On Friday, February 18, 2011 11:55:35 AM UTC-5, Phil Jessop wrote:> "bharat pathak" <bha...@n_o_s_p_a_m.arithos.com> wrote in message > news:wsudnUpvtrkzH8PQnZ2dnUVZ_uOdnZ2d@giganews.com... > > With ARM processors becoming increasingly fast and low power, > > I was wondering which audio processing algorithms are still > > done in hardware? > > > > Bharat > > An ARM processor is hardware - so all of them.Indeed. Our friend Bharat might have done better to ask about audio processes implemented without software (or firmware). Jerry -- Engineering is the art of making what you want from things you can get.
which audio processing algorithms are still done in hardware?
Started by ●February 18, 2011
Reply by ●February 18, 20112011-02-18
Reply by ●February 18, 20112011-02-18
On 02/18/2011 06:31 AM, bharat pathak wrote:> With ARM processors becoming increasingly fast and low power, > I was wondering which audio processing algorithms are still > done in hardware?Is your real question "What audio processing functions are done in analog circuitry?", or "What audio processing algorithms are implemented in dedicated hardware, as opposed to general-purpose processors?" But I don't know. Like most things, I expect that some companies and individuals are very slow to adopt the new technology, and others are all too quick. I do expect that a lot of the processing is, and will remain, on lower-end DSP chips -- ARM processors are good at a lot of things, but in my experience they're a lot less efficient at DSP than DSP chips. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com Do you need to implement control loops in software? "Applied Control Theory for Embedded Systems" was written for you. See details at http://www.wescottdesign.com/actfes/actfes.html
Reply by ●February 19, 20112011-02-19
On Fri, 18 Feb 2011 14:15:11 -0800, Tim Wescott <tim@seemywebsite.com> wrote:>On 02/18/2011 06:31 AM, bharat pathak wrote: >> With ARM processors becoming increasingly fast and low power, >> I was wondering which audio processing algorithms are still >> done in hardware? > >Is your real question "What audio processing functions are done in >analog circuitry?", or "What audio processing algorithms are implemented >in dedicated hardware, as opposed to general-purpose processors?" > >But I don't know. Like most things, I expect that some companies and >individuals are very slow to adopt the new technology, and others are >all too quick. I do expect that a lot of the processing is, and will >remain, on lower-end DSP chips -- ARM processors are good at a lot of >things, but in my experience they're a lot less efficient at DSP than >DSP chips.The ARM Cortex-M3's (and M4 ?) are getting pretty close... Close enough that we use them for FFTs at least since we were already using ARM chips and have the software tools. Don't know about the parallel processes though. That may take a while. boB
Reply by ●February 19, 20112011-02-19
On 02/19/2011 12:56 AM, boB wrote:> On Fri, 18 Feb 2011 14:15:11 -0800, Tim Wescott<tim@seemywebsite.com> > wrote: > >> On 02/18/2011 06:31 AM, bharat pathak wrote: >>> With ARM processors becoming increasingly fast and low power, >>> I was wondering which audio processing algorithms are still >>> done in hardware? >> >> Is your real question "What audio processing functions are done in >> analog circuitry?", or "What audio processing algorithms are implemented >> in dedicated hardware, as opposed to general-purpose processors?" >> >> But I don't know. Like most things, I expect that some companies and >> individuals are very slow to adopt the new technology, and others are >> all too quick. I do expect that a lot of the processing is, and will >> remain, on lower-end DSP chips -- ARM processors are good at a lot of >> things, but in my experience they're a lot less efficient at DSP than >> DSP chips. > > > The ARM Cortex-M3's (and M4 ?) are getting pretty close... Close > enough that we use them for FFTs at least since we were already using > ARM chips and have the software tools. > > Don't know about the parallel processes though. That may take a > while.I'll back a 150MHz DSP with a one-cycle-per-tap FIR filter loop over a 50MHz ARM ten-cycle-per-tap FIR filter. Any day. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com Do you need to implement control loops in software? "Applied Control Theory for Embedded Systems" was written for you. See details at http://www.wescottdesign.com/actfes/actfes.html
