-=> Quoting Dan Bridges to Richard Town <=-
 DB> Is the fallback based on both time and error-rate i.e. if X errors/sec
 DB> for 2 secs causes a fallback, will 2X errors/sec cause a fallback
 DB> after 1 or 2 secs?
But a Maestro's firmware may have made this different.

** An EQM measurement is only directly comparable to other EQM
** measurements at the same data rate and symbol rate.   If you have an
** EQM of 19 at 26.4Kbps and the modem rate renegotiates up to 28.8Kbps
** the new EQM value is not going to be 19.  It might be 25, it might be
** 45.  (It depends on the exact imparements on the line.)
V.FC and V34 may have fairly high EQM values (10 to 30) on a
good connection.  V.32bis typically has EQM values below 15 on a good
connection.
The EQM thresholds are the ones used by the firmware to decide when a
rate renegotiation up or down should be intiated.  Lines with EQM values
below the good EQM threshold for 60 seconds (expected to be reduced to
20 seconds in forthcoming ROM) will rate renegotiate upward.
Lines with EQM values above the bad EQM threshold for about one second
will rate renegotiate downward.
The good and bad EQM thresholds are dependant on the current data rate,
the good threshold ranges from 16 to 25, the bad threshold ranges from
60 to 70.  (These values are for V.34 and V.FC connections, V.32 and
V.32bis connections use different thresholds)
The EQM thresholds used during a retrain are different than those used
during rate renegotiation.  They are associated with a different
constellation than the data phase constellation, so the magnitude of the
EQM measured is very different.
%Q readings are based from the root-mean-square of the
ERROR VECTOR value currently generated in the modem.
The larger the vector the larger the %Q value reported.
The ERROR VECTOR is the average of the vector error for all points in the
type of carrier encoding pattern currently in use (either TCM or QAM -see
below).  The number of points in the carrier increases with the carrier
rate.
The error basically measures the amount of drift between the current
pattern of points on the carrier and the ideal location of these points.
The more the carrier signal is damaged, the greater the drift (or total
loss) of these points will become.
Causes of Damage to Carrier:
----------------------------
The carrier is degraded by echoes, signal loss, phase shifts, frequency
shifts, background noise, and other issues that normally plague
transmission lines.
Also harmonics from out-of-band signals may prove damaging to the carrier
signal.  Basically, the better the transmission lines is able to faithfully
transfer the original signal, the lower the error level rate will be.
The %Q value is a measure of the damage to
the carrier single when compared against the ideal carrier pattern.  The
measurement is in no way a direct measurement of the telephone connection.
The telephone connection from one modem to the other is the major
source of damage to the carrier signal.
%Q command reports the instantanious EQM (eye quality
measurement) of the connection.  Exactly what this means varies with
the protocol being used by the connection.  In non-trellis code modes
it is the square of the error vector and represents the average signal
energy in the error component.  Basicly this is the square of the
radius of a point in the signal constellation (eye pattern).  In
trellis code modes it is the minimum trellis path length.
EQM is not so much a measurement of noise on the line, but a
measurement of the magnitude of the errors being caused in the signal
constellation.  The constellation errors can be caused by many things,
noise, limited bandwidth, etc.  Small errors in the signal
constellation will not cause errors in data transfer.  Errors in data
transfer occur when the signal constellation errors are large enough
that two signalling states begin to overlap.  When this happens either
an error correcting protocol (eg. LAP-M) can be relied upon to fix
things or a lower data rate with a simpler constellation can be used.
If only an occasional data error occurs then trusting LAP-M is the best
choice, this lets data transfer at a higher average rate.  But if lots
of data errors are occuring then more time may be lost in
retransmissions by LAP-M than is gained by the higher data rate, so
decreasing the data rate will result in higher throughput.  At the
higher data rates a limited amount of signalling state overlap is
expected even on perfect lines (in other words don't use V.34 without
error correction, you won't like what happens otherwise).
*Note that Functions/Addresses described below canNOT be accessed
from your computer's interface.
Function 46:    Eye Quality Monitor     Acc. Method: 4  Addr.: 20C
In V.32 4800 bps, V.29, V.27, V.22 bis, V.22 and Bell 212A modes, EQM
is the filtered squared magnitude of the error vector.  However, for
all TCM modes (V.34, V.FC, and V.33 modes, and V.32 12k, 9k6, and
7k2 bps modes), EQM is the filtered minimum trellis path length (or
metric).  This gives a better indication of signal quality for trellis
modes.
