Coherent Raman microscope
Two synchronised picosecond beams beating at a molecular vibration, read out three ways: stimulated Raman loss, forward CARS, and epi-CARS.
Open in the canvas →Background
Spontaneous Raman scattering identifies molecules by their vibrational frequencies without any label at all, but it is desperately weak — perhaps one photon in ten million — which makes it far too slow for imaging. Coherent Raman techniques fix that by driving the vibration rather than waiting for it. Two beams, a pump and a Stokes, are overlapped in space and time; when their frequency difference matches a Raman-active vibration, they beat at exactly that frequency and drive the whole ensemble of molecules in phase. The signal that follows is coherent and can be orders of magnitude stronger than the spontaneous one.
Which vibration is addressed is set entirely by the difference between the two wavelengths, so tuning one beam sweeps the spectrum. Written as wavenumbers, the vibration addressed is simply the difference of the two beams' reciprocal wavelengths, and the anti-Stokes light that CARS produces appears at 2/λpump − 1/λStokes — shorter than either input, which is what makes it separable by a filter rather than by lock-in detection.
What this setup demonstrates
Two synchronised picosecond sources: a tunable pump, drawn at 780 nm and tunable across 750–950 nm, and a fixed 1030 nm Stokes. Picosecond rather than femtosecond pulses are the deliberate choice here — their narrower bandwidth matches the width of a Raman line, giving chemical selectivity that a femtosecond pulse would wash out by driving many vibrations at once.
At 780 and 1030 nm the pair addresses a vibration near 3100 cm⁻¹; retuning the pump moves that anywhere across the fingerprint and CH-stretch regions. A retroreflector delay line on the pump arm sets the temporal overlap the whole technique depends on — without it the two pulses simply miss each other — and a short-pass dichroic recombines the beams before the scanner.
The setup reads the interaction out in three independent ways:
- Stimulated Raman loss. An electro-optic modulator and polarizer impose a 20 MHz intensity modulation on the Stokes beam. Where the vibration is driven, energy transfers from pump to Stokes, so the pump comes out slightly depleted and carries that 20 MHz modulation as a copy. The SRS channel therefore sits at 780 nm — the pump — and a beamsplitter picks off part of the modulated 1030 nm Stokes onto a second detector as the reference. Moving the modulation onto one beam and detecting it on the other is what lifts a signal of one part in ten thousand out of the laser noise.
- Forward CARS. The same driven vibration also radiates a new anti-Stokes beam. Selecting the CARS channel shows 628 nm, which is exactly where 2/λpump − 1/λStokes puts it — blue-shifted of both inputs, which is why it can be separated from them by filtering alone.
- Epi-CARS. A second PMT collects the backward-radiated fraction through the excitation objective. It reads the same 628 nm at a much smaller amplitude, because CARS is a coherent process that phase-matches strongly in the forward direction; what returns backwards comes from small or interface-like structures, which is exactly what makes the epi channel informative rather than redundant.
Comparing the two mechanisms in one setup is the real lesson. SRS is a change in a beam you already have, so it scales linearly with concentration and carries no non-resonant background; CARS is a new colour you can filter for cleanly, but sits on a non-resonant background that distorts its lineshape.
The Raman interaction is not computed. The sample generates its CARS and SRS channels as configured signals with set efficiencies, so nothing here derives a lineshape from a susceptibility, and the non-resonant background that complicates real CARS is absent. Retuning the pump changes the anti-Stokes wavelength through the energy relation, but no vibrational resonance is modelled, so the signal does not rise and fall as you tune across a line. Temporal overlap is likewise not enforced: the delay line is drawn and is physically meaningful, but mistiming the two pulses will not extinguish the signal the way it would on a bench. The 20 MHz modulation is applied to the beam and detected, but no lock-in demodulation happens, and the shot-noise-limited sensitivity that makes SRS work in practice is outside what a ray tracer can express.