WO2023144582A1

WO2023144582A1 - Range estimation method for an amcw lidar

Info

Publication number: WO2023144582A1
Application number: PCT/IB2022/050713
Authority: WO
Inventors: Pedro Nelson SAMPAIO BARBOSA; Miguel Vidal Drummond; Daniel António MACEDO BASTOS; Abel LORENCES RIESGO; Paulo Miguel NEPOMUCENO PEREIRA MONTEIRO
Original assignee: Bosch Car Multimedia Portugal, S.A.
Priority date: 2022-01-25
Filing date: 2022-01-27
Publication date: 2023-08-03

Abstract

The present invention disclosure describes a new method for a light detection and ranging (LiDAR) estimation that is alternative to the mainstream time of flight (ToF) LIDARs in the particular use of autonomous driving vehicles. The present invention describes a method for range estimation of a received AMCW LiDAR signal comprising the steps of: calculating a spectrogram of an input signal x comprising a frequency bin ƒ RF, n ; calculating a ToF estimation, T ToF , from the spectrogram; calculating a phase Φ 1; calculating an AMCW estimation; calculating a final estimation based on the ToF estimation and the AMCW estimation; calculating a final range dest·

Description

Range estimation method for an AMCW LIDAR

Technical Field

The present invention describes a range estimation method for an AMCW LIDAR.

Background art

Amplitude modulated continuous wave (AMCW) and pulsed light detection and ranging (LIDAR) both resort to time-domain signal processing. A pulsed LIDAR emits a short light pulse and estimates range from the time that a reflected version of such a pulse takes to be detected, i.e., the Time of Flight (ToF).

An AMCW LIDAR does not use short pulses, but periodic signals, typically continuous sinusoidal waves. The ToF is estimated from the phase delay introduced by the channel. In more detail, the emitted signal by the LiDAR is given by

where A is the amplitude and ω_RF is the RF frequency. The received signal by the LiDAR is given by

where α_ch is the attenuation introduced by the channel, caused by partial reflection in objects and by free-space path loss, and τ is the ToF, given by

where d is the distance to target and c is the speed of light. The phase of the received signal is given by which results in

As the estimated phase is bound to

distance can be estimated without ambiguity for For instance,

for a maximum range of d = 300 m, this requires an RF frequency of at most 500 kHz.

This basic description of AMCW range estimation presents three important details. First, the signal has a very well- defined spectrum, especially if a pure sinusoidal wave is considered. In ideal conditions, this means that if interfering LIDARs operate with sufficiently different RF frequencies the impact of interference is none. Second, the signal can be purely continuous over time, which means that peak power can be identical to average power. This has a significant impact on laser driver design, as peak current is much more relaxed than for pulsed operation. Third, whereas for a pulsed LIDAR the maximum estimation range can be in theory infinite without any ambiguity, such is not true for an AMCW LIDAR. AMCW operation results in an unambiguous range that is inversely proportional to the RF frequency. Such is not a problem for ideal conditions. However, in practice, there is an error in phase estimation. This results in a trade-off between maximum range and accuracy .

US patent US7791715B1 describes two methods for overcoming the trade-off between maximum range and accuracy. Both methods, illustrated in Figure 1 and Figure 3 of said document, resort to two RF tones (instead of one alone) to achieve both long range and high accuracy. The method illustrated in figure 1 resorts to one low-frequency tone and one high-frequency tones. The low-frequency tone results in long unambiguous range but low accuracy; the high- frequency tone results in high accuracy but in short unambiguous range. In the case of figure 3, an estimation obtained from the high-frequency tone results in up to 5 possible phases. The estimation obtained by the low- frequency tone is sufficiently accurate to rule out 3 of the 5 possible phases, leaving 2 possible phases. From the 2 possible phases, the selected one results in range estimations obtained from low- and high-frequency tones that are as close as possible, i.e., z2 in the left figure. This method can be summarily described as follows: the estimation obtained from low-frequency tone is used to unwrap the estimation obtained from the high-frequency tone.

The described method is presented as prior art in the US patent US7791715B1. The patented method is the one described in figure 3 of the referred document. In the disclosed method, both tones are high-frequency, however with slightly different RF frequencies. As a result, both tones provide the same number of possible phases. From all possible phases, the selected one results in range estimations obtained from both tones that are as close as possible. As both tones are high frequency, high accuracy is guaranteed. The unambiguous range is given by (ƒ₁ - ƒ₂)^-1, where ƒ₁ and ƒ₂ are the tone frequencies; given that these are only slightly different, unambiguous range is large.

