The Doppler Aerosol Wind (DAWN) Airborne, Wind-Proﬁling Coherent-Detection Lidar System: Overview and Preliminary Flight Results

826 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 31

MICHAEL J. KAVAYA,JEFFREY Y. BEYON,GRADY J. KOCH,MULUGETA PETROS, PAUL J. PETZAR,UPENDRA N. SINGH,BO C. TRIEU, AND JIRONG YU NASA Langley Research Center, Hampton, Virginia

(Manuscript received 19 December 2012, in ﬁnal form 25 October 2013)

ABSTRACT

The first airborne wind measurements of a pulsed, 2-mm solid-state, high-energy, wind-profiling lidar system for airborne measurements are presented. The laser pulse energy is the highest to date in an eye-safe airborne wind lidar system. This energy, the 10-Hz laser pulse rate, the 15-cm receiver diameter, and dual- balanced coherent detection together have the potential to provide much-improved lidar sensitivity to low aerosol backscatter levels compared to earlier airborne-pulsed coherent lidar wind systems. Problems with a laser-burned telescope secondary mirror prevented a full demonstration of the lidar’s capability, but the hardware, algorithms, and software were nevertheless all validated. A lidar description, relevant theory, and preliminary results of flight measurements are presented.

1. Introduction overviews have been published (Baker et al. 1995; Singh et al. 2005); and the measurement requirements have a. Background been developed (NASA 2006)]. Here we report on the The primary motivation for wind measurement tech- first aircraft flights of a pulsed, 2-mm, coherent-detection nology development at the National Aeronautics and lidar system having basic laser parameters matching those Space Administration (NASA) is to prepare for an required for a space mission. Because of the advanced Earth-orbiting lidar system to globally measure vertical laser, the lidar system can provide the best sensitivity to profiles of horizontal wind magnitude and direction aerosol backscatter levels to date for this type of airborne (horizontal vector winds). This mission is one of 15 earth wind lidar. science missions recommended to NASA by the National b. Lidar system Research Council (NRC), which they call ‘‘3D Winds’’ (NRC 2007). Required preparation for this space mis- A block diagram of the Doppler Aerosol Wind sion includes formulation of the wind measurement (DAWN) lidar system is shown in Fig. 1. The pulsed requirements, technology development and validation, solid-state laser uses the Ho:Tm:LuLiF lasing crystal measurement technique development and validation, with laser diode array (LDA) side pumping. The pulsed atmospheric characterization, theoretical advancement, laser resonator closely follows the wavelength of the computer simulation of measurement performance ad- injected seed laser. The laser emits nominally 250-mJ vancement, and ground and airborne measurement pulses at a 10-Hz rate. The linearly polarized laser pulse is demonstration. Computer simulations began over three converted to circular polarization by a transmit–receive decades ago [Huffaker et al. 1984; atmospheric charac- switch comprising a polarizing beam splitter and a terization has been advanced, especially for global aero- quarter-wave plate. The 15-cm beam-expanding tele- sol backscatter levels (Menzies and Tratt 1997); relevant scope expands the input pulse to approximately the theory has been developed by many groups; mission optimum 12-cm diameter (Rye and Frehlich 1992). The expandedbeamisdirectedtonadirandentersanop- tical wedge that deflects the pulse by 30.128.Theazi- Corresponding author address: Michael J. Kavaya, NASA Langley Research Center, 5 N. Dryden St., Mail Code 468, Hampton, VA muthal direction of the wedge’s deflection is chosen by 23681. a computer-controlled ring motor that can turn the E-mail: michael.j.kavaya@nasa.gov optical wedge about a nominal nadir axis. As the pulse

DOI: 10.1175/JTECH-D-12-00274.1 APRIL 2014 KAVAYAETAL. 827

FIG. 1. Block diagram of the DAWN lidar system. travels away from the airplane, the aerosol and cloud 2. Theory particles it intercepts scatter the light in many di- a. Geometry rections, including a small fraction directed back (backscattered) to the lidar system. The backscattered Our wind-measurement flight geometry is depicted light as measured from the moving aircraft is Doppler in Fig. 2, using the right-handed Cartesian geographic frequency shifted by an amount proportional to the northeast-down (NED) coordinate system. Consider any projection of the airplane and wind velocity into the vector V. In NED, the polar angle u goes from the posi- direction of the pulse. The backscattered light, which tive geographic down (z) axis to the three-dimensional reenters the lidar system, is directed to the optical (3D) vector. The azimuth angle f lies in the north–east signal detector, where it is heterodyned or mixed with (x–y) horizontal plane, and goes from the positive geo- a portion of the injection laser. The injection laser also graphic north (x) axis, moving toward (and possibly serves the role of a local oscillator (LO) laser. Part of past) the positive east (y) axis, to the projection of the the transmitted pulse is directed to an optical monitor 3D vector into the north–east plane. The lidar-carrying detector where it is mixed with a frequency upshifted aircraft is flying approximately level, with the 3D air- portion of the injection laser. The frequency shifter, an craft nose (body forward) direction vector FA (unitless) 21 acousto-optic modulator (AOM), permits determina- and the 3D aircraft velocity vector VA (m s ). The tion of the magnitude and sign of the frequency dif- projections of these vectors into the horizontal plane are ferent between the injection laser and the pulsed laser. vectors FAH and VAH, where the subscript AH stands Knowing this frequency difference, the projected air- for ‘‘aircraft horizontal’’. In aviation, FAH defines the craft velocity, and the AOM frequency, permits cal- aircraft heading angle fF (rad) and VAH defines the culation of the projected wind velocity. When this is aircraft track angle fV (rad). The angle fVF 5 fV – fF repeated for several scanner azimuth angles, the three (rad) is the aircraft drift angle caused by horizontal components of the wind velocity may be determined. crosswinds. For the crosswind blowing primarily toward Major lidar system subsystems not shown in Fig. 1 include the east (westerly wind, Fig. 2) or west (easterly wind), power supplies, control circuits, laser cooling equip- either FAH or VAH may be closer to geographic north, ment, rigid mechanical structure for inter-component meaning the drift angle is positive or negative, respec- alignment stability, processing algorithms, and software. tively. The wind, carrying natural aerosol particles along 21 Many more details about the lidar system will be given in at the same velocity, has 3D velocity vector VW (m s ) section 3. and horizontal projection VWH; VWH has azimuth angle The rest of this paper will describe the theory used for fW (rad). geometry, Doppler shift, coordinate system conversions, During wind measurements, the airborne lidar sets the alignment error compensation, and signal processing; lidar scanner azimuth angle fS (rad) and then fires 21 provide more details on the Doppler lidar system; and several laser pulses with 3D velocity vector cIL (m s ; present preliminary flight measurement results from or possibly unequal vectors cIL,i for each pulse i), where NASA’s hurricane Genesis and Rapid Intensification IL is a unit length vector in the laser pulse direction and 21 Processes (GRIP) aircraft campaign in 2010. c (m s ) is the speed of light; ILH and fL are not shown 828 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 31

FIG. 2. Geometry for the airborne lidar measurement of wind. in Fig. 2. The illuminated aerosol particles backscatter pulse power temporal shape with a 1D Gaussian function a small fraction of the laser pulse’s energy along 3D truncated at 63 standard deviations, where it is about direction vector cIL and into the lidar system’s receiver 1.1% of the center height. For the Gaussian shape, the for detection. In our case, as is typical for coherent lidar, pulse will reach 50% of its center height at 2(2 ln2)0.5 5 the transmit telescope is also the receive telescope. 21.1774 standard deviations; and we arbitrarily let t 5 Because of special relativity, translation of the airplane 0 (s) be at this point in the pulse. The pulse duration is does not cause any transmit–receive phase front mis- deﬁned as the full width at half maximum (FWHM), and alignment (misalignment of the received phase front has duration tL (s) 5 2.355 standard deviations. The pulse with respect to the transmitted phase front; Gudimetla rate is frequency fL (Hz). and Kavaya 1999). Airplane rotations do cause mis- The lidar receiver’s signal at time t is primarily due to alignment. The laser velocity vector cIL makes 3D angles all the aerosol particles illuminated by the laser pulse VLA and VLW (rad) with VA and VW,respectively. and positioned between minimum range c(t 2 tL)/2 and The airplane has a conceptual, right-handed, forward- maximum range ct/2 (Kavaya and Menzies 1985). Range right-belly (FRB) coordinate system attached to its is proportional to time. The best or smallest possible airframe. This coordinate system is used in two ways. range resolution is approximately the length of this

