On 03/03/2013 03:03, [email protected] wrote: > Message: 4Date: Sun, 03 Mar 2013 10:32:05 +0800 From: Steve Underwood > <[email protected]> Subject: Re: [Freetel-codec2] codec2 float to > fixed To: [email protected] Message-ID: > <[email protected]> Content-Type: text/plain; > charset=ISO-8859-1; format=flowed On 03/03/2013 04:22 AM, David Rowe > wrote: >> Hello Samuel, >> >> Actually there are $3 32-bit microprocessors with floating point, and >> many embedded processors (e.g. those used in android phones) no have >> FPUs. So I think the need for fixed point is diminishing, its only a >> few thousand more transistors on a chip, Moores law etc. >> >> A year ago I worked up an estimate for the labour reqd for a fixed point >> port - it was about $80k. >> >> Cheers, >> >> David >> > The M4 is really making floating point hardware widely available in > modest prices MCUs. Look at how many M4 based MCUs actually have an M4F > core. Its more than a few thousand transistors, though.Its 10's of > thousands. Still, its a tiny part of a typical MCU die. > > A fixed point version of a DSP algorithm will typically take around half > the energy of a floating point implementation on the same generation of > silicon. That is often a good argument for a fixed point version in low > power applications. I'm not sure how applicable that might be for > codec2. It depends what it is used for. In most wireless applications > the energy for the wireless channel should dominate. > > You'll never achieve predictable bit exactness without fixed point, and > some people love bit exact implementations of algorithms. There really > aren't any good arguments for their existence, though. Easy conformance > testing is a pretty weak argument. > > Having a fixed point implementation so you can use a dsPIC, or other > small DSP seems a negative argument for most people. In most cases you > would need to add the cost of the dsPIC to a system that already has a > processor capable of doing the work in floating point. :-) Devices like > the dsPIC are struggling to survive. Their only real hope is > applications like digitally controlled power supplies. ARMs are pushing > them out of most motor control applications, which had looked like a > good market for them until recently. > > Regards, > Steve > > Hi Steve,
I have not yet looked at the ARM processors, but having a quick look at the specs, they do seem light years away from a dsPIC. I am sure there are a lot of people out there who may be interested in interfacing a Codec2 "codec IC" into their own circuits, I know that I would be. Does anyone fancy the challenge of making a small plug in board that could be used in circuits which has all the necessary Codec2 encoding/decoding, with audio in/out and data in/out? Effectively then people can input the data from their modem and then the decompressed audio comes out of the board (and vice-versa for TX). Any ideas? Samuel Hunt ------------------------------------------------------------------------------ Everyone hates slow websites. So do we. Make your web apps faster with AppDynamics Download AppDynamics Lite for free today: http://p.sf.net/sfu/appdyn_d2d_feb _______________________________________________ Freetel-codec2 mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/freetel-codec2
