency variation can be achieved, for example, by modulating a variable frequency onto the constant frequency of the laser source (this is difficult due to the large frequency range required), or by using a laser source with directly controllable frequency. Such lasers are typically controlled via mechanical variables (e.g., piezoelectric devices) or electrical variables (voltage, current). The electrical control variable can be generated digitally on a computing unit, such as a processor, converted to the analog range using a digital-to-analog converter (DAC), and optionally low-pass filtered. Since the frequency profile, and therefore the profile required for the control variable, is relatively low-frequency from a signal-theoretic perspective, delta-sigma modulation can be used to determine the input value of the DAC, particularly reducing the requirements on the DAC in terms of resolution. In addition to direct control, a control loop in the form of a phase-locked loop (PLL) can also be used. If the laser source does not have a sufficiently large frequency tuning range (partially to accommodate the frequency changes required for scanning, such as 1 THz, and partly to account for temperature, aging, and tolerances), multiple laser sources can be used and switched between to select the appropriate laser source. Alternatively, the frequency changes required for scanning can be significantly reduced by using waveguides with different phase differences at adjacent coupling points for the parallel transmit and receive paths. Frequency characteristics - at frequency f, the waveguides radiate in different directions, and the small scanning areas they realize are connected to each other (with a certain overlapping area if necessary); this method is further explained later.

[0366] Finally, it is worth mentioning that Figure 16 The waveguide shown not only realizes scanning, but also realizes focusing in the corresponding spatial direction, that is, in each section through the waveguide (to be precise, through the straight line where its coupling point is located), the wave has a flat wavefront portion.

[0367] Scanning in the second spatial direction

[0368] The following explains how to achieve focusing and scanning in the second spatial direction (the spatial direction perpendicular to the waveguide definition). Figure 19 As shown, focusing can be performed with the help of a lens 19.2; Figure 19 The arrangement is shown in the upper part, extending from the waveguide 19.1. The waveguide is in the focal plane of the lens, but is offset from the focal line (as will be the case later, considering multiple waveguides adjacent to each other), so that after being focused by the lens, the radiation constituting the plane wavefront is tilted by a radiation angle γ relative to the optical axis of the lens. 2，1 (exist Figure 19 Above, γ 2，1 The underscore symbol is used γ 2，1, because only the projection angle is indicated there). Since the waveguide itself is already focused in the first spatial direction, the lens only needs to focus in the second spatial direction perpendicular to the first spatial direction in order to have a constant cross section in the direction of the waveguide; Figure 19 As shown below. There, we can also see the two radiation angles γ1 and γ 2，1 are defined in a three-dimensional environment—thus, they are not based on angles in spherical coordinates; for simplicity, Figure 19 The lower part combines the refraction of the lens at both surfaces into one.

[0369] It should be noted that lenses with a constant cross-section in one dimension are easier to manufacture than lenses without this property, in particular focusing lenses for two spatial directions. Instead of a single lens, a lens system can also be used, where the system must also maintain the constant cross-section in the waveguide direction.

[0370] To achieve scanning along the second spatial direction, a material that is transmissive to the used wavelength can be used, whose dielectric constant can be changed by applying a voltage (which generates an electric field in the material) or by applying an electric current (in particular to generate a magnetic field in the material); liquid crystals and ferroelectric materials are examples of materials with such properties. Figure 20 Above (with Figure 19 (same view from above) shows an exemplary arrangement in which the body 20.3 made of this material, similar to the lens 20.2, has a constant cross section (triangular) in the direction of extension of the waveguide 20.1 and is therefore arranged in a prismatic shape; Figure 20 The side view shown below is a 90° rotation, where the voltage U applied to both sides of the prismatic body 20.3 can also be seen. This voltage changes the dielectric constant of the body directly or through the current generated by this voltage and flowing through the body. By varying the applied voltage U, the angle of incidence γ of the prism 20.3 can be changed. 2,1 The difference γ between the exit angle γ2 2,2 , thereby changing the radiation angle γ2 itself, thereby achieving scanning in a second spatial direction orthogonal to the first spatial direction. It is also worth mentioning that the prismatic body is preferably larger than required for the actual radiation path, in order to obtain an electric field that is as uniform as possible in the relevant area, and thus to obtain a spatial dielectric constant that is as constant as possible. In principle, it is also possible to consider forming a common body with a constant cross-section in the waveguide direction by lenses and prisms; however, the change in the dielectric constant cannot remain constant over the entire cross-section, otherwise the focusing characteristics and thus the focal plane will be changed (hence the need for a suitable field distribution that is not constant over the cross-section). Instead of such a prismatic body, a planar transmissive liquid crystal element can also be used, in particular in the shape of a lattice structure, in which the radiation is deflected into the second spatial direction by applying a voltage between the upper side and the lower side.

[0371] As an alternative to focusing and scanning, a liquid crystal array can be used. Figure 21 Shown is a possible solution in which a one-dimensional array of transmissive rod elements 21.3 is provided; Figure 21 The top view shows the array from above, the middle view shows the arrangement and radiation path from the direction of waveguide 21.1, and the bottom view, rotated 90°, shows the arrangement and radiation path from the side. By applying a corresponding voltage to a single rod element (between its top and bottom sides), the phase difference between the incident wave on one side and the outgoing wave on the other side can be varied—the phase difference achieved depends on the applied voltage. This achieves a linear phase curve for the upper wave, simultaneously achieving focusing (i.e., generating a plane wave) and a predetermined radiation angle γ2. By varying the applied voltage, the radiation angle γ2 can be achieved, thereby enabling scanning in a second radiation direction.

[0372] Liquid crystals typically react slowly to control changes, meaning they take a long time to stabilize to a new control state. This is particularly critical when using a two-dimensional liquid crystal array for two-dimensional scanning, as this requires switching (control state) pixel by pixel. In contrast, a one-dimensional liquid crystal array only needs to switch after scanning in the first spatial direction, approximately 4.5 milliseconds in this example, which is not a problem at all. Compared to liquid crystal arrays for two-dimensional scanning, this also has the advantage of requiring much lower control voltages (because there is only one dimension).

[0373] By using an additional lens, which is primarily responsible for focusing in a second spatial direction and again has a constant cross-section in the waveguide direction, the number of required rod-shaped liquid crystal elements and thus the required control voltages is reduced; the liquid crystal array therefore essentially only needs to achieve a deflection of the radiation direction, especially with a small deflection range, the permitted distances being much greater than the free-space wavelength λ0 = 1550 nanometers.

[0374] A disadvantage of the first approach using lenses and controllable prisms is the need for high precision in the lens geometry and its distance from the waveguide (otherwise optical blurring can occur due to, for example, focal plane offsets); this required high precision can increase sensor manufacturing costs and / or component prices due to tighter mechanical tolerances. In contrast, in the second approach based on a liquid crystal array, positional errors can be simply compensated for by controlling the liquid crystal array components accordingly to correct for the corresponding phase errors. This also applies to errors in the position and cross-section of the lenses that may be used.

[0375] Apart from Figure 21As shown in the figure, in addition to the previously considered transmissive liquid crystal array, a reflective liquid crystal array can also be used; for the reflective liquid crystal array, the phase difference between the incident wave and the outgoing wave can also be effectively achieved by applying a corresponding voltage, which can also be regarded as a change in the local outgoing angle relative to the incident angle. The corresponding settings are as follows Figure 22 The top view shows a layout with a waveguide 22.1, a polarization-dependent reflector 22.4 (reflecting one polarization and allowing transmission of radiation perpendicular to that polarization), a one-dimensional reflective array 22.2, and rod elements 22.3, with a 90° polarization rotation from above. The middle view shows the layout and radiation path from the direction of the waveguide 22.1. The bottom view, rotated 90°, shows the layout and radiation path from the side. The reflector 22.4 for deflecting the radiation allows the photonic chip 22.5, which houses the waveguide, and the liquid crystal array 22.2, including its control devices, to be located on the same circuit board 22.6, thereby reducing complexity and manufacturing costs. Of course, a layout without the reflector 22.4 is also possible, i.e., with a direct radiation path between the waveguide and the reflective liquid crystal array, in which case, for example, two circuit boards would be required.

[0376] One or more additional mirrors can be used to deflect the radiation, for example if the optical axis of the sensor is perpendicular to the printed circuit board, for example.

[0377] All other considerations described above for transmissive LC array arrangements also apply to reflective LC array arrangements - in particular, manufacturing tolerances of the relevant optical components, and of their relative arrangements, can be compensated for by corresponding control of the LC array.

[0378] Previously, liquid crystal arrays were considered to be one-dimensional. Due to tolerances (e.g., the thickness or optical properties of the liquid crystal array itself are not constant over the extension of the rod elements), it may not be possible to generate an absolutely flat wave in the first spatial direction (i.e., the direction defined by the waveguide). To avoid this, e.g. Figure 23 As shown, the rod element can be divided into multiple individual elements and phase errors can be compensated by controlling these elements accordingly. However, such a two-dimensional liquid crystal array is more complex and requires more control voltages. Compared to a two-dimensional liquid crystal array for two-dimensional scanning, the number of devices required in this dimension (where the rod element is divided to compensate for errors) is significantly smaller.

[0379] Instead of a reflective liquid crystal array, it is also possible to use reflective liquid crystal elements, in particular one-dimensional grid structures, which deflect the radiation into a second spatial direction by applying a voltage. For this purpose, the radiation must be focused beforehand; Figure 22 Compared to the setup in FIG, a focusing mirror (approximately parabolic), a reflecting lens (i.e. one side with a reflective coating) or a combination of a mirror and a lens is required, wherein these components still have a constant shape in the waveguide direction.

[0380] Finally, it is worth mentioning that for materials whose optical properties can be changed by applying electrical control parameters, in addition to the above-mentioned structure, other schemes can also be used to achieve scanning in the second spatial direction.

[0381] Waveguide setup and scanning mode

[0382] As mentioned several times above, there are, for example, 32 parallel transmit and receive paths, and therefore 32 waveguides; the arrangement and design of these waveguides on the photonic chip will be explained below. In principle, these waveguides can be used to achieve additional parallelism in detection in a first spatial direction (scanning via a frequency-modulated waveguide) and / or in a second spatial direction (scanning via a material with electrically controllable optical properties, such as in the form of a liquid crystal array).

