February 14, 2023 at 11:03 pm
#4914
For capture 24 bit integer is probably more than most physical devices will handle.
For computation,however, 32 bit float is very close to allowing some intrusion of calculation into the lsb’s of the final 24 bit integer product, so while it’s most often fine, sometimes using double precision float is not a bad idea. If you’re using recursive filters, maybe stick to double precision, anyhow.