While the trade-off between maximum range and accuracy is indeed overcome, there are associated costs: light power must be split between both tones, and both tones have to be individually processed. The former cost is particularly relevant for automotive LIDAR, as light power is capped by eye safety regulations, and long-range specifications force the LIDAR to operate at very low input light powers. A typical AMCW signal is continuous over time, i.e., a laser is continuously intensity modulated by at least one RF tone, and a receiver samples the received light. In fact, such a continuous operation mode was found to be the only followed paradigm in both academia and industry, with a notable exception of US patent application US20170329011A1. Continuous modulation is fine for scenarios in which background light (e.g., sunlight) power can be considered as negligible, as is the case for short-range AMCW LIDARs. In order to observe the impact of background light power, let us consider that the LIDAR projects a light beam on a flat obstacle, located at a given distance from the LIDAR. First, it can be proved that the received background light power is approximately constant regardless of the distance of the obstacle. However, the same does not apply to the signal (laser power): the received signal power is inversely proportional to the square of the distance. This means that, from a certain distance on, more background light power is received than signal power. Such a limitation is very severe for automotive LIDARs, as light power is capped by eye safety regulations, and long distances are usually targeted. A simple method of overcoming such a limitation is to use non- continuous AMCW signals, i.e., the laser only transmits a limited number of tone periods at a time. This enables decoupling signal peak power from average power. For instance, if the laser is off half of the time, in the other half the peak power can be twice as high as the targeted average power. The longer the laser is off, the higher can the peak power be. As a result, during a short time window the receiver captures all the signal light, but only a fraction of the average background light power. In terms of performance, if the laser is on for a given fraction of time (e.g., 50%), then the effective average power is the inverse of such a fraction of time (i.e., 1/50%). The consequence of using an AMCW signal complying with such a method is that, in general, one ends up with a non-continuous multi-tone AMCW LIDAR.

The problem to be solved is therefore how to obtain an accurate range estimation from a received non-continuous multi-tone AMCW signal, herein also referred to as AMCW burst.

Summary

The present invention describes a method for range estimation of a received AMCW LiDAR signal comprising the steps of: calculating a spectrogram of an input signal x comprising a frequency bin ƒ_RF,n; calculating a ToF estimation, τ_ToF, from the spectrogram; calculating a phase Φ₁; calculating an AMCW estimation; calculating a final estimation based on the ToF estimation and the AMCW estimation; calculating a final range d_est .

In a proposed embodiment of present invention, the input signal x is comprised of a tone AMCW burst with a set of N_t samples, each N_t sample of the set comprising only one frequency bin ƒ_RF,n .

Yet in another proposed embodiment of present invention, the AMCW burst comprises M samples, said M samples being smaller than the N samples of the tone frequency of ƒ_RF,n, and a first sample of the burst is sample m which is further represented within the range of {1,2,...N — M + 1}. Yet in another proposed embodiment of present invention, the spectrogram comprises a function of the tone frequency of ƒ_RF,n obtained through wherein

being F(s,ƒ) the discrete Fourier

transform (DFT) of s calculated only for a frequency corresponding toƒ_RF , and

Yet in another proposed embodiment of present invention, the ToF estimation, τ_ToF, comprises a function of

where ƒ_s is the sampling frequency.

Yet in another proposed embodiment of present invention, the phase Φ₁ comprises a function of

Yet in another proposed embodiment of present invention, the AMCW estimation comprises a function of where

ω_RF is the angular frequency of the AMCW tone.

Yet in another proposed embodiment of present invention, the final estimation (14) comprises a function of

Yet in another proposed embodiment of present invention, the final range d_est is determined through the relation between the final estimation (14) and 2-c^-1, wherein the variable "c" represents the light speed.

Yet in another proposed embodiment of present invention, the method for range estimation of a received AMCW LiDAR signal comprises an estimation correction operation over the final estimation resulting in a without systematic errors. Yet in another proposed embodiment of present invention, the estimation correction operation comprises calibration means through the estimation of the range of multiple targets located at known distances from the LIDAR.

Yet in another proposed embodiment of present invention, the spectrogram comprises filtering the input signal x with a narrowband bandpass filter centred at the frequency bin ƒ_RF,n and estimating a power over time of the resulting filtered signal.

