A.

General principle

This part describes the spectrometer studied at CNES which was based on a dispersive optical element: the échelle grating spectrometer [2].

The instrument is divided into three parts; see Fig. 2:

Figure 2.

Components of the instrument’s functional system

The field of view at the entrance of the instrument’s spectrometer section is limited by a physical slit. The spectrometer forms sharp monochromatic images of this slit on three matrix detector array. The three detectors are associated with the three spectral bands: B1 (0.76 microns), B2 (1.6 microns) and B3 (2.05 microns). These spectral bands are relatively narrow, but are observed with a high-power resolution, typically R = 25000 (R = λ/Δλ, where Δλ is the spectral width at half transmission along a monochromatic spectral line).

Using a long slit provides a spatial dimension on each detector which is perpendicular to the direction of spectral scattering. The information acquired along this axis can be added together (binning) to increase radiometric resolution. Typically, 50 pixels (Nbin) are summed in this way (Fig.3). The long axis of the slit is itself approximately perpendicular to the movement of the satellite at the nadir. Fig. 3 illustrates this geometric configuration and the terminology used in this case.

Figure 3.

Illustration of spectral axis versus spatial axis

The spectrometer must have a high-power resolution within a restricted space. It is known that the power resolution of a grating spectrometer used in near Littrow conditions is described by one of the following equations (1) or (2):

Where:

Diameter d1 is necessarily small due to the size issue. The value of entrance pupil diameter D is determined by the radiometric resolution. The spectral power resolution for a given instrument volume will therefore be optimised by increasing the angle of incidence α, diffraction order number k, or groove density m.

One of the fundamental characteristics of the MicroCarb instrument is its compactness, because it has to fly on a Myriade Evolution microsatellite bus. For this reason, we chose an optical concept based on double pass Tri Mirror Anastigmat (TMA) or Four-Mirror Anastigmat (FMA). Double pass means that the same optical components are used for the collimation function (light rays are made parallel before they arrive at the dispersive component) and for the camera lens function (the rays dispersed by the dispersive component are focused on the focal plane). The addition of a fourth mirror allows an additional degree of freedom to improve the quality image without increasing the size of the instrument. An initial implementation of this concept (see fig. 4) has been studied, taking into account a calibration system and active and/or passive thermal control. The volume is compatible with that available on a Myriade Evolution bus (H: 412mm x w: 700 mm x L: 864mm).

Figure 4.

Typical optical configuration. The diffraction orders for the desired spectral bands are separated and sent to the corresponding detectors by a set of dichroic filters.

B.

Zoom on dispersive component: echelle grating

This part describes the main component of the spectrometer studied at CNES: the echelle grating.

Spectral coverage requirements (bands B1, B2 and B3) led us to favour an increased incidence angle and diffraction order to the detriment of groove density. The type of grating with these characteristics is known as an “echelle” grating. This kind of component allows spectral multiplexing. By carefully choosing the grating parameters, central wavelengths of spectral bands can coincide with optimal echelle grating diffraction efficiency in specific diffraction orders. To further increase performance, the echelle grating is operated in near-Littrow conditions, i.e. the diffraction angle at the diffraction peak is nearly identical to the angle of incidence of rays on the grating. Using a high diffraction order (13 < k < 50) along with a carefully selected pitch m can provide diffracting combinations in the same directions as the rays associated with the centre of bands B1, B2 and B3.

Useful information is distributed in distinct diffraction orders, so the signal carried by these orders must be separated so it can be sent to each detector optimised for one band (B1, B2 and B3).

Conventionally, diffraction orders in an echelle spectrometer are physically separated through a slightly dispersive auxiliary element functioning perpendicularly to the main dispersive element (cross disperser). We rejected this solution because we are only interested in a relatively small number of spectral bands at the same time, these bands being covered by distinct diffraction orders. The orders are separated by a set of dichroic filters and spectral bandpass filters centred on the bands of interest (Fig. 4). As previously seen, the characteristics of the grating are carefully chosen so that the diffraction angle is identical for the central wavelength of each band.

