WO2023002242A1

WO2023002242A1 - Interference-resilient lidar waveform and estimation method thereof

Info

Publication number: WO2023002242A1
Application number: PCT/IB2021/056785
Authority: WO
Inventors: Miguel VIDAL DRUMMON; Daniel António MACEDO BASTOS; Arnaldo S.R. OLIVEIRA; Paulo Miguel NEPOMUCENO PEREIRA MONTEIRO; Dionísio ALVES PEREIRA
Original assignee: Bosch Car Multimedia Portugal, S.A.
Priority date: 2021-07-23
Filing date: 2021-07-27
Publication date: 2023-01-26

Abstract

The present application describes a method to process a signal received by a LiDAR sensor, adapted to determine whether the signal has been jammed by an interfering signal, and to correctly estimate the time of flight (ToF). The proposed signal processing method comprises several steps that ensure that the ToF estimation method is accurate, leading to an increased sensor range, while also enabling interference detection, ensuring that interference detection sacrifices range as little as possible.

Description

Interference-resilient Lidar waveform and estimation method thereof

Technical Field

The present application describes a new LiDAR waveform and a method for processing the herein disclosed waveform.

Background art

To the best of the authors ' knowledge, many different waveforms have been proposed for LiDAR sensing, namely Short pulses; Amplitude modulated continuous wave (AMCW) ; Optical code division multiple access (OCDMA) ; and Frequency modulated continuous wave (FMCW) .

Of all the mentioned waveforms, only FMCW is able to address both long range and interference, as it requires a coherent receiver . This means that the range is proportional to the received signal power, P_sig, instead of P² _sig, as is generally the case of incoherent detection. A coherent receiver also inherently attenuates signals that are not coherent with its local oscillator (LO) , which means that incoherent interfering signals are attenuated, whereas the LiDAR' s own transmitted signal is not .

Short pulses are best in maximizing range; however, they are helpless when jammed by other interfering pulses, as the receiver is unable to distinguish signal from interference pulses . AMCW signals are the opposite case : excellent against interfering pulses or even AMCW tones with different frequencies, but fail to maximize range . An OCDMA signal can be regarded as a particular case of an AMCW signal, in which only a few periods are effectively modulated in order to generate the right code . As a result, OCDMA signals shares the same pros and cons as AMCW signals .

Summary

The present invention discloses a solution to the problem of devising a waveform and a corresponding estimation method that maximize range while also enabling interference detection, however guaranteeing that interference detection sacrifices range as little as possible . This can be achieved through an innovative method of processing the received waveforms . The method to estimate the time of flight of an emitted waveform comprises the steps of : acquiring a received signal resulting from the reflection of the emitted waveform on an obj ect; applying aa cross-correlation step to the received signal; applying a find peaks step to the resulting signal of the preceding cross-correlation step; applying a sieve-one step to the resulting signal of the preceding find peaks step; applying a sieve-two step to the resulting signal of the preceding a sieve-one step; applying a crop waveform step to the resulting signal of the preceding sieve-two step; applying a separate edges step to the resulting signal of the preceding crop waveform step; applying a estimate delay step to the resulting signal of the preceding separate edges step; and applying a correction step to the resulting signal of the preceding estimate delay step .

In a proposed embodiment of present invention, the resulting signal of the separate edges step comprises a cropped pulse including a rising edge . Yet in another proposed embodiment of present invention, the resulting signal of the separate edges step comprises a cropped pulse including a falling edge .

Yet in another proposed embodiment of present invention, the estimate delay step determines the delay from the rising edge of the cropped pulse .

Yet in another proposed embodiment of present invention, the estimate delay step determines the delay from the falling edge of the cropped pulse .

Yet in another proposed embodiment of present invention, the correction step corrects the delay from the rising edge of the cropped pulse using a first calibration function, and the correction step corrects the delay from the falling edge of the cropped pulse using a second calibration function.

Yet in another proposed embodiment of present invention, the method comprises applying an averaging step to the resulting signals of the correction step .

Yet in another proposed embodiment of present invention, the emitted waveform comprises a first message pulse, at least one second signature pulse and a predefined relative delay between the first message pulse and the at least one second signature pulse, wherein the pulse power allocated to the first message pulse is greater than the pulse power of the at least one second signature pulse .

