Open Access
Add to BibSonomy
Abstract
Shifted Excitation Raman Difference Spectroscopy (SERDS) is a non-destructive chemical analysis method capable of removing the fluorescence background and other disturbances from the Raman spectrum, thanks to the independence of the fluorescence with respect to the small difference in excitation wavelength. The spectrum difference is computed in a post-processing step. Here, we demonstrate the use of a lock-in camera to obtain an on-line analog SERDS spectra allowing longer exposure times and no saturation, leading to an improved Signal-to-Noise Ratio (SNR) and reduced data storage. Two configurations are presented: the first one uses a single laser and can remove excitation-independent disturbances, such as ambient light; the second employs two-wavelength shifted sources and removes fluorescence background similarly to SERDS. In both cases, we experimentally extrapolate the expected SNR improvement.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Raman spectroscopy is a widely used spectroscopic technique that enables label-free detection of chemicals as well as their quantification [1–3]. The analysis of the spectrum of light after inelastic scattering events within a sample can yield significant information regarding its structure and composition [4,5]. Raman spectroscopy is available as consumer products [1,2,6,7], and it is widely accepted as a standard analytical tool. Moreover, a number of Raman techniques have been developed over the years to increase the signal [1], such as surface enhanced Raman spectroscopy, coherent anti-Stokes Raman spectroscopy, and stimulated Raman scattering, though with limited translation to the consumer market, as well as ongoing research to improve spontaneous Raman [8–10].
One well known challenge in Raman spectroscopy is the presence of a strong fluorescence background when visible light is used as the excitation wavelength [11,12], particularly when compared to typical Raman efficiency, that can be as low as 10−8 of the incident excitation power [1,13]. Fluorescence light can be emitted by the molecules of interest themselves, as well as the surrounding medium, and is particularly problematic for biological investigation [11], which is also one of the main drivers for research and further development of the technique [14]. The presence of this background causes several limitations [12]: it is difficult to identify Raman peaks over the fluorescence background and thus obtain quantitative estimations. The background noise is increased and the Signal-to-Noise Ratio (SNR) is reduced. Detector saturation may limit the exposure time and again reduce the SNR.
Several solutions to enhance the Raman signal with respect to the fluorescence background have been proposed, either taking advantage of different characteristics of those types of emission (lifetime, spectral width, polarization), via interaction with other materials such as nanoparticles, or by leveraging the relation to the excitation wavelength [11]. Each of these methods has its advantages and disadvantages, and the appropriate choice is dictated by the application.
One way to reduce the fluorescence signal is to use excitation wavelength in the near infrared, where the fluorescence emission is typically weaker. On the other hand, the Stokes emission then takes place at longer wavelength, and silicon based detectors are not efficient above 1000 nm. Alternatively, the use of differential excitation has been proposed [15,16]: the idea in this case is to use two or more wavelengths of excitation to produce multiple spectra. Raman scattering occurs at fixed offsets to the excitation wavelength, so its wavelength shifts in lockstep with the excitation source. The fluorescence component instead, has a much lower dependence on the wavelength of excitation when small shifts are considered. Computing the difference between spectra generated by two closely spaced excitation wavelengths generates background free differential spectra that can be analyzed more effectively [17–20]. This technique is usually referred to as Shifted Excitation Raman Difference Spectroscopy (SERDS). It is important to note that all these steps occur in post-processing. Each spectrum still contains the high-intensity background, which is digitally removed by subtraction. This means that the exposure time might have to be reduced in order to avoid detector saturation from strong backgrounds, and that the single-shot SNR of the Raman peaks remains low. Some previous works use synchronized illumination and detection, such as with charge-shifting CCDs [21], but the two spectra are still separately digitized, thus not solving the issue of saturation. The same applies to schemes that include multichannel lock-in [22], as the signal comes from a standard CCD and is thus bound to the well capacity of its pixels.
Taking inspiration from our previous work on lock-in Differential Phase Contrast (DPC) [23], in this manuscript we show how it is possible to use a lock-in camera to obtain on-line, analog SERDS spectra. With this method, it is possible to obtain background free spectra with longer exposure times and no saturation. Here, we demonstrate two case scenarios for the use of lock-in Raman spectroscopy. In the first example we use a single wavelength of excitation, and show how the ambient light, that would otherwise saturate the detector in normal conditions, can be removed. This allows the use of Raman for in-vivo or in-vitro applications without the need to shield the detector or operating in dark ambient conditions, facilitating the use of Raman instrumentation in non-conventional configurations. In a second example, two distinct excitation wavelengths are used to obtain a SERDS measurement whereby the fluorescent background is removed analogically in the lock-in camera. To encompass both the single source and double source configuration, we will refer to this technique as lock-in Raman Difference Spectroscopy (RDS).
