When I use Spectre to perform Pnoise simulation, I was really confused with the reference side band setting. Referred to the SpectreRF User Guide, a typical down conversion mixer should be simulated with the "-1" ref side band to calculate IF noise. When I design a harmonic mixer which uses the 2nd harmonic of LO to mix with the RF signal, is it correct to set the ref sideband to "-2"?
 Again, when I design a divide-by-2 circuit, and I want to calculate its phase noise, which ref side band should I use?
And how about the pfd?

The Jitter Man is right. The refsideband parameter is used to determine the input frequency. It is used when computing either the input referred noise or the noise figure. The PNoise analysis directly computes the output noise, and to compute either the input referred noise or the noise figure, it must divide through by the gain. To compute the gain, it must know the location and frequency of both the output and the input. To do the noise analysis you had to specify the output, which you generally do by specifying a pair of output nodes. You specify the output frequency when you specify the noise analysis frequency sweep range. To compute the input referred noise or the noise figure you also need to specify the input sources. If the source is either a voltage or current source, you get the input referred noise. If the source is a port, you get the noise figure. However, just specifying the name of the source is not enough. It must also know the input frequency, which you specify using refsideband. The input frequency is related to refsideband as follows ...
   |f_in| = |refsideband*f_PSS + f_out|
In other words, refsideband is the amount that the input frequency differs from that of the output frequency and is measured in multiples of the PSS fundamental frequency (the clock or LO frequency). It is constrained to be an integer.

Consider a circuit that has a PSS fundamental frequency of 1GHz. Further consider performing a PNoise analysis of that circuit and focus on one point in that noise analysis, the point where the output frequency is 1MHz.

• If refsideband=0, then f_in=1MHz. This would be the case in circuits where the input and the output are at the same frequency. This is generally the case in switched-capacitor filters, chopper-stabilized amplifier, sample-and-holds, etc.
• If refsideband=1, then f_in=1.001GHz. This would be the case in an high-side fundamental mixer.
• If refsideband=-1, then f_in=999MHz. This would be the case in an low-side fundamental mixer.
• If refsideband=2, then f_in=2.001GHz. This would be the case in an high-side second harmonic mixer.
• If refsideband=-2, then f_in=1.999GHz. This would be the case in an low-side second harmonic mixer.
• Etc.

Consider a frequency divider. If the input clock frequency is at 2GHz and it is a divide-by-two divider, then the output frequency will be at 1GHz and the PSS analysis will be performed with a fundamental frequency of 1GHz. Now, if you want to know the output spot noise at 1.001GHz, you would specify the noise analysis frequency to be 1.001GHz. If you also want to know the input referred spot noise at 2.001GHz, then you would specify refsideband=1 because the input frequency is one multiple of the PSS fundamental frequency above the output frequency.

In this example, the quantity of interest is the single-sideband spot noise. If you want to know the phase noise, you would use the procedure I described in my paper on phase noise, which involves computing the strobed noise and dividing through by the slew rate at the threshold crossing. Here you are only using the ouput noise, which is unaffected by refsideband.

-Ken