Yet in another proposed embodiment of present invention, the sieve-one step comprises discarding peaks whose pulse width is not within a given interval . Yet in another proposed embodiment of present invention, the sieve-two step comprises determining and discarding pairs of peaks that fail to have the predefined relative delay between the first message pulse and the at least one second signature pulse .

Yet in another proposed embodiment of present invention, the cross-correlation step comprises cross-correlating the received signal with an ideal version of a devised waveform.

Yet in another proposed embodiment of present invention, the find peaks step comprises estimating the intensity and position of N peaks in the cross-correlation step resulting signal .

Yet in another proposed embodiment of present invention, the crop waveform step comprises cropping the correct pair of peaks of the first message pulse in the received signal .

Yet in another proposed embodiment of present invention, the sieve-two step does not output any signal, thus classifying the received signal as corrupt .

Yet in another proposed embodiment of present invention, the sieve-two step outputs more than one pair of peaks .

The present invention further describes a computer program, configured to carry out every step of one of the methods described in the above-mentioned possible embodiments .

The present invention further describes a machine-readable storage device, on which the computer program is stored. The present invention further describes a data processing system, comprising the necessary physical means for the execution of the computer program.

The present invention further describes an electronic control unit, configured to carry out every step of one of the methods described in the above-mentioned possible embodiments .

General Description

The present invention is related with incoherent LiDAR sensor receivers, to which short pulses, AMCW and OCDMA waveforms apply.

The main goal of the present invention is to provide the best overall waveform for an incoherent LiDAR sensor, and also the best overall estimation method for an incoherent

LiDAR as well .

The combination of using the best overall waveform with the best overall estimation method means that the resulting LiDAR signal will have near-maximum range as well as the capability of detecting interference .

The present invention distinguishes itself from other approaches particularly due to the use of a devised waveform.

The emitted beam based on the devised waveform comprises a more accurate estimation. The proposed method allows the

LiDAR sensors to operate correctly despite the existence of jamming signals, fending off interfering pulses . Brief description of the drawings

For better understanding of the present application, figures representing p prreeffeerrrreedd embodiments aarree herein attached which, however, are not intended to limit the technique disclosed herein.

Fig. 1 - illustrates the proposed waveform ( 1 ) , also referred to as emitted waveform, which comprises a first short strong pulse (2 ) , herein referred to as "the message", and followed by a second weaker pulse (3) , herein referred to as "the signature" .

Fig. 2 illustrates the proposed devised waveform ( 1 ) estimation method ( 100) and processing steps for the captured/received waveform, where the references are related to :

10 - cross-correlation step;

11 - find peaks step;

12 sieve-one step check whether peaks have the expected pulse width;

13 - sieve-two step - identify pairs of peaks that have correct relative delay;

14 crop waveform step removes all samples except the ones of a single pulse (2 ) ;

15 - separate edges step;

16 estimate delay step estimate delay from the rising edge of the cropped pulse;

17 estimate delay step - estimate delay from the falling edge of the cropped pulse;

18 - correction step - correct the estimated delay from the rising edge of the cropped pulse using a first calibration function; 19 - correction step - correct the estimated delay from the falling edge of the cropped pulse using a second calibration function;

20 - averaging step;

22 - cross-correlation signal;

23 - resulting signal after cross-correlation step ( 10) and find peaks step ( 11 ) ;

24 resulting signal after sieve-one step ( 12 ) and sieve-two step ( 13) ;

25 - resulting signal after crop waveform step ( 14 ) ;

26 resulting signal with a rising edge after the separate edges step ( 15) ;

27 resulting signal with a falling edge after the separate edges step ( 15) ; x - received signal;

Fig. 3 - illustrates an example of a received message pulse

(2 ) , indicating the cropped samples from the rising and falling edges, aass in resulting signal (25) after crop waveform step ( 14 ) .

Description of Embodiments

With reference to the figures, some embodiments are now described in more detail, which are however not intended to limit the scope of the present application.

Based on Figure 1, which illustrates the proposed devised waveform ( 1 ) comprising a message pulse (2 ) and a signature pulse (3) , the time-of-flight (ToF) is estimated from the initial message pulse (2 ) , whereas the interference detection is performed by comparing both pulses, i . e . , the message pulse (2 ) and the signature pulse (3) . Such a task is herein referred to as signature verification, as it resorts to the signature pulse (3) .

Optimizing the devised waveform ( 1 ) boils down to choosing the waveform, relative power, and delay of the signature . A short pulse is preferred mainly for a matter of simplicity, and because it can be detected over a longer range than an

AMCW burst .

