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

Re: [802.3_100GCU] Item 3: Port type. 802.3bj Consensus Building Group - Channel Specifications



1. The RX filter is 4th order (4 poles). It is very realistic that an analog front end will have a bandwidth limiting filter of order 4 or larger. Butterworth or Bessel filters are simple idealizations; implementations will vary, and we will need margin to cover that.

2. The bandwidth limiting is important also for noise calculations (crosstalk, thermal, EMI), not just for limiting the desired energy. Without some filtering, NEXT will be dominant in many channels.

3. For editorial purposes, there are compact ways of specifying filters, such as pole values (two complex pairs) or even "4th order Butterworth" which is a complete description. For Matlab code, I think shortness is not as important.



</Adee>

-----Original Message-----
From: Charles Moore [mailto:charles.moore@xxxxxxxxxxxxx] 
Sent: Wednesday, June 13, 2012 11:20 PM
To: STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GCU] Item 3: Port type. 802.3bj Consensus Building Group - Channel Specifications

rich,

What you say is absolutely true, but the complex form is (about) 4 times as long. Maybe i am too lazy.

charles

|--------------------------------------------------------------------|
|       Charles Moore 
|       Avago Technologies
|       APD
|       charles.moore@xxxxxxxxxxxxx
|       (970) 288-4561
|--------------------------------------------------------------------|



Mellitz, Richard wrote:
>
> This filter
>
> rr_filter=(1+bw1*1i*f/fc+bw2*(1i*f/fc).^2+bw3*(1i*f/fc).^3+(1i*f/fc).^
> 4).^-1;
>
> has the EXACT power spectrum as was used in moore_01_0311 (and clause 85).
>
> rr_filter_POWER= (1+ (1i*f/fc).^8).^-1;
>
> To me these are simple single line equations.
>
> ...Rich
>
> *From:* Charles Moore [mailto:charles.moore@xxxxxxxxxxxxx]
> *Sent:* Wednesday, June 13, 2012 3:48 PM
> *To:* Mellitz, Richard
> *Cc:* STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx
> *Subject:* Re: [802.3_100GCU] Item 3: Port type. 802.3bj Consensus 
> Building Group - Channel Specifications
>
> The 4 pole (Tx) and 8 pole 0 phase Butterworth filters in
> moore_01_0311 were used strictly for mathematical convenience and were 
> not intended to be particularly realistic. There is little enough 
> energy at the 3dB frequency of the filters in most channels so that we 
> (adam and i) felt that the lack of realism would not be significant.
> Adding phase pretty much does away with the mathematical simplicity 
> with little gain in realism: a 4 pole Butterworth filter has higher Q 
> poles than anyone is likely to implement on a chip.
>
> charles
>
> |--------------------------------------------------------------------|
> |       Charles Moore 
> |       Avago Technologies
> |       APD
> |       charles.moore@xxxxxxxxxxxxx <mailto:charles.moore@xxxxxxxxxxxxx>
> |       (970) 288-4561
> |--------------------------------------------------------------------|
>
>
>
> Mellitz, Richard wrote:
>
> A 4 pole Butteworth-Thompson minimum phase filter might be:
>
> >> f=10e6:10e6:20e9; bw1=2.613126; bw2=3.4142136; bw3=2.613126;
> fc=13e9*1.5;
>
> >> 
> rr_filter=(1+bw1*1i*f/fc+bw2*(1i*f/fc).^2+bw3*(1i*f/fc).^3+(1i*f/fc).^
> 4).^-1;
>
> >> subplot(2,1,1); plot(f,20*log10(abs(rr_filter))); subplot(2,1,2);
> plot(f,unwrap(angle(rr_filter)));
>
> Should we use something like this for the Rx filter?
>
> A 1 pole is minimum phase. So Tx filter should be OK unless we would 
> like to add more poles for some physical reason.
>
> ... Rich
>
> *From:* Ran, Adee [mailto:adee.ran@xxxxxxxxx]
> *Sent:* Sunday, June 10, 2012 2:50 PM
> *To:* STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx
> <mailto:STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx>
> *Subject:* Re: [802.3_100GCU] Item 3: Port type. 802.3bj Consensus 
> Building Group - Channel Specifications
>
> RX bandwidth filter and TX edge filter should naturally be lower for
> PAM4 - about half the bandwidth or NRZ - otherwise crosstalk 
> generation and sensitivity are unnecessarily high.
>
> These filters should include phase, not just magnitude as in 
> moore_01_0311. Phase is significant in the time domain. It is still 
> multiplication, but with complex values.
>
> For the RX filter phase and magnitude we can assume Butterworth 
> filter, which has a rather simple transfer function (see Wikipedia or 
> more professional sources) and is an approximation for realizable filters.
>
> As for the TX filter, can anyone suggest which filter is adequate? Is 
> there a justification for 2^nd order and are the poles real or complex 
> conjugate?
>
> </Adee>
>
> *From:* Mellitz, Richard [mailto:richard.mellitz@xxxxxxxxx]
> <mailto:%5Bmailto:richard.mellitz@xxxxxxxxx%5D>
> *Sent:* Friday, June 08, 2012 5:21 PM
> *To:* STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx
> <mailto:STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx>
> *Subject:* Re: [802.3_100GCU] Item 3: Port type. 802.3bj Consensus 
> Building Group - Channel Specifications
>
> Should filters be different per port type?
>
> ...Rich
>
> *From:* Mellitz, Richard [mailto:richard.mellitz@xxxxxxxxx]
> *Sent:* Friday, June 08, 2012 10:19 AM
> *To:* STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx
> <mailto:STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx>
> *Subject:* Re: [802.3_100GCU] Item 2: preprocessing s-parameters. 
> 802.3bj Consensus Building Group - Channel Specifications
>
> Perhaps a convenient way to look at TX/RX filtering is convolution of 
> the following.
>
> Tx Voltage Amplitude scaling (for NEXT, FEXT, and THRU not coding 
> which will be done in a later algorithm )
>
> Tx edge filter (moore_01_0311)
>
> Rx bandwidth filter (moore_01_0311)
>
> Tx/Rx block return loss filters (gamma,moore_01_0311)
>
> The first three are a straight convolution and with is multiplication 
> in the frequency domain.
>
> The lasts is a chain matrix convolution of RL under worst +1 or -1 
> reflection coefficient phase conditions and in the frequency domain 
> may look like this
>
> sdd21=sdd21.*(1.-gamma_rx)./(1.- sdd11.*gamma_tx + sdd22.*gamma_rx
> +sdd21.^2.*gamma_tx.*gamma_rx -sdd11.*sdd22.*gamma_tx.*gamma_rx);
>
> ...Rich
>