Hall rotary encoder tempco--wisdom?

In this case, it was just measuring the right thing. Which is good, because the clinical trial starts in a week. ;)

Cheers

Phil Hobbs

All good news. But for the future, the cheapest workable optical encoders I know of are from US Digital: .

Joe Gwinn

The mag encoder is from US Digital as well, thanks.

Cheers

Phil Hobbs

Golly, is it supposed to be an absolute shaft encoder? I would have never tried to use a Hall device for that, only a very expensive optical encoder.

spectrometer. The SNR is pretty good, way over 60 dB (which is good for a SWIR spectrometer).

sort of swimming around a bit, i.e. the gross shape stays roughly the same, and the small scale noise is low, but there are small systematic variations on scales of 1/10 to 1/2 of the scan range, where multiple spectra don't quite line up with each other.

unstable with temperature, something like 3 arc min per kelvin. Since the grating only has to rotate about 8 degrees for the whole measurement, that winds up being, like, 3 nm/K, which is _horrible_.

tried to use a Hall device for that, only a very expensive optical encoder.

The data sheet says that's exactly what it is, and since it is now working fine in Phil Hobb's application (now that Phil's software is looking at the output in the right way) one has to suspect that your instincts need re-calibrating.

Yes, and once you measure the right thing, it works great--it stays stable to less than 1 count in 4096 even when you really bomb it with freeze spray. There's probably more sinusoidal error than you'd get in a sufficiently-carefully-aligned optical encoder, but for the price, these things are actually magic.

Cheers

Phil Hobbs (Coming to you live from the Delta Sky Club at ATL.)

stable to less than 1 count in 4096 even when you really bomb it with freeze spray. There's probably more sinusoidal error than you'd get in a sufficiently-carefully-aligned optical encoder, but for the price, these things are actually magic.

I once had to put together a do-it-yourself optical encoder, and alignment wasn't a problem. The moving disk had a nice regular pattern of lines, and the fixed moire-grid that we stuck between the LEDs (on one side, so that t he light went from the LEDs, through the fixed grating and the moving disk before it hit the photo-diodes) had an in-phase area and a quadrature area, both imprinted at the same time

The lines on the fixed grating had to be close to parallel to the lines on the rotating disk, but that wasn't difficult to get right, and the two grat ing had to stay close enough that the shadows of the lines on the fixed gra ting didn't have much of a penumbra on the rotating grating, but that wasn' t too hard either.

I'd been exposed to a system earlier where the in-phase and quadrature dete ctors had to be separately aligned, and the system stopped working wherever anybody knocked either one, but that wasn't a particularly clever design. We dumped that system as soon as we had the chance.

Eccentricity is the main source of sinusoidal error. As a thought experiment, if the wheel rotates through a point near its edge, the apparent rotation is twice the true rate on the side opposite the pivot and zero on the side closest to it. For smaller eccentricities you get an error of

Delta_theta ~ delta_x/R sin theta

so for a 20 mm diameter wheel, to get the sinusoidal error down to (2 pi)/4096 you have to centre it to an accuracy of

delta_x < (2 pi/4096)(10 mm) = 15 microns,

including all the runouts and angular eccentricities. That's far from impossible, but doesn't happen by accident.

Cheers

Phil Hobbs

You must be referring to VRS type sensors ? If so, we deal with that by a machinist do in-place turning to correct for the problem.

Jamie

able to less than 1 count in 4096 even when you really bomb it with freeze spray. There's probably more sinusoidal error than you'd get in a suffici ently-carefully-aligned optical encoder, but for the price, these things ar e actually magic.

ent wasn't a problem. The moving disk had a nice regular pattern of lines, and the fixed moire-grid that we stuck between the LEDs (on one side, so th at the light went from the LEDs, through the fixed grating and the moving d isk before it hit the photo-diodes) had an in-phase area and a quadrature a rea, both imprinted at the same time.

s on the rotating disk, but that wasn't difficult to get right, and the two grating had to stay close enough that the shadows of the lines on the fixe d grating didn't have much of a penumbra on the rotating grating, but that wasn't too hard either.

detectors had to be separately aligned, and the system stopped working wher ever anybody knocked either one, but that wasn't a particularly clever desi gn. We dumped that system as soon as we had the chance.

iment, if the wheel rotates through a point near its edge, the apparent rotation is twice the true rate on the side opposite the pivot and zero on the side closest to it. For smaller eccentricities you get an error of

)/4096 you have to centre it to an accuracy of

mpossible, but doesn't happen by accident.

A very useful and helpful comment, but that is "centering", rather than "al ignment". If you were using "alignment" in the broad-brush sense of getting all the geometry right, it does - broadly - makes sense, but I understand "alignment" in the narrow sense of "getting all the linear elements paralle l and in the right places" and your original post left me with a sensation of cognitive dissonance.

Any optical sensor based on multiple lines exhibits this effect. You probably aren't making the wheel and the encoder on the same lathe, so tolerance buildup is still a problem.

Cheers

Phil Hobbs

Glad to be of service. ;)

The generally unknown mutual orientation of the shaft and encoder produces a zero-order error, i.e. an offset. We usually expect to calibrate that out.

Eccentricity is the only sort of misalignment that causes a first-order error term with good encoders. Tilts and axial translation are second-order terms with encoders such as the Avago ones we've been talking about. Being a physicist, I know that only the leading order term matters. ;)

Cheers

Phil Hobbs

Well, I made it back from my trip to SoCal. For a simple trip from NY to LA and back, I wound up having boarding passes or seat requests for _nine_ airplanes. Two snowstorms back to back, then fog delaying the RJ three hours on the last leg.

However, it was well worth it, because the pre-production gizmos work really well now, even better than my original prototype.

We did a bunch of lab tests for glucose, 14 of us doing two scans before lunch and two after, on each of three machines (168 scans in all), compared with the finger stick + strip reader method. From essentially zero correlation last time, we've now got about the best r**2 ever seen for a noninvasive glucose tester, and that's against the strip tester rather than the true blood number. (Strip testers are a bit on the scattergun side.)

There were about half a dozen things wrong with the design, mostly in the Labview 'software'. The only remaining significant bit of buried treasure that we couldn't fix in time for this week's real trial is to redesign the grating drive to get more resolution and more repeatable motion.

So we may actually have the first successful transcutaneous glucose detector ever, which would be pretty cool.

Thanks again, all.

Cheers

Phil Hobbs

Wow! If that is small and cheap enough it would be fantastic! Congratulations!

A lot of people will appreciate that. People have told me that they don't mind the insulin shots as much as the finger pricks.

You'd owe me a beer if you hadn't already given us some really nice beers.

I definitely owe you a beer. Or maybe two, then I'd get to see you fall over. ;)

Cheers

Phil Hobbs

Thanks. The idea is to start from the higher-value applications like hospitals and doctors' offices, then things like visiting nurses, and eventually individual diabetics.

Cheers

Phil Hobbs

Yeah! Big Grins!

Too bad about labview.. (too expensive) I was dreaming about a monochromater into a fiber today. My problem has always been how to do the rotation stage cheaply. I always get stuck at a worm gear onto a big spur.. and then backlash. (or is the spur the small gear?)

Well of course it's a pleasure to have both your questions and solutions.

George H.

Where the hell is Global Warming when you need it? ;-)