LIVE lidar has been designed for 3D wind profiling from an aircraft for meteorological applications. The wind measurement requirements are a horizontal resolution of 3 km, a vertical resolution of 100 m and an accuracy on each of the three wind vector components better than 0.5 m.s−1.
2.1. Fiber Laser Design
The laser source of the LIVE lidar is a pulsed 1.5 µm all-fiber laser based on master oscillator power amplifier architecture. To achieve the required accuracies for wind velocity measurement, the laser pulses should have an energy higher than 300 µJ and a duration of 700 ns. The coherent Doppler lidar application has additional requirements such as polarized laser emission, near-diffraction limited spatial beam quality and Fourier-limited spectral linewidth for coherent detection.
In fiber amplifiers, this last specification on linewidth limits the achievable laser energy, or more precisely, the achievable peak power. Indeed, to avoid nonlinear effects (particularly stimulated Brillouin scattering (SBS)), peak power must remain below the SBS threshold. Many methods have been experimented with to overcome this limitation, i.e., special fibers [11
] or the use of a temperature or strain gradient along the amplifying fiber [12
]. The simplest one from the lidar system point of view is the use of large mode area (LMA) fibers. In this case, the technical challenge is the fabrication of the amplifying fiber. The core diameter must be increased while reducing the numerical aperture to minimize the number of guided modes and maintain a quasi-single mode operation, since high beam quality is essential for lidar efficiency. In commercial pulsed power amplifiers at 1.5 µm, amplifying fibers are double-clad polarization maintaining Erbium–Ytterbium co-doped fibers. Two years ago, commercially available fibers of this type limited the peak power of a 700 ns pulse to 250–300 W when no SBS mitigation technic was implemented. A new glass matrix was proposed and after several tests, a new Erbium–Ytterbium co-doped fiber has been fabricated. This fiber enables us to reach a higher peak power, around 600 W, with all the necessary characteristics in terms of polarization maintenance, beam quality and amplification efficiency [13
For the LIVE lidar, we developed the last power amplifier stage of the laser source with this new fiber and evaluated the laser performance. A picture of the laser is presented in Figure 1
. The emission is linearly polarized, and we measured a polarization extinction ratio of 17 dB. An excellent beam quality is also observed with a measured beam quality factor M2
= 1.1. Pulses were temporally shaped thanks to the acousto-optic modulator to compensate for distortions in the different amplification stages in order to obtain a duration of 715 ns and the pulse shape is presented in Figure 1
. The pulse repetition frequency is 14 kHz and the pulse energy is limited by the SBS to 410 µJ. A double-pass architecture in the preamplifier enabled us to minimize ASE (Amplified Spontaneous Emission) generation and amplification in the power amplifier; a low level of ASE (<2%) is measured [14
2.2. Lidar Design
LIVE Lidar has been designed to be compact, robust and suited to ATR42 aircraft integration. It was composed on an aircraft adapted bay that includes (see Figure 2
A rack for various Lidar power supplies.
A control rack for the scanning system.
A rack for the emission of the Lidar beam including the Onera laser source.
A rack for the Lidar signal reception and its coherent detection.
A rackable computer for real time processing.
The total power consumption is about 900 VA (Volt Ampere) and the weights of the scanning head and the lidar bay are 18 kg and 58 kg, respectively. The lidar bay is sufficiently compact to occupy a volume of 55 × 65 × 130 cm3.
The telescope is in a carbon tube positioned above the hatch of the aircraft on a metal interface plate that also supports the scanning system. Scanning for the analysis of the wind fields involves a system consisting of two mirrors to allow conical addressing. The beam emitted by the telescope is sent back to a fixed mirror and then to a mirror mounted on a gimbal-type scan turret as represented in Figure 2
For a flight altitude of 5000 m, and an aircraft speed of 100 m.s−1, the 3 km horizontal resolution is obtained with conical scanning with a total opening angle of 30° and one round in 17 s. After each lap, the lidar points in the nadir direction in 2 s in order to achieve good accuracy on vertical wind measurements.
The lidar is installed inside the SAFIRE ATR42 looking through a trapdoor in the floor of the plane in while aiming towards the nadir (Figure 3
2.3. Lidar Signal Processing
Lidar raw data are first processed into spectrograms which show power spectral density as a function of lidar range on one axis and Doppler frequency on the other axis. Within a short time (0.3 s), called the line of sight time, the motion of the laser beam (6°) is considered small enough to be neglected. All computed spectrograms within this line of sight time are averaged in order to reduce noise. This task is made in real time using GPU (Graphics Processing Unit) computing. Such averaged spectrograms are computed for every line of sight of a full 360° scan. Thus, after a full scan, spectrogram data show power spectral density as a function of line of sight angle, lidar range and Doppler frequency.