[0383] First consider the way waveguides are often used to achieve parallelism, i.e. parallel detection in the second spatial direction. Figure 24 As shown, 32 waveguides 24.2 are arranged side by side on a photonic chip 24.1 (for ease of illustration, the possible wave paths are not shown); all waveguides have the same design and therefore achieve the same radiation angle γ1 at the same frequency. Depending on the distance between the waveguides, they can directly address adjacent pixels or pixel lines, or more distant pixel lines with other pixel lines between them, for the second spatial direction; at least approximately, the distance between the waveguides is proportional to the distance between the pixel lines they address (if larger angles are ignored). It is first assumed that the distance between the waveguides is such that they address 32 directly adjacent, equidistant pixel lines. Figure 25 shows how to construct a two-dimensional pixel field; the terms "w, f, u" used therein are defined as follows:

[0384] -w represents the number of the corresponding waveguide: w=1, ..., 32,

[0385] -f means scanning by means of the waveguide with continuous frequency variation, i.e. continuous scanning in a first spatial direction and thus continuous scanning of the radiation direction component γ1; this continuous scanning will scan all 500 pixels in this first spatial direction, where f=1, ..., 500 is defined as the number of the respective frequency (more precisely, the respective center frequency, since the frequency varies continuously within a pixel),

[0386] -u refers to the step scanning performed with the aid of a material having electrically controllable optical properties, for example, in the form of a voltage-controlled liquid crystal array, i.e., step scanning is performed in the second spatial direction, thereby scanning the radiation direction component γ2; here, u represents the number of the step scanning (and thus also represents the electrical control parameter that is changed in steps, such as the applied voltage) - after a continuous frequency scan of approximately 4.5 milliseconds in the first spatial direction, a new γ2 scan is performed, i.e., a continuous frequency scan is performed under the new γ2; since each of the 32 waveguides itself realizes a different γ2, only 10 scanning steps u = are required for the 320 pixels in the second spatial direction, i.e., 320 different γ2.

[0387] 1, ..., 10, and the radial direction jumps by 32 pixels from one step to the next.

[0388] This setup requires a smaller waveguide spacing, which may be difficult to achieve, especially when the waveguide is strongly corrugated. As an alternative, the spacing can be increased by a factor of 32, so that the waveguide can achieve a pixel line with a spacing of 32; Figure 26 As shown. When scanning in the second spatial direction, only one pixel is jumped per step; this means that the required scanning range in the second spatial direction is significantly reduced by using materials with electrically controllable optical properties. This approach is beneficial and can expand simpler or more implementation possibilities. This arrangement also allows the above-mentioned method to be used, i.e., by selecting different frequency modulation bandwidths B in adjacent scanning planes, to determine the sign of the received frequency in the case of a real-valued mixer, provided that the object can be seen in at least two adjacent pixels (i.e., adjacent scanning planes, i.e., pixel lines); for this purpose, the alternating direction of the continuous frequency scanning for the first spatial direction must be applied over 10 scanning steps u = 1, ..., 10, thereby achieving an alternating sign of the modulation bandwidth B.

[0389] It was previously assumed that the waveguides were at least approximately equidistant from one another, which would also result in equidistant pixels from one another in the second spatial direction. However, it is often desirable to achieve maximum resolution only in the central angular range, while the resolution in the outward direction can be gradually reduced - at the same time, the gaps between pixels in this range can also be larger. This can be achieved by a non-equidistant arrangement of the waveguides. As a simple case, it is assumed that the distance a between the 16 intermediate waveguides is relatively small and equal (so that there are no gaps between the pixel lines they generate), while the distance between the 8 outer waveguides is doubled, i.e. 2·a, which is larger than the distance between the groups of 16 intermediate waveguides, i.e. 146·a. Figure 27 The resulting two-dimensional pixel field is shown. During scanning in the second spatial direction, each step jumps by 16 times the pixel distance (relative to the pixel distance in the middle range). In the upper and lower thirds of the pixel field, the detection density is only half that of the middle.

[0390] Unlike the previous setup, the waveguides can also be used for parallelism, i.e. for parallel detection in the first spatial direction. For this purpose, the waveguides are designed differently so as to achieve different radiation angles γ1 at the same frequency; the waveguides are still arranged parallel to each other (e.g. Figure 24 As shown), the distance between the connection points remains the same, but the length of the corrugated waveguide between the two coupling points needs to be designed accordingly. Figure 28 The two-dimensional pixel field produced when the waveguide spacing is small and constant is shown; here, the spacing is chosen so that the pixel lines realized are one pixel apart in the second spatial direction. The different designs of the 32 waveguides enable them to realize radiation direction components γ1 (in the first spatial direction) that differ by 16 pixels at the same frequency; in order to detect the entire first spatial direction (now consisting of 16·32=512 pixels), only a continuous frequency scan across 16 pixels is required, so that the frequency numbering covers the range f=1,...,16. The step scan in the second spatial direction (with the help of a material with electrically controllable optical properties) must now completely cover this spatial direction, i.e. all 320 pixels, so 320 steps are required. As Figure 28 As shown, it can be seen that the two-dimensional pixel field is no longer a precise rectangle, but instead forms a parallelogram. By slightly varying the frequency range used to scan the first spatial direction in each of the 320 steps, a slightly rotated, approximately rectangular shape can be achieved. This slight rotation of the rectangle can be compensated for by correspondingly rotating the photonic chip on the circuit board or by rotating the waveguide arrangement of the chip itself.

[0391] The advantage of this approach is that the frequency range required for scanning in the first spatial direction is much smaller (about 1 / 32, which is beneficial for realizing laser sources), or the phase difference of the waveguide is reduced when the frequency range is the same as before. The frequency sensitivity of the image is greatly reduced, which greatly shortens the waveguide length between the two coupling points and reduces the waveguide loss; however, the frequency change of each pixel will be much higher at this time, so continuous scanning will lead to too high a sampling frequency, requiring a step-by-step change of the frequency.

[0392] A disadvantage of this approach is that the increased number of steps leads to slightly higher costs for reconfiguring the control parameters for scanning (for materials with electrically controllable optical properties and frequency modulation of the laser source).

[0393] By continuously scanning in the second spatial direction instead of step-by-step scanning, the above-mentioned disadvantages of increased consumption for reconfiguration of control parameters and excessively high scanning frequency can be solved - that is, the control parameters of materials with electrically controllable optical properties are not changed in steps but continuously. During the continuous scanning of the entire detection range in the second spatial direction, the continuous scanning at the frequency in the first spatial direction only moves one pixel; Figure 29As shown. Therefore, a total of 16 consecutive scans in the second spatial direction are required. With this scheme, the speed of continuous frequency scanning is much slower. Therefore, when using phase difference In the case of simple waveguides with low frequency-dependent sensitivity, the temporal frequency change (i.e., the approximately linear frequency modulation superimposed on the phase modulation) is small enough to allow for a moderate sampling frequency. During continuous scanning in the second spatial direction, the frequency sweep in the first spatial direction can also be stepped, i.e., the frequency is held constant during each sweep in the second spatial direction and then set to a new value during the next sweep in the second spatial direction; thus, 16 different frequencies are used. Instead of a frequency-adjustable laser source, for example, 16 laser sources with different constant frequencies can also be used, with switching between these laser sources being performed.

[0394] The continuous scanning in the second spatial direction can be performed at a non-constant scanning speed, for example, for a lidar sensor detecting in the direction of travel, a higher angular resolution can be achieved in the direction of travel than the edge of the detection range.

[0395] Of course, the parallel detection via 32 waveguides can also be divided into two spatial directions to combine the advantages obtained (of course, the advantages will be weakened accordingly).

[0396] The scheme of using multiple different waveguides for different radiation direction ranges in the first spatial direction can also be implemented in a simpler lidar system with fewer pixels in the following way: in particular, the lidar system has only one transmit and receive path, which can be switched to multiple waveguides; thereby, in particular, reducing the frequency tuning range required for the laser source.

[0397] Alternatives for scanning in a second spatial direction

[0398] Previously, in the scanning in the second spatial direction, it was assumed that a material with electrically controllable optical properties was used. Of course, other methods are also conceivable.

[0399] In the previous example of continuously scanning the second spatial direction (i.e. Figure 29 In the scanning mode shown, for example, a mechanical method (for example using a rotating mirror or prism) can also be used.

[0400] according to Figures 25 to 27A scanning pattern utilizing step-by-step switching in the second spatial direction can be implemented as follows: Instead of just one waveguide, for example, 10 waveguides are provided on each of the 32 parallel transmit and receive paths, with switching between them. This results in a device with 320 parallel waveguides, of which only 32 are used during the frequency sweep in the first spatial direction. It is also worth noting that in this switching method, cascaded single-pole switches (i.e., with two inputs and one output) are typically used, so the inputs to the entire switching matrix are a power of 2, e.g., 8. In the example above, this means 256 waveguides and, consequently, 256 pixels in the second spatial direction. If the pixels are distributed more sparsely outward in the second spatial direction, i.e., the angular spacing is larger, the outward spacing between adjacent waveguides should be increased. If the pixels are distributed more sparsely outward, it is generally (but not exclusively in this embodiment) preferable to increase the radiation width in the corresponding spatial direction to avoid detection gaps. This is often inherent to optical focusing errors, or, in this case, can be achieved by arranging the waveguides outside the focal plane.

[0401] Finally, it is worth mentioning that for all the above solutions, the two mutually perpendicular spatial directions are preferably set in the horizontal and vertical detection directions, that is, roughly corresponding to the azimuth and elevation angles. As mentioned above, the two radiation direction components γ1 and γ2 do not conform to the definition of azimuth and elevation angles in spherical coordinates. In principle, both allocations are possible, that is, both spatial directions can be used for vertical or horizontal detection directions; depending on the detection range and / or resolution, a more beneficial allocation may be found.

[0402] Determining angular error

[0403] The direction of the radiation with the components γ1 and γ2 depends on the control parameters and the hardware characteristics, in particular on:

[0404] - Relationship between the waveguide radiation angle γ1 and the frequency,

[0405] -The relationship between the laser frequency and its control parameters,

[0406] - Relationship between the radiation angle γ2 and the control parameter of the material with electrically controllable optical properties.

[0407] These relationships vary, in particular, depending on the sensor, temperature, and aging. Accurate knowledge of these relationships is essential for correct detection of the environment; otherwise, angular errors (discrepancies between the measured and actual angles) can occur, particularly leading to distortions and offsets in the detection image. While initial determination of sensor-to-sensor differences can be made during production, temperature dependence is often impossible to account for (as the measurement workload in production would otherwise be prohibitive), and aging effects can only be determined over the lifetime of the sensor. Therefore, a method is required to determine angular errors during operation; based on this, angular errors can be compensated by varying the control, i.e., the selection of control parameters.

[0408] In addition to sensor angular errors caused by variations in hardware characteristics, sensors can also experience errors due to misalignment in the vehicle (especially due to mechanical tolerances, vehicle loads, and aging effects). This misalignment has a permanent effect on all detection directions in the relevant spatial direction, while hardware-induced imaging errors can be angle-dependent.

