Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

[10GBT] Models for Calculating System Margin


Most of us agree that having a reasonable system
margin is essential for a realistic implementation.
There are two areas that still need some
investigation: the THP and the power back off. In both
cases, we need to verify that we have sufficient
system margin, and if not, we need to modify our
requirements to guarantee required system margin.
Though we have reasonable model for channel insertion
loss and alien cross talk there are two areas that are
still seems to be open to discussion: the model for
noise floor, and method of SNR calculations.  However,
to come to an agreement about the available system
margin we need to establish a common understanding
about these two issues.  To expedite the process it
may be useful to discuss the reasonable models on the

1. The noise floor: Following the models used in most
presentations, I have used -104dBm/Hz to model over
all noise floor at the receiver for my calculations.
However, it seems there are some objections to the
-140dBm/Hz noise floor model being too pessimistic. I
am open to changing the model as long as we understand
the implication on the design and there is a wide
agreement. The second problem with the noise floor
model is the effect of power back off. I have used an
optimistic model for my calculation. I think  a more
realistic model is required. One possible model may be
as follows:

noise_floor (in dBm/Hz) = 10*log10(
10^(noise_floor_fixed/10)  + 10^(noise_floor_var/10) *

where noise_floor_fixed models the noise sources that
are independent of transmit power such as back ground
noise and quantization noise, noise_floor_var models
the noise sources that are dependent on transmit
power, and transmit_pow_dBm is the transmit power in
dBm. Once we agree on the two constants, we have noise
floor model for all cases including power back off and
transmit power variation.

2. The SNR calculation approach is also seems to be
open. I have used bandwidth of 400MHz for my
calculation without folding and without any receiver
filter. I think a more reasonable approach is to
consider a receiver filter. I would ask the interested
people to suggest what are the reasonable assumptions
for calculating the SNR. These assumptions must come
from realistic implementation issues and not the need
to get certain SNR.

To reach any agreement, we should at the least,
clearly specify all the assumptions for the
calculations so that others can verify the results.

As for required margin, I would leave it up to the
group. If the group is comfortable with 1dB system
margin, I have no argument. However, we should be
honest and tell that to the group so that they can
make an informed decision. We should keep in mind that
if we are serious about 100m operation, we should be
careful about specifying a system that can operate
over 100m under real conditions.