Abstract
Hyperfine wind structure detection is important for aerodynamic and aviation safety. Pulse coherent Doppler wind LIDAR (PCDWL) is a widespread wind remote sensing method with tunable spatial and temporal resolutions. However, meter scale and sub-second resolution are still challenging for PCDWL. This is because of the constraints among short laser pulse duration, spectral broadening, detection accuracy, and real-time processing. In this Letter, to further improve the spatial and temporal resolution of PCDWL, we optimize the optical design of a nanosecond fiber laser and telescope and adopt a new, to the best of our knowledge, algorithm called the even-order derivative peak sharpening technique. During the experiment, all-fiber PCDWL with spatial and temporal resolutions of 3 m and 0.1 s, respectively, is demonstrated. Two-day continuous observation of the wakes of the Chinese high-speed train shows detailed hyperfine wind structures. This is similar to a computational fluid dynamics simulation.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
Pulse coherent Doppler wind LIDAR (PCDWL) is of great importance and has been widely used in various deployments over the past four decades. These include wind turbine optimization [1], airport surveillance for wind fields [2], meteorological guarantee for aviation safety [3], city air pollution monitoring [4], research on atmospheric dynamics and gravity wave propagation [5], and validation of numerical models for weather prediction [6]. Moreover, PCDWLs with spatial resolution of tens to hundreds of meters and a long detection range of over 10 km are mature. Nevertheless, considering special fields such as aerodynamic experiments and helicopter landing in complex terrain, continuous measurement at meter spatial and sub-second temporal resolutions is still a challenge owing to the limitations of hardware and signal retrial algorithms.
The temporal and spatial resolutions, wind detection range, and deviation are key parameters for PCDWL, and they are constrained by each other. Theoretically, a better spatial resolution requires shorter pulse duration; nonetheless, PCDWL relies on spectral narrowband backscattering. The overall spectral width of backscattering is the convolution of the laser pulse spectral width with the bandwidth of the atmospheric returns. A shorter pulse duration will bring wider pulse spectrum [7], which will lead to deterioration of other parameters without better hardware design and signal retrieval algorithms. Significant efforts have been made to realize PCDWL with a spatial resolution of less than 10 m. In 2015, NASA proposed a PCDWL with a minimum spatial resolution of 7.5 m by adopting a laser with adjustable pulse width duration between 50 and 400 ns [8]. In 2019 and 2021, the University of Science and Technology of China (USTC) proposed a PCDWL with a minimum spatial resolution of 6 m and 3.3 m, respectively, using pulse-coding technology [9,10].
To further improve the spatial and temporal resolution using physical parameters instead of algorithms, such as smooth interpolation or deconvolution, three predominant problems in PCDWL should be considered. (1) The performance deterioration induced by short pulse duration is more serious. (2) The relative intensity noise induced by pulse coding must be resolved. (3) More refined data results require more data calculations, making real-time calculation of wind speed a difficult problem. Recently, by optimizing optical designs, manufacturing a nanosecond laser, and adopting even-order derivative peak sharpening technology (EDPST) algorithms, continuous wind detection using hyperfine PCDWL (HPCDWL) at spatial and temporal resolutions of 3 m and 0.1 s, respectively, has been realized. Considering further performance verification, the HPCDWL was calibrated with an anemometer (Visalia WXT536) and was applied to a two-day observation of the wakes of the Chinese high-speed train. A train fluid dynamics simulation was performed, and the simulation result was similar to the actual observation result from the LIDAR.
The system layout of the HPCDWL is shown in Fig. 1. Considering the dashed lines, the PCDWL is split into three parts: nanosecond laser, adjustable telescope, and receiver. Regarding the nanosecond laser, a temperature-controlled narrow-linewidth seed laser emits a continuous-wave (CW) laser beam with a 3 kHz linewidth and is split into a local oscillator and a transmitted CW laser. The transmitted CW laser is modulated into a nanosecond pulse laser with a 20 ns full width half maximum (FWHM) and a 600 MHz frequency shift by two acousto-optic modulators (AOMs). To acquire a high-power nanosecond pulse laser with a lower amplified spontaneous emission (ASE) noise and a better pulse extinction ratio, AOM1, EDFA1, AOM2, and EDFA2 are used separately to build two-stage laser modulators and amplifiers. After EDFA2, the output pulse laser is sent and received by a circulator. A filter with a bandwidth of 0.1 nm (FWHM) is used to filter out other noise. The shape of the laser output is shown in Fig. 1(b).
