Noise Floor Change

Bill Bailey

As I expand the horizontal frequency range on the bottom of the screen I have noticed that the baseline of the spectrum display (noise floor) gets lower. This happened on the last release also. Is this a problem? Bill Bailey AE6EQ

Sergey R5AU

Hi Bill, pls look at http://community.flexradio.com/flexradio/topics/band_gain_noise where explanations completed. I guess it is OK for you.

Al K0VM

Bill, That is normal for this radio.. There are a fixed number of bins ( narrow filters ) across the width of the spectrum display. As you change the span of the display the the bin size ( filter width ) changes. The narrower the filter the less noise in the filter and thus the lower the spectrum display noise level.. Note. Changing the spectrum display span dose not have any effect on the width of the receiver filter and thus the noise floor of the receiver. AL, K0VM

Bill Bailey

Is there an implication that in order for the spectrum display noise floor display to accurately reflect the noise floor of the receiver then the bin size would have to be equal to the receiver filter width ? How do you know what the real noise floor is IF the spectrum display noise floor can move around based on bin size ? I do not recall seeing this behaviour on my 5000. Maybe I just never noticed it. B.B. AE6EQ

Al K0VM

In PowerSDR ( the FLex 5000 ) the panadapter binsize was fixed ( for a given sample rate ) and did not cahnge with the zoom. It did however change with the sample rate and so did the panadapter noise floor change with sample rate. PSDR had a maximum of 8:1 zoom factor and a 4:1 sample rate.. The FLex 6000 is a different beast.. The zoom factor is 1000:1 and there would likely be to much data to keep the same bin size through out the whole zoom range. You have likely noted that the receiver slice s-meter does not indicate a change in noise floor when zooming but it does indicate a change when changeing the receiver bandwidth. The panadapter noise floor and the receiver noise floor do share antenna noise and preamp noise figure but not much more. AL, K0VM

Steve-N5AC

Bill, All the answers here are correct. In the ham community, the term "noise floor" is tossed around a lot without an explanation for what it really means. When you hook an antenna to your radio, there is broadband noise from the atmosphere, etc. You hear this noise in your receiver on sideband, but the noise level goes down when you switch to CW mode. This is because the filters in your radio switch from, say, 3kHz, down to let's say 500Hz. By filtering out 5/6 (83%) of the noise with the narrower filter, the amount of noise -- the power level of the noise -- is reduced. For receivers, the noise level is all sent right out your speaker and to your ears. On CW, you notice the reduction of noise as a loss of noise at certain frequencies because your brain can measure signals by frequency. When switching to CW, you hear all of the high frequency noise go away. Because your ear (brain) are used to filtering out signals, you do not perceive anything other than this -- the CW signal may seem to have about the same noise on it. Your brain is doing the equivalent of an FFT on the audio. The panadapter simply measures the signal in a given bandwidth and draws what it hears. If you look at any given pixel, it represents a certain amount of bandwidth. We call this the "bin size" of the FFT that is used to produce the display. If you cut the bin size into two pieces, the amount of noise in each piece goes down by half (3dB). In PowerSDR, the bin size is generally fixed for any given setup and does not change when you zoom. This is why the resolution gets worse as you zoom in on PowerSDR -- you begin to show one bin with multiple pixels. But for SmartSDR, we knew we wanted to have a larger range of zoom and this method was no longer acceptable. So we vary the bin size across a 1000:1 range. So the noise in each bin also varies. 1000:1 is a change of 30dB so from min zoom to max zoom, the noise in a bin will lower by 30dB and you see this change in the panadapter as you zoom in and out. When people talk about noise floor in ham radio they are generally talking about the noise level with a 500Hz bandwidth. When the panadapter is zoomed in to the max level, the bin size today is about 5.8Hz. This is a 19dB difference in noise from where a ham would say the noise floor is to what you can see on the panadapter. This means that the panadapter can see 19dB below what most hams would call the noise floor. Your ear and brain are also able to hear below the noise floor in 500Hz because of how they work. But there are limits to how well you can hear. If you've ever worked JT65 or another long-term integrating mode, you have noticed that your computer can copy signals that you cannot hear. So if you ask another ham "where is your noise floor on 80 meters" and he says "S5," what has he told you? Well with most hams, you don't know because you don't know the answer to these questions: 1. What bandwidth are you using to measure the signal? 2. Is your S-meter calibrated? An S5 signal corresponds to -97dBm. And if he's getting this on sideband set to, say 2.8kHz bandwidth then the actual noise floor in 500Hz would be -97 - 10*log(2800/500) = -104dBm. There's nothing magical about 500Hz, it just happens to be the convention for measuring noise in the ham radio world. In SmartSDR, if you set the passband filter to 500Hz, the S-meter in the slice will show you the 500Hz noise floor. If you start at maximum zoom and begin zooming out, you can see a point where the noise reading of the panadapter equals this number. What do you think this point is? ... if you've been following along, you will realize that this is the point where the FFT bin size is 500Hz. To get a rough idea if this is right, you could measure the width of your panadapter window and divide the amount of frequency displayed by this number. It should be in the 250-1000Hz range. The answer will not be exact because we do not continuously vary the FFT bin size -- we adjust it in steps and don't tell you where the steps are or what size they are. We do what's right for what you are viewing. I know that was a long-winded answer, but I hope it provides some insight into noise and how it changes what you see and hear. Steve

