Chirped laser dispersion spectroscopy (CLaDS) enables optical sensing of vapor phase chemicals using continuously frequency-tuned lasers. CLaDS provides the full information of the electromagnetic field interacting with a gaseous medium including both absorption and dispersion, respectively related to the imaginary and real parts of the medium’s complex refractive index [1]. Measuring molecular dispersion spectra, derived from the phase of the optical field, inherits characteristics and benefits of phase measurements: a robust immunity to intensity fluctuations, and linearity of the magnitude of dispersion with molecular concentration. These aspects contrast with and complement tunable laser absorption spectroscopy driven by the exponential Beer–Lambert law, solely related to the optical power incident upon a detector [2,3].

To suppress common mode noise, CLaDS measures the differential anomalous dispersion produced by molecular resonances. In its most straightforward form, two fully coherent fields, separated by a fixed frequency difference Ω, simultaneously probe the gaseous medium. In the near infrared part of the spectrum, relevant to the work reported here, a simple implementation consists of exploiting first-order sidebands produced by intensity modulation of a laser source.

Optical gas sensing using CLaDS has been demonstrated and characterized for in situ measurements [4,5] and remote measurements, both in the middle infrared and the near infrared [6,7]. However, so far only applications to narrowband absorbers, in other words low molecular mass species such as carbon dioxide and methane, have been reported. Many real time optical sensing scenarios require the identification of heavier molecules dispersed in air at atmospheric pressure, whose spectra are characterized by broad absorption features where individual rotational–vibrational transitions are no longer resolvable. This is the case for heavy hydrocarbons or volatile organic compounds related to industrial and air quality applications, but also of heavy threat molecules, such as improvised explosives or chemical warfare agents relevant to security and defense [8,9].

In this Letter, we report on the extension of CLaDS optical sensing to molecules with broadband resonance features. First, the system design choices are described, as well as the experimental implementation and its rationale. Then, experiments conducted for optical sensing of propyne, used as a test molecule, are reported before analyzing and discussing the results.

Optical fingerprinting using unresolved rotational–vibrational bands requires a wide frequency tunability of the single mode laser source. An external cavity semiconductor laser was therefore selected. To facilitate the demonstration and leverage the maturity of photonics components at telecommunications wavelengths, the near infrared part of the spectrum was selected [10,11]. This choice comes with a cost that only the harmonics of resonances can be probed, with the associated sensitivity loss compared to the stronger oscillator strength of middle infrared fundamental bands.

A time-dependent release of propyne (methyl acetylene, C₃H₄) was used for the demonstration. With the molecule’s three carbon atoms, at atmospheric pressure the first overtone of the acetylenic CH stretch (ν₁) centered at 6565 cm^-1 does not exhibit resolved rotational features [12]. The band covers a total range of 50 cm^-1 and has a classic P, Q, and R branch structure. With these choices, a mode-hop free fiber laser tunable from 6250 cm^-1 (1600 nm) to 6579 cm^-1 (1520 nm) was selected as the optical source (New Focus Venturi).

The frequency separation, Ω, of the two probing fields used for differential dispersion measurement is the second critical parameter. To maximize dispersion signals, Ω should be of the same order of magnitude as the spectral widths of the resonance features to be probed [1]. A requirement of Ω > 1 cm^-1 was set for the demonstration, approximately corresponding to the width of the propyne Q branch. As a result, the intermediate heterodyne frequencies to be analyzed are above 30 GHz, significantly larger than in the cases of smaller, lighter molecules.

The experimental apparatuses are shown in Fig. 1. Throughout the optical system, polarization-maintaining single mode fibers are used. The output of the laser source is connected to an electro-optical intensity modulator (OptiLab). The modulator is driven by a custom-built low phase noise radio frequency (RF) source operating at ${f_M} = 1\mathrm{6\; GHz}$, producing sidebands at $\nu \pm k \cdot {f_M}$, with $\nu $ the input laser frequency and k = (1, 2, …, n) an integer. The RF source consists of a phase-locked dielectric resonator oscillator (Ultra Electronics Herley PDRO14-16G-E100) with an external quartz crystal reference (Wenzel 100MHz-SC). The specified phase noise is below –100 dBc at an offset frequency of 1 kHz from the carrier. To prevent unwanted sideband mixing and to approximate closely the ideal case of two fully coherent fields separated by a fixed frequency, the DC bias applied to the intensity modulator is adjusted to ensure that the fundamental power, k = 0, is suppressed by 25 dB compared to the power at k = ±1. In addition, the RF power of the modulator excitation signal is also adjusted to ensure that higher order sidebands, |k| > 1, are at least 20 dB down. The relative contributions of the carrier and sidebands were measured using an optical wavemeter. With this condition, as a first approximation, the output from the modulator is a coherent dual frequency field, with a fixed frequency separation of $\mathrm{\Omega } = 2{f_M}$. With this sideband optimization, the modulator converts 16% of the incident laser power into the two desired tones.

 

Fig. 1. Schematic of the chirped laser dispersion spectrometer for broadband absorber sensing. Except for the open space where the gas release occurs between two collimators (Coll.), optical power is transmitted in polarization-maintaining fiber. PD, Amp, and Sub H mixer stand for photodiode, low noise amplifier, and sub-harmonic mixer, respectively.

