Full Precision option in Fixed-Point Converter MATLAB

In the HDL Coder and Fixed-Point Converter tool in MATLAB there is an option to set the precision as LSB (least-significant bit), MSB (Most-significant bit) and FullPrecision. What does FullPrecision actually refer too? How precise is full precision? What does it do?

Answer 1

[Walter Roberson]

https://www.mathworks.com/help/fixedpoint/ref/embedded.fimath.html#mw_deb9573c-8a01-4afb-b87a-e89586bebe1d

How the product data type is determined, specified as one of these values:

'FullPrecision' – The full precision of the result is kept.

'KeepLSB' – Keep the least significant bits. Specify the product word length. The fraction length is set to maintain the least significant bits of the product.

'KeepMSB' – Keep the most significant bits. Specify the product word length. The fraction length is set to maintain the most significant bits of the product.

'SpecifyPrecision' – Specify the word and fraction lengths or slope and bias of the product.

So if you were multiplying a 11 bit value by an 11 bit value by a 53 bit value, then the first part (11 x 11) could need as much as 22 bits to get exact output; then 53 bits more would take you up to 75 bits to get the exact output.

Full precision means full precision: number of bits of output is the sum of the number of bits of the inputs. No rounding, no truncation.

It is like doing long multiplication by hand: every digit of the inputs contributes to the length of the results.