PRODUCT SPECIFICATION
TMC2330A
R
θ
X
Y
SADR
SADR
SOURCE
IMAGE BUFFER
TMC2330A
COORDINATE
TRANSFORMER
R
θ
DATA OUT
TMC2246A
PIXEL INTERPOLATOR
(2) TMC2302A IMAGE
Σ
RESAMPLING
SEQUENCERS
U
V
U
V
TADR
TADR
DATA IN
(4) TMC2011A
DELAY
REGISTER
TARGET
IMAGE
BUFFER
TWR
Figure 1. Figure 1. First Quadrant Coordinate Relationships
Figure 2. Block Diagram of Scan Converter Circuit Utilizing TMC2330A and TMC2302A Image Resampling Sequencer
The results of the 10,000-vector study were as follows:
Arithmetic Error for Two’s
Complement Rectangular to Polar
Mean Error (X)
+0.0052LSB
0.0031LSB
0.662LSB
0.664LSB
1.025LSB
1.020LSB
+4/ -5 LSB
+5 -4 LSB
Conversion
A random set of 5000 input vector coordinate pairs (X,Y),
uniformly spread over a circle of radius 32767 was converted
to polar coordinates.
Mean Error (Y)
Mean Absolute Error (X)
Mean Absolute Error (Y)
Root Mean Square Error (X)
Root Mean Square Error (Y)
Max Error (X)
Radius Error Range
Mean Radius Error
–0.609 to 0.746 LSB
0.019 LSB
Max Error (Y)
Mean Absolute Radius Error
0.252 LSB
Since this is a double conversion (rectangular to polar and
back) which includes a wide variety of “good case” and “bad
case” vectors, the chip should perform even better in many
real systems. Repeating the experiment and restricting the
original data set to an annulus defined by 8196<R<32768
reduced the mean square error to 0.89 LSB and the peak
error to 4 LSB (x or y). These latter results are more ger-
mane to synthesizer, demodulator, and other applications in
which the amplitude can be restricted to lie between quarter
and full scale. The largest errors tend to occur in the angle
component of small radius cartesian-to-polar conversion.
Phase Error Range
Mean Phase Error
Mean Absolute Phase Error
–1.373 to 1.469 LSB
0.058 LSB
0.428 LSB
Statistical Evaluation of Double
Conversion
In this empirical test, 10,000 random Cartesian vectors were
converted to and from polar format by the TMC2330A. The
resulting Cartesian pairs were then compared against the
original ones. The un-restricted database represents uniform
sampling over a square bounded by -32769<x<32768 and
-32769<y<32768.
12
REV. 1.1.8 10/31/00