[0409] Now we will explain how to determine the sensor imaging error in the direction of travel. It is well known that the angular error of a radar sensor (especially the error caused by misalignment) can be determined by comparing the measured radial relative velocity of a stationary object with the vehicle's own velocity. The radar sensor measures the radial component of the object's relative motion; a stationary object moves relative to the sensor at a speed equal to the vehicle's own velocity v. 本车速度 (v ego ) moves in a direction parallel to Figure 30 As shown, the radial component v of the motion is measured m It is still necessary to multiply by the cosine of the stationary object's azimuth angle α:

[0410] v m = v ego ·cos(α). (59)

[0411] However, this relationship only holds when the elevation angle is zero; since radar sensors have only a small detection range when the elevation angle is approximately zero, and the above relationship is also a good approximation for small elevation angles, radar sensors usually only use this relationship. If the radar sensor is misaligned, the azimuth angle α measured by the sensor will be m is inconsistent with the actual azimuth angle α, that is, the expected measured velocity v ego ·cos(α m ) and the actual measured speed v egocos(α) does not match. The actual angle α can in principle be determined using equation (59), and the difference from the measured angle represents the misalignment. It should be noted that radar sensors can usually only detect objects within a small azimuth range (at least with high quality), and the cosine at small angles (i.e. close to zero) is very flat, and the vehicle's own speed measured and transmitted to the sensor is usually of poor quality. Therefore, the relative speed v measured for many stationary objects is m (α m ) performs a parabolic regression (small values ​​of the cosine approximate a parabola), and then deviates from the parabola to get the angle of the error.

[0412] Similar to radar systems, coherent lidar systems can directly measure the Doppler effect, i.e., radial relative velocity (both systems operate in a coherent manner and actually differ only in the frequency domain). Therefore, the above method can be applied to lidar systems. For lidar systems, in addition to misalignment, angle-dependent imaging errors caused by changes in hardware characteristics can also be detected, including horizontal and vertical spatial directions; in addition to equation (59), the radiation direction component α in the horizontal spatial direction and the radiation direction component β in the vertical spatial direction are also considered, as follows: Figure 31 shown.

[0413] v m = v ego ·sqrt(1-sin 2 (α) -sin 2 (β)). (60)

[0414] If one of the two angles is zero, then use cosine 2 (γ)+sin 2 (γ) = 1, we get:

[0415] v m =v ego ·cos(α) applies to β=0, (61a)

[0416] v m =v ego ·cos(β) applies to α=0, (61b)

[0417] This also corresponds to the above formula (59). When the absolute values ​​of the angles α and β are small, the following formula is used, and the approximation effect is very good:

[0418] v m =v ego ·cos(α)·cos(β) applies to small |α| and |β|. (61c)

[0419] According to equations (60) and (62c), the measured relative velocity depends on the two angles α and β of the corresponding pixel; therefore, it cannot be determined based on a single pixel without knowing the angular errors in the two spatial directions. However, the angular errors in the two spatial directions are independent of each other (they are determined only by hardware characteristics or control parameters in the corresponding spatial directions); in a pixel field of size 500x320, there are a total of 820 error variables (assuming the extreme case that the corresponding errors of adjacent pixels in the two spatial directions are independent of each other). If each pixel has a stationary object, then there are 500×320 measurement values ​​for determining 820 error variables, which is far more than the required number (820 measurement values ​​are sufficient). In practice, on the one hand, the number of pixels with stationary objects is relatively small, and on the other hand, the errors of adjacent pixels in a spatial direction are not independent, but the distribution of the angular error can be described with sufficient accuracy by an error curve with several parameters (such as a polynomial or a multi-segment broken line or a parabola), so only these few unknown parameters need to be determined; generally, there are enough measurement values ​​for each detection cycle - if necessary, the determination range can be extended to multiple cycles or many cycles, because temperature effects and aging effects are relatively slow in comparison.

[0420] Static reflections / stationary echoes obtained from the road surface at short ranges are very reliable; they cover most or all of the range in the first spatial direction. Since the sensor measures the reflection distance r from the road surface and the installation height h is known sen , so the actual angle β in the vertical space direction is also known, according to Figure 31 As shown, it satisfies the following conditions:

[0421] sin(β) = h sen / r. (62)

[0422] The distribution of the angular error α in the horizontal spatial direction can be determined using equation (60) or (62c), with only one remaining unknown. Given a generally small elevation angle and using equation (62c) to divide by cos(β), the road reflections with different β values ​​(i.e., originating from different distances) can be simply averaged, and the angular error can then be determined from the resulting distribution of α in the horizontal spatial direction. If the angular error in α in the horizontal spatial direction and the actual α value for each pixel are known, the angular error β in the vertical spatial direction can be determined from each pixel (including those that cannot be assigned to the road surface) (since only one unknown remains in equation (60) or (62c)).

[0423] To determine the angular error with the aid of a stationary object, a high-precision vehicle velocity v is required. ego, while the vehicle's own velocity measured and transmitted to the sensor usually cannot guarantee such accuracy. Therefore, the lidar sensor must determine its own velocity itself. In the simplest case, this is achieved by determining the maximum value of the measured radial relative velocity, because according to equation (60), the measured relative velocity can only reach the vehicle's own velocity at most - this is the case in the direction of travel (i.e., α = β = 0°). However, this method assumes that there is a static reflection in the direction of travel or at least close to the direction of travel and that the relative velocity is measured very accurately (i.e., in particular, there is no noise caused by a poor signal-to-noise ratio). As mentioned above, the angular error distribution in each spatial direction can be described with sufficient accuracy by an error distribution with several parameters (e.g., polynomials or multiple broken lines or parabolic segments), so these few unknown parameters only need to be determined from the measured values ​​(i.e., the relative velocity measured for the stationary object); if the own velocity is unknown, this velocity will be another unknown and will be determined at least implicitly.

[0424] If the angular error is known, it can be corrected in the detection list determined by the sensor, thus specifying the actual angle for the detection. Furthermore, the history of the control variables used for scanning can be adapted for subsequent detection cycles (if necessary iteratively) to eliminate the angular error – in particular, to precisely achieve the desired detection range of the 2D pixel field.

[0425] Determine road surface position at longer distances

[0426] As mentioned above, the road surface can be used to determine the angular error; here the reflections from the road surface at close range are used.

[0427] The reflection information from the road surface at a longer distance can reveal the exact position of the road surface at the corresponding distance (i.e., at which pixel the road surface is and at which corresponding vertical angles, which can be further improved to sub-pixel accuracy by interpolation if necessary), which is very helpful for interpreting the detection data. For example, the height of objects on the road can be accurately determined; if not only the reflection point of the measured object is known, but also the exact position of the road surface where the object is located, the height determination will be more accurate. It should be noted that the position of the road surface can only be known in advance at close range based on the known sensor height and the assumption that the road direction is not curved over a short distance; at longer distances, this assumption no longer holds (for example, in a low-lying area, the curved road direction will have a strong impact). In addition, small changes in the vertical direction of the sensor (such as a slight tilt of the vehicle during braking and acceleration) will also have a significant effect. However, lidar sensors are generally unable to detect reflections from road surfaces at long distances because the reflections are too weak; Figure 32The main reason for this is shown in the figure (the vertical direction is not drawn to scale for ease of understanding). The figure shows a radiation with a width of 0.05°, the center of which intersects the flat road at a distance of 100 meters, where the sensor is installed at a height h sen = 60 cm; Since the incident angle on the road is very small, the radiation is projected onto the road over a distance of about 15 meters. This means that in the two-dimensional correlation E m，k In the 2D correlation, the road surface is visible at approximately 30 discrete distances m (not just at m0), so the received power is divided into a corresponding number of values ​​in the 2D correlation. This results in a very poor signal-to-noise ratio, which can hinder detection. The reduction in received power for each distance value m in the 2D correlation can also be explained by the fact that only about 1 / 30 of the radiation width, and therefore the transmitted power, is effective at each distance value.

[0428] To better detect the road surface, we can use the fact that the road surface is visible at multiple pixels with respect to the horizontal spatial direction α, that is, there are multiple pixels where the radiation center intersects the road surface at the same distance; according to equation (62), these pixels have the same radiation angle β, that is, with respect to the vertical spatial direction. A first approach is to accumulate the two-dimensional correlation E across these pixels in an incoherent manner, in particular by using their modulus or power values. m，k , where at least all values ​​of the discrete distance m at which the road may be located are accumulated. The distance m at which the accumulated value significantly exceeds the noise level is the road location; the single two-dimensional correlation E m，k If the noise level in is known (e.g., determined according to the calculation logic in Figures 8 and 10), the noise level after incoherent integration can then be calculated based on theoretical relationships—or estimated directly from the numerical values ​​of the incoherent integration. This requires evaluating a sufficiently large range of m and, if necessary, different values ​​of the discrete frequency k. As mentioned above, the road can be observed over the entire range of discrete distances m, which in the example above covers approximately 30 values; the midpoint of this range, m0, represents the approximate actual distance of the road. Instead of performing incoherent integration only for the values ​​of the discrete propagation time m, this can also be extended to multiple adjacent values ​​of m, since the road itself spans a large number of m.

[0429] It is not necessary to perform the above evaluation for each discrete frequency. In principle, only the frequency k0 containing the road reflection needs to be evaluated. Since the vehicle's own speed is known (with the help of high-quality detection data from the lidar sensor), the discrete frequencies can usually be determined very accurately. Therefore, only one or a few discrete frequencies k need to be correlated. When using the calculation logic shown in Figure 8 or Figure 10, it is necessary to output the two-dimensional correlation E of these discrete frequencies. m，kIf these discrete frequencies can be varied, the output is very complex to implement; therefore, when using the calculation logic shown in Figure 10, the rotation factor d is selected by depending on its own speed. n，m , it is more beneficial to always set the desired discrete frequency at the same output of the calculation logic.

[0430] Instead of the two-dimensional correlation E m，k In order to perform non-coherent integration on the values ​​of , a second solution can be selected: the received pixel sequence e(n) is divided into two half-length sequences, where the first sequence consists of even numbers n and the second sequence consists of odd numbers n. The two-dimensional correlation is determined for each of the two subsequences, and then for each correlation distance m and frequency k, the correlation value is multiplied by the complex conjugate of the other correlation. There is a phase difference between the two subsequences, which is determined by the received frequency f e And the time interval between the two corresponding sampling values, that is, the sampling time T s Decision; the product of the two correlations has a complex value at the road reflection at distance m and frequency k0, whose phase corresponds to the phase difference, wherein this conclusion of course only applies to the signal part and not to the superimposed noise.

[0431] For a particular value of m, the phase is at least approximately constant across all pixels with road reflections, since the relative speed and reception frequency f e In a small angle range around 0°, the road angle can be considered constant (cos(α) in equation (62c) is approximately equal to 1). Therefore, the product can be coherently integrated, i.e., accumulated, between pixels, which can improve the signal-to-noise ratio more effectively than incoherent integration. If integration is also performed for different m, the received frequency f e The frequency component f related to distance r The correction must be performed according to equation (12), preferably as the rotation factor d in the calculation logic shown in FIG10 n，m Using this computational logic, the two-dimensional correlations for odd and even n in the e(n) subsequence also exist inherently—appearing as alternating sequences at the output of the penultimate fast Fourier transform stage. Further considerations and evaluations apply similarly to the first approach for incoherent integration.

