The biggest problem with fixed point is reciprocals behave very poorly; ie. x·(1/x)=1 and 1/(1/x)=x fail very badly. In order for these to work you obviously ne

The biggest problem with fixed point is reciprocals behave very poorly; ie. x·(1/x)=1 and 1/(1/x)=x fail very badly. In order for these to work you obviously need the same amount of numbers above 1 as below 1, like in floating point. Using fixed point with 11 bits after the decimal, with numbers as low as 100 (18 bits) we already have errors as large as 100·(1/100)=0.977.