2. Theory
2.1 Basic SERDS theory
Let us assume that the Raman emission spectrum obtained from a sample of interest is described ideally by the function $R(\bar \nu ,{\bar \nu _e})$, where $\bar \nu $ is the emission wavenumber, and ${\bar \nu _e}$ is the excitation wavenumber. The sample may also emit light in the form of fluorescence upon excitation, which is described by the function $L(\bar \nu )$. Moreover, unwanted light may also come from external disturbances such as ambient illumination or stray light, and are described as an excitation-independent background $B(\bar \nu )$.
The whole spectrum can then be described as the sum of these contributions [19]:
The Raman spectrum responds to changes in the excitation frequency only by an equal shift in the emission, with no significant changes in the spectrum itself. For two excitation wavenumbers ${\bar \nu _1}$ and ${\bar \nu _2}$, where ${\bar \nu _2} = {\bar \nu _1} + \delta \bar \nu $, we can write:
If we measure the spectra $S(\bar \nu ,{\bar \nu _1})$ and $S(\bar \nu ,{\bar \nu _2})$, we can then compute their difference:
If the excitation wavenumber difference is small, then $\delta S$ can be expressed as the differential of the Raman spectrum, so the Raman spectrum can be retrieved through an integration:
2.2 Lock-in RDS
While SERDS helps in digitally removing the fluorescence and ambient background from Raman spectra by subtraction, it does so in a post processing second step. This means that no benefit is provided in terms of dynamic range: when the Raman signal is much weaker compared to the background, the exposure time has to be set to accommodate the whole intensity within the well capacity of the detector. As a consequence, the SNR of the Raman peaks is only partially improved, unless multiple spectra are recorded and averaged. Of course, the stronger the background intensity, the shorter the possible exposure time, which means more spectra need to be collected. This can obviously put a strain on the data-intensiveness of the experiments, especially when a scanning confocal microscope is used to map the Raman spectrum of a sample.
A clear parallel can be drawn between this case and that of DPC [23]. In both techniques, the difference between two similar measurement, be it images or spectra, is calculated to separate a small, periodically modulated signal of interest from a high-intensity constant background. In DPC, the signal of interest is the phase term, and the modulation is a change of sign caused by the switching of mirrored illumination sources. In SERDS, the signal of interest is the Raman spectrum, and the modulation is the shifting of Raman peaks caused by a change in wavelength of excitation.
In order to remove the background in a bit-efficient way, we propose the use of a lock-in camera as a detector for differential Raman spectra. In our experiments, we used the helicam C3, produced by Heliotis: this camera features a detector with pixel-level circuitry that allows to perform I-Q direct detection [24–26], which employs In-phase (I) and Quadrature (Q) reference periodic signals to demodulate the received signal. A schematic explanation of the mode of operation of lock-in RDS is given in Fig. 1. The hardware elements of the setup are shown schematically in Fig. 1(a). The source of excitation wavelength is modulated periodically over a period T, switching between two closely spaced wavenumbers ${\bar \nu _1}$ and ${\bar \nu _2}$. This may be done by modulating a single source, or alternatively switching on and off two different lasers. This wavenumber switching source is used to excite Raman emission in the sample, and the emission is collected by a spectrograph. The wavelength-spread light is sent to the detector of the lock-in camera, and due to the periodically shifting nature of the excitation, the spectrum is also time-dependent. In particular, Fig. 1(a) shows two example simulated spectra where a single Raman peak is present over a strong background. For the first half period, ${\rm S(t)}={\rm S}_{\bar{\nu}1}$, and for the second half period ${\rm S(t)}={\rm S}_{\bar{\nu}2}$, where the background portion of the spectrum is the same as in ${\rm S}_{\bar{\nu}1}$, and the Raman peak has shifted by $\delta \bar \nu $. Thanks to the pixel-level demodulation, the lock-in camera outputs directly the difference spectrum δS.
