Which Spectrum analyzer would you get?

[rf-loop]

   Gain was set to 3 for all tests.

Why? With R&S there is no preamplifier. You need use gain 0 for same.

[joeqsmith]

The HP3589a in narrow band mode.

[rf-loop]

The specs say: <-145dBm, typ. -150dBm for >10MHz  center frequency, 10Hz RBW, 1Hz VBW and 0dB attenuation.

You still have 10dB attenuation, so without that the displayed noise floor would be somewhere near -150dBm. Since you’re using 1Hz RBW, we’d expect <-155dBm - maybe there’s still some other setting left to optimize.

The 1st order dynamic range is specified as 165dB for 1Hz RBW and the 1dB compression point is +10dBm (which is 10~15dB higher than the usual candidates and this is what makes the FSE analyzers so great), which confirms that the noise floor should indeed not be higher than -155dBm in this scenario.

But even if your particular unit doesn’t quite meet its specs and the noise floor is some 5dB higher than expected and the 1st order dynamic range is “only” 160dB – don’t worry. There are still not many other analyzers that could compete.

The usual Chinese budget SA have some -5dB CP – and consequently would have to display a noise floor of -165dBm/Hz (or -170dBm/Hz to meet the R&S specs for the 1st order dynamic range) with preamp off.

Not to mention the FSEA30 phenomenal 115dB 3rd order dynamic range (above 50MHz), which is the one that really counts. And this doesn’t require the lowest possible noise floor, but exceptional linearity in the entire signal path up to the point, where the narrow band filtering occurs, which for narrow RBWs is the digital processing after the ADC in the final IF…

One problem with atten settings is that different SA have different design. (example real mixer level)
If really want compare there need be some comparable settings. If change atten of course noise floor level drops but what is noise floor  distange from useable maximum signal level without clipping or compression.   
Example If in Siglent drop attenuator to 0 there can not at all measure 0dBm "high level" signal.  There is not rules exatly for this and every manufacturer can do what they want.
Overall  FSEA30  is very good quality SA and belongs to high class. Also its new price tell something..  

[Muxr]

The specs say: <-145dBm, typ. -150dBm for >10MHz  center frequency, 10Hz RBW, 1Hz VBW and 0dB attenuation.

You still have 10dB attenuation, so without that the displayed noise floor would be somewhere near -150dBm. Since you’re using 1Hz RBW, we’d expect <-155dBm - maybe there’s still some other setting left to optimize.

The 1st order dynamic range is specified as 165dB for 1Hz RBW and the 1dB compression point is +10dBm (which is 10~15dB higher than the usual candidates and this is what makes the FSE analyzers so great), which confirms that the noise floor should indeed not be higher than -155dBm in this scenario.

But even if your particular unit doesn’t quite meet its specs and the noise floor is some 5dB higher than expected and the 1st order dynamic range is “only” 160dB – don’t worry. There are still not many other analyzers that could compete.

The usual Chinese budget SA have some -5dB CP – and consequently would have to display a noise floor of -165dBm/Hz (or -170dBm/Hz to meet the R&S specs for the 1st order dynamic range) with preamp off.

Not to mention the FSEA30 phenomenal 115dB 3rd order dynamic range (above 50MHz), which is the one that really counts. And this doesn’t require the lowest possible noise floor, but exceptional linearity in the entire signal path up to the point, where the narrow band filtering occurs, which for narrow RBWs is the digital processing after the ADC in the final IF…

Thanks for the explanation. Learning my way around it still. I did notice the mixer level has something to do with the noise floor as well. For instance if I use the low noise preset it sets it to -20 dBm. Also things are a bit better at higher frequencies.

[Performa01]

We need to distinguish between sensitivity, signal handling capability and dynamic range. The corresponding specs for an SA are DANL (Displayed Average Noise Level), 1dB CP (1dB Compression Point - the input level where the gain decreases by 1dB due to signal limiting/clipping) and 1st order dynamic range, which is usually defined as the difference between CP and DANL. But one can easily make mistakes there…

Quite obviously, we can increase the CP as much as we like by adding internal and/or external attenuation. With 20dB attenuation, the CP will increase by 20dB as well. Of course, there is only so much power a certain attenuator can handle and the internal ones will usually only tolerate up to +30dBm at best, but it depends whether these attenuators are mechanically or electronically switched (be careful, without studying the specs, +20dBm would be a reasonably safe value). With the appropriate external attenuators, we can indeed go up as high as +50dBm and even more. But the attenuator will raise the noise floor by the amount of its attenuation.

We also can increase the sensitivity close to the limits of thermal noise by means of a preamplifier, once again either internal or external. With a low noise, high gain external preamp (NF <3dB), we can theoretically make any SA perform a DANL of <-160dBm/Hz. But the preamp will lower the CP by its gain at least – some preamps might even have a lower output CP than the input CP of the SA, thus further reducing the total system CP.

So even though we can have a very low noise floor as well as a very high CP, we cannot have them both at the same time. Consequently the 1st order dynamic range is the difference between CP and DANL within one setup. The whole point of a dynamic range definition is the ability to analyze weak signals in presence of strong ones. For example, if you want to verify the harmonics of your newly designed HF-PA to be <120dBc, you either need an SA that has a dynamic range in excess of 120dB or you have to use some tricks, like using a notch filter to suppress the carrier.

