Rigol MSO5074 or Siglent SDS2104X-Plus ?

[rsjsouza]

I'd say the Siglents (especially the SDS1000X) are quite impressive for their price.

[Electro Fan]

ditto what rsjsouza said.

With the SDS1204X-E, for example, this was done with a sweep on one channel of an external generator which was output through a filter to one channel of the oscilloscope which you setup to show amplitude and phase across the designated freq span?

[Howardlong]

ditto what rsjsouza said.

With the SDS1204X-E, for example, this was done with a sweep on one channel of an external generator which was output through a filter to one channel of the oscilloscope which you setup to show amplitude and phase across the designated freq span?

AWG 50ohm output to Fixture input + ChA 1Mohm input.

Fixture output to 50 ohm thru terminator, then into ChB 1Mohm input.

MY fixture has 270 ohm series resistors on its input and output, it was originally used for testing ceramic filters which matched to that setup. The crystal filter I used here has a higher impedance, about 1.8kohm//4pF, so this test doesn't really properly show the filter's performance, I'd need to build a fixture with an appropriate pair of L filters to do that.

[Howardlong]

These are some results with a more reasonably matched fixture using the manufacturer's specs of 1800//4pF. IME it takes a bit of experimentation to get the match right, and the manufacturer's specs for SAW and crystal filters are rarely particularly accurate, I think they must go off the modelled specs, not the empirical specs. However, they are usually pretty uniform in production.

Anyway, I used L matches on both sides, the 1800//4pF gives 1458-j706 ohms @ 10.7MHz. For both the L matches, I used 4.7uH series and 43pF shunt (impedance matching calculator came up with 4.4uH & 44.9pF).

The most obvious thing to see is that the inline losses are a lot less when a reasonable match is provided.

Unfortunately, the resolution of the marker readout on the Siglent is only 100kHz which is something that a software fix could easily resolve. I didn't try the Keysight for markers.

HP 8753A VNA


DSA815+Tracking generator


SDS2054X+


MSOX3104T

[Martin72]

(..) but the MSO5000 doesn't appear to have the frequency resolution to even start measuring such a device, being limited to 300 points/decade, so completely useless for crystal measurement.

Wow, they (rigol) told me BP will come in may 2019 - You all know how long it takes in real to come at last.
And then this result..

[rf-loop]

These are some results with a more reasonably matched fixture using the manufacturer's specs of 1800//4pF. IME it takes a bit of experimentation to get the match right, and the manufacturer's specs for SAW and crystal filters are rarely particularly accurate, I think they must go off the modelled specs, not the empirical specs. However, they are usually pretty uniform in production.

Anyway, I used L matches on both sides, the 1800//4pF gives 1458-j706 ohms @ 10.7MHz. For both the L matches, I used 4.7uH series and 43pF shunt (impedance matching calculator came up with 4.4uH & 44.9pF).

The most obvious thing to see is that the inline losses are a lot less when a reasonable match is provided.

Unfortunately, the resolution of the marker readout on the Siglent is only 100kHz which is something that a software fix could easily resolve. I didn't try the Keysight for markers.

HP 8753A VNA
(Attachment Link)

DSA815+Tracking generator
(Attachment Link)

SDS2054X+
(Attachment Link)

MSOX3104T
(Attachment Link)

Yes this Siglent marker displayed resolution is poor. It is also weird why they do this. Whole all in BP is based to data table what it produce during sweep and there data is full resolution. Display draw and all displayed things are based to this primary table. And if you take also table display open in same window with draw also there data resolution is rounded for 2 decimals. But if you take data table out, there is full resolution. So why they do not just pick up more decimals from table and without any single reasons.


As can see there is lot of room in displayed data table for higher resolution what also is already available there in primary data table what can download from scope (and also upload back to scope). Or is it so that there is two programmers (one who did BP primary engine and other who then did BP display/UI things) and they did not say even hello to each others. Sorry I do not have here available any data table out from scope so I can not show this real resolution.



