co processeor

In 100ns !!! ?

Just the time taken alone to transfer the data in and out via the SPI/I2C interface will take many microseconds.

Graham

But then, he's not really describing a coprocessor, which integrates will with the CPU and is handling functions by itself.

Michael

The chip you mention is much too slow to give you the performance you need. It is only a math co-processor with a clock speed better suited for calculators than fast applications. Besides it lacks ADC and interrupts. To get the speed you want would likely require custom IC's of the type used to digitize and process video and radar signals in real time. Assuming you could procure the chips a working system would cost hundreds of thousands to millions of pounds, dollars, euros (take your pick).

If you can get by with an arc tangent conversion in 6 microseconds with a hardware cost of $99 USD I would refer you to an article in Circuit Cellar issue #202 May 2007. The article of interest is titled "Proton Precession Magnetometer" by James Koehler. The author describes a hardware and software solution for measuring the phase slip angle of magnetically precessed protons in water or kerosene. Using a $99 Keil evaluation board with an NXP LPC2138 ARM 32 bit microprocessor running at 60 MHz, two 10 bit ADC, two serial ports, JTAG (although the USB debugger is an extra cost item) he manages to collect 8000 10 bit analog samples and perform 2000 arc tangent calculations in 250 milliseconds. The arc tangent calculation method is described in the article and employs straight line approximation along with table lookup trigonometric functions to achieve the 6 microsecond speed mentioned. If you or your library don't have access to Circuit Cellar you can order a back issue (print or download) or download for purchase a single article for a few dollars from their website

Hope this helps, I'm curious to know the intended application of your query.

David

Eeyore a écrit :

Hi, the serial bus in only intended to setting parameters of the time which the interrupt will be released according to the angle calculated.

Le Tue, 04 Sep 2007 03:22:22 +0100, Eeyore a écrit:

Uh?

100ns is just 10MHz. These days it's POC. For ex. AD has the AD9254 and AD9640-150 which are resp. a single and a dual 150MHz 14b ADC. They also have three 16b >100MSPS ADCs. They have fifteen 14b ADCs and eight 16b ADCs that fit the > 10MSPS criteria. Just stuff the output stream to a small FPGA, let it manage the other functions too and it's done. Not a big deal really. The only possible question is the arctan latency. But even 16b accuracy isn't much at all and a simple table driven + interpolation algorithm will easily take care of that. Or even a simple dumb 64K size table.

But... but the OP didn't tell what the ADC resolution has to be, and so on... This and a lot of details seem to indicate that he's just throwing some ideas without any understanding of what he's doing and what he really needs.

--
Thanks,
Fred.

ok, I' m going to looking for other method regards to the price.

In fact, I don' t need a good precision, 1° is enough. With a TI TMS320F2812 DSP at 150MHz I can do it in less than 1µs with a

0.2° precision and a limited development method, but it' insufficient.

So with this development method, maybe I can do it with an FPGA and fast ADCs.

It' a an industrial application with a high precision optical angle encoder. We do image acquisition thanks to a photo multiplier with a scan method.

Thanks.

Fred Bartoli a écrit :

Yes, but I will have preferred a little component because I don' t have experience in FPGA and I think it will complicate the design.

Ok, the solution is sufficient I think.

Not needed a big reolution. The final 1° angle resolution is enough. So regards to the electronic an 12b ADC seems to be appropriate.

How exactly do you think the data that needs to be converted gets into the co-processor (and how does the result get outputted) ?

Graham

Eeyore a écrit :

The data will get into the co-processor by ADC(s), and when compare to the angle parameter will generate an output on a GPIO. I have seen that the Micromega UM-FPU V3.1 have this peripherals. I haven' t the ADC coversion time of this component but I suppose it will be longer that I needed.