Fig. 1. Scheme of operation of lock-in RDS. (a) The laser source is periodically driven between two wavenumbers ${\bar{\nu}_1}$ and ${\bar{\nu}_2}$ over a period T. The spectrum S(t) emitted by the sample is consequently time dependent and periodically modulated over the same period, switching from ${\rm S}_{\bar{\nu}1}$ to ${\rm S}_{\bar{\nu}2}$. The lock-in camera demodulates this signal and outputs directly the difference spectrum δS. (b) Internal operation of a pixel of the lock-in camera. The local oscillator (LO) generates two periodic signals in quadrature, which are multiplied with the input modulated spectrum. The resulting signals are low-pass filtered (LPF) and summed, giving the two outputs I and Q. If the modulation of the input spectrum takes place at the same period and phase of the local oscillator signals, then I and Q are identical, and equal to the difference spectrum δS.
The internal operations performed at the pixel level are showed in more details in Fig. 1(b). A local oscillator generates two periodic signals in quadrature, which are made of a sequence of [1, 0, -1]. The modulated optical signal received by each pixel, which represents a portion of the emitted spectrum, is multiplied to each local oscillator signal. The resulting signals are then low-pass filtered and summed, giving the two output frames I and Q. It is possible to see that, if both the modulated spectrum and local oscillator are periodic over a period T, and in phase, both I and Q are given by the difference between the two shifted spectra, so effectively they are the difference spectrum δS.
What is described in Fig. 1 is a scheme that is capable of removing the fluorescence and ambient light background similarly to SERDS. The same scheme could also be used with a single laser, which is switched on and off over the period T. In this case, the spectrum during the first half interval is given by the Raman peaks and any background present, while during the second half period the spectrum would correspond only to the ambient light. As a consequence, the difference spectrum output by the camera would be the Raman spectrum without ambient light contributions. Obviously this cannot remove emissions that depend on the presence of laser excitation, such as fluorescence, but if this source of background is negligible compared to ambient light, this scheme of operation would be advantageous since it directly gives the Raman spectrum rather than the difference spectrum.
Another possible way to use this camera is with four wavelengths, each on during a quarter period. Two differential spectra would be obtained, providing extra information to improve the recovery of the Raman peaks. While the use of multiple wavelengths has been explored by other researchers [27,28], the use of multiple wavelengths with lock-in would require a different approach to reconstruct the Raman spectrum, since the lock-in method can only provide differential spectra. Thus, the methods proposed in the literature to extract the Raman spectrum from multiple measurements would not be directly applicable in this case.
The advantage of analog demodulation of the difference spectrum at the pixel level is that the background information is never stored, and thus it does not contribute to filling the well capacity. As a consequence, the exposure time can be increased, causing only the Raman peaks’ intensity to rise. Ideally, the exposure time could be set up to the point where one of the Raman peaks almost fills half of the bit-depth, which would give the best possible SNR with the given detector. The overall intensity of the spectrum still generates noise within the pixels, and it is possible to show that SNR of lock-in RDS improves with respect to the standard case as [23]:
3. Experimental
3.1 Raman setup
The setup used for the experiments is shown in Fig. 2. Since two lasers were used, they needed to be aligned onto the same path using a beam splitter. In order to maximize power efficiency, we used a polarizing beam splitter (Thorlabs, CM1-PBS251) and two half-wave plates (Thorlabs, WPH05M-633) to rotate the polarization of each laser. A band-pass filter (BrightLine, FF01-640/14) is used to remove some amplified spontaneous emission from the lasers at longer wavelengths. The laser light is then transmitted through a dichroic mirror (Thorlabs, DMSP650) and focused onto the sample by a 60x objective with 0.8 NA. The same objective is also used to collect light emitted by the sample, and the dichroic mirror reflects it toward the spectrograph. While the dichroic mirror mostly doesn’t reflect the laser light, the remaining small percentage is still stronger than the Raman emission, so a long-pass filter (Thorlabs, FEL0650) is used to remove this remaining disturbance. The spectrum is obtained with a Horiba grating spectrograph with 25 cm focal distance; the grating (Horiba, 510 14-R) is blazed at 750 nm and has 600 grooves/mm. A cylindrical lens is used to focus the spectrum over a thin line at the lock-in camera’s detector.