One question. Why you use this low freq resolution. Nearly 1.67kHz/step. (if I understand right your previous message where you tell 60 points)
When look images it looks like these stop band dips before and after pass band may be more deep if you use more resolution. Specially when look Siglent image it looks it "jumps over" this dip. It is not rejected due to dynamic rage. It looks like least 10dB is missing if not even more.  Also If look HP VNA image it is clear (but this first dip look also weird in HP VNA image, are there also lack of freq step resolution or what... ).  Overall this filter is very easy for these machines except these sharp dips (what are somehow nonsense in practice but can give some knowledge about test equipment performance.  If have better filter then start real game... and then can differentiate toys and tools and there some HP VNA wins hands down these oscilloscope bodeplot things where -80... -90dB  down from 0dBm start be impossible totally or start be limiting factor when take Siglent BP. (in around 10MHz freq)

...as can see in next OLD image where is one other typical filter freq (455kHz) tested using step attenuator.
As can see -90 still usable and then -100dBm level ref 0dBm level start be game over limit totally (noise level). Note!! In image 0dB level in scale is +17.9dBm due to reason what is not important here. (SDG1000X limit in 10MHz is +17.9dBm so I use this level what was input to step attenuator)
Step attenuator (of course HP) stepped manually during one sweep. Bit messy but it was not for beauty image but for data..

[Electro Fan]

Or is it so that there is two programmers (one who did BP primary engine and other who then did BP display/UI things) and they did not say even hello to each others.

rf-loop - you do very good work with your analysis and your annotated images are top notch, and you have a great sense of humor    

[tautech]

Some Bode plots from SDS5000X that I have no reason to believe is any different to SDS2000X Plus other than SDS5000X is running latest firmware and we are waiting for the SDS2kX Plus update.

As used before my same passive 1 KHz to 30 KHz bandpass filter with a 0 dBm stimulus from SDG1032X.
In Decade display 2pts/dec to 216pts/dec are available with sweeps very fast to some 6 1/2 minutes for max pts.

5pts/dec is the minimum displayed below as lower settings results in poor interpolation at low frequencies.
Markers set to -3 dB levels. ~6s sweep



Max 216 pts/dec ~6 1/2 minutes sweep


Configure page


Measurements page

[rf-loop]

Nice to see BP UI is bit improved (settings)

If you open dual display, trace + table do this displayed table still show these highly truncated rounded frequency values for data points.
Btw, other question, is it possible you take out table data and upload it so that we can see what is real BP engine data resolution in primary table behind displayed traces and table in SDS5k or SDS2kplus. Siglent  really need pick up these  values also to displayed table. It is important because also cursors resolution is based to this displayed table instead of true primary table. Of course it is not problem with wide span sweeps over decades but  as seen previously, utterly useless with narrow span sweeps. 

[Howardlong]

One question. Why you use this low freq resolution. Nearly 1.67kHz/step. (if I understand right your previous message where you tell 60 points)

No particular reason, the first unit I tried was the Keysight, it offered up 60 points, so I kept that setting so as to compare like for like.

The VNA and SA+TG plots have far more points, ISTR the VNA is 801 points for example.

[2N3055]

One question. Why you use this low freq resolution. Nearly 1.67kHz/step. (if I understand right your previous message where you tell 60 points)

No particular reason, the first unit I tried was the Keysight, it offered up 60 points, so I kept that setting so as to compare like for like.

The VNA and SA+TG plots have far more points, ISTR the VNA is 801 points for example.

Howard,

With do all respect, i kindly disagree, that is exactly the point, the fact that they don't compare like for like..
Advantage is the fact that on some equipment you can set it to 1000 points across very narrow frequency span, and that will give you resolution.
They will all have similar results when working with limited points, a bad one...

Also, on my Keysight I can set a span from 10,65 -10,75 MHz with 1000 points in between.. It takes 6-7 minutes to run. Step was roughly 100Hz.
Are you sure you couldn't set it to more than 60 points? What firmware are you running?
Regards,

Siniša

[rf-loop]

One question. Why you use this low freq resolution. Nearly 1.67kHz/step. (if I understand right your previous message where you tell 60 points)

No particular reason, the first unit I tried was the Keysight, it offered up 60 points, so I kept that setting so as to compare like for like.

The VNA and SA+TG plots have far more points, ISTR the VNA is 801 points for example.

