1. Introduction

Several mid-infrared laser-based absorption spectrometers based upon the promising technology of difference frequency generation (DFG) have been demonstrated over the past few years [1–9]. To date, DFG developments dedicated for Airborne atmospheric applications have only been pursued by our group [6–9]. In our laboratory, this new laser technology will eventually replace existing airborne spectrometers utilizing liquid nitrogen cooled tunable diode lasers (TDLs) to measure the important trace gas formaldehyde (CH₂O), a significant hydrogen radical precursor and the most abundant aldehyde found in the atmosphere. Formaldehyde plays a critical role in hydrocarbon oxidation processes in the lower atmosphere, and as such is a key species in testing photochemical models of these processes [10–13].

As discussed by Weibring et al. [8], DFG technology has the potential to significantly improve upon TDL measurements of trace gases in general, and CH₂O in particular, during airborne operation. Specific advantages of DFG systems in this regard include: 1) readily available room-temperature single frequency near-infrared pump and signal laser sources; 2) eliminating the need for cryogenic cooling and the associated issues of temperature cycling and concerns about changing laser wavelength and tuning characteristics; 3) eliminating the need for frequent dewar pump-out to reduce the deleterious effects of scattering from condensed dewar contaminants [14]; 4) improved mid-IR laser beam quality, which improves the IR-beam coupling and transmission through the entire optical system, reduces scattering and its associated optical noise, and minimizes the optical alignment effects by cell distortions caused by differential pressure changes across the cell; 5) allows for close coupling of the optical collection element with the DFG source, which simplifies the transfer optics from the laser source to the multipass absorption cell and reduces required beam magnification, both of which improve the system opto-mechanical stability; 6) eliminating the need for a large and heavy dewar system containing liquid nitrogen, which presents challenges in maintaining robust opto-mechanical laser alignment stability, particularly when one experiences cabin pressure changes and/or aircraft accelerations; 7) allows for direct mounting of the DFG stage directly to the absorption cell, which also improves beam pointing stability; 8) allows one to place all the optical components in a pressure-controlled environment, eliminating beam steering from pressure-induced flexures of the dewar, cell, and detector windows; and 9) allows for a significantly smaller, lighter, more compact, and more versatile multi-species detection system.

However, to date DFG measurements of atmospheric constituents have been only demonstrated in the laboratory and in selected ground-based applications [1–9]. Airborne measurements by contrast present additional challenges such as changing cabin pressures and temperatures, severe vibrations, and accelerations, all of which can affect the spectrometer performance. The present paper, to our knowledge, presents the first DFG instrument to successfully acquire measurements of a trace gas in the atmosphere on an airborne platform. This new instrument, employing second harmonic detection and sweep integration, successfully acquired ambient CH₂O measurements during three consecutive airborne field missions (MIRAGE, IMPEX and TexAQS) in rather harsh operating environments on two different turbo-propeller airplanes in 2006. This instrument is an improvement over our airborne TDL system [14], in terms of reduced weight and size, elimination of cryogens, and robustness. As we will discuss in detail in this paper, our developed DFG technology performed with sensitivity comparable to our TDL system [14]. Performance analysis during these three campaigns identified several areas of improvement, allowing for a DFG performance that will eventually exceed that of airborne TDL systems. In addition, such efforts will ultimately lead to autonomous operation and extend the number of detectable species by utilizing the capability to wavelength multiplex the DFG laser source

The instrument employed in this study is based upon the laboratory DFG spectrometer described in [7,8] but with major modifications for airborne operation. In the laboratory the DFG spectrometer achieved a typical CH₂O sensitivity of 13 pptv (A_min~4.3^*10^-7) for 60 s averages and 5 pptv (A_min~1.7 × 10^-7) for 260 s averages. Foremost in achieving this performance was a systematic effort to precisely capture the amplitude, shape, and time dependence of the wavelength modulated background structure detected at the output of the multipass absorption cell by the cell detector (CD). As discussed by Weibring et al. [8], this structure resulted from the DFG laser source and was successfully captured prior to the absorption cell with a second detector, designated as the amplitude modulation detector (AMD), with a contrast signal to noise ratio of ~1500 or better. To achieve this goal, we devoted considerable effort to minimize the time dependent differential signals between the two detectors between zero air background acquisitions. Efforts included, 1) matching of the illuminated detection areas and beam shapes on the two detectors, 2) elimination of scattered light that may not be equally captured by the two detectors; 3) accurate matching of the two detection channels in terms of ac and dc gains, including the detectors and lock-in amplifiers; 4) minimizing temperature variations in the PPLN crystal used for DFG; 5) applying active wavelength control and; 6) computer based balancing of the two detector signals. Furthermore, as our system employs second harmonic detection using phase sensitive detection, the phase of the detected signals must be closely matched in the two detection chains.

The present paper is a continuation of our laboratory DFG characterization [8], with specific emphasis on airborne performance. As mentioned, these platforms present a particularly challenging environment for instruments due to severe vibrations and large temperature and pressure variations. This is particularly true on turbo-propeller aircraft. Such perturbations affect the ability to maintain stable intensity output, precise beam pointing alignment, and polarization in optical fibers. In anticipation of such challenges our laboratory instrument was modified and enhanced in several areas, including: improvements in the mechanical stability of the transfer optics, improved balanced power matching on the two detectors using an innovative beam splitter configuration, and placement of the PPLN crystal in an improved temperature controlled environment. One particular concern was the impact of vibrations in inducing fiber noise, which was addressed by applying vibration damping to all optical modules.

This paper is organized as follows. The instrument layout and the airborne platforms are described in Sect. 2, including the laser spectrometer, the gas handling system, the electronic and the data systems. In Sect. 3, we introduce methods to characterize and track instrument noise. We discuss several examples of encountered “noise” issues during airborne operation as well as modifications that were implemented in the field to mitigate such noise issues. Sect. 4 highlights the instrument’s scientific performance in demonstrating high sensitivity, and fast time resolution measurements of formaldehyde. Finally, a summary and an outlook section concludes the paper.

