Only the new design allows for a correction of response linearity, but even then it's not a trivial task as you'd have to adapt the shape of the flux concentrators and it's hard to predict what the actual result would be. Essentially you could obtain a linear response working by attempts, building new concentrators and measuring each new shape. However, as Mike pointed out, this approach introduces new drawbacks that require better, more expensive materials. There's "drop in" replacement hall sensors, normally used in the car industry, that will cost around 30 US$ each, so if you factor in the time and the bill of materials eventually you are better off buying the rotary hall sensors already available. There's also quadrature chips that offer you perfect linearity, since they compare the readings from sensors placed at 90 degrees rather than reading an absolute. I suspect that these have a digital output, but there might be some that offer an analogue output instead. The older arrangement cannot be linear, it will by necessity read the central portion of a sinusoidal signal. the magnets can be mounted on a non-ferrous material, the classic setup uses a portion of a Bic pen with the magnets glued on the outside and the sensor in the middle. What I am trying to do is to increase the flux by providing a shorter magnetic path for the magnet's free poles.