The devised waveform ( 1 ) departs from a simple principle but comprises advantageous properties . This waveform ( 1 ) allocates most of the power to the initial short message pulse (2 ) , in order to maximize range, and allocates just enough power to the trailing short signature pulse (3) , to make it ddeetteeccttaabbllee within the entire predefined range, therefore enabling interference detection within the entire range .

The devised estimation method ( 100) of a received signal waveform (x) as shown Figure 2 comprises the following outlined steps .

In the cross-correlation step ( 10) , the received waveform

(x) is cross-correlated with an ideal version of the devised waveform (22 ) . This step procedure is equivalent to a matched filtering. TThhee resulting signal (22 ) has at least two triangular peaks : a strong one at the message (2 ) , and a weaker one at the signature (3) .

Afterwards, a find peaks step ( 11 ) is performed to the resulting cross-correlation signal (22 ) , where the intensity and position of N non-overlapped peaks are estimated in the resulting signal (23) . For regular estimation of an isolated short message pulse (2 ) , where no signature (3) is present, it is sufficient to set N=l . For the devised hybrid waveform

(1 ) one must set N=2 to capture message (2 ) and signature

(3) . If there is the possibility of occurring interference, there is a possibility that an interfering pulse stronger than the signature ((33)) is captured. For this reason, one should set N=3 to capture message (2 ) , interfering pulse and signature (3) . In general, it should be considered that

N≥2+N_int, where N_int is the number of interfering pulses that may be present on the received waveform (x) . As illustrated in Figure 2, the resulting signal (23) results in N=4 peaks were found.

The Sieve-one step ( 12 ) ensures that the right pair of peaks, message pulse (2 ) plus the signature pulse (3) , is found. Accomplishing such a task can be done simply by estimating relative delay between all pairs of pulses and comparing such values with the predefined delay between message (2 ) and signature pulses (3) . However, before doing so one should first check whether each peak has the correct shape , The simplest way of doing so is to check whether peaks have the correct pulse width, being this performed in the Sieve-one step ( 12 ) . Note that discarding individual peaks before analysing pairs of peaks is important ffoorr reducing computational effort .

The Sieve-two step ( 13) is responsible for discarding pairs of peaks that fail to have the predefined relative delay between message ((22 )) and signature (3) pulses . In order to account for noise, the tolerance of Sieve-two step ( 13) can be adjusted by defining upper and lower thresholds . In most cases only one pair of peaks should pass through Sieve-two step ( 13) , which correctly includes the message (2 ) and signature (3) pulses . If not, there is the possibility of either choosing one of the pair of peaks, or of cautiously discarding the present estimation altogether . As illustrated in Figure 2, aass the resulting signal (24 ) comprises four identified peaks, there is a total of six pairs of peaks ,

However, only the pair of peaks comprising the first and third peaks has the correct relative delay, which means that only the correct pair passed through Sieve-two processing step .

The use of the Sieves ( 12, 13) in the devised waveform (x) to perform signature verification and discard interfering short pulses comprises aa completely new and disruptive solution when compared with similar approaches .

In the crop waveform step ( 14 ) , the samples of the earlier peak of the pair of peaks that is output by Sieve-two step, i . e . , the message (2 ) pulse, are cropped.

The crop waveform step ( 14 ) is also of crucial importance for two reasons :

1. CIRCUMVENTING THE NONLINEAR RECEIVER.

In practice, the receiver unavoidably behaves linearly only within a limited range, which means that it will essentially compress aa strong received pulse (x) , enlarging the pulse and flattening its top . Circumventing the nonlinear receiver should focus on not processing the samples of the flattened top of the pulse, cropping them out . This is actually fine, as for LIDAR the most important part of the waveform is the one that contains (most of ) timing information: the rising and falling edges .

2. IMPROVING PRECISION. Using as least samples as possible is very important for enhancing precision. First, only the samples with high signal-to-noise ratio (SNR) should be used, i . e . , the edge samples . Second, and as we will describe further on, the estimate delay steps ( 16, 17 ) , produces the following estimations : and τ_{fall ing}, with an identical expression. a. The total estimation range is given by Ns/ƒ_s, where Ns is the total number of captured samples per received waveform and ƒ_s is the sampling frequency. b. A coarse delay estimation is given by t_rising, as t_rising is a multiple of ƒ_s ^{- 1}. As a result, the remaining need is a precise delay estimation, with an estimation range that needs not be higher than a single sample, i . e . , ƒ_s ^{- 1} c . Such a precise estimation is given by the first term and has an estimation range given by N_rising / ƒ_s, which is always higher or equal to ƒ_s ^{- 1} To make the estimation range as small as possible, thereby enhancing precision, oonnee sshhoouulldd ideally have N_rising =1 - However, such is not possible as the DFT of one sample is the value of such a sample itself, which is useless . In practice, one needs N_rising ≥3.

