Understanding why panadapter noise floor & S-meter are different, why the atten. makes no difference, and S-meter cal.

User avatar
Posts: 1041
Joined: Fri Mar 10, 2017 1:47 pm

Understanding why panadapter noise floor & S-meter are different, why the atten. makes no difference, and S-meter cal.

Postby w-u-2-o » Tue Jun 13, 2017 1:55 am

Summary: the panadapter noise floor will NOT match the S-meter reading of noise power and there are good mathematical reasons for this.


Understanding the difference between what the panadapter is showing you and what the S-meter is showing you is a relatively new problem for hams but old hat to RF engineers.

In the RF engineering world the same kind of confusion often exists for new engineers when they try to make sense of measurements using a spectrum analyzer (the panadapter equivalent) and an RF power meter (sort of the equivalent of the S-meter).

Panadapter/Spectrum Analyzer Displayed Average Noise Level (DANL):

What you are seeing on the panadapter (spectrum analyzer) is known as the Displayed Average Noise Level (DANL). Some people call it Displayed Average Noise Floor. By either name it's the same thing.

Before discussing the DANL further, understand that these days there are two different kinds of spectrum analyzer measurement methods. Older analog spectrum analyzers used a swept superhet receiver with a filter after it. The bandwidth of that filter is the "Resolution Bandwidth". Narrow resolution bandwidths require longer sweep times because the local oscillator in the spectrum analyzer can only be moved in an accurate fashion so quickly and because the risetime of a signal is inversely proportional to its bandwidth. Newer, digital spectrum analyzers perform FFTs and construct the spectrum using math. They can make instantaneous snapshots of a band because the data is gathered in a very wide filter and nothing needs to be tuned or swept. The Bin Width of the FFT is analogous to the Resolution Bandwidth of the swept superhet.

Now obviously our modern SDR radios are essentially inexpensive digital spectrum analyzers using FFT methods. When comparing the DANL on any combination of spectrum analyzers or SDR radios the Bin Widths and/or Resolution Bandwidths MUST be matched in order to get the same results. The reason for this is because ALL RF power measurements MUST be referenced to some bandwidth. Clearly, there is more noise power in 10Hz of bandwidth than 1Hz, so this should make some intuitive sense. Assuming the noise is normal Gaussian noise, and that the Resolution Bandwidth/Bin Widths are all matched, any differences between measurement instruments or radios are going to be the result of differences in the noise figure of those instruments or radios.

The bottom line here is that the DANL is the noise power that exists within a single Resolution Bandwidth/Bin Width on the panadapter or spectrum analyzer.

S-Meter Noise Measurement:

Now looking at the S-meter, we can just consider it a single channel, non-swept, spectrum analyzer with an adjustable Resolution Bandwidth/Bin Width. In our case, that adjustment is the receive filter passband. Let's look at an example of how that compares to the panadapter.

If the panadapter Bin Width is 3Hz and the S-meter bandwidth is 3KHz, then the S-meter will be taking in 1000 times more noise energy than a single panadapter bin. A factor of 1000 equals 30dB (10log1000). So clearly the DANL will appear MUCH lower than the S-meter reading of the same noise condition. You can easily demonstrate this to yourself by fooling with the Bin Widths and receive passband settings to obtain any difference you want within the range of adjustment of the radio. Indeed, most RF engineers will measure and report broadband noise by referencing it to the bandwidth of a single Hz. For instance the thermal noise floor used in many engineering calculations is -174dB/Hz. That way any other RF engineer can calculate the total noise power in any bandwidth he wants.

To calculate the difference between two bandwidths in decibels, use the following formula: 10log(bandwidth 1 in Hz/bandwidth 2 in Hz). Example: 10log(3000Hz/3Hz) = 30dBHz.

Better example: your panadapter bin width is 2.93Hz and you see a DANL of -120dBm. Your receiver passband is set to 2.9KHz. 10log(2900/2.93) = 29.96dB. With just noise in the passband (no signals), the S-meter will display -90dBm (approx. S6).

