calculation of equivalent input noise

[nishanthpv]

Hai,
Can anybody tell me how the cadence tool calculates the total equivalent input noise ? If we plot equivalent input noise with frequency (with spectre ) we will get a graph. What the software meant by the y axis variable ?

 

[FvM]

Should be noise voltage or current density, I assume.

 

[nishanthpv]

sir,
When i plotted equivalent input noise i got a graph with negative slope( x1=0 Hz:y1=13.5e153 volts/sqrt(Hz), x2=2 Hz:y2=0 ). But when i plotted squared input noise which is simply the square of the above function, i got a graph with negative slope(x1=0 Hz:y1=182e-6 volts square, x2=2 Hz:y2=0 volts square).Can you give an explanation to this or a detailed explanation to the equivalent input noise which cadence plots (that justifies my graphs )

 

[FvM]

I don't understand the numbers. Neither x nor y can be identical zero in the noise plot.

 

[zorro]

In the first case, at least the units [volts/sqrt(Hz)] are consistent, but the values are strange.
In the second, the unit should be [volts^2/Hz], unless you are seeing the integrated spectrum.
Regards

Z

 

[nishanthpv]

Sir, here i am adding the images. Now i believe that somebody can give an explanation

 

[zorro]

Hi,

"Squared input noise" doesn't look bad, but it would be good to represent it in a log-log- scale in order to better appreciate the rate of variation. Looking at 10 Hz and 100 Hz points, it seems to variate at 20 dB/decade. (The y-axis label should be volts^2/Hz rather than volts/Hz .)
"Total equivalent input noise" looks so bad probably because the first point is at a very low frequency, the linear scale doesn't allow to see the values of the other points, and the first x-axis points are not close enough.
My advice: use log-log scale for both graphics and use log-spaced frequency points.
Regards

Z

 

[philcorb]

Noise is calculated at the output then referred back to the input via a transfer function. It is good to check the transfer function is meaningful. The "Total equivalent input noise" plot may be trying to refer noise to a disconnected net with a transfer function of 0 (-inf dB). This would create a meaningless less value on the plot. With SpectreRF is is possible to plot the transfer function used in the noise analysis. I think it may be possible with Spectre too.