Fig. 2. Setup for lock-in RDS. Two lasers are combined on the same path using a Polarizing Beam Splitter (PBS). In order to maximize transmission and reflection respectively, the polarization of each laser is tuned using a half-wave plate (λ/2). A Band-Pass Filter (BPF) is used to remove some spurious emissions from the lasers at longer wavelengths. The laser light is transmitted through a Dichroic Mirror (DM) and focused on the sample by a 60x objective (Obj) with 0.8 NA. The light emitted by the sample is collected by the same objective, and reflected by the DM. A Long-Pass Filter (LPF) rejects remaining laser light. The different wavelengths are then dispersed by a Horiba spectrograph with focal length of 25 cm. A cylindrical lens (Cyl) focuses the spectrum to a line that occupies few pixels on the detector of the lock-in camera.
The laser diodes used in this setup are Thorlabs’ HL63142DG, with nominal wavelength of 637 nm and maximum power of 100 mW. In order to obtain slightly spaced wavelengths, we used the LDM56 temperature control mounts and set their temperatures to different values, observing experimentally optimal differentiation of the Raman peaks of interest. The switching of the lasers was controlled by a National Instruments acquisition card, which also generated the triggering signal for the lock-in camera.
3.2 Adapting the lock-in camera for Raman experiments
As mentioned previously, it is expected that the analog demodulation of the difference spectrum would bring the advantage of longer exposure times over multiple short exposure frames, in line with what has been demonstrated for DPC [23]. On the other hand, the current model of the lock-in camera was not developed for this application and, as such, is not optimized for the low signals related to Raman spectroscopy: even the fluorescence component is typically several orders of magnitude weaker than the excitation source, and is subject to saturation and quenching. As a consequence, the maximum period of one demodulation cycle on the C3 detector is limited to approximately 470 µs. This is extremely short compared to typical Raman spectroscopy experiments, which can last up to several minutes. Moreover, the aspect ratio of the camera is square, since it is meant for imaging, whereas only a thin line of pixels is actually being used for this spectroscopy application. It is possible to set the detector to average up to 510 cycles internally, practically increasing the effective exposure time. However, in our experiments, the performance in terms of SNR seemed to deteriorate for more than 32 cycles, so we assumed this to be the maximum effectively usable.
In this manuscript, we will use samples that are known to be strong emitters, like diamond powder, so that a Raman signal may be picked up even in such short exposure time conditions. Where necessary, we will sum up multiple frames of lock-in RDS until saturation of the 10-bit scale is virtually close to being reached. Assuming that the camera were to allow such long exposure times, this would be the limit at which we would need to stop our measurement, and thus it would give the SNR to compare with standard RDS [23]. In the manuscript, extrapolated results will be pointed out for clarity, with respect to purely experimental results. It is important to note that the results extrapolated from summing multiple frames are expected to be slightly worse than the equivalent single long exposure, due to extra digitization noise present in each frame, so these may be interpreted as worst case scenario.
The goal of the present manuscript is to demonstrate a proof of concept to improve single-shot SNR in RDS, given the high interest present around Raman imaging [14]. The benefits of this technique may be enhanced in the future with a tailored version of the helicam detector.
3.3 Single-laser RDS
The first experiment demonstrates the use of the lock-in camera for the analog removal of background light that is not excited by the laser source itself, such as ambient disturbances. The spectrum resulting from this experiment is expected to show normal Raman peaks.
The sample used to emit Raman signal was diamond powder. We focused in particular on the 1332 cm-1 peak [29]. For this experiment, we introduced an external disturbance in the form of a smartphone torch light shined directly onto the detector.
In this experiment, we first collected a measurement of the spectrum by setting the lock-in camera in intensity mode, which essentially operates as a standard detector. The exposure time was limited to 0.512 ms in order to avoid saturation of pixels caused by the external disturbance. This spectrum is shown in Fig. 3(a). The configuration of the sources, the laser and the torch, is depicted in the box above the graph. The dashed black spectrum shows the reference spectrum of the diamond powder, but it is not actually possible to see the diamond line at the same location in the blue spectrum because of the torch light which is much more powerful. The peak at 1700cm-1 is not a Raman peak but a hot pixel. Moreover, as the torch light did not go through the spectrograph, it is important to highlight that we are not observing its spectrum, but just a disturbance in the form of additional power. A mechanical holder partially shadowed some of the torch light, which is why the power is lower around the pixels at 1600 cm-1. We then collected an image of the disturbance only, with the same exposure time, and the laser being turned off, which is shown in Fig. 3(b). Since the Raman emission was very small compared to the background, Fig. 3(a) and Fig. 3(b) look almost the same.
