Thanks, that makes things a lot clearer.

The basic problem seems to be that 16 bits aren't enough to hold the result of the product of a 16 bit number (the limit) and an 8 bit number.

If you have the memory and the CPU cycles to spare, and want the most accurate result, do the whole calculation with a long. After shifting, you will end up back in the integer range.

For example:

unsigned long HC_Gain_temp

HC_Gain_temp = HC_GAIN_HIGH - HC_GAIN_LOW;
HC_Gain_temp = HC_Gain_temp * pot_level;
// Divide by 256 with rounding
HC_Gain_temp = (HC_Gain_temp + (1 << 7)) >> 8;

HC_Gain_scaled = (unsigned int) HC_Gain_temp + HC_GAIN_LOW;

If you don't care about getting the best accuracy and want to fit everything in the existing variables, use only the upper 8 bits of the limit. Beware (!!!), though, that you might run into trouble if the upper and lower limit are too close to each other.

HC_Gain_scaled = ((((HC_GAIN_HIGH - HC_GAIN_LOW) >> 8) * pot_level)); //Average and divide by 256

HC_Gain_scaled += HC_GAIN_LOW;

You might want to round before the right-shift to avoid at least some of the precision loss associated with division.



That sounds like the rounding was left out at some point during the calculations.