Table of Contents
Advertisement
In cycle 2.0, the first element of the rule list (a 16-bit address) is read from memory.
Due to propagation delays, this value cannot be used for calculations until two cycles
later (cycle 4.0). The X index register, which is used to access information from the
rule list, is incremented by two to point at the next element of the rule list.
The operations performed in cycle 4.0 depend on the value of the word read from the
rule list. $FFFE is a special token that indicates a transition from antecedents to con-
sequents, or from consequents to antecedents of a new rule. The V bit can be used to
decide which transition is taking place, and V is toggled each time the $FFFE token is
detected. If V was zero, a change from antecedents to consequents is taking place,
and it is time to apply weighting (provided it is enabled by the C bit equal one). The
address in TMP2 (derived from Y) is used to read the weight byte from memory. In this
case, there is no bus access in cycle 5.0, but the index into the rule list is updated to
point to the next rule element.
The old value of X (X
cess a rule word in cycle 7.2. The read of the rule word is timed to start two cycles be-
fore it will be used in cycle 4.0 of the next loop pass. The actual multiply takes place
in cycles 6.2 through 8.2. The 8-bit weight from memory is incremented (possibly over-
flowing to $100) before the multiply, and the upper eight bits of the 16-bit internal result
is used as the weighted result. By using weight+1, the result can range from 0.0 times
A to 1.0 times A. After 8.2, flow continues to the next loop pass at cycle 4.0.
At cycle 4.0, if R
x
ents of a new rule is taking place, so accumulator A must be reinitialized to $FF. During
processing of rule antecedents, A is updated with the smaller of A, or the current fuzzy
input (cycle 6.0). Cycle 5.0 is usually used to read the next rule word and update the
pointer in X. This read is skipped if the current R
is a weight multiply pass, the read is delayed until cycle 7.2. During processing of con-
sequents, cycle 6.1 is used to optionally update a fuzzy output if the value in accumu-
lator A is larger.
After all rules have been processed, cycle 7.0 is used to update the PC to point at the
next instruction. If weights were enabled, Y is updated to point at the location that im-
mediately follows the last rule weight.
9.6 WAV Instruction Details
The WAV instruction performs weighted average calculations used in defuzzification.
The pseudo-instruction wavr is used to resume an interrupted weighted average op-
eration. WAV calculates the numerator and denominator sums using:
MOTOROLA
9-22
) is temporarily held on internal nodes, so it can be used to ac-
0
is $FFFE and V was one, a change from consequents to anteced-
System Output
FUZZY LOGIC SUPPORT
is $FFFF (end of rules mark). If this
x
n
∑
S
F
i
i
i
=
1
=
--------------------- -
n
∑
F
i
i
=
1
CPU12
REFERENCE MANUAL
Advertisement
Table of Contents