First, the lidar scanner is mounted and calibrated with range interval, which is RRES,MIN ; ctL/2. The transmitted the goal that its setting provides the transmitted laser laser pulse intensity spatial cross section typically has a beam direction relative to the FRB coordinates. Second, shape similar to the modeled temporal shape, except the the lidar system’s inertial navigation system–global po- Gaussian is 2D with a spatial standard deviation differ- sitioning system (INS–GPS), which is mounted to the ent from the temporal standard deviation. It is useful to lidar (see Fig. 1), provides the yaw, pitch, and roll angles conceptually think of a circular cross section that includes of the physical case of the INS–GPS with respect to the 86.5% of the intensity. This circle extends to the radius 2 NED coordinate axes. The goal is that these three angles that has e 2 5 13.5% of the on-axis intensity (i.e., extends also represent the rotations of the FRB axes (airplane radially to two spatial standard deviations from center). body) with respect to the NED coordinate axes. In prac- In general, combined with the temporal shape, this ‘‘cy- tice, neither of these goals is perfectly met, and extra lindrically’’ shaped volume of air may contain different calibration steps are required. values of laser pulse power, laser pulse frequency (chirp), Both the transmitted laser pulse and the backscattered aerosol concentrations, wind velocity, and wind turbu- light travel at approximately the vacuum speed of light lence. For our DAWN lidar system, typical values are tL ; 21 c 5 299 792 458 (m s ). We model the transmitted laser 180 ns and RRES,MIN ; ctL/2 ; 27 m. APRIL 2014 KAVAYAETAL. 829

In operation, the lidar receiver detector signal is dig- We next combine Eqs. (2) and (3), use the vector dot itized by an analog-to-digital converter (ADC) at sam- product expressions, 21 pling rate fADC 5 1/TADC (sample s or Hz), so that further signal processing may be performed by a com- V (cI ) cI V cosV 5 A L and cosV 5 L W , puter. This is easier and has more options than performing AL [jV jj(cI )j] LW [j(cI )jjV j] signal processing with optical or analog electronic com- A L L W ponents. The continuous string of ADC signal samples (4) from each laser shot is grouped into sets of NRG (sample) consecutive signal samples, which are called range gates and use the spherical coordinate system definitions for (RG). Successive range gates provide a range profile the polar uL and azimuth fL angles (rad), of the wind. The range gate observation time is TRG 5 5 u f 5 u f NRG/fADC (s), and the range resolution (m) of a range ILX sin L cos L, ILY sin L sin L, ; 1 gate is RRG (c/2)(tL NRG/fADC). DAWN typi- 5 u u 82 8 f 82 8 6 ILZ cos L , and L:0 180 , L:0 360 , (5) cal values are fADC 5 500 3 10 samples per second (106 samples 5 1 Msample), T 5 2 ns, N 5 512, ADC RG to obtain NRG/fADC ; 1.02 ms, and RRG ; 180.5 m. 2 ni 2 ni ) 5 [sinui cosui (Vi 2 V ) b. Doppler shift equation ( R L l L L AX WX L The coherent-detection Doppler lidar system is de- 1 sinui sinui (Vi 2 V ) signed to measure the change in light frequency between L L AY WY 1 ui i 2 the transmitted laser pulse and the backscattered (re- cos L(VAZ VWZ)], (6) ceived) light. Referring to Fig. 2, the equation linking the light frequency change, the velocities, and the angles where l 5 c/n (m) is the transmitted laser pulse is (Taff 1983) L L wavelength and x, y, and z represent any arbitrary right- 2b cosV 2 2b cosV handed coordinate reference frame. In practice the (n 2 n ) 5 n A AL W LW , (1) R L L(1 1 b cosV )(1 2 b cosV ) user must select the coordinate system in which to use W LW A AL Eq. (6)—see section 2c below for a discussion of dif- jV j jV j ferent coordinate systems. A superscript i has been b 5 A and b 5 W . (2) A c W c added to variables expected to change with different lidar scanner azimuth angles fL. The reason that the u We use n to indicate very high-optical frequencies as superscript is on the laser beam polar angle L is to allow f opposed to the English letter f for lower frequencies, the possibility that changes in azimuth angle L cause subscript L to indicate the laser’s transmitted frequency, changes in the polar angle. The three wind velocity subscript R to indicate the backscattered received opti- components are assumed to be constant for all azimuth cal frequency, subscript A for the aircraft, and subscript angles in a scan pattern. W for the wind. Equations (1) and (2) include special All possible directions of the transmitted laser beam u f relativity and are valid for any airplane (lidar platform) have a unique pair of angles ( L, L)—these angles and wind velocities. Equation (1) has never been pub- with subscript L represent the actual direction of the lished in a widely available journal to our knowledge, laser beam outside the airplane and are not necessar- u f and so we include it here. For the airplane flights de- ily equal to the settings of the lidar scanner ( S, S)— scribed herein, we may use the limit of small velocities because imperfections in the lidar scanner, lidar assem- compared to light speed. Both terms in Eq. (2) approach bly and alignment, and lidar mounting to the aircraft 0, and Eq. (1) becomes frame may introduce angle changes. These angle changes may not be simple angle offsets but may be functional, n 2 n ’ n b V 2 b V u f 5 u f ( R L) L(2 A cos AL 2 W cos LW). (3) ( L, L) f( S, S); that is, changes in the lidar scanner azimuth angle may change the laser beam polar angle. (The tangential velocity of a spacecraft will be much For the flights reported here, scanner setting values larger than an aircraft’s velocity. A recent investigation were approximately uS 5 30.128,andfS 52458, 222.58, showed the approximation in (3) could lead to a wind 08, 22.58, and 458. The five azimuth angles comprised a 2 velocity error of about 0.1 m s 1 (Ashby 2007), which is scan pattern. probably too large to ignore with overall wind error The unknown quantities in Eq. (6) are the three com- 21 goals of 1–3 m s from all contributions. ponents of the wind velocity vector VW. Therefore, for 830 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 31 wind data processing, at least three copies of Eq. (6) with the SWD system. For example, if air is blowing east different values of the lidar scanner azimuth angle fS like in the NED discussion, then the SWD azimuth must be simultaneously solved for the wind components. angle is 2708. Reporting the number 2708 in NED

If we assume the vertical wind VWZ 5 0, then at least two means the direction west. The meteorological com- copies of Eq. (6) are needed. munity calls the wind a westerly wind and will in- We initially solved the data processing algorithm us- terpret the 2708 in NED correctly. ing only two azimuth angles and equations, and assumed One processing option is to use Eq. (6) entirely in the vertical wind 5 0. We used fS 52458 and 458, and did not use the recorded data from the other three azimuth ENU coordinate system. The laser frequency differ- angles. Later we solved the data processing algorithm ence and wavelength do not change with the coordinate using all five azimuth angles and were able to calculate system. The aircraft velocity components VA are al- vertical wind also. ready provided in ENU. The lidar scanner settings and functional equation provide the laser beam direction c. Using four different coordinate systems with respect to FRB, which is also with respect to NED before aircraft rotations. The laser direction vector The angles and velocities in Eq. (6) must all be outside the aircraft is then rotated within NED by ap- referenced to the same coordinate system. Then the plying the appropriate rotation matrices using the air- solution from Eq. (6) must be converted to the co- craft body rotations. (The next section describes the ordinate system desired by the meteorological wind rotation matrices.) Further conversion of the laser di- user community. We had to deal with four coordinate rection vector to ENU is straightforward. Several systems: copies of Eq. (6) arethenusedtosolveforthethree 1) The FRB coordinates of the airplane. The laser beam components of wind velocity in ENU. The vertical wind direction outside the airplane is known with respect component is reported in ENU. The horizontal wind to FRB using the lidar scanner settings and the angle- direction is reported either in NED or SWD as dis- converting functional equation. cussed above. 2) The NED coordinates. Our lidar-mounted INS–GPS unit provided yaw, pitch, and roll angles of the air- d. 3D rotation matrix due to roll, pitch, and yaw in plane body in NED. INS–GPS roll and pitch errors NED are specified as 1.0 mrad–1s. Heading error 5 yaw error is specified as 1.5 mrad and up, depending on As stated above, a key step before using Eq. (6) is to the recent aircraft horizontal accelerations. When all rotate the laser beam direction outside the airplane us- three aircraft rotations equal zero, the FRB axes are ing the aircraft rotation angles provided by the INS– aligned with the NED axes. (Heading equals yaw GPS unit and staying in the NED system. We begin with because we are dealing with an airplane and the the three Cartesian 3D rotation matrices (Goldstein aviation word for yaw in NED is heading.) et al. 2001, section 4.2): If the horizontal wind direction (HWD) is plotted as 2 3 arrows in a top-view plot, then the meteorological 10 0 community wants the arrow to point in the direction 6 7 R 5 6 0 cosC 2sinC 7 of the air movement. This indicates the NED system. X 4 X X 5 C C For example, if air is blowing east, then the NED 0 sin X cos X 8 2 3 azimuth angle is 90 . The community wants the arrow cosC 0 sinC to point with a heading of 908 in the plot. 6 Y Y 7 P 5 6 7 3) The east–north-up (ENU) coordinates. Our INS– Y 4 0105 GPS unit provided the three components of the 2 C C sin Y 0 cos Y aircraft velocity in ENU. The velocity component 2 3 21 s cosC 2sinC 0 errors are all specified as 0.1 m s –1 . The meteo- 6 Z Z 7 rological community prefers the vertical wind to be Y 5 6 C C 7 Z 4 sin Z cos Z 0 5. (7) labeled positive if the air is moving up, which in- 001 dicates the ENU system. 4) The southwest-down (SWD) coordinates. If the horizontal wind is reported with a numerical azimuth These matrices represent, from top to bottom roll (R)