[0432] This second approach, consisting of the product of two correlations and summing across pixels, can also be generalized. It is important that the two correlations at least partially use received signals from the same reflection point on the road, i.e., the same area, in order to achieve a defined phase relationship. For example, two correlations can be calculated using the first and second parts of the received sequence for each pixel, or, if they overlap, the correlations of two adjacent pixels. Since the received values ​​belonging to the two correlations are separated in time, they may be sensitive to changes in speed (in this case, the phase difference will vary over the pixel sequence).

[0433] In the case of real-valued mixers, the sign of the received frequency is determined by means of non-binary phase modulation.

[0434] In the binary phase modulation considered above with two phase values ​​0° and 180°, when using a real-valued mixer, the two-dimensional correlation E m，k In the example, two modulus peaks appear at (m0, +k0) and (m0, -k0), so only the modulus of the received frequency can be determined. As mentioned above, superimposed linear frequency modulation can at least partially resolve this ambiguity. The following describes an alternative approach to determining the sign of the received frequency using a real-valued mixer.

[0435] Now, phase modulation It should not consist of two values ​​0° and 180°, but a set of J phase values where j = 0, ..., J-1, and general phase values ​​may be assumed; moreover, the phase modulation varies in an irregular manner, in particular in a pseudo-random manner, over these J phase values. The phase modulation sequence can be described by the following complex value:

[0436]

[0437] Using formula (3a), for the received sequence e(n) of a single object:

[0438]

[0439] Which is consistent with the above derivation, The negative sign of is due to the fact that the complex-valued mixer is assumed to constitute a phase difference between the transmitted and received signals (in other cases, i.e. if the mixer constitutes a phase difference between the transmitted and received signals, the sign is positive). In a real-valued mixer the real part is obtained:

[0440]

[0441] The two-dimensional correlation E in the form of formula (8) m，k By product sequence

[0442]

[0443] The fast Fourier transform of , where b(n) is calculated according to equation (63) and b(nm) is not complex conjugated, which leads to the association corresponding to the first part of e(n) at frequency +k0 in equation (65) (using the complex conjugate of b(nm) would lead to the second part corresponding to frequency +k0). According to equation (65), it can be obtained:

[0444] p(n)=p1(n)+p2(n) (67a)

[0445] in

[0446]

[0447] In the case of discrete propagation times, i.e., object distance m=m0, the first part p1(n) represents the continuously rotating pointer of the object discrete frequency +k0 (because ), so after the fast Fourier transform, that is, in the two-dimensional correlation E m，k In the result, it constitutes the modulus peak at frequency +k0 and discrete propagation time m0.

[0448] The second part p2(n) generates a modulus peak at frequency -k0 and discrete propagation time m0 in the case of binary phase modulation (phase values ​​0° and 180°) because Only two practically identical values ​​can be taken: 0° and 360° (the complex pointers formed by them are both 1). When other phase values ​​are selected, a phase jump will occur in p2(n) because It is assumed that the phase value is not just an integer multiple of 360°. 90°, 180°, 270° are taken as examples (the modulation sequence b(n) is assumed to be 4 values ​​0, -1, ). Then, assuming The values ​​of are +1 and -1 and jump pseudo-randomly between them. Since all 4 phase values ​​are used with the same probability or frequency, The average value of is at least very close to zero; at the discrete frequency -k0 of the object, p2(n) has a two-dimensional correlation E m，k The contribution of has disappeared, so the modulus peak no longer appears. Since there is now only one modulus peak (at the positive frequency +k0), the set receive frequency is clear and correct.

[0449] The disappearance of the modulus peak at the error frequency -k0 can be achieved by any set of J phase values ​​uniformly distributed over a circular period, i.e., 360°.

[0450]

[0451] Therefore, at least J=3 phase values ​​are required 120°, 240°. When other phase values ​​are selected, the error modulus peak at frequency k0 will not be completely eliminated, but will only be reduced; when only J=2 phase values ​​are selected In the example of 90°, it is reduced by 3dB. To identify the correct modulus peak, the larger of the two modulus peaks at ±k0 should be selected; except in the case of extremely poor signal-to-noise ratio, even a small nominal difference in modulus is sufficient for correct identification. In general, when the J phase values ​​used are In order to determine the sign of the received frequency in the case of a real-valued mixer, at least two phase values ​​that are neither in phase nor in phase opposition are required.

[0452] It is usually not possible to achieve arbitrarily precise phase values, especially when the phase values ​​are not limited to inversion. For example, with J = 4 equally spaced phase values, the nominal position considered is 90°, 180°, 270°; its actual value should be 75°, 195°, and 255°, resulting in a significant ±30° error between them. In this case, a ±30° phase jump occurs in the first component p1(n), resulting in phase instability (i.e., phase jitter). This causes the modulus peak at the correct frequency +k0 to decrease by only 0.3dB, resulting in minimal sensitivity loss (the lost energy is transferred to the small noise at other Doppler frequencies k, but this noise level is far lower than the noise level generated by the phase modulation itself in the two-dimensional correlation). At the erroneous frequency -k0, the modulus peak is not completely eliminated. Instead, a small DC component in the p2(n) phase causes the modulus peak to be 11.4dB smaller than the modulus peak at the correct frequency +k0, allowing correct identification. This example shows that even a significant error in the phase value does not pose a serious problem.

[0453] Finally, possible implementations are discussed. One solution is to switch between line segments of different lengths; Figure 33 An example of the case where J=3 phase values ​​is shown, i.e. switching is performed between three line segments of different lengths, wherein phases differing by 120° or 240° are achieved by the different lengths. For the above example where J=4 equidistant phase values, Figure 34 As shown, a combination of a switchable inverter 34.1 and a changeover switch 34.2 can be used between two line sections with a phase difference of 90°.

[0454] Overall system

[0455] According to the above design, the LiDAR system preferably includes only the following three main components:

[0456] -Photonic chip, which includes:

[0457] a frequency-tunable laser source (preferably only one; more than one is needed only if the frequency tuning range is too small),

[0458] a phase modulation unit, which includes a switchable inverter,

[0459] 32 parallel transmit and receive paths, including optical amplifiers, circulators, real-valued summation mixers, and photodiodes,

[0460] 32 waveguides (or 320 waveguides if a switch matrix is ​​used for scanning in the second spatial direction),

[0461] Output end, with analog receiving signal of 32 high frequency MHz band;

[0462] -Digital chip, which includes:

[0463] 32 analog-to-digital converters for receiving signals output by the photonic chip,

[0464] Hard-wired calculation logic for determining two-dimensional dependencies and subsequent evaluation,

[0465] a microcontroller and / or a digital signal processor for further signal evaluation (in particular for determining the detection list) and for calculating the laser frequency and the control parameters of the second spatial direction scanner,

[0466] A control output for the laser frequency and a scanner for the second spatial direction (which can be analog or digital);

[0467] A scanner in one spatial direction (for a second spatial direction) is implemented by:

[0468] Materials with electrically controllable optical properties, in particular liquid crystal elements or liquid crystal arrays, or

[0469] A switch matrix for each of the 32 parallel transmit and receive paths, or

[0470] Mechanical components such as mirrors or prisms that can vibrate or rotate,

[0471] Some solutions also require a lens unit and / or a radiation deflection unit.

[0472] In the optimal case, all electronic components are located on one circuit board; for some configurations, two circuit boards may be necessary. The proposed approach exploits the potential of semiconductor integration, significantly reducing costs and structural size.

[0473] To reduce the required hardware consumption, the data acquisition time per pixel can be halved, which in turn halves the number of parallel transmit and receive paths, thereby halving the size and current consumption of the fixed-wiring computational logic. The general guiding principle in this approach is that all components are active (i.e., used) the entire time (full time), and all generated and radiated power is used for detection (i.e., specifically, power is radiated only in the direction of the simultaneous signal reception).

[0474] Alternative forms of phase modulation

[0475] Previously, Figure 1 In the coherent laser radar system shown in the figure, the phase modulation is considered to be Figure 2 A repeating pseudo-random binary sequence with a period N (i.e., consisting of two phase values ​​0° and 180°) is shown. In addition to pseudo-random sequences, irregular sequences with different definitions can also be used, such as sequences with small autocorrelation sidelobes (e.g., Gold Codes known from the literature). At least when high sensitivity is required, irregular sequences always require a two-dimensional correlation E according to equation (8). m，k This requires a complex determination, which in particular requires special calculation logic, as shown in Figures 8 and 10. If such calculation logic is not available, but only a digital signal processor (especially with a parallel vector calculation unit) is used, then modulation sequences that can be evaluated more simply should be used (but this may also bring other disadvantages).

[0476] As an example, first consider the paper “Phase-Coded-Based Modulation for Coherent Lidar” by Sebastian Banzhaf and Christian Waldschmidt, in IEEETRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 70, NO. 10, OCTOBER 2021, vorgeschlagen80". For this purpose, a binary modulation sequence b(n) of length N is synthesized from two parts: the first subsequence b1(n) has a length N1, during which the phase is constant, that is, for example

[0477] b1(n)=1, where n=0, ..., N1-1, (69a)

[0478] The second subsequence b2(n) of length N2=N-N1 is irregular, for example, pseudo-random:

[0479] b2(n) = ±1, where n = N1, ..., N-1. (69b)

[0480] The lengths N1 and N2 can be the same, i.e.

[0481] N2=N1=N / 2, (69c)

[0482] The modulation sequence can target multiple detection directions, i.e., the pixels are repeated periodically; Figure 35 This type of modulation sequence is shown, in which the sequence synthesized from the subsequences is repeated with a period of N=4096.

[0483] According to formula (3a), for the received subsequence e generated by an object i 1，i (n), for the first constant modulation subsequence b1(n), it satisfies:

[0484] Where n = m 0，i ,......,m 0，i +N / 2-1, (70)

[0485] That is, it only carries the Doppler frequency of the corresponding object, which can be determined by discrete Fourier transform (DFT) or fast Fourier transform. However, due to the unknown time shift m 0，i , and don't know e 1，i (n) The corresponding accurate time position in the received sequence e(n); For simplicity, for example, it can be assumed that the time shift is zero, that is, on the received subsequence

[0486] Where n=0, ..., N / 2-1 (71)

[0487] Fast Fourier Transform

[0488] Where n=0, ..., N / 2-1. (72)

[0489] However, the zero time shift assumption leads to the following: 0，i , especially for long-distance targets, in receiving subsequence The initial m 0，i -1 values, there is no constant modulation subsequence b1(n) value, but the tail value of the modulation subsequence b2(n) of the previous cycle; therefore, the height of the corresponding modulus peak of the fast Fourier transform is reduced, and thus its distance from the noise is reduced, thereby reducing the sensitivity. The peak value of the fast Fourier transform Check on a detection threshold. The frequency k of J modulus peaks above the detection threshold 0，jis used for further processing; these modulus peaks are generally observed between two adjacent FFT values ​​(because they are not at integer Doppler indices k0 as previously considered), so their exact positions can then be determined accordingly by interpolation, i.e. at non-integer frequencies k 0， j. These frequencies k 0，j corresponds at least approximately to the Doppler frequency k of the object 0，i or a subset thereof (for objects with very low reflectivity, this may not result in any modulus peaks above the detection threshold).