Howard,

With do all respect, i kindly disagree, that is exactly the point, the fact that they don't compare like for like..
Advantage is the fact that on some equipment you can set it to 1000 points across very narrow frequency span, and that will give you resolution.
They will all have similar results when working with limited points, a bad one...

Also, on my Keysight I can set a span from 10,65 -10,75 MHz with 1000 points in between.. It takes 6-7 minutes to run. Step was roughly 100Hz.
Are you sure you couldn't set it to more than 60 points? What firmware are you running?
Regards,

Siniša

Least in 4000 series Keysight can as can see in previous video where they tell improvements in BP. It can do 1000step also with more narrow span than 100kHz. And it clearly is sign that there is demand for these better resolutions (and also for better dynamic what is totally other thing what also need know and compare.). They do not without reason just for checkbox features in shining sales brochures. There is demand and they respond. As also Siglent do in different price group.

[Howardlong]

One question. Why you use this low freq resolution. Nearly 1.67kHz/step. (if I understand right your previous message where you tell 60 points)

No particular reason, the first unit I tried was the Keysight, it offered up 60 points, so I kept that setting so as to compare like for like.

The VNA and SA+TG plots have far more points, ISTR the VNA is 801 points for example.

Howard,

With do all respect, i kindly disagree, that is exactly the point, the fact that they don't compare like for like..
Advantage is the fact that on some equipment you can set it to 1000 points across very narrow frequency span, and that will give you resolution.
They will all have similar results when working with limited points, a bad one...

Also, on my Keysight I can set a span from 10,65 -10,75 MHz with 1000 points in between.. It takes 6-7 minutes to run. Step was roughly 100Hz.
Are you sure you couldn't set it to more than 60 points? What firmware are you running?
Regards,

Siniša

I find that I am regretting sharing information again :-(

There seems to be some projection on the purpose of the tests I ran.

My test was to see if these scopes would be even capable of measuring a relatively narrow band device such as a 10MHz crystal filter, nothing more. Before running the tests, I was doubtful, so didn't see much point in even attempting to run a ten minute FRA when I could get what I was after in a minute or so.

The Keysight may well be capable of more points, I never claimed otherwise. As I already said, 60 points happened to be the default setting when I ran my first test which was on the Keysight, and I used that 60 points for the rest to compare like for like (except VNA and SA+TG).

It's not unusual to use a VNA or SA+TG to dynamically adjust filters, which typically will sweep ITRO a second or so. If you have to wait even a minute to get a response graph it's going to be a very painful time in that use case!

I've found that the problem with documenting such tests here is that different people have different demands, and there are assumptions I must have had the same demands. By publishing findings, I'm finding that I regret it because what should be a half hour or an hour's tinkering to see if something can even be achieved, I end up being pushed to go back and set up tests again and again to fit with someone else's demands, not mine, and that takes time.

[2N3055]

I find that I am regretting sharing information again :-(

There seems to be some projection on the purpose of the tests I ran.

My test was to see if these scopes would be even capable of measuring a relatively narrow band device such as a 10MHz crystal filter, nothing more. Before running the tests, I was doubtful, so didn't see much point in even attempting to run a ten minute FRA when I could get what I was after in a minute or so.

The Keysight may well be capable of more points, I never claimed otherwise. As I already said, 60 points happened to be the default setting when I ran my first test which was on the Keysight, and I used that 60 points for the rest to compare like for like (except VNA and SA+TG).

It's not unusual to use a VNA or SA+TG to dynamically adjust filters, which typically will sweep ITRO a second or so. If you have to wait even a minute to get a response graph it's going to be a very painful time in that use case!

I've found that the problem with documenting such tests here is that different people have different demands, and there are assumptions I must have had the same demands. By publishing findings, I'm finding that I regret it because what should be a half hour or an hour's tinkering to see if something can even be achieved, I end up being pushed to go back and set up tests again and again to fit with someone else's demands, not mine, and that takes time.

