linear Hall IC up to 5000 Gauss ?

I think you might reconsider the problem. There are likely many ways to reduce the flux in the vicinity of the sensor itself that would not affect it's function as an accurate sensor of the relatively larger value that is available locally.

Getting repeatable results from one unit to the next has always been a matter of careful positioning. This would not change.

RL

That field value is above the range of permanent magnets, so most preamplified Hall effect devices won't have enough dynamic range; they are built with too much gain.

I've seen raw Hall sensors, but don't know any source of 'em in commerce. Have you considered magnetoresistive sensors instead? DigiKey has half-bridge sensors in that technology, part# AAV001-11 from NVE.

legg a =E9crit :

I agree but it is not that easy to reduce the B field while keeping the response linear. I think it is very difficult in fact, at least in my application which is a current transducer. We measure the intensity of B in four air gaps inside a magnetic core surrounding the cable where the current is to be measured. Moreover, it is not satisfactory to use a very sensitive cell and being forced to reduce the magnetic field. It is like reducing a signal and have it amplified. This would probably inject noise in the system. Thank you for your suggestion anyway

val

val a =E9crit :

Melexis have it in fact ! The part is MLX90251-0, and -0 is for option

0 which are for the less sensitive cells with ranges going up to 800 mT =3D 8000 Gauss. First time I read the datasheet I thought it was value in Gauss and didn't realized it was in mT.

val