<div dir="ltr">Hey Phil,<div><br><div class="gmail_quote"><div dir="ltr">On Mon, Jul 30, 2018 at 1:39 PM Phil Karn <<a href="mailto:karn@ka9q.net" target="_blank">karn@ka9q.net</a>> wrote:</div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
You're talking about a purely logarithmic signal, then. That has its own<br>
problems; how would I do an FFT or FIR that requires addition?<br>
Multiplication (even floating point) is now quite fast on modern CPUs,<br>
and there are even fused multiply-adds on newer hardware. So I don't see<br>
any problems working with conventional floating point formats that<br>
consist of a (linear) mantissa plus an exponent that simply says where<br>
to put the decimal (binary, actually) point.<br>
<br>
I would only use half-precision floats as an interchange format for high<br>
speed I/Q streams where network bandwidth is a factor. Also, these<br>
streams will undergo further bandwidth reduction to that of the signal<br>
being demodulated, so the required SNR of the representation is only<br>
moderate -- somewhere between that of the A/D and the baseband signal<br>
itself.<br></blockquote><div><br></div><div>My thought here is that, as you point out, it's an interchange format - it's not a format to do any real work on, and targeting 16-bits per number similarly to half-precision floats. Conversions fit easily in a small LUT to turn into single or double precision floating point after interchange.</div><div><br></div><div>The benefit I saw with doing a logarithmic notation is the addition of different external gain stages turns into an extra input when being converted to the interchange format which will then just be added or subtracted to the number on the remote processor closest to the radio/ADC since most gain stages are already in dB. So processing could still be done in fixed point, and only at the interchange point would the gain modify the signal itself.</div><div><br></div><div>For something standardized, half-precision makes sense. I just wonder if there's something that may work better than IEEE-754 half-precision for a radio interchange format when dealing with numbers at an ADC that are normalized within +/- 1.0.</div><div><br></div><div>Brian</div></div></div></div>