There are significant potential scientific benefits of O₂ A-band multi-angle measurements. For example, they can accurately measure cloud properties such as cloud top height and cloud physical thickness (Fischer and Grassl, 1991; Min and Harrison, 1999; Koelemeijer et al., 2001; Cornet et al., 2010; Schuessler et al., 2014; Yang et al., 2021), surface atmospheric pressure (Barton and Scott, 1986; Vanbauce et al., 1998; Tran and Hartmann, 2008; Fischer et al., 2009), aerosol properties (van Diedenhoven et al., 2005; Hollstein and Fischer, 2014; Nanda et al., 2018; Davis and Kalashnikova, 2019; Gao et al., 2019; Remer et al., 2019), and chlorophyll fluorescence (Perez-Priego et al., 2005; Frankenberg et al., 2011; RayChaudhuri, 2012; Sanders and de Haan, 2013). These and other scientific benefits have been demonstrated from O₂ A-band measurements from satellites, such as POLDER (POLarization and Directionality of the Earth’s Reflectances), SCIAMACHY (The SCanning Imaging Absorption spectroMeter for Atmospheric CartograpHY), and OCO-2 (Orbiting Carbon Observatory-2).

Until now, O₂ A-band measurements have not been demonstrated from instruments on CubeSats. In this study, we demonstrate an innovative photonic integrated circuit concept that potentially supports high spectral resolution (0.01 nm) for O₂ A-band measurements suitable for CubeSats missions. Compared with grating spectrometer system equipped with satellites (i.e OCO-3), our 10 p.m. linewidth integrated MRR filter perform more than 25 dB suppression of photons in off-resonance wavelengths. This device highlights the potential applications of integrated photonics in satellite-based remote sensing systems for high detection sensitivity, superior resolution, and significantly reduced size, weight, and power (SWaP). Therefore, our vision is to be able to launch tens of CubeSats that carry these new O₂ A-band instruments to provide high spatial and temporal resolution measurements of aerosols, clouds, ocean and land surfaces.

We calibrate the MRR filter with a tunable laser (New Focus, TLB-6700). The atmospheric absorption spectrum between 760 to 770 nm is simulated using the HITRAN 2020 (high-resolution transmission molecular absorption) database (Gordon et al., 2022). We use an ultra-narrow bandpass filter (Alluxa 769.89-1 OD7) to define a ±1 nm spectral window with 769.89 nm center wavelength to detect an Oxygen doublet line shape between 769.72 nm to 770 nm. To characterize the MRR, an electric current is applied to the micro-heater from an external power supply (GPD-4303S, Instek), whose power is increased from 24 mW to 38.4 mW at a 0.8 mW interval. Figure 5A shows the MRR spectrum in different wavelengths by tuning the applied electric power from 0 mW to 29.1 mW. The fitting results are shown in Figures 5A,B linear redshift of the filter’s center wavelength from 769.72 nm to 770 nm at a 13 p.m. step size is measured. This good linearity is the key to the proposed spectroscopic measurement. Currently, the coarser thermal tuning step size of the MRR is limited by the tunable laser used for its’ characterization. Finer tune step size can be achieved using a narrow linewidth tunable laser system with higher resolution.

The experimental setup for measuring the atmospheric O₂ absorption lineshape is outlined in Figure 6. It consists of a free-space-to-fiber beam reducer (Thorlabs-GBE10-C) to collect the solar irradiation into a multi-mode fiber (50-micron diameter) spliced into a single-mode fiber. A bandpass filter blocks the transmitted solar radiation to a ±0.6 nm spectral window centered at 769.89 nm. A polarizer is introduced to couple the polarized solar photons into the single-mode MRR filter. The polarized light splits into two channels with a 99:1 coupler, with 99% of the light coupled into the MRR filter on the LNOI chip and the remaining 1% directly into the reference channel Si APD. This configuration allows us to account for the photon count variation in real-time due to the fluctuations in the fiber coupling and the dynamic atmosphere attenuation induced by, for example, cloud coverage, solar zenith angle, or atmospheric scattering. The MRR filter’s output is fiber-coupled into the signal channel Si APD to count the photon numbers while the filter is swept from 769.72 nm to 770 nm. A FPGA is employed as the central processor for tuning the center wavelength of the MRR filter by varying the electric heating power on the Pt micro-heater and acquiring synchronized photon counting data from the two Si APDs channels. Table 1 lists the parameters of each part of the system.

