Post by thread:FFT width PICs

At 16:10 28.01.97 +0000, Antonio wrote: > Has anyone tried to implement FFT width PIC? > I'm tryng to do some guitar effects and i don't want to use a DSP (to >expensive). At first glance, the worst problem is the memory. Any ideas ? > I think also computing-power is a problem. For a reasonable sound quality, you should at least calculate with 16bit numbers. Then you need multiplication and addition in 16bit and even the 17C44 would not be that fast (though it has hardware-multiplication 8bit). regards Wolfram +-----------------------------------------------------+ | Wolfram Liebchen | | Forschungsinstitut fŸr Optik, TŸbingen, Deutschland | | spam_OUTliebchenTakeThisOuTffo.fgan.de | +-----------------------------------------------------+

At 16:10 28.01.97 +0000, you wrote: > Has anyone tried to implement FFT width PIC? > I'm tryng to do some guitar effects and i don't want to use a DSP (to >expensive). At first glance, the worst problem is the memory. Any ideas ? > > Thanks, > Antonio Almeida > > I think it is not so expensive. Analog Devices ADSP2105 have price ~5-15$. 2115 also near that price. Bye, Arnold S. Galoustov

Antonio Almeida wrote: > > Has anyone tried to implement FFT with PIC? > I'm tryng to do some guitar effects and i don't want > to use a DSP (to expensive). At first glance, the > worst problem is the memory. Any ideas ? The problem is border line computing power. It is right on the edge. The 17c42a and 17c44 have hardware multiplies and that helps. We added a switch in MPC (It may be undocumented) to generate multiplies in line. That was enough implement some basic FFT filters. As you have stated RAM in some of the other parts may be a problem. Depending on the effect the amount of computing power may be less of a problem. Some filters can be implented with very little code. In general exchanging code size for speed will be a big win in this type of application. Walter Banks

Hi Don't know much about this but... If you have a PC why not let it do the processing and use the PIC as a semi intelligent interface. Something to think about. Tim ------------------------------------------------------------------ If you can read this, it is the end of the message! My web pages are at http://web.ukonline.co.uk/members/tim.kerby/ My PIC site is at web.ukonline.co.uk/members/tim.kerby/pic/ It needs your projects! ------------------------------------------------------------------

You should plan on processing data at about 3 to 4 times the highest frequency you'll be working with. (Example: For 10KHz signal bandwidth use 30KHz to 40KHz data sample rate.) Distortion will be significant with only 8 bits of resolution. Dynamic range (volume range) will also be very limited with only 8 bits of resolution. I highly suggest you move to a fast 16-bit architecture for this audio application. ______________________________ Reply Separator _________________________________ Subject: FFT width PICs Author: pic microcontroller discussion list <.....PICLISTKILLspam@spam@MITVMA.MIT.EDU> at smtp Date: 1/28/97 4:10 PM Has anyone tried to implement FFT width PIC? I'm tryng to do some guitar effects and i don't want to use a DSP (to expensive). At first glance, the worst problem is the memory. Any ideas ? Thanks, Antonio Almeida

In message <Pine.OSF.3.91.970128160758.19838C-100000KILLspamtom.fe.up.pt>,> Antonio Almeida <.....deec352KILLspam.....TOM.FE.UP.PT> writes > Has anyone tried to implement FFT width PIC? > I'm tryng to do some guitar effects and i don't want to use a DSP (to >expensive). At first glance, the worst problem is the memory. Any ideas ? The Microchip Embedded Systems Handbook has an application note about this. Leon -- Leon Heller, G1HSM EraseMEleonspam_OUTTakeThisOuTlfheller.demon.co.uk Tel: +44 (0) 118 947 1424 (home) +44 (0) 1344 385556 (work)

> The problem is border line computing power. It is right > on the edge. The 17c42a and 17c44 have hardware multiplies > and that helps. We added a switch in MPC (It may be > undocumented) to generate multiplies in line. That was > enough implement some basic FFT filters. As you have stated > RAM in some of the other parts may be a problem. Another question is what you need the "FFT" for? For some applications (such as a frequency-range level meter, etc.) where you don't need very high precision, it's possible to fudge the maths so as to eliminate any multiplies. In particular, instead of integrating , T1= | sin(t)f(t) dt ' and , T2= | cos(t)f(t) dt ' and then assuming amp=sqrt(T1*T1 + T2*T2) it may be much faster to integrate three functions: [nb: modstep(x) is defined as 1 whenever sin(x) is positive and 0 elsewhere]: , T1= | modstep(t)f(t) dt ' and , T2= | modstep(t+2pi/3)f(t) dt ' and , T3= | modstep(t-2pi/3)f(t) dt ' and then assume amp=max(abs(T1-T2),abs(T2-T3),abs(T3-T1)). This will not be as accurate as a real foorier transform, but it is much faster. Its primary failing is that it will pick up fifth harmonics at 1/5 of their original strength and seven harmonics at 1/7 of their original strength. All harmonics which are multiples of 2 or 3, however, will be filtered out. Using this approach, it is possible to do continuous simultaneous detection of all seven DTMF tones on a mid-class (e.g. 16Cxx) PIC with a decent amount of RAM.

In message <2ee7e700spam_OUTmailb.harris.com> @spam@PICLISTKILLspamMITVMA.MIT.EDU writes: > You should plan on processing data at about 3 to 4 times the highest > frequency you'll be working with. (Example: For 10KHz signal bandwidth > use 30KHz to 40KHz data sample rate.) Distortion will be significant > with only 8 bits of resolution. Dynamic range (volume range) will > also be very limited with only 8 bits of resolution. I highly suggest > you move to a fast 16-bit architecture for this audio application. Fujitsu has just/is bringing out a DSP chip with 16 bit adc and dac on board. I think it would make a great part for a guitar FX unit and may have a play around with it when I can get hold of one. The part is named the "whisper". Regards, Mike Watson