angle, then the meteorological community prefers the about the x axis by angle CX (rad), pitch (P) about the y direction from which the air is coming. This indicates axis by CY,andyaw(Y) about the z axis by CZ. The capital APRIL 2014 KAVAYAETAL. 831 roman letter indicates the popular name of the rotation, Our particular INS–GPS unit [Systron Donner In- while the subscript indicates the axis of rotation. The ertial C–Miniature Integrated GPS/INS Tactical System matrix element indices with negative signs are 23, 31, (MIGITS) III] assumes that 1) the rotation order is yaw, and 12, respectively. Defining the number sequence then pitch, then roll; 2) the rotation angles it provides 123123, the indices with negative signs all follow the will be used in rotation matrices that rotate the co- sequence order. These matrices rotate a vector direction ordinate axes; and 3) all rotations are counterclockwise while holding the coordinate axes constant in inertial (CCW) when looking down the rotation axis from in- space, giving the three new vector components of the finity. (Caution is advised since these assumptions may rotated vector in the original coordinate system. For the differ with different INS–GPS units.) Because of the inverse matrices of the three rotation matrices, which second assumption by the INS–GPS, we must use the are their transposes since they are orthogonal matrices, inverse matrices of those in Eq. (7) to match the INS– the indices of the negative components would be 32, 13, GPS. The result of doing that will be an overall 3D ro- and 21, respectively. They are in the reverse order of the tation of the coordinate axes. But we need to rotate the 123123 sequence. They would rotate the coordinate axes laser beam vector and not the coordinate axes, so one while holding the vector direction constant in inertial space, final inverse of the total is needed. To properly combine giving the three vector components of the original vector in the rotation matrices in Eq. (7) to rotate the laser beam the new rotated coordinate system. vector, we must use the following rotation equation:

2 3 2 3 2 3 2 3 ROT ILX ILX ILX ILX 4 ROT 5 5 4 5 5 R21 P21 Y21 214 5 5 Y P R 4 5 ILY (M) ILY [( X )( Y )( Z )] ILY ( Z)( Y )( X ) ILY . (8) ROT I I I ILZ LZ LZ LZ e. Finding the functional equation and its parameters scanner settings to the actual direction of the laser beam outside the aircraft. This equation is found by modeling Before processing the wind data, it is necessary to ﬁnd the possible mounting and optical alignment imperfec- the parameter values of the mounting and alignment tions. Without showing the lengthy derivation, an ex- imperfection functional equation that converts the lidar ample of a functional equation for DAWN is

5 f 1 A4 S A3 , 2 u 5 cos 1[2sinu sinA cos(A 2 A ) 1 cosu (cosA )], L ( S 1 4 2 S 1 ) 2 2 2 [2sinu cosA sinA cosA (1 2 cosA ) 1 sinu sinA (cosA sin A 1 cos A ) 1 cosu (sinA sinA )] f 5 tan 1 S 4 2 2 1 S 4 1 2 2 S 2 1 . L u 2 1 2 2 u 2 1 u [sin S cosA4(cosA1 cos A2 sin A2) sin S sinA4 sinA2 cosA2(1 cosA1) cos S(cosA2 sinA1)] (9)

Both laser beam angles are functions of both scanner must be properly accounted for. This is accomplished in setting angles. Equations (9) have three unknown pa- many computer codes by using the atan2 function in- rameters: A1, A2,andA3 (rad). To derive these equa- stead of the inverse tangent function. tions, we assumed the rotation axis of the wedge To determine the three unknown parameters, lidar scanner (see Fig. 1) is unintentionally tipped from data are taken over land with NAZ,CAL azimuth angles vertical by angle A1 in NED azimuth direction A2.This per scan pattern, and the range gates with land return tipping may come from the lidar mounting in the air- signals are identiﬁed. For these range gates, we assume craft and/or from the aircraft ﬂight attitudes, We also the land is nonmoving, and Eq. (6) becomes assumed that when the scanner motor is electronically commanded to home position, the thickest part of the ni 2 ni 5 2 ui ui i 2 ( R L) l [sin L cos L(VAX 0) optical wedge is unintentionally offset from the NED L north direction by azimuth angle A3.Weassumedthe 1 ui ui i 2 sin L sin L(VAY 0) laser beam perfectly aligns with the rotation axis of the 1 cosui (Vi 2 0)]. (10) optical wedge. The correct quadrant for the angle fL L AZ 832 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 31

The ‘‘wind’’ velocity components for the ground re- g. Pulsed coherent lidar power carrier-to-noise ratio turns have been zeroed. The aircraft velocity com- equation ponents may differ between different lidar scanner We define the power (P) carrier-to-noise ratio (CNR ) azimuth angle measurements. We assume they are P to be the ratio of the ensemble-mean desired signal power constant during shot accumulation at one azimuth an- to the ensemble-mean undesired—hence, noise, power— gle. The number of unknown parameters of the func- and the power signal-to-noise ratio (SNR )astheratio tional equation—in our case, 3—is the minimum value P of the ensemble-mean desired signal power to the en- for N .TheN copies of Eq. (10), one for AZ,CAL AZ,CAL semble standard deviation of the desired signal power. each value of the lidar scanner azimuth angle, are si- (Some publications refer to CNR as SNR ,andsome multaneously solved to find the three unknown pa- P P publications define SNR as the square of CNR .) Since rameter values. These parameter values are later used P P the backscattered light is from many randomly oriented when solving for the three unknown wind components. aerosol particles that are randomly distributed over As a minimum, the parameter values should be de- the illuminated volume, the received signal is a random termined following each integration of the Doppler lidar process. Coherent detection lidars are designed so that the hardware into the aircraft. It would be better to determine postdetection noise is dominated by the vacuum quantum them more often in case wind turbulence or landing fluctuations (numerically equal to the local-oscillator- bumps change the functional equation. As an example, induced shot noise model; Shapiro and Wagner 1984). one determination of our parameter values yielded A1 5 2 2 Since the dominating noise is added during the optical 22.37 3 10 5 rad, A 523.79 3 10 5 rad, and A 5 2 3 detection process, adding another random process, there 0.0186 rad (24.9 and 27.8 arc s, and 1.18, respectively). is no need to track predetection optical CNRP —only the postdetection electrical CNR . The pulsed, coherent- f. Solving for the unknown wind components P detection lidar CNRP equation can be written in many AcopyofEq.(6) is created for every azimuth angle forms representing the trade-offs between the desire to to be used in solving for the unknown wind compo- provide physical intuition versus the desire to include nents. The functional equation is used with the ap- as many lidar, target, and environment variations as propriate parameter values discussed in the previous possible. We refer the reader to any number of excellent section. The set of Eqs. (6) is solved simultaneously sources, such as Gagliardi and Karp (1976, chapter 6), for the three wind components, or, in the case of us- Kingston (1978, chapter 3), Rye (1979), Frehlich and ing two azimuth angles, for the two horizontal wind Kavaya (1991),andHenderson et al. (2005). components. For example, Eq. (7.40) in Henderson et al. (2005) is

