Re: idt reset issue

From: Peter Liebmann <peter_liebmann_at_.....>
Date: Thu Dec 07 2006 - 17:42:39 PST
Ken,

I did this work a number of years ago.  If I remember correctly, I 
checked the reset value after convergence.  I then reset if on, and 
continued to the next time point (I used "cross" so reset was set "on" 
at the present step then off at the next step).  I think this fits into 
the rising edge model.
I'm not sure if this messes up other applications, so use your judgment.

Peter Liebmann



Ken Kundert wrote:
> Peter,
>     You bring up a very important point. Allow me to reframe your point 
> into the following question:
> Should assert be level sensitive or edge triggered?
> 
> Either is valid, but there are some important consequences to the choice.
> 
> Assume it is level sensitive. Then you can model the idt function as 
> follows:
> 
> 
> 
> In this case,
>     V(out) <+ idt(0, V(in), 1);
> is a valid implementation of a unity gain buffer. As a result, the 
> implementation must include the path from the assert input to the output 
> in the Jacobian when assert is high. Since the Jacobian includes all the 
> right terms the basic issue with convergence is avoided, but this 
> approach may be result in surprising behavior when /ic/ is dependent on 
> the /out/. Consider your example. If you make the /ic/ = -V(/out/), then 
> when /assert/ goes high the output will immediately jump to 0 because 
> that is the only value of V(/out/) with satisfies V(/out/)// = -V(/out/).
> 
> Now consider the case where it is edge triggered. In this case the idt 
> function keeps integrating as /assert /goes high. When /assert/ then 
> goes low, say at/ t/_a ,
> 
> where /y/ is the output, /x/ is the input, and
>  
> meaning that the /ic/ input is sampled at/ t/_a and /c/ becomes that 
> value for all time beyond (but not at) /t/_a .
> 
> I would expect that this would be the preferred implementation. If so, I 
> can add this description to the changes I proposed yesterday.
> 
> -Ken
> 
> 
> Peter Liebmann wrote:
>> I came across a problem (the bouncing ball) which works in VHDL_ams.  It
>> is related to the example given by Ken for reset.  One has the case
>> similar to:
>>
>>      reset=0;
>>      @(timer(1,1))
>>         reset=1;
>>      V(out) <+ idt(1,-V(out), reset);
>>
>> V, in this case, behaves like a velocity which reverses when a ball hits
>> the ground.
>> If, at convergence, we reverse V and continue with NR iterations, we
>> will never converge.  I wasn't sure if Ken's proposal meant to continue
>> with the NR after convergence.
>> NOTE: this situation will converge if, at convergence, we reset and then
>> go to the next step, or reset, freeze the reset value and then 
>> continue with the NR.
>>
>> Peter Liebmann
>>
>> Ken Kundert wrote:
>>> Sri,
>>>     Yes, I think I can make it. I have created a proposal and Geoffrey
>>> posted it on the website:
>>> http://www.eda-stds.org/verilog-ams/htmlpages/public-docs/idt-proposal.pdf 
>>>
>>>
>>> -Ken
>>>
>>> Sri Chandra wrote:
>>>> Hi Ken,
>>>>
>>>> Would you be available 9pm Pacific this thursday (Dec 7th) to discuss
>>>> the idt proposal along with the proposed changes. If you can send the
>>>> proposal across before the meeting that will be good.
>>>>
>>>> cheers,
>>>> Sri
>>>>
>>>> Sri Chandra wrote:
>>>>> Ken,
>>>>>
>>>>> By the way just thought will let you know that the timings of the
>>>>> meeting has changed (from this week onwards). Now-a-days the meetings
>>>>> are on thursday evenings at 9pm pacific.
>>>>>
>>>>> Please go ahead and send out a proposed change to the committee based
>>>>> on the feedback that you received on the reflector. We can review that
>>>>> as part of the meeting. Since idt is a very key item, and a fairly
>>>>> major feature in Verilog-A quiet widely used, I would be very
>>>>> uncomfortable changing the LRM without actually having a live
>>>>> discussion in the call. So hopefully you would be able to attend one
>>>>> of the upcoming meetings at this new time and we can discuss the new
