Ken Kundert
|
Vincent, What I am proposing is something very similar to using AC analysis to compute power supply rejection. If you were simulating something simple, like an opamp, you could determine the effect of small variations in the supply voltage by computing the transfer function from the supply to the output, and then multiplying the spectrum on the supply by the transfer function to determine the portion of the output spectrum that results from supply noise.
In this case, you cannot use AC because it does not account for frequency conversion. Of particular concern is noise at low frequencies being up converted to the output frequency. However, you could use PAC as it does properly handle frequency conversion.
However, it is even more efficient to use PXF, which is like PAC except that it computes the transfer function from every source in the circuit to the output in a single analysis. To run PXF you need to specify a pair of nodes. This defines the output port. You also need to specify a start and stop frequency. This defines the output frequency range of interest. Finally, you need to assure that there are sources present in the circuit to model noise sources. If the needed sources already exist in the circuit, as it would for the power supply, then you do not need to touch them. If they do not exist, add sources in such a way so as not to affect the behavior of the circuit. In other words, add zero volt voltage sources in series or zero amp current sources in shunt. We don't need these sources to do anything really. The mere fact that they exist and are wired into the circuit is sufficient. Then run the PXF analysis. It will compute the transfer functions from each source to the output.
One last detail is that you are probably primarily interested in up-converted noise. In particular, if the output is at fo=fc+fm, where fc is the carrier frequency and fm is the modulation or offset frequency, then you are probably interested in noise on the supplies at fm up-converting to fo. This is a conversion from the -1 sideband to the 0 sideband (in PXF the output frequency defines the 0 sideband, in otherwords; the index of the sideband is defined relative to the output frequency). Thus, maxsideband must be set to be at least 1, and the transfer function of interest would be the one associated with the -1 sideband. (On my last post I mistakenly gave you the sidebands that would be appropriate for PAC rather than PXF).
So, for example, if you had a PLL configured as a frequency synthesizer that produced a 1GHz output, and you were concerned about power supply noise from 1kHz to 1MHz corrupting your output signal, you could do the following: 0. Assume the output is at node "out" and that the power supply source is named "Vdd". 1. run PSS with the reference clock applied. It will compute the steady-state response with the 1GHz output. At this point, there is no noise signal injected in the supplies. 2. perform a PXF analysis with "out 0" defined as the output nodes and sweep the frequency from 1GHz+1kHz to 1GHz+1MHz (you could do this using either an absolute or relative frequency range). 3. Plot the signal Vdd[-1] from the PXF results. This is the transfer function from fo-fc to fo, where in this case fc=1GHz.
-Ken
|