What really is the S21 (or S12) noise floor of the PicoVNA 106?
Pico specifies its noise floor with a substantial margin. The specification is –100 dB to 10 MHz, –110 dB to 4000 MHz and then –100 dB to 6000 MHz.
The specification limits apply to a noise quantity known as "displayed average noise". The user in need of a deep noise floor will however need to know more about the actual noise level likely to be seen and perhaps what the numbers mean. That is what we shall discuss here.
All VNA manufacturers quote, in addition to a specification limit, a typical noise floor level. However, different definitions of noise are seen and there can be a lack of clarity in meaning.
So, first of all, let us take a "typical" PicoVNA 106 and either of the Pico test leads. We calibrate an S21 measurement at 10 Hz bandwidth (Isolation and Through) using either of the Pico calibration standards. We then terminate the two ports and sweep S21 from 300 kHz to 6 GHz in 1001 points, and we will see something like this:
We have used 1001 points to ensure that we see sufficient spectral detail; to do so is disadvantageous in a
comparison with the common practice of using 201 points. Statistically this will reveal higher noise peaks.
Meaningful comparison is the intent here, not seeming to be the best.
Pulling that noise trace into Excel (blue) and then adding 15 more sweeps at exactly the same settings (yellow) we see noise statistics at work as more and more data points are added. Noise peaks are sitting at around –104 dB with 16 000 data points in the picture opposite:
1. RMS noise level (dark blue). Here, the log scale values are converted to a linear attenuation factor and then the root mean square of the 16 noise traces is taken and converted back to the log scale of the plot. This trace is a measure of noise power that is present and will be important to those wanting to make noise measurements on the VNA.
2. Average noise level (light blue). Again we converts to linear attenuation factor, average the 16 traces and then convert back to log. This is the true average level of noise present in the measurement. Often quoted, but not necessarily stated to have been calculated accurately.
3. Displayed average noise (purple). To be honest, this is not a meaningful measure at all; it has simply become common practice to quote and specify it (not least because it gives lower numbers). It is a linear average of the 16 traces as they appear on the display – irrespective of the logarithmic scale that they are plotted against. There is validity here in that most VNAs do include a noise averaging feature that works in this way to give a quieter trace.
We have added two specification lines and a guide line:
Solid red – PicoVNA 106 guaranteed specification – displayed average noise
Dashed red – PicoVNA 106 typical specification – displayed average noise
Dashed black – typical level for noise power in the measurement – RMS noise
Median and average difference between RMS and average noise is 1.0 dB in these plots. Max 2.1 dB.
Median and average difference between RMS and displayed average noise is 2.3 dB. Max 4.7 dB.
Finally, the relatively high specification margin applied by Pico takes into account that the PicoVNA 106 is an all-new design for which statistical multiple batch data is being compiled. While no unit will be worse than the limit, there is no implication that a given unit is likely to be close to the limit. The typical performance really is typical of what you should expect from the product.