1. Introduction

Vibrational spectroscopy has proven to be a powerful method for chemical selective imaging because each molecule has a set of vibrational frequencies unique to the species. In particular, biological imaging based on Raman spectroscopy has the advantages of being non-invasive, label-free, and highly spatially resolved [1–3]. Among such spectroscopy methods, coherent anti-Stokes Raman scattering (CARS) spectroscopy is particularly sensitive; therefore, it is one of the most promising methods for biological imaging, where the samples tend to be vulnerable to optical or thermal damages [4, 5].

Conventional CARS spectroscopy uses three laser fields with frequencies ω_pump, ω_Stokes, and ω_probe. When ω_pump − ω_Stokes approaches the vicinity of an active molecular vibrational frequency, the vibrational mode is strongly driven, leading to the emission of an anti-Stokes field (ω_probe + ω_pump − ω_Stokes) through a four-wave mixing process (three-color scheme). The CARS process can also be triggered by only two laser fields with frequencies ω_pump and ω_Stokes, where ω_pump also works as the third field (two-color scheme).

A broad Raman spectrum is required for identifying biological molecules because these molecules generally have complex structures and thus complicated vibrational spectra that span a wide spectral range. Moreover, because CARS imaging requires a large number of point-by-point measurements, single-shot measurements are preferred for obtaining the spectra in view of data rates. To conveniently obtain a broad spectrum with a single-shot measurement, the multiplex CARS method [6, 7] uses a combination of a broadband femtosecond laser and a monochromatic probe field. However, multiplex CARS spectroscopy has several drawbacks. First, the obtained signals are generally weaker than those obtained by the three- or two-color schemes because the amplitude of each spectral component of a broadband laser is relatively small. Second, a narrowband pulse, which is necessary to improve the frequency resolution, must be both spatially overlapped and synchronized with a broadband pulse. This indicates that multiplex CARS requires precise control of two independent laser systems with respect to each other, making the scheme difficult to apply for practical purposes. Finally, a nonresonant background (NRB), which deteriorates the signal-to-noise ratio, is particularly prominent in femtosecond-laser-based multiplex CARS spectroscopy.

Oron et al. proposed single-beam CARS spectroscopy in which broad Raman spectra are obtained with a single femtosecond laser by enhancing signal contrast using heterodyne detection [11]. Instead of a narrowband laser pulse, they employed a femtosecond pulse shaper to generate a narrowband phase modulation. A narrowband phase modulation produces interferometric marks where the resonant CARS signals appear on the NRB. Here, the NRB functions as a local oscillator for heterodyne detection. Single-beam CARS spectroscopy is very powerful and has therefore been improved [12–14] and applied to biological imaging [15, 16] by several groups. However, the method still has a few weaknesses. First, the NRB is removed by subtracting the spectrum obtained without phase modulation from the modulated spectrum. However, because the NRB spectra with and without phase modulation are different, the subtraction cannot be perfect; hence, additional structures appear in the spectrum, making the identification of molecules difficult. Therefore, further mathematical processing is necessary to extract resonant Raman signals.

In this study, we improved the single-beam CARS measurement to achieve the effective subtraction of NRB while maintaining high signal contrast. We propose a new approach for multiplex CARS with a simple setup and operation that realizes heterodyne detection while suppressing the NRB. To subtract the NRB perfectly, we produced temporal displacement between the narrowband component (probe) and the broadband component (pump). Temporal displacement eliminates the effect of the phase modulation on the NRB spectrum and enables us to perfectly subtract the NRB. Such time-resolved method [17,18] in combination with phase sensitive detection and heterodyne detection [19] enables us to achieve an extremely high signal to noise ratio for multiplex CARS spectroscopy.

2. Experimental

The goal of the method presented here is to extract the resonant signal via phase sensitive detection. The probe is applied after the pump has passed through the sample while the coherent molecular vibrations still remain. Phase modulation on the probe therefore selectively modulates the resonant signals, which may then be detected. The signal contrast is further enhanced by the heterodyne detection using the NRB generated by the pump as the local oscillator.

To achieve this, we developed a Michelson interferometer in which a band-pass filter is used in place of the half-mirror to divide the probe from the pump. This spatial separation enabled us to simultaneously generate sufficient delay and apply phase modulation. A schematic of the experimental setup for our proposed CARS spectroscopy is shown in Fig. 1. We did not use an SLM as in the previous study [11–16] because both pixelization of an SLM and diffraction limit of light caused difficulties in generating a sufficiently large delay between the pump and the probe to temporally separate them. Moreover, the use of a femtosecond pulse shaper is not preferred for practical applications because the slow response speed of a liquid crystal spatial light modulator (LC-SLM) generally used in a pulse shaper limits the acquisition speed. In fact, special modifications are required to achieve quick response with an SLM [20]. This is a large disadvantage in imaging studies.