Howard,
I apologize if my comment was offensive or harsh to you. I must emphasise  English is not my first language, and nuances of expression are not easy.
I didn't mean to slight your effort in any way, but merely to state a difference in opinion. My comment to number of points on Keysight comes from fact that Keysight upgraded FRA last year and it wasn't as good as today..
I certainly don't want to push you to do my bidding of any sort.  I 'm sorry if it came out that way.

All the best,
Siniša

[tautech]

Nice to see BP UI is bit improved (settings)

If you open dual display, trace + table do this displayed table still show these highly truncated rounded frequency values for data points.
Btw, other question, is it possible you take out table data and upload it so that we can see what is real BP engine data resolution in primary table behind displayed traces and table in SDS5k or SDS2kplus. Siglent  really need pick up these  values also to displayed table. It is important because also cursors resolution is based to this displayed table instead of true primary table. Of course it is not problem with wide span sweeps over decades but  as seen previously, utterly useless with narrow span sweeps. 

Further info for you to inspect.....note from SDS5kX and as previous posts using 0 dB stimulus from SDG1032X for 1-30 KHz BP filter.
Points, Decade vs Linear view, decade max is 216/dec, linear max is 500 points.......with settings used for this BP filter....not yet tried other settings.
Settings = Decade, points max.

Table enabled as requested and Data file attached, remove txt extension. 
BTW, widen the csv columns and 7 digits past the decimal point are visible ! 

[mawyatt]

Looks like this is what rf-loop wanted regarding using the finer grain data for display and markers 

Hopefully will trickle down to the lower grade DSOs soon 

Think it's impressive Siglent has followed Keysight's lead and produced a useful and powerful Bode Plot feature, then expended more time and $ to expand the dynamic range (clever use of input/output scaling) and now the display. They are squeezing more useful information out of these 8 bit ADCs than I thought possible, nice work

Best,

[tautech]

Looks like this is what rf-loop wanted regarding using the finer grain data for display and markers 

He'll ask for more if he needs it. 

Hopefully will trickle down to the lower grade DSOs soon 

Typically 2kX Plus development has followed in the footsteps of 5kX yet as Performa01 has indicated in time their features will be closely aligned.

Think it's impressive Siglent has followed Keysight's lead and produced a useful and powerful Bode Plot feature, then expended more time and $ to expand the dynamic range (clever use of input/output scaling) and now the display. They are squeezing more useful information out of these 8 bit ADCs than I thought possible, nice work

Yet when BP was implemented in KS's 1k series they had a good amount of catchup to do to equal the capability of X-E models BP version 1. Interesting a DSO 1/2 the cost could do a much better job of BP.

Anyways, off to find some crystals to do some narrow span BP investigations. 

[cutthebluewire]

I have been trying to keep up with the discussion, but lack the experience to see the strengths/weaknesses most of you do. Though I understand some of the obvious bits like steps/points per span etc.

As a hobbyist the last time I did an FRA was a 7.1Mhz filter in Technical College 5 years ago, using an old HP VNA (don't know the model), as part of an assignment. I sort of get the impression that bode plot capability in scopes, though handy, (and some seem to do it better than others) if I am wanting to do more than just fiddle around I would be best served getting an actual VNA or SA + TG.

@Howardlong I for one would be saddened if you stopped sharing this sort of information. To all that take the time to produce and share such info I must say thanks, it is invaluable to the likes of me. Especially the inevitable discussions that arise, I find them very informative and view such as a learning experience. Although sometimes I am more confused than ever ,but hey.
Though it does remind me a bit of when I worked in IT and discussing the merits of the latest Cisco kit vs Juniper, and others with the various vendors. Sometimes "pistols at dawn" type stuff.

[rf-loop]

Anyways, off to find some crystals to do some narrow span BP investigations. 

SDS5kX BP UI is improved but it works as BP in all models.

For XTAL you need first build (simples possible) test fixture for it for load it right way. 
It is bit tricky to first find ok settings.
https://www.giangrandi.org/electronics/crystalfilters/xtaltest.html
No need switch, just use this 10 - 27pF series capacitor what ever you find or is suitable for your xtal. For training and demo it is not so important. With good setup you can easy get over 80 - 100dB difference between high and low peaks depending of course also used xtal.

