transient noise analysis?
Nor Hisham Hamid
Andrew Beckett
colleagues for this discussion below which describes why implementing
a transient noise analysis is problematic):
There are number of
difficulties with this approach. First, to adequately model white noise
requires a small, and very well controlled time step. Second, to
adequately model flicker noise requires long simulation intervals.
Third, to accurately see the effect of the noise during a transient
analysis requires extremely tight tolerances. Consider trying to see the
effect of noise from a 50 Ohm resistor during a phase noise analysis of
an oscillator. Assume that the oscillator operates at 2.5 GHz, and you
are interested in the noise from 100 Hz to 10 MHz from the carrier. To
model the noise appropriately at 10 MHz requires the noise sources to be
updated every 100 ns at a minimum (not a problem here because the period
of the oscillator is 400 ps, meaning that the time step would be on the
order of 2-4 ps). However, trying to see flicker noise at 100 Hz would
require that the simulation interval be 10 ms, and to achieve any
accuracy would require at least 100 times that, or 1s. With the 4 ps
time step, one would need 250 billion time points, which is impractical.
Furthermore, the noise voltage produced by our 50 Ohm resistor over a 10
MHz bandwidth is only 3 uV. If we have signals on the order of 1 V in
our oscillator, then we will need reltol set to no more that 1e-6 to
force the simulation errors to be comparable in size to the noise. I
would feel the need to set it to 1e-7 or 1e-8 to feel comfortable with
the accuracy. Of course, this will further shrink the time step. You
could imagine artificially inflating the size of the noise to get it
above the simulation error, but you have to be very careful not to make
it so large that it starts changing the behavior of the circuit. For
example, if we increase the noise produced by all of the noise sources
in our oscillator by a factor of 100 so that we could use reltol=1e-6
rather than 1e-8, then we will have also increased the line width of the
oscillator to the point where it is probably affecting the results above
100 Hz.
We do however have a number of small-signal noise analyses, including
those which work with periodic circuits. These work by adding a number
of small signal noise sources to each device and computing the transfer
function of each to the output, then summing up all the noise powers at
the output.
That said, we do have something called "tdnoise". What this does is
allow you to look at how noise varies in the time domain in a periodic circuit,
as simulatable with PSS analysis in spectreRF. This allows you to see how
the noise changes over the PSS fundamental period, and can be quite
instructive for periodic circuits.
Regards,
Andrew.
On Tue, 25 Mar 2003 13:28:29 +0000, Nor Hisham Hamid <Hisham...@ee.ed.ac.uk>
wrote:
>What method available to do transient noise analysis?
--
Andrew Beckett
Senior Technical Leader
Custom IC Solutions
Cadence Design Systems Ltd
b
HTH
"Nor Hisham Hamid" <Hisham...@ee.ed.ac.uk> wrote in message
news:3E80597D...@ee.ed.ac.uk...
Erik Wanta
---
Erik
"b" <b@f.g> wrote in message news:<b5v10c$8ge$1...@info4.fnal.gov>...