Reply by ●February 19, 20112011-02-19
On Sat, 19 Feb 2011 09:16:07 -0800, Tim Wescott <tim@seemywebsite.com> wrote:>On 02/19/2011 12:56 AM, boB wrote: >> On Fri, 18 Feb 2011 14:15:11 -0800, Tim Wescott<tim@seemywebsite.com> >> wrote: >> >>> On 02/18/2011 06:31 AM, bharat pathak wrote: >>>> With ARM processors becoming increasingly fast and low power, >>>> I was wondering which audio processing algorithms are still >>>> done in hardware? >>> >>> Is your real question "What audio processing functions are done in >>> analog circuitry?", or "What audio processing algorithms are implemented >>> in dedicated hardware, as opposed to general-purpose processors?" >>> >>> But I don't know. Like most things, I expect that some companies and >>> individuals are very slow to adopt the new technology, and others are >>> all too quick. I do expect that a lot of the processing is, and will >>> remain, on lower-end DSP chips -- ARM processors are good at a lot of >>> things, but in my experience they're a lot less efficient at DSP than >>> DSP chips. >> >> >> The ARM Cortex-M3's (and M4 ?) are getting pretty close... Close >> enough that we use them for FFTs at least since we were already using >> ARM chips and have the software tools. >> >> Don't know about the parallel processes though. That may take a >> while. > >I'll back a 150MHz DSP with a one-cycle-per-tap FIR filter loop over a >50MHz ARM ten-cycle-per-tap FIR filter. Any day.Unfortunately for DSP chips, ARM SOCs are getting a lot more attention and now there are chips out there and announced which run at 1GHz+ with multiple cores and SIMD instructions. With all the infrastructure development and diversity of ARM SoCs are getting, it's going to be quite difficult to justify using any other chip type pretty soon. One can buy ARM+DSP, ARM+baseband, ARM+graphics and ARM+ etc and it's just going to get worse (or better depending on which side of the fence you sit). -- Muzaffer Kal DSPIA INC. ASIC/FPGA Design Services http://www.dspia.com
Reply by ●February 19, 20112011-02-19
On Sat, 19 Feb 2011 09:16:07 -0800, Tim Wescott wrote:> I'll back a 150MHz DSP with a one-cycle-per-tap FIR filter loop over a > 50MHz ARM ten-cycle-per-tap FIR filter. Any day.While the DSP will be much easier to get to that performance level, thanks to lack of issue slots for the memory operations, it is not too difficult to get well less than 1.5 cycles/tap for FIR filters on modern ARM cores, not counting the various SIMD options available now. And there might be as many as 1GHz available, these days. So the comparison is (these days) often not 150MHz DSP vs ARM for some purpose, but 150MHz DSP at extra cost vs some chunk of an existing application-processing ARM, at close to zero marginal cost. Of course it's not as straightforward as that, but the DSP marketplace is clearly in a state of flux. There are FPGA/ASIC options in there too. The target environment that I have to deal with these days is radically different to the one I encountered fifteen or so years ago. Cheers, -- Andrew
Reply by ●February 19, 20112011-02-19
On Fri, 18 Feb 2011 08:31:10 -0600, bharat pathak wrote:> With ARM processors becoming increasingly fast and low power, I was > wondering which audio processing algorithms are still done in hardware? > > BharatI was at a pub gig last night (Swervedriver). Normally I'm right up the front so I can avoid the FoH, but last night I was standing behind the mixer. The mixing desk was a great big Yamaha thing (56 inputs, 8 groups). All analog. (This includes functions like input gain, panning, group assignments, various sends (but not foldback; they used a separate mixer (again all analog) to the side of stage for that), simple "shelving" EQ, metering, etc. BTW, it