2. Measurement system

The measurement system consists of a laser spectrometer, a gas handling system, an air conditioning system, an uninterruptible power system (UPS), and a computer system for data acquisition and instrument control. A schematic overview of the system components and their interconnections is shown in Fig. 1. These components are housed in two FAA certified aircraft racks developed for NCAR’s Gulfstream G-V airplane (HIAPER), with a combined weight of ~570 lbs (including. racks) and occupying 1×1.5 m² floor space. Figure 2 shows a photograph of the system in the cabin of one of NOAA’s P-3 aircraft. The outboard rack to the left contains the permeation calibration system, air compressor, pressure and flow controllers, power supplies, clean air generator and UPS. The inboard rack to the right houses a LCD monitor, service oscilloscope, control and data acquisition computer, laser spectrometer enclosure, and an air-conditioning unit. The insert shows the inlet system, including the inlet pressure controller, mounted in an access port of the aircraft hull. The laser spectrometer system is under full computer control and allows the operator to monitor both system status and all measurement parameters simultaneously from either the LCD monitor or remotely through an Ethernet/Video-link connection.

 

Fig. 1. Schematic overview of the airborne system consisting of the laser spectrometer (lasers, DFG module, Multipass cell (MPC), and detectors), the gas handling system (inlet, calibration system, vacuum pump), the air conditioning system (AC), the uninterruptible power system (UPS) and the computer system for data acquisition and instrument control.

Download Full Size | PDF

 

Fig. 2. DFG system mounted in the cabin of NOAA’s P-3 aircraft during the TexAQS mission. The rack to the left contains; permeation system, air compressor, pressure and flow controllers, power supplies, clean air generator and UPS. The rack to the right houses; operator monitor, service oscilloscope, control and data acquisition computer, spectrometer enclosure (lasers and DFG module, fiber amplifiers, multipass cell, detectors), and air-conditioning unit. The insert shows the inlet system mounted in an access port of the aircraft hull

Download Full Size | PDF

2.1 Laser spectrometer

As alluded to earlier, high sensitivity measurements become particularly challenging on airborne platforms where severe vibrations and variable sampling and cabin environmental conditions of temperature, pressure, and relative humidity are encountered. To mitigate the vibrations in the C-130/WP-3 aircraft cabins, all components in the laser spectrometer enclosure were vibration isolated with Barry mounts^TM, which damp the aircraft vibrations, turbulence, and impacts of landing. To maintain a stable temperature of all spectrometer components, the laser spectrometer enclosure is insulated with Gillfab^TM 5040 and equipped with a two stage air conditioning system, consisting of a compressor system for rough cooling and a Thermo Electric Cooler system (TEC) under PID control for fine temperature control. In combination with ten air circulation fans mounted inside the enclosure, the air conditioning unit is capable of keeping the enclosure temperature stable to within ± 0.2 °C in a cabin temperature range of +5 to +40 °C. As will be shown, such temperature stability was particularly important in the performance of the fiber amplifiers and detectors. By over pressurizing the spectrometer enclosure with dry air from a clean air generator, the enclosure sub-systems are protected from dust and moisture, even in harsh environments.

Building upon experience gained from the laboratory prototype [8], the spectrometer was improved as shown in Fig. 3. The optical layout is comprised of four modules, the laser module, the DFG module, the detection module, and a multipass cell module. The laser module is based on two fiber coupled laser sources and fiber amplifiers, which are mounted on a vibration damped base plate inside the spectrometer enclosure. The signal laser, a 1562 nm DFB diode laser, is controlled by the computer for wavelength scanning, modulation and stabilization purposes (25 Hz scan, 50 kHz modulation). The signal laser output is spliced to a commercial rare-earth-doped erbium fiber amplifier (EFA) to increase the optical power to ~500 mW. The pump laser source consists of a DFB fiber laser (1083nm), which is pumped by a Bragg grating stabilized diode laser (976 nm). The output of the fiber laser is spliced to a commercial rare-earth-doped ytterbium fiber amplifier (YFA) to increase the 1083 nm power to ~800 mW. For amplifier protection, a faraday isolator (ISO) is spliced in after the YFA. By comparison, the EFA includes a faraday isolator. The two mixing wavelengths are combined using a wavelength division multiplexer (WDM), and spliced to a single mode fiber, which connects the pump and signal beams to the DFG module, (see Fig. 3). To achieve optimal conversion in the PPLN crystal, the polarization of the fiber and the DFB lasers are adjusted by in-line polarization controllers (PC) placed in the signal and pump beam fiber trains before the WDM.

 

Fig. 3. DFG optical system mounted to the multipass cell. The components are: DL (diode laser), ISO (optical isolator), PC (polarization controller), BD (beam dump), WDM (wavelength division multiplexer), HF (hybrid fiber ball lens), FL (focusing lens), PPLN (periodically poled lithium niobate), BD (beam dump), Ge (germanium filter), IL (imaging lens), P (prism reflector), OS (optical shutter), BS (coated beam splitter), OAP (off axis parabolic focusing mirrors), RD (reference detector), CD (cell detector or sample), AMD (amplitude modulation detector), MPC (multipass cell), multipass cell front mirror (MPM) and CW (cell window). A reference cell (RC), containing pure CH₂O at low pressure precedes the RD.

Download Full Size | PDF

In the DFG module the pump and signal beams are launched into the PPLN crystal (maintained at 40 °C) by a two stage lens design introduced by Richter and Weibring [7]. Schematically, the DFG module (mid-IR power ~400 μW) is similar to our laboratory design [8]. However, the new design has improved PPLN temperature control through dual oven compartments, permanent non-adjustable optics, reduced near-infrared scattering and enhanced vibration isolation characteristics. The unconverted signal and pump radiation are removed by a Germanium filter (Ge) and directed into a beam dump (BD). The remaining idler beam is imaged by a CaF₂ lens (IL) into a Multipass absorption cell (MPC, 100-m pathlength astigmatic Herriott cell). The mid-IR beam enters the detection module through an aperture to prevent scattered pump and signal light to reach the detectors in the detection module. The aperture is periodically blocked by an optical shutter (OS) to enable detector dark current measurements.

