Re: [10GBT] THP Fixed Set TF
While I agree with Gottfried that the effect is minimal, in practical
systems, the receive filter does have an effect, for two reasons:
1) Finite length effects of the feedforward equalizer: the receive
filter convolves with the channel to influence the loss associated with
finite length effects of the FFE. To the extent that the precoder
doesn't match the channel in the tails (which it generally won't, both
because of quantization, mis-match, and finite length effects in the
precoder, and because since in the tails of impulse responses for
practical cables you see various random reflections and imperfections,
which vary from installation to installation), you must rely on the FFE
to remove the additional ISI from these effects, and therefore, the
finite length of the FFE must be taken into consideration. The receive
filter effects this.
2) While I agree that IDEALLY the SNR(f) function is not changed by the
receive filter, PRACTICALLY it will be. If ALL of the sources of
receiver noise were prior to the receive filter, Gottfried's statement
about the SNR being unchanged with respect to the receive filter would
be true. However, as has been pointed out on several occasions by Dr.
Ungerboeck's colleagues at Broadcom, the A/D converters are a
significant source of noise. The receive filter between the sources of
noise coupling on the line, and those coupling in the receiver (e.g.,
ADC noise) effectively weights the receiver noise (including, but not
limited to ADC) in the frequency domain with respect to the noise
sources on the line. (Prof. Spencer showed this nicely both in his
study-group presentation in July'03 (spencer_1_0703.pdf), and later in
his journal paper.) This changes the SNR profile, particularly for long
lines, in a way that IS dependent on the receive filter, although not
hugely.
All this, in the end, winds us up in a fine argument reminiscent of
angels dancing on the head of a pin. I'm beginning to think that a
simple, fully programmable precoder with coefficient exchange (as was
used in HDSL2 and other systems) would be the right approach, and this
"set of fixed precoders" is either doomed to a host of practical
interoperability problems, or necessitating significant extra complexity
in the rest of the system to compensate for cutting a small corner on
the TH precoder. I'm not ready to abandon it right yet, but we are
carefully evaluating the small set of fixed precoder concept for
robustness across front-end component design, component variation, and
line variation.
-george
-----Original Message-----
From: IEEE P802.3an Reflector [mailto:stds-802-3-10gbt@ieee.org] On
Behalf Of Gottfried Ungerboeck
Sent: Wednesday, December 22, 2004 5:05 AM
To: STDS-802-3-10GBT@listserv.ieee.org
Subject: Re: [10GBT] THP Fixed Set TF
Hi Pedro and fellow 10GBASE-T-ers,
please note that the receive filter has two purposes: (1) attenuate the
received signal at frequencies beyond 1/2T in order to suppress
out-of-band
noise and minimize the effect of sampling-phase dependent aliasing, and
(2)
support the T-spaced FFE in its task of shaping the received signal
towards
the target symbol response (=TH precoding polynomial).
The spectral shape of the receive filter has **no** influence on the
optimum
target response to be employed for TH precoding (if one ignores effects
of
sampling-phase dependent aliasing). The optimum target response is
determined by the spectral signal-to-noise function SNR(f) of the
received
signal. Ideally, the receive filter does not change the SNR(f) function.
Regards,
Gottfried
-----Original Message-----
From: IEEE P802.3an Reflector [mailto:stds-802-3-10gbt@ieee.org] On
Behalf
Of Reviriego, Pedro (Pedro)
Sent: Donnerstag, 16. Dezember 2004 13:06
To: STDS-802-3-10GBT@listserv.ieee.org
Subject: Re: [10GBT] THP Fixed Set TF
Hi Albert,
So far several precoder proposals have been presented: one is your
proposal
and another one is Ungerboeck's.
They assume different Rx Architectures, namely analog high pass boosting
is
used in yours but not in Ungerboeck's. I think we need to assess the
performance loss of your precoders with a receiver that does no boosting
and
the performance of Ungerboeck precoders with a receiver that includes
boosting.
This will show if setting the precoder coefficients limits the options
for
Rx implementation in which case we should consider making the precoder
programmable.
Regards,
Pedro
-----Original Message-----
From: Albert Vareljian [mailto:albertv@ieee.org]
Sent: Wednesday, December 15, 2004 7:36 PM
To: Reviriego, Pedro (Pedro); STDS-802-3-10GBT@listserv.ieee.org
Subject: Re: [10GBT] THP Fixed Set TF
Pedro,
I did not use the Rx filter you mentioned below primarily due to
aliasing
associated with its not very steep roll-off.
(Also pointed by Gottfried during Nov meeting.)
Regards,
Albert
Reviriego, Pedro (Pedro) wrote:
>Albert,
>
>Have you tried using the receive filter presented in
"ungerboeck_1_1104"
>with your precoder? If so what kind of performance do you get?
>
>Regards,
>
>Pedro
>
>-----Original Message-----
>From: IEEE P802.3an Reflector [mailto:stds-802-3-10gbt@IEEE.ORG] On
>Behalf Of Albert Vareljian
>Sent: Saturday, December 11, 2004 2:14 AM
>To: STDS-802-3-10GBT@listserv.ieee.org
>Subject: [10GBT] THP Fixed Set TF
>
>Dear All,
>
>Further to our Nov meeting discussions on fixed THP TF set, pls find
>for evaluation attached material outlining 3 TF sets.
>This is aimed in combination with THP bypass option -- i.e.
>four THP modes in total.
>
>All simulations were performed in time domain as per "vareljian_1_1104"
>and showed reasonably good performance with relatively short (24-Tap
>and shorter) FFE. Performance impact due to 7-bit coefficient
>quantization effects was found to be very small.
>
>Similar results were observed in simulations for a 2nd order Tx LPF @
>200 MHz as per "ungerboeck_1_1104" in place of the 3rd order Tx LPF @
>400 MHz.
>
>Hope, the posting will stimulate discussion on our, otherwise
>relatively quiet, reflector.
>
>Happy Holidays to everyone.
>
>Regards,
>
>Albert
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>