DIY DSP Engine

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I'd like to educate myself on DSP by making a DIY digital equalizer and loudspeaker crossover. So far, in my research into getting started, I've run across the TI TMDSDSK6713 development board (see http://focus.ti.com/docs/toolsw/folders/print/tmdsdsk6713.html) as a good and affordable DSP hardware platform I've also found a board that plugs into the TI board (see http://www.mds.com/Products/product.asp?prod=ADK) that provides for audio input and output.

Is anyone else doing this sort of thing? Does anyone have other suggestions for hardware? Lastly, what would it take to build something like the MDS board? I have an BSEE degree, but I've been a software guy for the last 20 years. What does it take to do DIY board design and assembly these days?

The MDS product costs around $1000, so building something would have to be cheaper enough to make it worth the while.

Thanks.

Paul Ebert
 
There was a guy doing just that sort of thing at the Dutch DIY audio day, sorry i can't remeber his name:confused:. He had got analogue and digital active X-overs in the same box. He could switch between the two from his pc/listening seat and alter the digi parameters from there too. Driving some pretty high class drivers from it too, although, the digi side of things wasn't fully operational yet. Very impressive!
Mark
 
I'm going to use Alesis (Wavefront) 1 KS DSP with Alesis DAC's and Crystal receiver. Also a microcontroller is needed to upload the firmware (size is just 1024 x 3 bytes) - maybe you could use the parallel or even serial port (control lines) of PC for the job. I wish someone could desing the PCB (for DSP, S/PDIF receiver and DAC's, with a header for DSP serial port to wire to external MCU card / parallel port etc.) and arrange a group order for them. These are really inexpensive chips and easy to use. The 1 KS (1 KM, the parallel interface version) DSP has 1024 instructions / sample to process, so it's possible to implement a decent FIR xover with them. For EQ like BSC (baffle step compensation) an analog circuit suits much better (to conserve the dynamic ration of DAC's).

Alesis Semiconductors
 
Digital EQ and DAC dynamic range

Let me see if I understand your point. If digital EQ is applied, say to provide some bass boost or bsc, then, for those frequencies that are boosted at least, the DAC is 'biased' toward higher bits. This then reduces the dynamic range for those frequencies. Is that what you are saying?

Seems like a good point that hadn't occured to me before. I'll have to think about that some. Probably wouldn't stop me from using digital EQ, though.

Paul Ebert
 
Yes, but of course you could apply the BSC just to the lowpass channels so you loose some (6 dB with full BSC, that's one bit of resolution) dynamic range from the f-6 point (like @450 Hz with 25 cm wide baffle) up till crossover frequency. Sadly that's the area where the human ear is most sensitive. Anyway, as a hybrid filter you could implement the analog BSC and a digital allpass with opposite phase response to keep the phase as linear as possible (if you think the phase linearity is important). The AL1201 DAC's have 107 dB SNR so 101 dB is still pretty good, still above 16-bit performance (shouldn't propably matter).
 
mhelin said:
I'm going to use Alesis (Wavefront) 1 KS DSP with Alesis DAC's and Crystal receiver. Also a microcontroller is needed to upload the firmware (size is just 1024 x 3 bytes) - maybe you could use the parallel or even serial port (control lines) of PC for the job. I wish someone could desing the PCB (for DSP, S/PDIF receiver and DAC's, with a header for DSP serial port to wire to external MCU card / parallel port etc.) and arrange a group order for them. These are really inexpensive chips and easy to use. The 1 KS (1 KM, the parallel interface version) DSP has 1024 instructions / sample to process, so it's possible to implement a decent FIR xover with them. For EQ like BSC (baffle step compensation) an analog circuit suits much better (to conserve the dynamic ration of DAC's).

Alesis Semiconductors

I'd recommend the TI c671x series. The chip is somewhat more expensive at $22 ($5 for Alesis), but the DSP starter kits are not far off (TI=$395 vs Alesis=$250). TI wins on CPU power - they claim from 600 to 1350 MFLOPS vs 49 MIPS for the Alesis, so for a 60 % increase in prototype system cost you get at least 10x performance. The TI chip, by the way, does 32 bit floating point as well as 24 bit integer. 32 bit integer is also available, but I haven't checked to see whether that involves an efficiency hit.

http://focus.ti.com/docs/toolsw/folders/print/tmds320006711.html
http://focus.ti.com/docs/toolsw/folders/print/tmdsdsk6713.html


Francois.
 
I forgot: I *like* mhelin's idea of a daughtercard design & group buy. Momentum Data Systems wants a grand for their 4-in 8-out card; that's a bit pricy for something hanging off a $400 dev board. I'd be all over putting together a DIY board but my analog days are way behind me. Hmmm. Now how do we get some of that fancy mixed-signal design talent on the other boards to show up here?


Cheers,
Francois.
 
tiroth said:
DIYing the BGA package is a bit...impossible.

What we could also really use is some software. There are a lot of people that could assemble the working hardware, but a bare handful who have the skill to write the software.

No no no no, not the DSP in BGA. We can buy the DSKs for $400, I'm happy with that. I was talking about DIYing DACs hanging off the daughterboard connector. I have enough mixed-signal chops to make vague recommendations but the heavy lifting, well, I'd be really pleased to admire someone experienced.

