Notice that the break-away in the late side window margin data (due to the errant pulses in the histogram) is causing a poor curve fit for the late portion of the plot. (The dashed line is the extrapolation). While the margin extrapolation algorithm assumes that the distribution is purely Gaussian, this is a case where it is clearly not. You may therefore wish to experiment with fitting or weighting the extrapolation more towards the �tails� of the distribution rather than the center of the distribution. Conversely, you may wish to ignore certain outliers which may be due to media defects, or other known abnormalities. The HP E1741A allows you to modify the data range over which the extrapolation is fitted, as shown next. 3 Click on the right marker, and while holding the left mouse button down, drag the marker on the late portion of the window margin curve to where the data portion of the curve starts to break away; about �3.5 (look at the y-axis, right marker readout at the bottom of the display). By modifying the default extrapolation values, you will be able to cause the extrapolation to ignore this break-away in the curve. 4 5 6 Select Window Margin>Extrapolation levels. Highlight the ��10� value for the extrapolation. Enter ��3.5� and click the OK button. This directs the software to use only error-rate data in the range of 10�1.5 to 10�3.5when generating the extrapolation curve.
