Re: Digital Telephony costs $2

[jim bell <jimbell@pacifier.com>]

At 01:50 PM 8/4/96 -0700, Bill Stewart wrote:

>One advantage of higher-speed modems is that you can get away with
>16kbps ADPCM coding, which is dirt-simple computationally;
>your 386 probably has enough horsepower to do it, though a PC's
>interrupt structure may make it tough to shove all the data in and
>out in real time.  You still need a sound card that'll do the
>A/D and D/A conversion simultaneously if you want full-duplex;
>that wasn't part of the original market vision of Soundblaster,
>so vanilla sound cards don't all do it.

What is unclear, however, is WHY they "had to" build a card that couldn't do 
full-duplex.  I mean, would there have been a problem implementing that?  Or 
was this just another one of those stupid design decisions which could have 
been easily fixed if it had been realized in time?


> It also has the advantage
>that the data is being moved through your CPU, so encryption is
>an easy add-on, rather than having one combined modem/voiceblaster
>card which doesn't have any hooks for crypto or other processing.

Well, I assume that if implemented as a new type of modem card, the 
processor can be used to do the data transfer.


>
>>Sure,  it may not be necessary to compress voice audio all the way down to 
>>2400 bps, since the current modem standards allow 28.8kbps and beyond, but I 
>>suggest that decreasing net traffic by a factor of 12 (28.8k to 2.4k) is a 
>>desirable goal.  
>
>One problem is that tighter compression methods are far more sensitive
>to network latency than crude ones, and need to process more milliseconds
>of speech before putting out a packet on the net (e.g. a 64-byte tinygram
>is 200ms of speech at 2400bps, vs. 32ms at 16kbps.)  For modem-to-modem
>communications, this is no problem; for Internet random delays it is.

I see what you're saying; this makes sense.  Maybe what the industry is 
going to have to do is to start out at 16kpbs, reserving full 2400 bps 
compression for a (near?) future time when network latencies are low and 
predictable.


>Also, another big difficulty with full-duplex transmission is that you
>need echo-cancelling, especially with high-latency circuits.
>Half-duplex is annoying, but it doesn't echo, and it's more tolerant
>of delay because you're not expecting it to have natural timing...

Fortunately, this is the kind of thing that DSP's are good at...


>>The reason I think a system I've described has a future is that modem 
>>manufacturers have a PROBLEM.  Their problem is that they've pretty much run 
>>out of room to improve the bit-pushing through a 3 KHz bandwidth.  
>
>Given that the "3KHz" is almost universally transmitted over 64kbps
>digital channels, there's really no point in pushing past 33.6 with
>analog-based coding; better to just do ISDN. 

The local phonecos still want to overcharge for ISDN, however.  Major 
bigtime problem.  ISDN looked great back in about 1980 when the fastest 
common modem was 300 baud, but it's lost much of its lustre competing 
against 33.6 kbps.  Maybe if ISDN were available at a premium of $5 per 
month or so...

Jim Bell
jimbell@pacifier.com