Rodney Thomson

Hi Steve, The best approach when you have varying FFT bin sizes (for differing time/frequency resolution requirements) is to normalise the resulting power by the FFT size squared. In MATLAB it would be something like below (written verbosely): -------------------------------------- data_FFT = fft(raw_rf_time_series, fft_size); power = real(data_FFT)*real(data_FFT) + imag(data_FFT)*imag(data_FFT) power = power ./ (fft_size*fft_size) -------------------------------------- The end result if you then take the log scale of the data is you have power in units of dB re 1 count / Hz. So as your FFT size decreases (and bandwidth increases) the power should remain similar. And if you are aware of the transfer functions within the radio you can convert to real units (ie dBm/Hz). And you could also extrapolate a 500Hz 'noise floor' measurement at all resolutions to then remain consistent with ham convention. (Note I am not a ham person, but have a lot of acoustic visualisation experience and a need for an SDR!) Rod

Steve-N5AC

This is a great idea to show a relative power FFT. But, we are actually showing an absolute power FFT where we read the output power in dBm at the antenna connector. Also, we scale the FFT by your window size on the client anyway, so this scaling would get thrown away in that process.

John Dalrymple

I agree with Rod that you should provide the ability to normalize as he suggests above, so you display dBm based on "absolute watts per Hz bandwidth" at all bin sizes. Otherwise you end up with situations like the attached images show. Which "noise floor display" is correct? Your answer: both of them.

When I first saw this, I thought it was a bug. Now ater browsing this board, I find you're saying it's a feature, however, it's a confusing feature and since you don't tell us the bin size of each display, we can't compute the watts/Hz even if we wanted to.
.



John K7JCD

Steve-N5AC

Couple of things here -- the panadapter shows three main things: 1) signals, 2) Johnson noise from the receiver, 3) atmospheric noise.  In any given bin, only one of these things is visible because we measure the absolute level of the bin and this is the magnitude of the strongest of these three.  We also have no idea which of the three we are showing, generally.  If a signal is visible, you can read the power of the signal directly off of the panadapter and it is accurate within a couple of dB.  If there is noise, you can't really tell by looking and I can't tell you if you are looking at Johnson noise from the circuit or atmospheric noise.  So I can tell you the absolute power of any signals you see and I can tell you the level of the combined atmospheric and Johnson noise.

I could normalize the panadapter to the RBW of the bins.  If I did this, the noise floor would remain a constant level and you could always read the noise floor level in something like dBm/Hz, W/Hz or the more common nV/rootHz.  How would these numbers benefit you (I bet only one in a thousand of our customers really understand what a nV/rootHz measurement conveys)?  Even worse, if I normalize, now you can no longer read the signal strength of real signals -- they will vary based on the bin width.  For example a S9 signal is -73dBm.  But if we convert this into dBm/Hz and my bins are 10Hz, it will read as -83dBm/Hz.  Zoom out to 100Hz bins and it will read -93dBm/Hz.  How does this help?

Every spectrum analyzer I have used shows power just like we do.  Most people are interested in either the signal strength or the SNR and not the level of the noise -- the noise simply prevents a reading if it is above the signal level.  Have I missed something in your comment?  

I also would not convey what we are doing as a "feature."  We are simply showing the true value of the noise or signal in the bin.  To me, scaling the panadapter by the bin width to yield dBm/Hz or similar is like changing all of the gas pumps in the US to read in "dollars per liter."  Yes the information would still be factual and accurate, but few would understand what it means nor embrace the change.

Simon Lewis

learnt a lot from that - much appreciated - I noticed this curve on a few videos .. is there not a simpler cause?? resonance of the antenna for example? or we talking the small peaks and spikes in the noise baseline only .. keen to understand what we are looking at here

Steve-N5AC

Well, there are two things being discussed now.  If you look at the top panadapter and compare to the bottom, John is showing that they both show two different noise levels.  We display power in a given bandwidth so the noise floor moves as the resolution bandwidth (RBW) is changed.  This is what I cover above.  

But the **** in the top panadapter just indicates that there is more noise in some of those bins than others.  Generally when I see this it is as you say, the resonance of the antenna.  The antenna is resonant in this range and so it pull in more signals and noise and so the total power in that bin is higher than in adjacent bins where the antenna is not resonant.  It could also represent just more power in those bins -- for example there might be several strong signals that add together to raise the bin, but generally a nice shape like this is really the antenna.

DH2ID

Steve, I like your explanation about noise floor levels.
I'll take it to our club meeting this friday and explain it
in German ;-) That'll show some of these OKL (Old Knob Lovers)
73, Alex DH2ID