Considering the adjustable telescope, the optical diameter is optimized to improve the carrier-to-noise ratio (CNR) between 6 m and 1000 m. Moreover, the focal length can be adjusted. According to the LIDAR equation, keeping parameters related to the telescope diameter design and other parameters as constant, the optimal diameter can be found using a detection range:
where ${\eta _a}$ is the heterodyne antenna efficiency, only related to telescope diameter D and traction efficiency $\rho $, here $\rho \approx 0.8$, and R is the detection distance [11]. The trend of the CNR is shown in Fig. 1(c).Regarding the receiver, the intermediate frequency (IF) is downconverted to 300 MHz using AOM3, which is sufficient for measuring a radial wind velocity of −230 to 150 m/s. The receiver bandwidth is limited to within twice the IF to reduce the amount of calculations and to achieve real-time signal processing using an onboard field programmable gate array (FPGA). However, the bandwidth of the balance detector and analog-to-digital digitizer is 1 GHz, which is compatible with other special dynamic range application environments.
The PCDWL power spectrum is usually composed of a Doppler signal and various noises such as the background noise of the detector, relative intensity noise of the laser, electromagnetic noise of the environment, and so on. Considering HPCDWL, owing to the spectral width broadening and window effects [12], most Doppler spectra have at least two peaks after subtracting the detector background noise. A quasi-Gaussian peak is always added with multi-amplitude harmonics, and the following paramount analysis is to determine the information of the main peak. As mentioned previously, the spectrum can be expressed as
Moreover, EDPST is an algorithm that artificially improves the apparent resolution of peaks. Mathematically, this technique is a simplified version of a Taylor series expansion. However, only the even-order derivative terms are considered, and their coefficients alternate in sign [16]. The expression can be written as
where ${R_{}}$ is the sharpened function, ${Y_{}}$ is the original function, ${Y^{(2i)}}$ are even-order derivatives, and ${k_{2i}}$ is the derivative weighting factor. A larger i leads to faster convergence and better performance in searching for hidden peaks. Nonetheless, this increases complexity because of the more adjustable factors, and i = 1 is enough for this application. Empirically, ${k_2}$ = 200 is selected. The equation can be rewritten as As shown in Fig. 2(a), there is an obvious difference between the normalized R and Y at approximately 230 MHz, where the EDPST finds a hidden peak. The parameters of the main peak and hidden peaks will be calculated using R and then applied to fit Y. Figures 2(b)–2(d) show the comparison results between the traditional algorithm and the EDPST. The EDPST is more accurate and robust, especially under low-CNR conditions. Details of data processing can be found in Ref. [17].After the hardware and algorithm optimization, three experiments were conducted to demonstrate the performance of this HPCDWL. First, to evaluate the accuracy of this HPCDWL system, an experiment comparing the LIDAR and Vaisala WT536 ultrasonic anemometer was conducted on the USTC campus on November 22, 2021. The ultrasonic anemometer was installed on a 1.8 m mast, which was approximately 50 m away from the LIDAR and at the same altitude. Considering this experiment, the LIDAR made continuous horizontal observations with a fixed azimuth of 300° and a temporal resolution of 10 s. The LIDAR records only the radial wind velocity, whereas Vaisala records both the wind velocity and direction. The mean radial wind velocities of the LIDAR from 48 to 54 m were compared with the corresponding projection results of Vaisala. To reduce the influence of the wind direction error caused by turbulence, only wind velocities greater than 1 m/s from Vaisala were considered. Figure 3(a) shows the wind speed and direction distribution, with strong northeast winds throughout the day. Moreover, Fig. 3(b) shows the linear regression results between the two devices and 7246 data points are fitted. The slope and correlation coefficient are 0.942 and 0.883, respectively. Furthermore, Fig. 3(c) shows the distribution of residual between LIDAR result and projection result. The mean value of the residual is −0.06 m/s, whereas the root mean square error is 0.482 m/s. The comparison results between 4 a.m. and 6 a.m. on November 22 are shown in Fig. 3(d). All the comparisons demonstrate that the HPCDWL detection accuracy is within 0.5 m/s.
