propplot ?

Sawmiller

another thing comes to mind...... the stampplot with the propellor ? hmmmm
will have to play with that on my day off [noparse]:)[/noparse]
dan

Ken Gracey

Dan,

Considering the UART libraries with BS2 compatibility are written this should be quite feasible. Even within my reach.

And after you get that working, toss the UART and go for the real thing using the video drivers. Sounds like a great way to spend the day off.

Ken Gracey
Parallax, Inc.

cgracey

One thing that we wanted to do with the SX-Key software, but never did, was to make graphical terminals, like triggered scopes, X-Y scopes, RGB buffers, FFTs, Spectographs, etc. With the Propeller, we intend to add this to the Propeller.exe program. Once you download a program over the USB serial port, you can turn around and transceive data at 3M baud over the FTDI chip. Jeff is very busy with documentation right now, but once he's got more time, this is the next big thing on the list.


▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔


Chip Gracey
Parallax, Inc.

Jim C

Chip:

Speaking of FFT's and plotting, how long do you think it would take a propeller to do a 1024 point FFT? Seems to me an object like that would open up some interesting application possibilities.

Jim C

Jim C

Chip:

Speaking of FFT's and plotting, how long do you think it would take a propeller to do a 1024 point FFT? Seems to me an object like that would open up some interesting application possibilities.

Jim C

cgracey

Well, I think it would take N*log₂N computations, so 1024*10, or 10,240 complex multiply-accumulates. At present, it takes ~32 instructions or 1.6us at 80MHz to do a 16x16 multiply. Two (complex) would take 3.2us. 10,240 of those would take ~32,768us, or ~32ms. So we could update at ~30Hz. That's just one COG working, though.

The next generation Propeller chip will be able to do this FFT at over 1KHz (using a single COG).

Jim C said...
Chip:

Speaking of FFT's and plotting, how long do you think it would take a propeller to do a 1024 point FFT? Seems to me an object like that would open up some interesting application possibilities.

Jim C

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔


Chip Gracey
Parallax, Inc.

Jamesx

Chip:

Hmm... 30Hz, going to 1K Hz. Impressive!

I'll keep my eye on this forum for such an FFT object. I'd try it myself, but I'm afraid that's a little over my head.

Thanks.

Jim C

Paul Baker

A key caviat to what Chip has said is that these estimates are for a single cog FFT, FFT algorithms are based on butterfly networks to speed things up (it's where the first F (standing for "Fast") comes from), portions of the network can be computed seperately using the concept of Decimation in Time (DIT, or breaking the signal into smaller windows of time), meaning you can parallelize the computation of the FFT (use more than one cog to perform the calculation).

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Life is one giant teacup ride.

Jim C

Another question about FFT computation speeds: What if the data were floating point? Would that change the time it takes?

Thanks,

Jim C.

Paul Baker

Yes it would, its best to use fixed point notation.

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Life is one giant teacup ride.

johnnyairplane

Great discussion on the ability to do FFT's in 'real - time' on the prop!
Reading it I remember using Motorola DSP chips to do FFT's and then,
I remember doing a fourier transform by hand back in the 80's in an
advanced engineering mathematics course, it took 3 weeks, required every
chalkboard in the room and the cooperation of the janitors to not erase
the previous days work!

My point is, sometimes, 16ms may seem like forever.....
Most of the time a FFT is 'overdone' - too much information for the application...
Keep it simple, keep it application oriented...

3 weeks to transform a pulse function into a fourier equation was sure harder....
We all need to be thankful we have these little chips to do all the dirty work for us...

I may just brush off my HP-41CV and see if it still works!

John....