[0490] According to equation (3a), the second modulation subsequence b2(n) causes a time shift of about m0.i in the received sequence e(n) and is multiplied by the corresponding Doppler frequency k0.i, i.e., by the modulation component

[0491] Among them, nm 0，i =N / 2, ..., N-1. (73)

[0492] In order to eliminate the corresponding modulation caused by the Doppler frequency, the receiving sequence e(n) is rotated around the frequency k 0.j Reversal Where n = N / 2, ..., N-1 + M-1 and j = 1, ..., J-1; (74)

[0493] Here, the range of received sequences e(n) within which the received second modulated subsequence b2(n) can lie must be considered: n=N / 2, ..., N-1+M-1, where M-1 corresponds to the maximum target distance assumed or of interest. It should be noted that the range of received signals n=N, ..., N-1+M-1 can only be used for continuous scanning, not for switching scanning (in which case the signals are incoherent due to the different detection directions).

[0494] Sequences modified in this way Contains the modulated subsequence a shifted by the corresponding marker j i b2(nm 0，i ); with other Doppler frequencies k 0，i (i.e. k 0，i ≠k 0，j ) is represented as a modulation sequence that is uncorrelated with b2(n) because they can also be expressed as the difference frequency k 0，i -k 0，j Therefore, the sequence Can be related to the second modulation subsequence b2(n):

[0495] Where m = 0, ..., M-1 and j = 1, ..., J-1 (75)

[0496] In these one-dimensional correlations In the figure, the modulus peak appears at position m=m 0，i At the discrete propagation time of the object; at the non-integer discrete propagation time m 0，i In general, if appropriate measures are taken (e.g., the modulated pulse is modified into a triangle shape), the modulus peak value spans two adjacent values ​​m, and its non-integer position can be determined by interpolation. The label j and the Doppler frequency k belonging to j 0，j , the Doppler frequency k of the object can be obtained 0，i =k 0，j . From this, we can see that the correlation above the detection threshold The modulus peak value determines the distance and radial relative velocity of the object in the corresponding detection direction. It should be noted that each phase has a different Doppler frequency k. 0，i (i.e. k 0，i ≠k 0，j ) will only cause noise in the corresponding correlation due to its uncorrelation with b2(n), and no modulus peaks above the detection threshold will appear; when the signal strengths reflected by the objects are roughly the same, this noise is significantly lower than the modulus peaks of interest and therefore will not mask these peaks - only when the signals reflected by the objects differ greatly can an object with a strong reflection signal mask weak reflection signals with different Doppler frequencies through its noise.

[0497] Utilization basis Figure 35 The modulation sequence and the evaluation means explained above, in the general case where there is only one object in a pixel, it is necessary to perform a fast Fourier transform on the reception subsequence e1(n) of the first modulation subsequence b1(n) and on the reception subsequence e1(n) of the second modulation subsequence b2(n) with the frequency inverted corresponding to the second modulation subsequence The correlation between them is Fast Fourier Transformed. This correlation in the time domain can also be achieved by multiplying the Fast Fourier Transform in the frequency domain with a subsequent Inverse Fast Fourier Transform (the Inverse Fast Fourier Transform requires the same amount of computational effort as the Fast Fourier Transform itself); the Fast Fourier Transform of the modulated subsequence b2(n) can be determined a priori once and for all, so it is only necessary to determine the received subsequence Therefore, a total of three fast Fourier transforms are required, and according to formula (8), if the two-dimensional correlation E m，k, a total of M = 250 Fast Fourier Transforms must be calculated, whereby, assuming the same pixel duration, the length of these Fast Fourier Transforms is approximately twice as long. This significantly reduces the computational requirements to an order of magnitude that can be implemented by modern DSPs with parallel vector computing units.

[0498] and Figure 2 The modulation sequence shown and the two-dimensional correlation E of formula (6) or (8) m，k Compared with the analytical evaluation conducted, the following points are particularly disadvantageous:

[0499] - If the same duration of a pixel is assumed, the sensitivity is more than 3 dB lower; the FFT (from the received subsequence e1(n)) is half the length and Half the time correlation length with the modulated subsequence b2(n)) will result in a 3dB loss. In addition, as mentioned above, there are the following effects, especially for long-range targets, the received subsequence The first values ​​of do not originate from the constant modulation subsequence b1(n), but from the tail values ​​of the modulation subsequence b2(n) of the previous period, so these values ​​do not actually contribute to the corresponding modulus peak of the fast Fourier transform.

[0500] Since the length of the Fast Fourier Transform is halved, the resolution and accuracy of the Doppler determination, ie the radial relative velocity determination, deteriorates by a factor of two.

[0501] - for distance determination The temporal correlation with b2(n) is only half the length. Therefore, in the case of multiple objects per pixel, the dynamic range is reduced by 3 dB. That is, the noise level of an object with a strong reflection signal is increased by more than 3 dB compared to an object with a weaker reflection signal. Therefore, the probability of missing such a second object is higher.

[0502] - In a continuous scanning system, pixels can in principle overlap, i.e. part of the received value e(n) can be used for two adjacent pixels, in particular in order to extend the data acquisition time per pixel and thus increase sensitivity. However, Figure 35 The modulation sequences shown can only overlap by 50% because two modulation subsequences are required for each pixel. Often the preferred smaller overlap is not possible.

[0503] Figure 35 These shortcomings of the modulation sequences shown and the related analytical evaluation explained above are prior art, and Figure 36 The new modulation sequence shown can largely eliminate these disadvantages without significantly increasing the computational effort required for the analysis and evaluation. Figure 36 In the modulation sequence b(n) shown, the previous two modulation subsequences b1(n) and b2(n) are nested alternately with each other:

[0504] b1(n)=1, where n=0, 2, 4, ..., N-2, (76a)

[0505] b2(n) = ±1, where n = 1, 3, 5, ..., N-1. (76b)

[0506] As a result, the two modulation subsequences each extend over the entire modulation duration, wherein the two modulation subsequences each only take on every second grid value.

[0507] The received subsequence e generated by object i for the first constant modulation subsequence b1(n) 1，i (n) can be an even value or an odd value, depending on the discrete propagation time m of the corresponding object. 0，i Is it an even number or an odd number (here we first assume that m 0，i is an integer):

[0508] Where n = 0, 2, 4, ..., N-2 or n = 1, 3, 5, ..., N-1, (77)

[0509] In the range marked n, the first value originates from the end of the previous modulation period (due to the propagation time of each object) - Figure 35 Compared to the modulation sequences considered in , these values ​​also originate from the first constant modulation sequence, so they also make a coherent contribution to the subsequent fast Fourier transform. 1，i (n) has two possible positions, so the Fast Fourier Transform used to determine the object's Doppler frequency must now be calculated twice—once by the first sequence,

[0510] Where n = 0, 2, 4, ..., N-2, (78a)

[0511] Another pass through the second sequence calculation

[0512] Where n = 1, 3, 5, ..., N-1. (78b)

[0513] Two Fast Fourier Transforms

[0514] Where n = 0, 2, 4, ..., N-2, (79a)

[0515] Where n = 1, 3, 5, ..., N-1 (79b)

[0516] It relates to the output dimension, i.e., the discrete frequency k = 0,......, N / 2 - 1, and this frequency is related to the discrete frequency k considered previously due to the effective half-scan rate

[0517] k = mod N / 2 (k)(80)

[0518] In Figure 37 , the two fast Fourier transforms according to Equation (79) are shown in modulus form: Two objects have the same received amplitude at (m 0，1 , k 0，1 ) = (300, 3846) and (m 0，2 , k 0，2 ) = (101, 1000); The first object forms a modulus peak at the Doppler frequency 0，1 m = 300 in the first fast Fourier transform k 0，1 = k 0，1 - N / 2 = 1798 The second object forms a modulus peak at the Doppler frequency 0，2 m = 101 in the second fast Fourier transform k 0，2 = k 0，2 = 1000

[0519] Check the detection threshold for the modulus peaks and [[ID= for the two fast Fourier transforms. The frequencies of the modulus peaks J above the detection threshold k 0，j are used for further processing; Usually, modulus peaks can be seen in the values of two adjacent fast Fourier transforms (because as shown in the above example, they are not in an integer Doppler index k 0，j 0，j 0，j k 0，j ), so their corresponding accurate positions, i.e., non-integer frequencies, can be determined by interpolation k 0，j 0，i 0，j These frequencies at least approximately correspond to the Doppler frequency k of the object or its subset (for objects with extremely low reflectivity, they may not cause modulus peaks above the detection threshold). For further processing, it must also be taken into account that k 0，jThe modulus peak at is detected at which of the two fast Fourier transforms, that is, whether the associated discrete propagation time is even or odd; for this purpose, the following variables are introduced:

[0520] m j ∈{0,1} (81)

[0521] If a modulus peak appears in the first fast Fourier transform but m j =0; if a modulus peak appears in the second fast Fourier transform In m j =1.

[0522] According to equation (76b), the second modulated subsequence b2(n) will result in approximately m in the received sequence e(n). 0，i The time shift is multiplied by the corresponding Doppler frequency k 0，i , that is, multiplied by the modulation component

[0523] Among them, nm 0，i =1,3,5,...,N-1. (82)

[0524] To eliminate the corresponding modulation caused by the Doppler frequency, the receiving sequence e(n) changes the frequency k 0. j inversion:

[0525] Where n = 1, 3, 5, ..., N-1 and j = 1, ..., J-1; (83)

[0526] At the same time, the variables introduced in formula (81) m j It is taken into account whether the corresponding received subsequence is in an odd or even grid (i.e. has an even discrete propagation time or an odd discrete propagation time); it is also noted that, as determined in the fast Fourier transform, within the range of mark n, the first received value (due to the propagation time of the corresponding object) originates from the tail of the previous modulation period, but this does not violate the consistency of the subsequent correlation since the entire modulation sequence is periodic (with a period of N).

