ltf@killer.Dallas.TX.US (Lance Franklin) (04/30/89)
I am interested in any information on chips that would facilitate construction of an audio-range spectrum analyzer. The ideal solution would be a single chip that could be accessed n-times a second and provide a computer a word or words in which each bit represents a frequency range, and the "on" condition would represent the signal at that range going above a preset threshold during the time-slice involved....however, any chip or set of chips that could do the same for a single frequency range would probably do. The application, by the way, is a computer-displayed voiceprint...Fourier Analysis is just too slow. Thanks......... -- +-------------------------+ +-----------------------------------------------+ | Lance T Franklin | | I never said that! It must be some kind of a | | ltf@killer.DALLAS.TX.US | | forgery...I gotta change that password again. | +-------------------------+ +-----------------------------------------------+
tomb@hplsla.HP.COM (Tom Bruhns) (05/02/89)
>I am interested in any information on chips that would facilitate construction >of an audio-range spectrum analyzer. The ideal solution would be a single >chip that could be accessed n-times a second and provide a computer a word or >words in which each bit represents a frequency range, and the "on" condition >would represent the signal at that range going above a preset threshold during >the time-slice involved....however, any chip or set of chips that could do >the same for a single frequency range would probably do. > >The application, by the way, is a computer-displayed voiceprint...Fourier >Analysis is just too slow. I submit that it isn't the Fourier Analysis that's too slow -- it's your implementation of it. Try it on a DSP chip. You should be able to do this easily with a Texas Instruments TMS320 or Motorola 56000 or ... . BTW, easily refers to the capacity of the chip to do it, not the ease of programming it or building it! Either chip should be able to do an FFT of a 1024-point input record fast enough to keep up with over 10000 samples per second; another way to do it is to program digital filters for the bands you want, to avoid the FFT's constant-width-per-bin. You can also get cards to slip into a PC to do this sort of thing. Ariel in New Jersey comes to mind; I don't have their address handy. Or you can rent or buy an FFT analyzer from my employer or (heaven forbid)) a competitor. > >Thanks......... > >-- >+-------------------------+ +-----------------------------------------------+ >| Lance T Franklin | | I never said that! It must be some kind of a | >| ltf@killer.DALLAS.TX.US | | forgery...I gotta change that password again. | >+-------------------------+ +-----------------------------------------------+ >---------- Tom Bruhns tomb%hplsla@hplabs.hp.com
todd@ivucsb.UUCP (Todd Day) (05/03/89)
In article <7958@killer.Dallas.TX.US> ltf@killer.Dallas.TX.US (Lance Franklin) writes:
~I am interested in any information on chips that would facilitate construction
~of an audio-range spectrum analyzer.
Well, I'm not sure if this will do, but here goes. I recently saw an ad
for an FFT chip from TRW that does a 1024 point FFT in 0.5ms. But, best
of all, you can get a FREE sample! Just tell them about your application
and they'll send you a free TMC2310.
TRW LSI Products Inc.
P.O. Box 2472
La Jolla, CA 92038
(619)457-1000
--
-Todd Day-
UUCP: {pyramid, ucbvax}!ucsbcsl!nessus!ivucsb!todd
Internet: todd%ivucsb.UUCP@anise.acc.com
ee5391aa@hydra.unm.edu (Duke McMullan n5gax) (05/03/89)
>I am interested in any information on chips that would facilitate construction >of an audio-range spectrum analyzer. The ideal solution would be a single >chip that could be accessed n-times a second and provide a computer a word or >words in which each bit represents a frequency range, and the "on" condition >would represent the signal at that range going above a preset threshold during >the time-slice involved....however, any chip or set of chips that could do >the same for a single frequency range would probably do. > >The application, by the way, is a computer-displayed voiceprint...Fourier >Analysis is just too slow. Discrete Fourier analysis is NOT too slow for most audio work; you're just using the wrong hardware. Even a 33 MHz '386 with numeric coprocessor and fast cache memory is probably too slow, but a DSP chip optimized for FFT calculations isn't. There are PC plug in cards available with the DSP IC, cache memory, support software, etc. on the market TODAY. They ain't cheap. You probably want one of the type that does floating point calculations: the fixed point variety is a nuisance to use, since you have to worry about scaling problems. On this same note, if you haven't yet picked up a copy of the June "Radio- Electronics," go out & get one. Now. Don't even finish reading this article. Don Lancaster's "Hardware Hacker" column discusses three new ICs from Crystal Semiconductor in Texas which are fast, high resolution, CHEAP A/D converters quite sutable for audio work. You probably need to know.... Best o' luck, d "To be or not to be -- that is the square root of 4 B^2." -- Anon. Duke McMullan n5gax nss13429r phon505-255-4642 ee5391aa@hydra.unm.edu
knudsen@ihlpl.ATT.COM (Knudsen) (05/04/89)
In article <7958@killer.Dallas.TX.US>, ltf@killer.Dallas.TX.US (Lance Franklin) writes: > The application, by the way, is a computer-displayed voiceprint...Fourier > Analysis is just too slow. Too slow on what? Have you considered one of the DSP microprocessor chips made by TI, AT&T, and various branches of JapanInc? These are not all that expensive or hard to connect to RAM, ROM, and A/D chips. I forget the published FFT times for such beasties, but you should have no trouble getting 10-50 FFTs per second. Since you need to monitor all frequency bands at once, the FFT really is the right way to go. Analog circuits just don't time-mux worth a damn. You might consider hacking up a cheap stereo equalizer with a built-in spectrum display. My RadShack unit uses 10 discrete op-amp filters to estimate each octave, but the freq resolution isn't as good as you want. -- Mike Knudsen Bell Labs(AT&T) att!ihlpl!knudsen Round and round the while() loop goes; "Whether it stops," Turing says, "no one knows!"
harrison@sunny.DAB.GE.COM (Gregory Harrison) (05/04/89)
TRW has released a hot new FFT chip, capable of performing a 1024 complex FFT in approximately 1/2 a millisecond. Part number:TMC2310. Greg Harrison, The opinions expressed are not necessarily those of my employer.