Fig. 3. Single laser background removal experiment. (a) Standard measurement of the Raman spectrum with external disturbance. The box above the spectrum shows the measurement configuration, with both the laser source and torch on. (b) Standard measurement of the external disturbance. The box above the spectrum shows the measurement configuration, with only the torch on. (c) Comparison between the standard Raman spectrum after digital background subtraction, and the lock-in Raman measurement. The blue standard spectrum was obtained from the difference between spectrum (a) and (b). The orange lock-in spectrum was obtained by switching the laser on and off periodically, synchronously to the lock-in camera local oscillator. In all graphs, the dashed black spectrum shows the reference diamond powder spectrum measured with a Renishaw inVia Raman microscope using a 600 gr/mm grating, 532 nm laser at 0.5 mW, and 10 integrations of 1s each.
The spectrum resulting from the difference between these two measurements is shown in Fig. 3(c), in blue. Due to the very short exposure time, the Raman peak is almost imperceptible in this measurement.
For the lock-in measurement, we set the duration of one quarter period to the maximum allowed of 0.119 ms, and the number of cycles averaged by the detector to 32. In total, this gives an equivalent exposure time of 3.8 ms. The laser was on during the first half period, and off during the second half period. In this way, the lock-in procedure effectively removes the external disturbance at the analog level. The Raman peak was thus the only signal present in the measurement, and its magnitude was only 7 gray levels out of the 1024 available. This means that, if the camera allowed a longer cycle duration, it would be possible to integrate for a much longer time without saturation. This extrapolated result is shown by the orange plot in Fig. 3(c), which is given by the sum of 80 frames, effectively an exposure time of 228 ms. The quality of this spectrum is clearly improved with respect to the standard measurement. The diamond line is quite broad, but its wavenumber width appears consistent in both standard and lock-in measurements as well as the reference measurement of the same sample obtained with the Renishaw inVia Raman microscope, which means it must be linked to the sample itself since we used diamond powder instead of a whole crystal.
These results are summarized in Table 1. The exposure times, their square roots, and the SNR for both lock-in and standard RDS are given in the table, in the first two rows. The last row indicates the ratio of the lock-in and standard measurement results. Note that two values are reported for the lock-in case. In the white cells are the experimental values from a single frame. In the orange cells, the values are obtained from the sum of 80 frames, showing the experimentally extrapolated result with a potential longer exposure time. As expected from the theory and Eq. (5), the ratios of the SNRs correspond to the ratios of the square roots of the exposure times. The same result is also shown in Fig. 4, that shows the SNR with respect to the exposure time, for both experiments and the extrapolated point. It is possible to see that indeed, the slightly longer exposure time of the lock-in measurement allowed to improve the SNR by the expected value given in Eq. (5). The extrapolated point (in green) could be reached in a single shot in lock-in, with a detector that allows a longer exposure time, without saturation.
Fig. 4. SNR graph with respect to the exposure time. The orange line shows the curve proportional to the exposure time square root. The yellow data point refers to the standard measurement. The blue datapoint refers to the lock-in measurement. The green datapoint refers to the extrapolated result for a potential longer lock-in measurement. The error bars on the experiment are from the relative error over 50 measurements.
Table 1. SNR comparison for single laser RDS
3.4 Double-laser RDS
In this section, we demonstrate the use of the lock-in camera for fluorescence removal in RDS. The sample used for this set of experiments is a mixture of diamond powder and oil red dye diluted in sunflower seed oil as a fluorescent disturbance. Two laser sources were used in this case. The diode is the same for both, but their temperature was set to 22°C and 30°C, respectively. In this way, the wavelength shift between the two lasers is enough to clearly distinguish a positive and negative peak in the difference spectra.
For the standard case, we collected a spectrum for each laser, with an exposure time of 2.56 ms to avoid saturation caused by the fluorescence background. The two standard spectra are shown in Fig. 5(a) and (b), which also show the corresponding configuration in the boxes above. The two spectra are very similar, as expected, and the theoretical location of the diamond peak appears to shift slightly with respect to the fluorescence curve. The standard RDS spectrum is shown in Fig. 5(c), where it is compared to the lock-in RDS spectrum.