ðð 2 " # [h (x, y)u~ (x, y)]*u~ (x, y, t) dx dy P (t)h h P QE LOD SD CNR (t) 5 ST OR BSR N 3 ðð ðð . (11) P hnB P 1 P ( N EN) h j~ j2 j~ j2 QE(x, y) uLOD(x, y) dx dy uSD(x, y, t) dx dy

In Eq. (11) all the quantities are ensemble averages, the analytic signal representation of a real-valued op- 21 superscript asterisk (*) refers to the complex conjugate of tical field u, u~SD(x, y, t)(V m ) is the received signal a complex number, t (s) is the time, x and y (m) are the complex optical field at the detector plane (SD), and 21 coordinate axes perpendicular to the optical beam di- u~LOD(V m ) is the local oscillator (LO) complex optical rection z, PST(t) (W) is the desired backscattered signal field at the detector plane (LOD). Equation (11) assumes h optical power collected by the telescope (ST), OR is the the LO power is constant and is large enough that h optical receive (OR) path intensity efficiency (0–1), BSR shot noise dominates other light-caused noise, that total is the beam splitter receive (BSR) path intensity efficiency noise is white over the bandwidth, and that any quantum 2 (0–1), h 5 6.626 3 10 34 (J s) is Planck’s constant, B (Hz) efficiency dependence on the intermediate frequency is the receiver bandwidth, PN (W) is the noise (N)power ÐÐ(IF) signal can be neglected. The denominator term ~ 2 2 due to ‘‘shot noise,’’ PEN (W) is the excess noise (EN) juSD(x, y, t)j dx dy(V ) equals the signal optical power h « 2 5V power in addition to shot noise, QE(x, y) (photoelectrons– at the detector plane, PSD (W) times 2/(c 0)(V/W ), 21 photon, effectively dimensionless, and 0–1) is the optical where «0 is the vacuum permittivity (F m ). detector’s quantum efficiency (QE), u~ is the complex We may reorganize Eq. (11) to be APRIL 2014 KAVAYAETAL. 833 ðð 2 [h (x, y)u~ (x, y)]*u~ (x, y, t) dx dy P (t)h h P QE LOD SD CNR (t) 5 ST OR BSR N 3 ðð ðð P hnB P 1 P ( N EN) jh ~ j2 j~ j2 QE(x, y)uLOD(x, y) dx dy uSD(x, y, t) dx dy ðð jh ~ j2 QE(x, y)uLOD(x, y) dx dy 3 ðð . (12) h j~ j2 QE(x, y) uLOD(x, y) dx dy

In Eq. (12), we have divided the integral term into an of the constant quantum efficiency and CNRP(t) } h equivalent product of two integral terms. Equation (12) PST(t) QE,CONSTANT. now closely matches Eq. (3.3b) in Kingston (1978,chap- h. Can airplane rotations cause significant reduction ter 3), who identified the first integral term as the mixing in CNR ? efficiency (now usually referred to as the heterodyne ef- P ficiency because of the presence of the detector quantum The coherent detection process requires that the efficiency spatial profile), and the second integral term as ensemble-averaged return signal optical field be matched the effective detector quantum efficiency. to the LO optical field in wavelength, direction, trans- The second integral term, the effective detector quan- verse size, curvature, and polarization. This leads to the hEFF tum efficiency QE , does not involve the signal optical question of whether CNRP can be significantly reduced field. The numerator uses the quantum-efficiency-weighted by aircraft rotations between the laser pulse emission and ~QEW (QEW) LO optical field in the detector plane, uLOD ,to the detection of the backscattered light, leading to a di- calculate the area-integrated irradiance yielding the opti- rection difference. The sensitivity of coherent lidar CNRP QEWFIELD « cal power PLOD times 2/(c 0). It is this field that to direction difference has been published (Frehlich produces the intuitive and useful, but only conceptual, time- 1994). For GRIP flights, our maximum signal range was reversed back-propagated LO (BPLO) outside the lidar 16 270 m. This corresponds to a round-trip time of light of system. The denominator uses the quantum-efficiency- 108.5 ms. Using the DAWN laser wavelength and receiver weighted LO irradiance in the detector plane and area mirror diameter, a 3-dB loss of CNRP is caused by a di- averaging to calculate the ensemble-averaged power rection difference of 10.3 mrad. The airplane rotation rate QEWIRRADIANCE « PLOD times 2/(c 0). It is this weighted irra- that would create this direction difference for the maxi- 2 diance that produces the detection shot noise, the domi- mum range signal is 5.48 s 1. Normal-level flight aircraft hEFF nating noise source. The value of QE does not change if rotation rates are much below this amount. However, ~ uLOD is multiplied by a constant, and so it is indepen- during aircraft turns, a reduction of CNRP is expected. dent of the LO intensity. If h (x, y) 5 h , QE QE,CONSTANT i. Two figures of merit for pulsed, coherent-detection then hEFF/h . QE QE,CONSTANT wind lidar The numerator of the first integral term, which is the 21 heterodyne efficiency hH, also uses the quantum- In addition to random velocity error sV (m s ) near efficiency-weighted LO optical field in the detector the true velocity, there is a second and more important ~QEW plane uLOD . It then calculates the magnitude squared figure of merit (FOM) for pulsed, coherent lidar mea- of the area-integrated product of this weighted field surement of wind. This second FOM is the probability with the received signal optical field in the detector PG that the velocity estimate is obeying the error sta- plane. This quantity is proportional to the detector output tistics and is near to the true value—that is, it is ‘‘good’’ electrical power at the intermediate frequency. The de- (Frehlich and Yadlowsky 1994). When one is finding the QEWFIELD nominator normalizes the numerator by PLOD times frequency of the signal peak in the frequency domain, it is 2/(c«0), and by the ensemble-averaged optical power of easy to visualize that occasionally one or more noise the signal field in the detector plane PSD(t) times 2/(c«0). If peaks will be higher than the signal peak. In this case the h ~ 5 ~ QE(x, y)uLOD(x, y) KuSD(x, y, t), with K aconstant, frequency of the selected highest noise peak has no re- h 5 then the heterodyne efficiency H 1. The heterodyne lationship to the true wind value, and we say that the efficiency does not change if u~SD or u~LOD or both are velocity estimate is ‘‘bad.’’ Bad estimates are uniformly multiplied by constants, and so it is independent of the distributed across the detection bandwidth. h 5 h signal and LO powers. If QE(x, y) QE,CONSTANT, As postdetection CNRP decreases, the random veloc- then the heterodyne efficiency becomes independent ity error of the good wind estimates increases, typically 834 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 31

FIG. 3. Depiction of an ensemble-averaged, single-RG, shot-averaged periodogram. going from 20% to about 90% of the backscattered signal length of the range gate. This is an estimate of the signal’s spectrum width as CNRP decreases by several orders of ensemble-averaged power spectrum. [A general signal’s 2 magnitude (Frehlich et al. 1997). In most cases, this re- x(t) ‘‘statistical power’’pffiffiffiffiffi is considered to be jx(t)j (W), so mains a low velocity error measurement. However, the the unitspffiffiffiffiffi of x(t)are W and the units of the DFT of x(t) more dramatic effect is the simultaneous transition of PG are Ws. The periodogram–power spectrum then has 2 21 21 from 100% to a very small number. The change of PG units W s s 5 Ws5 WHz 5 J, so areas under the from 99% to 50% can occur with only a factor of 4 drop in periodogram have units J Hz 5 W].