[rf-loop]

Nice to see BP UI is bit improved (settings)

If you open dual display, trace + table do this displayed table still show these highly truncated rounded frequency values for data points.
Btw, other question, is it possible you take out table data and upload it so that we can see what is real BP engine data resolution in primary table behind displayed traces and table in SDS5k or SDS2kplus. Siglent  really need pick up these  values also to displayed table. It is important because also cursors resolution is based to this displayed table instead of true primary table. Of course it is not problem with wide span sweeps over decades but  as seen previously, utterly useless with narrow span sweeps. 

Further info for you to inspect.....note from SDS5kX and as previous posts using 0 dB stimulus from SDG1032X for 1-30 KHz BP filter.
Points, Decade vs Linear view, decade max is 216/dec, linear max is 500 points.......with settings used for this BP filter....not yet tried other settings.
Settings = Decade, points max.

Table enabled as requested and Data file attached, remove txt extension. 
BTW, widen the csv columns and 7 digits past the decimal point are visible ! 

Bode Plot engine primary work is to do primary table using every sweep step precise frequency, measured level and measured angle using full resolution from meas math.  This is all what it produce.
But because users need see something on screen this data must use for produce visible things. 
This means also that you can set what ever vertical scale for display level draw and for display angle draw. Also you can vertically shift these where ever and force it stay with these settings. If signals image go out from display no problem at all... because all is still there in primary table... you movw display and it draw from table...

Then there is table what user can take to display. This table do not show full primary table(s) resolution. Yes there is tables... if you use all three channels there is three primary tables.

Now, when user narrow or very narrow band SFRA this displayed table display resolution have been and is very poor and what more bad also cursor use this poor resolution. But behind TFT screen all time there is available this full resolution primary table what is  heart of whole BP.  Btw, if you dowload it... you can see.
If user download primary table out from scope he can analyse it more. But more. For later needs it can also load back to scope and look as what ever just done BP.

@tautech  .csv  have example about this full resolution. Here attached clipped image where can see what it include.

This displayed resolution NEED improve urgently in SDS1000 and 2000 models.
Btw I do not like how they now show level!  It is much better if tightly stay in dB and not jump to m (mdB) without real reason. Again, what you thing there in Siglent or what you smoke. -0.667 is well enough and good for readability because somewhere is it jump 10.76, 666m, 1.365, -75m, -11.57 and so on...  s t u p i d. keep it like 000.000  and same for freq, do not jump dimensions ninside one table... and keep decimal point place.

Here is primary table what BP engine do. BP Displayed all graphic and data, all is derived from this primary table. BP have resolution. More than enough.
But also users need see resolution....please do not hide it in cases when they are useful. 
Also SDS1004X-E do this same kind of full resolution proimary table as also SDS2000XPlus and other Siglent models what have BP.



And small edit for cursors data and for table. Of course not as nice as I like but l do not have image editor software, just irfan.
Keep fixed decimal, keep decimal point position, Hz, space for 000 separate...  all right alingn, so all do not jump here and there when values change in table rows...  readability, ergonomic, including visual ergonomic and it need include also colors selection... 
Perhaps they can finally read even one old HP journal before 2025. Why this learning curve is so slow, why it is so difficult... WHY?

[tautech]

Anyways, off to find some crystals to do some narrow span BP investigations. 

SDS5kX BP UI is improved but it works as BP in all models.

For XTAL you need first build (simples possible) test fixture for it for load it right way. 
It is bit tricky to first find ok settings.
https://www.giangrandi.org/electronics/crystalfilters/xtaltest.html
No need switch, just use this 10 - 27pF series capacitor what ever you find or is suitable for your xtal. For training and demo it is not so important. With good setup you can easy get over 80 - 100dB difference between high and low peaks depending of course also used xtal.

Yes, thanks for these tips. 

Some digging through my scrap bin of old PCB's found a 24 MHz ATL brand crystal and enough bits to knock together a rats nest BNC test fixture wired the same as in the link above with ~10pF and the result was certainly better than first tried by with a stimulus of 0 dB and series wiring of the crystal as in this first screenshot.
.



Done properly and capacitively coupled the result is much different. Such a simple test fixture could easily be cobbled together on some blank PCB Manhattan style or with a little Dremeling.