Fig. 5. Comparison of standard and lock-in double laser RDS, with a sample made of a mixture of diamond powder and oil red. (a) Spectrum of the diamond powder and oil red mixture obtained with the first laser. (b) Spectrum of the diamond powder and oil red mixture obtained with the second laser. (c) Comparison of the difference spectra in lock-in (orange) and standard (blue) mode. The standard RDS spectrum is obtained as the difference between spectrum (a) and (b). The lock-in RDS spectrum is obtained by switching periodically the two laser sources synchronously to the local oscillator of the camera. In all graphs, the black dashed spectrum is the reference diamond powder spectrum.
In the lock-in experiment, the quarter period was again set to 0.119 ms with 12 cycles being averaged. This gives an equivalent exposure time for one laser of 1.41 ms. The orange plot in Fig. 5(c) shows the sum of 100 lock-in RDS spectra. As in the previous set of experiments, even at the maximum exposure time, the lock-in camera is not saturated, thanks to the analog removal of the background. If it were possible to set a 100 times longer exposure time, the 10 bit detector would still not be saturated, as appears from Fig. 5(c).
With discrete integration we obtain the reconstructed Raman spectra for lock-in and standard RDS, which are shown in Fig. 6. The quality of the diamond peak in terms of noise appears to be higher for the lock-in case as expected, despite the uncovering of a spurious signal at 1500 cm-1. Most likely, this is related to the presence of some residual fluorescence signal and caused by a mismatch between the two lasers emission. More spectrally pure lasers would not incur such issue. In terms of quantitative analysis, Table 2 summarizes the exposure times and SNRs of the difference spectra, for a single standard RDS, a single lock-in RDS, and the extrapolated result for a 100 times longer exposure in lock-in RDS. In this case, there appears to be a 10% mismatch between the expected values of SNR ratio and calculated ones. This might be due to some additional quantization noise, as the Raman peaks were only a few gray levels in this case.
Fig. 6. Normalized, integrated spectra for lock-in and standard RDS, obtained from the difference spectra of Fig. 5(c).
Table 2. SNR comparison for double laser RDS
The same results are again shown in Fig. 7, which that for double-laser RDS we still obtain the same SNR behavior with respect to the exposure time. Of course, since the exposure time was slightly shorter for the lock-in measurement, the SNR is degraded. On the other hand, the extrapolated point shows a possible great improvement with a longer exposure time.
Fig. 7. SNR graph with respect to the exposure time. The orange line shows the curve proportional to the exposure time square root. The yellow data point refers to the standard measurement. The blue datapoint refers to the lock-in measurement. The green datapoint refers to the extrapolated result for a potential longer lock-in measurement. The error bars on the experiment are from the relative error over 50 measurements.
4. Discussion and conclusion
In this manuscript, we have demonstrated the use of a lock-in camera paired with synchronized wavelength-shifted switching lasers to obtain difference Raman spectra. In particular, we showed how this method can effectively remove both external disturbances and fluorescence background through analog demodulation. This allows to obtain difference spectra that are not plagued by saturation, and the whole bit depth available can be used to encode the relevant Raman peaks.
The lock-in camera available for this work, the helicam C3, is not optimized for low signals and long exposure times, and its square aspect ratio is intended for imaging applications rather than spectroscopy. Despite this limitation, we have demonstrated that the SNR of lock-in RDS does scale compared to the standard measurement as the square root of the exposure times ratio.
In the first set of experiments in Section 3.3, we have obtained a factor of 2.78 improvement in SNR, limited by the maximum cycle duration of the camera. On the other hand, with longer exposure time, it would have been possible to improve to SNR by up to a factor of 22, as shown by the extrapolated result in Table 1.
In Section 3.4, we also demonstrated a more classic RDS application for fluorescence removal, where a long exposure time would have allowed a factor of 6.8 improvement.
Next to our experimental results, the extrapolations show how an effective system for single-shot, high-sensitivity Raman can be constructed based on the principles presented here. Indeed, the limited exposure time in standard RDS due to saturation means that in order to reach better SNRs, multiple frames have to be averaged [23]. In the case of an SNR improvement by a factor of 20, such as in Section 3.3, 400 frames are required. For a single point measurement, this may be feasible, but current trends in biological imaging show that Raman spectroscopy is most valuable when paired with microscopic imaging, providing hyperspectral information with a high spatial resolution. The method proposed in this manuscript allows practical shifted excitation measurements even in this case, thanks to data processing requirements that are several orders of magnitude lower than digital difference methods.
