I am using the ADS1232 but I think this might be a more general question regarding load cell design.
Almost all designs I see will tie Vref to the analog voltage rails. This gives a max range of +/-19.5 mV with maximum gain on the ADS1232.
My load cells will give more like 5 or 6 mV (total swing) over the range I expect to use. The load cell is rated at 37.5 lbs, I only need to go up to 20 lbs. I have a 39mV range to cover a ~7 mV swing. I would think it is rare for a load cell to cover close to the perfect 19.5 mV max value without some kind of analog tinkering to scale things properly.
I would think it should be possible to design a voltage divider for Vref that narrows down the ADC range to span just what I need (plus appropriate margin, of course). A simple voltage divider with three series resistors across the excitation voltage should work, with Vref+ and Vref- connected across the center resistor. In this case, I need my Vref to span about -0.5mV to 6.5 mV, so the ADC result will go from almost the minimum count value to almost the maximum count value over the desired range. This change would give me around two more bits of usable resolution for the price of three resistors.
Likewise, my scale has a dynamic range of 20 pounds, but I expect the pan will weigh something like 6 pounds. If I was really smart I would bias Vref so the scale to covers a range of 6 to 26 pounds on the load cell, because I don't need to count the pan in my dynamic range. The total of load+pan weight must still be under the 37.5 lb limit of the load cell, but that's not a problem here.
Is this a valid way of thinking? I know there are limits as to how narrow the Vref span should be, but 5 volts all the time seems to ignore the fact that scales don't report negative values. Nothing weighs below 0 pounds. I don't need the full negative range, so why not repurpose it into a usable area and normalize it all with calibration and tare? I don't recall seeing anybody mention doing this to properly match load cells ranges to the ADC range.