Hello all,
I am investigating whether it is worth buying, distributing and servicing a certain DSP quite suitable for DIY audio purposes.
Below are the specs:
multiplier 24 x 12 bit
dual harvard
4 x i2s in
4 x i2s out
1 x DA
Programmable PLL
2 x SPDIF In (muxed)
70 MIPS
1.5k word Program memory
2 k word data memory (24 bit)
1.5 K word coefficient memory (12 bit)
With 2 multiplications dual precission calculations are possible.
Available tool is the assembler only.
My idea is to :
- supply you with IC's
- make kits with IC, board and tool
- make a full digital crossover board
What do you think ?
grtz
Simon
I am investigating whether it is worth buying, distributing and servicing a certain DSP quite suitable for DIY audio purposes.
Below are the specs:
multiplier 24 x 12 bit
dual harvard
4 x i2s in
4 x i2s out
1 x DA
Programmable PLL
2 x SPDIF In (muxed)
70 MIPS
1.5k word Program memory
2 k word data memory (24 bit)
1.5 K word coefficient memory (12 bit)
With 2 multiplications dual precission calculations are possible.
Available tool is the assembler only.
My idea is to :
- supply you with IC's
- make kits with IC, board and tool
- make a full digital crossover board
What do you think ?
grtz
Simon
One opinion.
It seems to fall between two stools. For those skilled in the art as they say, it's too little. One could do more on a low cost FPGA or DSP EVB. For those not skilled in the art, assembly language programming may well be too much to ask and one of those join-up-the-boxes type tools would put it beyond the reach of most, assuming one is even available for said chip.
It seems to fall between two stools. For those skilled in the art as they say, it's too little. One could do more on a low cost FPGA or DSP EVB. For those not skilled in the art, assembly language programming may well be too much to ask and one of those join-up-the-boxes type tools would put it beyond the reach of most, assuming one is even available for said chip.
Depending on the DSP itself i would say that assembler is maybe not enough, i could probably cope but it takes time to learn a new processor at assembler level.
As for processing power does the multiply take more than one cycle and how about memory fetches?
Are there any MAC operations available in the processor or does it take the extra add cycle.
All in all it would be a cost and ease of integration issue if i was to buy one.
A slight suggestion if you are to make boards, reclock the I2S outputs in an intelligent manner.
As for processing power does the multiply take more than one cycle and how about memory fetches?
Are there any MAC operations available in the processor or does it take the extra add cycle.
All in all it would be a cost and ease of integration issue if i was to buy one.
A slight suggestion if you are to make boards, reclock the I2S outputs in an intelligent manner.
I can assure you that creating Filters in an FPGA is far more complex than using an DSP !
Ever wondered how to create a Multiplier / adder and etc ????
How to create a circular buffer for coeffiecents and data ?
Not talking about I2s receivers and transmitters !
Or are you willing to buy such mega functions ? I am not !
Idea is to create an open library wherein knowledgable will contribute to the lesser knowledgable.
It is 56000 a like
All 1 instruction cycle.
You can do a multiplication add and 2 moves (coef and data) in 1 cycle. But not in all cases.
OK. Good suggestion
grtz
Simon
apologies if i offended you !
I investigated different options my self, since i became intrigued by digital (dsp) x-overs.
FPGA's was one of them.
Problem with FPGA's is that it becomes fastly more complex.
You need to keep in mind the package and amount of gates, knowledge of verilog/vhdl and the more indepth knowledge of multipliers/adders versus speed and logic size.
But if i had more time............... i would certainly try it.
grtz
Simon
I've been using these things since the days of the XC3000 and I have some measure of what they can do. HDL's are a side issue. One has to first understand the logic being described and if one does then it becomes apparent some things are more suited to FPGA's than DSP chips but each to his own.
Well, having some information on what DSP you're talking about (a part number or datasheet) would certainly help. I'm not going to buy a random DSP until I know exactly what it has for packaging, required voltages and power requirements, I/O & peripherals, RAM architecture and addressing capabilities, some idea of what the instruction set looks like, yadayadayada...
Until I get this info, I'll stick with my SHARC and Blackfin evaluation boards.
And I will agree, doing FIR filters in a FPGA is harder - hand-coding a FIR in Verilog is a pain in the *** (been there, done that) and using an easy FIR filter block invariably involves buying a $$$ set of design tools. But the only place when you'll really want to do that is when you're using excruciatingly high data rates where a DSP just doesn't have the processing capabilities.