Later, a fine wind structure observation experiment was performed at the Suzhou High-speed Railway Station in Anhui Province. Figure 4(a) shows a schematic diagram, where two horizontal azimuths are chosen for different aims. The first angle is 155°, which can cover the railway platform as much as possible. Figure 4(b) exhibits a continuous observation of 480 s. The background wind is light air (grade 1) in the range of 0.1 to 1.0 m/s. It is obvious that at the distances of 80 to 110 m and 190 to 220 m, there are two streaks with a negative wind velocity, which correspond to the inside of platforms No. 1 and No. 4, respectively. These may be topographical winds caused by the drop height between the railway and platform. The second angle is 25°, which is used to measure the unsteady gusts generated during high-speed train wakes. A total of 116 events were captured during the experiment in 36 h. Figure 4(c) shows two wake flows under different conditions. When the crosswind is light air, the wake flows can reach approximately 50 m downstream and exist over 30 s. Moreover, the corresponding maximum and minimum winds are 10.25 and −7.11 m/s in downwind and upwind conditions, respectively. The maximum wind velocity in the downwind wake flow is slightly less than the model value owing to the radial measurement [18]. Nonetheless, the upwind flow is impacted by the background wind to a certain degree.
Considering the application experiment, the HPCDWL had a fine wind structure with spatial and temporal resolutions of 3 m and 1 s, respectively. Nonetheless, there are still some problems to be solved. A major concern is that the maximum wind velocity in the wake flows is only 2 or 3 s, corresponding to two or three data points. The higher the temporal resolution, the more detailed is the wind structure information captured. There are two challenges in resolution enhancement. First, a higher resolution implies a huge computational cost in real time. Second, a higher resolution will reduce the CNR, and a lower CNR corresponds to a large detection variance. To increase the backscatter signal level, a third experiment was performed during a cold frontal passage. Humid air conditions obviously reinforce the CNR, as shown in Fig. 5(b). The minimum of CNR is −20 dB when the temporal resolution is 0.1 s, which makes the wind detection variance to be below 0.5 m/s. Figure19]. The near-wake flow lasted about 4 s with an obviously quasiperiodic structure as shown in Fig. 5(a). There are 6 beads and each continues for about 0.5 s. This structure is similar to Von Karman vortex shedding [20] and may originate from the last bogie. In the far-wake area, the wake flow interacts with background wind and was severely disturbed. Meanwhile, a simulation result obtained with Simcenter STAR-CCM+ 15.06 is shown in Fig. 5(c) (details of the simulation are shown in the Supplement 1). The speed trend and quasiperiodic structures in wake flows are similar to those in Fig. 5(a).
In conclusion, we have demonstrated continuous hyperfine wind structure observation using the HPCDWL with spatial and temporal resolutions of 3 m and 0.1 s, respectively. The all-fiber HPCDWL is compact and adjustable by optimizing the nanosecond laser, optical antenna, and wind retrieval algorithms. Particularly, the EDPST algorithm is adopted in PCDWL for the first time, and it enhances the quality and efficiency of wind retrieval processing. The observation and simulation results are similar to each other, and both show a structure similar to Von Karman vortex shedding. This is a vital first step toward the integration of computational fluid dynamics (CFD) and real-time hyperfine wind measurement.
Funding
State Key Laboratory of Pulsed Power Laser Technology; National Natural Science Foundation of China (42125402, 42188101); Innovation Program for Quantum Science and Technology (2021ZD0300300); Key-Area Research and Development Program of Guangdong Province (2020B0303020001); Shanghai Municipal Science and Technology Major Project (2019SHZDZX01); Joint Open Fund of Mengcheng National Geophysical Observatory (MENGO-202106); Fundamental Research Funds for the Central Universities.