had lots and lots of knobs (which should make Jerry happy). The effects racks were full of gear: The multiband EQs were analog. The dbx compressor and limiter thingies were analog. The various delay and reverb units were digital. I couldn't see them, but I know that there are racks of amplifiers just to the side of stage. All analog. So, in this professional audio installation, the only digital processing used was in the delay and reverb effects units. The only ARM processors I could locate were in the iPod used for the house music between bands and the phones in the punters' pockets. Regards, Allan
Reply by ●February 19, 20112011-02-19
On 02/19/2011 09:50 AM, Muzaffer Kal wrote:> On Sat, 19 Feb 2011 09:16:07 -0800, Tim Wescott<tim@seemywebsite.com> > wrote: > >> On 02/19/2011 12:56 AM, boB wrote: >>> On Fri, 18 Feb 2011 14:15:11 -0800, Tim Wescott<tim@seemywebsite.com> >>> wrote: >>> >>>> On 02/18/2011 06:31 AM, bharat pathak wrote: >>>>> With ARM processors becoming increasingly fast and low power, >>>>> I was wondering which audio processing algorithms are still >>>>> done in hardware? >>>> >>>> Is your real question "What audio processing functions are done in >>>> analog circuitry?", or "What audio processing algorithms are implemented >>>> in dedicated hardware, as opposed to general-purpose processors?" >>>> >>>> But I don't know. Like most things, I expect that some companies and >>>> individuals are very slow to adopt the new technology, and others are >>>> all too quick. I do expect that a lot of the processing is, and will >>>> remain, on lower-end DSP chips -- ARM processors are good at a lot of >>>> things, but in my experience they're a lot less efficient at DSP than >>>> DSP chips. >>> >>> >>> The ARM Cortex-M3's (and M4 ?) are getting pretty close... Close >>> enough that we use them for FFTs at least since we were already using >>> ARM chips and have the software tools. >>> >>> Don't know about the parallel processes though. That may take a >>> while. >> >> I'll back a 150MHz DSP with a one-cycle-per-tap FIR filter loop over a >> 50MHz ARM ten-cycle-per-tap FIR filter. Any day. > > Unfortunately for DSP chips, ARM SOCs are getting a lot more attention > and now there are chips out there and announced which run at 1GHz+ > with multiple cores and SIMD instructions. With all the infrastructure > development and diversity of ARM SoCs are getting, it's going to be > quite difficult to justify using any other chip type pretty soon. > One can buy ARM+DSP, ARM+baseband, ARM+graphics and ARM+ etc and it's > just going to get worse (or better depending on which side of the > fence you sit).Subtract 10 years, replace "ARM" with "Intel x86" and your statement is otherwise unchanged. Yup. Very believable. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com Do you need to implement control loops in software? "Applied Control Theory for Embedded Systems" was written for you. See details at http://www.wescottdesign.com/actfes/actfes.html
Reply by ●February 19, 20112011-02-19
On Sat, 19 Feb 2011 18:18:33 -0800, Tim Wescott <tim@seemywebsite.com> wrote:>>> I'll back a 150MHz DSP with a one-cycle-per-tap FIR filter loop over a >>> 50MHz ARM ten-cycle-per-tap FIR filter. Any day. >> >> Unfortunately for DSP chips, ARM SOCs are getting a lot more attention >> and now there are chips out there and announced which run at 1GHz+ >> with multiple cores and SIMD instructions. With all the infrastructure >> development and diversity of ARM SoCs are getting, it's going to be >> quite difficult to justify using any other chip type pretty soon. >> One can buy ARM+DSP, ARM+baseband, ARM+graphics and ARM+ etc and it's >> just going to get worse (or better depending on which side of the >> fence you sit). > >Subtract 10 years, replace "ARM" with "Intel x86" and your statement is >otherwise unchanged. > >Yup. Very believable.You seem to be confusing Intel's marketing foil with reality. x86 is not a licensable core which can be integrated into SoC. It can not be produced on any fab but Intel's (AMD might be going to different foundries but they're legally prohibited from licensinsg their cores to other because of the agreements they have with Intel). Here are just two examples of thing which exist today: http://focus.ti.com/paramsearch/docs/parametricsearch.tsp?family=dsp§ionId=2&tabId=2688&familyId=1985 http://www.qualcomm.com/documents/snapdragon-msm8x60-apq8060-product-brief -- Muzaffer Kal DSPIA INC. ASIC/FPGA Design Services http://www.dspia.com
