[plug] Re: [plug] modems (was Re: [plug] Gut modem für Linux? [those blinkinglights again])

[Jason Nicholls]

G'day,

On Sun, Aug 20, 2000 at 11:25:28AM +0800, BillK wrote:
> Hi Brett,
> 	modem modulation at 33.600 and below uses various analog techniques to
> sustain the analog signal.  V.90 speeds above this actually try to use
> the line as a digital line with compensation and other tricks to sustain
> the signal (and lots of computational power needed at both ends).  Yes
> the theoretical maximum speed of an analog line using current modem
> techniques is 33.6, but this limit is bypassed whith V.90 techniqes, not
> increased if you see the difference. 

The increase in connection speed is more a function of removing losses
associated with DA <-> AD conversion (over distance) and increasing the S/N
ratio. Previously with old setups the system would be something like this:


 modem (DA) <-> (AD) telco infrastructure (DA) <-> (AD) ISP
 	   ^^^^^				  ^^^^^^
	   local loop (house to exchange)	  local loop (exchange to ISP)


Current 56k modem technology uses the ISDN infrastructure which wipes out one
of the DA/AD (over distance) stages and significantly reduces losses:

                  |<- no loses here ------>|        
 modem (DA) <-> (AD) telco [idsn] <-> ISP (DAD) 
 					 ^^^^^^^ cust -> ISP 
					 	 isdn to the original analogue 
					 	 signal and then demodulated.
						 (or reverse)

This is a little simplified but the loses associated with the ISP end of the
local loop are gone. In the end what happens is that the ISP is able to pump
out a maximum of 56k out to the customer and receive a maximum of 31.2k back
(yes v.90 is 56k in, 31.2k max out).

The maximum theoretical data carried on a line using limited bandwidth is
governed by:

	maximum bits/sec = H log2(1 + S/N)

	where H = bandwidth (roughly 3.6k over the telephone system)
	and S/N is roughly the signal to noise ratio experienced.

The S/N achievable via the ISDN system is much larger compared to the old
method. I don't have an idea what this would be over the crappiest leg (local
loop home, exchange), but traditionally this would have been around 30DB.

	so for about an SN of 30DB we get:

		= 3600 * log2(1 + 30DB)
		= 3600 * log2(1001)
		= 35882
		  which is roughly what is expected for 33.6k communications.

	with a larger SN the value increases. but as I mention above I
	wouldn't be able to guess as to what this number would be (but let's
	try something like 60DB).

		= 3600 * log2(1 + 60DB)
		= 3600 * log2(1000001)
		= 71753

	(btw: 50DB = 59794) etc...

These figures represent the absolute max the line can carry, it doesn't matter
if you try sampling quicker, or add additional signal levels etc...

Anyways, it should also be noted that this is an upper bound and a real system
would only be able to reach it in ideal situations.

Hehe, I just used the iiNet modem diag for myself and see the following:

	Signal Noise Ratio:  39 dB
	(ie: 44k, currently connected crappily with v.34+ at 33.6k)


Many of the benefits with the new system are felt even for older modem users:

	- lower latency (down from 200+ms -> around 110ms)
	- better S/N which means potentially faster connects

A v.90 modem will allow you to make the most of the increased S/N.


PS: Phew, I hope nothing above is outright incorrect ;)


Later,

Jason Nicholls
--------------------------------------------------------------------
Jason Nicholls    icq: 11745841    email: <jason at mindsocket.com.au>
Proprietor                        mobile: 0417 410 811
Mind Socket [web services]          http://www.mindsocket.com.au/
--------------------------------------------------------------------