gfs@abvax.icd.ab.com (Greg F. Shay) (10/31/88)
I posted the following proposal on rec.music.synth and have received a good response. I thought others here at comp.sys.next might be interested from the DSP music standpoint. Please follow up on rec.music.synth, I do not read comp.sys.next. This will be my only posting to comp.sys.next. ------------------------------------------------------------- Shall we start a discussion group on DSP for audio/synthesis applications? For starters, we can use rec.music.synth, but perhaps a rec.music.dsp could be created. I envision swapping of algorithms, assembly code, etc, of which I have some to share already. With the Next system coming out, I predict an explosion of activity in this area. If Next will document their DSP library, this could form some of the foundation of what we can work with. One of the stepping stones for this would be a common hardware platform that we could share code for. Algorithms are processor-type independent, and actual code may be translatable (DSP assembler tends to be simple move, add, multiply instructions). As a hardware engineer with experience in DSP, I offer to design the 'DSPNET' board if we can come to a consensus on what features we want. I would suggest the 56000 for obvious compatibility with Next software, although the Texas Instruments is another option. I will be evaluating them head to head soon. (I presently use the TI 32010, but it may not be the best choice.) For host interface, an IBM plug-in board is one choice, allowing DMA bandwith to/from the hard disk. For Mac's (and Atari), I suggest a SCSI interface. A SCSI interface on the DSP board, and then a simple SCSI interface plug-in for the IBM would not be out of the question, keeping the DSP board constant for all systems. There are 56000 plug-in boards available commercially, but they are in the $1000 price range. I read somewhere that the 56000 will sell for about $19 and I know the Texas Instruments range from $10 (32010) to about $80 (320C25). I have a very usable 16bit D/A and 14bit A/D design for less than $50 total. Local DRAM for audio storage local to the DSP will probably be the main cost. A Meg of DRAM costs $300 nowdays. We can start with 256K to keep costs down. the remaining design should be under $50, with an additional $30 for the printed circuit board. Let's see... all added up, about $225 ballpark. No guarantees until the requirements are figured out and the design finalized. One missing piece needed is the DSP cross-assembler running on each of the hosts. I have written simple assemblers myself, and I have seen 'customizable' assemblers (basically macro assemblers) for not much money. Perhaps a software person (or two) out there will sign up to bring up a simple cross assembler for the DSP code on each target host. --------------------------------------- And my follow up: Well, it seems the right idea at the right time. So far, there seems to be at least a dozen interested. I will bet more will come as word spreads. Hardware is cheaper the more you buy, so spread the word, and when the time comes, we can all save a little more money. Summary so far: 1) Unanimous agreement that the board shall be external to any given computer system. 2) A very good point from Dave Bell, that a midi interface is a logical choice for a compatible interface. DSP programs tend not be large by the simple fact that they need to execute in real time (therefore not too much code). For DSP sample data, I have heard the many complaints about downloading over midi. Therefore, I will persue both midi and scsi, making midi standard and scsi an option (can be deleted to save cost). Since a midi interface requires a UART anyway, it doesn't cost much to have RS232 also, (with selectable baud rates). This should satisfy even computer owners without midi interfaces. (A serious question: How many interested computer owners do not have midi interfaces? Perhaps more than we guess. Of course, you could then use a timeshared mainframe too.) 3) My guess is this project will generate a good deal of net traffic, so I think we need rec.music.dsp (comp.music.dsp?). I am not a net guru, so we need a volunteer who knows how to get new groups administrated to do this. 4) I would like to have a mailing list of US mail addresses eventually of those interested. Heaven forbid, but if we would lose our news feed (it has happened once), or any other problem (for any of you too), a periodic newsletter could keep us all in touch. Once you spend some $$, it is nice to have old reliable (US mail). 5) We need a census of host machines people want to use so we can coordinate cross assembler development. Michael Cripps volunteered one so far. Any others? Obviously work cannot start until the DSP is decided, which should happen soon. I will be out of town this week. Be back 11/4. Go to it. Greg Shay (216)646-5206 day ..pyramid! (216)942-5759 eves ..decvax !abvax!gfs 7508 McKinley St. Mentor Oh 44060