Technology
Dual-comb spectroscopy and asynchronous optical sampling
Both techniques rest on one idea: two pulse trains with slightly different repetition rates sample each other automatically. That removes the mechanical delay stage from ultrafast measurement and the moving grating from spectroscopy, and it maps femtosecond structure onto a timebase ordinary electronics can read.
The shared mechanism
Take a laser at repetition rate f and a second at f + Δf, with Δf small. The first pulse pair arrives coincident. The next pair arrives with the probe displaced by a small amount, because its period is slightly shorter. The pair after that is displaced further. The relative delay therefore walks across the entire interval between pulses, one small step per period, and then repeats.
Three numbers follow immediately. The delay window is 1/f, the full interval between pulses. The window is traversed Δf times per second. And time is magnified by the factor f/Δf, so femtosecond-scale structure arrives at the detector stretched onto a microsecond timebase — which is what makes the detection electronics ordinary.
Two applications of one idea
Time domain
ASOPS — asynchronous optical sampling
Used for pump–probe. The sweeping delay replaces a translation stage. A 60 MHz single-cavity system scans roughly 20 ns of optical delay at high speed — a range no practical mechanical stage reaches, and a scan rate that makes heavy averaging affordable.
Frequency domain
Dual-comb spectroscopy
The interference of two combs down-converts an optical spectrum into the radio-frequency domain, where it is measured with a single fast photodetector. High resolution and broad bandwidth, with no moving parts and no dispersive spectrometer.
The stability problem, and the shared cavity
Everything above assumes Δf is stable. If the two lasers drift relative to each other, the time or frequency axis stretches and compresses unpredictably, and averaging smears the result instead of improving it. The conventional answer is a fast feedback loop locking the two lasers — which reintroduces the cost, complexity and failure modes that removing the delay stage was meant to eliminate.
The alternative is to generate both pulse trains inside one shared cavity. Both combs then experience the same vibration, the same thermal drift and the same environmental noise, so their fluctuations are highly correlated and largely cancel in the difference. The relative timing becomes passively stable, with no active lock and no optical amplification stage. That is the architecture K2 Photonics builds around, and Paeonia represents in Asia.
Where it fits, and where it does not
- Fits: long delay ranges — carrier recombination, acoustic echoes, THz time-domain traces, thermal transport.
- Fits: measurements that need heavy averaging, because thousands of complete scans per second are available.
- Fits: compact or OEM systems that must stay aligned without a specialist on site.
- Fits less well: dynamics complete within a few picoseconds that need the finest available time resolution, where a short mechanical scan may still win.
The trade is moving parts for a fixed step size. The advantage is largest exactly where the mechanical approach hurts most.
The three ways this is built, side by side
The top row is the classic arrangement: one ultrafast laser, a beam splitter, and a variable delay line on a translation stage, with a modulator and a lock-in amplifier to recover the signal. The middle row removes the stage by using two lasers — and pays for it with the locking electronics that keep them synchronised. The bottom row is the shared-cavity approach: one source, no stage, no locking stack, and a signal chain short enough to fit in an instrument.
What "shared cavity" means physically
Both pulse trains are generated inside the same resonator, so they travel through the same optics and see the same vibration and the same thermal drift. Their repetition rates differ by a small offset Δf, which is what sweeps the delay; but because the noise is common to both, it largely cancels in the difference. That is why the mutual coherence is passive and needs no feedback loop.
Specifying an ultrafast system?
Tell us the delay range, the time resolution and the integration constraints, and we come back with the configuration that fits.