The separation of diffraction orders by the echelle grating through bandpass filters is well-suited to the use of a long slit, which maximises instrument scope.

The main features of the echelle grating are as shown in table 1 below:

TABLE I.

ECHELLE GRATING CHARACTERISTICS

C.

Experimentation of the instrument principle

We set up an experiment to trial the principle of an echelle grating spectrometer operating in MicroCarb’s spectral bands. This breadboard uses a standard (Richardson) echelle grating. The assembly was calculated and set up so as to attain a power resolution of up to R = 40000 in the CO₂ 1.6 micron band (i.e. a spectral width of 0.04 nanometres, see Fig.5).

Figure 5.

Example of an experiment profile for the CO2 spectrum in band B2 (uncalibrated)

Fig.6 shows the simplified optical configuration used for this feasibility breadboard. The equipment can be used to check the principle behind order selection through bandpass filters, measure the polarimetric specifications of gratings, specify the processing and calibration algorithms, and enhance the modelling of this kind of instrument

Figure 6.

MicroCarb Engineering Breadboard

A.

Parametric relationship (level 1)

The level 1 requirements for MicroCarb instrument are such that:

The signal/noise ratio, spectral resolution and bandwidth are the instrument’s driving parameters for CO2 retrieval accuracy. As very different combinations of these parameters might give similarly good Level 2 performance, we define a parametric relationship:

Where:

α, β, γ, δ are optimal values fitted for a set of different instrument configurations.

The requirements asks p to below 0.4 (goal) or 0.55 (threshold)

A minimum and a maximum value are also specified for each parameter (cf table 2).

TABLE II.

SPECIFICATIONS OF PARAMETER

A footprint on the ground of the sounding elementary pixel is specified between 9 km² and 50 km².

For a typical diameter of the entrance pupil of 26 mm, a spectral resolving power of 25000 and the mean observation radiance, table 3 gives the modelled radiometric resolution. The latter is the signal to noise ratio (SNR) in the spectrum by the resolution element after binning 50 pixels along the spatial axis (Fig.3). This binning corresponds to a linear length along the ground and at the satellite nadir of 5 km. Integration time is such that the elementary FOV footprint covers 5 km x 5 km.

TABLE III.

PERFORMANCES LEVEL 1

The echelle grating is an etched model with 69.33 lines/mm, used with an incidence angle of 67.9°. The effective size of the grating is 48 mm x 140 mm.

In its reference operating mode, the instrument simultaneously acquires five contiguous, independent FOV with the radiometric resolution indicated in table 3. The instrument’s ground swath when the slit length is perpendicular to the satellite velocity axis is therefore 25 km. Within this swath, the pixel binning areas can be chosen as needed to avoid cloudy areas.

B.

Pseudo-noises (level 1)

Along with the key parameters considered in the important function described above, instrument defects will affect the spectrum measured in the spectral bands, and therefore the precision of the estimated CO2 concentration. Such defects have a similar effect on radiometric noise, which is why the term “pseudo-noise” is used. Different kinds of pseudo-noise include:

The goal is to design an instrument for which pseudo-noise is limited and noise mainly consists in radiometric noise.

In the case of non-linearity defects, one of the main contributors is the detector, especially in the low signal dynamic. Let us examine this defect more closely.

In the preliminary study, to reduce development and qualification costs, we used performances of HgCdTe off-the-shelf detectors from Sofradir for B2 and B3 bands to perform our radiometric calculations [5]. The size of the detector is 500x256 pixels, with a 30µm pitch and uses Capacitive TransImpedance Amplifier (CTIA) readout circuit. This choice means that we have to deal with non-customised components. For instance, the capacitance is larger than we need, which induces read-out noise and non-linearity issues.

Existing data concerning detector response show an increase in pixel response dispersion under 10% of full well capacity and a non-linearity shape different from one pixel to another.