Reply by ●February 21, 20112011-02-21
On 20/02/2011 03:18, Tim Wescott wrote:> On 02/19/2011 09:50 AM, Muzaffer Kal wrote: >> On Sat, 19 Feb 2011 09:16:07 -0800, Tim Wescott<tim@seemywebsite.com> >> wrote: >> >>> On 02/19/2011 12:56 AM, boB wrote: >>>> On Fri, 18 Feb 2011 14:15:11 -0800, Tim Wescott<tim@seemywebsite.com> >>>> wrote: >>>> >>>>> On 02/18/2011 06:31 AM, bharat pathak wrote: >>>>>> With ARM processors becoming increasingly fast and low power, >>>>>> I was wondering which audio processing algorithms are still >>>>>> done in hardware? >>>>> >>>>> Is your real question "What audio processing functions are done in >>>>> analog circuitry?", or "What audio processing algorithms are >>>>> implemented >>>>> in dedicated hardware, as opposed to general-purpose processors?" >>>>> >>>>> But I don't know. Like most things, I expect that some companies and >>>>> individuals are very slow to adopt the new technology, and others are >>>>> all too quick. I do expect that a lot of the processing is, and will >>>>> remain, on lower-end DSP chips -- ARM processors are good at a lot of >>>>> things, but in my experience they're a lot less efficient at DSP than >>>>> DSP chips. >>>> >>>> >>>> The ARM Cortex-M3's (and M4 ?) are getting pretty close... Close >>>> enough that we use them for FFTs at least since we were already using >>>> ARM chips and have the software tools. >>>> >>>> Don't know about the parallel processes though. That may take a >>>> while. >>> >>> I'll back a 150MHz DSP with a one-cycle-per-tap FIR filter loop over a >>> 50MHz ARM ten-cycle-per-tap FIR filter. Any day. >> >> Unfortunately for DSP chips, ARM SOCs are getting a lot more attention >> and now there are chips out there and announced which run at 1GHz+ >> with multiple cores and SIMD instructions. With all the infrastructure >> development and diversity of ARM SoCs are getting, it's going to be >> quite difficult to justify using any other chip type pretty soon. >> One can buy ARM+DSP, ARM+baseband, ARM+graphics and ARM+ etc and it's >> just going to get worse (or better depending on which side of the >> fence you sit). > > Subtract 10 years, replace "ARM" with "Intel x86" and your statement is > otherwise unchanged. > > Yup. Very believable. >There are a number of /very/ big differences between the x86 ten years ago, and ARM today. The most obvious ones are the size, power and cost of the devices themselves and the other required components. Ten years ago, making an x86-based board was a lot harder and more expensive than making a DSP board with the same FIR filter rate, and the power consumption would be orders of magnitude higher. But with ARM now, that's not the case. It is still the case that DSP's - like most specialist devices - will give you better value for money and for power when doing work that fits with their strengths. But when you are doing many other things as well, as you often are, then why choose a DSP? It is getting easier and easier to do fast DSP on general purpose ARM's (and other general purpose cores like MIPS or PPC), while it is still just as hard to do general purpose work with most dedicated DSP devices (with their 32-bit "char", fragmented memory, awkward and expensive tools, etc.). For some tasks, the more DSP speed you have, the better. And high-end DSPs certainly remain the speed kings. But there are a lot of tasks that simply don't need that much speed. Ten years ago, if you wanted 5 million FIR taps per second, you needed an expensive DSP. Now you can get it on a 50 MHz ten-cycles-per-tap ARM. And if 5 million taps per second is all I need, I'll take the ARM over the 150 MHZ one-cycle-per-tap DSP any day.