CNRP. The location of the PG ‘‘cliff’’ compared to the We accumulate NL laser shots at each laser beam changing value of sV depends on the shot accumulation direction. This shot accumulation permits wind mea- number (Frehlich et al. 1997; Frehlich 2004). surements to have a higher success rate in lower con- The goal of the flights reported below was to measure centrations of aerosols. We could also accomplish this wind profiles and not to determine the values of PG and sensitivity increase by accumulating adjacent range sV. However, the theory has been confirmed using gates, but this would degrade the range resolution. The ground-based measurements with a different lidar sys- NL periodograms for a specific LOS direction are reg- tem (Frehlich et al. 1997). istered to each other along the frequency axis by using the monitor detector in Fig. 1 to label each laser shot j. Frequency domain signal processing with the frequency difference between the injection and The coherent Doppler lidar system is designed to mea- pulsed lasers, the pulsed laser’s frequency jitter. The sum sure the change in light frequency between the transmitted of the frequency-registered periodograms divided by NL laser pulse and the backscattered (received) light (see is a shot-averaged periodogram. The standard deviation Figs. 1 and 2). These frequency differences are then used of the averaged periodogram estimate at each frequency in Eqs. (6) and (10). The received signal is digitized and is approximately equal to the true power spectrum value processed in the frequency domain with a computer as at that frequency divided by the square root of NL. The discussed in section 2a. For the wind measurement results DAWN receiver generates real values for the ADC, so presented below, the periodogram peak finding approach the negative frequency values of the DFT output simply to frequency estimation was employed. A discrete Fourier repeat the positive frequency values. The negative fre- transform (DFT) is performed by a computer on each quency values contribute no additional information and range gate’s set of NRG samples. A periodogram is then we ignore them here. Figure 3 depicts the ensemble- calculated by taking the square of the magnitude of the averaged, single-range-gate, shot-averaged, positive complexDFTandthendividingbyNRG 3 TADC,thetime frequency portion of a periodogram. APRIL 2014 KAVAYAETAL. 835

The positive portion of the DFT–periodogram output TABLE 1. Pulsed laser parameters for DC-8 flights. shown in Fig. 3 has values every DfDFT 5 fADC/NRG 5 2 Crystal Ho:Tm:LuLiF (500 Msample s 1)/(512 samples) 5 0.977 MHz, going Wavelength l 2.053 472 mm from direct current (dc) to fMAX 5 fADC/2 5 250 MHz. 5 1 Architecture Master oscillator power amplifier There are NRG/2 256 frequency intervals and NRG/2 Number of amplifiers One, single pass 1 5 257 frequency values. The periodogram ensemble- Cavity Ring, 3.1 m long 21 averaged noise level is LN (W Hz ), the noise plus Free spectral range 96.7 MHz Pumping source LDA signal level is LD, and the level of the signal is LS 5 LD 2 L . Assuming detector bias and amplification elec- LDA WAVELENGTH 792 nm N LDA pulse length 1 ms tronics are linear, the area under the signal peak but Pumping geometry Side, 1208 spacing around above the mean noise level is proportional to the signal laser rod power S (W). The signal peak’s width wS (Hz) may be Injection seeding technique Ramp and fire LDA cooling Conductive smaller or larger than DfDFT. It is caused by factors such as the laser pulse duration, the range gate length com- Laser rod cooling Liquid (water) Pulse energy, U 250 mJ bined with the atmospheric wind shear and wind tur- L Pulse rate, fL 10 Hz bulence, laser-pointing jitter during shot accumulation, Pulse duration, tL 180 ns and error in registering the periodograms in frequency. Pulse frequency spectrum Approximately transform It is this latter contribution that leads us to zero pad our limited 2 , 512 sample range gates out to 2048 samples, which Pulse spatial quality M 1.2 Output polarization Linear, parallel to bench provides a frequency spacing of 0.244 MHz for better Faraday isolators Free space periodogram frequency registration. Design and fabrication NASA LaRC For simplicity, we now assume that wS 5DfDFT in the following, so that S 5 LSDfDFT. The wideband (WB) noise is the entire area under the LN line, namely, NWB 5 subsystems. The lasers for DAWN comprise a pulsed laser LNfADC/2 (W). The wideband CNRP 5 CNRP,WB 5 and a CW laser. The optics may be grouped as small versus S/NWB 5 (LSDfDFT)/(LNfADC/2) 5 (LS/LN)(2/NRG). The large optics, or as transmit path, receive path, and both narrowband (NB), ‘‘matched filter’’ noise is the area un- paths. The electronics include power supplies, control der the LN line and under the signal, NNB 5 LNfADC/NRG. circuits, heat removal components, and data acquisition The narrowband CNRP 5 CNRP,NB 5 S/NNB 5 equipment. The data acquisition electronics perform the (LSDfDFT)/(LNfADC/NRG] 5 LS/LN 5F. Referring functions of acquiring, processing, displaying, and ar- to Fig. 3, the narrowband result is a better figure of merit chiving the data. The term lidar transceiver refers to the for the ability to find the signal peak in the frequency natural subassembly of the lasers, small optical compo- domain. The ratio of the two is CNRP,NB/CNRP,WB 5 nents, and the receiver that includes the optical detectors. NRG/2, which is typically two orders of magnitude. The The large optics transform the transceiver’s output laser Frehlich parameter F was recently defined (Frehlich and beam into a larger beam for transmitting into the atmo- Yadlowsky 1994) and was interpreted as the average sphere, aiming the laser beam in the desired direction and number of ‘‘coherently detected’’ photoelectrons per receiving the return signal. observation time. a. Pulsed laser One way to make more good wind estimates for

ﬁxed CNRP is to narrow the region of looking for the The 2-mm solid-state, pulsed laser is the heart of the signal peak to a frequency interval fSB smaller than coherent lidar system. It is composed of lasing media, fADC/2. This is accomplished by using contextual in- laser-pumping components, small optics to make the formation and frequency estimates from nearby range resonator, and electronics to provide power and to con- gates. This ‘‘search band’’ (SB) CNRP 5 CNRP,SB 5 trol the laser. Our pulsed laser has been developed at the (LSfADC/NRG)/(LNfSB) 5 (LS/LN)(fADC/NRG)(1/fSB) 5 NASA Langley Research Center (LaRC) over the last CNRP,WB[(fADC/2)/fSB]. two decades (Jani et al. 1995, 1997; Yu et al. 1998; Walsh et al. 2004; Yu et al. 2006). During this time, both the requirements on the pulsed laser parameters and their 3. Lidar system hardware priorities for the future space mission have changed The components of a pulsed, coherent-detection, wind slightly due to wisdom gained in space instrument and lidar system may be grouped in several ways. The lidar mission studies. Laser requirements include pulse energy, system consists of lasers, optics, electronics, mechanical pulse rate, and pulse duration. The relative priorities that structure, and software. Figure 1 shows some of these changed over time were between pulse energy, mass, 836 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 31

TABLE 2. Lidar system parameters. TABLE 3. Data acquisition parameters.

Transmit–receive Off axis, reﬂective, afocal Receiver analog 10–180 MHz telescope category bandwidth Telescope type Dall–Kirkham, beam expanding LOS frequency capture 680 5 160 5 20–180 MHz Telescope primary Concave elliptical, 15-cm bandwidth 21 mirror, DM diameter LOS velocity capture 682.1 5 164.3 m s Telescope secondary mirror Convex spherical bandwidth 2 Telescope magniﬁcation 20 Horizontal velocity 6163 5 327 m s 1 56366 mph Detection Coherent (heterodyne) capture bandwidth 56318 kt Coherent detection Dual balanced, reverse biased ADC sample bits 10 bits per sample 6 21 arrangement ADC sample rate, fADC 500 3 10 samples s Signal optical detector InGaAs 32, room temperature, ADC sample interval 2 ns, 0.299 79 m in range, 75-mm diameter 0.259 17 m in height Signal optical detector Alternating current Total recorded 55 000/54 999, 16 488.3 m in coupling samples/intervals range, 14 262.0 m in height Monitor optical detector InGaAs, room temperature Prelaser shot sample 1–512 Scanner type Rotating optical wedge numbers Scanner motor aperture 17.5 cm Monitor detector sample 1–1024 Scanner motor mechanical Yes numbers

home function Number RG samples NRG 512 Scanner wedge material Single crystal silicon, n 5 3.45 Signal detector sample 1025–55 000 Scanner wedge diameter 15 cm numbers Scanner wedge angle 11.0158 Number signal samples 53 976 Scanner wedge beam ;30.128 Start, center, and end 1025 1 512(n 2 1); 1280.5 1 deﬂection uS sample number of 512(n 2 1); 1536 1 512(n 2 1); Lidar optics canister Fused silica, 31.75-cm RG n (e.g., n 5 1) (1025/1280.5/1536) pressure seal window diameter, 2-in. thick Start, center, and end [1025 1 512(n 2 2] 0.299 79 m; Aircraft pressure seal Fused silica, 43.2-cm range of RG n [1280.5 1 512(n 2 2)] 0.299 79 m; window diameter, 2-in. thick (e.g., n 5 1) [1536 1 512(n 2 2)] 0.299 79 m; Eye safety Eye safe at any range when (153.8/230.4/307.0 m) canister closed for ﬂight RG length without 153.2 m in range, 132.5 m in height INS–GPS Hard mounted to lidar system; laser pulse provides 3D orientation in RG length including 180.2 m in range, 155.8 m in height

