FEEDFORWARD,ENOUGH-RKC

From Bob Clark (931209.1430 EST)

Recent postings from Hans Blom (931207.01:59) and Bill Powers
(931207.20:55) are getting a bit contentious.

To me, there seem to be two quite different viewpoints, each of which
is relevant to the viewer, but becomes contradictory to the other.

Rather than trying to discuss either viewpoint, I'd like to suggest a
real, and relevant, problem to which the term (and concept) of
FeedForward can be applied.

Consider the case of an incoming ballistic missile. The question is
one of defense. Deeper bomb shelters are of limited use, and running
in some other direction isn't much better. Interception in flight is
a very attractive suggestion.

From a theoretical engineering standpoint, the form of the solution

is obvious. From a small amount of initial observations, the
incoming trajectory can be estimated. Assuming that the defense is a
ballistic missile, its trajectory is known. Aiming and timing for
interception is readily computed. It may or may not be successful,
but the concepts and methods are straightforward.

For a hardware solution, it is assumed that the unit has the ability
to measure the early performance of the incoming missile. Aiming and
timing parameters are needed. There are three ways they can be
obtained. 1) They can be computed in real time and adjusted as more
data are acquired. 2) They can be computed ahead of time and
tabulated. The table can be entered on the basis of the initial
observations. 3) Recordings of actual test trajectories could be
compared with the observed initial portions of the trajectories.

The problem, here, is not which is "better," nor how to complete the
engineering, design and construction.

Here the problem is how to describe the sequence of events.

For this kind of problem, time is used as the independent variable.
Thus:

t = t(0) observation begins
t = t(1) analysis begins -- whatever method of analysis is used
t = t(2) analysis complete
t = t(3) aiming complete, including specification of firing time
t = t(4) missile fired
t = t(5) impact!
t = t(6) champagne!

All of this sequence occurs in real, current time, but not in a
single instant. It is necessary to allow for temporal variables.

It is possible to describe this in greatly condensed terms such as,
"shooting down a ballistic missile. The perceptual variable to be
controlled is the probability of impact by the defense missile. This
is a statistical matter, with a reference level set by a higher level
of control. The commander's viewpoint.

It can be described in terms of the hardware required. The job shop
viewpoint.

It can be described in terms of the data required at each of the
steps in the analysis. The computer programmer's viewpoint.

It can be described in terms of the details of the sequence of
events, which parts move, in what manner, which sub-programs do what
in which sequence and so on. The system analyst's viewpoint.

It can be described in terms of a single, complex control system with
internal connections unusual for a basic simple feedback system.
Here is where various ad hoc connections, feedback, feedforward,
integrating, delaying, etc are introduced. A more detailed system
analyst's viewpoint.

It can be described in terms of a multi-level combination of feedback
control systems. This is the HPCT viewpoint

I think no one would assert that any of these descriptions in
incorrect. However, some may be better for the engineer building the
unit. Others, perhaps, better for the commander. Still others,
perhaps, better for those interested in general properties and
descriptions of negative feedback control systems.

ยทยทยท

=====================

The disagreements begin when applications are made to human
performance. There are many situations of this form:

the skidding car
the walking to the bed in the dark
the baseball fielder
the skeet shooter
the duck hunter
the fox chasing the rabbit
the football pass receiver
the boxer and the punch
the tennis player
etc etc

======================

The key concept here is "anticipation." The concept of "lead time"
seems especially relevant. In addition, humans lack high speed
computation ability, and rarely are equipped with suitable tables of
measured data. However they generally are equipped with quite
remarkable memory capability, used to estimate the future location of
moving targets. These terms and concepts are very familiar and
communicate effectively. They are hard to apply to engineering
designs, but seem to be very useful for human purposes.

I hope this rather detailed presentation will help resolve the
developing conflict. To some extent, this might be considered an
attempt to "move up a level."

Regards, Bob Clark