A custom coated beam splitter (BS) [15], is used to split off and image a portion of the input beam onto the AMD. This allows us to closely match the beam intensities impinging on the AMD and CD detectors, an important requirement and improvement over our laboratory system for removal of optical noise from the laser source. The BS ratio between CD and AMD detection paths is close to unity, which allows us to obtain the best signal to noise ratio for available photons. The BS also relays a small fraction of the beam through a reference cell (RC) to the reference detector (RD) for computer controlled active wavelength locking (to be discussed) using the target CH₂O absorption feature at 2831.6417 cm^-1. In addition, apertures attached to the AMD and CD detectors further minimize the effect of scattered light. Even though the detection module concept is similar to the laboratory prototype [8], the new detection module was completely re-designed. The optical components and mounts are both significantly smaller and mounted permanently in place by UV activated epoxy after alignment with external jigs. To further minimize the impact of accelerations and vibrations, the optics base plate is mounted vertically and flush with the cell endplate instead of protruding from it. Although the detection module is fully enclosed in an aluminum structure, it was not pressure-sealed. An AR-coated Sapphire window mounted rigidly onto the cell optical input-output port separate the detection module and the multi-pass cell.

The CD, AMD and RD are matched photovoltaic HgCdTe detectors (D^*~3.4×10¹⁰@1kHz, d=1mm), providing almost identical response and noise characteristics. Four stage Peltier elements cool the detectors down to -80°C. The heat from the Peltier elements is conducted through the detection module housing and is dissipated into the spectrometer enclosure compartment using three cooling fans mounted on heat sinks, which are attached to the module outside walls, leaving the module temperature only a few degrees above the enclosure temperature. Reverse biased, low noise trans-impedance amplifiers directly connected to a high precision, differentially coupled analog to digital conversion card in the computer, along with software based lock-in algorithms allowed for very high precision in matching the CD and AMD responses. The 0.7 m long Multipass/Herriott cell (~24 lbs and 4.3 liters), is aligned for an effective absorption path-length of 100 m. Four carbon fiber rods connect the cell endplates to minimize flexure of the endplates and the cell mirrors.

2.2 Gas handling system

The gas handling system (see Fig. 4) is comprised of an inlet, zero air generator, permeation calibration system, spectrometer, multipass cell, air compressor, vacuum pump and an air cylinder. The system has essentially three operational modes; Calibration, Background and Ambient. In all three modes, air is drawn through the multipass cell and inlet tubing at a flow rate of ~8.7 standard liter per minute (slm, where standard conditions are 273 K and 760 Torr). This yields a combined inlet and cell residence time of about two seconds. A pressure controller mounted in the inlet maintains the cell pressure at 50 Torr. The zero air generator, employing a heated Pd/Al₂O₃ catalyst, supplies the inlet and the permeation calibration system with cabin air scrubbed of CH₂O. A pressure controller and glass capillaries regulate the pressure of the two temperature controlled permeation tube ovens housing the CH₂O permeation tubes, ensuring constant CH₂O permeation rates independent of ambient pressure. When power was not available, an air cylinder and a critical orifice in the calibration system established a low flush flow through the calibration system to prevent build up of formaldehyde on the tubing walls. During the Calibration mode, either one of the permeation tubes provides a standard CH₂O mixing ratio into the zero air stream, which is then added to the inlet. During all other measurement modes a ‘suckback’ flow is applied, which removes the CH₂O standard before addition to the zero air. In the Background mode, ‘zero air’ from the zero air generator is directed to the inlet tip at a flow rate a few slm higher than the cell intake flow, thereby excluding ambient air from the system. In the Ambient mode, the ‘zero air’ solenoid valve downstream of the zero air generator is switched and the ‘zero air’ is dumped into the cabin, allowing outside ambient air to be drawn through the multipass cell. A small continuous ‘flush’ flow purged the zero air/calibration line in the inlet, thus ensuring that no ambient air contaminants were deposited. Comprehensive discussions of this calibration approach and its accuracy have been published [10,11,16] and references therein. CH₂O calibration mixtures from the laboratory system were cross compared with those from our second well-characterized airborne system, showing an agreement to within <2% for the two calibration sources [8].

 

Fig. 4. The gas handling system encompasses inlet, zero air generator, permeation system, multipass cell, air compressor, vacuum pump and an air cylinder. The zero air generator supplies the inlet and the permeation system with cabin air scrubbed from CH₂O during powered operation. A pressure controller regulates the pressure of the two heated permeation tubes ensuring constant CH₂O permeation rates. Air is drawn through the multipass cell and inlet tubing at a flow rate ~8.7 slm yielding a cell residence time of less than two seconds. A pressure controller mounted in the inlet maintains the cell pressure around 50 Torr.

Download Full Size | PDF

2.3 Electronic and data systems

The spectrometer system uses 15A of 110V/60Hz single phase power, which is distributed to a vacuum pump, an air compressor, a zero air generator, an enclosure air conditioning system and a dual conversion UPS. The UPS regenerates power for the power supplies powering lasers, fiber amplifiers, detectors, PPLN oven, permeation system and computer, allowing the laser spectrometer to be fully powered during power outages of up to 7 minutes.

The data control and processing system is comprised of a single board computer with dual-Xenon^TM processors running National Instruments LabVIEW^TM on a Windows XP^TM operating system platform. The DFG spectrometer interfaces with the computer through four National Instruments computer boards. An arbitrary waveform generator board provides scan/modulation waveforms and a master trigger for the entire system. A field programmable gate array (FPGA) board serves as a delay generator and laser wavelength stabilization control system. A high-speed acquisition board records the detector signals and a low speed data acquisition board records housekeeping functions such as temperatures, flows, pressures, etc. in the system. Due to the strong wavelength dependence of the background spectra from the DFG source, an active wavelength locking system is essential to eliminate additional noise from environmentally-induced wavelength drifts. This is in contrast to TDL measurements, where the background structure is typically 1 to 2 orders of magnitude smaller and as a result laser feedback control is only required about once every minute. The wavelength locking system utilized the RD detector signal from the formaldehyde reference cell and PID algorithms to actively lock the center of the wavelength scan of the laser to the formaldehyde feature of interest on every scan. The center of gravity of the second derivative signal was determined and used to apply a correction voltage directly to the laser driver. The wavelength locking system updates at 25 Hz and keeps the wavelength drift to less than ± 5 MHz or 0.15% of the scan window. Without feedback compensation, the instrument wavelength drift can be as large as 50–100% of the scan window during a 12 hour flight. In case of malfunction, a back-up system applies a more traditional approach utilizing the same reference cell for co-alignment of each scan in acquisition memory before co-averaging [14]. However, this approach is less desirable since it can generate artificial features on the sides of the averaged scan, which can be erroneously incorporated into the fit of the absorption feature of interest. Comparisons of the various wavelength stabilization approaches for DFG have been investigated using our laboratory system [8].

