<DD7A9A95166BF4418C4C1EB2033B6EE20227F23D@xxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <5c023dc83727c256d78a8ba0181a9291@xxxxxxxxxxx> X-Sender: ckumar@xxxxxxxxxxx User-Agent: RoundCube Webmail/0.3.1 Content-Transfer-Encoding: 8bit Content-Type: text/plain; charset=UTF-8 I believe Fangyi is pointing out cases where modified (impulse) response exists, However it is not the result of convolution with a filter The larger more issue still stands. i.e the potential of getting two different results for the same model On Wed, 21 Apr 2010 20:20:42 -0700, "Muranyi, Arpad" <Arpad_Muranyi@xxxxxxxxxx> wrote: > Fangyi, > > I am sorry but I don't quite understand the second half of your > 1st sentence (after the word "because") because there is some > problem with English in it and I am not sure what you mean. > Please clarify it. > > Regarding the 2nd sentence about assuming that Rx Init modifies > the input, I am not sure how that makes a difference, since the > convolution #2 box seems to be there to do that modification in > case the Init function didn't do it. (This is also true for > the Tx Init function and the convolution #1 box). > > The question that we didn't seem to address in the November > flow is what Init will return when Init_Returns_Impulse is False. > This is not shown on the drawings and I can't find anything in > the current spec that says what is returned in that case. > > If it returns what goes into it, then this setting may be a duplicate > of the Use_Init_Output = False setting, at least as section 2.3 > describes that in the current spec, unless we meant to distinguish > how the Init Output is passed to the GetWave functions vs. the > next Init function, but this is not spelled out in the spec > either as far as I can tell. > > Or does it mean that it should return nothing (i.e. uninitialized > memory, or noise)? Or should it mean that it should return the > filter only? (In that case we do not need the new Init_Returns_Filter > Boolean). This must be defined because it fells like we have another > loose end dangling here. > > Regarding de-convolution, I thought the reason we added the > Init_Returns_Filter Boolean was to eliminate the need for > de-convolution. As far as I can tell, according to the November > flow de-convolution would only be needed with older models > which do not have the ability to return filter only to the > EDA tool. This would hopefully die out as new models come > into existence and old models fade away. > > Thanks, > > Arpad > ===================================================================== > > -----Original Message----- > From: fangyi_rao@xxxxxxxxxxx [mailto:fangyi_rao@xxxxxxxxxxx] > Sent: Wednesday, April 21, 2010 9:36 PM > To: Muranyi, Arpad; ibis-macro@xxxxxxxxxxxxx > Subject: RE: [ibis-macro] Re: AMI_Flows_6.pdf for today's ATM > teleconference > > Hi, Arpad; > > Moving the Selector to the right side does not save time-domain > simulation because it makes the modified impulse returned by Rx Init > based on the modified impulse returned by Tx Init and Tx Use_Init_Output > is False. Moreover, you can't assume that Rx Init modifies the input > impulse by convolving it with a filter. > > In my opinion (I pointed out last year) the November flow does not > really address this issue. It works only if above assumption is true. In > addition, de-convolution is still required in certain cases in the > November flow. > > Regards, > Fangyi > > -----Original Message----- > From: ibis-macro-bounce@xxxxxxxxxxxxx > [mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad > Sent: Wednesday, April 21, 2010 7:08 PM > To: ibis-macro@xxxxxxxxxxxxx > Subject: [ibis-macro] Re: AMI_Flows_6.pdf for today's ATM teleconference > > Fangyi, > > Thanks for your comments. I agree with your observations. > > In fact this is why Walter was right when he wrote in his > recent comment that the Tx Init Output Selector box on my > latest drawing should be on the right side instead of the > left side. That way both statistical and time domain > simulations would be correct without having to sacrifice > either one of them over the other. > > The problem is that making that change would deviate from the > flow in more than just changing the order of steps 4 and 5 as > described in the existing specification in section 2.3 and it > would also bring us back to the issue of having to do a > deconvolution for certain configurations, which is why it > was suggested to me to move that selector to the left side a > week ago. I am beginning to see why Walter said that the > entire flow is flawed in the existing specification... > > The more I push and shove the blocks around in this flow diagram, > the more I feel that the November flow is the right way to go. It > seems that we have already figured it all out in the fall how to > do this right, and no matter how hard we try to be minimalistic > in fixing the spec, the best way to do it is as described in the > November flow. > > Thanks, > > Arpad > ==================================================================== > > > > -----Original Message----- > From: fangyi_rao@xxxxxxxxxxx [mailto:fangyi_rao@xxxxxxxxxxx] > Sent: Wednesday, April 21, 2010 8:32 PM > To: Muranyi, Arpad; ibis-macro@xxxxxxxxxxxxx > Subject: RE: [ibis-macro] Re: AMI_Flows_6.pdf for today's ATM > teleconference > > Hi, Arpad; > > In the current BIRD only section 2.3 defines the control of Tx > Use_Init_Output on input impulse response to the Rx Init call. The > excerpt quoted in Walter's email is actually for the control of input > waveform to the Rx GetWave call. Your flow presented in the last meeting > is consistent with section 2.3. > > A key difficulty arises from the condition when Tx has GetWave, Tx > Use_Init_Output is False and Rx Use_Init_Output is True. If the input > impulse to Rx Init is the original impulse input to Tx Init, then the > time-domain bit-by-bit simulation is fine but the statistical simulation > is broken. If the input impulse to Rx Init is the modified impulse > returned by Tx Init, then the statistical simulation is fine but the > time-domain simulation is broken. Given the fact that time-domain > simulation captures more effects of nonlinear models, it seems we should > give it higher priority over statistical simulation. > > Regards, > Fangyi > --------------------------------------------------------------------- > 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 > > --------------------------------------------------------------------- > 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 --------------------------------------------------------------------- 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