160 baud ALDL data stream format - reference URL?

[steve ravet]

IMO you guys are making this complicated.  The intent is to sample at a
certain time after the falling edge.  The spec has a range of possible
times to sample, but that's only because the code stream may vary
between ECMs.  Having more or fewer instructions between the one that
sets the falling edge and the one that sets the bit value would change
the time from t1 to t2.  Apparently that doesn't really happen according
to Ludis.  Just sample in the middle of the sample region and you're
good to go.

--steve

jsg at donet.com wrote:
> 
>    In message <B0B0BB04D8ABD111937800805FEA5B1D02123384 at exchou-prod1001.eng.hou
> .compaq.com>, you write:
> 
> <delete>
> 
> | should this be:
> |                S   0        1    S
> |            ..__     ........_____    ...
> |                |   ^        ^    |   ^
> |                V___|........|    V___|..
> |               T0   T2       T2  T3
> |
> |                 t2-t0 > t3-t2 => 0
> |                 t2-t0 < t3-t2 => 1
> 
> yes, t2 (the rising edge time) should be used in the equation.
> 
> | I would think that the transition from low to high would be where the
> | timestamp
> | should happen.  If it occurs early it is a zero and if it occurs late it is
> | a one.
> 
> It's the ratio of high time versus low time that is important.
> Without going through the math, I believe an optimal Bayes detector
> for this problem would integrate the "1" and "0" time and compare the
> ratio to a fixed value (a threshold based on expected duty
> cycle... assumed 1 in this case). The above differences are an
> implementation of the integration given transition times. Other
> solutions would not be optimal in this sense.
> 
> john gwynne

-- 
Steve Ravet
steve.ravet at arm.com
Advanced Risc Machines, Inc.
www.arm.com