Issue link: https://resources.pcb.cadence.com/i/1480210
29 Advanced Topics Resolution of the Impulse Response For frequency independent R and G, OrCAD PSpice calculates closed form impulse responses [1]. But, when R and G are Laplace expressions, a numerical impulse response is calculated using FFT. The minimum number of points for the FFT is 256 and the maximum is 65536. Decreasing RELTOL will increase this frequency resolution. Here are some points to keep in mind: Large G/C and R/L ratios (especially > ~1E8) can give inaccurate results, and the simulation may diverge. If one of these ratio exceeds 1E10, PSpice will issue a warning for this condition: WARNING – G/C for T_T1 is 1e+015. Results may be inaccurate for G/C > 1e10. For these high loss lines, use a lumped modeling approach as discussed previously. The time step size will decrease as you increase the R/L or G/C ratio (for highly lossy lines). For these cases the controlling factor for time steps becomes the time resolution of the impulse responses used in convolutions for the loss branches of the transmission lines. To increase resolution for fast inputs set the Print Step to Tr/10. When Laplace expressions are used for R and G, numerical impulse responses are computed based on FFT. The number of points used, and hence the resolution varies between 256 and 65536. PSpice tries to estimate the best resolution for the simulation, by examining the transmission line model and length, RELTOL, the Print Step, the Step Ceiling, and the Final Time. There is, however, a tradeoff between resolution and simulation time. The number of points can be increased by setting a smaller Print Step, which increases the resolution of the impulse response. Setting the Step Ceiling smaller will also increase the resolution, but may also impose an unnecessarily small maximum time step during the transient analysis. Essentially the reason for this "user control" is that when the impulse response is computed for a lossy tline, it has no knowledge of the input that will be applied during the transient analysis. Example: 10 m of RG58/U coax is used for a 60ns and 600ns simulation. The impulse response is ~4% non-causal, but the results are still valid. The 60ns simulation shows good resolution at far end of line, however, the 600ns simulation requires setting RELTOL=0.0001 to obtain the same good resolution. Non-Causality And The Numerical Impulse Response PSpice uses a numerical convolution to obtain the impulse responses for lossy transmission lines with frequency dependent losses. It is an assumption fundamental to the convolution method [1] that the impulse response is a causal function of time. Unfortunately, not all Laplace expressions have causal impulse responses. What to do when PSpice reports that the impulse response is non-causal? If an impulse response is partially non-causal, PSpice will write a message to the output file: WARNING -- 10.9038 percent of T_T1 impulse response is non-causal.