[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[HLL] Re: Celp

Status: RO
Content-Length: 2262
Lines: 47

>Hy !
>> > I suggest you look at some of the freely available CELP codecs. They are
>> > very forgiving and work at  low bit rates. [No doubt that both  properties
>> > relate to the fact that these codecs have been developed by the US
>> > Department  of Defense :-] The major drawback of the CELP codecs is that
>> > they require more  CPU power. But the HLB should have enough.
>> I don't think so, or you must know something that I don't. The CELP codecs
>> I found couldn't do realtime on a 486DX4-100. More like 10% of realtime.
>> You need pretty cool DSP to keep up with that. Neat codecs though. Sounded
>> amazing still at 7200bps.
>Our HLB uses a 30MHz/32-bit risc-cpu. This gives us about 25 Mips, i expect.
>Standard-gsm needs about 10Mips (full-duplex), crypto needs about 5 Mips.
>Let's have 5 Mips for everything remaining - then we use about 20 Mips.
>So there is not much room left to complicate things...
>It's not that we could not build an even more powerful HLB-hardware ;-)
Yeah. Let's thake the new Hitachi CPU with 300Mips/1.3Gflops :-)

>But this would make it much more complex/expensive and should be avoided !!!
>We would probably have to use 'real DSPs' again.
>Does CELP usually need fixed or floating-point arithmetic ?
Usually floating-point!!! There is an implementation for TI-Fixedpoint DSPs 
but not for free. The normal C-Source (Fortran too) that could be found on 
the Net is Floating-point and not useable for us.

We could use ITU G.723.1 (5.3 or 6.3 kbit/s) if we order the Source 
(fixed-point C Reference Implementaion), but I'm not shure if we have enough

So lets think more practical. PGPfone uses GSM-Halfrate on low Speedconnects
we should do either. People that have only a 14400 Modem could live with that.

Anyway. Has anybody the Source of PGPfone? It seems they have any of the
Codec/Linespeed-Problems solved. I'd like to have a closer Look on that.

CU Hacko

Hartmut Schroeder             Email:hacko@mail.hh.provi.de
http://www.mms.de/~hacko      32U0564185 5937869 (WGS84)

Theory: "Computers are based on Smoke." Argument: "If the Smoke escapes from
the Chips,(attempt to false inserting your CPU) they don't function any more."