It might be caused by interpolating the simulation data (which is not equally spaced) before doing the FFT. Did you try to do something with tstep, tstart, and tstop ? I see similar problems (less severe but distracting) with another simulator. It becomes less visible when increasing the simulation duration and/or decreasing tstep. -marcel
https://arxiv.org/pdf/1908.00204 uses a cheap 190$ Titan X card and achieves a peak 16 times speedup. However, it appears to only speed up the solver part of the simulation, and therefore needs very large circuits (more than 40,000 rows in the MNA) before the acceleration rises above 2x. Maybe Francesco can share some recent insights here ... -marcel PS: I couldn't find https://github.com/sheldonucr/GLU_public but saw https://github.com/chenxm1986/cktso-gpu/blob/master/results.pdf. This one seems...
https://arxiv.org/pdf/1908.00204 uses a cheap 190$ Titan X card and achieves a peak 16 times speedup. However, it appears to only speed up the solver part of the simulation, and therefore needs very large circuits (more than 40,000 rows in the MNA) before the acceleration rises above 2x. Maybe Francesco can share some recent insights here ... -marcel PS: I couldn't find https://github.com/sheldonucr/GLU_public but saw but https://github.com/chenxm1986/cktso-gpu/blob/master/results.pdf. This one seems...
https://arxiv.org/pdf/1908.00204 uses a cheap 190$ Titan X card and achieves a peak 16 times speedup. However, it appears to only speed up the solver part of the simulation, and therefore needs very large circuits (more than 40,000 rows in the MNA) before the acceleration rises above 2x. Maybe Francesco can share some recent insights here ... -marcel
The S-parameter data supplied for the vast majority of RF devices was generated under "small-signal" test conditions using a VNA. S-parameter data for a RF power device isn't valid when the device is actually operating at high power levels. That means that the time-domain model derived from that data is also only valid for small signal (i.e. linear) analysis. I don't think that makes it less useful. How did large-signal S-parameters handle ring-modulator / mixer circuits? -marcel
As mentioned above, the manual states in its Backus-Naur diagram for .param that an '=' sign is required. There is no exception to this rule mentioned in any of the compatibility mode descriptions (Chapt. 12.14.5 - 12.14.10).
Do you have access to something like rational() ? Use the rational object to perform rational fitting on complex frequency-dependent data. This object uses a non-iterative interpolatory algorithm to construct a fit with complex frequencies s. The fit of the each element of the rational object is given by this equation: n Ck F(s)= ∑ ---- + D, where: s=j×2πf, C=Residues, A=Poles, D=Direct term k=1 s−Ak https://nl.mathworks.com/help/rf/ref/rfckt.rational.html?s_tid=doc_ta The F(s) is your FREQ data....
You can say '2Ampere' or '2 A' or '2 ampere' if it is clarity you're after. I like to always write '2uH', etc., but I never needed '2aH'.