I have recently started using an 8753C and I am trying to understand the accuracy limits of my measurements. I am a bit uncertain as to how the noise can be included in the uncertainty calculations.

There are two types of noise, Low-Level (Noise Floor) and High-Level (Phase Noise effects etc.), and the 8753C specifications provide values for each, e.g.

Low-Level Noise: -100 dBm

High-Level Noise: 0.004 dB

The 8753C specifications also has some nice equations for the uncertainty of transmission and reflection measurements, but these ignore both noise sources.

Looking on the web-site I have found, in the PNA help, similar equations that do include the noise sources:

Noise^2 = (Nt x S11)^2 + Nf^2

Where Nt is the trace (high-level) noise and Nf is the noise floor.

The difficulty that I have is this:

In all the specifications that I can find the noise floor figure is a dBm value. Does it make sense to include and absolute power (units dBm or mW) value in a calculation in which all other values are unit-less?

I'd be grateful for any thoughts,

Peter

There are two types of noise, Low-Level (Noise Floor) and High-Level (Phase Noise effects etc.), and the 8753C specifications provide values for each, e.g.

Low-Level Noise: -100 dBm

High-Level Noise: 0.004 dB

The 8753C specifications also has some nice equations for the uncertainty of transmission and reflection measurements, but these ignore both noise sources.

Looking on the web-site I have found, in the PNA help, similar equations that do include the noise sources:

Noise^2 = (Nt x S11)^2 + Nf^2

Where Nt is the trace (high-level) noise and Nf is the noise floor.

The difficulty that I have is this:

In all the specifications that I can find the noise floor figure is a dBm value. Does it make sense to include and absolute power (units dBm or mW) value in a calculation in which all other values are unit-less?

I'd be grateful for any thoughts,

Peter

High level trace noise, especially in the 8753, is due, principally, to the "noise pedestal" of the source, close-in to the signal. This noise pedestal is something like 60 dB below the signal. When the signal in the receiver is high, this noise dominates. As signal from the source is attenuated, this noise pedestal is also attenuated, leaving you with the low level noise floor of the receiver.

So from the high level trace noise value, you can compute the noise pedestal in linear noise units (sort-of like noise voltage, but refrenced to sqrt of impedance). This adds to the the other uncertainties. You can also compute such a noise from the noise floor, and see at what level the noise floor starts to dominate. Where both terms are significant, it is typical to RSS (root-sum-square) the contributions, since it is assumed that they won't add up in a worst case way.

In the normal uncertainty specs on the analyzer, the assumption is that the measurement is made in a 10 Hz IF bandwidth, so the trace noise effects are de minimis and are ignored. The noise floor effect is included in the uncertainty curve at the higher attenuation points, but -100 dBm noise floor should show a 6 dB uncertainty at -100 dB S21. I'd have to look back at the curves to see if it does, but I'm guessing that it does. I think (it's been a long time since I looked) the dynamic accuracy curve includes noise effects.