5 Replies Latest reply on Jul 26, 2011 7:52 AM by weiming.qian

    Sythesized Tap values swapped in FaseEye Channel Analyzer?

    weiming.qian

      I am running FastEye Channel Analyer, the resulted channel pulse response is shown below. I denote two pre-cusors (X-2, X-1), main cursor (X0) and two post-cursors (X1, X2) on the curve. The value is the difference between the corresponding point and the baseline at -480mV.

      FIR_coefficient.jpg

       

      When I set Tx pre-emphasis Taps to 3, the synthesized optimal values is following:

      FIR_coefficient1.PNG

       

      When I set Tx pre-emphasis Taps to 5, the synthesized optimal values is following:

      FIR_coefficient2.PNG

      It seems to me that somehow the Tool got confused about Pre-Taps with Post-Taps and swapped them. My understanding is that, to first order, Pre-Taps are used to reduce/minimize the pre-cursor of the channel pulse response, and Post-Taps are used to reduce/minimize the post-cursor. The channel pulse response got a much bigger post-cusor than pre-cursor, so my naive thinking is the Post-Tap should have a much bigger absolute value than that of Pre-Tap.

       

      In 3-Tap case, using ZFS (Zero Forcing Solution) method with following matrix,

       

      [y(-1)]      [X0 X-1  X-2]   [C-1]

      [ y(0)]  =  [X1  X0  X-1] * [C0]

      [Y(1)]       [X2  X1  X0]    [C1]

       

      Setting y(-1)=y(1)=0 and C0=1, we got C-1 = -0.11, C1=-0.4635. Compare these two values to those synthesized in 3-Tap case, they are close but swapped.

       

      In 5-Tap case, the tool synthesized result gives 3 Pre-Taps and 1 Post_Tap, whereas the channel pulse response got a much longer post-cursor than pre-cursor. So this also suggests the Tool swapped Pre-Tap with Post-Tap.

       

      To verify my suspicion, I set the Xilinx GTX Tx pre-emphasis settings (3-tap) according to the tool synthesized Tap values, the resulted eye is closed. If I set Xilinx GTX Tx pre-emphasis settings using swapped synthesized Tap values or my own calculations (C-1, C1 above), the resulted eye is open.

       

      So my real questions are: a) Are those tool sysntesized Tap values just swapped or completely wrong? b) Am I wrong in above analysis?

       

       

      My HyperLynx version is:

       

      LineSim v8.1.1

      [V8.1 build 510(510,0)]

      Jan 20 2011

       

       

      Regards,

      Weiming

        • 1. Re: Sythesized Tap values swapped in FaseEye Channel Analyzer?
          david.royle

          Weiming,

          I noticed this as well.

          I was using an Altera tool for estimating best tap coefficients (PELE) and comparing these results with fast eye and noticed the swap.

          Hopefully Hyperlynx can confirm.

           

          By the way if you are using AMI models, I believe the the fast eye channel analyzer will not include it.

          Dave

          • 2. Re: Sythesized Tap values swapped in FaseEye Channel Analyzer?
            weiming.qian

            Thanks for confirming this problem.

             

            Acctually, what I was trying to do is to run FastEye on a channel including Xilinx V-6 GTX IBIS-AMI models to get optimal Tap coefficients and then feed those settings into IBIS-AMI model to run IBIS-AMI channel analysis flow. In the FastEye analysis, it ignores AMI models and just uses IBIS models for analog front-end. I compared the channel pulse responses generated in both FastEye flow  and IBIS-AMI flow, and they are the same. This means my above method is valid.

             

            Regards,

            Weiming

            • 3. Re: Sythesized Tap values swapped in FaseEye Channel Analyzer?
              david.royle

              Weiming,

              I am finding that the channel pulse responses from FastEye and IBIS-AMI appear to be doing the same thing, characterizing the passive channel only. My receiving device has an analog front end amplifier with some peaking at 3GHz, and the opportunity to add more peaking using equalization controls. This is all within the AMI code and will not effect either pulse response or FastEye tap selection.

              That said, I am finding the FastEye tap selection quite useful. I am not able to invoke RX equalization at the datarate that I am running, and the FastEye selected taps give me a nicely opened eye.

              Dave

              • 4. Re: Sythesized Tap values swapped in FaseEye Channel Analyzer?
                weston_beal

                I checked with the developer of this code and found that the indexing of the taps is reversed compared to many other notations. If you use the optimized taps within HyperLynx, the index order is consistent, but if you use the HyperLynx taps in another tool, or taps from another tool in HyperLynx, then you need to reverse the order of the taps. HyperLynx 8.2 will reverse this order to be consistent with common notation.

                 

                On the point of channel characterization, the simulation takes into account the driver and receiver IBIS models. It is a bit more than just the passive portion of the channel. The problem is that some IBIS AMI models move even the analog (traditional IBIS) effects into the compiled AMI model. In this case, the channel characterization cannot include all the effects of the driver and receiver.

                • 5. Re: Sythesized Tap values swapped in FaseEye Channel Analyzer?
                  weiming.qian

                  Following figure is from Linesim help file:

                  FIR_diagram.PNG

                  So the tap indexing is reversed compared to its own menu, not just many other notations.