PE>> From Bill's point of view, the receive levels are most

 PE>> important.



 BL>   Yes. I assume the dB figures are dBm in 600 ohms (db below one

 BL> milliwatt in 600 ohms). 0dBm is 0.77v, -12dB is 0.2V, and -35dB is

 BL> 14mV, which is getting pretty small.



 PE>> I presume these protocols are designed so the roundtrip delay

 PE>> does not impact on performance.



 BL>   I'm not sure about this, so I didn't say anything. If we are talking

 BL> about 28,800 baud or one bit every 35uS, then we are already miles

 BL> faster than the delay. The delay will set the ultimate speed, every

 BL> time both ends have to exchange information. I guess that's why they

 BL> use CRC... it doesn't need confirmation.



 PE>> I have another phone line which I could experiment with at a

 PE>> later date when all other possibilities have been exhausted, if

 PE>> that would make a difference.



 BL>   It'd come through the same exchange, and probably the same junction

 BL> box outside the flats.



 BG>> 2700 36 2850 37 3000 37 3150 38 3300 40 3450 42 3600 48 3750 56



 BL>> This is a frequeny response plot of the signal received by Bill

 BL>> from NetComm. Yours is about the same. You need roughly

 BL>> 1.5-times the carrier frequency (2700Hz) for reliable 28800

 BL>> connects. You can see that it is 



 PE>> 1.5 * carrier frequency = 2700Hz, or 1.5 * 2700Hz = reliable?

 PE>> Or 1.5 times the db rating of 2700Hz = something special? I

 PE>> really have no idea what the above says.



 BL>   The information is sent as a series of tones on top of a carrier.

 BL> The carrier used is 1829 Hz (according to the modem info), and I

 BL> took the highest tone on that as roughly half the carrier (900Hz). I'm

 BL> just picking that out of the air as normal modulation practice - I

 BL> dont;l have th actual protocol. The highest frequency used is the sum

 BL> of the tone and the carrier, or 2729Hz.



 BL>   The dB rating is just a measure of the amplitude of the signal. The

 BL> way it drops off with frequency gives you an idea of how the line will

 BL> handle high-speed transmissions.



 PE>> Ok, I'll assume that "31" is the start of "not
very good" and

 PE>> the higher the more and more worser.



 BL>   Most likely. It will also depend on any interference on the line,

 BL> and how quickly the level drops off once it starts to go. The modem

 BL> can correct for overall level differences at the carrier frequency,

 BL> say, but it is much more difficult to correct for changes with

 BL> frequency.



 BL>> This is the carrier frequency they used, and you can see it at

 BL>> -33dB which is what they quoted for the connect... 33/12.



 PE>> 1829 * 1.5 = close enough to 2700.



 BL>   That's what I used. I should have been more explicit.



 PE>> 33/12?



 BL>   33/12 means -33dB receive level and -12dB transmit level... what it

 BL> gets and what it sends back. 



 PE>> Anyhow, although I have a rough idea what a carrier frequency

 PE>> is for a radio signal, I'm not really sure how I'm meant to

 PE>> apply it for a modem connection, I thought they were meant to

 PE>> use the whole bandwidth not just the bit with the best rating.



 BL>   A modem is fiendishly clever. Perhaps RodS can tell you better; I

 BL> don't know the actual protocols, but it's a bit like colour TV.



 BL>   Original modems used two tones; one for 0 and one for 1, but there

 BL> is a limit to how fast you can switch a tone on or off... roughly the

 BL> bandwidth of the line. For a 3000Hz line this is 300uS or 3300bps. You

 BL> can switch the tone instantanteously, especiually at zero, but it

 BL> decays at the rate of the line anyway.



 BL>   A way around this is to use a carrier with the tone modulated on

 BL> top, and switch the phase of the carrier 90 degrees. You then remove

 BL> the carrier, and what remains is a signal that can be resolved into

 BL> two, 90 degrees apart, by putting the carrier back in again.



 BL>   There is no disturbance on the line to decay, expecially if you pick

 BL> a precise time in the cycle to switch, and it if works for two

 BL> carriers at 90 degrees, then why not three at 60, or six at 30? Of

 BL> course, it gets harder and harder to sort out one from the other. And

 BL> it it works for one tone, then why not two or the or six? The more

 BL> bits you can send at once, the faster.



 BL>   This is the state of play for 28800 modems. They use multiple tones

 BL> to fill the full 3000Hz bandwidth of the line, and they swithc the

 BL> carrier phase in tiny 30-degree steps... using incredibly clever

 BL> software filters to sort it out (not to mention the hardware). It's a

 BL> miracle it works at all, and most of the time, it doesn't. 



 PE>> What makes you think the Netcomm modem has a lower transmit

 PE>> level than a courier? 



 BL>   I was looking at Bill's results comparing DD's local JabberWocky

 BL> with yours.



 PE>> I'd have to put the courier online to see that, I don't think

 PE>> Bill took any line readings whilst I had it online.



 BL>   Oh, perfect! I'd forgotten you still had the Courier. Do that! If I

 BL> am right, the Courier will give lower dB figures overall and let us

 BL> put a true dB figure on the difference between NetComm and Courier

 BL> (if one exists). That aftercon thing of Bill's is brilliant.



 BL>   So far, Bill has assumed that the difference is Rockwell software v.

 BL> USR software. He may be right, but it could be just a simple level

 BL> difference.



 BL>   The only way to solve problems like this is to collect hard data. 

 BL> It's not good postulating causes with no data. The first step is to

 BL> measure anything you can, and try to find a pattern. 



Well BoB you had me sucked in. As I read your message I thought, "this

man knows his stuff" and I was believing everything you said.

Then God.... Sorry Rod, Shot you down in flames. Oh well, It happens to

the worst of us.







Russell



--- Terminate 3.00/Pro