It is important to determine whether non-linearity observed at low levels is due to measurement precision, and can be treated as a noise, or if it is a signed response of the CTIA that has to be taken as a bias.

In order to keep to the specified objective of not being dominated by pseudo-noise, the design has been studied to increase the signal (integration time and optics diameter).

C.

Polarisation

Incident solar radiation at the top of the atmosphere is nonpolarised. Two mechanisms polarise it: scattering throughout the atmosphere, and reflection by the surface.

Polarisation by the atmosphere is caused by radiation scattering, particularly Rayleigh scattering (by molecules), as well as scattering by aerosols (the smaller they are, the greater their polarising effect will be). The rate of polarisation by the atmosphere is variable and can be high: especially in absorption lines because of Rayleigh polarisation).

In the case of polarisation by specular-type reflection, the size of the IFOV considered is an important parameter. For kilometric resolution (MICROCARB or POLDER type), the phenomenon of unpolarized lambertian reflection clearly prevails for a view of the nadir on earth. For a view of the sea, however, due to the glint, the polarisation rate can be nearly 100% with low winds (specular surface, see Fresnel’s Law), which confirms the signal’s potentially high polarisation level when it enters the instrument.

For MicroCarb, the instrumental polarisation is specified so as to ensure that its impact on the signal measured by the detector is negligible compared to the radiometric noise. Other strategies can also be used. The polarisation can be measured by the instrument, for example, which is what the GOSAT instrument does. The measurement can also be taken using fixed polarisation, as in the case of the OCO instrument.

Specifying polarisation with the MicroCarb approach has been established as follows: by rotating the polarisation plane of the polarised monochromatic incident flux, the instrumental polarisation appears as a variation in the recorded signal. In quantitative terms, this leads to specification of the defined polarisation ratio as p=(Imax−Imin)/(Imax+Imin), in which Imax and Imin are the maximum and minimum intensity readings when the polarisation plane is rotated by 180°.

In glint mode: Tp < 0.1% at the goal and 0.7% at the threshold

In nadir mode: Tp < 0.25% at the goal and 5% at the threshold

With these values, considering the expected polarisation rate values for scenes on entry, the pseudo-noise level is below 1:1000.

A component which makes up a considerable share of the overall instrument budget is the grating, and special care is taken to ensure good grating polarisation performance. This is why we have developed a grating polarisation model (see paragraph IV) and studied very precise characterization of instrument polarisation and its main contributing factors. Nevertheless, to meet the required specification level, a polarisation scrambler (optical depolarising component) needs to be inserted in the MicroCarb design.

A.

Requirements

It is tricky to find a diffraction Grating with low polarisation and good efficiency, and we know that this will be the main contributor to the instrument polarisation.

As the instrument polarisation requirement is <0.1% we are forced to add a depolariser to the instrument, as this value is out of the state of the art for gratings. But a depolariser only reduces the instrument polarisation; it is thus important that the grating itself should have the lowest possible polarisation.

We are working with both Richardson Gratings and Horiba Jobin Yvon companies to find the best grating for polarisation and efficiency. The different gratings ordered from both will then be mounted on our breadboard and characterized for comparison.

B.

Modelling Polarisation

A development was made with Institut Fresnel [6] to build a grating model in order to predict grating performances for our instrument.

The model was achieved and compared with real measurements on several different gratings, including gratings with uncommon profiles, and immersed gratings.

The figure 7 shows the comparison for the Richardson Grating (ref 53_*_453E) that we currently use on the MicroCarb Breadboard.

Figure 7.

Comparison between grating measurement and simulation using real groove profile. 31.6 grooves/mm, Blaze = 71.45°, diffraction order 38, const deviation = 0.6°

There is a good agreement between performance simulations and measurements, as well as for absolute efficiency and for polarisation. The model is able to take into account real groove profiles, high orders of diffraction, and a high angle of incidence.

