How to reduce the precision of calculations in Matlab?

Hello everyone! I'm writing an algorithm, which is then required to perform in the microcontroller. Usually ADC has 14-bit. Matlab uses a data type of double (64-bit, double -precision) and float (32-bit, single-precision) when using real numbers: -12,234 or 742,0263 etc. How to reduce the precision of calculations in Matlab (maybe use half-precision, but i don't know how), or reduce the bit my variables (e.g. 16-bit) to work with the ADC?

[Stephen23]

You are right about the number of bits, but in MATLAB the floating point number classes are called double and single. The name float is not used in MATLAB, and is very misleading because both double and single are floats!

You can see a list of all inbuilt numeric types (including the floating point double and single, and the different integer classes) here:

http://www.mathworks.com/help/matlab/numeric-types.html

[Vitaliy Chistyakov]

Thanks Stephen! But may be possible solve this question differently?

[Star Strider]

You can use int16 or uint16, depending on what you need.

[Guillaume]

@Stephen, the float type in C languages normally means a 32-bit floating point number and matlab keeps this convention, as 'float' is an accepted precision for fread (and is equivalent to 'single').

Answer 1

[Guillaume]

If you are really limited to 14 bits for real numbers, then probably you should be used fixed-point numbers instead of floating point numbers. As far as I know matlab does not support half-precission IEEE-754 numbers and anyway these are 16 bits.

[Walter Roberson]

https://www.mathworks.com/matlabcentral/fileexchange/23173-ieee-754r-half-precision-floating-point-converter

Also there might be some GPU support for half precision.

Answer 2

[Walter Roberson]

When you read a sample from an ADC that uses has 14 bit integer resolution, the result is a 16 bit number that has the value in the most significant 14 bits.

Answer 3

[Krishna Bindumadhavan]

MATLAB now supports half precision floating point numbers via the half precision object, which is available from the fixed point designer toolbox. For more information, please refer to:

https://www.mathworks.com/help/fixedpoint/ref/half.html