NED and 3D velocity in ENU laser pulse, RRES Number consecutive 105 5 53 760 samples (216 samples

NRG RG possible not used) Start, center, and end 16 117.1, 16 193.7, 16 270.3 m volume, electrical efﬁciency, spatial beam quality, spec- range of RG 105 2 tral purity, shot lifetime, and component count. The LaRC Maximum periodogram 250 MHz, 6125 MHz, 6128.3 m s 1 21 laser development team used parallel development efforts, frequency, fMAX LOS, 6255.8 m s horizontal 2 concentrating on logical subsets of the requirements, to Periodogram frequency 0.976 562 5 MHz, 1.00 m s 1 range, 21 make faster progress. In the area of pulse energy, the spacing without zero 2.00 m s horizontal padding, DfDFT LaRC team increased the demonstrated energy from 2 Periodogram frequency 0.244 140 6 MHz, 0.25 m s 1 range, 2 20 mJ to 1.2 J (Yu et al. 2006). For the ﬂights reported spacing after zero 0.50 m s 1 horizontal here, the laser parameters are provided in Table 1. padding to 2048 Typical RG overlap 256 samples b. Lidar system Number of overlapped 209 5 53 760 samples (216 samples RG not used) The entire lidar system consists of the pulsed laser, continuous-wave (CW) laser and the other subsystems named above. Measurements from the ground in a mo- bile trailer have been described (Koch et al. 2007, 2010, Table 3 presents the data acquisition parameters used 2012). Table 2 presents additional parameters of the li- in the GRIP campaign. The signal frequency is permit- dar system. The lidar scanner is a step-stare conical ted to fall anywhere in the range 10–180 MHz. If the scanner, permitting any laser beam direction that lies on guard band from dc was smaller than 10 MHz, then the the surface of a cone extending below the airplane and analog electronics would be more complex. The ADC having a 30.128 half angle. The telescope is off axis to sample rate must be at least twice the upper frequency, 2 2 prevent backscattering of a large fraction of the laser pulse or 360 Msample s 1. We used 500 Msample s 1 to be into the laser or detectors. The telescope is afocal to pre- well above the Nyquist frequency. The DAWN wind vent arc breakdown of the atmosphere by the laser pulse. velocity algorithm found the frequency of the peak APRIL 2014 KAVAYAETAL. 837

FIG. 4. Optical frequencies layout of the DAWN lidar system. value in the shot-accumulated periodograms. Our wind sent directly to optical detector 2 (signal detector) and to measurements had line-of-sight (LOS) resolutions of optical detector 1 (monitor detector) after frequency 21 180 m and 1.0 m s . The vertical resolution was 156 m. upshifting by fAOM 51100 MHz. The AOM frequency 2 The horizontal wind resolution was 2.0 m s 1, ampliﬁed shift permits the sign of the jitter frequency to be deter- 2 from 1.0 m s 1 due to the nadir angle. Specialized signal mined. The light backscattered by aerosol particles at processing algorithms to improve data processing per- frequency nR (as a function of range) is Doppler shifted by formance have also been investigated (Beyon and Koch the projections onto the laser beam direction of the air-

2006a; Beyon et al. 2006; Beyon and Koch 2006b, 2007). craft velocity fAC and wind velocity fW (see also Fig. 2). The block diagram of the optical part of the lidar sys- Both optical detectors are used to heterodyne two tem used to track frequencies is shown in Fig. 4.The sources of light. The heterodyne process causes the fre- stable seed laser provides injection seeding for the pulsed quency of the detector output current to be at the abso- laser at frequency nSEED. The laser pulse frequency nL lute value of the difference frequency between the two differs from the seed laser frequency by a positive or light sources. The lhs of Eqs. (6) and (10) is found in the negative jitter frequency fJITTER. The seed laser is also following way:

5 j n 1 2 n 1 j 5 2 fDET1 ( SEED fAOM) ( SEED fJITTER) fAOM fJITTER . . j j assuming fAOM 0 and fAOM fJITTER 5 j n 2 n 1 1 1 j 5 1 1 fDET2 ( SEED) ( SEED fJITTER fAC fWIND) fJITTER fAC fWIND . . j 1 j assuming fAC 0 and fAC fJITTER fWIND n 2 n 5 1 5 1 2 R T fAC fWIND fDET1 fDET2 fAOM . (13)

Equation (13) shows that the lhs of Eqs. (6) and (10) are c. Calculation of wind velocity the sum of the two detector output frequencies minus the

AOM frequency, that is, minus 100 MHz. Note that fDET1 For the ﬂights reported here, a scan pattern consisted does not change with the range gate for a given laser di- of uS ; 30.128 and ﬁve directions fS 52458, 222.58,08, rection and shot, and that fAOM does not change with the 22.58, and 458. We employed land backscattered signals range gate or the laser shot or the laser direction. to determine the mounting offset angles of DAWN in 838 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 31

TABLE 4. DAWN hardware list and DC-8 locations.

Location Component Subcomponent Comments Roof GPS antenna DC-8 supplied Passenger level Laser chillers Three each in 19-in. rack Laser operator station Electronics in 19-in. rack Data acquisition operator station Electronics in 19-in. rack Cargo level Optics canister Injection laser CW Pulsed laser Optical detector Two each Telescope Scanner Canister window Laser diode drivers Three each Laser control electronics Analog signal electronics Scanner driver Power distribution circuit Belly Pressure window Window port cover Motorized; DC-8 supplied the DC-8. Our flight paths over water limited the avail- direct communication. The DC-8 required the window able land returns to islands. We used the calculated offset port cover to be closed for takeoff and landing. The laser angles and our model for the mounting errors to find the operator only started the laser after the port was open. five pairs of laser angles outside the airplane. These five The data acquisition operator occasionally stopped the directions were then modified by the three rotation an- software in flight, changed the number of laser shots ac- gles of the airplane to obtain the laser beam directions in cumulated, and restarted the software. The stopping of the NED coordinate system. Finally, the directions were the software was done both to match atmospheric condi- converted to ENU coordinates and inserted into two or tions and to learn about the DAWN performance. It was five copies of Eq. (6) for solving. We report only two not done near a high science priority target, and only a few azimuth angle results in this paper. minutes of data were lost when it was done. The DC-8 supplied electricity, the GPS antenna, and the motorized d. DC-8 accommodation port cover. The DAWN lidar components were accommodated in both the passenger and cargo levels of the DC-8. The hardware list and the locations in the DC-8 are provided 4. Preliminary flight results in Table 4. The pressure window at the skin of the air- The NASA GRIP campaign involved the flights of craft is required. It must have a high transmission and three aircraft in August–September 2010 (Braun et al. a minimum polarization effect on the 2-mm laser beam 2013). The DAWN lidar flew on the DC-8. Of the 25 passing through it at ;308. It must also not deform under total DC-8 flights, 15 were science flights to areas of the pressure difference to the extent of causing optical possible hurricane development, developing storms, and aberrations to the laser beam. The optics canister also mature hurricanes. Table 5 provides some statistics of has a pressure window that is currently used to keep dust DAWN’s participation in GRIP. and dirt out of the canister. In the future, it may provide a pressure seal for flight in an unpressurized compart- ment of a different aircraft. Having the laser chillers on the passenger level and the lasers on the cargo level re- TABLE 5. Statistics of DAWN during GRIP. quired routing the cooling water between levels. This Science flights All flights caused no problems. We began the GRIP flights with Number of flights 15 25 three operator stations in the passenger level. During DAWN scan patterns* 11 685 13 062 GRIP, the ability to control the laser chillers was imple- DAWN laser shots 2 058 520 2 243 620 mented at the laser operator station, which permitted only DAWN emitted photons** 5.3 3 1024 5.8 3 1024 two operators with no requirement to move around in the * Each scan pattern yields a vertical profile of horizontal vector airplane. The laser operator and data acquisition operator wind. were in adjacent seats to permit both headphone and ** Assuming 250-mJ laser pulses. APRIL 2014 KAVAYAETAL. 839

FIG. 5. Comparison of DAWN lidar (dots) and dropsonde (line) showing both (left) horizontal wind speed (HWS) and (right) HWD.