The integration time of photon counting per sample point is 125 ms in the current system, rendering the total data acquisition time around 2.6 s. Here, an adequate integration time is crucial to attaining statistical significance by suppressing the Poissonian noise with single-photon counts. This ensures the signal-to-noise ratio (SNR), N Solar / N Solar + N DC , to be greater than 253 across all data points (Quatrevalet et al., 2017). Here, N _Solar is the registered photon counts in the signal channel. In contrast, N _DC is the total noise count of the entire detection system, measured by disconnecting the optical fiber from the beam expander. Our chip-to-fiber coupling efficiency is more than 7 dB per facet due to the higher mode field size mismatch compared with telecommunication band wavelengths (i.e., 1,550 nm). By designing and fabricating on-chip edge couplers (i.e., inverse taper), the chip-to-fiber coupling efficiency can be greatly enhanced to 2 dB per facet (He et al., 2019; Li et al., 2020). With 10 dB improvement chip-to-fiber coupling, three times higher SNR of solar photon counting can be achieved (Quatrevalet et al., 2017). A shorter integration time is needed to achieve the same SNR with a lower noise level or reduced insertion loss of the MRR filter. The shortest thermo-optic response time for thermal-optical tuning is 60 μs Liu et al. (2020). An even shorter time is possible by using electro-optical tuning on thin-film lithium niobate (Zhang et al., 2019).

We plot the normalized photon counts to retrieve the atmospheric O₂ absorption spectrum. The result is shown in Figure 7 along with the calculated dry-air O₂ absorption spectrum from the HITRAN 2020 (temperature: 296 K, pressure: 1 atm). An good agreement is seen, both on the absorption linewidth and depth. Specifically, the measured linewidth between two absorption line is about 104.5 p.m., compared with 102 p.m. as extracted from the HITRAN database. This small discrepancy is within the error caused by the 10 p.m. resolution of the MRR filter. The depth extinction is 0.906 by fitting the measurement results with a Lorentzian function, which agrees with the HITRAN database value of 0.901, too. The solar photon counting rate of the signal channel is equal to 141 kHz at 769.795 nm. The maximum photon counting rate outside these doublet absorption lines is equal to 169.9 kHz. The photon counting rate from the reference channel fluctuates between 4.1 to 4.3 MHz during the data-acquiring time. The dark counts for signal and reference Si APDs equal 500 Hz, much lower than the detection photon counts. Benefiting from the ultra-low noise of our system, only several tens of thousands of photon counts were needed per spectral point to retrieve the fine lineshape of the atmospheric O₂ band around 769.795 nm and 769.898 nm. Also, atmospheric solar photons around 770 nm are much more than mid-inferred wavelengths (i.e., 1,570 nm); therefore, the solar photon counting of visible light has a higher photon counts rate to obtain better SNR in remote sensing.

Our results show that the combined LNOI-based integrated photonics and single-photon detection promise photon-efficient remote sensing. With two channeled photon counts simultaneously recorded in the Si APD channel, we could acquire the visible absorption spectrum with high resolution and accuracy amid the imperfections in sunlight couplings and dynamic atmospheric conditions without adaptive optics. The present spectroscopy can prove helpful in pervasive deployments in satellite, airborne, and wide-field missions with fast scanning, single-photon sensitive, high resolution, and ultra-low detection noise. Real-time, dynamic monitoring of the atmospheric O₂ concentration can be realized by using dual MRR filters on a single nanophotonic chip, with one tuned into the absorption line (i.e., on-line) and the other off (i.e., off-line) while recording the ratio of their photon counts using Si APDs. Also, a similar system can be applied to monitoring the sun-induced fluorescence of vegetation using two-dual MRR filter sets, with one centered at the oxygen B-band (i.e., 687 nm) and another one at the oxygen A-band (i.e., 760 nm), to maximize the detectability of fluorescence photons through the deepest solar Fraunhofer lines of O₂ (Cendrero et al., 2019). This approach is similar to the IPDA lidar but without restriction in laser wavelength. Moreover, space-borne spatial and temporal monitoring and measurement of aerosols, clouds, ocean, and land surfaces are possible by leveraging the photonics chip’s scalability and broadband nature of solar radiation.

Our new system highlights the potential of adopting integrated photonics and single-photon counting in remote sensing systems for high detection sensitivity, superior resolution, and significantly reduced device size and weight. However, to be compatible with the current remote sensing spectral measurements and provide more precise spectroscopic results, we are looking forward to the data training from current ground-based and airborne databases in the future.