FPGA lossless fixed-point

CoffeeOrTea

FPGA lossless fixed-point [Copy link]

I took a DJI written test a few days ago. One of the questions was how to do lossless fixed-point conversion of 12.918. What is the minimum bit width? I looked at other people's answers online, but they all seemed wrong, such as the following one. With 12 bits, the minimum precision that can be represented is 1/(2^12), 12.918/(1/(2^12)) = 12.918*(2^12), the result is not an integer, so it cannot be losslessly fixed-pointed. But according to this, no matter how many bits, this number cannot be losslessly fixed-pointed, right? So did I understand something wrong?

coyoo

It depends on the specific definition of "lossless". If no error is allowed, then the method you mentioned will not work. However, fixed-point conversion is generally achieved by sampling the method you mentioned. It can only be said that the larger the bit width, the higher the accuracy.