2.4 Signal processing

To reveal detailed operating characteristics of the DFG technology during airborne operation, an extensive array of monitoring capabilities is included in the software, allowing the user to get a complete real-time diagnostic picture of the system. The system processes the raw data in several steps to compute the trace gas concentration. In the first step, the demodulated CD and AMD detector signals are matched by applying a correction factor ‘R×’ minimizing the ‘R×CD-AMD’ expression during background acquisition. In the second step, the raw data, both spectral and housekeeping, are averaged and grouped into data blocks of high (1s) and low (60s) time resolutions. These data are streamed to the hard drive, allowing different evaluation algorithms to be applied and compared post mission. In the final step, the actual trace gas concentration is computed using Singular Value Decomposition (SVD) [14].

Four figures of merit are used to estimate the data quality and help in determining the Limits Of Detection (LOD), the Background Amplitude Residual (BAR), the Integrated Difference (ID) and the Fitted Background Absolute Areas (FBAAs) [14]. The BAR figure is purely instrument dependent, while the FBAA is also sensitive to interference from other trace gases in the wavelength scan window. The BAR figure is determined for each ambient measurement cycle to quantify the signal magnitude of the background changes. It is computed by subtracting the pre- and post-background spectra point by point over the fit window and then computing the absolute integrated residual voltage values. By empirically relating the BAR values to LODs and stability times, it was possible to estimate a BAR threshold value under which the instrument achieved a certain LOD or better. During a background acquisition, the Integrated Difference (ID) measures the absolute integrated difference between the detector signal voltages using the (RxCD-AMD) expression. A low ID implies a good matching between CD and AMD signals and success in removing the laser source noise. Other indirect parameters such as enclosure temperatures at various locations and cabin pressure provide additional important means to flag compromised data. The concentration estimation is derived from the background subtracted ambient spectra using an SVD algorithm [14]. To accommodate residual background structure after background subtraction, three background terms (offset, linear, and quadratic) were added as basis functions to the regression fit. These terms respectively account for a dc offset, a linear slope, and quadratically-curved background structure between ambient and calibration spectra. The FBAA is determined by taking the absolute area under the calculated background function as determined from the background terms, to yield a ‘relative concentration’ area product. During time periods of perfect background subtraction, these three terms should all be close to zero, which in turn should result in a near-zero calculated background spectrum with a near-zero FBAA value. In the presence of optical perturbations, residual background structure in the time period between background acquisitions can change significantly. This will produce non-zero fitted background spectra with large FBAA values for each associated ambient 1s data point.

As discussed in the introduction, the laboratory system reached a typical CH₂O LOD of 60 pptv for 1s averaging and 13 pptv for 60s averaging [8]. All performance levels in this paper refer to the 1σ level. An environmentally controlled laboratory is not representative of an airborne measurement platform exhibiting severe vibrations and large temperature and pressure variations, therefore instrument LODs were continuously determined in-flight. Almost all precision testing of the laboratory system was carried out using the Allan variance concept [17,18], which unfortunately is extremely time consuming and requires a long stability time (15–60 minutes) to yield enough data points to determine the real sensitivity in the 10-200s stability range. Most importantly, Allan variance measurements prohibit simultaneous ambient measurements from being acquired, which, during expensive airborne campaigns is ill-advised. To maximize ambient measurements, only a limited number of Allan variance measurements were performed in-flight. In lieu of Allan variance measurements, LODs were determined for each measurement cycle by indirect means and justified by the “figure of merit” parameters discussed previously. Each measurement cycle was comprised of an ambient acquisition, a background (zero air) acquisition, and a few seconds of flush before and after each background acquisition. At the end of each cycle, the average of the surrounding background spectra were subtracted from the raw ambient spectra, yielding the optimum estimated ambient spectra. The 1s LOD was calculated from the standard deviation of all 1s measurements during each ambient measurement cycle. An average value over a large number of such measurement cycles was employed to arrive at a single 1s LOD determination for each flight. Assuming a constant ambient concentration and instrument stability during a given measurement cycle, the full cycle LOD should improve the 1s LODs by the square-root of the ambient averaging time, and this generally has been shown to be the case out to 60 seconds of averaging. Thus, the 1s LODs divided by the square-root of the number of 1s ambient measurements yield one estimate of the measurement cycle LOD. In the case of ambient variability, which can be identified in the 1s concentration data, the 1s LOD represents an upper LOD limit determined by the ambient concentration variation. Under conditions where the ambient CH₂O levels are low and stable for sustained time periods longer than say 5–10 minutes, the standard deviation of replicate 30 to 60-second ambient measurements yield a direct determination for the cycle LOD. As ambient variability can not be ruled out, this method also yields an upper limit to the ambient LOD. Nevertheless, such direct determinations, where applicable, were generally in close agreement to those calculated from the 1s precisions normalized by the square-root of the number of 1s ambient averages.

3. Airborne operation

The DFG instrument discussed herein was deployed and airborne for ~300 hrs in total during three consecutive field campaigns in 2006: 1) the Mega cities Impact on Regional and Global Environment (MIRAGE) study; 2) the Intercontinental and Megacity Pollution Experiment (IMPEX); and 3) the Texas Air Quality Study 2006 (TexAQS). The MIRAGE field campaign (March 2006) was designed to examine the chemical and physical transformations of gases and aerosols in the polluted outflow from Mexico City. Formaldehyde plume concentrations ranged from a few hundred pptv to tens of ppbv, depending on sampling distance and altitude from the emission source, requiring 1s LOD in the 200–400 pptv range. The IMPEX (Seattle, April-May 2006) campaign targeted transport and transformation of gases and aerosols on transcontinental/intercontinental scales. An important goal was to quantify the transpacific transport and evolution of Asian pollution to North America and assess its implications for regional air quality and climate. Low formaldehyde levels in the aged Asian air masses required formaldehyde 30s LODs in the 20–40 pptv range. The TexAQS (September-October 2006) study examined regional air quality over Houston, Texas to enhance our understanding of the impact of local, regional, and distant pollution sources. Here, high plume concentrations of formaldehyde in the 10’s of ppbv range required 1s LODs similar to those over Mexico City. Our 1s and 60 s time resolutions throughout all three campaigns, allowed us to resolve both fast mixing ratio variations in highly polluted air and to determine low mixing ratio levels in aged intercontinental transported air masses.