Why The Differences in Noise Power Don't Match Differences in Signal Power:

So far so good, at least until somebody tries to compare noise power to signal power. Then they get all confused. But that's because they are comparing apples to oranges. Broadband noise is different than a narrowband signal. Consider a CW signal with an amplitude of -73dBm, which is equal to S9. IF the bandwidth of the CW signal fits ENTIRELY within a single FFT bin then it will appear on the panadapter with an amplitude of -73dBm. AND it will appear on the S-meter with an amplitude of S9 (actually a tiny bit more because there is noise power in there too, but the difference the noise power makes is negligible when compared to such a large signal). Now imagine if the CW signal is twice as wide as an FFT bin (unlikely, but please bear with me). In such a case the amplitude of the CW signal on the panadapter will appear to be -76dBm. That is 3dB, or a factor of 2 less, because now the power of the signal is spread over TWO panadapter bins. Meanwhile the entire CW signal still easily fits within the S-meter passband so the S-meter still reads S9. If the CW signal spread over 4 bins it would appear to be 6dB lower on the panadapter. If it spread over 10 bins it would appear to be 10dB lower. And so on and so forth. Take this to the limit and you can see how the wider a signal is compared to a Bin Width the lower it appears on the panadapter compared to the S-meter. And noise is infinitely wide!

Now obviously CW signals are not that wide. But SSB phone signals are! Since the energy is spread over the entire SSB signal bandwidth no individual peak will ever be as high as the S-meter reading, unless the operator whistles a pure tone and the bandwidth of that tone fits entirely within a single panadapter bin. Thus it is very, very difficult to make comparisons of SSB power levels on the panadapter vs. SSB S-meter readings. They are not so neat and clean as a CW or noise signal.

S-meter Accuracy:

Finally, do not put too much faith into the accuracy of the typical Kenwood/Yaesu/Icom S-meter where noise is concerned. They are, in fact, not that accurate. It would appear that the marketing departments at those companies have had the engineering departments arrange things so that the S-meters are highly non-linear below S9, such that they can claim to have "quiet" receivers. Those receivers are not any quieter than the receivers on our ANAN series radios, or Flex radios for that matter. But their S-meters are junk compared to our honest ones. So don't feel bad when your ANAN is telling you that you have S7 noise on the band and your trusty FTDX5000 is telling you S3. Guaranteed the ANAN is correct and the Yaesu or whatever is lying. You can prove this by using an expensive spectrum analyzer to measure the correct channel power. The channel bandwidth will either be that of the S-meter or of the panadapter bin width. When testing S-meter accuracy set the Resolution Bandwidth to be EQUAL to your receiver passband. When testing panadapter accuracy set the Resolution Bandwidth to be EQUAL to your FFT bin width.

See also: http://rfmw.em.keysight.com/spectrum-analyzer/



User avatar
Posts: 1041
Joined: Fri Mar 10, 2017 1:47 pm

Postby w-u-2-o » Sat Aug 19, 2017 2:44 pm

Why the S meter and panadapter don't change readings when you change the attenuator setting:

The software that forms our radios is smart. When it shows you an S meter reading, or a signal level on the panadapter, that is referenced to the antenna port on the rear panel.

In other words, the software adjusts for the attenuation you have selected. Thus, if you are seeing an S9 signal at the ADC input with 0dB of attenuation, and the S-meter is reading S9, and then you add in 30dB of attenuation, yes, the signal gets smaller at the ADC, but the software knows what you did so it adds the 30dB back in on the S-meter. That way it is always reporting the conditions at the rear panel connector, i.e. what is coming out of your antenna.

Note also that the pre-amp in our radios is not defeatable/selectable. The software also accounts for the pre-amp gain automatically. This is 20dB in all radios except the 8000 where it is 14dB.

