Comparison of FMCW-LiDAR system with optical- and electricaldomain swept light sources toward self-driving mobility application

Transcription

1 P1 Napat J.Jitcharoenchai Comparison of FMCW-LiDAR system with optical- and electricaldomain swept light sources toward self-driving mobility application Napat J.Jitcharoenchai, Nobuhiko Nishiyama, Tomohiro Amemiya, and Shigehisa Arai Dept. of Electrical and Electronic Engineering, Tokyo Institute of Technology Abstract: The suitable configuration of FMCW-LiDAR to realize compact LiDAR system on Si-photonics platform for self-driving mobility application is investigated. Especially, signal comparison with two types of wavelength swept light sources is discussed. One is an optical-domain thermally wavelength-tuned laser by sweeping injection current. The other is an electrical-domain wavelength-tuned source by the combination of CW laser with an external modulator which superimposes electrical frequency shift. Signals and 3- dimentional images were taken and compared using two light sources with fiber-based coherent detection setup. The results revealed similar signal-to-noise ratio and images for two configurations if applying k-clock resampling interferometer. Keywords: FMCW-LiDAR, Swept light source, Self-driving application 1. Introduction Self-mobility applications got high attentions and developments of technologies for such applications are going in worldwide. Light detection and ranging (LiDAR) sensor is one of the crucial components for these technology as an environment mapping and sensing device. The commercial grade LiDAR usually comes with the bulky size and consisted of many parts of mechanical movement results in high physical disturbance and high production cost. To find an alternative to a bulky LiDAR system, the compact size LiDAR system was expected to be developed [1-3]. We are also proposing to realize the one-chip size Frequency-modulated Continuous-wave (FMCW) LiDAR with the slow light and Si-photonics technology with hybrid III-V laser integration to minimize the unit size and making more stable light sweeping while lower the fabrication cost [4-6]. The sweeping FMCW system employ the different in frequency or the beat frequency of transmitted and received light signal to measure the distance value. Thus, the system required a linear relationship of time and frequency in order to recover back the correct and accurate distance value. To realize high performance FMCW-LiDAR, wavelength swept light source is a key device. Although the required optical power level of FMCW-LiDAR is much lower than that for time of flight (ToF)-LiDAR, narrow line width, low-noise and linear wavelength sweeping by time are required. To tune the wavelength, several methods can be considered. One of the method is direct thermal tuning of lasers [7]. Another method is using an external modulator to superimpose swept frequency. In this paper, we compared acquired images from FMCW- LiDAR setup using two types of swept light sources. 2. System setup in Optical-Domain Method In order to design the system, we have to verify the performance with the large components setup firsts before making one-chip LiDAR. Therefore, all experiments in this paper used optical fiber based system. The scanning of light was realized by Galvano scanner. Figure 1. shows the setup with thermal modulated swept laser diode (we called optical-domain light source). To compensate k-clock linearity of the wavelength sweeping, a Mach-Zehnder interferometer (MZI) was installed. Figure 2. shows the time average spectrum of lasers in the setup. We used distributed feedback (DFB) lasers with total output power of 10 dbm. Injection current was modulated with the repetition rate of 100 khz. By changing the injection current, refractive index was changed, resulting wavelength change. Tuning bandwidth was 0.03 nm (3.7 GHz). However, in this thermal tuning, the relationship between the time and frequency of the sweeping is CLRC 2018, June

2 distorted. This is the reason for the k-clock MZI in the system. It works as a calibration system to recover the non-linear relationship back to the linear one.. Figure 1. LiDAR system setup in optical-domain method. Figure 2. Direct thermal modulated spectrum of the Laser Diode (time averaged). Since the wavelength (or frequency) is single in this case, the result of beat signal is simple. Let this modulation light signal be A exp (jω a t) and the reflected signal be B exp {j(ω a + ω)t}, the intensity of the signal after interference will be I = A exp(jω a t) + Bexp{j(ω a + ω)t} 2 = A 2 + B 2 + 2ABcos( ω) (1) The beat frequency ω is yielded and some DC portion is remained. Therefore, this modulation method has a simple setup; however, it also comes with unwanted problems. The modulation speed is limited by the thermal distribution in the modulation process. Also, k-clock calibration is needed as mentioned above. 3. System setup in Electrical-Domain Method To solve the heat distribution and non-linearity problem in modulation, the electrical-domain method was introduced. The system was set based on the former setup; however, the laser was not getting modulated in this setup, but the external modulation method was employed. An arbitrary wave generator (AWG) and a Lithium-Niobate (LN) modulator were introduced in the system to sweep the wavelength. The system setup is shown in Fig. 3. CLRC 2018, June

