Spectral watermarking
approach to stimulated Raman spectroscopy – background free femtosecond
vibration spectra
Miroslav Kloz
ELI-Beamlines, Institute
of Physics, Na Slovance 2, 182 21 Praha 8, Czech Republic,
E-mail: miroslav.kloz@eli-beams.eu
Abstract:
Femtosecond stimulated Raman spectroscopy (FSRS) is
plagued by serious baseline issues that lead to spurious peaks. Here we present
a robust approach to FSRS that finally turns it into a versatile tool in
structural biology.
FSRS
experiment was introduced already nearly 20 year ago, despite its numerous
benefits it is still recognized as a relatively exotic technique with only a
handfull of groups worldwide that put it into practice. The reason is in
inseparability of FSRS signal from the transient absorption and other non-liear
backgrounds that bring numerous adjacent technical dificulties such as fixed
patter noise. We developped a “spectra watermarking method” that borows
concepts form the field of information procesing but essentialy function as spectral
modualtion based lock-in detection. Since Raman signal is a form of hihgly
defined spectra shift, such technique can safely extract the Raman signal from
the numerous unwanted nonlinear signals.
Digital watermarking is a technique of
inscribing a subtle signal (a watermark) on top of another, typically much
stronger, signal such as music or video stream (carrier signal). The key
problem is in developing a reliable method to inscribe the watermark in such a
way that it does not substantially obstruct the original signal, and at the
same time can be retrieved with high fidelity independent of the carrier signal
structure, and ideally even sustain signal alteration during transcription to
various formats. This can be achieved by inscribing the watermark as
pseudorandom wavelets. When data mixed with a watermark are convoluted by a
second identical watermark in post-processing, the position of the watermark is
manifested as a sharp localized peak due to constructive interference. With
this information the separation of the watermark from the carrier signal becomes
trivial matrix multiplication.
We recognized an important analogy between
digital watermarking and the Raman experiment. In both cases the goal is to
detect a fine structure on top of a strong, broadband and generally unspecified
background. In the Raman experiment, fluorescence or stimulated emission can be
treated as the carrier signal while the Raman signal itself can be seen as a
watermark. The key problem in current femtosecond frequency domain Raman
spectroscopy (FSRS) is that it depends on employing a spectrally-narrow Raman
excitation that leads to a single specific manifestation of the Raman signal
(upper part of fig. 1). The advantage of such approach is that the recorded
spectrogram represents a direct image of the vibrational spectra. Nowadays,
however, in the era of digitized detection this benefit dropped in importance.
Implicit data can be automatically converted to an explicit signal provided
that the correct routine exists. The traditional approach of improving signal
contrast by repeating the same experiment with a fixed narrow Raman excitation
leads to improvement of the signal, but the background signal is constantly
accumulated as well. If we instead accumulate data by cycling the Raman signal
as pseudorandom watermarks (bottom part of fig. 2) we can in principle recover
only the desired signal. This is indeed possible experimentally by watermarking
the pulses used to generate Raman signal (Raman pulse-pump: “Rp”), which
results in a direct watermarking of the Raman signal as illustrated in Fig. 1-2. Figure 2 show
that when watermarkign is generated as a difference of two complementary
watermarks, it yields aditional benefit
of fixed pattern noise suppression and aditional baseline removal. The
method was already proven valid in applications such as investigation of
carotenoid S* state origin1 or inspecting cofactors in proteins
(fig 3).
Figure
1 (left) Peaks in
the Raman spectra are in fact replicas of spectral shape of the Raman pulse.
When a Raman experiment is performed with broad watermarked pulses the Raman
peaks are manifested as defined watermarks.
Figure 2 (right) Procedure of differential
watermark inscribing: the chosen watermark with zero integral (W0) is divided
into positive (W+) and negative (W-) components. Two experiments are performed
with Rp pulses W+ and W-. The difference of these experiments results in a
signal where Raman peaks carry the W0 watermark while the broadband baseline
and fix pattern noise are simultaneously suppressed.