To conclude, the results in this manuscript demonstrate the technical feasibility of lock-in RDS in at least two configurations. Experimental results validate the expected scaling of SNR, while extrapolated result show the future potential of this technique for Raman spectroscopy in biology.
Acknowledgements
The authors would like to thank Dr. P. Lambelet for the helpful discussions and support in the use of the helicam.
Disclosures
The authors declare no conflicts of interest.
Data availability
Lock-in and standard Raman spectra will be made available upon publication.
References
1. R. R. Jones, D. C. Hooper, L. Zhang, D. Wolverson, and V. K. Valev, “Raman Techniques: Fundamentals and Frontiers,” Nanoscale Res. Lett. 14(1), 231 (2019). [CrossRef]
2. P. Rostron, S. Gaber, and D. Gaber, “Raman Spectroscopy, a review,” International Journal of Engineering and Technical Research 6, (2016).
3. C. V. Raman and K. S. Krishnan, “Polarisation of Scattered Light-quanta,” Nature 122(3066), 169 (1928). [CrossRef]
4. V. D’Ippolito, G. B. Andreozzi, D. Bersani, and P. P. Lottici, “Raman fingerprint of chromate, aluminate and ferrite spinels,” J. Raman Spectrosc. 46(12), 1255–1264 (2015). [CrossRef]
5. A. Kudelski, “Analytical applications of Raman spectroscopy,” Talanta 76(1), 1–8 (2008). [CrossRef]
6. N. Emmanuel, R. B. Nair, B. Abraham, and K. Yoosaf, “Fabricating a Low-Cost Raman Spectrometer to Introduce Students to Spectroscopy Basics and Applied Instrument Design,” J. Chem. Educ. 98(6), 2109–2116 (2021). [CrossRef]
7. H. J. Butler, L. Ashton, B. Bird, G. Cinque, K. Curtis, J. Dorney, K. Esmonde-White, N. J. Fullwood, B. Gardner, P. L. Martin-Hirsch, M. J. Walsh, M. R. McAinsh, N. Stone, and F. L. Martin, “Using Raman spectroscopy to characterize biological materials,” Nat. Protoc. 11(4), 664–687 (2016). [CrossRef]
8. R. B. Versteeg, J. Zhu, P. Padmanabhan, C. Boguschewski, R. German, M. Goedecke, P. Becker, and P. H. M. van Loosdrecht, “A tunable time-resolved spontaneous Raman spectroscopy setup for probing ultrafast collective excitation and quasiparticle dynamics in quantum materials,” Struct. Dyn. 5(4), 044301 (2018). [CrossRef]
9. B. Durrant, M. Trappett, D. Shipp, and I. Notingher, “Recent developments in spontaneous Raman imaging of living biological cells,” Curr. Opin. Chem. Biol. 51, 138–145 (2019). [CrossRef]
10. Z. Q. Li, Z. Li, Q. Chen, A. Ramos, J. Zhang, J. P. Boudreaux, R. Thiagarajan, Y. Bren-Mattison, M. E. Dunham, A. J. McWhorter, X. Li, J. M. Feng, Y. P. Li, S. M. Yao, and J. Xu, “Detection of pancreatic cancer by convolutional-neural-network-assisted spontaneous Raman spectroscopy with critical feature visualization,” Neural Networks 144, 455–464 (2021). [CrossRef]
11. D. Wei, S. Chen, and Q. Liu, “Review of Fluorescence Suppression Techniques in Raman Spectroscopy,” Appl. Spectrosc. Rev. 50(5), 387–406 (2015). [CrossRef]
12. J. Kostamovaara, J. Tenhunen, M. Kögler, I. Nissinen, J. Nissinen, and P. Keränen, “Fluorescence suppression in Raman spectroscopy using a time-gated CMOS SPAD,” Opt. Express 21(25), 31632–31645 (2013). [CrossRef]
13. A. J. Driscoll, M. H. Harpster, and P. A. Johnson, “The development of surface-enhanced Raman scattering as a detection modality for portable in vitro diagnostics: progress and challenges,” Phys. Chem. Chem. Phys. 15(47), 20415–20433 (2013). [CrossRef]
