what is the number representation for sample_in and _out ?

[matpalm]

for a custom gateware core configured with, say, W=16 what's the representation that sample_in and _out would be?
is it a particular fixed point value? or some other quantised integer? ( was poking around code and couldn't quite work it out sorry )

also are other values for W available? say 8, or 32? and is this then configured in some more global config? ( since i imagine it effects the config of the ADCs/DACs too? )

[matpalm]

oh, nevermind, i think i was just confusing myself; it's just a signed value, -32768 to 32767 representing -8V to +8V (?)

[schnommus]

For user-defined cores the representation is 14.2 fixed point by default, for 4 counts per mV. This means +8V becomes +32000 and -8V becomes -32000. The hardware actually supports a wider range than this, but this is the default configuration.

The calibrator module turns raw samples into this representation, and is basically the source of truth for this representation. See here for the implementation.

I have not tried different sample widths other than W=16 for this demo repository, but the CODEC itself is using 32-bit slots per sample (ak4619.sv), we are just clocking in the top W bits in the driver and ignoring the rest. So for different sample widths you would need to change W in the top.sv, making sure the new W is propagated down to all submodules. Then, one would need to modify the fixed-point scaling in cal.sv to take advantage of the extra bits.

I created an issue to try this, as it should work but I have not tested it :) #39