Download Full Size | PDF

To record spectra of the absorption band of propyne, the laser frequency is tuned at a rate of 87,000 cm^-1/s between 6452 cm^-1 and 6579 cm^-1. Each trace requires a scan time of 1.5 ms. This mechanical tuning of the laser wavelength can be repeated at a rate up to 27 Hz.

The optical field is coupled to free space propagation using an aspheric lens collimator (focal length =18.4 mm) with a path length of about 30 cm, defining the zone for gas plume release and gas/laser interaction. After the free space propagation, the light is focused back into fiber, using an identical collimator, before being coupled to the optical detector.

This is a fast InGaAs photodiode, 40 GHz bandwidth, acting as a heterodyne optical mixer. The ∼32 GHz beat signal from the k = ±1 sidebands feeds from the detector into a low noise amplifier with a power gain of 24 dB. A 3 GHz wide bandpass filter, centered at 32 GHz, reduces noise components.

The RF demodulation architecture was kept simple with a direct baseband conversion using a sub-harmonic RF mixer (Analog Devices). The 16 GHz local oscillator comes directly from the reference RF source. A power coupler (Marki C13-0126) divides the RF power and provides both the drive input into the electro-optic intensity modulator and the local oscillator of the demodulator. The sub-harmonic approach uses the internal nonlinearity of the device to beat a virtually frequency-doubled local oscillator with the RF signal of interest. This approach ensures full coherence and reduction of potential phase drifts. The mixer outputs both the In-phase (I) and Quadrature (Q) baseband signals, conditioned by amplifiers (90 × voltage gain and 150 kHz bandwidth) and digitized by a 16-bit acquisition card (Adlink USB-1210 operating at 2 MS/s). This configuration avoids any additional frequency downshifting or demodulation [7] and remains fully analog without the need for high-speed digitizers.

The amplitude (AM) and phase (ϕ) of the heterodyne beat note are derived from the measured I and Q outputs, using rectangular to polar coordinate conversion. The frequency is obtained by numerically differentiating the phase with respect to time.

During a gas-release experiment, a plastic tube of ∼30 cm in length, 5 cm diameter, and open at each end is installed around the beam to provide loose confinement of the gas. Pure gas is then released inside the tube while the instrument operates.

For quantification of gas concentrations, amplitude (absorption) and frequency (dispersion) signals were measured and compared. Reference cross sections for propyne at 25°C and 1 atm from the Pacific Northwest National Laboratory (PNNL) spectral database [9] were used to calculate the expected signals for forward modeling. For amplitude measurements, the optical transmission for each trace τ(ν) was calculated from the measured AM signal, V_AM(ν), using an absorption-free trace recorded immediately before the gas-release V_AM0(ν),

The absorption model includes a cubic polynomial baseline (a, b, c, d coefficients) to account for large baseline variations introduced by the fluctuations of the received power, and a translational frequency offset ${\nu _t}$. The wavenumber dependent cross section reference data C(ν) are used to forward-model the absorption assuming a concentration of x and a path length of L.

The frequency f(ν) is defined in Eq. (2) as the time derivative of the signal phase ϕ corrected from the phase ϕ₀ measured before a gas release. Ideally ϕ₀ should be constant. However, a residual slight phase unbalance between the two sidebands generated by the optical modulator means a residual dependence on the chirp rate exists. For the forward model, the derivative of the real part of the complex refractive index n is taken with respect to the angular frequency of the radiation ω. In

The real part of n is derived [13] using the Kramers–Kronig equation via the real part of a Hilbert transform $\mathrm{{\cal H}}$ of the absorbance α(ν) calculated from the reference cross section data,

Fixed parameters of the models include the absorption path length taken as the length of the “confinement” pipe (L = 27 cm), the laser tuning rate (S = 87,000 cm^-1/s), the frequency difference between the +1 and –1 sidebands (Ω = 32 GHz) and the bandwidth of the final I/Q amplifiers (150 kHz). This final term is a low-pass temporal filter applied to both the amplitude and frequency outputs of the model.

With the forward models in place, concentrations are fitted (among other fitted parameters described above) using a non-linear iterative method based on Levenberg–Marquardt optimization [14]. Once the fitting procedure has converged on a solution, the residuals between the experimental and the fitted forward model are used for diagnostics. Examples of fitted spectra and residuals for both beat note frequency deviation and transmittance are shown in Fig. 2. These traces were obtained during a single frequency sweep of the laser lasting 1.5 ms.

 

Fig. 2. Experimental spectra obtained during a propyne release (black crosses) together with fitted models (red lines). From top to bottom: beat note frequency deviation and corresponding residual; transmittance and corresponding residual.

