[asterisk-users] [Zaptel] Why no driver for PCI voice modems?

Steve Underwood steveu at coppice.org
Wed Apr 1 12:47:00 CDT 2009


Tim Nelson wrote:
> ----- "Wilton Helm" <whelm at compuserve.com> wrote:
> >
> >If half-duplex audio is good enough for you, sure.
>  
> You've lost me there.  I am not aware of a modem that is for sale 
> today that is half duplex. (OK some support a couple of minor half 
> duplex modes).  All state of the art modem protocols send and receive 
> simultaneously using the full 300 - 3000 Hz bandwidth in both 
> directions with adaptive equalization and echo cancellation to make it 
> work, which is pretty much what a voice circuit need.  There are two 
> differences:  1) The response and quality of a current modem must be 
> considerably higher than what is needed for voice use or it would 
> never achieve the throughput expected of it, and 2) the adaptive 
> equalization algorithm is designed around modem specific techniques.  
> The latter is (especially for a softmodem) a software issue, not a 
> hardware limitation.
The hardware modems are usually half-duplex. If you use the hardware of 
winmodems directly, bypassing their normal drivers, you can get a good 
quality bidirectional channel.
>  
> >Only a fraction of the hardware available is actually capable of full 
> duplex audio.
> >
>  
> Absolutely not the case.  Particularly the softmodems (the most 
> inexpensive) contain little else than what is required for placing and 
> answering full duplex audio calls.  Everything else is in the driver.  
> The OP is 100% correct, that they would be an excellent candidate for 
> FXO use in low volume applications.
Its only really the winmodems that are of interest, so you are quite right.
>  
>  
> >What it really comes down to is a value proposition: 
>  
> Quite true.  This is the real issue.  As mentioned, these drivers 
> require considerable skill and knowledge to write.  While there is no 
> doubt that the result would be very cost effective, the business model 
> is lacking.  The modem manufacturer is going to see the potential 
> market for this as somewhere down in the noise compared to their 
> normal modem sales, so isn't inclined to invest.  A third party 
> developer with the skills would have a difficult time recouping 
> development costs (let alone any profit) because they don't control 
> the hardware, and therefore have no leverage.  A user with enough 
> volume to justify paying for the development (or doing it if they had 
> the skill) probably has enough volume to use T1s instead.  If everyone 
> that could benefit from using a modem card were to pitch in $10 
> towards the development, it would probably be quite possible.  But how 
> to make that happen?
Its straightforward to achieve, but nobody bothers. The Linux drivers 
for most winmodems have the DSP (which you don't need) as a binary blob, 
and the kernel driver (which you'll need to modify) as source code. As a 
modem they normally run at 9600 samples/second. Most of the chips can be 
programmed for 8000 samples/second, though, so they'll do what you need. 
Use them with OSLEC, and you could get great results. There are only a 
few suppliers of these winmodem chips - some USB and some PCI. You 
wouldn't need a lot of drivers to cover practically the whole market.

As I said. Its all possible. The necessary hardware info is mostly out 
there in downloadable source code. Its just that nobody has bothered.

Regards,
Steve




More information about the asterisk-users mailing list