Float to 24-bit int conversion

Hemanth Sundaresh

Other Parts Discussed in Thread: TMS320F28069

Hello All,

I am using TMS320F28069 with CCS ver 5.5.0

I having a Float value (32-bit)(range -1 to +1), this needs to be converted to 24-Bit value (signed data).

Could anyone let me know how can I perform this conversion in CCS environment?

Are there any method for doing this other than CLA?

regards,
Hemanth

over 11 years ago

0 Gautam Iyer over 11 years ago

Guru 192875 points

Hi Hemanth,

Hemanth Sundaresh said:
I having a Float value (32-bit)(range -1 to +1), this needs to be converted to 24-Bit value (signed data).

You can try exploiting the IQMath library.

Regards,

Gautam

0 Tjarco Boerkoel over 11 years ago in reply to Gautam Iyer

Expert 1725 points

Just a thought, can't you just typecast this and leave it for the compiler?

float fValue;

int32 iValue;

....

iValue = ( (int32)fValue) >>7;

0 Gautam Iyer over 11 years ago in reply to Tjarco Boerkoel

Guru 192875 points

Tjarco Boerkoel said:
iValue = ( (int32)fValue) >>7;

This should work too :)

0 Norman Wong over 11 years ago

Guru 26430 points

Casting a floating point value of -1.0 to 1.0 would result in an integer value of -1 , 0 or 1. Shifting right by by 7 would result in -1 and 0. You need to scale the floating point value to -2^24 to +2^24 first before casting.

iValue = (int32)(fValue * 8388608.0);

If you've got a FPU, this is fast. If you don't have an FPU, you would have to break apart the IEEE-754 binary representation and process the mantissa. A handful of integer operations but would be faster than a software floating point multiply.

0 Tjarco Boerkoel over 11 years ago in reply to Norman Wong

Expert 1725 points

Hi Norman,

Euh yes Sorry, I read over the fact that the maximums were +/-1. This will be shifted out. Well, leaving it to the compiler is the best way.

Why not just use IQ-format if you just have maximum values of +/-1? IQ24 has same resolution as 32bit float (right?). IqMath is also very powerfull if your numeric range is limited.

0 Hemanth Sundaresh over 11 years ago in reply to Tjarco Boerkoel

Intellectual 380 points

Hello all,

Thanks everyone for your response.

I modified one of the methods suggested in e2e forum

http://e2e.ti.com/support/microcontrollers/c2000/f/171/p/354060/1246582.aspx#1246582

    typedef union{
        struct{
            Uint16 lo;
            Uint16 hi;
        }u16;
        Uint32 u32;
        int32 i32;
        float  f;
    }Versatype32;
    
    void Float_to_I24bit(int *dest, float *src, Uint32 length)
    {
    register Versatype32 data;
    
    	while(length--){
    		data.f = *src;
    		data.i32 = ((int32)(data.f * 8388608) & 0x00ffffff);
    		*dest++ = data.u16.hi;
    		*dest++ = data.u16.lo;
            src++;
    	}
    }

0 Norman Wong over 11 years ago in reply to Hemanth Sundaresh

Guru 26430 points

Masking with 0x00ffffff will corrupt your negative values. For example -0.5 scales to -4194304 or 0xFFC00000. Masking with 0x00FFFFFF gives 0x00C00000 or +12582912.

The TMS320F28069 is Little Endian. Your conversion writes back out in Big Endian. Is that intended?

0 Hemanth Sundaresh over 11 years ago in reply to Norman Wong

Intellectual 380 points

It is fine for my application.

I actually mask with 0x00ffffff & then left shift by 8 then the data is fed to DAC.

DAC is configured to take left justified 24-bit data.

C2000™︎ microcontrollers

C2000 microcontrollers forum

Float to 24-bit int conversion