Applications of Emergent Phenomena - Systematic Design for Emergence in Cellular Nonlinear Networks - page 152

Fig. 8.17. Rate-distorsion curve giving comparison between our method in either simple
(CA-VQ) or with entropic encoder (CA-VQ-E) and the JPEG standard

8.4.4 Aspects on Hardware Implementation

The CA-Q encoding algorithm is the most critical in terms of execution time
while decoding is a fast process. However the proposed CA-Q scheme is ex-
tremely simple consisting in a comparison between a certain binary block in the
current bitplane and a D -sized codebook stored in a memory. Let us denote ' the
time required by the hardware unit to compute the Hamming distance between two
blocks and decide whether this is greater or not than a previously stored value. For a
sequential implementation of the Hamming unit (which is as simple as XOR-ing
two bits and consequently incrementing a counter if the two bits are different) and
assuming some two additional cycles for comparison and the eventual storage of
the closest code-word it follows that

CLK

w 2 ' , where w is the block size

and CL T is the clock period (determined by the implementation technology). For
a variable block size, and an image size of M×N (assumed that both M and N are
integer multiples of w ) the overall computation time may be evaluated as:

2 T

m

MN

m

¦ ¦

T T DwT 2

2

# ND .

(8.2)

encode CLK k k CLK k

2

w

k 1 k 1

k