14. R. Strack, “Imaging without the labels,” Nat. Methods 19(1), 30 (2022). [CrossRef]
15. A. P. Shreve, N. J. Cherepy, and R. A. Mathies, “Effective Rejection of Fluorescence Interference in Raman-Spectroscopy Using a Shifted Excitation Difference Technique,” Appl. Spectrosc. 46(4), 707–711 (1992). [CrossRef]
16. P. A. Mosier-Boss, S. H. Lieberman, and R. Newbery, “Fluorescence Rejection in Raman Spectroscopy by Shifted-Spectra, Edge Detection, and FFT Filtering Techniques,” Appl. Spectrosc. 49(5), 630–638 (1995). [CrossRef]
17. B. B. Praveen, M. Mazilu, R. F. Marchington, C. S. Herrington, A. Riches, and K. Dholakia, “Optimisation of wavelength modulated Raman spectroscopy: towards high throughput cell screening,” PLoS One 8(6), e67211 (2013). [CrossRef]
18. B. B. Praveen, P. C. Ashok, M. Mazilu, A. Riches, S. Herrington, and K. Dholakia, “Fluorescence suppression using wavelength modulated Raman spectroscopy in fiber-probe-based tissue analysis,” J. Biomed. Opt. 17(7), 0770061 (2012). [CrossRef]
19. M. A. D. Martins, D. G. Ribeiro, E. A. P. dos Santos, A. A. Martin, A. Fontes, and H. D. Martinho, “Shifted-excitation Raman difference spectroscopy for in vitro and in vivo biological samples analysis,” Biomed. Opt. Express 1(2), 617–626 (2010). [CrossRef]
20. M. Chen, N. McReynolds, E. C. Campbell, M. Mazilu, J. Barbosa, K. Dholakia, and S. J. Powis, “The use of wavelength modulated Raman spectroscopy in label-free identification of T lymphocyte subsets, natural killer cells and dendritic cells,” PLoS One 10(5), e0125158 (2015). [CrossRef]
21. K. Sowoidnich, M. Towrie, and P. Matousek, “Lock-in detection in Raman spectroscopy with charge-shifting CCD for suppression of fast varying backgrounds,” J. Raman Spectrosc. 50, jrs.5597 (2019). [CrossRef]
22. A. C. De Luca, M. Mazilu, A. Riches, C. S. Herrington, and K. Dholakia, “Online fluorescence suppression in modulated Raman spectroscopy,” Anal. Chem. 82(2), 738–745 (2010). [CrossRef]
23. C. Bonati, D. Loterie, T. Laforest, and C. Moser, “Lock-in incoherent differential phase contrast imaging,” Photonics Res. 10(1), 237–247 (2022). [CrossRef]
24. S. Beer and P. Seitz, “Real-time tomographic imaging without x-rays: A smart pixel array with massively parallel signal processing for real-time optical coherence tomography performing close to the physical limits,” 2005 PhD Research in Microelectronics and Electronics, Vols 1 and 2, Proceedings, 335–338 (2005).
25. S. Beer, P. Zeller, N. Blanc, F. Lustenberger, and P. Seitz, “Smart pixels for real-time optical coherence tomography,” in Electronic Imaging, Three-Dimensional Image Capture and Applications VI 2004), 21.
26. S. Beer, S. Waldis, and P. Seitz, “Video-rate Optical Coherence Tomography Imaging with Smart Pixels,” in Optical Coherence Tomography and Coherence Techniques, (SPIE, 2003),
27. J. B. Cooper, M. Abdelkader, and K. L. Wise, “Sequentially Shifted Excitation Raman Spectroscopy: Novel Algorithm and Instrumentation for Fluorescence-Free Raman Spectroscopy in Spectral Space,” Appl. Spectrosc. 67(8), 973–984 (2013). [CrossRef]
28. S. T. McCain, R. M. Willett, and D. J. Brady, “Multi-excitation Raman spectroscopy technique for fluorescence rejection,” Opt. Express 16(15), 10975–10991 (2008). [CrossRef]
29. R. S. Krishnan, “Raman Spectrum of Diamond,” Nature 155(3928), 171 (1945). [CrossRef]