3.1 Problems and solutions

It is important to note that the transition from laboratory to airborne operation for the DFG system discussed herein did not occur without challenges. In the sections that follow, the various problems and solutions are discussed allowing potential new DFG users to avoid such pitfalls when operating in similar harsh environments. To improve the sensitivity and reliability, the instrument was modified whenever possible throughout these campaigns, with major revisions occurring during the intermission breaks. In contrast to the first two field missions, the instrument required no modifications and adjustments during the entire TexAQS campaign. The operator only had to initiate start up/shut down procedures during each research flight and run calibrations. In the beginning of the airborne DFG deployment during the MIRAGE campaign, instrumental problems associated with excessive cabin temperatures (>35 °C), large temperature variations (~20 °C) and severe aircraft vibrations were experienced. Despite the fact that these perturbations were anticipated and taken into account in our design plans, their magnitude simply overwhelmed our design. The two main components affected were optical fibers and fiber amplifiers, as will now be discussed.

3.1.1 Vibrations, pitch, roll, and yaw maneuvers and their effect on performance

The first version of the airborne DFG spectrometer was constructed on 3 separate mounting plates, each shock mounted and vibration dampened and each moving independently of the other plates. The installation of the DFG spectrometer in the propline of the C-130 aircraft during MIRAGE and IMPEX exposed it to severe vibrations in the 10–400 Hz frequency range. As a result, the instrument fiber trays and fiber interconnections between diode lasers, fiber amplifiers, and detection module started to vibrate at the different resonance frequencies. These vibrations induced micro-bending and tensions in the fiber train, leading to changes in phase and polarization states of the transmitted signal and pump beams. The phase matching process for generating mid-IR radiation in the PPLN crystal is strongly dependent on the incoming beam polarization and phase. The contribution of these noise effects manifested itself most clearly as a time dependent intensity change of the second harmonic signal. Before fiber stabilization, the 1s LODs were degraded from 1–2 ppbv to 15 ppbv and from 400 pptv to 2–3 ppbv respectively for the CD and R×CD-AMD detector combinations for measurements before aircraft engine run up and during the take off roll. It is worth noting that an aircraft take off roll presents some of the strongest vibrations encountered. During level flight (MIRAGE), the spectrometer performed a factor of 4–5 times better, yielding 2–3 ppbv and 700–800 pptv (1s averaging) for CD and R×CD-AMD detector combinations respectively. However, the spectrometer was still sensitive to different propeller pitch settings, which resulted in various resonances affecting the fibers in trays and fiber interconnection lines.

Figure 5(a) shows a wavelength scan in terms of wavelength channel numbers (1s averaging) for the CD (white trace) and R×CD-AMD (red trace) detector combinations before the fibers were stabilized during research flight 6 MIRAGE at take off. Figure 5(b) shows the same wavelength scan during a later research flight 10 IMPEX after the fibers were stabilized. As seen in Fig. 5(a), not only was the CD detector affected by the aircraft vibrations, but the RxCD-AMD noise subtraction scheme failed due to a small imbalance between the CD and AMD detection arms. The effect of cabin pressure can also be seen as the overall shape change after the take off roll in the latter half of the Fig. 5(a-b) movies. Balancing the CD and AMD arms is the most effective technique for improving instrument performance. Mechanical design constraints made it difficult while on-board the aircraft to manipulate the optics to achieve a balance that was equal to the ~1/1500 CD and AMD matching possible in the laboratory. Instead efforts focused on making the instrument background (source noise) as stable as possible. The instrument’s fiber sections that were most susceptible to vibrations were located while on the ground using a vibrating toothbrush. The support and stabilization for the susceptible sections of fiber were re-enforced by adding additional foam support and adding additional tape to hold the fibers steady. This improved the system performance, yielding an in flight R×CD-AMD sensitivity of 200–300 pptv (1s averaging) and 35–55 pptv (30s averaging) for measurements during airborne operation (research flights 8–12, MIRAGE). Unfortunately, not all critical fiber paths and interconnects could be re-enforced during MIRAGE due to the limited accessibility of the spectrometer internal parts when mounted on the C-130 platform. During the break between the MIRAGE and IMPEX campaigns, the entire spectrometer enclosure was replaced and most optical fiber paths and fiber trays received extra support and were embedded in damping material. There were a few fiber bridges connecting the fiber amplifier and pump/signal modules which could not be fully supported and remained a problem. Movement between them caused slow but noticeable background changes during the taxi, take off, landing, pitch and roll maneuvers. Despite the fact that these were comparatively slow, a few seconds to tens of seconds, these changes were still fast enough to increase the instrument LODs by a factor of ~2 during their appearance. This effect was recorded during repeated parabolic vertical ascends and descents (Lenschow maneuvers) during MIRAGE. A 30–80% increase in the BARS and an increase of 1s LODs to the 400–500 pptv range was observed. The issue of differentially moving modules was solved during IMPEX, by coupling the individually vibration damped platforms together by makeshift metal brackets and cable ties, resulting in only 15–30% raised BARS and 1s LODs in 150–200 pptv range for the same type of Lenschow maneuvers. During the break between the IMPEX and TexAQS campaigns, all modules were merged to a common vibration damped platform. Unfortunately, no Lenschow maneuvers were performed on the WP-3 during TexAQS, but full turns, ascents and descents were performed without raised BARS, and this resulted in 1s LODs in the 100-120 pptv range, which is comparable to the instrument performance during nominal level flight.

 

Fig. 5(a). (0.83 MB) CD (cell detector, white trace) and R×CD-AMD (background corrected cell detector, red trace) detector combinations (arb. units) as a function of wavelength (channel numbers) before fiber stabilization during take off (research flight 6 MIRAGE) (3.85 MB version) [Media 2].

Download Full Size | PDF

 

Fig. 5(b). (1.39 MB) CD (cell detector, white trace) and R×CD-AMD (background corrected cell detector, red trace) detector combinations (arb. units) as a function of wavelength (channel numbers) after fiber stabilization implementations take off (research flight 10 INTEX-B) [Media 3].