Acknowledgment
Chong Wang thanks Qiankun Wang for assistance in conducting the experiments.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are available in Dataset 1, Ref. [17].
Supplemental document
See Supplement 1 for supporting content.
REFERENCES
1. Y. Käsler, S. Rahm, R. Simmet, and M. Kühn, J. Atmos. Ocean. Technol. 27, 1529 (2010). [CrossRef]
2. C. Shun and P. Chan, J. Atmos. Ocean. Technol. 25, 637 (2008). [CrossRef]
3. S. Rahm and I. Smalikho, J. Aircr. 45, 1148 (2008). [CrossRef]
4. H. Takashima, K. Hara, C. Nishita-Hara, Y. Fujiyoshi, K. Shiraishi, M. Hayashi, A. Yoshino, A. Takami, and A. Yamazaki, Atmos. Environ.: X 3, 100043 (2019). [CrossRef]
5. M. J. Jia, J. L. Yuan, C. Wang, H. Y. Xia, Y. B. Wu, L. J. Zhao, T. W. Wei, J. F. Wu, L. Wang, S. Y. Gu, L. Q. Liu, D. C. Lu, R. L. Chen, X. H. Xue, and X. K. Dou, Atmos. Chem. Phys. 19, 15431 (2019). [CrossRef]
6. M. Weissmann and C. Cardinali, Q. J. R. Meteorol. Soc. 133, 107 (2007). [CrossRef]
7. O. Reitebuch and R. M. Hardesty, in Springer Handbook of Atmospheric Measurements (Springer, 2021), pp. 759–797.
8. N. S. Prasad, R. Sibell, S. Vetorino, R. Higgins, and A. Tracy, Proc. SPIE 9465, 94650C (2015). [CrossRef]
9. C. Wang, H. Y. Xia, Y. B. Wu, J. J. Dong, T. W. Wei, L. Wang, and X. K. Dou, Opt. Lett. 44, 311 (2019). [CrossRef]
10. Y. P. Zhang, Y. B. Wu, and H. Y. Xia, Opt. Lett. 46, 5550 (2021). [CrossRef]
11. T. Fujii and T. Fukuchi, Laser Remote Sensing (CRC Press, 2005).
12. Z. C. Bu, S. Y. Chen, Y. C. Zhang, H. Chen, X. Y. Ge, and P. Guo, Chin. Opt. Lett. 12, S12801 (2014). [CrossRef]
13. J. L. Yuan, K. N. Wu, T. W. Wei, L. Wang, Z. F. Shu, Y. J. Yang, and H. Y. Xia, Remote Sens-Basel 13, 3815 (2021). [CrossRef]
14. M. Aoki, H. Iwai, K. Nakagawa, S. Ishii, and K. Mizutani, J. Atmos. Ocean. Technol. 33, 1949 (2016). [CrossRef]
15. M. F. Wahab, T. C. O’Haver, F. Gritti, G. Hellinghausen, and D. W. Armstrong, Talanta 192, 492 (2019). [CrossRef]
16. B. M. Romeny, Front-End Vision and Multi-Scale Image Analysis: Multi-Scale Computer Vision Theory and Applications, Written in Mathematica, Vol. 27 of Computational Imaging and Vision (Springer, 2003).
17. C. Liang, C. Wang, X. Xue, X. Dou, and T. Chen, “Data,” figshare (2022), https://doi.org/10.6084/m9.figshare.19890358.
18. C. Baker, J. Wind. Eng. Ind. Aerodyn. 98, 277 (2010). [CrossRef]
19. C. J. Baker, S. J. Dalley, T. Johnson, A. Quinn, and N. G. Wright, Proc. Inst. Mech. Eng. F. J. Rail. 215, 83 (2001). [CrossRef]
20. M. Matsumoto, J. Fluids. Struct. 13, 791 (1999). [CrossRef]