Received: from watserv1.uwaterloo.ca (watserv1.waterloo.edu) by karazm.math.UH.EDU with SMTP id AA18002
  (5.65c/IDA-1.4.4 for <glove-list@karazm.math.uh.edu>); Wed, 16 Oct 1991 16:38:29 -0500
Received: by watserv1.uwaterloo.ca
	id <AA22998>; Wed, 16 Oct 91 17:34:28 -0400
Date: Wed, 16 Oct 91 17:34:28 -0400
From: Dave Stampe-Psy+Eng <dstamp@watserv1.uwaterloo.ca>
Message-Id: <9110162134.AA22998@watserv1.uwaterloo.ca>
To: glove-list@karazm.math.uh.edu

I've been experimenting with the hi-res mode, and I want to compare my
results with others.

I have found a way to read the glove as fast as possible.  Simply read the 6
data bytes from the glove, then 2 more at the fast speed (discard these).
Now read a byte from the glove every 2 to 4 millisecond (with an interrupt?).
If the read returns $A0, read the next data set.  If 500 reads pass without
$A0, repeat the hires setup sequence.  This can run in the background while
graphics are drawn.

This results in speeds of 17 samples/sec (fingers open) to 12.5 (fist).
It appears that up to 25 samples/sec could be achieved by disabling finger
reads by the glove controller, and reading finger positions externally as
I suggested earlier.  (Lance Norskog suggested using an IBM PC's game port
as an analog input for this.  Sounds feasible, and could be made fast).

The data etc. read from the glove is in 6 bytes:
1) X (side-to side)
2) Y (up and down)
3) Z (distance)
4) rot (turning wrist)
5) fingers (2 bits for each finger)
6) keys on glove

For X and Y, the glove seems to return differences in distances to receiver
pairs from the transmitters.  This implies that the X and Y scaling to
position gain changes with position, being greatest closer to and centered
on the receiver array.	At 300 cm (10') the resolution is 3 mm per count.
Right and up are positive, left and down are negative.	Range is +/- 127,
which is achived at 45 degrees from and outside the receiver array (!).

Z resolution seems disappointingly low, being 1.6 cm (0.65") per step.
I KNOW that the internal resolution is 2 mm for this one! Away is positive,
toward is negative.

Rotation is also low resolution, and seems quite nonlinear.  This could
be caused by the room I'm in, though.  The values increase with clockwise
rotation, palm down being 0.  Steps are (about) 30 to 40 degrees, with a
range of 0 to 13.  Sensitivity seems greatest in palm-up position (6 to 9).

Finger positions are packed into 1 byte, with 2 bits per finger.  The
format is:
		 T t I i M m R r
for thumb, index, middle and ring fingers.  Open hand is 0, fist is 3.
Capital is MSB, lowercase is LSB.

The key byte can accept all number keys (0 to 9) and returns their number
codes (0-9).  If no key is pressed, this byte is $FF.
Other key codes are:

SEL	 $83
START	 $82
A	 $0A
B	 $0B
UP	 $0D
DOWN	 $0E
LEFT	 $0C
RIGHT	 $0F

Codes are returned as long as the key is held down.


NOTE ON NOISE:	I have found the glove position data tends to take
jumps of about 6 to 8 units in X and Y position (about an inch).
This is NOT a glitch or a misread, but a position-related discontinuity.

The reason for this can be seen by scoping the receiver detector inputs and
outputs (in the small box joining the receivers and the glove).  These are
peak detectors, and respond when the sound pulse has been received.

However, the glove uses piezoelectric transducers in its transmitters and
receivers, which have a VERY high Q.  This means that a 12-cycle pulse
driving the transmitter is stretched to several hundred pulses at the
receiver detector.  Also, the "rising edge" of the received envelope
increases linearly for at least 20 cycles before stablizing.

This means that 3 to 6 pulses actually arrive at the receivers before
the detector is tripped.  The detection point is always near the peak
of one of the 45 KHz wave cycles, and this adds a quantized offset
to the delay.  The power glove seems to use a delay resolution of
2 microseconds.

However, if the strength of the pulse at the receiver changes ( or
if a reflection of the pulse interferes) the detector will trigger on
an earlier or later cycle than it did on the last pulse.  This
causes a +/- 20 microsecond change in the delay time, or about
8 mm of distance.  Thus, the glove can resolve about 1.3 mm of
distance, but with errors of +/- 8 mm superimposed!
See this diagram for a possible result:
		__________
	       /	  \
	      / 	    \
	    /		      \
   ^	   |		       |  <-- GLITCH!
   |	   |		       |
   |	  /			\
   |  __ /			 \___
position			      time--->

This problem is built into the design, and can't be fixed.  If the user
moves the glove slowly enough, these glitches might be filterable
by software.  Have to try it, though.

-Dave Stampe