effect of data correlation on fir filter
DESCRIPTION
EFFECT OF DATA CORRELATION ON FIR FILTER. Deepaknath Tandur K Sumit Ahuja. OBJECTIVE. Implement an FIR filter as RTL netlist using VHDL in two forms: No shared resources Shared resources Select 5 input data streams characterized by different correlations. - PowerPoint PPT PresentationTRANSCRIPT
EFFECT OF DATA CORRELATION ON FIR FILTER
Deepaknath Tandur KSumit Ahuja
OBJECTIVE Implement an FIR filter as RTL netlist
using VHDL in two forms: No shared resources Shared resources
Select 5 input data streams characterized by different correlations.
Synthesize the two implementations down to the gate-level.
Estimate power consumption and discuss the relation between power and data correlation.
UNSHARED RESOURCE IMPLEMENTATION
u[k]
y[k]
u[k-5]u[k-4]u[k-3]u[k-2]u[k-1]
b0 h5h4h3h1h0
+
* * * * * *
+ + + +
h2
Fig1: Direct Realization of FIR filter with unshared resources.
SHARED RESOURCE IMPLEMENTATION
h3
Time 6
Time 5
Time 4
Time 3
Time 2
Time 1
h5 h4h2h1
*
*
+ *
+ *
+
+
*
*
+
u[k] u[k-2] u[k-3] u[k-4] u[k-5]u[k-1]h0
y[k]
Time 7
Fig2: Direct Realization of FIR filter with shared resources.
DATA CORRELATION & INPUT PATTERNS
Highly Correlated Input pattern00000000 00000001 00000011 and so on.
Correlated Data Pattern00000000 00000001 00000010 and so on.
Mid-correlated Data Pattern00000000 00000011 00001111 and so on.
Uncorrelated Data Pattern00000000 00001111 11111111 and so on.
Highly Uncorrelated Data Pattern00000000 11111111 00000000 and so on.
RESULTS
Effect of Data Correlation on Shared resources
4.95
5
5.05
5.1
5.15
5.2
5.25
5.3
5.35
highest high 2bit 4bit lowest
Correlation
Dyna
mm
ic Po
wer(m
w)
Series1
Effect of Data Correlation on Unshared Resources
0
1
2
3
4
5
6
7
8
9
10
highest high 2bit 4bit lowest
Correlation
Dyna
mic
Pow
er(m
w)
Series1
COMPARISON & CONCLUSIONS:
Data correlation more observable on parallel implementation.
In resource sharing, the data correlation is destroyed.
Parallel implementation is faster as compared to shared implementation.
Dynamic power consumption is highly variable in case of parallel implementation.
Thus power dissipation due to variance in resource sharing to be considered while designing.