Floating point vs. Fixed Point Computations

[chillin4ever]

cell-phone floating point

Which type of computation : Fixed Point or Floating point is preferred in real time applications such as Cell phones and why ?

 

[alzomor]

Hi

Floating point is better because of dynamic range and precision.
but it consumes a lot of computational power and time.
so it's an optimization problem up to you and your architecture and application

Salam
Hossam Alzomor
www(.)i-g(.)org

 

[rsrinivas]

in cell phones also computing comes in different phases.
say it's clarity of video then u may prefer floating point whereas conversation can be fixed point as it can provide a SNR which is good quality as the human ear is 120 db 20 bits fixed point would be far more sufficient
so it's a trade off between fixed and floating

 

[spweda]

All the DSP in the major cellular phone chip use fixed-point DSP, Qualcomm, Nokia and Freescale.

 

[chillin4ever]

Thanks a lot guys ! That makes sense !

 

[ddboy40]

All the DSP in the major cellular phone chip use fixed-point DSP, Qualcomm, Nokia and Freescale.

I think it should be mentioned that the preference of Fixed Poin DSPs to thier Floating Point counterparts in cell phones is only a Cost factor and power desipation. Since cell phones are mass products why spend about $300 per cell Phone per DSP e.g TMS67X when cheaper DSPs could realize thesame functions(but in Software which is slower) and even less baterrie power used. But in Aplications where price/power places no rule at that level say medical signal processing then it becomes clear that floating point DSPs are prefered.

 

[kastantin]

from the precision point float is better
but the floating point DSP is more expensive that floating point DSP
so we use fixed point DSP

 

[chillin4ever]

Thanks guys for all the input.
Appreciate ur comments.

 

[mukul]

In cellular communication only Fixed point is preferred due to ease of implementation(fast) and less power consumption. As all the new devices coming in market are getting smaller and smaller Fixed point is only solution. Very less no. of companies make ploating point DSP's and as per my point of view floating point DSP's will be obsolete in few years.
Now a days 16-32 and even 64bit DSP's are available so this much range is enough in fixed point for any king of signal in real scenarios.

 

[rmreddy]

hi,
its seems that lot of guys here have knowledge abt floating point to fixed point conversion and arihtmetic operations on fixed point data.

it will be helpful if anyone could share some material relating to floating to fixed point conversion and what are the criteria that sholud be taken into account before carrying out mathematical opearations like addition, subtraction, multiplication and division on fixed point data like truncation, round off, overflow, underflow.
a supplemented code is really appreciable ..

seeking ur help ...

 

[andre_luis]

...it will be helpful if anyone could share some material relating to floating to fixed point conversion...

If you use C language, just do a simple type casting.

+++

 

[rmreddy]

hi andre,

i want data to have same values as it had previously before conversion. typecasting actually truncates a decimal value to nearest integer value which i dont want to happen.
i need conversion of floating point data say 0.7071 into Qformat say Q10 like 724 ( 2^10 * 0.7071 = 724 )
and when i do this i am setting aside 10 bits for fractional representation and 5 bits for interger representation and 1 bit for sigh bit.