Nice work David. Maybe some one can let me know what the scope of topics of this forum is. Hopefully, forum users will accept this brief off topic discussion. I apologize in advance.
Agreed with all on development cost. Depends on the market I guess. The STM32 is actually right at the high end of complexities for the MCU market. It is a complex device. But effective. From my aristocratic POV, 60mA at 3V is a real power hog for a battery device, but- as you say David - IT MEETS THE OBJECTIVES , and that is really what it is all about . Something either meets objectives, or it doesn't . Anything else is just fluff. For a box that does not have to do much- simple sequential processing, IE convert ADC, filter microphone samples- encode audio - add framing bits - generate modulator samples, a simple interrupt driven job is fine. Now, throw in a bunch of asynchronous peripherals, peripherals that must be talked to loaded with data, waited up for events to finish and now you have a real problem with the sequential system, and the simple kernel that provides waitable semaphores and timers and some basic co operative multitasking is the answer, rather than a complex state machine. And a basic kernel will only cost less than 100 cycles on a context switch. In 2000 I wrote basic kernel for the AVR that did a context switch in about 60 cycles. So, OS need not have much overhead, and it is ideal when different people are writing for different jobs/peripherals on the one chip. glen english VK1XX "yes - I do this for a living" Altium- ModelSim, Matlab, Vivado, Rowley. On 21/08/2014 8:12 AM, David Rowe wrote: > Hello Glen, > > Yes I agree re the STM32F4's DSP capability. It doesn't have proper > single cycle MACs. However it's fast and cheap and has float so it does > the job nicely. Curiously, no operating system ends up being kinda > helpful on a CPU of this size. > > I measured 60mA at 3V on the STM32F4, that's the lowest power CPU I have > ever played with! I estimated 24 hours operation on a pair of AA's. > Plenty. > > I think if I was doing a multi-channel FreeDV device I'd use .... a PC. > Just throw MIPs at the problem. Dead easy to develop on. I/O would be > the only hassle. > > Cheers, > > David > > On 21/08/14 07:27, glen english wrote: >> indeed Bruce. >> >> Consider I might be aiming at a portable device, or other low power, low >> count silicon platform. (5$) >> and bear in mind, I might want a heap processing power left over for >> modulator/demod, error correction, some audio processing , noise >> reduction etc. >> >> For those that know real DSPs, they'll recognize that the Cortex M4 >> (stm32 4...) is NOT a real DSP. It has some handy instructions, they >> call DSP, sure. But start throwing it alot of filtering tasks and you'll >> run out of cycles. >> >> It is a very good general purpose processor though, excellent in fact. >> I use it for all sorts of things when I don't care about power >> consumption. The CODEC2 is not a simple DSP task, it is much more a >> complex algorithm that doesnt get alot of help from a real DSP. The >> STM32M4 is not a low power processor. >> A real FP processor like the latest low power SHARCs ($10) for similar >> money might do the job for less power, depending on the efficiency of >> the coder- that's the thing to get the advantage of the real DSP, you >> got to know what you are doing. The M4 will make fairly good throughput >> out of junk coding. I use the Rowley Associates toolchain. >> >> >> glen english >> VK1XX >> "yes - I do this for a living" >> Altium- ModelSim, Matlab, Vivado, Rowley. >> >> On 21/08/2014 5:44 AM, Bruce Perens wrote: >>> The reasoning is indeed that floating point is easier to develop and >>> that our development time is more expensive than CPUs.We don't know >>> the table sizes offhand. >>> >>> However, the assumption that both of the codec and modem would fit in >>> really small and relatively low-power floating point chips was >>> optimistic and as of this moment it's right on the edge of working in >>> the STM32F405 that David has built into his SmartMic project. The >>> STM32F405 is an ARM Cortex M4F at 168 MHz, 1 MB FLASH, 126K >>> instruction/data RAM, and 64K data RAM. >>> >>> Over the past weeks David has torn through the code working on >>> optimization, and at this moment the receive speed is "borderline". >>> >>> Thanks >>> >>> Bruce >>> >>> CPOn 08/20/2014 11:42 AM, Steve wrote: >>>> I think the reasoning is, that floating point and memory are so cheap >>>> now, that trying to fit a design into a restricted space would just >>>> lengthen the time to profit. >>>> >>>> Why design to a fixed point $30 DSP when you can buy a $5 CPU with >>>> hardware FP. >>>> >>>> 73,Steve >>>> >>>> >>>> ------------------------------------------------------------------------------ >>>> Slashdot TV. >>>> Video for Nerds. Stuff that matters. >>>> http://tv.slashdot.org/ >>>> >>>> >>>> _______________________________________________ >>>> Freetel-codec2 mailing list >>>> [email protected] >>>> https://lists.sourceforge.net/lists/listinfo/freetel-codec2 >>> >>> >>> ------------------------------------------------------------------------------ >>> Slashdot TV. >>> Video for Nerds. Stuff that matters. >>> http://tv.slashdot.org/ >>> >>> >>> _______________________________________________ >>> Freetel-codec2 mailing list >>> [email protected] >>> https://lists.sourceforge.net/lists/listinfo/freetel-codec2 >> -- >> - >> Glen English >> RF Communications and Electronics Engineer >> >> CORTEX RF >> & >> Pacific Media Technologies Pty Ltd >> >> ABN 40 075 532 008 >> >> PO Box 5231 Lyneham ACT 2602, Australia. >> au mobile : +61 (0)418 975077 >> >> >> >> ------------------------------------------------------------------------------ >> Slashdot TV. >> Video for Nerds. Stuff that matters. >> http://tv.slashdot.org/ >> >> >> >> _______________________________________________ >> Freetel-codec2 mailing list >> [email protected] >> https://lists.sourceforge.net/lists/listinfo/freetel-codec2 >> > ------------------------------------------------------------------------------ > Slashdot TV. > Video for Nerds. Stuff that matters. > http://tv.slashdot.org/ > _______________________________________________ > Freetel-codec2 mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/freetel-codec2 > -- - Glen English RF Communications and Electronics Engineer CORTEX RF & Pacific Media Technologies Pty Ltd ABN 40 075 532 008 PO Box 5231 Lyneham ACT 2602, Australia. au mobile : +61 (0)418 975077 ------------------------------------------------------------------------------ Slashdot TV. Video for Nerds. Stuff that matters. http://tv.slashdot.org/ _______________________________________________ Freetel-codec2 mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/freetel-codec2
