| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Thanks, Christensen for your reply. In answer to one of your questions below, yes I was measuring broadband jitter (ie with a scope) and not narrowband jitter.
I am also interested in finding out the frequency characteristics of these jitters. I can easily find RJ using the spectrum analyzer by putting in 1010...FYI, the spectrum, as expected, is like a VCO spectrum, sloping down from the center. Hence, it is concentrated at low frequencies.
Now, I wanted to do the same for DJ. However, it doesn't seem possible using a spectrum analyzer, right? Firstly, its a large scale phenemonon. Secondlyh, I am inputting 2^23-1, which does not have a spectrum which is easy to quantify accurately.
Any ideas? A theoretical analaysis would be welcome too otherwise I can do the experiment and feed the results back to this group
thanks
Rohit
-----Original Message-----
From: Christensen, Benny [mailto:benny.christensen@intel.com]
Sent: Saturday, April 07, 2001 4:12 AM
To: _Serial PMD Ad Hoc Reflector (E-mail)
Subject: RE: Jitter method for serial PMDs
Hi Rohit
Some comment (from my experiences) with DFB laser diodes.
I'm glad that you manage to feed more real data (though at 2.5 Gb/s)into
this jitter discussion, as my fear is that most negotiation and paperwork
compromises are based on performance expectations rather than real world
measurements and limitations.
More comments/answers below
Benny
--------------------------------------------
llllllll ii llllllll llllll
ll ll ll ll
ll llll ll ll llll llllllllll
ll ll ll ll ll ll ll
llllllll ll llllllll ll ll
GIGA, an Intel company
Benny Christensen, M.Sc.E.E, Ph.D.
Mileparken 22, DK-2740 Skovlunde, Denmark
Tel: +45 7010 1062, Fax: +45 7010 1063
e-mail: benny.christensen@intel.com, http://www.giga.dk
-----Original Message-----
From: Rohit Mittal [mailto:RMittal@oni.com]
Sent: 6. april 2001 20:26
To: _Serial PMD Ad Hoc Reflector (E-mail)
Subject: RE: Jitter method for serial PMDs
I am curious to know, has anybody tried to measure jitter out of a DFB etc.
since this is the technology we will be using for longer lengths.
In my measurements, I saw that a DFB adds a lot of Deterministic jitter but
no Random jitter. Most of the random jitter was the random jitter of the
preceeding electronics (Eg. CMU + driver).
>The (continuum spectrum) random jitter of a transmitter is mainly due to
the phase noise of the VCO (+ transferred jitter/phase noise from the
reference clock, due to the PLL, which typically have MHz BW in order
suppress as much phase noise as possible). The (continuum) thermal noise of
the following circuits are neglectable contributors.
>The DJ (discrete line spectrum on TIA) of the electronics is mainly pattern
jitter (or DCD) due to non-linear phase distortion and the corresponding
(finite) BW impacts (ie. group delay variations vs. frequency). Remember,
that for digital signals, the GD is a large signal swing characteristic,
that may differ from the small signal characteristic, which is normally the
case when measuring with a Network Analyser. Also a (non-symmetric)clipping
effect may also add to the DJ pattern jitter. Unwanted low frequency pattern
dependent cross-talk effects into the PLL (or reference clock) should be
observed for fully integrated modules - but not a problem in your set-up)
> The optical source is a seperate chapter (contributor). Commenting only on
DFB lasers for the following. Depending on how close one modulates down to
the lasing threshold (or even below, due to electrical undershoot from
driver or impedance mismatch or AC coupling) the DJ is significant. Also,
the laser has it own GD characteristics, which varies with temperature and
operation point and maybe drive voltage.
> After a laser turn off, the on-set of lasing (ie. coherent light) is
triggered by the spontaneous emission into the lasing mode - ie. the exact
turn-on is a random process, driven by /dependent on/correlated with the
pattern (effect). That is, the turn-off is probably deeper for two/several
consecutive zeros that one, so the turn-on process may take a little longer
after several seros. Also, the turn-on is a very fast process, as the lasing
medium pumped high above the steady state carrier level (which also give
rise to the large overshoot and relaxations oscillations), whereas the
turn-off process is governed by the carrier life-time in the laser cavity,
so this is a slower (decay-ing) process. Meanwhile, the laser chip local
temperature normally have time constants in the MHz range, so the low
frequency components of the pattern impact the temporal laser
characteristics for the following bits (ie. a pattern/memory effect
affecting the transparency point - lasing threshold (pattern dependent)
wander).
>The deep turn-off can be avoided by lowering the extinction ratio (min 3-4
dB or OMA) as in 10GE, but for SDH equipment (STM-16) normally 10 dB EX or
better is used/specified.
So, I had a bert driving an external laser. The bert random jitter was 8 ps
and the Bert + external laser gave random jitter 11 ps (All in
peak-to-peak). The bert DJ was about 14ps, whereas the Bert + external laser
DJ was about 50ps!
> Just to clarify: Is this broadband jitter (ie. measured on the
oscilloscope eyediagram, which is from DC to 20 GHz)??? or the GE TIA
bandpassed method, or SDH (20? kHz to 20 MHz for STM-16 - 2.5 Gb/s). If the
first method, consider the DCD contribution.
>The 8 ps pp is typical the limitation of the scope itself (ie. trigger and
sampling noise and may not be RJ from the TX electronics). At least my TEK
CSA803 has a broadband pp jitter figures in that range, even with a very
clean oscillator (<-140 dBm/Hz).
Can anyone shed light as to why a DFB adds so much DJ and is this low
frequency/high frequency phenomenon. And is it true for other laser
technologies, such as FP. I believe VCSEL is very different but have no
proof of that.
>Yes, different may mean even worse!!
ps: My measurements were done at Oc48. For RJ, I put in 1010... For total
jitter I put in 2^23-1. So DJ is got by subtracting one from the other,
roughly.