I assume that these are Micro-Electro-Mechanical Systems (MEMS) sensors integrated on a chip. The last time I looked at these, a year or two ago, the best once could do, with lots of integration/averaging, was a few arcseconds of precision. You could make the resolution finer but it would not repeat.
If you "average" the reviews that I studied, then everything looks something like this "sensitivity of the order of 1 micron, repeatability of several microns." Of course, we need to honestly say right away that in fact it will be a device that reliably measures with an accuracy of 10 microns, and not very confidently with an accuracy of 5-7 microns. But it's a plug and play chip!
The SCA103 data sheet says that the noise (DC - 100Hz) is 0.0004 degrees/root Hz. So to get 0.1 arcsec =0.00003 degrees of noise means an averaging time T that satisfies 0.0004 deg/sqrt(T/sec) = 0.00003 deg which implies T/sec = (40/3)^2 = 178 so that T~ 3 minutes.
thanks for the info, the commercial MEMS parts are too small, so the resolution and noise performance are not good enough. you can try paralleling several parts together, the noise performance will degrade very close to 1/squareroot N，if 16 parts parallel, the noise could be 0.0001degree, that could reach the digipas DWL9000(0.2arcsec @ <25s readtime), still cheap enough, only consumes more powerThis is a level from the Soviet Union, I don't think you will find any information about it. I haven't even been able to find instructions yet. The model is called Caliber 128. I have a photo of his wiring diagram in poor quality, but now it's all good only for a museum. On ready-made sensors, the levels turn out a little worse, but without problems - judging by the reviews. Just a few elements on the board and a cable, instead of a stack of electronic boards. Here is what my colleagues did with ready-made chips from the store:
№1, from Ukraine
№2, from Russia
Level:
adc:
Maybe the common mode noise are cancelled, like noise from powersource and natural resonance， while the other noises are averaged
thats a great idea. differential light sensors are sensitive, you can even use mirror or prism to extend the pendulum even longer. it is likely easy to achieve nano-rad sensitivity, but i think the repeat and long-term-drift are still difficult
Thats pretty complex. the servoed close loop clinometer could have wider range and bandwidth than the open loop ones, thats a pretty good electronic level that you haveI spent a little time doing the English captions for the Soviet level scheme that I talked about above. Even if you remove the right side, which is responsible for the ADC, averaging the result and displaying information, there are still a hell of a lot of elements. Compared to the Talyvel electrical circuit, this level is several times more difficult. Probably for self-repetition, the classic design of Talyvel will be the most correct prototype.