And I've been doing DSP code for a while now, so:

WILL PROGRAM FILTERS FOR FOOD, er, DACS.


Francois.
 
I'd be quite satisfied with the TI DSK as well. The price seems fine for the hardware and development software provided. What I'd like to see on the daughterboard would be two channels of high quality ADC, six or eight high quality DACs and some sort of high powered (24/96 would be nice) digital I/O as well.

BTW, what is JTAG (can you tell I'm a DSP newbie)?

I have no DSP software experience, but loads of rather intense embedded software experience (medical devices). I'm looking forward to writing the software, if I can get the hardware together.

I'm fairly new to this board. Where do those mixed signal gurus hang out?

Paul
 
:wave2:

I ve got TMDSDSK6713 board and TI DSK , and few D/A chips.
In near future i plan to do Dig XO with it. I never turned up the board (didnt have time). But i have DSP knowledge and programming experience C,asm,Matlab. It would be a good idea to get together about TI platform Dig XO.
Have to check I/O possibilites of the board and i have problem with extension connectors on it ,i`ll have to buy female ones for extension board.

http://cizzy.freeservers.com/software.html
http://cizzy.freeservers.com/Hi-Fi.html
 
netgeek said:
I'm very interested in the Analog Devices AD1954 for use as a crossover (with some minimal EQ thrown in as well). Have you looked at or considered that device?

It looks as if it only does IIR filters, so you're talking the same sort of filters you can get from analog active crossovers. Granted, the chip also has cute stuff like lookahead limiters and parametric EQ, but it's not my cuppa tea as I'd prefer trying FIRs (finite impulse response) filters for digital crossovers, not to mention playing with inverse filtering to flatten out driver response.


Francois.
 
ditter said:
IMHO DAE-7
will be better for DIY audio :D


tiroth said:
How much?


I emailed MDS and was told that the projected price for the DAE-7 would be about the same cost as the DAE-5 - around $2500. More than I'd like to spend. They also said that they would not be able to provide support since I was buying it for educational purposes (they are looking for OEM purchases, I guess).

Paul
 
price for the DAE-7 would be about the same cost as the DAE-5 - around $2500
:eek:
well, board DAE-7 + software =2500 ? will they supply Gerber/shematic/ to DAE-7 board development kit ?
to get TMS320C6713-PYP208 also from TI samples programm is not difficult . To create shematic & PCB is not difficult too, but we can`t receive Dolby algorithmes. The difficulties are in acknowlegement of these algorithmes.
 
Paul Ebert said:


I emailed MDS and was told that the projected price for the DAE-7 would be about the same cost as the DAE-5 - around $2500. More than I'd like to spend. They also said that they would not be able to provide support since I was buying it for educational purposes (they are looking for OEM purchases, I guess).

Paul

They're nuts. If they want $2500 for a board with $150 of parts, it'd make more sense for an OEM to design it and build it at any more than 100 lots, even with the annoyance of sending it out to a service bureau to solder the BGA. It shows they also don't understand the economics of OEM building, since OEMs need to price retail at 4xparts if they don't want to go broke. That means the $2500 MDS board turns into $10000 of retail price for whatever it's designed into.

The MDS website indicates the DAE-7 is aimed at, among other things, digital loudspeakers. You can buy an awful lot of driver goodness for $2500.


Francois.
 
This thread pretty much highlights why I gave up on DSP hardware for my experiments/projects and just ended up using a PC - the market is just not really conducive to one-offs. By the time you get a good DSP board and dev tools etc it's pretty expensive, and AFAIK *none* of them have decent audio sections, so you still have to build your own outputs. If your main interest is in the programming, it's just too much hassle.

Unfortunately, the PC approach doesn't really lend itself too well to production/semi-production use, so a modular DSP hardware project is still interesting. I'm not aware of any ideal candidates, but I haven't searched thoroughly - the Momentum stuff looks great, but they're obviously targetted at larger industrial customers. The Sharc boards from www.danvillesignal.com look pretty good and seem to run about $400, but info on their site is pretty sketchy - in particular the expansion connectors dont' seem to be documented, and it looks like dev software would still have to be purchased on top of the board cost.

One idea that I still think is worth persuing for a 'budget' solution would be to hook up the I2S lines from the dsp board into one of the Panasonic digital receivers, which use I2S internally in their Equibit power stages. It would obviously require a mod to the receiver, but should result in a very impressive package for ~$700 .
 
DSP_Geek said:
They're nuts. If they want $2500 for a board with $150 of parts, it'd make more sense for an OEM to design it and build it at any more than 100 lots, even with the annoyance of sending it out to a service bureau to solder the BGA.

The purpose of the dev boards is to allow developers to work on the code from a known good platform, without having to design and build hardware. While there are some reference platforms, most of these devices are never intended to be in an end-product. You write your code on the $2500 module and then build 10,000 boards with $25 chips on them.

If you are paying your engineers $50+/hour (all costs considered) the dev board pays for itself if you save a week's time. In all likelihood, that week is worth far more than just their time because it is a week shaved off of your production cycle.

The reference platforms are generally cheaper, but usually they are only interested in selling OEM quantities.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.