This again shows why processing only the samples from one edge is advantageous, as it minimizes to the fullest the number of samples considered in fine delay estimation, making it as precise as possible .

Note that processing both edges, and thus the entire pulse, would result in a number of samples given by N_rising +N_top+N_falling, which is much larger than N_rising . N_top is the number of samples between rising and falling edge that belong to the top of the pulse, which must be included in order to guarantee continuity, which again should be discarded for circumventing the nonlinear receiver .

The optimum method thus is processing both edges separately, and then averaging the resulting estimations, as proposed. However, before averaging, the resulting estimations should first be calibrated.

The edge cropping ( 14 ) allows to obtain two estimations from the edges of the message pulse (2 ) using frequency-domain methods (i . e . , DFT) , therefore guaranteeing maximum precision.

In the separate edges step ( 15) , the samples of the rising edge of the pulse (26) are separated and routed to one output, and the samples of the falling edge of the pulse

(27 ) are separated and routed to the other output .

The estimate delay step ( 16) performs the delay estimation from the samples of the rising edge of the cropped pulse

(26) , and the estimate delay step ( 17 ) performs the delay estimation from the samples of the falling edge of the cropped pulse (27 ) as follows . a. The discrete Fourier transform (DFT) of such samples is calculated; b. The phase of the frequency bin with lowest, non-zero frequency is calculated, given by Φ_rising . c. The delay estimated from the rising edge is given by where ƒ_s is the sampling frequency, N_rising is the number of samples of the rising edge (and thus also of the

DFT) , and t_rising is the time instant of the first sample.

The same procedure is performed to the falling edge, resulting in

On the correction step (18) and (19), calibration functions are applied to remove systematic errors. Each estimated delay from (16, 17), obtained from the rising or falling edge, has its own calibration function. The results are τ_rising, corrected fcal, rising ( τ_rising,) , and τ_falling, corrected= ƒ_{cal, falling} ( τ_falling), where ƒ_{cal, rising} is the calibration function applied in the correction step (18) to the estimation obtained with the samples of the rising edge (16) , and ƒ_{cal, falling} is the calibration function applied in the calibration function step (19) to the estimation obtained with the samples of the falling edge (17) . It is important to ensure that each calibration function also considers the average power of the input signal (x) , and also a rough distance estimation, as both parameters enabling achieving a higher accuracy. Correction step ( 18,

19) is very important in avoiding systematic errors as frequency-domain methods such as the DFT are applied to a very small number of samples . Suitable calibration depends on the power of the input signal (x) . A given calibration function should ideally be uusseedd for aa specific target reflectivity, which can be inferred from the rough distance estimation (obtained, for instance, from t_rising) and from the average signal power . However, if such turns out not to be feasible, the resulting estimation error produced from the rising edge partially cancels out the one produced from the falling edge .

Finally, the averaging step (20) ensures that a final estimation is obtained by averaging the τ_{rising , corrected} of the correction step ( 18 ) and the τ_{falling, corrected} of correction step (19) . Such step (20) averages noise out, and partially compensates for incorrect calibration stemming from receiver nonlinearities .

The devised estimation method ( 100) of a received signal (x) comprises sseevveerraall advantages . Properly processing the received waveform (x) is as important as defining the LIDAR waveform itself . The devised estimation method ( 100) has two key aspects :

1. SIGNATURE VERIFICATION.

By combining a set of sieves ( 12, 13) , a pair of pulses is chosen from a set of N pulses, N ≥2. The chosen pair

