Twigg
Science Advisor
Gold Member
- 893
- 483
Sorry for the confusion. I'm saying at 964nm, the linewidth of the cavity will be bigger than 30kHz. Since you have finesse of 10,000 at 1064nm, that means your reflectivity must be at least 99.99%, and probably much better. It depends what kind of mirror coating you use, but the reflectivity will probably decrease significantly as you go farther from the design wavelength. Lower reflectivity means broader linewidth.kelly0303 said:Sorry I am a bit confused. What I meant was, don't I need the linewidth of the laser to be smaller than the cavity linewidth in order to get a lock? Are you saying that there is a chance that at 964 nm the linewidth of the laser would be smaller than 30 kHz?
I recommended changing wavelengths not as a workaround, but just as a diagnostic tool to see if you could lock the cavity with lower finesse and a different (presumably well-behaved) laser. After all, at the non-ideal wavelength your finesse will be lower so the built-up optical power will be less.
To be honest, in hindsight my suggestion sounds like a lot of work just to do a diagnostic test.
You're scanning the laser much too quickly. Remember that the cavity has a finite bandwidth and acts like a lowpass filter. You're scanning 300kHz in 20us, so your (ideally) 30kHz cavity linewidth will be fully scanned out in ##30\mathrm{kHz} \times \frac{20\mathrm{\mu s}}{300\mathrm{kHz}} = 2\mathrm{\mu s}##. Now ask, will a ##2\mathrm{\mu s}## pulse be distorted by passing through the cavity (or equivalently, a 30kHz lowpass filter)? Well, to not distort the pulse you need a Fourier-limited bandwidth of at least ##\frac{1}{2\mathrm{\mu s}} = 500\mathrm{kHz}##. What you're seeing in your oscilloscope trace is actually a really crude cavity ringdown. You can tell because your transmission line has an exponential tail (totally not a Lorentzian or Gaussian shape, not even symmetric). Slow your scanning rate or your scanning amplitude by an order of magnitude or two (keep turning it down until you get that symmetric Lorentzian shape).kelly0303 said:I also tried to measure in my cavity, when applying the servo on the mirror piezo, the transmission (pink line below) and reflection signal (green light below). Each horizontal line in the grid on the oscilloscope corresponds to about 300 kHz (so the linewidth of the peak would be ~150 kHz, which is not too far from what I would expect theoretically?) and the actual time scale is 20 microsecond per grid.
Essentially, the company tuned the two lasers at some frequency difference and left them there. If the lasers were perfectly stable, you'd get a pure sinusoid beat signal. On a spectrum analyzer, like the image they sent you, a pure sinusoid would look like a single, very narrow peak. The width of this peak would be limited only by the resolution of the spectrum analyzer, which is 10kHz (that's what's meant by "RBW = VBW = 10kHz" at the top of that image). However, you can see that their beat spectrum is a few hundred kHz broad (remember FWHM means the width after a 3dB drop, not halfway down the curve since it's log scale). That means that the relative frequency difference between the lasers is drifting back and forth by a few hundred kHz. Assuming the two lasers drift by a comparable amount (since they're the same make and model of laser), you can conclude that each lasers drifts by an amount that is roughly the same few hundred kHz divided by ##\sqrt{2}##.kelly0303 said:As far as I understand, these beats happen when the 2 lasers have different frequencies, so that plot shows that the 2 lasers do indeed have different frequencies, but I am not sure how does that reflect the temporal stability of each one of them individually (which is what I care about).
I would argue that their measurement isn't definitive here because they have the wrong timescale. According to the image they sent, they did a 50ms acquisition time to get this few hundred kHz broadened beatnote. That doesn't really matter because your servo can handle variations on a 50ms timescale. What matters is the frequency variations at timescales beyond your servo's bandwidth. If they did this same measurement at ##33\mathrm{\mu s} = \frac{1}{30\mathrm{kHz}}## acquisition time (note: this acquisition time is not possible on a spectrum analyzer), I bet the result would be less. (Note: I chose ##33\mathrm{\mu s}## because that's the fastest time that the transmission photodiode signal could change, and that's what you've been locking to in the past.)
The reason why the laser spectrum depends on the acquisition time is because laser frequency noise isn't white noise, it's usually some kind of diffusion process and is pink (1/f). This is why in frequency metrology, we don't just quote linewidths for atomic clocks or ultrastable reference cavities, we give the whole Allan variation curve versus acquisition time (or equivalently, a power spectral density of phase noise).
The thing is, we kind of already have reason to suspect that your laser noise exceeds your locking range above the transmission error signal bandwidth. That's because we see your noise (as in post #11) shows up on timescales comparable to ##33\mathrm{\mu s}##.
I'm optimistic about your prospects for locking to the reflection signal. I found a theory paper that discusses the difference between the transient response of transmission locking and reflection locking. It turns out (different than I remembered) that the reflection signal still behaves a first-order lowpass filter with cutoff frequency at the linewidth, but the transmission signal behaves as a second-order lowpass filter.