charlie@oakhill.UUCP (Charlie Thompson) (05/02/89)
I agree with T.B., the DSP56001 from Motorola will do a 1024 point complex FFT in < 4 milliseconds. You can call the Motorola bulletin board "Dr. BuB" at 512 891-DSP1 and download an FFT for the 56001. I have used this FFT to create a real-time spectrum analyzer using a Blackman-Harris window and a log magnitude output display routine. The program will be operational at the ICASSP conference in Glasgow, if you're in the area drop by and take a look. For good resolution at lower frequencies I believe that a log scale DFT might be a better approach. This would give a constant number of frequency bins per decade as opposed to the linear frequency scale output from the FFT. While the DFT is an N-squared job as opposed to the Nlog(N) of the FFT it would take a fairly large FFT (with round off errors abounding) to get the equivalent low end resolution of a brute force N-squared log scale DFT. Has anybody out there done any log scale DFT's? -Charlie Thompson Motorola DSP Operation Austin, TX NBITS
aic@mentor.cc.purdue.edu (George A. Basar) (09/07/89)
Occasionally, I work with bands running sound. One of the nicer pieces of equipment to own is an audio spectrum analyzer, along with a pink noise source. I'd like to build one of these beasties(the SA, not the pink noise source) to interface with a PC (C64 to be precise). I've had a few ideas on how to go about this but I would appreciate any pointers to information on this subject. Ultimately, it should be 31-band, starting at 20Hz and ending at 20KHz, something like a sweepable notch filter connected to an A/D converter. Any help/comments is/are appreciated. Thanks in advance George A. Basar
seningen@oakhill.UUCP (Michael Seningen) (09/07/89)
use a super-hetrodyne principle
source ----X-----[fixed notch filter]---A/D---C64
|
|
------[Voltage Controlled Osc]---|
|
|
[ramp voltager generator]-----------|
the idea is that the fixed notch filter can be much more
accurate than a sweeping filter.
multiply the source out to a higher freq, filter it out,
then shift the source by a small ammount more (ie up the freq of VCO output.
You can use an oscope with the ramp voltage as x and
notch filter output as Y -- this will give you a description of your input waveform.
I learned this in a Signals course which covered fourier and laplace transforms. A good book with some practical circuits might have this circuit and would also give the math background as well.
Mike (dusting off the cobwebs) Seningenjgk@cbnewsc.ATT.COM (joseph.g.klinger) (09/08/89)
In article <3906@mentor.cc.purdue.edu> aic@mentor.cc.purdue.edu (George A. Basar) writes: > Occasionally, I work with bands running sound. One of the nicer >pieces of equipment to own is an audio spectrum analyzer, along with a pink >noise source. > I'd like to build one of these beasties(the SA, not the pink noise source) >to interface with a PC (C64 to be precise). I've had a few ideas on how > Ultimately, it should be 31-band, starting at 20Hz and ending at 20KHz, >something like a sweepable notch filter connected to an A/D converter. I built one of those Gold-Line 10 band spectrum analyzers many years back, it was simply ten parallel notch band filters with a little bit of logic to drive a half dozen LEDs per channel. That's probably the most direct (and conservative) approach to take. I don't think you can find the I/O bandwidth on a C64 to support 16 bit (or 12 bit for that matter) A/D at 44 kHz. One idea, that might save time without too much expense, is to get one of those BSR spectrum analyzer/equalizers from DAK ($100 ?). And use only the notch filters, case, and power supply. Using their filter output (TTL ?) should save a lot of the work. I would think that the interface logic would be strait forward enough. If you went with the A/D method, you would also have to write an FFT program, I found one for the C64 (in basic) on a BBS but it is sloooooow. Joe Klinger att!iexist!jgk Disclaimer - yes
peg@psuecl.bitnet (09/08/89)
In article <3906@mentor.cc.purdue.edu>, aic@mentor.cc.purdue.edu (George A. Basar) writes: > I'd like to build one of these beasties(the SA, not the pink noise source) > to interface with a PC (C64 to be precise). I've had a few ideas on how > to go about this but I would appreciate any pointers to information on this > subject. Hey! The sweepable notch filter sounds like a good idea--the fellow at Motorola should be thanked for that one--but I have another. It is my understanding that a lot of the commercial spectrum analyzers (especially the lower cost ones) are using switched-capacitor filter IC's. These are really nifty IC's that give excellent filter performance and are easy to use. Examples are the MF 4 and MF 10. These IC's are tuned by inputting a digital clock at a multiple of the desired center frequency. Typically, the clock is at 50 or 100 times fc. For audio, this would mean a clock of 2kHz to 200kHz. You would want to watch your signal routing, but you could probably keep the clock out of your audio signals. The big advantage I see of this approach is: by using a programmable divider IC, you could keep your CPU fairly free. You could load a new divisor, say, 60 times a second. That seems like you could be doing other things (especially if you had an Amiga computer!!!! :-). Of course, if you want really fast display update, this might not matter.... But most SA's I've seen have a peak hold type display. Well, I hope that idea might help you. I've been wanting to do this project for a while, but never find time.... Good luck! Paul P.S. The MF4 + 10 I have are by Texas Instr.
ray@ole.UUCP (Ray Berry) (09/09/89)
>The sweepable notch filter sounds like a good idea--the fellow at >Motorola should be thanked for that one--but I have another. This is a nice idea. Perhaps the LO could be log swept. If you could generate a quadrature LO, you could use a pair of mixers and a pair of LPF's to get your 'notch' at baseband (by computing analytic magnitude). If the LPF's were SCF's, perhaps you could even devise a means to make the cutoff freq track the LO such that the LP's had constant Q and approximated something like 1/3 octave bandwidth regardless of the LO frequency. >It is my understanding that a lot of the commercial spectrum analyzers >(especially the lower cost ones) are using switched-capacitor filter >IC's. These are really nifty IC's that give excellent filter >performance and are easy to use. Examples are the MF 4 and MF 10. Reticon makes these in a 1/3 octave flavor, 3 to a package. Each such chip will cover 1 octave total in 3 bands- with a single clock input to the chip. Dividing the clock by two (1 octave) produces the clock for the next chip down the chain. Dual 1/2 octave chips are also available, as are single 1 octave devices. >... The big advantage I see of this approach is: by using a >programmable divider IC, you could keep your CPU fairly free. >You could load a new divisor, say, 60 times a second. That seems These filters have a lot of "state" info in them. I.E., you can't instantly change their output by banging the clock frequency around. You have to wait some (non-trivial) number of clock periods for the output to stabilize after changing the clock freq. -- Ray Berry kb7ht uucp: ...ole!ray CIS: 73407,3152 /* "inquire within" */ Seattle Silicon Corp. 3075 112th Ave NE. Bellevue WA 98004 (206) 828-4422