[Fixed point model and floating Point]

[whizkid123]

Hi

What does it mean - when somebody says fixed point model and floating point model ?

How ti can be understood by a IC design engineer?

--BR

 

[Tunelabguy]

Floating point refers to the type of data representation of real numbers consisting of a mantissa and an exponent. It is the most common way for modern computers to represent and calculate with real-valued numbers. It allows wide dynamic range with a constant resolution. Fixed point refers to data that is represented as integers with an assumed scaling factor. Sort of like saying that we will represent a temperature where 223 corresponds to 22.3 degrees and 250 corresponds to 25.0 degrees. The computer hardware that handles fixed point calculations is just simple integer arithmetic hardware. This allow for fast operations on cheap hardware at the expense of requiring that the application interpret such numbers with the appropriate scaling factor.

 

[whizkid123]

Floating point refers to the type of data representation of real numbers consisting of a mantissa and an exponent. It is the most common way for modern computers to represent and calculate with real-valued numbers. It allows wide dynamic range with a constant resolution. Fixed point refers to data that is represented as integers with an assumed scaling factor. Sort of like saying that we will represent a temperature where 223 corresponds to 22.3 degrees and 250 corresponds to 25.0 degrees. The computer hardware that handles fixed point calculations is just simple integer arithmetic hardware. This allow for fast operations on cheap hardware at the expense of requiring that the application interpret such numbers with the appropriate scaling factor.

Thanks TunelabGuy.

I understand your point on fixed point model.
What if we need to compare two results - one from floating point and one from fixed point. What is the approach we need to follow?

 

[Tunelabguy]

What if we need to compare two results - one from floating point and one from fixed point?

I suppose you would have to convert the results of one to the other format. Conversion between integer and floating point is built-in to most processors, and accessible from a high-level language like C by doing a "cast". But you will have to keep in mind that due to the differences in precision, the converted results may not be exactly the same. There is the potential for round-off error whenever you convert from floating point to integer. Even when you use integers to implement fixed-point interpretations, there are still approximations involved in the calculations as certain low-order bits are discarded. So the result of any DSP calculation can contain some small errors.

 

[whizkid123]

Thanks Tunelabguy ... I got my doubt cleared