10.06.2016 Views

eldo_user

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Monte Carlo Analysis<br />

Primary Statistics of Uncertainty Analysis<br />

Tail Bndry/ Lower Pr. Value Upper CV(%) #Runs Speedup Name<br />

Dist. Sigma Tail / Comp of Meas<br />

--------------------------------------------------------------------------------<br />

Left 1.75e+09 7.69e-05 8.45e-05/1.0e+00 9.22e-05 9.05 3530 1.6e+03 FOSC<br />

-4.631 NQ=(-3.785 -3.761 -3.739)<br />

Right -1.10e+02 3.53e-06 3.88e-06/1.0e+00 4.23e-06 9.05 5560 2.2e+04<br />

21.303 NQ=( 4.492 4.472 4.453)<br />

PHASE_N<br />

These values may be compared with the results obtained using the standard Monte Carlo<br />

method (refer to “Probability Estimator using Standard Monte Carlo” on page 480).<br />

Using the standard Monte Carlo method, the upped bound for the phase noise response was<br />

5.99×10 −4 . The ISMC method returns 3.88×10 −6 as an estimation of the probability, with 9%<br />

precision. The confidence interval (3.53×10 −6 , 4.23×10 −6 ) for this estimator is given on the<br />

same line. Notice the major gap (almost two orders of magnitude) between the upper bound<br />

based on the binomial law and the value obtained using ISMC. This implies that the probability<br />

of the phase noise exceeding -110dBC/Hz is much less than the conservative estimation<br />

computed using the standard Monte Carlo method.<br />

An estimation of the speedup over the standard Monte Carlo method is given in the .mcm file. In<br />

the case of the phase noise response, approximately 5560 × 2.2 × 10 4 ≈ 120 million standard<br />

Monte Carlo runs would need to be carried out to obtain equivalent results (in terms of value<br />

and precision).<br />

The .mcm file contains the normal quantiles (NQ) corresponding to the probability estimator<br />

and its confidence interval. The normalized distance of the <strong>user</strong> specification to the mean of the<br />

distribution (q σ ) is also given. In the above example, the first response is slightly non Gaussian<br />

(compare NQ = -3.761 with q σ = -4.631), while the phase noise has a very right-skewed<br />

distribution which is clearly non-Gaussian (compare NQ = 4.472 with q σ = 21.303). In this<br />

second case, making the assumption that the output distribution is Gaussian can lead to severe<br />

underestimation of the true risk.<br />

The ISMC method produces an .mcm file containing the following data about quantile<br />

estimators:<br />

Tail Probability Lower Quantile Upper Prec(%) #Runs Speedup Name<br />

(Goal Value) Value of Meas<br />

--------------------------------------------------------------------------------<br />

Left 1.00e-04 1.750e+09 1.751e+09 1.752e+09 0.06% 3010 2.0e+02 FOSC<br />

Right 1.00e+00 -1.124e+02 -1.121e+02 -1.119e+02 0.27% 2510 1.1e+02 PHASE_N<br />

Two columns are added to the standard Monte Carlo results: the number of runs associated with<br />

the estimator and the speedup over the standard Monte Carlo method. The precision is obtained<br />

by inverting the confidence interval on the underlying probability estimator.<br />

484<br />

Eldo® User's Manual, 15.3

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!