To cut a long story short, it doesn’t mean a thing if someone shows the maximum sensitivity of their gear – simply because it still tells absolutely nothing about the dynamic range.

Here’s an example with the SA44.

With 0.8Hz RBW, preamp off and 10dB attenuation, the noise floor is about -133dBm, which appears at least 6dB higher than the R&S FSEA30 in the similar situation shown by Muxr in Reply #128 (SA44_Noisefloor_REF-80dBm_ATT10dB)



Still not too bad, but what is the dynamic range? What is the maximum signal it can handle, i.e. measure with reasonable accuracy? Since the reference level is set to -80dBm, we are limited to some 53dB anyway, since we don’t want to count on measurements outside the display range.

All in all, this measurement is almost meaningless for judging the capabilities of the instrument. In a situation like the one above, we would certainly use the preamplifier and the noise floor would drop by more than 20dB – and that would make perfect sense, since the preamp easily copes with signals in the range of -80dBm and lower, of course.

So we now set the reference level to +10dBm (SA44_Noisefloor_REF+10dBm_ATT10dB)



Oops – the noise floor has increased by some 12dB to -121dBm, but now at least we can hope to measure signals up to +10dBm. Let’s try -10dBm first (SA44_-10dBm_REF+10dBm_ATT10dB)



This works – I’ve just switched the vertical scale back to 10dB/div, because the noise floor isn’t visible anymore anyway, due to the strong phase noise of the analog synthesizer signal. Anyway, reference level and attenuation remain unchanged and we get an accurate measurement, so this already covers a dynamic range of  -10 - -121 = 111dB.

Now let’s try +10dBm (SA44_+10dBm_REF+10dBm_ATT10dB)



This still works, although we can see a slight hint on gain compression, as the marker reading is already 0.1dB low. This is in accordance with my previous experiences, that this analyzer has a 1dB compression point (at the mixer input) somewhere at +1.5dBm. With 10dB attenuation, it would be about +11.5dBm of course.

This gives a total first order dynamic range of +11.5 - -121 = 132.5dB.
Not too bad, but still nowhere near the R&S FSEA30.

[Muxr]

That makes perfect sense, thanks Performa01!

[rf-loop]

If I change the ref. level so that the attenuation goes down to 10dB (from 20dB):

This is how it looks with around same settings using Siglent SSA3021X. (exept that start is 0Hz and stop 20MHz (20MHz Span)
ATT 10dB, RBW=VBW=3kHz.  R&S RBW is 2kHz. This difference is around 1.8dB in noise level (R&S looks this amount better due to RBW) so in this case around nonsense because if we go more deep in various tests of course R&S is very different class SA.

[Muxr]

Just wanted to add one more thing. The reason why my measurements above were taken at -10 dB attenuation is because that's the lowest attenuation I could set with the dial. It tells me the limit was reached when I try to get lower. However I am able to punch in the 0 dB attenuation with the keypad. Just realized that (duh). I am guessing it's a safety measure. The noise floor drops further with attenuation off as expected.

[Wuerstchenhund]

Just wanted to add one more thing. The reason why my measurements above were taken at -10 dB attenuation is because that's the lowest attenuation I could set with the dial. It tells me the limit was reached when I try to get lower. However I am able to punch in the 0 dB attenuation with the keypad. Just realized that (duh). I am guessing it's a safety measure.

It is, pretty much all R&S SA's do that.

[rf-loop]

Just wanted to add one more thing. The reason why my measurements above were taken at -10 dB attenuation is because that's the lowest attenuation I could set with the dial. It tells me the limit was reached when I try to get lower. However I am able to punch in the 0 dB attenuation with the keypad. Just realized that (duh). I am guessing it's a safety measure. The noise floor drops further with attenuation off as expected.

I remeber this was same with my previous R&S also. 0dB Atten only using keyboard. And this is not alone R&S special. This is also least in some HP/Agilent models. For safety! 

But then, noise level drops but also max signal what can measure drops.





Example with Siglent SSA3000X Atten 10dB and maximum signal is around 0dBm. (or bit over)

Signal level is exactly 0dBm. (Far over screen top. Yes it is theere still in good condition. It can look adjusting Ref level. SSA3k max display window vertical range is 100dB) Peak measured, -0.27dBm.  With 10dB atten, level accuracy bit better.
RBW11kHz. Span 20MHz. Signal near 11MHz.  If I drop Atten to 8dB it start warning ADC overload and level result is bad and noise level rise lot of etc..
If I set atten to 20dB max good signal level is  +10dBm.  With Atten 10dB it can say max signal is 0dBm.
Noise level follow "exactly" Atten level, naturally.
Of course If now drop RBW also noise floor drop. If drop from 1kHz to 100Hz naturally noise floor drops 10dB but still maximum signal peak is same. And with 10Hz RBW agen 10dB more low noise level.
In image can see average noise floor around -115dBm and max signal level what can measure is 0dBm.  (RBW 1kHz)
If now drop RBW to 10Hz  then average noise floor is around -135dBm  and max signal peak is still 0dBm. But if there is signal level same as average noise level it can not measure so it can not claim at all that it can measure 135dB difference at once. But 110-120dB difference, yes.
(of course IF there is SSA3021X+ where can use 1Hz RBW it give bit more. But then ther come other limiting factor. Phase noise. With narrow spans phase noise floor is far higher than base noise floor. So, limit here, limit there and so on...