 

Fig. 1 Schematic of the experimental setup. Femtosecond oscillator output is sent into the frequency-resolved interferometer to perform phase modulation while a delay is simultaneously introduced.

Download Full Size | PDF

We used a single femtosecond laser oscillator (Tsunami, Spectra Physics Inc.) as the broadband laser source. The center wavelength and bandwidth are 800 and 30 nm, respectively. The output typically had an averaged power of 300 mW and a pulse repetition rate of 80 MHz; thus, the estimated pulse energy was 4 nJ. The pulses were sent into the Michelson interferometer, in which the pump was separated from the probe by the band-pass filter. The two components were retroreflected and recombined at the filter. The retroreflector used in the narrowband arm was mounted on a piezo-translational stage to modulate the optical path length (and therefore the spectral phase). The recombined pulse propagated through a longpass filter (> 780 nm) and was tightly focused by an achromatic lens into a sample cell. In this study, shorter pump duration is required for observation of Raman signals. To avoid undesired dispersion, we used minimum number of the optical elements in the optical path. The total dispersion of the setup is small enough to observe CARS spectra despite the existence of inevitable dispersions of the bandpass filter and the focusing lens. The beam was collimated after the cell, passed through a short-pass filter (< 770 nm) to cut the input components, and introduced into a spectrometer having a cooled CCD. Here, all the spectra were obtained with an exposure time of 21 ms, which is the minimum exposure of the CCD. Typical modulation speed was set to ∼ 20 rad/s, suitable for the exposure time, but 10⁴ rad/s is potentially achievable. We used chloroform in the cell as a test sample; it has distinct Raman modes in the frequency region of < 1000 cm⁻¹.

3. Results

We used two different band-pass filters with bandwidths of 4 nm (63 cm⁻¹) and 0.3 nm (4.7 cm⁻¹). As the first demonstration of the proposed CARS measurement, we used the wider bandwidth filter so that the resonant signals would be intense and thud easily identified. By using the interferometer, we simultaneously produced arbitrary delays and rapid modulation on the probe. Here, the probe was randomly modulated, so that the spectra were obtained shot-by-shot with the different modulation phases. The upper five panels in Fig. 2 show CARS spectra obtained with probe delays of −0.67, 0, 0.67, 1.3, and 2.0 ps. Each panel is an overlay of 100 shots of spectra with random modulation phases.

 

Fig. 2 Raw data (upper five panels) and difference CARS spectra (lower five panels) with different probe delays.

Download Full Size | PDF

If the phase-modulated probe is temporally overlapped with the pump, large variations in the NRB spectra appear because entire pulse shape and peak intensity change with the modulation phase. Large change on the temporal profile of the entire pulse should cause large variation on the NRB spectra. In fact, when the probe delay is 0 ps, the spectra vary substantially from shot to shot. The background for wavelengths longer than 770 nm was significantly reduced by a short-pass filter inserted immediately after the sample cell. The lower five panels in Fig. 2 show the differences between the nth and (n + 1)th spectra (0 ≤ n ≤ 99). The highest peaks appearing on the right correspond to the interference between the pump and the probe. At zero delay, the difference spectrum became noisy, reflecting variations in the NRB. For other delay values, variation in the spectra was clearly suppressed. For negative delays, the NRB and resonant peaks decreased. This indicates that the variations in the NRB and resonant peaks are decreased as the temporal overlap between the pump and the probe decreases. In contrast, for positive delays, the variation of the NRB was selectively suppressed, and resonant peaks were maintained at approximately 780 nm and 750 nm. This is because the vibrational coherences excited by the pump maintained by the probe arrived. When the temporal displacement is sufficiently large to divide the pump and the probe and is small compared to the coherence time of the vibrational modes, the NRB variation can be selectively suppressed.

We summarized the results by accumulating the absolute differences from the averaged spectrum at particular delay time as expressed by

 

Fig. 3 Difference spectra of CARS spectra with (a) a 4-nm-bandwidth filter and (b) a 0.3-nm-bandwidth filters. Black and red curves are obtained with and without delay of the probe respect to the pump.

Download Full Size | PDF

The largest double peak around 780 nm corresponds to the interference between the pump and the probe at the edges of the narrowband phase gate. Two peaks that appear below 770 nm correspond to the resonant CARS signals.

We improved the spectral resolution by using another custom-made band-pass filter with the bandwidth of 0.3 nm at 790 nm. E ^AC(λ) with the narrow bandwidth filter at delays of 0 and 3 ps are shown in Fig. 3(b). Even for the significantly narrower filter bandwidth, the NRB was influenced by the modulation at zero delay. Therefore, the baseline of Fig. 3(b) was not flat; hence, it was difficult to identify resonant peaks. On the other hand, for the 3-ps delay, three distinct peaks appeared, corresponding to the two resonant CARS signals (746.8 and 763.7 nm) and the interference on the input spectrum (∼785.7 nm), which is consistent with the result in Fig. 3(a). The achieved spectral resolution measured was 8 cm⁻¹ and was limited by the resolution of the spectrometer we used.

