Performance questions

Edgar

I was wondering whether the Propeller could do
DFT or Goertzel algorithm for signals less than 1kHz,
and how fast it could save a word to USB flash memory
or SD memory.

- Thanks.

Rayman

I think the Prop can do DFT on any signal.· I guess you want to ask how fast can this be done?· This would depend on the size of the sample vector and whether you needed floating point, or could use fixed point calculations.

Saving to SD is very fast.· This would not be a limiting factor.

Edgar

Because I had never done digital signal processing before,
I don't know how to translate a requirement in sampling
rate to a requirement on the code size or memory size.
That is the my missing knowledge.

I read a post on the Basic Stamps forum that the Basic Stamps
board could not do DFT. The problem was in both speed and
memory. I was also considering the Arduino processor, but
it seemed that the Arduino was similar to a Basic Stamp.

The acceptable sampling rate was 2000 per second.
I don't know how to translate this to hardware requirement.
It would be ok to do DFT on 1024 points. Although this could be
much fewer using Goertzel. Are these two parameters enough
to estimate the performance with or without floating point?

( I have just realized that I was asking for a benchmark. )

About saving:

Did you imply that saving to USB is a lot slower? and might be
a limiting factor?

- Thanks


Reference to Basic Stamp:
http://forums.parallax.com/forums/default.aspx?f=19&m=42878&g=42895#m42895


Reference to another post on FFT using Propeller
http://forums.parallax.com/forums/default.aspx?f=25&m=119048&g=120533#m120533

Post Edited (Edgar) : 3/26/2008 7:34:48 PM GMT

Rayman

Ok, so you want to·process 1024 points every 0.5 seconds?· That sounds pretty easy.


There is no native way to send large amounts of data via USB. You'd need some kind of 3rd party device.

(although, if you are using the PropPlug, that will let you send data up to 2MBPS or so to the PC over USB).

Post Edited (Rayman) : 3/26/2008 8:26:52 PM GMT

Rayman

You might want to look at implementing it in fixed point though because floating point would definitely slow things down. Also, fixed point could be converted more easily to assembly, if that winds up being required.

I ported some mDCT and iFFT routines into SPIN when I was looking into MP3 decoding, but these were only 32 samples...

Bob Lawrence (VE1RLL)

You may find the FFT info that epmoyer posted interesting:

Use this URL:
http://forums.parallax.com/showthread.php?p=711145

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Aka: CosmicBob

Edgar

From the link you gave me, I found this post

http://forums.parallax.com/forums/default.aspx?f=25&m=204569

on 1-bit delta-sigma ADC. I read the concept from this page

http://members.cox.net/brookhaven-north/Parallax/PropellerSpectrumAnalyzerforAudio.html

and tried different parameters using on Excel. It looks like it works,
but I am missing the theoretical background to justify that it works.

I am going to post the algorithm so that people can tell whether
something is wrong with it or how I estimate the performance.

- Thanks