Why the noise floor appears "so high" when compared to a <fill-in-the-blank> radio:

The noise floor you are seeing of S2 on the S-meter with no antenna connected is correct. Refer to the discussion on S-meter accuracy above.

On the 8000 I have here, into a dummy load, using the standard 2.7KHz filter preset, I get -125dbm with the meter on Sig Avg, and around -117dBm on Signal (peak reading). You are probably using the Signal meter mode (peak reading) and S2 = -115dBm (close enough!)

Looking at this a bit more closely, the thermal noise floor is -174dBm/Hz and 2700Hz = 34.3dBHz. This gives a total noise power of -174 + 34.3 = -139.7dBm in a 2.7KHz bandwidth. Using the Sig Avg meter reading from above, that implies a receiver noise figure of about -125 less -139.7 = 14.7dB, which is about what you'd expect. Note that approx. half of that noise figure, 6.2dB, is from the pre-amp, which is an LTC6400.


User avatar
Posts: 1041
Joined: Fri Mar 10, 2017 1:47 pm

Re: Understanding why the panadapter noise floor and S-meter are different and why the attenuator makes no difference

Postby w-u-2-o » Sun Mar 11, 2018 1:47 pm

Why you don't need to and probably shouldn't calibrate your S-meter:

If you're read everything above and still think you need to calibrate your S-meter because the S-meter doesn't read the same as your trusty YaeKenCom, consider the following additional facts.

Since thermal noise is a universal constant, and since most receivers have approximately the same noise figure (form follows function), any radio that is showing less than an S2 noise floor with a terminated input is demonstrating how poorly calibrated or linearized its S-meter is. So seeing S2, give or take, on your radio with a terminated or shorted antenna input is exactly correct, the math for this being discussed above.

There is no "factory calibration" required. A multi-conversion superheterodyne receiver, such as that in your trusty YaeKenCom, has a great many variables that affect front end gain and loss, including LO levels, mixer variability, multiple stages of IF amplification, etc. This level of complexity does not exist in a direct conversion receiver. The difference in input gain between radio serial numbers in a direct conversion receiver is minimal because there are only two active components in the receive signal chain (not including the 6M pre-amp): the ADC pre-amp and the ADC, and they are all essentially identical as delivered from the manufacturer. The variation in the remaining passive components in front of the pre-amp are minimal at HF frequencies. I'd hazard a guess that all Apache or Apache-like radios are +/-2dB in terms of front end gain/loss.

Lastly, a lot of radios use average reading S-meters. Change your S-meter in PowerSDR to "Sig Avg" to match them up.

You are doing yourself a disservice if you mess with the S-meter calibration on your Apache Labs radio. Unless you have a trustworthy and accurate reference signal source, calibrate your other radios, not your Apache-based radio, because the direct conversion radio is almost certainly more accurate than the multi-conversion superhet's that have had their S-meter linearity messed with by the marketing department.


User avatar
Posts: 1041
Joined: Fri Mar 10, 2017 1:47 pm

S-meter standard reference levels

Postby w-u-2-o » Wed Apr 18, 2018 2:41 pm

S-meter standard reference levels:

The S-meter standard reference levels are defined by IARU Region 1 Technical Recommendation R.1, as ratified at both the Brighton 1981 and Torrelmolinos 1990 meetings. This standardization is documented as follows:

Code: Select all

1. One S-unit corresponds to a signal level difference of 6 dB,
2. On the bands below 30 MHz a meter deviation of S-9 corresponds to an
available power of -73 dBm from a continuous wave signal generator
connected to the receiver input terminals,
3. On the bands above 144 MHz this available power shall be -93 dBm,
4. The metering system shall be based on quasi-peak detection with an
attack time of 10 msec ± 2 msec and a decay time constant of at least
500 msec.

Unfortunately, this leaves S-meter reference levels undefined for the 6M band! For 6M, PowerSDR currently uses the "VHF" reference level of S9 = -93dBm. Whether this is correct or not is a subject for another, separate discussion topic ;)

Return to “PowerSDR mRX”