We further improved the analysis of the data set presented in Figs. 3(a) and 3(b) by a phase-sensitive method. Because the amplitude of the spectrum with pump-probe interference reflected the modulation phase, we could phase-sensitively extract the CARS spectra E ^PS(λ) by weighted averaging according to the phase θ_i as

Figures 4(a) and 4(b) show E ^PS(λ) calculated from the same data sets used for Figs. 3(a) and 3(b). Both in the Figs. 4(a) and 4(b), spectral fringes appeared in the resonant peaks and the pump-probe interference peaks; the fringe interval corresponded to the reciprocal of the delay time, equivalent to spectral interference.

 

Fig. 4 Phase-sensitive detection for CARS measurement with (a) 4-nm bandwidth filter and (b) 0.3-nm bandwidth filter. Delays are 2 and 3 ps for (a) and (b).

Download Full Size | PDF

The envelopes of the spectra were extracted by the formula

 

Fig. 5 Comparison of two different analyses of data sets acquired by (a) 4-nm-bandwidth filter and (b) 0.3-nm- bandwidth filter. Black and red curves were the difference spectra E ^AC(λ) and the phase-sensitive spectra E ^AC(λ), respectively.

Download Full Size | PDF

4. Discussion

Chloroform is an asymmetrical tetrahedral pentatomic molecule that has six distinct vibrational modes at 262.0, 365.9, 668.3, 761.2, 1215.6, and 3018.9 cm⁻¹ [21]. Given the bandwidth of the pump and the cut-on wavelength of the short-pass filter after the sample, the second- and third-lowest modes can be observed in our current setup. The effective bandwidth of the probe was measured to be 0.5 nm and is limited by the resolution of the spectrometer we used. Thus, the limit of the spectral resolution of the present measurement is 8 cm⁻¹. The CARS signals shown in Fig. 5(b) were identified as vibrational modes at 366.6 and 664.6 cm⁻¹. Table 1 compares previously measured vibrational frequencies of the chloroform molecule [21] and the results of our measurement. The results of our measurement agree with the known frequencies within the frequency resolution of 8 cm⁻¹, demonstrating that our method works reliably.

Table 1. Comparison of Previously Measured Vibrational Frequencies of Chloroform Molecules [21] and those Measured with our Method

View Table

The delay time τ required for the probe is related to its duration t_p. To avoid the overlapping of the probe with the pump, the delay τ should exceed the duration of the longer pulse (generally the probe), which can be estimated from the bandwidths. This condition can be expressed by $\tau \hspace{0.17em}>\hspace{0.17em}{t}_p\hspace{0.17em}\approx {\omega }_p^{-1}$. Thus, a higher spectral resolution requires a longer time delay to suppress the NRB. On the other hand, the coherence of the molecular vibration excited by the pump should be maintained until the arrival of the probe. Therefore, the delay τ should be shorter than the coherence time of the vibrational modes Ω⁻¹ where Ω indicates the spectral linewidth of the mode. As a result, the delay τ is constrained as

In this paper, we show averaged difference spectra E ^AC(λ) and phase sensitive spectra E ^PS(λ) of CARS signals, both of which are mathematically constructed. However, these operations can, in principle, be achieved by electrical processing. For example, as the average of the difference spectra is equivalent to an AC-coupled spectrum, one method to observe the spectra in real time is to use a multichannel photodetector array with an electrical low-frequency cut filter followed by a spectrometer. In fact, this method can be implemented as an extension to an instrument developed by our group for quantum wavepacket measurements [22, 23]. Multi-channel phase sensitive detection is also practically available as shown in Ref. [24]

5. Conclusions

In previous studies of single-beam CARS spectroscopy, the NRB could not be subtracted perfectly; hence, further mathematical extractions were required. Moreover, the use of a femtosecond pulse shaper causes the setup to be complicated and difficult for practical use. In this paper, we proposed and demonstrated an improved CARS spectroscopy method that realizes perfect noise reduction with a relatively simple setup. To efficiently reduce the NRB, we applied a large delay to a phase-modulated narrowband component of a single broadband light source. By referring the modulation phase, we successfully obtained CARS spectra with extremely high signal-to-noise ratios by using a phase-sensitive method. To achieve the delay and modulation, we developed a Michelson interferometer in which an optical band-pass filter is used instead of the half mirror generally used. Compared to a femtosecond pulse shaper, our interferometer is quite simple and has a better response speed, which is an advantage for two-dimensional studies. In this study, we mathematically demonstrated the extraction of phase-sensitive spectra. In addition, the proposed method can be extended to real-time measurements by electrically achieving phase-sensitive detection.