On Mon, 20 Sept 2021 at 15:22, Kevin Townsend <kevin.towns...@linaro.org> wrote: > On Mon, 20 Sept 2021 at 15:52, Peter Maydell <peter.mayd...@linaro.org> wrote: >> >> >> Why "lsb" ? >> >> >> > >> > In my head, using LSB seemed more precise since I know exactly what value >> > will >> > be set to the registers, and exactly what I will get back when reading >> > versus passing >> > in a float that's needs to be conveted as a 'best-fit' scenario in 0.125°C >> > steps? >> >> My question was really, "what does 'lsb' mean here"? I would usually >> assume "least significant bit", but that makes no sense in this context. >> > Ha, sorry. Least significant bit, yes. It gets used in sensor and IC > datasheets all > the time and basically means '1 bit', so if the DS says 0.125°C per LSB it > means > that value for 1 bit, or to multiply the integer by the 1 LSB value. > > Conversion factors from raw units to standard SI units are almost always > indicated this way, though, such as 'LSB/Gauss', etc.
OK, that's not a convention I've encountered before. >> Well, given that the device specifically changes the value it >> shows the guest based on guest-programmable gain settings, >> it does seem to me to be more natural to specify the values >> in some way that represents the "real world data" that the >> sensor is measuring. Ideally we would then if/when we add more >> magnetometer implementations have them all use the same units, >> for consistency. This is the first magnetometer we have, so this >> is where we get to pick the convention. > > > Sounds reasonable. > > We have two options here: > > - Gauss (standard SI unit) > - micro Tesla (0.01 Gauss) > > They've both widely used; but I think uT might be slightly more common. > >> >> > In that case, should I accept floating point inputs, however, or stick to >> > integers? >> > When dealing with magnetometers the values can be very small, so it's not >> > the >> > same as a temp sensor where we can provide 2300 for 23.00C. >> >> What sort of range and precision requirements are we talking about >> here? If we can avoid having to use float that would be nice... >> > Well, taking the LSM303DLHC as an example, we have 1100 LSB per Gauss > at a range of +/- 1.3 Gauss, so 1 bit is: 0.0009090909091 Gauss. > > If we use micro Tesla (uT) we get 11 LSB per uT so the smallest value is > 0.09090909091 uT ... which we could represent with something like > 1000 = 1.000 uT > > That gives us +/- 1.3 G = +/- 130 uT = +/- 130,000, for example. > > This would require a 32-bit integer, though, to take into account the full > range of +/-8.1 G (+/- 810 uT) = +/- 810,000. That's OK -- our "int" properties are int64_t. So we could easily fit something like 10000 == 1.0000 uT, in case we might want the extra precision in future. That would be 1,000,000 == 1 G (assuming I haven't messed up my arithmetic ;-)) > There are devices with a much wider range, like the MLX90393, which can > measure up to +/- 50 mT (50,000 uT), so +/-50,000,000. > > That's the largest range I'm aware of personally, with +/- 1-8G (or 100-800 > uT) > the most common. thanks -- PMM
