Hello all, Recently, I simulated a Track and hold circuit in Xschem+NGspice and exported the simulation raw file for post processing fft. The signal in time domain seems to be perfect but in frequency domain it looks like something is wrong, if anyone can help, I'd be very thankful. attached the circuit + post processing script + output wave form in time and frequency domains.

clear all;
close all;
clc;
path = "D:\studies\SHA.raw";
[signals, signals_values] = parse_tran_ngspice(path);
time = signals_values(find(strcmp(signals, "time")),:);
y = signals_values(find(strcmp(signals, "v(hold)")),:);
plot(time,y);
N = length(y);
s = abs(fft(y,N))
%s=s(1:N/2);
f= (0:length(s)-1)/N;
sdb = 20*log10(s);
plot(f, sdb);
xlabel('Frequency [ f / f_s ]');
ylabel('Amplitude [ dBFS ]');

I'd suspect

plot(y)

. Transient has no constant timestep and

fft(y,N)

doesn't know about the

time

vector

What do you suggest?

Check if

time

does not have constant timesteps, if confirmed rebase y with interp1()

It actually doesn’t have constant time-steps, how can I rebase y?

Thanks.

@Aly you can run ngspice terminal mode and write the following: set lin-tstep=10n linearize v(VSH) wrdata txt v(VSH) then you will have a linearized version from your wave in the simulation files. However, something is still missing!

Thanks, Karim!

in ngspice there is a

linearize

command that transforms tran vectors to constant timestep: section in ngspice manual: 17.5.40 Linea`rize*: Interpolate to a linear scale.` This should be done before processing data for fft.

Are the voltage values captured at the constant time steps times or at the non linear then linearised by a math function? Thanks, Stefan.

As far as I understand the Linearize command will sample the vector to the first interval given in the

tran

command. Example: in my simulation I have a vector generated with a

tran 0.05u 1m

that is

25852

points long. Applying the linearize command i get a

20001

points vector, which is exactly a vector running from 0 to 1m with 50ns increments. Voltage values are interpolated if no exact X-value is present (I guess taking the point before and the point following and doing linear interpolation, but I have not checked that).

I understand, do you think that if the process is interpolating the values if no

y

values exist for the `x`(linearised) values will affect the spectrum assuming that the input signal is a sine wave. I wouldn’t say that it’ll be affected if the input is a ramp. Do you agree?

I agree. Verify that the sampling time interval is dense enough to cover the highest signal frequency. If not restart the tran command with a smaller interval.

Will try that! Do you also think that rectangular windowing causing discontinuity between periods of the fft and causing spectral leakage? Thanks so much, Stefan!

I think for this question you are much more expert than me !

That was very helpful, thanks.