[0527] Sequences modified in this way The corresponding marker j contains the periodically shifted modulated subsequence a i b2(mod N (nm 0，i )). With other Doppler frequencies k 0，i (Right now k0，i ≠ k 0，j ) represent modulation sequences that are uncorrelated with b2(n) since they are still expressed in terms of Doppler frequencies. k 0，i - k 0，j modulation; the same applies to m j The contribution of is because they are modulated with b1(n). Therefore, the sequence It can be periodically correlated with the second modulation subsequence b2(n):

[0528] in, and j=1, ..., J-1. (84)

[0529] In these one-dimensional correlations The peak value of the mid-mode value appears at The discrete propagation time m of the object at the location 0，i Subtract the displacement applied in equation (83) m j The discrete propagation time is thus obtained as For the above example with two objects, Figure 38 The modulus values ​​of two correlations are shown, one of which is ( k 0，1 =1798 and m 1=0) Another one is ( k 0，2 =1000 and m 2=1) 2, m The modulus peak is at and At the location of their discrete propagation time 1+ m 1=300 and Corresponding.

[0530] According to the correlation of the peak value of the modulus The marker J and the Doppler frequency belonging to the marker j k 0，j The Doppler frequency k of the object can be obtained 0，i , which only has a potential offset of no more than N / 2, see equation (80), where it is considered that the Doppler frequency is determined in a sequence with half the sampling rate, i.e., the additional half-cycle based on the full sampling rate is not recognizable. However, such half-cycles will lead to, in the fast Fourier transform or The complex value of the corresponding modulus peak and the correlation The complex values ​​of are rotated 180° relative to each other, while in other cases (i.e., when there is no extra half cycle in the Doppler frequency), they have the same phase. By this association, it is possible to solve the relative phase difference of each corresponding Doppler frequency k 0，i The ambiguity is corrected by adding two complex values ​​corrected for the possible phase shifts (0° and 180°), i.e. forming the sum and difference of the two complex values ​​- if the modulus of the sum is greater than the difference, then the Doppler frequency k 0，i = k 0，j , in other cases, k 0，i = k 0，j +N / 2; In the above example, for the first object (i.e., for k = k 0，1 and ),Difference Greater than and (Because the two components are approximately in anti-phase), the Doppler frequency is k 0，1 = k 0，1 +N / 2=3846, and for the second object (i.e. k = k 0，2 and Zehe Greater than difference (Because the two components are approximately in phase), the Doppler frequency is k 0，2 = k 0，2 = 1000. Since the Doppler frequency is determined over the duration of the entire modulation period, there is no longer Figure 35 A disadvantage of the modulation sequence shown (according to the prior art) is that the resolution and accuracy of the Doppler determination is reduced by a factor of two due to the fact that the fast Fourier transformation is determined over half the duration of the modulation.

[0531] Using the relationships and steps described above, the correlation above the detection threshold can be The modulus peak value determines the distance and radial relative velocity of the object in the corresponding detection direction. It should be noted that with corresponding other Doppler frequencies k 0，i (Right now k 0，i ≠ k 0，j ) and / or other values m jThe object contribution, due to its uncorrelation with b2(n), only results in noise in the corresponding correlation and therefore does not cause a modulus peak above the detection threshold. When the signal strengths reflected by the objects are roughly the same, this noise is significantly lower than the modulus peak of interest and therefore does not mask the modulus peak. Only when the signals reflected by the objects differ greatly can the noise of the object with a strong reflection mask weaker reflections at different Doppler frequencies.

[0532] Instead of correlations above the detection threshold The scheme for generating detection of modulus peak value can be based on the detection threshold of the fast Fourier transform introduced above. or The sum and difference of k = k 0，j ) and correlation Checking that the two fast Fourier transform components and the correlation are implemented with the correct phase and the calculation of the sum and difference, i.e. the coherent calculation, has a signal-to-noise ratio that is about 3 dB better than the correlation itself; this is because the sum or difference is integrated over the entire modulation sequence b(n), but in the correlation, it is only integrated over half of the modulation sequence. Thus, the optimal two-dimensional correlation E according to equation (6) is m，k The sum or difference also has the same signal-to-noise ratio and therefore the same sensitivity. However, this only applies at the respective positions, i.e. k = k 0，j The case where a modulus peak is already detected in the Fast Fourier Transform at position ; since the integration duration is halved, the signal-to-noise ratio of the Fast Fourier Transform is 3 dB worse, so a correspondingly lower detection threshold must be used. This lowered detection threshold increases the probability of detecting spurious noise peaks; however, these (noise peaks) are again eliminated at the more stringent detection threshold in the sum and difference of the Fast Fourier Transform and the correlation, so the only remaining disadvantage is a slightly increased computational effort (because the correlations must be calculated more frequently). ). Also note that it is not necessary to determine each The sum and difference of The test is performed using a detection threshold that is approximately 3 dB lower, and sums and differences are formed only where this detection threshold is exceeded. Figure 36 The modulation sequence shown and the new method of analysis and evaluation above have the best two-dimensional correlation E according to formula (6) m，k Same sensitivity, so eliminated Figure 35 The disadvantage of the modulation sequence shown (according to the prior art) is that the sensitivity is degraded by about 3 dB. Now we shall briefly discuss whether the new method of sum and difference of fast Fourier transform and correlation can also be applied to the method according to Figure 35 Modulation sequence shown: This is only possible if the received phase remains stable and linear over the entire modulation sequence (ie the instantaneous frequency is constant), but can only be used to a limited extent, especially due to speckle effects.

[0533] Fast Fourier transform and correlation sum and difference methods also avoid, Figure 35 The disadvantage of the modulation sequence shown (prior art) is that the dynamic range is reduced by approximately 3 dB in the presence of multiple objects in a single pixel, since the integration is then carried out over the entire modulation sequence.

[0534] Figure 36 Another advantage of the new modulation sequence shown is that, since each part of the periodic modulation sequence (period N) has the characteristics described according to equation (76), any overlap of two adjacent pixels can be achieved.

[0535] use Figure 36 The new method and use of the modulation sequence shown Figure 35 The difference in the computational effort required by the method for the modulation sequence shown (prior art) is that the former requires an additional fast Fourier transform of length N / 2 (independent of the number of objects in each corresponding pixel), while the correlation is only half the length. However, this does not offer any advantages when implementing the correlation in the frequency range (i.e., via fast Fourier transform and inverse fast Fourier transform). In the general case of only one object in a pixel, two fast Fourier transforms and one correlation must be determined. In principle, additional correlation calculations can also be performed in the above-mentioned method for reducing the fast Fourier transform detection threshold. Consequently, the required computational effort remains within the range achievable by modern digital signal processors with parallel vector computing units and can therefore be implemented in a cost-effective manner without the need for specialized computing logic.

[0536] For the output dimension of the fast Fourier transform, that is, the discrete frequency, as usual, the above (in this part) takes into account k =0, ..., N / 2-1, and the same is true for discrete frequencies after resolving the ambiguity in the asymmetric range k=0, ..., N-1; the actual relative velocity and Doppler frequency can assume two signs, so the upper part, especially the upper half of k=0, ..., N-1, is expressed as a negative value by subtracting N.

[0537] As previously considered Figure 36 As an alternative to the new modulation sequence shown, it is also possible to Figure 39As shown, alternating values ​​(i.e. alternating +1 and -1) are used for the modulated subsequence b1(n). The only significant difference is that the received subsequence for b1(n) has an additional frequency with a period of 2 (related to the modulation rate of b1(n)), i.e. two correlated Fast Fourier Transforms and The modulus peak of the image is offset by half the fast Fourier transform length N / 4, which must be corrected accordingly during further processing.

[0538] Previously in this section we considered the case where the discrete propagation time m 0，i is an integer, the modulation duration T m Equivalent to the sampling repetition time T s For non-integer discrete propagation time m 0，i In the case of an ideal rectangular modulated signal where the received signal also maintains its ideal shape, sampling may occur at the edge where no meaningful information is contained. To avoid this, the sampling repetition time T of the received sequence can be set to s Preset to be smaller, for example, preset to the modulation duration T m and / or modifying the shape of the modulated pulse directly during its generation or in the receiver to, for example, an approximately triangular shape (the latter being achieved by low-pass filtering in the receiver). 0，i In general, there will be modulus peaks in both fast Fourier transforms. and Therefore, for two m j =0,1 to determine the correlation The non-integer m is determined by numerical interpolation of the two correlation modulus peaks. 0，i (non-integer m 0，i It can also be determined by interpolating the values ​​of the two fast Fourier transform modulus peaks or by combining them, that is, by combining the sum or difference of the fast Fourier transform values ​​with the correlation. m Compared with) sampling time T s For example, in the half-method, the alternating nesting of the two modulation subsequences relative to the sampling time has a period of 4, and in the received signal, two consecutive sampling values ​​are derived from the first sequence b1(n), and the next two sampling values ​​are derived from the second sequence b2(n); therefore, on the one hand, for the received subsequence e 1，i (n), four possible positions must be considered, i.e. four Fast Fourier Transforms must be calculated On the other hand, the sample values ​​corresponding to b2(n) are set to zero here (because they are no longer in the grid of period 2 and can therefore be simply omitted), so that the Fast Fourier Transform can be calculated on all the values ​​of the received values ​​(in contrast to half of the values ​​previously). Then, in at least part of the Fast Fourier Transform, three modulus peaks appear for a single object: the maximum value at the correct position, i.e., the Doppler frequency of the object (unlike above, there is no ambiguity, which is an advantage), and two other modulus peaks of approximately 3 dB smaller around a quarter of the length of the Fast Fourier Transform (these should be ignored for further processing). Since four modulus peaks appear in the Fast Fourier Transform Non-integer discrete propagation times may be determined by interpolation (using Fast Fourier Transform values ​​or correlations or a combination thereof).

[0539] In addition to the two previously considered alternating cases of modulation subsequences b1(n) and b2(n), i.e., the alternating nesting with a period of 2, in principle, a longer period can also be used for alternating nesting. Optionally, the number of b1(n) and b2(n) elements in each period can also be unequal.

[0540] Based on the discussion previously in this section Figure 1 In the laser radar system, the mixer is complex-valued, which significantly increases the workload (almost twice) compared to the real-valued mixer in the receiving path. When using a real-valued mixer, only the absolute value of the relative velocity can be determined, but not the sign, because in the fast Fourier transform, + k 0 and - k There are two modulus peaks at the 0 position. To determine the symbol, a method of plausibility check and / or tracking is required, i.e., a method of tracking over multiple acquisition cycles. Alternatively, as described above, a complex-valued modulation sequence can be used, for example, consisting of four values, i.e., 0, -1 and (For 4 equidistant phase values 90°, 180°, 270°)——composed of a sequence. For the first modulation subsequence b1(n), it can be in four values ​​0, -1 and The first modulation subsequence b1(n) with a period of 4 is then alternated with the second modulation subsequence b2(n), which can also be assumed to be four zeros in a pseudo-random sequence. -1 and Although generating such a complex-valued sequence requires increased circuit work (see e.g. Figure 34), but this only needs to be done once, whereas for parallel receivers, the workload of the complex-valued receiver is generated multiple times. For complex-valued modulation sequences, the complex conjugate of the modulation sequence can be used for correlation. Alternatively, the complex conjugate of the received sequence can also be used, as long as it is complex-valued.