The DC-8 total flight time was 6712 min, and DAWN significantly lowered our CNRP due to photon loss collected wind data for 5987 min or 89%. Of the 5987 combined with heterodyne efficiency degradation. DAWN data minutes, 4634 min or 95% have been A second problem was the reduction of laser pulse en- processed and uploaded to the GRIP website (http:// ergy, as the cold DC-8 cargo section cooled the air airbornescience.nsstc.nasa.gov/grip/). Each DAWN around our laser optics. We mitigated this in real time by scan pattern consisted of five azimuth angles, and the adjusting laser chiller temperatures, but offsetting the number of laser shots per azimuth angle was varied mitigation was lower laser energy and some receiver between many different values from 1 to 300. The three misalignment caused by changing chiller temperatures. most common values of shot accumulation were 20, 60, A third problem was faulting by the laser diode array and 120. With a laser pulse rate of 10 Hz, these shot power supplies. This forced a laser restart each time and accumulation values have data collection times of 2, 6, a small amount of data (minutes) was lost. The vendor and 12 s per azimuth angle, or 10, 30, and 60 s per scan later fixed this problem. Finally, our initial airplane ve- pattern, respectively. The scanner azimuth change time locity removal equations used during the flights were not was 2 s or so, and the computing time after each scan correct, which prevented real-time knowledge of the pattern was 2 s or so. The minimum scan pattern time wind, but they did not corrupt the stored data. The total was therefore 22, 42, and 72 s, respectively. With a typical CNRP losses may have exceeded 10 dB, and our data 2 DC-8 ground speed of 450 kt 5 231.5 m s 1(1 kt 5 appear as if our laser had much lower pulse energy than 2 0.51 m s 1), the minimum scan pattern horizontal reso- 250 mJ. lutions were 5.1, 9.7, and 16.7 km, respectively. These are Preliminary data processing has used the outer two the repeat distances for the vertical profile of horizontal azimuth angles only from the scan patterns. Since GRIP wind—basically the same wind product as a conventional represented the first flights of DAWN, we were eager dropsonde. to compare DAWN winds with dropsonde winds. The cross-track horizontal resolution depends on the Figure 5 shows an example of this intercomparison from distance below the DC-8. Immediately below the DC-8, 1 September 2010, when the DC-8 was at 10 611-m the resolution is the 15-cm telescope aperture size. (34 813 ft) altitude. Laser shot accumulation was set to (The lidar does not obtain measurements for about 20. The DAWN single-scan pattern shown lasted from a 100-m distance from the airplane.) For the DC-8 flying 1720:00 to 1720:33 UTC, or 33 s. Data collection time for at 10.6-km altitude, the resolution at the earth surface is two azimuth angles was 4 s. The dropsonde was launched about 8.7 km. at coordinates (275.755 7118, 29.959 8888) at 1720:15 Several DAWN hardware problems were encoun- UTC and splashed in the ocean at coordinates tered during GRIP. The worst problem was a burn of the (275.833 5558, 29.968 8558) at 1733:31 UTC, which is telescope secondary mirror that was not discovered until slightly over 13 min later and 7.6 km away at heading after GRIP, and after deintegration from the DC-8. We 277.68. During the DAWN scan pattern, the DC-8 flew believe the burning is caused by dust settling on the 7.1 km at track angle 146.18. The dropsonde and DC-8 mirror and being burned by the pulsed laser. The burn travel angles differed by about 1328. During the fall of apparently occurred before the DC-8 flights, and it the dropsonde, the DC-8 flew 172.2 km. The agreement 840 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 31

FIG. 6. Panorama display of DAWN wind data from 0 to 12 km on 1 Sep 2010 showing slightly less than 4 h of data: color-coded (top) 2 2 signal energy (220 to 15 dB in 5-dB increments), (middle) wind speed (0–25 m s 1 in 5 m s 1 increments), and (bottom) wind direction (08– 3608in 508 increments.) between DAWN and the dropsonde is very good at DC-8 ﬂew over the eye of Hurricane Earl at approxi- altitudes just below the DC-8, and then at 0–4 km. The mately 1912:00 UTC. This time is about in the center good low-altitude wind data eliminate the possibility of of Fig. 6, where the wind direction rapidly changes from thick clouds, and we attribute the altitudes without blue to yellow, or by about 1808. The top panel shows wind measurements as due to our CNRP losses com- the laser beam reaching to lower altitudes within bined with the very low aerosol content ocean air. The the eye, and the middle panel shows that some wind agreement shown is not typical for our GRIP drop- magnitudes were measured at lower altitudes within sonde intercomparisons due to the combination of the eye. frequent thick clouds, clean ocean air, and our CNRP A third type of display of wind velocities is the lon- losses. The areas of excellent agreement in Fig. 5 gitude versus latitude ‘‘top view’’ display, as shown in demonstrate the fundamentally correct operation of Fig. 7. The data are from 21 September 2010, and the DAWN, and also indicate a stable wind ﬁeld during the DAWN lidar was shot accumulating 60 laser shots. The fall time of the dropsonde and over the ;10-km mea- unnamed storm did not become a hurricane or even a surement separation. tropical storm. Approximately 202 min of data are shown. Another display of wind data is the time–altitude Winds at altitudes of 2, 8, 9, 10, and 11 km are plotted. 2 ‘‘panorama’’ display shown in Fig. 6, showing slightly A10ms 1 wind magnitude reference vector is provided less than 4 h of continuous data from 1 September 2010. for scale. For the most part, the wind direction at dif- The laser shot accumulation was again 20. The areas ferent altitudes is approximately the same (easterly). The with only high-altitude wind measurements most likely low-altitude-only winds away from the storm are in- contained thick clouds. The areas with low-altitude or dicative of the clean ocean air and our CNRP losses. As multi-altitude wind measurements most likely are free the storm is approached, the thick clouds begin and the of thick clouds. It is usually the case that winds are lidar winds transition to the higher-altitude-only winds successfully measured near the aircraft altitude. The above the clouds. APRIL 2014 KAVAYAETAL. 841

FIG. 7. Mapping of horizontal vector winds at multiple altitudes (color coded) vs longitude and latitude.

5. Conclusions Technology Office; NASA Science Mission Directorate, Earth Science Division; NASA Langley Research Cen- We have described results of the first flights of an ter; NASA Langley Research Center, Engineering Di- airborne, 2-mm, high-energy, coherent-detection, wind- rectorate; and NASA Langley Research Center, Chief profiling lidar system developed at the NASA Langley Engineer’s office. Research Center. The combination of pulse energy, pulse We appreciate the technical assistance of R. Atlas, Y. Bai, rate, receiver diameter, and coherent detection have the B. Barnes, F. Boyer, J. Campbell, S. Chen, M. Coleman, potential to provide much improved sensitivity to aerosol L. Cowen, J. Cox, G. Creary, J. Cronauer, D. Emmitt, backscatter levels, and therefore greater coverage of the F. Fitzpatrick, P. Gatt, M. Grant, M. Jones, N. Massick, atmosphere. The long-term motivation is to measure E.Modlin,A.Noe,D.Oliver,K.Reithmaier,G.Rose, wind velocities from space for improved weather pre- A. Webster, T. Wong, and W. Wood. We appreciate diction and severe weather disaster mitigation. The in- the suggested improvements to the manuscript by Sammy terim steps of developing and testing the technology, and W. Henderson and the anonymous reviewers. then the initial validation of the technology with aircraft This paper is dedicated to the memory of Dr. Rodney flight, have been described. The lidar system was flown on George Frehlich, 7 August 1952–4 May 2012, a great the DC-8 during a NASA 2010 hurricane research flight scientist and friend. campaign. Although hardware problems reduced the

