RE: PAM-5, what are your BERs ?
I believe we agree most of whatever we have been reasoning. However, the
details of operations, and the quantitative figures are the areas we did not
have opportunities to discuss, which does not mean we disagree.
The single prohibitive task to perform here is to make all these complex
circuits to perform at 10 GBaud bit rate and 5 GHz clock rate with BER of
10^-12 or less.
As you have already identified that it is questionable the bandwidth of CMOS
technology and other parts can deliver the performance we need.
At 10 GBaud, the data pulse waveforms are quite departed from those ideal
square waveforms, and are quite distorted. Any normal stray capacitance and
line inductance of components and pc runs will cause visible distortion to
the waveforms. I am not sure if the waveform at the input of a A/D
converter is same as the one A/D converter is actually digitizing. The A/D
converter it self can alter the waveform. It is very hard to use Fourier
analysis, or a Gaussian pulse approximation to characterize the pulse
responses, which are quite distorted. As a result, the equalization result
may not be as predictable as expected. It means an incorrect restructuring
of the original waveform. It means high BER.
I just select one example to emphasize the difficulty of a data recovery at
The right approach is to keep circuit as simple as possible to avoid
additional waveform distortion caused by its own circuit.
Usually the waveform at the TIA out put is very sensitive to the S/N issue,
especially when the input power is near sensitivity level. One would
pre-amplify the incoming signal first before equalizing the waveform. This
amplifier better be a real good one, which is allowed to amplify the data
but not the noise.
If the waveform is equalized first, I bet, the passive equalizer may
attenuate the low amplitude signal to case insufficient S/N problem. It
means high BER.
The worst thing: if the eye is closed, we do not know how low is the signal
amplitude. It is possible, the signal amplitude of a narrow pulse (high
frequency pulse) is far below the minimum required optical power at the
There will be many new issues we did not have deal before.
While I agree, theoretically, a closed eye may be recovered depending on the
S/N ratio, it is highly recommended to adopt a approach that is easier for
designer to work on it. -- a wider eye the better.
Edward S. Chang
NetWorth Technologies, Inc.
[mailto:owner-stds-802-3-hssg@xxxxxxxx]On Behalf Of Vivek Telang
Sent: Tuesday, February 29, 2000 2:15 PM
Subject: RE: PAM-5, what are your BERs ?
Good points. Let me see if I can respond to them.
You're right about the fact that equalization has its limits. These limits
are well understood, and given the channel and noise characteristics, one
can easily determine the number of levels that can be supported at a desired
BER. My point was that this limit is far greater than the one that the open
eye requirement would lead you to believe.
The high-pass filter equalizer approach that you describe is not as optimal
as a minimum-mean-squared-error (mmse) equalizer. It is important to realize
that the narrow pulse does not simply get attenuated, but is dispersed over
time. A good mmse equalizer processes numerous samples of the dispersed
pulse and reconstructs the symbol while trying to minimize the noise energy.
DFEs are particularly good at doing this.
Your points about the linear amplifier (bandwidth, non-linearity) are
correct; however, they are not an issue if the equalization is done
I can hear you saying, well, if DSP can do all that, what's the catch? The
catch is that to do any DSP at all, you need an A/D converter up front. This
A/D needs to run at the symbol rate at least. Also, the A/D adds
quantization noise that needs to be low enough so as not to be the
performance-limiting factor. For Oscar's 10G DSP proposal, this means a
5Gsps A/D with 6 effective bits of resolution. No mean task in CMOS. Also,
all the DSP needs to run at 5GHz.
So in summary, I would reiterate my two comments:
1) A closed eye does not necessarily preclude low BERs.
2) DSP techniques which would be required to operate under "closed eye"
conditions are going to be a challenge to implement at 10G bit rates.
* Vivek Telang
* Cicada Semiconductor Inc.
* 901 MoPac Expressway South
* Building One, Suite 540
* Austin, Texas 78746
* 512-327-3500 x114 voice
* 512-327-3550 fax
From: NetWorthTK@xxxxxxx [SMTP:NetWorthTK@xxxxxxx]
Sent: Tuesday, February 29, 2000 12:15 AM
To: vivek@xxxxxxxxxxxxxxx; stds-802-3-hssg@xxxxxxxx
Subject: Re: PAM-5, what are your BERs ?
>From theoretical point of view, you reasoning makes some points. However,
from the real implementation point of view, it is not quite true. Before
starting analyzing the frequency response, just ask a question: "If we can
simply keep equalizing the receiving signals to bring them back to the
looking-alike to the original, transmitting signal, why we bother all those
bandwidth issues? There must be some limitations to the equalization
The eye closure is caused by the insufficient bandwidth of a receiving path;
as a result, the narrow pulse (higher frequency pulse) is much more
attenuated than the wider pulse (lower frequency pulse). We can cascade a
high-pass frequency response equalizer to suppers the amplitude of a wide
pulse, and keep the amplitude of the narrow pulse remain unchanged (but not
amplified) to open the eye. However, if the amplitude of a narrow pulse is
already too small to meet the minimum S/N requirement, the equalizer is
useless. Theoretically, a linear amplifier can be added to bring the signal
amplitude up to meet the minimum S/N requirement. The linear amplifier will
need a BW larger than the transmitting signal rise time. Furthermore, any
deficiencies in the linearity, will add both timing and amplitude distortion
to the received data. The additional distortion is not included in the
jitter specification; as a result, the link will cause higher BER.
Especially in a high data rate link, a linear amplifier may cause more
than the expected benefit. In practice, it is impractical to add a linear
The right way is to keep eye open at the receiver input.
NetWorth Technologies, Inc.