3 Figure 3. LiDAR system setup in electrical-domain method. The AWG generated the electrical sweep frequency starting from 8 GHz to 13 GHz with the repetition rate of 200 khz. The laser wavelength as carrier frequency was set to be 1550 nm. The electrical signal was applied to the LN modulator, resulting the electrical frequency was superimposed to the optical frequency. In this case, we have double sideband wavelength spectrum as shown in Fig. 4. Since the modulation is conducted in the modulator, the linearity can be directly controlled. Hence, the k-clock MZI can be disable (physically connect to the system but turned-off). Figure 4. Spectrum of signal after LN modulation (time averaged). Similar to the Optical-Domain Method, the beat frequency can be derived. Let the modulation light signal be A exp(j(ω c + ω 1 )t) + A exp(j(ω c ω 1 )t) and the reflected signal be B exp(j(ω c + ω 1 + ω 2 )t) + B exp(j(ω c ω 1 ω 2 )t), the intensity after the interference will be I = Aexp{j(ω c + ω 1 )t} + Aexp{j(ω c ω 1 )t} + Bexp{j(ω c + ω 1 + ω 2 )t} + Bexp{j(ω c ω 1 ω 2 )t} 2 (2) After conjugate squared and term distribution, 5 different frequency terms of sinusoidal function is written below I = {4ABcos ( ω 2 )} + {4A 2 + 4B 2 } + {2A 2 cos( ω 1 )} + {4ABcos(2 ω 1 + ω 2 )} + {2B 2 cos(2 ω ω 2 )} (3) Using the band-pass filter to omit the unwanted frequency components, left the beat frequency term. I = 4ABcos ( ω 2 ) (4) Hence, the beat frequency ω 2 = ω b ω a is yielded (When ω a and ω b is the start and stop sweeping frequency) Although the beat frequency ω a ω b for distance measurement is yielded, other beat frequencies appear unlike optical-domain one. If these frequencies are close to main beat frequency, they can be noise or ghost. CLRC 2018, June

4 So, enough separation to filter out may be needed. In our measurement, the frequency separation is 8 GHz and it should not affect to the image. 4. LiDAR Image Results First, we measured the object 1-m away from the LiDAR probe. The LiDAR system can capture the image in the resolution of 512x512 pixels. The 2-demesional image comparison between Optical- and Electricaldomain setup can be seen in Fig. 5. Figure 5. Image comparison of 1. Image from camera (left), 2. Image from optical-domain system (middle), 3. Image from electrical-domain System (right). These result shows the electrical-domain method can also achieved the similar image compared with that of the optical-domain method. The only difference is the image brightness due to the lower output power of the electrical-domain method. This is due to higher total optical insertion loss due to the LN modulator. The 1-dimensional intensity at a certain length of LiDAR scanning are shown in Fig. 6. Peak-to-noise ratio are similar for both methods. Next, the effect of k-clock compensation was investigated. The comparison can be seen in Fig. 7. This comparison shows that the image by the electrical-domain method can produce a non-distorted image even though the k-clock system is disable. Thus, k-clock MZI can be removed in from the chip fabrication design. Figure 2. 1-dimensional depth profile of the object. CLRC 2018, June

5 Optical-Domain Electrical-Domain (a) k-clock enable (b) k-clock disable (c) k-clock enable (d) k-clock disable Figure 7. Images of optical- and electrical-domain methods with/without k-clock. 5. Conclusion The image performance comparison is carried out for two-types of swept light sources, thermally modulated optical-domain method and electrical-domain method in FMCW LiDAR. The qualities of images were compared and the result shows that the both methods can achieved similar quality of image with same peakto-noise ratio. The electrical-domain one showed linear frequency sweep to eliminate k-clock compensation. 6. References [1] C.V. Poulton et al., Coherent solid-state LiDAR with silicon photonic optical phased arrays, Optics Letters, Vol. 42, No. 20 (2017) [2] P. Adany et al., Chirped Lidar Using Simplified Homodyne Detection, Journal of Lightwave Technology, Vol. 27 No. 16 (2009) [3] S. Arafin, Advance InP Photonic Integrated Circuits for Communication and Sensing, IEEE Journal of Selected Topics in Quantum Electronics, Vol. 24, No. 1 (2018) [4] T. Baba et al., Line Beam Scanner using Slow-Light Waveguides in Si Photonics, Solid State Devices and Materials International Conference, H-8-04 (2017) [5] T. Baba et al., Investigation of Si-Photonics Slow Light LiDAR, JSAP Autumn Meeting, 14p-B4-10, (2016) [6] Y. Hayashi et al., GaInAsP/silicon-on-insulator hybrid laser with ring-resonator-type reflector fabricated by N2 plasma-activated bonding, Japanese Journal of Applied Physics, Vol. 55, pp , (2016) [7] N. Nishiyama et al., Investigation on Application of OCT system for FMCW-LiDAR System, IEICE Society Conference, C-3-22 (2017) CLRC 2018, June