Yet in another proposed embodiment of present invention, filtering the input signal x with a narrowband bandpass filter centred at the frequency bin ƒ_RF,n comprises one of: a digital bandpass filter, implemented on either time- domain or frequency-domain, being the resulting filtered signal defined by P_ƒ=ƒRF(k); an analogue bandpass filter, and a sampling stage of the filtered signal, being the resulting filtered signal defined by P_ƒ=ƒRF(k); or an analogue bandpass filter and an edge detector connected to a timer for directly estimating τ_ToF•

The present invention further describes a computer program configured to carry out every step of the described method.

The present invention further describes a (non-transitory) machine-readable storage device, on which the computer program configured to carry every step of the described method is stored. The present invention further describes a data processing system, comprising the necessary physical means for the execution of the computer program configured to carry every step of the described method.

The present invention further describes an electronic control unit, configured to carry out every step of the method herein disclosed.

General Description

The present application describes an accurate range estimation method for received non-continuous multi-tone signals in amplitude modulated continuous wave (AMCW) light detection and ranging (LiDAR) sensors, herein also referred to as AMCW bursts.

A LiDAR range estimation could be obtained from the non- continuous multi-tone AMCW signal using pure AMCW estimation methods, but such methods have two important drawbacks:

1.They do not take advantage of the fact that the AMCW signal is non-continuous (i.e., bursty);

2.They require at least 2 tones, which forces sharing power among tones while at the same time increasing bandwidth .

The second drawback is particularly harmful given the extremely weak power that can be received by an automotive LiDAR system.

The signal to noise ratio (SNR) of the received LiDAR signal being proportional to where

P_sig is the average optical signal input power,

R is the responsivity of the photodetector,

P_BL is the background light power,

I_d is the dark current of the photodetector and

Δƒ is the bandwidth over which the SNR is calculated.

Such a bandwidth thus is at least as high as the spectral width of the input signal.

If N_t tones are used, the power allocated to each tone is at best P_sig/N_t. Assuming that the bandwidth allocated to a single tone is Δƒ, the SNR of a single tone is given by

The SNR is thus reduced with the square of the number of tones, which discourages using more than 1 tone.

The main advantages of the present invention allow to overcome the two above-mentioned drawbacks. A novel range estimation method for a non-continuous multi-tone AMCW signal is disclosed, which:

• Is compatible with any given number of tones, N_t, including 1;

• Takes advantage of the limited duration ("burstiness") of the non-continuous AMCW signal;

• Is simpler than previous known methods, and yet more robust to unwrapping failures, that in turn result in catastrophic errors; • By adding a simple estimation correction stage, the range-dependent systematic error of the provided range estimation can be corrected.

Brief description of the drawings

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

Fig. 1 - illustrates the proposed range estimation method for a single-tone AMCW burst, wherein the references relate to: x - input signal, N samples burst with a ƒ_RF tone frequency;

11 - spectrogram operation;

12 - range estimation τ_ToF given through the ratio between where ƒ_s is the sampling frequency and m

is the first sample of the single-tone AMCW burst;

13 - addition operation between range estimation τ_ToF and the AMCW estimation τ_AMCW, thus provided by

where ω_RF is the angular frequency of the AMCW

tone;

14 - resulting estimated τ_ƒgiven through step 13;

15 - correction operation;

16 - resulting estimation corrected with corrected systematic error;

17 - phase estimation operation;

18 - resulting phase estimation of the input signal x obtained through

19 - AMCW estimation τ_AMCW given through the ratio between where the ω_RF is the angular frequency

of the AMCW tone.

Fig. 2 - illustrates the graphical resulting estimation τ_ƒ, given by τ_ƒ= τ_ToF + τ_AMCW, wherein the reference numbers refer to:

20 - estimated τ_ƒrange in μs;

21 - channel delay in μs;

14 - estimated τ_ƒ;

23 - ideal value.

Fig. 3 - illustrates the graphical estimation error of τ_ƒ, wherein the reference numbers refer to:

30 - estimation error of τ_ƒ;

31 - channel delay in us.

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.

As previously anticipated, the present invention aims to provide a novel range estimation method for a non-continuous multi-tone AMCW signal.