Download Full Size | PDF

3.1.2 Temperature effects on fiber amplifier performance

The fiber amplifiers were adversely affected by temperature instabilities in the spectrometer enclosure. Temperature variations as low as a few tenths of a °C caused, the amplification characteristics (e.g. polarization, dispersion, gain, etc.) to change, leading to a time dependent signal change, which consequently altered the instrument background signal. The transfer function of both fiber amplifiers were affected, but the effect was more pronounced in the EFA amplifier due to the wavelength scanning and associated small input power variation of the current-tuned signal diode laser. Thus, the EFA amplifier transmission function became an order of magnitude noisier. During ideal stable temperature conditions, laboratory measurements showed a considerable noise contribution by the EFA, and as the room temperature was varied the EFA noise also showed strongly correlated temperature dependence. The application of second harmonic detection with appropriate modulation depth offered noise suppression, but was far from perfect. Figure 6 shows the CD and R×CD-AMD signals as the air temperature around the fiber amplifier in the enclosure reacts to an induced temperature variation of ~2 °C. Under perfect CD and AMD matching this effect should be removed. However, in the present system the measurements revealed a need for temperature stability better than a few tenths of a °C during a 30 – 60 second measurement cycle to achieve a LOD of 20–30 pptv. Weibring et al showed that all the components in the optical train, including the PPLN crystal contribute to the optical noise characteristics [8]. The net noise contribution of the PPLN crystal due to enclosure temperature changes was significantly smaller here, since the crystal was encapsulated in a separate, temperature stabilized oven (±0.04 °C). The instrument background signal behavior in Fig. 6 shows similarities to the signal behavior induced by the aircraft take off accelerations in Fig. 5(b). Both accelerations and temperature variations seem to induce polarization changes in the optical train before the PPLN crystal and consequently the DFG output signal background changes. At first sight, the background structure in Fig. 6 appears like an optical fringe, but it becomes evident that this is not a ‘normal’ fringe, as both period and shape drastically change during the temperature change.

 

Fig. 6. (2.07 MB) CD (cell detector, white trace, left scale, arb. units) and R×CD-AMD (background corrected cell detector, red trace, right scale, arb. units) detector combination signals and the air temperature at the fiber amplifier position in the enclosure during an induced 2 degree C temperature disturbance. [Media 4]

Download Full Size | PDF

To compensate for the amplifier temperature sensitivity during airborne operation, both the YFA and EFA amplifiers were placed inside the temperature controlled spectrometer enclosure. The TEC (500 BTU), controlling the temperature, was designed 50% over capacity, which was not sufficient here due to warm weather conditions over Mexico in combination with excessive cabin heat loads from the large payload of scientific instruments. The cabin temperature was never below 34 °C (MIRAGE) and usually varied between 35 °C to 39 °C, except during high altitude legs, that were initiated to cool down the overheated payload. By adding an external compressor air-conditioning unit to cool the hot side of the TEC, temperature control was regained and as a result an LOD of 50–80 pptv (30s averaging) could be maintained during a larger fraction of the MIRAGE flight time. Further investigations showed that temperature gradients and air stratification were present in all parts of the enclosure, and the temperature sensitive fiber amplifiers were located too close to the TEC output. Figure 7 shows the temperature monitored at four different locations inside the enclosure. Every time the cabin temperature changed, an enclosure temperature fluctuation was induced and the TEC temperature compensation heavily affected the components in its close vicinity, especially the fiber amplifiers. This problem was suppressed by installing ten fans in the enclosure mounted in a series fashion creating well mixed air circulation. The air was circulated starting from the TEC output, throughout the whole enclosure and ending at the fiber amplifiers before entering the TEC air intake again. This allowed a much more stable air temperature (± 0.2 degrees C) around the fiber amplifiers and as a result the fiber amplifier induced background variations were greatly suppressed. Further improvements were made by installing insulation panels between the fiber amplifiers and TEC to allow proper heat transfer away from the fiber amplifier heat sinks. These improvements resulted in typical 1s LODs in the 80–120 pptv range and 30s precisions that were typically in the 20–30 pptv range (IMPEX). Not only are these results comparable to our latest lead salt performance on the DC-8 platform, but are also within a factor of ~2.5 of that achieved in the laboratory. In addition, the improved temperature stabilization also enabled the wavelength stabilization system to operate in a higher resolution mode, utilizing a smaller control range without railing and consequently yielding better wavelength locking performance.

 

Fig. 7. Temperature stratification monitored at four different locations in the spectrometer enclosure before modifications were made to improve temperature stability. Diode laser platform (blue trace), DFG and detection module (black trace), Multipass cell (green trace) and Fiber amplifier site (red trace).

Download Full Size | PDF

3.1.3 YFA noise

During airborne operation, the YFA (1083 nm) amplifier output showed periodic noise bursts, which did not correlate to vibrations or temperature fluctuations, but rather originated from the amplifier itself. The YFA was therefore replaced, during the intermission break (MIRAGE/IMPEX), by a new low noise YFA amplifier, which reduced the average noise level in the raw data by a factor of ~10. To protect the new YFA amplifier from optical feedback and decouple the fiber train to suppress Stimulated Brillouin Scattering (SBS) buildup, an optical isolator (35 dB) was spliced to the output of the YFA amplifier. The optical isolator also reduces potential amplifier noise that may be induced by minute amounts of backscattered light originating from optical components down the optical train.

3.1.4 The effect of cabin pressure changes on detector matching

When all effects of vibration and temperature were minimized, it was discovered that cabin pressure variations also contributed to limit the instrument performance, especially during rapid climbs and descents. The impact of such pressure changes can be seen in the movies of Fig. 5(a-b); here one observes complete shape changes in the spectral scans. Neither the DFG nor the detection modules were pressure sealed and ad-hoc modifications during the field deployment were not practical given the current instrument design constraints. Cabin pressure variations monitored during several flights reinforced the fact that pressure variations showed strong correlation with degraded instrument performance, resulting in temporarily degraded 1s LODs and high BAR and FBAAs diagnostic values. This is illustrated in Fig. 8, which shows the cabin pressure variations, concentrations, IDs, FBAAs, BARs, and 1s LODs during several climbs and descents for research flight 7 (IMPEX).

