How to use half.typecast

I have to read in a 16-bit float from a binary file. I used

x = fread(fid,1,'*uint16','ieee-be') since there is no half data type for fread.

I then tried half.typecast(x) to get the number. The number being put in is uint16 65444. It should return -0.640625. I'm getting NaN back. What could I be doing wrong?

Answer 1

[James Tursa]

k = uint16(65444)

k = uint16
65444

dec2bin(k)

ans = '1111111110100100'

half.typecast(k)

ans = 

  half

   NaN

The half precision data type has 1 sign bit, 5 exponent bits, and 10 explicit significand bits. So your bit pattern is:

1 11111 1110100100

When all exponent bits are set and the significand is non-zero, that is a NaN. Note that the payload gets preserved in this conversion as demonstrated by the fact that we can recover the original integer bit pattern:

storedInteger(ans)

ans = uint16
65444

Whereas when you create a half NaN from scratch MATLAB uses a predifined payload with only the leading bit set:

dec2bin(storedInteger(half(NaN)))

ans = '1111111000000000'

Note that when you swap bytes on your original value, you still don't get what you expect:

half.typecast(swapbytes(k))

ans = 

  half

   -0.0195

Are you sure you are working with half data types and not bfloat16 data types?

[Anthony]

What I'[m doing is reading data from a flight recorder (not the comercial airline type). I found some code my predecessor wrote that looks like this:

Roll = fread(fid,1,'int16=>double','ieee-be'. This angle is in radians. He then multiplies it by 0.00549316 to get the angle in degrees. Any idea how he came up with that formula? It works, I just don't know how.

[Anthony]

I figured it out, it's using a scalar. Thanks for all the help, this thread is done.