# fixed point implementation of 1/x

I am trying to use fixed point to implement the function 1/x for the integer from 1 to 2048. My plan is that I use fixed point representation for some picked integers as entries in one lookup table and use linear interpolation obtain missing integers. About picking integers, more integers will be selected at high curvature regions. I am using Matlab to achieve the goal. I have few experience of converting one function from floating point to fixed point. Currently I am working on the first step. It seems it is not an easy task. Can anyone give some ideas or links to useful information?

• It sounds like you have a plan. You need to be more specific about what you're having problems with. May 7, 2015 at 11:43
• My first problem is to how to pick some points from 1 to 2048. I don't hope to use fixed point representation for all integer from 1 to 2048 because it will use too much storage space. May 7, 2015 at 11:51
• It depends on the accuracy you need, the amount of memory you have available, the precision of the fixed-point format that you're using, and so on. If you don't know all of these, then you need to do some more analysis first to make those decisions. May 7, 2015 at 11:54
• It's the first time for me to converting one function into fixed point representation. I hope the error is less 5% of real value. May 7, 2015 at 12:06
• What is the range of values? Are you looking for all values smaller than 1? As example 01 may mean binary 0.1= decimal 1/2=0.5. 001 in this case will be 0.25. Do you want it as two's complement for negative numbers. In other words - are you dividing all values by 2 to the power of 1024 or 2048, dividing all values?
– Moti
May 7, 2015 at 18:35