From a simplicity analysis point of view, firstly let's consider only 1 tone, and later generalize to more than 1 tone. The range estimation method for a single tone non- continuous AMCW signal can be described with descriptive support on Figure 1. The input signal (x) is a one tone AMCW burst with a tone frequency of ƒ_RF comprised of N samples. The AMCW burst comprises M samples, where M < N, and the first sample of the burst is sample m, m E {1,2,...N — M + 1}. The method comprises the following steps:

1.calculating the spectrogram (11) of an input signal (x) comprising the frequency bin of the AMCW tone (ƒ_RF ), such power given by

where F(s,ƒ) is the discrete Fourier transform (DFT) of s calculated only for a frequency bin corresponding to ƒ_RF , and k ∈{1,2, — M +

2. will increase as more burst samples are considered, wherein the first burst sample, m, is the one that maximizesP_ƒ=ƒRF(k) , i.e., arg max P_ƒ=ƒRF(k)= m ;

3.The ToF estimation (12) from the spectrogram (11), i.e., the τ_ToF, is defined by the relation where

ƒ_s is the sampling frequency;

4.The phase Φ₁ (18) is estimated (17) as Φ₁ =

5.The AMCW estimation (19) thus is where

ω_RF is the angular frequency of the AMCW tone;

6.The final estimation is given by τ_ƒ= τ_ToF + τ_AMCW (14);

7.The range is estimated as

In case the final resulting estimation τ_ƒ (14) has a systematic error (30), a estimation correction operation (15) can be applied resulting in a τ_ƒ,corrected (16). A better understanding can be achieved by looking at figure 2 and 3. In that case estimated τ_ƒ (14) has a range-dependent systematic error (30). Nonetheless, as the estimated τ_ƒ (14) is an absolutely monotonic function of the ideal value (23), such a systematic error (30) can be corrected, enabling an accurate estimation, in theory capable of zero error. For that, the correction function τ_ƒ,corrected (16) can be obtained by calibration means, for instance by estimating the range of multiple targets located at known distances from the LIDAR.

The precise range estimation of τ_ToF (12) as described in previous steps 1 to 3, and also illustrated in Figure 1, results in discrete estimation values, imposed by the discrete nature of the input signal (x). As a result, even in ideal conditions, the estimated range τ_ToF (12) can have an error of half a sample, i.e.,

By looking to the inset at the right of figure 1, one can see that P_ƒ=ƒRF(k) describes a triangular shape for samples that are near to m. This means that one can refine the range estimation of τ_ToF (12) as follows:

1.Given that the AMCW burst comprises M samples, fit a Such a line has a

positive slope; Such a line has

a negative slope;

3.Find the point at which the first and second lines cross. Such a point has coordinates (m_refined,P_ƒ=ƒRF(max). Note that m_refined is a real number, and not necessarily integer;

4.The refined estimation of τ_ToF is given by τ_ToF = m_refined.

Note that could be lower than 1 (e.g., if m = 1),

and that could be greater than N — M + l. If such is

the case, fitting should (and must) include only available samples. The extreme scenarios are m = 1, for which only the second line can be fitted and m = N — M + 1, for which only the first line segment can be fitted.

Alternatively, and instead of fitting two lines as the previous method suggests, an alternative and simpler method to obtain a more precise estimation of τ_ToF is directed to run a moving average of length L over P_ƒ=ƒRF(k) The position of the maximum of the output signal is then calculated, estMaxPos, and the peak maximum is estimated as m = estMaxPos+ L/2+ 1.

In general, a non-continuous AMCW signal may comprise N_t tones, each with a frequency of If N_t > 1,

the disclosed method can be generalized to a non-continuous multi-tone AMCW signal as follows:

1.N_t spectrograms of the input signal (x), each considering only one frequency bin corresponding to one AMCW tone are calculated, such that P_ƒ=ƒRF(k) = where k ∈ {1,2,...,N — M + 1};

2.All spectrograms are added into a final spectrogram, i.e.,

3.P_ƒ(k) increases as more burst samples are considered. Therefore, the first burst sample, m, is the one that maximizes P_ƒ(k , i.e.,

4.The ToF estimation (12) from the spectrogram (11) thus is τ_ToF = where ƒ_s is the sampling frequency;

5.An AMCW estimation flow taking into account up to N_t tones is obtained from samples x(rn:m + M — 1), resulting in τ_AMCW (19);

6.The final estimation is given by τ_ƒ= τ_ToF + τ_AMCW (14);

7.The range is estimated as

For a single tone, the AMCW estimation flow τ_AMCW (19) can be as previously described; for more than one tone, the AMCW estimation flow τ_AMCW (19) can be based on already known methods such as the ones presented in the prior art (see, e.g., figure 1).

The precise estimation of τ_ToF (12) as described above is also applicable for this general method, however applied to P_ƒ(k) instead of P_ƒ=ƒRF(k).

The spectrogram operation (11) as described in Figure 1 can be regarded as filtering the input signal (x) with a narrowband bandpass filter centered at a given frequency bin and estimating the power over time of the filtered signal.

