Index Home About Blog
From: (Floyd Davidson)
Newsgroups: comp.dcom.modems
Subject: Re: Slow Throughput--Modem, Phone Line or ISP?
Date: 3 Jul 1999 04:04:09 GMT

Kelly Loch <> wrote:
>What can affect signal to noise ratio?  Modem getting too warm?
>Earlier today I was getting 45 db, 45,333:
>Preemphasis (-dB)        4/6
>Recv/Xmit Level (-dBm)   20/10
>SNR (dB)                 45
>Near Echo Loss (dB)      8
>Far Echo Loss (dB)       77
>Roundtrip Delay (msec)   5
>Status       uu,5,12N,0.0,-0,0N,0,0.0,0.0
>Then later tonight I get disconnected and "unable to retrain" at 33,600.
>Preemphasis (-dB)        2/6
>Recv/Xmit Level (-dBm)   20/11
>SNR (dB)                 30
>Near Echo Loss (dB)      22
>Far Echo Loss (dB)       60
>Roundtrip Delay (msec)   5
>Status    uu,5,12N,0.0,-0,0N,0,0.0,0.0

The first reading is taken while the modem is operating with
v.90 protocols, and measures the SNR between the D/A conversion
at the line card for your POTS line and your modem.  The second
reading is taken when the modem is operating with v.34
protocols, and measures the SNR between the server modem's
virtual analog input and your modem.  The first SNR figure is
basically the SNR of a cable pair (commonly that might be 55 dB
or better), while the second includes that cable pair plus an
analog 4Khz voice grade circuit transported via a digital
carrier system that has a best case SNR of 37 dB.

An SNR of 45 dB is not exactly great for a cable loop, and any
other added impairment would then prevent a v.90 connection
perhaps.  An SNR is 30 dB over a digital system is not great,
but it isn't terrible either.  I'd be surprized that you get
better than about 26Kbps with an SNR of 30 dB though.

However, modems are not known to provide accurate measures of
SNR.  The effect is that you can compare readings taken with the
same modem when it uses the exact same configuration.  You
cannot reliably compare readings taken with different modems
(even the same model) or with even with the same modem using a
different configuration.  Hence while I said above that an
analog channel over a digital carrier system will get 37 dB at
best, it is very hard to say what the modem reading will be when
that is the SNR.  It might, for example, say 40 dB!  Or 31 dB.
We don't know, and therefore we don't know what your modem's 30
dB report means.

All we can say is that tomorrow if you do exactly the same test
with the same configuration and it reads 26 dB we will know that
it is worse than it was when the 30 dB reading was taken.

As far as heat goes, A modem that has that problem when it is
warm is a defective modem.  That could happen of course, but it
isn't likely.


Floyd L. Davidson                      
Ukpeagvik (Barrow, Alaska)             
     North Slope images: <>

From: (Floyd Davidson)
Newsgroups: comp.dcom.modems
Subject: Re: 45333, no digital impairments, 43 SNR
Date: 28 Nov 1999 19:57:52 GMT

Hooda Gest <> wrote:
>Floyd Davidson wrote:
>>Hooda Gest <> wrote:
>>>I'll toss something else in now...
>>>In my opinion, the SNR reported by these modems is neither
>>>accurate or reliable. Why would the SNR change radically
>>>between V.34 and V.90 (or x2 or Flex)? Is the V.90 signal that
>>>much stronger than V.34's that the noise (a nominal constant on
>>>a good line) "drops" by 20 points? I tend to look at the SNR in
>>>V.34 as a better indication of the "real" SNR. Then factor in
>>>Receive level and rolloff.
>>The two SNR reports are for two distinctly _different_ channels!
>>The SNR reported for a v.34 connection is for an analog channel
>>from one modem to the other.  That includes any PCM analog
>>channel that is transported via a digital link, and of course
>>any such channel will have an absolute maximum SNR of 37 dB due
>>to quantization noise added on the receive end when the PCM
>>codes are converted back to output levels which don't exactly
>>match the input levels.  Actual measured SNR on such channels is
>>very likely to be 35 or 36 dB; however, that will drop by 1/2 dB
>>or so for every link using robbed bit signaling that grabs
>>another bit, and will drop by 3 dB every time two codecs are
>>placed back to back, as is commonly done with subscriber line
>>units.  Hence what might be commonly seen in the real world will
>>vary from about 30 dB to 36 dB, and occasionally a dB higher or
>So I can apply my use of the V.34 SNR to get an idea of
>potential quality of the line since it will reflect things
>which will adversely affect the use of 56K technology. Yes?

It may or may not.  A v.34 SNR modem measures the noise
resulting from conversion of an analog signal to and from
digital PCM codes.  That conversion doesn't happen with v.90,
and part of the SNR measurement which is affected by that
quantization noise is immaterial to how well a v.90 connection
might work.

Given that v.90 needs something like a 45-50 dB SNR on the
cable, but v.34 will mask that with the minimum of 37 dB SNR
from the unused PCM analog conversion...  it rarely will be
useful.  But probably any measurement that indicates more than
say 4 dB degradation from the 37 dB best case, means that
something else is adding at least as much noise, which will
almost guaranteed to be enough to affect a v.90 connection.

Hence I'd guess that readings of 34 to 38 dB SNR over a v.34
channel are not indicative that the channel will work for v.90
also.  But values worse than about 34 dB probably mean too much
impairment for v.90.  Of course with any given modem a 34 dB SNR
might give a reading of 30, or 50.  Or something in between.

>>Obviously either these modems do not accurately measure SNR or
>>what I'm saying is typical is not typical.  Since we _know_ that
>>I would never tell a lie, the _modems_ must be lying to us!
>Well, I wouldn't say they lie to us but I would say that these
>numbers are not accurate measurements in telco terms, as you
>seem to agree. I look at the modem reports as indicators rather
>than actual measurements.

Exactly.  With the typical test equipment used in a telco
office, we can line up 5 different brands, and 5 of each model,
and they will all read the same thing within about 1 dB for SNR.
Try that with modems and the SNR values will be +/- 6 or 7 dB at

I had an early experience, years ago, with using a "modem" to
measure SNR that put it all into perspective for me.  We
acquired an (expensive, about $12 grand worth) HP piece of test
equipment (HP4847 or 4849???) that would lock up on leased line
modem protocols and do diagnostics similar to what modern
v.34/v.90 modems do.  This thing would read gain dropouts, phase
shifts, SNR, signal levels, and all kinds of neat things.  But
the first time I wanted to read the SNR on a private line
circuit, it didn't seem to be very good.  Then I discovered that
a "circuit" consisting of a 3 foot patch cord going to TX out to
RX in, had a SNR of 28 dB.  (Specs are 24 dB minimum on most
circuits, and 28 dB on D1 conditioned data lines.)

Clearly, using 1980's technology, it was not possible for the
test set to distinguish well enough between signal and noise
to separate them better than 28 dB.

I haven't regained much confidence in modems as SNR measuring
tools... :-)


Floyd L. Davidson                
Ukpeagvik (Barrow, Alaska)

Index Home About Blog