CNRP, the lidar successfully measured winds throughout the campaign. Flights of the lidar on a smaller UC-12B REFERENCES airplane at Langley have already begun with the purpose Ashby, N., 2007: Relativistic effects in Earth-orbiting Doppler lidar of optimizing the lidar and developing a greater un- return signals. J. Opt. Soc. Amer., 24A, 3530–3546, doi:10.1364/ derstanding of its characteristics. The Doppler lidar is JOSAA.24.003530. generally available to collect wind data as a contribution Baker, W. A., and Coauthors, 1995: Lidar-measured winds from space: to airborne science campaigns. A key component for weather and climate prediction. Bull. Amer. Meteor. Soc., 76, 869–888, doi:10.1175/1520-0477(1995)076,0869: LMWFSA.2.0.CO;2. Acknowledgments and dedication. The authors Beyon, J. Y., and G. J. Koch, 2006a: Resolution study of wind pa- acknowledge support from NASA Science Mission rameter estimates by a coherent Doppler lidar system. Laser Directorate, Earth Science Division, Earth Science Radar Technology and Applications XI, G. W. Kamerman and 842 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 31

M. D. Turner, Eds., International Society for Optical Engi- ——, N. Barnes, K. E. Murray, D. W. Hart, G. J. Quarles, and

neering (SPIE Proceedings, Vol. 6214), 621403, doi:10.1117/ V. K. Castillo, 1997: Diode-pumped Ho:Tm:LuLiF4 laser at 12.663416. room temperature. IEEE J. Quantum Electron., 33, 112–115, ——, and ——, 2006b: Wind profiling by a coherent Doppler lidar doi:10.1109/3.554905. system VALIDAR with a subspace decomposition approach. Kavaya, M. J., and R. T. Menzies, 1985: Lidar aerosol back- Signal and Data Processing of Small Targets 2006, E. E. scatter measurements: Systematic, modeling, and calibra- Drummond, Ed., International Society for Optical Engi- tion error considerations. Appl. Opt., 24, 3444–3453, doi:10.1364/ neering (SPIE Proceedings, Vol. 6236), 623605, doi:10.1117/ AO.24.003444. 12.663423. Kingston, R. H., 1978: Detection of Optical and Infrared Radiation. ——, and ——, 2007: Novel nonlinear adaptive Doppler-shift es- Springer Series in Optical Sciences, Vol. 10, Springer-Verlag, timation technique for the coherent Doppler validation lidar. 140 pp. Opt. Eng., 46, 016002, doi:10.1117/1.2434768. Koch, G. J., J. Y. Beyon, B. W. Barnes, M. Petros, J. Yu, ——, ——, and Z. Li, 2006: Noise normalization and windowing F. Amzajerdian, M. J. Kavaya, and U. N. Singh, 2007: High- functions for VALIDAR in wind parameter estimation. Laser energy 2mm Doppler Lidar for wind measurements. Opt. Radar Technology and Applications XI, G. W. Kamerman and Eng., 46, 116201, doi:10.1117/1.2802584. M. D. Turner, Eds., International Society for Optical Engi- ——, and Coauthors, 2010: Field testing of a high-energy 2-mm neering (SPIE Proceedings, Vol. 6214), 621404, doi:10.1117/ Doppler lidar. J. Appl. Remote Sens., 4, 043512, doi:10.1117/ 12.663420. 1.3368726. Braun, S. A., and Coauthors, 2013: NASA’s Genesis and Rapid ——, J. Y. Beyon, E. A. Modlin, P. J. Petzar, S. Woll, M. Petros, Intensification Processes (GRIP) Field Experiment. Bull. J. Yu, and M. J. Kavaya, 2012: Side-scan Doppler lidar for Amer. Meteor. Soc., 94, 345–363, doi:http://dx.doi.org/10.1175/ offshore wind energy applications. J. Appl. Remote Sens., 6, BAMS-D-11-00232.1. 063562, doi:10.1117/1.JRS.6.063562. Frehlich, R. G., 1994: Heterodyne efficiency for a coherent laser Menzies, R. T., and D. M. Tratt, 1997: Airborne lidar observations radar with diffuse or aerosol targets. J. Mod. Opt., 41, 2115– of tropospheric aerosols during the Global Backscatter Ex- 2129, doi:10.1080/09500349414551991. periment (GLOBE) Pacific circumnavigation missions of 1989 ——, 2004: Velocity error for coherent Doppler lidar with pulse and 1990. J. Geophys. Res., 102, 3701–3714, doi:10.1029/ accumulation. J. Atmos. Oceanic Technol., 21, 905–919, 96JD03405. doi:10.1175/JTECH1596. NASA, 2006: Working group report. J. Neff and A. Valinia, Eds., ——, and M. J. Kavaya, 1991: Coherent laser radar performance Lidar Technologies, 171 pp. [Available online at http://esto. for general atmospheric refractive turbulence. Appl. Opt., 30, nasa.gov/files/LWGReport2006.pdf.] 5325–5352, doi:10.1364/AO.30.005325. NRC, 2007: Earth Science and Applications from Space: National ——, and M. J. Yadlowsky, 1994: Performance of mean-frequency Imperatives for the Next Decade and Beyond. The National estimators for Doppler radar and lidar. J. Atmos. Oceanic Academies Press, 456 pp. Technol., 11, 1217–1230, doi:10.1175/1520-0426(1994)011,1217: Rye, B. J., 1979: Antenna parameters for incoherent backscatter POMFEF.2.0.CO;2. heterodyne lidar. Appl. Opt., 18, 1390–1398, doi:10.1364/ ——, S. M. Hannon, and S. W. Henderson, 1997: Coherent Doppler AO.18.001390. lidar measurements of winds in the weak signal regime. Appl. ——, and R. G. Frehlich, 1992: Optimal truncation and optical Opt., 36, 3491–3499, doi:10.1364/AO.36.003491. efficiencyofanaperturecoherentlidarfocusedonanincoherent Gagliardi, R. M., and S. Karp, Eds., 1976: Optical Communications. backscatter target. Appl. Opt., 31, 2891–2899, doi:10.1364/ Wiley-Interscience, 432 pp. AO.31.002891. Goldstein, H., C. P. Poole Jr., and J. L. Safko, 2001: Classical Shapiro, J. H., and S. S. Wagner, 1984: Phase and amplitude un- Mechanics. 3rd ed. Addison-Wesley, 680 pp. certainties in heterodyne detection. IEEE J. Quantum Elec- Gudimetla, V. S. R., and M. J. Kavaya, 1999: Special relativity tron., 20, 803–813, doi:10.1109/JQE.1984.1072470. corrections for space-based lidars. Appl. Opt., 38, 6374–6382, Singh, U. N., S. Ismail, M. J. Kavaya, D. M. Winker, and doi:10.1364/AO.38.006374. F. Amzajerdian, 2005: Space-based lidar. Laser Remote Sensing, Henderson, S. W., P. Gatt, D. Rees, and R. Milton Huffaker, 2005: T. Fujii and T. Fukuchi, Eds., CRC Taylor & Francis, 781–881. Wind lidar. Laser Remote Sensing, T. Fujii and T. Fukuchi, Taff, L. G., 1983: Relativity and radars. MIT/LL Tech. Rep. 660, Eds., CRC Taylor and Francis, 469–722. AD-A133791, 22 pp. Huffaker, R. M., T. R. Lawrence, M. J. Post, J. T. Priestly, F. F. Hall Walsh, B., N. Barnes, M. Petros, J. Yu, and U. Singh, 2004: Spec- Jr., R. A. Richter, and R. J. Keeler, 1984: Feasibility studies for troscopy and modeling of solid state lanthanide lasers: Ap- 31 31 a global wind measuring satellite system (Windsat): Analysis of plication to trivalent Tm and Ho in YLiF4 and LuLiF4. simulated performance. Appl. Opt., 23, 2523–2536, doi:10.1364/ J. Appl. Phys., 95, 3255, doi:10.1063/1.1649808. AO.23.002523. Yu, J., U. Singh, N. Barnes, and M. Petros, 1998: 125-mJ diode- Jani, M. G., F. L. Naranjo, N. P. Barnes, K. E. Murray, and G. E. pumped injection-seeded Ho:Tm:YLF laser. Opt. Lett., 23, Lockard, 1995: Diode-pumped long-pulse-length Ho:Tm: 780–782, doi:10.1364/OL.23.000780. YLiF4 laser at 10 Hz. Opt. Lett., 20, 872–874, doi:10.1364/ ——, and Coauthors, 2006: 1 J/pulse Q-switched 2 mm solid-state OL.20.000872. laser. Opt. Lett., 31, 462–464, doi:10.1364/OL.31.000462.