As a result, instead of doing this with a discrete Fourier transform (DFT), one can use several alternatives, such as:

1.Use a digital bandpass filter for filtering the input signal (x), implemented on either time-domain or frequency-domain; the resulting filtered signal being defined by P_ƒ=ƒRF(k);

2.Use an analogue bandpass filter instead, and sample the filtered signal; the sampled signal being defined by P_ƒ=ƒRF(k);

3.Use an analogue bandpass filter and an edge detector connected to a timer for directly estimating τ_ToF.

The quality of the obtained range estimationτ_ToF (12) depends on multiple factors, namely noise and interference. In an ideal scenario, without noise nor interference, the peak of the spectrogram, given by is much higher than

the minimum or the mean value of A simple quality

metric of the obtained estimation thus is the peak-to-average power ratio (PAPR) of P_ƒ=ƒRF(k). For instance, for PAPR values below a predetermined threshold, the obtained estimation (12) can be classified as unreliable, and thus should be discarded.

Please note that analytics other than PAPR can be applied as well. In general, the behaviour of P_ƒ=ƒRF(k) should be carefully analysed to check whether an appropriate symmetric triangular peak is present and whether multiple peaks are present (which indicates in-band AMCW interference). Spectrogram-based analyses can also be made to check for out-of-band AMCW interference and for pulsed interferers.

Claims

1. Method for range estimation of a received AMCW LiDAR signal comprising the steps of: calculating a spectrogram (11) of an input signal (x) comprising a frequency bin (ƒ_RF,n); calculating a ToF estimation (12), τ_ToF, from the spectrogram (11); calculating a phase Φ₁ (18); calculating an AMCW estimation (19); calculating a final estimation (14) based on the ToF estimation (12) and the AMCW estimation (19); calculating a final range d_est.

2. Method for range estimation of a received AMCW LiDAR signal according to the previous claim, wherein the input signal (x) is comprised of a tone AMCW burst with a set of N_t samples, each N_t sample of the set comprising only one frequency bin (ƒ_RF,n)•

3. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, wherein the AMCW burst comprises M samples, said M samples being smaller than the N samples of the tone frequency of ƒ_RF,n, and a first sample of the burst is sample m which is further represented within the range of {1,2,...N — M + 1}.

4. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, wherein the spectrogram (11) comprises a function of the tone frequency of ƒ_RF,n obtained through wherein

being F(s,ƒ) the discrete Fourier

transform (DFT) of s calculated only for a frequency corresponding toƒ_RF , and k E {1,2, N — M + 1}.

5. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, wherein the ToF estimation (12), τ_ToF, comprises a function of τ_ToF = m - where ƒ_s is the sampling frequency.

6. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, wherein the phase Φ₁ (18) comprises a function of

7. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, wherein the AMCW estimation (19) comprises a function of

where ω_RF is the angular frequency of the AMCW tone.

8. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, wherein the final estimation (14) comprises a function of τ_ƒ= τ_ToF + TAMCW •

9. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, wherein the final range d_est is determined through the relation between the final estimation (14) and 2c^-1.

10. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, comprising an estimation correction operation (15) over the final estimation (14) resulting in a τ_ƒ,corrected (16) without systematic errors.

11. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, wherein the estimation correction operation (15) comprises calibration means through the estimation of the range of multiple targets located at known distances from the LIDAR.

12. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, wherein the spectrogram (11) comprises filtering the input signal (x) with a narrowband bandpass filter centred at the frequency bin (ƒ_RF,n) and estimating a power over time of the resulting filtered signal.

13. Method for range estimation of a received AMCW LiDAR signal according to any of the previous claims, wherein filtering the input signal (x) with a narrowband bandpass filter centred at the frequency bin (ƒ_RF,n) comprises one of: a digital bandpass filter, implemented on either time- domain or frequency-domain, being the resulting filtered signal defined byP_ƒ=ƒRF(k) ; an analogue bandpass filter, and a sampling stage of the filtered signal, being the resulting filtered signal defined by P_ƒ=ƒRF(k); or an analogue bandpass filter and an edge detector connected to a timer for directly estimating τ_ToF .

14. Computer program configured to carry out every step of one of the methods described in claims 1 to 13.

15. (Non-transitory) Machine-readable storage device in which the computer program of claim 14 is stored.

16. Data processing system comprising the necessary physical means for the execution of the computer program of claim 14.

17. Electronic control unit, configured to carry out every step of one of the methods of claims 1 to 13.