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

Re: [10GBT] PAM8 and PAM12 sys time domain analysis


Thanks for your feedback. Pls see discussion below.

sailesh rao wrote:

> Albert,
> Thanks for the input.
> Please note that I'm not particularly religious about which transmit
> filtering method to use so long as we can meet emissions requirements
> comfortably. Therefore, I'm considering co-opting the fine transmit
> filtering method proposed in powell_1_0704.pdf to use in a PAM8 system
> - I
> think it's about time I take something from Scott, for a change :-).
> Having
> said that, I've several questions and comments on your simulation
> results.
> 1. In the first paragraph, you state "Study systems including PAM8
> @1GHz...., and an exemplary PAM12 system@825MHz. The latter was
> derived by a
> simple frequency scaling of respective elements in the PAM8 system". My
> question is why didn't you study the actual PAM12 system proposed in
> powell_1_0704.pdf?

The main intent of the comparative study was to investigate how the
system performance evolves w.r.t. baud rate and PAM modulation
index, having the system structure largely preserved. The approach
allows to "weight" an incremental contribution of a block or function
by observing its effect on the particular system characteristic.
Another important aspect for comparative analysis is the MMSE

Focusing on only few key system elements is usually sufficient to get
meaningful results in reasonable time.

Choosing, as a base, the multi-vendor proposal "powell_1_0704" with
a relatively wide scope and considerable volume (diversity) of
material, perhaps, would not have served the study goals just
as well.

> 2. On the forward equalizer frequency response plots on Page 3 (Figure
> 1),
> why are there such huge ripples in the FFE responses? In contrast,
> there are
> no such ripples in the FFE frequency response plots shown on slide 24 of
> ungerboeck_1_0704.pdf.

The observed "wavy" behavior is not unusual in discrete-time
adaptive systems, especially with insufficient degrees of
freedom, when attempting to approximate a wider band
continuous-time model. Ripple means that the system is not
able to achieve the "optimal" solution for all frequency points
in the band, but few.

One would expect a fractionally-spaced equalizer (an over-sampled
system) to produce "smother" results in the studied examples.

As far as mentioned Dr. Ungerboeck's results are concerned -- it
would be difficult to get one exact answer without spending some
time for the due analysis...

One likely reason for a smoother FFE gain could be the much more
aggressive band limiting in the system -- a zero at Nyquist is
forced in the DFE, plus the 5th BW LPF at 200 MHz.

Another "helping" factor is the fixed "smooth" DFE IIR. Under
this constraint, the receiver FFE effectively becomes quasi-MMSE.
Yet, some ripple still could be detected by a sharp eye:) -- on
the mid lower graph.

(With Frequency-Time dualism it is not surprising to find that
relatively smooth pass-band type FFE has now its ripple in the
time domain -- see ibid., p. 24, right corner top and bottom.)

> 3. On the same figure, why does the PAM8 FFE frequency response tail off
> very slowly towards fs/2 whereas the PAM12 FFE frequency response drops
> about 10dB lower at fs/2?

The MMSE-FFE tries to achieve maximum gain in the frequency band
where SNR is high, reducing the gain as SNR drops. This could be
observed in Fig. 1. Two systems -- PAM8F and PAM12 show steeper
than PAM8 gain slope in the upper band, as they "run" out of
SNR "faster".

Hope, this addresses the questions.



> 4. For Model 3, the DFE SNR Margins computed using solarsep_varlen7a.m
> are
> PAM8: 5.2dB relative to 19.9dB for 1E-12 BER
> PAM12: 5.7dB relative to 23.8dB for 1E-12 BER
>> From the SNR at the slicer that you are computing (Table 2 on page
>> 2), I get
> PAM8: 1.0dB relative to 19.9dB for 1E-12 BER
> PAM12: 0.27dB relative to 23.8dB for 1E-12 BER
>> From this, it does appear that your simulation results are supporting my
> contentions on PAM8 vs. PAM12.
> Regards,
> Sailesh Rao.
>> From: Albert Vareljian <albertv@IEEE.ORG>
>> Reply-To: "IEEE P802.3an" <>
>> To:
>> Subject: [10GBT] PAM8 and PAM12 sys time domain analysis
>> Date: Mon, 26 Jul 2004 19:57:28 -0700
>> Hi All,
>> Pls find attached pdf report on PAM8 and PAM12 systems
>> time domain simulation and comparative analysis.
>> Regards,
>> Albert Vareljian
>> << PAM8_PAM12.pdf >>
> _________________________________________________________________
> Express yourself instantly with MSN Messenger! Download today - it's