Until I get this info, I'll stick with my SHARC and Blackfin evaluation boards.
And I will agree, doing FIR filters in a FPGA is harder - hand-coding a FIR in Verilog is a pain in the *** (been there, done that) and using an easy FIR filter block invariably involves buying a $$$ set of design tools. But the only place when you'll really want to do that is when you're using excruciatingly high data rates where a DSP just doesn't have the processing capabilities.
IMHO, the mathematical precision of the DSP in question is quite bad. Multiplier is 24x12 and coefficients are 12 bits?
If a person wants simply to make a digital crossover, a much much better and easier solution is to use a product like TI's TAS3103 DSP. It is not a general purpose DSP, so you can only do what it does, but what it does is 16 IIR filters per each of 3 channels, plus a whack of mixing/3D effects including delay (good for lip sync), dynamic range exp/comp, and a lot more. It has almost infinitely better precision than the aforementioned DSP, with 48x28 multiplies and 28 bit coefficients. All that you need to do is configure it with EQ coefficients, etc., which is a lot easier than writing code from scratch. Its inputs and outputs support 16 to 32 bits in just about every format you've ever heard of, and it interfaces directly with DACs and ADCs or DIRs.
If a person wants simply to make a digital crossover, a much much better and easier solution is to use a product like TI's TAS3103 DSP. It is not a general purpose DSP, so you can only do what it does, but what it does is 16 IIR filters per each of 3 channels, plus a whack of mixing/3D effects including delay (good for lip sync), dynamic range exp/comp, and a lot more. It has almost infinitely better precision than the aforementioned DSP, with 48x28 multiplies and 28 bit coefficients. All that you need to do is configure it with EQ coefficients, etc., which is a lot easier than writing code from scratch. Its inputs and outputs support 16 to 32 bits in just about every format you've ever heard of, and it interfaces directly with DACs and ADCs or DIRs.
Unfortunatly i can not go in further detail since i am in negotiations with the company
I am investigating if there is a need ! and interest !
What about double precision multiplications ?
That is 24 x 24 !
Is more tah sufficient.
Package is QFP44.
VCC 3v3 , IO's 5 V tolerant.
4 x I2S in
4 x I2S out
1 x DA Out
2 x SPDIF In (muxed internally)
Dual harvard architecture (program, data and coef memories)
grtz
Simon
I am investigating if there is a need ! and interest !
IMHO, the mathematical precision of the DSP in question is quite bad. Multiplier is 24x12 and coefficients are 12 bits?
What about double precision multiplications ?
That is 24 x 24 !
Is more tah sufficient.
Package is QFP44.
VCC 3v3 , IO's 5 V tolerant.
4 x I2S in
4 x I2S out
1 x DA Out
2 x SPDIF In (muxed internally)
Dual harvard architecture (program, data and coef memories)
grtz
Simon
I must say, the TQFP44 package is kinda sexy, and being able to process multiple I2S streams is nice.
- Are the four I2S ports separately clocked, or do all four share a common clock and frame sync? (eg, can I make this chip do interpolation or ASRC?)
- How many 24x24 MAC per clock, and what's the clock speed?
- does it allow full zero overhead loops or just a REP MAC instruction?
- You said internal program memory and internal coefficient memory... is this onboard flash, or is the chip somehow booted from external parallel or serial ROM?
I'm being a bit inquisitive here, but it's because DSP programming is most of my day job... there's hundreds of questions I could ask, but I'd still be much happier if you just posted a datasheet for this DSP.
- Are the four I2S ports separately clocked, or do all four share a common clock and frame sync? (eg, can I make this chip do interpolation or ASRC?)
- How many 24x24 MAC per clock, and what's the clock speed?
- does it allow full zero overhead loops or just a REP MAC instruction?
- You said internal program memory and internal coefficient memory... is this onboard flash, or is the chip somehow booted from external parallel or serial ROM?
I'm being a bit inquisitive here, but it's because DSP programming is most of my day job... there's hundreds of questions I could ask, but I'd still be much happier if you just posted a datasheet for this DSP.
That many I2S with that little processing power just does not seem to hold much value for DIY audio. Perhaps for low end purposes it holds some value, but for what most of us want to do on this site, I don't think this is where we want to go. Blackfin, Shark, TI C6x, etc. are all relatively low cost and support 32 bit easily which makes for easier arithmatic.
Just some thoughts...
Alvaius
Just some thoughts...
Alvaius
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.