[0541] Previously, the case in which the FFT was considered was one in which no window function was used, meaning that the FFT input values ​​were not multiplied by a bell curve. This is only necessary or meaningful when two objects with similar relative velocities but significantly different reflection intensities appear at the same distance from a pixel and need to be separated. In particular, if no window function is used at the FFT input, the sensitivity of the FFT output is reduced (i.e., the ability to detect weakly reflective, distant objects is reduced) if the Doppler signature corresponding to the relative velocity is not an integer, i.e., the modulus peak is spread over two adjacent FFT values. This effect can be minimized by choosing a longer FFT length than the input signal, i.e., by adding zeros to the input signal, a process known as zero padding.

[0542] It should also be noted with respect to the second modulation subsequence b2(n) that it can consist not only of pseudo-random switching between discrete phase values, but also of codes with low autocorrelation sidelobes—for example, the gold code known from the literature. This results in a higher dynamic range for correlation, provided that there are no signal components present as noise in the signal to be correlated (e.g., signals from other objects at relatively high speeds).

[0543] The modulation sequence introduced in this section consists of two subsequences, which are arranged sequentially according to the prior art or nested in the new method. According to the present invention, they can also be used in combination with the linearly varying frequency in equation (9); Figure 40 An example of this is shown. The only effect, in addition to the Doppler shift f D In addition, the receiving frequency f e It also includes a component f related to the propagation time r (See equation (12)); therefore, all the above considerations still apply; the only consideration is that, when determining the relative velocity, a component that depends on the propagation time must also be subtracted from the received frequency, where the propagation time is known from the correlation. Superimposed frequency modulation enables the advantages of the above-described system approach (especially the continuous spatial scanning in frequency) and (especially with regard to determining the received frequency sign in a real-valued mixer). The methods described in the previous sections are also applicable to this type of phase modulation or can be applied to it.

[0544] It should also be noted that for the recognition of road surfaces at long distances, the advantage of a constant or periodic subsequence b1(n) is that it does not distribute the received signal over discrete distances, so that the entire radiation width can be effectively used, and the signal-to-noise ratio is greatly improved, resulting in a high probability of detecting the road surface in a single pixel.

[0545] Functional check of the scan

[0546] If scanning in one or two spatial directions fails, particularly due to hardware failures, high energy densities may occur in certain radiation directions (because the system stops there significantly more frequently than normal), potentially exceeding the permissible limit for eye safety. Therefore, the scanning must be monitored, and the LiDAR sensor must be deactivated if a failure occurs.

[0547] If the sensor no longer scans in one spatial direction, the received signal for all pixels in that spatial direction (assuming the other spatial direction is the same) remains unchanged, excluding system noise and moving objects. Therefore, it is necessary to examine the changes in the received signal in both spatial directions. A first approach involves comparing object reflections, i.e., modulus peaks exceeding the detection threshold. As mentioned above, this requires the exclusion of all moving objects, which can be identified by their measured relative velocity and known intrinsic velocity. A second approach involves using the received signal components at approximately zero distance, caused by internal reflections, coupling, and cover reflections. As mentioned above, these components are already used to compensate for correlation effects (via the correction value c1(n) in the hard-wired calculation logic shown in Figures 8 and 10). This utilizes the fact that these received signal components generally change during scanning.

[0548] Summary

[0549] According to the above application examples, the concepts and design solutions according to the present invention can be easily transferred to general detection and parameter configuration, that is, they can also be used for other values. Therefore, general parameters are often given in the formulas and drawings.

[0550] Compared with the existing technologies, the proposed new scheme is not only novel in combination with other methods, but also novel in itself. Examples of this include:

[0551] - A scheme for determining and implementing correction values ​​to compensate for coupling and reflection effects within the lidar system or its immediate surroundings (in particular the cover plate), even without superimposed frequency modulation.

[0552] -The scheme of frequency scanning in a first spatial direction by a transceiver unit with one or more waveguides and scanning in a second spatial direction by a scanner can also be used in combination with other modulation forms (especially in the case of frequency step changes) and can even be used in incoherent lidar systems.

[0553] The angle error determination scheme (e.g. due to hardware characteristic variations or sensor misalignment) can also be used for other modulation forms, as it is essentially based only on the inherent Doppler measurement capability of the coherent lidar system.

[0554] - The long-range road surface position determination scheme can also be used for other modulation forms (such as pure linear frequency modulation - usually consisting of two frequency ramps with opposite slope signs); the correlation value used in this case is based on the corresponding different correlation calculations (implemented in the form of a fast Fourier transform of each frequency ramp in the case of pure linear frequency modulation).

[0555] A scheme for determining the sign of the received frequency in a real-valued mixer by means of non-binary phase modulation, which can also be used without superimposing frequency modulation.

Claims

1. A coherently operating lidar system for environmental detection, which emits a phase-modulated signal, which signal comprises, in particular, irregular switching at discrete phase values, preferably at only two phase values ​​differing by approximately 180°, wherein: The frequency of the phase modulation varies at least continuously in sections, preferably with an at least approximately linear course, The laser radar system receives a signal reflected from an object, wherein the reflected signal is delayed relative to the transmitted signal due to a propagation time that is related to the distance, and the frequency of the reflected signal is shifted by both the Doppler effect related to the relative velocity and the propagation time that is related to the distance caused by the linear frequency change, and the reflected signal is converted into a low-frequency signal by frequency mixing and digitized into a received sequence. - performing a two-dimensional correlation filtering in a digital signal processing device to determine the variable dimensionality of the signal reflected by the object, That is, time shift and frequency shift, - determining the distance of the object from the determined time shift and subtracting the frequency shift contribution caused by the time shift from the determined frequency shift to determine the radial relative velocity of the object.

2. The laser radar system according to claim 1, characterized in that At least a portion of the two-dimensional correlation filtering is implemented by a fixed-wired digital circuit implemented as a pipeline, wherein for each clock cycle of the digital circuit, multiple output values ​​or all output values ​​are determined in one of the two dimensions, and multiple output values ​​or all output values ​​are determined in the other dimension over a series of clock cycles.

3. The laser radar system according to claim 2, characterized in that The fixed-wire digital circuit has a pre-stage in which a signal sequence or the complex conjugate of the signal sequence is multiplied by a phase modulation sequence or the complex conjugate of the phase modulation sequence at a relatively offset position, the sequence obtained by the multiplication is then decimated if necessary and / or zero-padded if necessary, and then a multi-stage fast Fourier transform is performed, wherein each single calculation operation is implemented in a dedicated circuit, in the first stage the offset between the signal sequence and the modulation sequence is changed with each clock cycle, and the Fourier transform result is generated at the output of the subsequent stage, wherein the result respectively relates to the output data generated by the pre-stage over a plurality of cycles.

4. The laser radar system according to any one of claims 2 or 3, characterized in that Due to their high transmission rate, fixed-wire digital circuits are used for a plurality of pixels, preferably for all pixels, wherein these pixels can be generated in particular by scanning the optical radiation and / or by parallel reception paths.

5. The laser radar system according to any one of claims 2 or 3, characterized in that: A fixed-wire digital circuit has at least one of the following characteristics: - substantially eliminating the coupling and reflection components contained in the digital received signal from the laser radar system or its immediate surroundings, in particular the cover, by adding or subtracting a compensation sequence; Using binary phase modulation, in particular phase modulation consisting of only two phase values ​​differing by approximately 180°, the multiplication with the modulation sequence value being performed by a switchable inverter; - the multiplication of the signal required for the fast Fourier transform by the rotation factor is implemented using a small number of additions and / or subtractions of the shifted signal values, preferably using at most one addition and / or subtraction to implement a real-valued multiplication; The bit length used is varied at each pipeline stage of the fixed-wire digital circuit, the bit length preferably being of such a size that the quantization noise generated in the digital circuit does not significantly contribute to the system noise generated in the analog portion of the receiver; - the Fast Fourier Transform is implemented in the form of a structure with frequency domain decimation, thereby avoiding the need to reorder the input data in the form of long lines and placing the longest lines of the structure and non-trivial multiplications in the preceding stages with lower bit length; - truncation and / or pure bit inversion are used for quantization, the effect of the resulting average error being compensated by adding a correction value in one stage of the digital circuit; The fixed-wire digital circuit is expanded with one or more additional stages for evaluating the results of the two-dimensional correlation, in particular for modulus formation or power formation and subsequent summation and / or maximum value search.

6. The laser radar system according to claim 5, characterized in that The coupled and reflected components of the interior of the lidar system or its immediate surroundings, in particular the cover, contained in the digital received signal are determined by using at least one of the following methods: - determining one or more average values ​​by calculating the product of the received sequence and the modulation sequence at the same or similar points in time, wherein the plurality of average values ​​is formed by observing a plurality of segments of the phase modulation sequence and / or different phase value combinations; - to form the average value of all values ​​of the product of the received sequence and the modulation sequence, using a fixed-wire digital circuit output value corresponding to the zero frequency point of the Fourier transformation in the first stage and the zero offset between the signal sequence and the modulation sequence, preferably pausing the clock cycle of the first stage during the calculation of this value; - forming an average value over a plurality of detection cycles and / or forming an average value over different, preferably mutually adjacent, pixels.

7. The laser radar system according to claim 6, characterized in that The coupling and reflection components contained in the digital received signal from within the lidar system or its immediate surroundings, in particular the cover, are compensated by a sequence that is formed in a fixed-wire digital circuit by correctly assigning the signs of one or more average values ​​determined by the product of the received sequence and the modulation sequence.

8. The laser radar system according to any one of the preceding claims, characterized in that - multiplication of the received sequence or its complex conjugate by a relatively offset phase modulation sequence or its complex conjugate, wherein the respective offset corresponds to the propagation time, - multiplying the resulting product sequence by a complex twiddle factor having an at least approximately linearly varying phase, which twiddle factor at least partially compensates the frequency contribution of the linear frequency modulation for the respective propagation time, wherein a frequency shift is optionally additionally achieved, - subsequently decimating the sequence if necessary, - a subsequent discrete Fourier transformation is performed, which is preferably implemented by means of a fast Fourier transformation, This operation is performed for all propagation times of interest, wherein the sequence of complex rotation factors varies with the corresponding time shift.

9. The laser radar system according to any one of the preceding claims, characterized in that Frequency modulation is not exactly linear. Linearity errors are taken into account in the following way: In two-dimensional correlation filtering, essentially - multiplying the received sequence or its complex conjugate by a relatively offset phase modulation sequence or its complex conjugate, wherein the respective offset corresponds to the propagation time, - multiplying the resulting product sequence by a sequence of complex twiddle factors, wherein the twiddle factors at least approximately correct the non-constant frequency contribution of the nonlinear frequency modulation for the corresponding propagation time, - a subsequent discrete Fourier transformation is performed, which is preferably implemented by means of a fast Fourier transformation, This operation is performed for all propagation times of interest, wherein the sequence of complex rotation factors for correcting frequency modulation linearity errors is varied with the corresponding time shift.