The echelle grating we are currently using on the MicroCarb breadboard has a polarisation under 5% and an efficiency > 65% on a 15 nm spectral range in band B2 (Fig. 7), which is a really good performance. But this echelle grating is not a custom grating, and does not exactly match MicroCarb spectral bands, even if it is close. Thus, through CNES Research and Development program, we are looking for a grating that fulfils the MicroCarb project needs, with both Richardson and Jobin Yvon. We aim at a polarisation under 5% in each of the 3 MicroCarb spectral bands, along with an efficiency above 70%.

C.

Measuring efficiciency and polarisation of gratings with breadboard

This part outlines the echelle grating measurement method with the MicroCarb breadboard.

Figure 8 illustrates the measurement of efficiency in band B1. This measurement is made directly with the MicroCarb breadboard. The results are thus fully representative of the grating performance with the instrument conditions (angle of incidence, beam size, wavelengths, order of diffraction…).

Figure 8.

Example of spectral response of Richardson echelle gratings measured with the breadbord in band B1

A first measurement is made with the echelle grating, and a second measurement with a reference mirror in order to be able to correct the grating measurement from the instrument spectral response.

The measurement is made with a collimated beam in quasi Littrow, with a constant angular deviation of 0.6° (Fig.6). This means that the incidence angle on the grating is adjusted for each wavelength, so that the angular deviation is 0.6°. This is necessary to ensure a rigorous comparison of the measurement with the reference mirror.

A polarizer in the breadboard is used to measure the efficiency under TE or TM polarisation. Figure 8 illustrates the measurement of polarisation in band B2.

The sources are 2 tuneable lasers ([760-780nm] and [1460-1610 nm])

This setup makes it possible to test and compare many different gratings, and to use real measurements in the instrument simulator.

A.

Description of the instrument simulator

Calculating the performance of a spectrometer is fairly complex and requires modelling of various types of instrumental defects (spectral, electronic, spatial, etc.) as well as a calibration strategy. A model of the dispersive spectrometer as described in Figure 4. physics equations, can be used to obtain the signals measured on the detection matrix (science measurements and calibration measurements) using a sun spectrum, atmospheric spectrum or specific sources (blackbody, gas cell, laser…). The instrumental defects from the optical, detection or electronic systems, etc. are then added. The kinds of defects typically encountered include:

This way, we can model the level 0 measured by the instrument and level 1. Level 0 corresponds to non-calibrated raw digital signals picked up as they exit the detection lines. All of the raw measurements will be used to check level 1 products, calibrated spectral products. Setting up an instrument modelling software programme has proved usefulness in establishing or verifying the validity of the instrument specification.

The flow diagram for the MicroCarb instrument model is presented in Fig. 9.

Figure 9

Flow diagram for the MicroCarb instrument model

B.

Illustration with breadboard simulation

The characteristics of the specific MicroCarb mock-up set out in paragraph II above have been taken into account in the instrument simulator.

Since there is a slit image projection for each wavelength, by using the slit defects and studying their position in the field when they are projected on the detector, we are able to estimate the misalignment of the optics. This way, we have been able to estimate certain defects more precisely:

After correcting the misalignment, we proceed to estimate the optical distortion.

The image of a laser line through the spectrometer is distorted by the optics. To evaluate this, we calculate the barycentre of the laser line as a function of the position in the field (spatial axis), and we then make this curve a fit curve. This gives us the distortion associated with the wavelength of the laser line (Fig. 10).

Figure 10

Estimated distortion on the image obtained using the laser

Using the results of the measurements taken with the mock-up (laser line spectrum at 770nm, Fig. 10), the shape of the line on the detector was estimated as illustrated in Figure 11.

Figure 11:

Example of the measured shape of a laser line (dots) and the modelled shape (continuous line).

For this estimation, we correct the distortion and then combine the spectrums from different positions in the field in order to obtain the average shape.

The shape above corresponds to a resolution power of 31000. A supergaussian shape of around 1.6 fits best with the shape’s reconstruction.

The dispersion seen here is due to the presence of speckles relating to the use of a laser source for the measurement, rather than the instrument itself. Some improvements are in progress on the breadboard to reduce these speckles.