Over the last few weeks I've been trying to diagnose the poor v.90 connections to my ISP here in the UK.

I found this website a few weeks ago and it's been very, very helpful indeed. Definetly one of the better technically oriented sites.

However, I'm still not 100% happy with my connect speeds. I was getting 38-44k connections, and now I'm getting 42.6k connections. I think the reason (other than the reliability factor of v.90) is the poor quality of my phone line. Yet I'm not 100% sure if this is really the problem, because the 3Com diagnostic data is not giving me a clear picture of the problem.

I have a Dell Inspiron 3700 Laptop, dual booting Win98SE and Win2KPro. I have been trying different ISPs (all UK-based), and different modems. All modems have been PCMCIA or PC-Card models.

The latest modem I've tried, an older (1-2 year old) 3Com Megahertz 10/100 LAN+56k Modem, has given me the most consistent connect speeds of 42.6k. It has also allowed me to view all the 3Com diagnostic data covered on this site.

Firstly, lets look at the ATI6 output:

ATI6 FEM556BI LAN + 56K Modem Link Diagnostics...

Chars sent 210965 Chars Received 704600
Chars lost 0
Octets sent 0 Octets Received 0
Blocks sent 0 Blocks Received 0
Blocks resent 0

Retrains Requested 0 Retrains Granted 0
Rate Switches 5
Line Reversals 0 Blers 100
Link Timeouts 0 Link Naks 3

Data Compression NONE
Equalisation Long
Fallback Enabled
Protocol NONE
Speed 42666/31200
V.90 Peak Speed 42666
Last Call 00:06:13

Disconnect Reason is DTR dropped

OK

A good value for Link Naks is 0, but what is a bad value for, say, a 1 hour session?

Also, I don't understand what 'Rate Switches' are. Are they significant? If so, what is a good range?

There is nothing too bad here as far as I can tell, although I could have done this with a longer session.

Next, I checked for the presence of more than 1 A/D Conversion - here is the output from ATY11:

ATY11
Freq Level (dB)

150 32
300 29
450 27
600 26
750 26
900 25
1050 25
1200 24
1350 24
1500 24
1650 24
1800 24
1950 23
2100 23
2250 23
2400 23
2550 23
2700 23
2850 23
3000 23
3150 23
3300 24
3450 24
3600 26
3750 28
OK

As you can see the difference between the 3750 and 3300 values is only 4dB, and all the db readings are relatively low, so it looks like there is only 1 A/D conversion and/or there is little or no distortion or noise on my phone line.

So nothing bad here either as far as I can tell.

Moving right along, here is the ATI11 output:

ATI11 FEM556BI LAN + 56K Modem Link Diagnostics...

Modulation V.90
Carrier Freq (Hz) None/1920
Symbol Rate 8000/3200
Trellis Code None/64S-4D
Nonlinear Encoding None/ON
Precoding None/ON
Shaping ON/ON
Preemphasis (-dB) 0/4
Recv/Xmit Level (-dBm) 24/15
Near Echo Loss (dB) 14
Far Echo Loss (dB) 0
Carrier Offset (Hz) NONE
Round Trip Delay (msec) 6
Timing Offset (ppm) -1815
SNR (dB) 41.3
Speed Shifts Up/Down 0/0
Status : aa,5,13Y,13.2,-0,0N,0,39.2,32.5
OK

This is where is gets interesting:

The Recv/Xmit Level values seem a little high (especially when looking at the The maximum transmit level of the server modem value in the Status line ), although this could be normal for the UK - I don't know.

The Near Echo Loss value seems a little high - or is this normal?

The SNR seems a little low - what range of values is 'normal'?

The SNR ratio of the 2-point training seems a little low too - what range of values is 'normal'?

The Codec distortion seems far too high, based on the suggested normal range of teens to early 20s. What is Codec distortion? Is this a reasonable measure of line quality? This value not always the same, but it never drops below 30!

Reviewing all this, the ATY11 outputs suggests that I have a 56k compatable line and that my line quality is fine. However, some data from the ATI11 output seems to suggest otherwise.

So its not clear whether I have a poor line quality problem. The only other explanation would be that all the ISPs I've tried are crap, and I need to try some more.

But before I hassle British Telecom about my phone line, or start signing up with half a dozen new ISPs, I'd like to know what you think. Am I interpreting everything correctly?

On a slightly different note, I've noticed a recent, disturbing, trend of PCMCIA and PC-Card developers to use non-hardware modems in almost all of their current line up of modem based products.