From this stage, real time processing and post processing will differ. The aim of real time processing is to quickly deliver information about lidar measurement. The carrier to noise ratio (CNR) and radial wind speed are computed from spectrogram data at all lidar ranges and line of sight angles, using fast computing moments estimators along the Doppler frequency dimension. This CNR and radial wind speed data (functions of line of sight angle and lidar range) are displayed in real time by the lidar interface for monitoring purposes.
The aim of post processing is to offer the best performance. Since the aircraft will operate above the planetary boundary layer (PBL), the backscatter coefficient is expected to be a hundred times lower than usual at ground level. Therefore, the best low CNR estimators must be used for wind vector retrieval. According to [15
], the usual algorithms such as velocity azimuth display (VAD) are limited at low CNRs; in essence, these algorithms use radial velocity data computed from the spectrogram in a way that is similar to real time processing. The limitation at low CNRs lies in the fact that each column of the spectrogram (along the Doppler frequency dimension) has been computed within a short time (of a line of sight). Thus, retrieving information at a column scale is not efficient in terms of low CNR performance. A more efficient algorithm would work with spectrogram data as a whole along the line of sight angle dimension.
According again to [15
], two algorithms show excellent low CNR performance: the maximum of the function of the accumulated spectra (MFAS) and the wind vector maximum likelihood (WVML). For each lidar range, both algorithms compute the wind speed vector directly from spectrogram data along the Doppler frequency dimension and line of sight angle dimension. Both algorithms are complementary and are applied to our data during post processing.
The MFAS is, in essence, well suited for signal detection.
shows an example of low CNR spectrogram data obtained at high altitude (first range gate from the aircraft), which at first sight seems to contain only noise. The MFAS algorithm has detected the lidar signal, and its presence is highlighted by a trail of black dots. Lidar signal center frequency as a function of scan angle has a sinusoidal shape due to the projection of a constant wind vector on laser beam direction with conical motion. The phase of the sinusoidal shape is the direction of horizontal wind, its amplitude is proportional to the modulus of horizontal wind, and its offset from zero Doppler frequency is proportional to vertical wind. The MFAS algorithm makes rough assumptions about the value of horizontal wind (within the range of expected values). There is a sinusoidal shape for each assumption. As shown in Figure 5
on the left hand side, the MFAS algorithm shifts the spectrogram data of frequency in order to compensate for lidar signal center frequency dependence on scan angle. Then, the MFAS algorithm averages the compensated spectrogram data over all scan angle values, as shown in Figure 5
on the right hand side. If the assumption is roughly equal to the real value of horizontal wind, the accumulated spectra will exhibit a signal peak. The closer the assumption is from the real value, the stronger the signal peak, hence the name of the algorithm: maximum of the function of accumulated spectra. The remaining frequency offset from zero Doppler frequency leads to an estimate of vertical wind. Accumulated spectra also allow us to estimate the lidar CNR.
The accuracy of the MFAS algorithm is related to the number of assumptions being computed, i.e., the computing time allocated to the algorithm. It is wise to use a rough MFAS estimate in order to initiate a WVML algorithm dedicated to wind accuracy. The WVML algorithm is a more common algorithm which fits a wind parametric model over spectrogram data thanks to an optimization algorithm. We then proceeded with wind vector estimation for all lidar ranges.
At this stage, the wind has been estimated relative only to the aircraft. In order to obtain the wind relative to the ground, we must add aircraft speed relative to the ground (or cancel ground speed relative to the aircraft). The accuracy of aircraft attitude from FTI (Flight Test Instrumentation) and lidar orientation relative to the aircraft is not high enough to safely add aircraft speed. For an aircraft speed of 100 m.s−1, an error of orientation of a single degree will lead to a residual aircraft speed of 1.75 m.s−1 on axes perpendicular to aircraft motion. The desired wind speed accuracy is 0.5 m.s−1, and this goal requires a precise ground speed relative to the aircraft cancelling procedure. We have chosen to work with ground speed measured from the lidar ground level return. Since both measurements are done within the same frame of reference, there is no lidar orientation issue. As the signal level from the ground return is stronger than the signal level from the atmospheric return, we use a more common VAD algorithm to retrieve ground speed relative to the aircraft.