 

Fig. 8. Comparison of measurement data and instrument diagnostic data. The scale on the left hand side applies to; the formaldehyde concentration (Amb. Conc.) with a 1400 pptv offset, the cabin pressure (CPressure) and the LODs with no offset. The scale of the figure of merit parameters IDs, BARs and FBAAs are arbitrary.

Download Full Size | PDF

Relying solely on 1s ambient measurement variability (LOD) is not the best indicator of instrument performance in all cases since true ambient CH₂O variability will also have a contribution here. However, if the 1s LOD correlates well with the BAR diagnostic it may be an indication that the instrument background is changing over a measurement cycle. Under these circumstances, the cycle background subtraction scheme will not faithfully capture the true background structure underlying each of the ambient spectra in that period, and as a consequence this will introduce background structure and noise in the retrieved ambient concentrations. Figure 8 shows two regions exhibiting the most rapid cabin pressure changes during steady ambient CH₂O concentrations and they are highlighted in green and red rectangles. In the red rectangle, a change of cabin pressure of ~100 torr induces high LODs, BARs, and FBAAs. A variation of the power matching (ID) between the CD and AMD detectors can also be seen, suggesting a beam steering effect due to pressure.

During other times, as illustrated in the blue rectangle, the change of pressure was relatively slow and thus did not produce any significant changes in the retrieved ambient variability and FBAAs, while the BARs did indicate instrument background changes. Since such background changes are directly tied to our current fitting approach, any background structural changes not represented by our offset, linear, and quadratic background fitting functions will create erroneous variations in the retrieved concentrations, which are shown in the red rectangle. On the other hand, if a linear combination of the above functions can describe the induced changes of shape in the background spectrum (blue rectangle), then the 1s retrieved spectra are not significantly affected. In further consideration, the green rectangle represents an intermediate case of pressure change, where one can deduce a moderately fast change of the IDs and thus indicate a change in the CD/AMD matching. As a matter of fact, the entire ID trace tracks the cabin pressure variations. Due to the sensitivity of the fitting routine to the spectral shape of the background, the FBAAs in conjunction with the BARs were considered as the best figures of merit to determine if the fitting approach was retrieving less reliable results. In the high formaldehyde concentration region (Time 20:00–23:00) one can see low BARs, yet the FBAAs vary by a large amount. This case indicates a stable instrument background with an ambient spectral structure, which is not formaldehyde, and which can not be fully captured by the offset, linear and quadratic terms in the fitting algorithm. The most likely explanation is an interfering trace gas in the scan window. In the case of the 2831.6 cm^-1 formaldehyde line employed in this study, the only known spectral interference is from weak methanol features on both sides of the CH₂O line. Fried et al. [19] have determined that equal methanol and CH₂O mixing ratios will affect the retrieved CH₂O values by 0.3%, and thus under most circumstances appreciable spectral interference should not be a problem.

3.1.5 Allan variance measurements

To validate the system performance and the figure of merit parameters, a limited number of Allan variance measurements were performed throughout the campaigns. Figures 9(a) and 9(b) shows Allan variance measurements for the MIRAGE and IMPEX campaigns respectively. The lower (black) trace in Fig. 9(a) shows the spectrometer performance on the ground when the aircraft engines were turned off. The upper (blue) trace and the middle (red) trace show the airborne performance before and after fiber vibration stabilization. Even though effective, the vibration damping efforts during MIRAGE did not damp out all vibrations and consequently the airborne LODs are higher and the stability time is shorter compared to the ground based performance. The performance improvement as a result of the modifications (new YFA, optical isolator, fiber and interconnections support) during the MIRAGE/IMPEX intermission break becomes evident if we compare Figs. 9(a) and 9(b) (red trace airborne, black trace ground operation). The airborne operation 1s and 30s LODs are improved by a factor of ~2 and the instrument stability time is extended beyond 60s. The same improvement can also be seen during the ground based operation 1s LODs, but not in the 30s LODs. The main reason for this behavior is that the optical isolator installed to protect the YFA amplifier from optical feedback introduced a 3 dB transmission loss and therefore the pump power to the EFA amplifier had to be increased to compensate for the power loss. As a consequence, the EFA amplification noise increased considerably, which can be seen as a less steep LOD slope in Fig. 9(b) compared to Fig. 9(a). Note, that the Allan variance requires long instrument stability times (15–60 minutes) to yield enough data to determine the real sensitivity in the 10–200s stability range. In flight we acquire backgrounds every 30 to 60 seconds, which resets the stability criteria. Thus, the Allan variance constitutes an estimate of an ‘upper’ limit for the LODs and a ‘lower’ limit of the stability time.

 

Fig. 9(a). Allan variance measurements showing the spectrometer LODs for the MIRAGE campaign. Ground based operation (black trace) airborne operation before (blue trace) and after (red trace) fiber stabilization.

Download Full Size | PDF

 

Fig. 9(b). Allan variance measurements showing the spectrometer LODs for the IMPEX campaign. Ground based operation (black trace) airborne operation (red trace).

Download Full Size | PDF

3.1.6 Solutions before the TexAQS Campaign

The instrument was modified during the intermission break between IMPEX and TexAQS campaigns to further improve on instrument robustness, weight reduction and semi-autonomous operation, as well as to address the temperature and vibration issues discussed above. In addition to employing a common vibration damped platform as mentioned earlier; an improved thermally insulated spectrometer enclosure was constructed that included a new compressor air-conditioner and PID controlled TEC capable of maintaining the enclosure temperature at 26 °C ±0.2 °C even at cabin temperatures as high as 40 °C. In addition, the commercial Aadco^TM zero air system was replaced by a compact light weight zero air generator using the same type of Pd/Al₂O₃ catalyst repackaged in a new horizontal tubular arrangement without the molecular sieve towers (weight saving of ~ 71 lbs); new electronics and FPGA algorithms for improved wavelength stability were implemented; and new monolithic flexure mounts improved the alignment and stability of the detector focusing mirrors. All the above improvements yielded a lighter, more compact and semi-autonomous instrument, requiring only start-up/shut-down user intervention at the beginning/end of each research flight.

