while some standards (i22 802.3) specifies particular training pattern (prbs 11) others like pcix do not. Most modern devices are multi standard and the rx is robust enough and iare designed to be not predicated and dependent on particular training pattern. From: ibis-macro-bounce@xxxxxxxxxxxxx [ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Walter Katz [wkatz@xxxxxxxxxx] Sent: Tuesday, April 08, 2014 4:39 PM To: Dmitriev-Zdorov, Vladimir; ibis-macro@xxxxxxxxxxxxx Subject: [ibis-macro] Re: Back-channel BIRD discussion Vladimir, First, the Tx Init must be called to get the correct amplitude of the Impulse Response (Tx Peak to Peak Voltage), and some high frequency affects (details on the shape of the eye). Tx Peak to Peak voltages are programmable in each standard and can range by as much as 50%. Second, it depends on what you mean by “Rx training”. If you want the Rx training to emulate the standard (and the silicon implementation of the standard), then the Rx would need to know the PRBS pattern to use. If the IC vendor chose to use Rx Init to implement this training pattern and “time domain simulation”, then it would need that pattern hard coded in the DLL, or it would need to be specified in a Usage In parameter. But it is also true that training can be done with any reasonable input. If the Tx has 3 FFE taps, and the Rx has 4 DFE taps, then any PRBS7 would probably be reasonable enough. (Note that when I describe an Rx as having 4 DFE taps, I am really saying that it can optimize internally 4UI of the impulse response, not that it represent DFE equalization specifically. Walter From: Dmitriev-Zdorov, Vladimir [mailto:vladimir_dmitriev-zdorov@xxxxxxxxxx] Sent: Tuesday, April 08, 2014 4:22 PM To: ibis-macro@xxxxxxxxxxxxx Cc: wkatz@xxxxxxxxxx Subject: Back-channel BIRD discussion Hi Walter, A couple comments/questions about the presentation. I agree that Rx optimization (e.g. made in INIT) can do almost everything since Rx knows the channel’s response. If Rx were to know how to set Tx equalization, it could find the optimal tap coefficients and send them to Tx. This statement remains true regardless of the backchannel support. Therefore there is a question. Why in our regular flow we call Tx INIT prior to Rx INIT? Maybe we should allow Rx INIT be first? Then, it can update impulse response accordingly. If we have a mechanism to pass tap values from Rx to Tx, that’s fine. If we don’t, then Tx can receive the impulse response modified by Rx optimization and do the best it can. Second is about making BCI file a property of Rx, a tempting idea, considering that in the current BIRD BCI file should be read/understood by Tx, Rx and EDA tool. The last is required in the current spec to allow EDA tool to generate the training pattern, as required by the protocol. Making your proposal, do you mean that Rx training can be made on any reasonable input, not only by that specified by the back-channel communication protocol? Vladimir From: ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx> [mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Ambrish Varma Sent: Wednesday, April 02, 2014 6:56 AM To: ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx> Subject: [ibis-macro] Revised Backchannel BIRD Hi All, The latest version of the Backchannel BIRD (draft 8) has been uploaded on the ATM website and is available here: http://www.vhdl.org/ibis/macromodel_wip/archive/20140401/ambrishvarma/Backchannel%20support%20BIRD%20147%20draft%208/Backchannel_BIRD147_Draft8.docx Or http://tinyurl.com/o8cqyqr This is the version that was discussed at the 04/01/2014 ATM and includes support for AMI_Init/Statistical Backchannel training. Please let me know if there are any questions. Thanks, Ambrish. Ambrish Varma. Cadence Design Systems. --------------------------------------------------------------------- IBIS Macro website : http://www.eda.org/pub/ibis/macromodel_wip/ IBIS Macro reflector: //www.freelists.org/list/ibis-macro To unsubscribe send an email: To: ibis-macro-request@xxxxxxxxxxxxx Subject: unsubscribe