::: Bug in 2105 DSP serial port :::

  We are experiencing failure of the serial receive port on the 210x DSP used
in SPACE.  The failure mode is that after successfully running for a while, the
DSP suddenly stops generating RCV interrupts even though the signals on the inputs
seem to be fine and internal control registers seem to be correctly set up.

  I spoke with Dan Ash (the A/D field app engineer) and he confirmed that other
customers have reported similar failures.  After a failure occurs, there is a way
for the DSP to reset its internal state to reset the port (assuming that the DSP
detects the failure with a watchdog).  Also, he recommended a filter circuit to
minimize the glitches that create the problem.

  The hardware should have an RC filter with a 68pF cap. and a 100 Ohm resistor
on the RCV SCLK line of the DSP.

  The software needs to reset the serial ports by disabling/re-enabling them when
the failure occurs.  It might be possible to prevent the error by resetting the
serial port whenever the DSP switches from INTERNAL SERIAL CLOCK to EXTERNAL SERIAL
CLOCK (which it does frequently).

  I am currently upgrading the software from STEEL TALONS to implement the new
resets;  we probably should investigate the RC filter for inclusion on any
new boards.

Peter Lipson's VAX Mail "210X DSP serial failures" implies that there is a
newly discovered bug in the 2105.

I talked to Dan Ash and it appears the "bug" in question is really that a
glitch on the Serial Clock input that is shorter than the minimum serial
clock specified in the data sheet can cause the state machine that performs
the serial operations to get screwed up so that it needs to be reset.

The Party Line Bus used in Steel Talons and in the Police Trainer allow the
bus to float after one unit has released it and before another unit
takes it.

The work we did on the Police Trainer System showed that when the bus is
floating, the output of the Bus Receivers will be one or zero, depending
on its own input offsets. (The receivers are actually op-amps in disguise.)   

This led to a real problem on the Frame Sync signal, which we fixed by 
introducing just enough bias so that a floating bus would be pulled to the
correct inactive polarity. 

It appears that something like this should be done with the Serial Clock.

Note: 1. The way this is done on a terminated bus is different from the way
         it is done on a non-terminated bus.

      2. I have not seen the SPACE circuit. It may or may not be the same as
         in Steel Talons.

      3. The 'filter' Dan recommended is about 9 ns and may or may not help
         glitches caused by the switching between Bus Masters.


 Jed