Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT
Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT
Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
of these simulations fail to meet requirements all together. In contrast, all of the<br />
simulations meet requirements for both of the RPTT designs, and only 34.5% and<br />
24% of the simulations require tuning for the α =0andα =0.1 designs, respectively.<br />
Since there is a cost associated <strong>with</strong> hardware tuning it is perferable to design the<br />
system such that it is likely to meet requirements nominally. In this sense, the RPTT<br />
design <strong>with</strong> α =0.1 performs the best at this requirement and uncertainty level since<br />
all simulations meet the requirement and only 24% require hardware tuning.<br />
A plot of the nominal, worst-case and tuned performances of RPTT designs at<br />
various values of α at this uncertainty level (∆ = 21.5%) is shown in Figure 5-11 along<br />
<strong>with</strong> the requirment at 330µm. Note that the tuned worst-case performance is above<br />
the requirement for values of α just over 0.1. The results of the hardware simulation<br />
indicate that to produce the best RPTT design α should be set to the maximum value<br />
at which the tuned worst-case performance meets the requirement. This robustness<br />
weighting produces a design that is robust to uncertainty, yet is still tunable to meet<br />
requirements. As indicated by the bar chart (Figure 5-10(b)), including maximum<br />
allowable robustness in the cost increases the number of uncertainty realizations in<br />
which hardware tuning is not necessary.<br />
5.4 Summary<br />
<strong>Robust</strong> performance tailoring for tuning blends the concepts of robust design and<br />
hardware tuning to produce a design <strong>with</strong> good nominal performance that is both<br />
robust to uncertainty and has sufficient authority to meet requirements through hard-<br />
ware tuning if necessary. The RPTT formulation extends the idea of robustness from<br />
the design to the physical system by optimizing to facilitate hardware tuning by mini-<br />
mizing the worst-case tuned performance over the uncertainty space. During tailoring,<br />
the performance prediction accounts for the fact that the tuning parameters can be<br />
changed when the hardware is built and the uncertainty is fixed (and known) i.e. the<br />
value of �y is different depending on the value of �p. This knowledge adds extra degrees<br />
of freedom to the problem and results in a less constrained optimization.<br />
177