Claims

of pulses comprises the message (2 ) and the signature(3) pulses of the devised waveform ( 1 ) . As a result, thanks to signature verification, all interfering pulses end up not impairing the estimation process . 2. ESTIMATION. The message pulse (2 ) is cropped from the remaining waveform (24 ) , and samples are carefully chosen from its rising and falling edges are then processed using frequency-domain digital signal processing methods, resulting in a precise distance estimation. For longer distances/ranges, at which the receiver behaves linearly, one should remember that the SNR is proportional to the Psig2. As precision is proportional to the SNR, then the precision obtained from producing and estimate from the signature (3) would be of only x2 of the message (2 ) , where x is the relative power of the signature pulse (3) with regard to the message pulse (2 ) . As a conclusion, for longer distances it may not be worth the effort . However, for shorter distances the maximum SNR is capped by the nonlinear behaviour receiver, namely saturation. As a result, the precision provided by the signature pulse (3) may be the same as the one provided by the message (2 ) , which means that one may be able to halve the estimation error . In summary, if one intends to further reduce the error for shorter distances, a good strategy is to process the signature pulse (3) as well . CLAIMS

1. Method (100) to estimate the time of flight of an emitted waveform (1) comprising the steps of: acquiring a received signal (x) resulting from the reflection of the emitted waveform (1) on an object; applying a cross-correlation step (10) to the received signal (x) ; applying a find peaks step (11) to the resulting signal of the preceding cross-correlation step (10) ; applying a sieve-one step (12) to the resulting signal of the preceding find peaks step (11) ; applying a sieve-two step (13) to the resulting signal of the preceding a sieve-one step (12) ; applying a crop waveform step (14) to the resulting signal of the preceding sieve-two step (13) ; applying a separate edges step (15) to the resulting signal of the preceding crop waveform step (14) ; applying a estimate delay step to the resulting signal of the preceding separate edges step (15) ; and applying a correction step to the resulting signal of the preceding estimate delay step.

2. Method (100) according to the previous claim, wherein the resulting signal of the separate edges step (15) comprises a cropped pulse including a rising edge.

3. Method (100) according to any of the previous claim, wherein the resulting signal of the separate edges step (15) comprises a cropped pulse including a falling edge.

4 . Method ( 100) according to any of the previous claims, wherein the estimate delay step ( 16) determines the delay from the rising edge of the cropped pulse .

5. Method ( 100) according to any of the previous claims, wherein the estimate delay step ( 17) determines the delay from the falling edge of the cropped pulse .

6. Method ( 100) according to any of the previous claims, wherein the correction step ( 18 ) corrects the delay from the rising edge of the cropped pulse using a first calibration function, and the correction step ( 19) corrects the delay from the falling edge of the cropped pulse using a second calibration function.

7 . Method ( 100) according to any of the previous claims, comprising applying an averaging step (20) to the resulting signals of the correction step .

8. Method ( 100) according to any of the previous claims, wherein the emitted waveform ( 1 ) comprises a first message pulse (2 ) , at least one second signature pulse (3) and a predefined relative delay between the first message pulse (2 ) and the at least one second signature pulse (3) , wherein the pulse power allocated to the first message pulse (2 ) is greater than the pulse power of the at least one second signature pulse (3) .

9. Method ( 100) according to any of the previous claims, wherein the sieve-one step ( 12 ) comprises discarding peaks whose pulse width is not within a given interval .

10. Method ( 100) according to any of the previous claims, wherein the sieve-two step ( 13) comprises determining and discarding pairs of peaks that fail to have the predefined relative delay between the first message pulse (2 ) and the at least one second signature pulse (3) .

11. Method ( 100) according to any of the previous claims, wherein the cross-correlation step ( 10) comprises cross- correlating the received signal (x) with an ideal version of a devised waveform (22 ) .

12. Method ( 100) according to any of the previous claims, wherein the find peaks step ( 11 ) comprises estimating the intensity and position of N peaks in the cross-correlation step ( 10) resulting signal .

13. Method ( 100) according to any of the previous claims, wherein the crop waveform step ( 14 ) comprises cropping the correct pair of peaks of the first message pulse (2 ) in the received signal (x) .

14. Method ( 100) according to any of the previous claims, wherein the sieve-two step ( 13) does not output any signal, thus classifying the received signal (x) as corrupt .

15. Method ( 100) according to any of the previous claims, wherein the sieve-two step ( 13) outputs more than one pair of peaks .

16. Computer program, configured to carry out every step of one of the methods described in claims 1 to 15.

17. (Non-transitory) Machine-readable storage device, on which the computer program of claim 16 is stored.

18. Data processing ssyysstteemm,, comprising the necessary physical means for the execution of the computer program of claim 16.

19. Electronic control unit, configured to carry out every step of one of the methods of claims 1 to 15.