you need to design the fixed point format for any stage of your process, especially for floating point. It could vary based on the process you conduct. Better to design your possible / thinkable format of fixed point format. Then compare for their performance, regarding quality vs efficiency.

Hope it helps

cheers

In order to convert any real, signed number into binary (I assume you mean binary fixed point initially), you need to tell the algorithm what the real number is, obviously, but you also have to tell the algorithm how many bits the digital number will use, and the value of the most significant bit. With that information, the conversion works flawlessly.

The attached text file is a subroutine of a program I wrote to convert real numbers into binary and hex, and vice versa. This one subroutine does what you are asking, which is convert a real number into binary. You should be able to follow it, because it's QuickBasic. Almost looks like pseudocode.

Note that any negative values are coded in binary two's complement, which is pretty much the standard way of doing this. (Although I have seen other techniques used.)

So for example, you ask about converting 0.367. First, let's assume you use 8 bits. If the MS (most significant) bit value of that 8-bit number is 128, then ooops, the least significant bit value comes out to 1. No way can you accurately code 0.367 using that scaling! The algorithm I provided will give the best possible result, which is zero.

Okay, so let's stick with 8-bit numbers, but let's make the MSbit value 2. Now the least significant bit value becomes 2 / 27 = 1/64 = 0.015625, so that's more like it.

Using 8 bit words, with MSbit value of 2, your 0.367 = (binary)

0001 0111, the closest value, which comes out to 0.359375

Not close enough? Well then, either use more bits, to widen the range, or redefine the value of the MSbit. Keeping 8 bits, but assigning a value of 0.5 to the MSbit, you get:

0101 1101, so now the closest value comes out to  0.36328125

And so on. You decide how wide the binary number can be, and assign a value to the MSbit. The algorithm computes the best binary equivalent.

Best of luck.