10. The laser radar system according to any one of claims 5 to 9, characterized in that: The multiplication by the complex rotation factors for correcting frequency modulation linearity errors and / or for propagation-time-dependent frequency shifts is also implemented—if necessary in combination—in a fixed-wire digital circuit, preferably by means of a single programmable structure with low precision in the low bit range and therefore low cost.

11. The laser radar system according to any one of the preceding claims, characterized in that Since, in particular for objects at greater distances, only one relative velocity assumption is possible or at least more reasonable based on the known, propagation-time-dependent frequency shift components, the use of real-valued mixers enables a sign-correct determination of the received frequency and thus a unique determination of the relative velocity of the object.

12. The laser radar system according to any one of the preceding claims, characterized in that When using real-valued mixers, the symbol-correct determination of the received frequency, and thus the unique determination of the relative velocity of the object, is achieved by varying the slope of the frequency change, acquiring and evaluating data with different slopes for the same object, and by exploiting the fact that the two frequency-shifting effects, relative velocity and propagation time, have different relationships to one another.

13. The laser radar system according to claim 12, characterized in that The sign of the frequency change slope is changed, rather than the modulus, so that the two frequency shift effects of relative speed and propagation time are superimposed with different signs, thereby enabling the sign of the received frequency to be determined.

14. The laser radar system according to any one of the preceding claims, characterized in that The frequency variation serves to vary the radiation direction, in particular in an at least piecewise continuous manner, in order to be able to detect data from a plurality of pixels in different directions.

15. The laser radar system according to any one of the preceding claims, characterized in that The change in radiation direction by frequency change is achieved in that a waveguide with a plurality of coupling points is used for transmission and reception, which are preferably located in an equidistant grid on a line.

16. The laser radar system according to claim 15, characterized in that The radiation direction change caused by the frequency change is used for a first spatial direction, and another radiation direction change device is provided for a second spatial direction, which is at least essentially perpendicular to the first spatial direction. Preferably, one of the two spatial directions is the horizontal direction and the other is the vertical direction.

17. The laser radar system according to any one of claims 15 or 16, characterized in that The change in the radiation direction is achieved by material transmission or reflection, and at least one optical material property is changed by a control variable, in particular by applying a voltage or current, wherein the control variable and the resulting change in the optical material property can be locally varied.

18. The laser radar system according to claims 16 and 17, characterized in that For changing the radiation direction in the second spatial direction, a body made of a monolithic material is preferably transmitted, which body has a constant, in particular triangular, cross-section in the orientation direction of the waveguide and can, if necessary, be combined with a lens which also has a constant cross-section in this dimension for focusing the radiation in the second spatial direction, the electrical control parameters for changing the optical properties being applied with reference to the orientation direction of the body with constant cross-section.

19. The laser radar system according to claims 16 and 17, characterized in that For changing the radiation direction in the second spatial direction, a planar transmissive or reflective liquid crystal element is used which has a one-dimensional lattice structure and is controlled by voltage.

20. The laser radar system according to claims 16 and 17, characterized in that For changing the radiation direction in the second spatial direction, a transmissive or reflective liquid crystal array is used, which has only a small number of elements, preferably only one rod-shaped element, in the waveguide orientation direction. Preferably, the liquid crystal array is also used for radiation bunching alone for the second spatial direction or for radiation bunching of an auxiliary lens, and preferably, the manufacturing tolerances of the relevant optical components and their relative arrangement are compensated with the help of the liquid crystal array.

21. The laser radar system according to any one of the preceding claims, characterized in that There are multiple transceiver paths working in parallel, each having a waveguide, which is used to change the radiation direction in the first spatial direction by changing the frequency. Preferably, they are powered by a common laser source through subsequent phase modulation. The multiple transceiver paths open up different radiation directions in parallel in at least one of the two spatial directions.

22. The laser radar system according to claim 21, characterized in that The different radiation directions in the second spatial direction are generated by multiple parallel transceiver paths, preferably by multiple parallel transceiver paths having waveguides with the same radiation characteristics and arranged in parallel, and also by electrically controlling the parameters to change the optical properties of the transmissive or reflective material. Through multiple parallel transceiver paths, for the corresponding control of the transmissive or reflective material, the radiation direction either addresses a group of directly adjacent pixels or addresses a group of pixels that are at least partially spaced farther apart, wherein the control of the transmissive or reflective material is gradually changed to cover the entire target radiation direction range, and if necessary, the angular spacing of the pixels increases outward on non-equidistantly arranged side-by-side waveguides, in particular spaced waveguide groups, the spacing within the group increases outward, and preferably the optical radiation width increases outward in the second spatial direction.

23. The laser radar system according to claim 21, characterized in that The waveguides arranged in parallel and adjacent to one another are preferably different from one another in order to open up different radiation direction regions in the first spatial direction by the same frequency change, so that only part of the radiation direction region in the first spatial direction is covered by a single waveguide.

24. The laser radar system according to claim 23, wherein: The change in the radiation direction in the second spatial direction is achieved continuously electronically or mechanically, thereby achieving a slow scan in the first spatial direction by a frequency change, in particular in such a way that during a complete scan in the second spatial direction, the scan is advanced only by approximately one pixel in the first spatial direction by the continuous frequency change.

25. The laser radar system according to any one of the preceding claims, characterized in that There are multiple transceiver paths working in parallel, preferably powered by a common laser source, each of which has a switch matrix for sequential switching within the corresponding waveguide group, wherein preferably the same type of waveguides are used to change the radiation direction in the first spatial direction by changing the frequency, all waveguides are arranged side by side so that they completely open up the second spatial direction, and if necessary, the angular spacing of the pixels is increased outward by the non-equidistant arrangement of the side-by-side waveguides, and preferably the width of the light radiation in the second spatial direction is increased.

26. The laser radar system according to any one of the preceding claims, characterized in that A transmitting and receiving path is provided which includes a switching matrix for sequentially switching between a plurality of waveguides which are preferably parallel and adjacent to each other, wherein the waveguides are different so as to open up different radiation direction areas in the first spatial direction by corresponding frequency changes of the same type, so that a single waveguide only covers part of the radiation direction area in the first spatial direction and reduces the frequency tuning range required for the laser source, and the radiation direction change in the second spatial direction is achieved in a continuous manner by means of electronic or mechanical methods, so that slow scanning is achieved in the first spatial direction by frequency change, in particular, during a complete scan in the second spatial direction, the scan is advanced by only approximately one pixel in the first spatial direction by continuous frequency change.

27. The laser radar system according to any one of the preceding claims, characterized in that The change of radiation direction achieved by frequency change has different speeds. It is characterized in particular by the fact that the radiation direction changes more slowly in the central region than in the outer regions.

28. The laser radar system according to any one of the preceding claims, characterized in that The frequency change for achieving the change in radiation direction is generated by one or more frequency-tunable laser sources. In the case of using multiple laser sources, each laser source only covers a part of the frequency range required for the change in radiation direction.

29. The laser radar system according to any one of the preceding claims, characterized in that When using a real-valued mixer, the correct determination of the received frequency and thus the unique determination of the relative velocity of the object is achieved in the following way: the frequency change used for the change of the radiation direction in the first spatial direction has different, in particular alternating, signs in different radiation directions in the second spatial direction. These different signs cause the two frequency shift effects of relative velocity and propagation time to be superimposed with different signs on different, in particular adjacent, radiation direction planes in which the same object is detected, so that the sign of the received frequency can be determined.

30. The laser radar system according to any one of the preceding claims, characterized in that In particular, deviations in the radiation direction may occur due to misalignment, inaccurately known frequencies and / or inaccurately known assignments of optical properties to control variables, wherein the deviation is determined based on a measured radial relative velocity to a stationary object in order to subsequently take it into account and / or perform a correction.

31. The laser radar system according to claim 30, characterized in that The road surface is used as the stationary object, the angle of which is preferably determined in the vertical direction based on the measured distance and the sensor installation height.

32. The laser radar system according to any one of the preceding claims, characterized in that The device for changing the radiation direction is used to compensate for misalignments and / or to adaptively adjust the detection range, in particular depending on the traffic situation.

33. The laser radar system according to any one of the preceding claims, characterized in that The position of the road surface, in particular at a greater distance, is determined by accumulating the complex-valued product of the correlation powers or the conjugate complex numbers of a first correlation and a second correlation at the same vertical angle for a plurality of pixels belonging to different, in particular adjacent, horizontal angles and / or for different, in particular adjacent distances within a pixel, in order to better distinguish between system noise; for the product of two correlations, the received signal originating at least partially from the same reflection point on the road, in particular by an alternating distribution of the received sequence values.

34. The laser radar system according to any one of the preceding claims, characterized in that When using a real-valued mixer, sign-correct determination of the receiving frequency, and thereby unique determination of the relative velocity of the object, is achieved in the following manner: among the discrete phase values ​​used in an irregular sequence for phase modulation, there are at least two phase values ​​that are neither in phase nor in phase opposition; the correlation between the frequency-shifted receiving sequence and the complex modulation sequence or its complex conjugate is calculated for different receiving frequencies, and the sign of the receiving frequency is identified by utilizing the characteristic that this correlation has different levels at positive receiving frequencies and negative receiving frequencies.

35. The laser radar system according to claim 34, characterized in that A value set of at least three phase values ​​is used that are nominally evenly distributed over a period, but may deviate significantly from the nominal position.

36. The laser radar system according to claim 34 or 35, characterized in that The phase value is achieved by switching between lines of slightly different lengths; or by a combination of switching between lines of slightly different lengths and a switchable inverter.

37. The laser radar system according to any one of the preceding claims, characterized in that In particular, to ensure eye safety, the device for changing the radiation direction is monitored by checking the change in the received signal in the corresponding spatial direction with respect to object reflections or with respect to internal reflections and couplings as well as components of cover reflections.

38. The laser radar system according to any one of the preceding claims, characterized in that The laser radar system includes the following three main components, the electronic components of which are preferably located on the same or at most two circuit boards: - a photonic chip, in particular comprising a frequency-tunable laser source, a phase modulation unit and a waveguide, wherein the phase modulation unit has a switchable inverter and parallel transmit and receive paths; a digital chip, in particular comprising an analog-to-digital converter, a hard-wired computing logic for determining the two-dimensional correlation with subsequent evaluation means, one or more microcontrollers and / or digital signal processors, and means for generating control signals; A scanner for a spatial direction, which is realized by a material having electrically controllable optical properties, in particular a liquid crystal element or a liquid crystal array, or is realized by a switch matrix for each parallel transmit and receive path, or is realized mechanically.

Images