Re: [BP] signal to noise enery

[Charles Moore <charles.moore@xxxxxxxxxxxxx>]

steve,

      Here are the slides.  As adam suggests the details need cleanup so 
just
pay attention to the overall idea.

                           charles

Steve Anderson wrote:

>	Charles,
>
>	Are the slides that Richard M. is talking about (from January
>12)
>posted somewhere?
>
>	Steve A.
>
>
>-----Original Message-----
>From: Charles Moore [mailto:charles.moore@avagotech.com] 
>Sent: Wednesday, January 18, 2006 5:00 PM
>To: Healey, Adam B (Adam)
>Cc: Mellitz, Richard; Spagna, Fulvio; Radhakrishnan, Prakash K
>Subject: Re: signal to noise enery
>
>adam, et al,
>
>I can't argue with points 1-3, those issues need fixing.
>
>Point 4 is really an invitation do discussion and i will take it:
>
>Clearly rich intended that this or something else very similar "is the 
>way to go"
>or he would not have presented the slides. I for one would like to go 
>this way at
>least in the sense that i want to see crosstalk should be measured as a 
>power (or
>energy) integral and a ratio of crosstalk to some measure of signal.
>
>I do not see the crosstalk integral as very contoversial, clean up 
>rich's, or use one
>from healey_c1_0505, or even one from one of my presentations, they all 
>will give
>close to the same result.
>
>The signal part is harder. A few suggestions, starting with the more 
>difficult but
>most accurate and working to easier:
>
>1. Simulate (with suitable Tx equalization) pattern1= 11111000 and
>pattern2=
>11110111, through the channel. Look at difference between the 2
>patterns at the Rx end, and find the positive peak of
>(response1-response2).
>This is the difference between the response to a 1 and to a 0 (with
>maximum
>precursor interference) and is what the Rx needs to discriminate,
>regardless
>of how it does the discrimination. Call this difference (scaled by the
>input
>amplitude) the signal gain.
>
>2 Simulate a pulse going through the channel and call the peak of the 
>output
>the signal gain.
>
>3. Integrate sinc(f/10.3125 GHz)/10.3125 GHz * |Sdd21| (which will give
>an
>upper bound on 2.) and call it signal gain.
>
>4. Evaluate -A(f) at some fixed frequency and call it signal gain.
>
>These are all amplitude rather than power numbers so either the signal 
>gain will have
>to be squared or the square root taken of the crosstalk power.
>
>charles
>
>
>Healey, Adam B (Adam) wrote:
>
>  
>
>>One more thing, from the practical standpoint, data is measured from 
>>fmin to fmax (nominally 50 to 15000 MHz). Do you propose extrapolation
>>    
>>
>
>  
>
>>to DC or to simply ignore missing data points between DC and fmin 
>>(assume they are zero?).
>>
>>    -----Original Message-----
>>    *From:* Healey, Adam B (Adam)
>>    *Sent:* Wednesday, January 18, 2006 4:10 PM
>>    *To:* 'Mellitz, Richard'; charles.moore@avagotech.com
>>    *Cc:* Spagna, Fulvio; Radhakrishnan, Prakash K
>>    *Subject:* RE: signal to noise enery
>>
>>    Rich et al.,
>>    I have looked at the proposed signal-energy calculations and I
>>    have following questions/comments...
>>    1. You have taken the square-root of the energy terms but then
>>    converted them to decibels with a factor of 10. I think you need
>>    to pick one or the other (e.g. 20*log10( sqrt( E_sig/E_xtk ) ) or
>>    10*log10( E_sig/E_xtlk ).
>>    2. Maybe I'm too much of a mathematician for my own good, but I am
>>    troubled by the use of a "PDF" in the frequency-domain integral. I
>>    think you are really looking for the power spectral density (PSD)
>>    which already has the appropriate units and does not need to be
>>    squared. For that matter, you really want to square the
>>    _magnitude_ of s21, and that is not clearly shown. You may also
>>    clarify the limits on your integrals and you can dispose of the
>>    1/(2*pi) if you just elect to integrate over f rather than omega,
>>    3. What PSD do you intend to use? Will it be a basic sinc( )
>>    function or will you add roll-off to represent transmit and
>>    receiver bandlimiting. What amplitude will you assume?
>>    4. I don't see anything in these slides indicating that "this is
>>    way to go". Please note that this methodology is basically an
>>    extension of what I proposed in May (refer to
>>    http://ieee802.org/3/ap/public/channel_adhoc/healey_c1_0505.pdf).
>>    You are essentially proposing that we also integrate the signal
>>    energy and consider the ratio, rather than just examine the
>>    crosstalk in isolation. What does this ratiometric view add in
>>    terms of value, and how does it relate the expected equalization
>>    capability and noise enhancement properties of actual devices?
>>    Thanks,
>>    -Adam
>>
>>        -----Original Message-----
>>        *From:* Mellitz, Richard [mailto:richard.mellitz@intel.com]
>>        *Sent:* Thursday, January 12, 2006 9:24 AM
>>        *To:* Mellitz, Richard; charles.moore@avagotech.com
>>        *Cc:* Healey, Adam B (Adam); Spagna, Fulvio; Radhakrishnan,
>>        Prakash K
>>        *Subject:* RE: signal to noise enery
>>
>>        Put pdf inside of integral.
>>
>>        ... R
>>
>>
>>    
>>
>------------------------------------------------------------------------
>  
>
>>        *From:* Mellitz, Richard
>>        *Sent:* Thursday, January 12, 2006 9:11 AM
>>        *To:* 'charles.moore@avagotech.com'
>>        *Cc:* 'Healey, Adam B (Adam)'; Spagna, Fulvio; Radhakrishnan,
>>        Prakash K
>>        *Subject:* signal to noise enery
>>
>>        Here's a quick set of slides that suggest energy may be the
>>        way to go. I think this is the gist of where we were going
>>        last night. We will need data for all the channel to support
>>        this for the Feb meeting.
>>
>>        ...Rich
>>
>>    
>>
>
>
>  
>


-- 
|--------------------------------------------------------------------|
|       Charles Moore 
|       Avago Technologies
|       Image Solutions Division
|       charles.moore@avagotech.com
|       (970) 288-4561
|--------------------------------------------------------------------|

mellitz_01_0106.pdf