How does the Hspice calculate the DFT of a signal ?

[jiangwp]

When i do the DFT of a signal by hspice such as :

.FFT v(x) NP=25536 window=bartlet FMIN=1K FMAX=20MEG

I do not known how hspice transfer a continuous time signal v(x) to a discrete frequency signal . However, the DFT mathematic equation can transfer a signal from the time domain to the frequency , how does it approximate a period to a
frequency unit ? and how does it present the spectrum ?

 

[andy1]

I believe it uses FFT to do discrete Fourier transform. FFt treats the entire window of signal as one period. For example, given x(t) for t = t1 to t2, then one period equals t1 to t2 (t2 - t1). FFt assumes that x(t) repeats after t2 with the same period.

 

[jiangwp]

Yeath , you are right.
In the mean while , i know the step frequency (approximating delta frequency=2/(Tstop)) , by it , we can calculate the point number for frequency range estimated by fft .
But the frequency step effects the time of estimation and the accauracy. How does we select the step as resonable as possible . Is there equation for it.

 

[lakeoffire]

1/Time window will determines the frequency step in DFT. So for the sake of accuracy, you need to run longer transient simulation.

 

[andy1]

Yeath , you are right.
In the mean while , i know the step frequency (approximating delta frequency=2/(Tstop)) , by it , we can calculate the point number for frequency range estimated by fft .
But the frequency step effects the time of estimation and the accauracy. How does we select the step as resonable as possible . Is there equation for it.

I am not quite sure what you mean and it may have been answered by lakeoffire but if you're looking for finer step or better resolution in the frequency domain then just take more point during transient analysis but keeping the same period.