Download Full Size | PDF

The average propyne concentrations returned by the fit were 18,092 ± 50 ppm and 16,734 ± 39 ppm when using the frequency and transmittance spectrum, respectively. Using the propagated errors on concentration determined by the fit, the one sigma detection limit corresponds to 0.74 ppm.m.√Hz and 0.58 ppm.m.√Hz for the frequency and transmittance spectrum, respectively. Because about half of the spectral range, 6450–6520 cm^-1, is free from propyne interactions, the residuals can also inform on the nature of the errors affecting the measurement. Away from the propyne absorption features, the frequency signal is truly random with a standard deviation of 53 Hz. Using the maximum peak to trough frequency amplitude of the signal as a reference, 6 kHz, this corresponds to ∼0.8%. In contrast, the transmittance shows structures, but with an overall standard deviation of 0.001. Using the maximum absorbance, 0.085, as a reference, this corresponds to 1.2%. Considering the spectra within the P, Q, and R branches of propyne (6540–6580 cm^-1), the spectroscopic model errors dominate, as witnessed by the non-random residual features. Following the same analysis, in this spectral region the relative errors come to 2.3% and 3.5% for frequency and transmittance spectrum, respectively.

A 20 cm gas cell filled with a mixture of propyne in air at atmospheric pressure was introduced in the beam path to carry out noise analysis using Allan variance. Dispersion (absorption) signals were found to be white noise limited until at least 300 s (100 s). In contrast, after the fits, concentration measurements for both cases followed white noise statistics up to ∼20 s.

Systematic errors also originate from uncertainties in model parameters, primarily the interaction path length. However, these are the same for absorption and dispersion spectra. Periodical zero mean structures in the dispersion spectra are also perceptible. These are inherent to the residual fringing (∼15 cm^-1 free spectral range) of the laser source.

The concentration measurement methodology suggests using dispersion signals provides more precise estimates on a single trace. The case of the dynamic of a propyne release was considered by recording traces before, during, and after a 5 s gas injection. Individual spectra are recorded over 1.5 ms durations, at a rate of 15 spectra per second. The non-uniform time-dependent nature of the gas release and diffusion afterwards was seen to strongly influence the transmittance signal, owing to the significant power variations of the beat note amplitude introduced by scintillation. Traces sampled at different phases of the release show these effects in Fig. 3. The five traces shown were recorded before (black trace), during (red and blue), and after (pink and green) the gas release.

 

Fig. 3. Frequency (left) and transmittance (right) spectra recorded at different stages of a propyne gas release. Each spectral trace was recorded over a time duration of 1.5 ms. The black trace was recorded before the gas release, the red and blue traces were recorded during the release, and the pink and green traces were recorded after the gas release.

Download Full Size | PDF

Effects of signal variation occurring during the experiment are further illustrated in Fig. 4. In this figure, the contrast of on- and off-resonance individual spectral channel monitored over time is considered for the frequency and transmittance. The difference in behavior between the two types of signal is striking, particularly the immunity of the dispersion-based signal to the large power variations affecting the transmittance measurements.

 

Fig. 4. Temporal traces of single wavelength transmittance and frequency signals recorded during a gas-release measurement. The red line depicts an “on-resonance” point at the Q branch peak; the black line is an “off-resonance” point away from any propyne features.

Download Full Size | PDF

The recovery of the off-resonance transmittance to its expected value of 1 indicates the turbid phase of the release is finished at ∼25 s.

Using the quantitative approach of spectral fittings, the measured temporal evolution of propyne concentrations can be scrutinized to provide further elements of comparison between dispersion- and absorption-based measurements. The evolution of the fitted concentration over time is shown in Fig. 5. During the injection phase and the turbid phase, the frequency measurements are more reliable and stable, as indicated by the relative error in concentration, in line with the far superior stability of the dispersion-based measurement in this condition. Using frequency spectra, concentrations up to ∼310,000 ppm are measurable. Considering the one sigma precision of 50 ppm for a single spectrum, measured with an optical power of 0.16 mW, a dynamic range of 6000 is achievable with the demonstrator.

 

Fig. 5. Top: temporal evolution of the propyne concentrations derived from frequency and transmittance spectra; bottom: corresponding relative error of the measurements.

Download Full Size | PDF

After 25 s have elapsed and stable conditions are re-instated, concentrations and errors derived from transmittance and frequency spectra converge to similar values. Well after the release, as the gas has dispersed, the concentrations measured from frequency spectra consistently give ∼100 ppm, close to 2× the one sigma precision of the measurement.

This demonstration of a broadly tunable laser dispersion spectrometer operating in the near infrared shows the relevance of the technique to the detection and quantification of large molecules presenting broadband features. Using propyne as a test molecule, a one sigma precision of ∼0.7 ppm.m.√Hz is reported. A dynamic range of ∼6000 is demonstrated with only 0.16 mW of optical power used for the λ ≈ 1.55 µm measurement.

The demonstrator reported in this Letter was implemented in the most straightforward and economical manner: it uses an analog RF architecture performing direct baseband downconversion from a 32-GHz carrier, suitable for Q branch molecular features.

The results obtained with the demonstrator confirms the benefits of a zero-baseline frequency signal, requiring fewer fitting parameters for quantitative interpretation, reducing model errors and cross-correlations. The dispersion spectra approach is shown to be more robust for turbid gas release, through its immunity to received power variations. Precision on concentration measurements is also improved, though still limited by model errors, which are about three times the measurement errors in the case of the gas studied.