According to the 3Com website, all but two products in their current line up of PC-Cards with some form of modem functionality, is a WinModem, including their top of the line combo (ie combined with Ethernet Functionality) product.

As it happens, I tried their 'top of the line' combo product, with mixed results. It was connecting at 40k, but occasionally would connect at 44k. The older model, that it replaces and which I am now using, is a full hardware modem combo card, and it consistently connects at 42.6k. I think the newer winmodem model may have given me slightly better throughput, but my testing was not exactly scientific. Unfortuneatly I no longer have access to the newer one to check this theory out. Any differences may have been due to the difference architectures involved - the newer model was a 32-bit CardBus PC-Card (ie the newer PCI based standard for PCMCIA devices), whereas the older model is a 16-bit PCMCIA model. I'd have to say I've have more 'piece of mind' using the older model as it is a proper hardware modem.

The Xircom mobile range also follows this trend. Their range of 16-bit PCMCIA and 32-bit PC-Card modems has more hardware based modems in their range than 3Com do, but the hardware modems are still outnumbered by WinModems.

I've tried a Xircom 32-bit PC-Card CardBus WinGlobal56 (a WinModem, but I don't know whose chipset they used) and a Xircom 16-bit PCMCIA CreditCard 10/100+Modem56 (a Lucent-based hardware modem). While Xircom's support website is better than 3Coms for the models I've been testing, neither Xircom product could compare in performance to the 3Com products I tried.

Obviously this trend in mobile modem connection technology is driven by cost and power consumption. This is where Winmodems and fully software-based modems are significantly more attractive to the developers in this market, and, if you believe the product marketing from 3Com and Xircom, the mobile consumer as well.

Anyway, sorry for rambling on so much , but TIA for any feedback you can give.

Cheers,

The failure to achieve a better 56k rate may be more of design/implementation flaws in the modem(s) you have tried. Have you tried a Lucent or Rockwell-based modem? I would also suggest trying (if available) an external 56k [usr] modem, and comparing the diag results: form-factor limitations for PCMCIA cards introduce some compromise in their design/performance.

Note that the receive/xmt levels reported in i11 & y11 are negative numbers - so a bigger number is smaller. The receive level of the signal from the remote modem is lower (bigger numbers) than I would expect to see (at least in the US - I'm not sure about UK). A lower signal level means higher s/n ratio.

Link naks - a failure in transmission/reception of protocol info = worse reduction in throughput than a bler; rate changes = speed shifts - the modem thinks it can do better, then it can't, and keeps shifting between rates - this is 'normal' behaviour with 3Com/USR modems. I would be concerned if consistently getting 10 or more link NAKs/hr.

I would suggest this w/ current modem: disable 3429 & 3200 symbol rates (which will reduce your V.34 upstream rate) and see if that has any effect on your downstream rates/reliability. ATs$ produces help screen of s-register settings. s33=48 disables 3429 & 3200 on most USR models.

I'd also be interested to know (if you can find out), what kind of facilities your telco serves you with. As is common outside the US, no digital pad is shown in the i11 status line, and the server modem is transmitting a signal at a -13dbm level, yet you are receiving the signal at -24dbm - an 11db loss. If you were on straight copper to the telco, I'd expect a run that produced 11db loss to produce y11 results with greater loss at the high end. Which is why I'd guess that you are served by some sort of digital subscriber loop that isn't handled well in the V90 code....

Yes I've tried at least 1 Xircom Lucent-based hardware modem, and one other Xircom modem called a 'WinGlobal56'. I would assume that all Xircoms are Lucent based? I haven't tried a Rockwell based modem yet. Of the 2 3Com's I've tried one is a WinModem and the other, which I'm using now is a 3Com Megahertz hardware modem. At this stage I haven't bothered with any external modem, due to the mobile nature of my setup, and nor do I have ready access to one at this stage.

I'll have to keep an eye on the Link Naks.

I tried disabling the 3429 & 3200 symbol rates using s33=48, but could only get 28.8k connections, which took ages to negotiate.

I think you're right about the presence of some sort of digital subscriber loop which is non-friendly to v.90. British Telecom have been upgrading all the London area gear for a while now, in preparation for ADSL etc.

It sounds like I'm stuck with 42.6k connects unless I can try some other modems. Such is life . Maybe I'll just have to make the jump to ADSL when it becomes available here (supposedly in a month or 3... )

Thanks for your feedback and keep up the good work. If I find out anymore I'll let you know.