The error vector formed by the decision logic can be used to indicate
relative signal quality.  As signal quality deteriorates, the average
error vector increases in magnitude.  By calculating the magnitude of
the error vector and filter the results, a number inversely
proportional to signal quality is derived.  This number is called the
eye quality monitor (EQM).  Because of the filter time constant, EQM
should be allowed to stabilize for approximately 700 baud times
following RLSD going active.
The EQM value for the non-trellis configurations is the filtered
squared magnitude of the error vector and represents the average
signal power contained in the error component.  The power is directly
proportional to the probability of errors occurring in the received
data and can be used to implement a discrete Data Signal Quality
Detector circuit (circuit 110 of CCITT Recommendation V.24 or circuit
CG of the RS-232-C standard) by comparing the EQM value against
experimentally determined criteria (Bit Error Rate curves).
One can illustrate the relationship of the EQM number to an eye pattern
created by a 4-point signal structure (e.g., V.29/4800 bps) in the
presence of high level white noise.  The EQM value is proportional to
the square of the radius of the disk around any ideal point.  The
radius increases when signal to noise ratio (SNR) decreases.  As the
radius approaches the ideal point's boundary values, the bit error
rate (BER) increases.  Curves of BER as a function of the SNR are used
to establish a criteria for determining the acceptability of EQM
values.  Therefore, from an EQM value, the host processor can determine
an approximate BER value.  If the BER is found to be unacceptable, the
host may cause the modem to fallback to a lower speed to improve BER.
It should be noted that the meaning of EQM varies with the type of
line disturbance present on the line and with the various
configurations.  A given magnitude of EQM in V.29/9600 does not
represent the same BER as in V.27/4800.  The former configuration has
16 points that are more closely spaced than the four signal points in
the latter, resulting in a greater probability of error for a given
level of noise or jitter.  Also, the type of line disturbance has a
significant bearing on the EQM value.  For example, white noise
produces an evenly distributed smearing of the eye pattern with about
equal magnitude and phase error.  While phase jitter produces phase
error with little error in magnitude.
Since EQM is dependent upon the signal structure of the modulation
being used and the type of line disturbance, EQM must therefore be
determined empirically in each application.
Note that the eye pattern
is not displayed when 2800 or 3429 baud is selected in V.34 or V.FC
modes.  The use of precoding and shaping in V.34 and V.FC modes will
distort the eye pattern's appearance, even under ideal conditions.
The EQM value should be monitored in V.34 and V.FC modes to determine
the quality of the connection.
Note:  Many modem manufacturers only include the UPPER byte in
       their EQM report... so   300(hex) would be reported 3
Function 69:    EQM Scale Factor (Gain) Acc. Method: 3  Addr.: A29
                EQM ARA Bias (Offset)   Acc. Method: 2  Addr.: 2A4, 2A5
The automatic rate adaptation (ARA) algorithm adjusts the data rate
based on the level of EQM.  The algorithm is twofold, one being used
for initial train and retrain, the other for rate-renegotiation. In
both cases, ARA is enabled by setting the EARC bit (15:0), which
defaults off.
Upon initial train and retrain, the EQM is checked towards the end of
the training, just before the rate negotiation.  The 4-point EQM is
compared against a table of values representing the necessary levels
to achieve the corresponding data rates with an EQM of around 2000h.
once the maximum achievable rate is determined, the CONF register is
changed to reflect the estimate, and is then used to suggest a data
rate in the following negotiations.
In rate-renegotiation, only the instigating modem implements the
algorithm.  The responding modem, as usual, indicates availability of
all rates.  The instigating modem will attempt to go to a rate which
will result in an EQM of 1800h-3000h.  To do that, it will check the
current EQM (at the start of rate-renegotiation), and change CONF
according to the following table:
EQM Before Rate Change  Rate Change
Above 3000h             Down 1
1800h-3000h             No change
C00h-1800h              Up 1
600h-C00h               Up 2
300h-600h               Up 3
Below 300h              Up 4
The rates suggested in the rate-renegotiation will then be reflected
to the CONF.
 
Hope that answers your queries, please come back if needing any more info
rgdZ
Richard
      +===============================+
      | PackLink / Zoom Modem Support |
      |     BBS +44(0)1812972486      |
      |      FidoNet 2:254/235        |
      |     city.view@emarkt.com      |
      +===============================+
... Why pay more?
--- FMail/386 1.02
---------------