>>>>> document that you plan to post.
>>>>>
>>>>> Hope that is agreeable.
>>>>>
>>>>> Regards,
>>>>> Sri
>>>>>
>>>>> cheers
>>>>>
>>>>> Ken Kundert wrote:
>>>>>> Sri,
>>>>>>     I have an on-going conflict that prevents me from calling into 
>>>>>> the
>>>>>> meetings. I have published a description of the problem and the 
>>>>>> desired
>>>>>> behavior. There has been some discussion on the reflector and 
>>>>>> nobody has
>>>>>> presented any objections. Perhaps I should just send out a proposed
>>>>>> change to the LRM and see what everyone thinks?
>>>>>>
>>>>>> -Ken
>>>>>>
>>>>>> Sri Chandra wrote:
>>>>>>> Ken,
>>>>>>>
>>>>>>> I haven't been able to attend the recent calls but I hope to be back
>>>>>>> online from this week onwards. I dont think this issue has been
>>>>>>> discused
>>>>>>> in the recent meetings. The committee has been reviewing independent
>>>>>>> chapters and we are currently in the process of reviewing chapter 7
>>>>>>> being edited by Marq Kole.
>>>>>>>
>>>>>>> The last I remember in the reflector was you were planning to 
>>>>>>> present
>>>>>>> this item to go over it to the committee but you were unable to 
>>>>>>> attend
>>>>>>> the meeting. If you are available and if you can present the
>>>>>>> proposal at
>>>>>>> one of the the committee meetings that will be great and we can
>>>>>>> schedule
>>>>>>> it in one of the upcoming calls.
>>>>>>>
>>>>>>> Regards,
>>>>>>> Sri
>>>>>>>
>>>>>>> Ken Kundert wrote:
>>>>>>>> Sri,
>>>>>>>>      Was there any decision made on the idt issue? Would you like
>>>>>>>> me to
>>>>>>>> a cut at refining the description of idt in the LRM to avoid the
>>>>>>>> ambiguity in its behavior?
>>>>>>>>
>>>>>>>> -Ken
>>>>>>>>
>>>>>>>> Ken Kundert wrote:
>>>>>>>>> All,
>>>>>>>>>    I apologize for missing the call this morning. It turns out 
>>>>>>>>> that
>>>>>>>>> Thursday mornings are just too busy for me to attend.
>>>>>>>>>
>>>>>>>>> I have updated the document to include an model that patterns the
>>>>>>>>> desired behavior. You can find the updated version at
>>>>>>>>> http://designers-guide.org/private/vams-extensions/idt-issue.pdf
>>>>>>>>>
>>>>>>>>> Also, I would like to offer the use the my online forum for use by
>>>>>>>>> the
>>>>>>>>> Verilog-AMS committee. We used it when defining the compact model
>>>>>>>>> extensions and I found it to be a very convenient way to carrying
>>>>>>>>> on the
>>>>>>>>> conversations about particular issues. It naturally separates the
>>>>>>>>> discussion threads and makes them easy to follow. If you wanted 
>>>>>>>>> to do
>>>>>>>>> this, I would give you a private board, so only invitees would be
>>>>>>>>> allowed to see the board or contribute.
>>>>>>>>>
>>>>>>>>> -Ken
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Geoffrey.Coram wrote:
>>>>>>>>>> Resending for Ken Kundert; original message bounced (too long).
>>>>>>>>>> Attachment has been saved as
>>>>>>>>>> http://www.verilog.org/verilog-ams/htmlpages/public-docs/idt-issue.pdf 
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> ----------------- Original Message -------------
>>>>>>>>>> All,
>>>>>>>>>>     I'd like to join the meeting tomorrow and discuss the reset
>>>>>>>>>> feature
>>>>>>>>>> of the idt function. I have not had much luck using this feature
>>>>>>>>>> through
>>>>>>>>>> the years, and recently had a situation where I really needed it.
>>>>>>>>>> Unfortunately, I found the Cadence implementation unsuitable once
>>>>>>>>>> again,
>>>>>>>>>> and when I dug in to it I found the LRM silent on critical
>>>>>>>>>> aspects of
>>>>>>>>>> this feature. I have attached a very short document that 
>>>>>>>>>> illustrates
>>>>>>>>>> the
>>>>>>>>>> issue and proposes what I believe to be the desirable behavior.
>>>>>>>>>> If you
>>>>>>>>>> all agree I will work on coming up with the needed 
>>>>>>>>>> modifications to
>>>>>>>>>> the LRM.
>>>>>>>>>>
>>>>>>>>>> -Ken
>>>
>>
Received on Thu Dec 7 17:43:10 2006

This archive was generated by hypermail 2.1.8 : Thu Dec 07 2006 - 17:43:23 PST