3.2 Performance summary

The vibration and temperature problems described above required sampling of background (zero air) for a larger fraction of the measurement cycle during the MIRAGE and IMPEX campaigns. This measurement cycle consisted of 30s ambient, 15s background, and two 5s flush periods, which reduced the ambient data duty cycle to ~55%. During the TexAQS campaign the instrument upgrades allowed for a ~70% ambient duty cycle without sensitivity degradation. Table 1 shows sensitivity, duty cycle and data coverage for all three campaigns. One must keep in mind that our measurement cycle includes background acquisition time, i.e. 100% data coverage of “flight minutes” translates to 55% ambient coverage for MIRAGE/IMPEX and 70% ambient coverage for TexAQS. As seen in Table 1 the instrument improvements throughout the campaigns not only improved our LODs but also increased our data duty cycle.

Table 1. Instrument performance in terms of sensitivity, duty cycle and data coverage for MIRAGE, IMPEX and TexAQS campaigns. The instrument measurement cycle includes background acquisition time, i.e. 100% data coverage of “flight minutes” translates to 55% ambient coverage for MIRAGE/IMPEX and 70% ambient coverage for TexAQS.

View Table

4. Sample results

4.1 Measurement intercomparison

The instrument systematic errors were evaluated by laboratory instrument comparisons and during MIRAGE by intercomparison with the NCAR-TDL formaldehyde instrument flown on the NASA DC8 aircraft [19]. Figure 10 displays an example of a blind intercomparison time series of field data for formaldehyde (60s averages) between the C-130 and the DC-8 over Mexico City on March 19, 2006. The DC-8 results were produced from our lead-salt TDL instrument, with the blue symbols representing the 1s TDL results averaged over the 30s DFG time base. The C-130 results were from the DFG instrument and are shown in red. A least squares regression of the data is shown in Eq.(1).

 

Figure 10. Airborne intercomparison of DFG and Tunable Diode Laser (TDL) measurements of CH₂O flying on respectively, the NCAR C-130 and the NASA DC-8 airplanes. The two aircraft were flying in close formation on March 19, 2006 over Mexico City. The 1-second TDL measurements were averaged over the DFG time base.

Download Full Size | PDF

Based on these results, the TDL and DFG retrieved CH₂O mixing ratios were in agreement to within ± 13% over the 1 – 4 ppbv range, which is well within the mutual uncertainties of both instruments. The accuracy of the TDL instrument is ± 13% and this has been verified by comparison with other instruments during numerous campaigns. The study by Wert et al. [20] is one such example.

4.2 TexAQS formaldehyde concentration map

The present paper focuses on the airborne performance characterization of the spectrometer and describes modifications to reduce the effects of noise induced during airborne operation. Although many modifications were made throughout the campaigns, the instrument was operated nearly continuously and the following example illustrates one measurement and its scientific value. Figure 11 shows a map of Texas and flight track with formaldehyde boundary layer measurements from a research flight 7 on Sept.21, 2006 during the TexAQS2006 campaign. The flight path (lat/long) is overlaid with the formaldehyde concentration scaled by point size and color. The map shows a northward moving plume originating from urban Houston and the ship channel area. The instrument’s 70% measurement duty cycle can be seen in the flight track, where the gaps occur during instrument background acquisition, which excludes it from taking ambient data. The insert shows a sample of high resolution (1s) data from a high altitude leg, from which a 1s LOD of ~90 pptv can be deduced.

 

Fig. 11. Research flight 7 TexAQS2006, September 21, 2006 over Houston and surroundings. The flight path (lat/long) is overlaid with the CH₂O concentration scaled by point size and color. The insert shows a sample of high resolution (1s) data from a high altitude leg, indicating a 1s LOD of ~90 pptv.

Download Full Size | PDF

5. Summary and outlook

A DFG spectrometer for airborne measurements of trace gases in the mid-IR wavelength region has been developed and operated for ~300 hours during three airborne missions throughout 2006 (MIRAGE, IMPEX, TexAQS). To our knowledge, this is the first time a DFG spectrometer has successfully acquired trace gas data on an airborne platform. The detection of formaldehyde, which is an important trace gas, is representative of the capabilities of this technology, which may be extended to other species in the mid-IR wavelength region. The 1s LODs for formaldehyde were typically between 80–120 pptv and the 30s LODs were typically in the 20–30 pptv range. Not only are these results comparable to our latest lead salt (TDL) performance on the DC-8 platform, but they are also within a factor of ~2.5 of that achieved in the laboratory. Turbo propeller platforms, such as the NCAR C-130 and NOAA WP-3, present a particularly challenging environment due to severe vibrations and large temperature and pressure variations, requiring robust spectrometer characteristics. As we have learned from these studies, thermal and vibration management together with pressure management and improved CD/AMD detector matching are extremely important requirements for high DFG performance. Improvement in all of these areas will be a high priority in future developments to achieve even better instrument sensitivity and robustness. Based on our overall experience with both airborne TDL and DFG systems, measurements with the latter appear to be much more stable in the presence of aircraft accelerations (pitch, roll, and yaw maneuvers) than the former once the various precautions discussed in this paper were implemented. Quantitative improvements in this regard will be the subject of another paper. In addition, since the DFG source, transfer optics, and cell coupling can be packaged in a more compact configuration than those in a TDL system, airborne DFG systems are more readily amenable to pressure control than TDL systems. We anticipate that these advantages and those enumerated in the introduction will eventually lead to an airborne DFG performance that is a factor of 2 to 3 higher than that achievable with the best TDL systems.

The next generation instrument will also be more compact, lighter weight and autonomous. During the three campaigns the instrument size and weight was reduced from 2 to 1.5 racks and from 750 lbs to 570 lbs. Even further size and weight reduction will be realized by more compact custom electronics, miniaturization of the gas handling and air-conditioning systems and a laptop based computer platform. The goal is to fit the complete spectrometer system in a single rack weighing less than 430 lbs (inc. rack), capable of operation on NCAR’s high altitude aircraft platform (Gulfstream G-V airplane HIAPER) without a human operator. The next generation spectrometer software will be autonomous and incorporate the ability of intelligent pre-processing of the detector signals, background adaptive robust fitting algorithms and support multi-species measurements (formaldehyde and methanol). The prime efforts to achieve true autonomous multi-species operation will be in software development, electronics and gas handling systems.