<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<br>
<div class="moz-cite-prefix">Hi Eric,<br>
<br>
I thought this approach was creative; so, I read part of the
paper. As he notes, it's somewhat difficult to implement;
however, I'm sure we could overcome that. The real problem with
using this algorithm with what we're doing on the Jetson board
appears to be the amount of computation it requires. If you look
at Figure 6, and then project those lines out to, for example,
1,000,000, you can see that there's a huge difference in the
amount of computation required versus the "large block" approach.<br>
<br>
By the way, I also disagree with his position the
direct-time-domain-convolution FIR and his approach are "zero
delay." The delay of a linear-phase direct-convolution filter is
Sample_rate*(N-1)/2 where N is the number of taps. His filter
approach will be the same.<br>
<br>
Window functions are sometimes used in the generation of the
filter coefficients. I think he just didn't really discuss that
process.<br>
<br>
73,<br>
Warren NR0V<br>
<br>
On 9/9/2014 10:01 AM, Erik Anderson wrote:<br>
</div>
<blockquote
cite="mid:CAAeZAZ6EXVdFVLoW=XgN91VUFezwud4OnQYBRDSAA5Q+JC-hBg@mail.gmail.com"
type="cite"><br>
<div class="gmail_extra">
<div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
<div dir="ltr">
<div class="gmail_extra">A random article I ran into
yesterday seems like it would have value here, it is
focused on "zero-latency convolution" and uses a hybrid
of convolution and FFT calculations. It talks about
overlap-keep, but if you ignore all references to h<font
size="1">o</font> it also is creating an
arbitrary-latency FFT computation by disassembling a
"decimate-by-frequency" FFT formula and
precomputing/reusing as many of the calculations as
possible.</div>
<div class="gmail_extra"><br>
</div>
<div class="gmail_extra">I haven't fully parsed everything
in here yet, but I do have two questions:</div>
<div class="gmail_extra"><br>
</div>
<div class="gmail_extra">(*) Would this approach have any
value in say the Jetson board people have been talking
about? It could be used to create a heck of a lot of
overlapped FFT results.</div>
<div class="gmail_extra"><br>
</div>
<div class="gmail_extra">(*) How come there is no mention
of window functions in here? They seem rather important
for getting good outputs, but as soon as we start
tinkering with the inputs they no longer exist as even
something to struggle with.</div>
<div class="gmail_extra"><br>
</div>
<div class="gmail_extra">Erik KM2G</div>
</div>
</blockquote>
</div>
</div>
</blockquote>
</body>
</html>