[rf-loop]

Anyways, off to find some crystals to do some narrow span BP investigations. 

SDS5kX BP UI is improved but it works as BP in all models.

For XTAL you need first build (simples possible) test fixture for it for load it right way. 
It is bit tricky to first find ok settings.
https://www.giangrandi.org/electronics/crystalfilters/xtaltest.html
No need switch, just use this 10 - 27pF series capacitor what ever you find or is suitable for your xtal. For training and demo it is not so important. With good setup you can easy get over 80 - 100dB difference between high and low peaks depending of course also used xtal.

Yes, thanks for these tips. 

Some digging through my scrap bin of old PCB's found a 24 MHz ATL brand crystal and enough bits to knock together a rats nest BNC test fixture wired the same as in the link above with ~10pF and the result was certainly better than first tried by with a stimulus of 0 dB and series wiring of the crystal as in this first screenshot.
.



Done properly and capacitively coupled the result is much different. Such a simple test fixture could easily be cobbled together on some blank PCB Manhattan style or with a little Dremeling.

You used 100kHz span. 500 steps responds 200Hz steps.
It is very rough step for xtals. This sharp dip down need much fine resolution...  if you use 10kHz span you get 20Hz step. Even it is lot but this xtal or setup is loose so...

Look this image what I previously show some days ago what was from test using some 84kHz xtal. There was 500Hz span and so 1Hz step and even this is bit too rough... (sharp dip is true even when I have draw it out from image area... it there is primary table and if just change display vertical position it is there or change scale.)

But as can see also in your image, this table freq resolution is total shame... what have this Siglent programmer thinking... are they there just for playing some fun like kids. Shame.  Yes they can repair this later but this must not happen even in first place if programmer have any idea what he is doing. This tell total lack knowledge, lack of enough education and total lack of enough experience with test instruments... why they do not select better workers. Perhaps they do not know how to select...  too much "child level" mistakes in every second corner. Of course this is not severe error but ... they do not think reputation damage when whole part of world are laughing. Perhaps they think fun is fun and show must go on...

Here is totally loose 3.58MHz surplus XTAL from some recycle basket.



And here is 2kHz span in use with 500 steps.  But because xtal is crap AND setup is crap, 4Hz step is enough for this.


Of course if look 100kHz trace there in your image can see small changes far before and far after resonances- these parts are mostly nonsense but what can first look and after then more narrow, depending what is doing, example if is selecting best xtals for ladder filter from big lot of candidates. With this 24MHz xtal 10kHz or max 20kHz is enough for get more resolution. Normally when we do xtal oscillators osc freq is just there between these fs and fp resonances. 
But yes nice test and show this displayed table "fun". But for show limits... there need find "thousand times" more challenging things.

[tautech]

 
Do I ever wish you lived next door !

Some further experiments with xtals applying tips from rf-loop.
Using the same 24 MHz xtal and test fixture but with SVA1032X on SA mode using TG as the stimulus. Trace A = 30Hz RBW, Trace B = 100Hz  RBW (2nd half)
Note the sweep time @ 30Hz RBW ! 



Back to Siglent's Bode plot SDS5000X/SDS2000X Plus where similar settings have been applied and many sweeps are made in the same time for ONE spectrum analyser sweep !
Mind you, the narrower RBW of the SA makes for a cleaner trace at the expense of much longer sweep times.



 
Note the horizontal scaling: 24.00, 24.00 (CF), 24.02 MHz....shaking head. 
School boy errors.

[tautech]

New SDS2000X Plus firmware released:
https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg3321796/#msg3321796

[rf-loop]

Do I ever wish you lived next door !

Some further experiments with xtals applying tips from rf-loop.
Using the same 24 MHz xtal and test fixture but with SVA1032X on SA mode using TG as the stimulus. Trace A = 30Hz RBW, Trace B = 100Hz  RBW (2nd half)
Note the sweep time @ 30Hz RBW ! 



Back to Siglent's Bode plot SDS5000X/SDS2000X Plus where similar settings have been applied and many sweeps are made in the same time for ONE spectrum analyser sweep !
Mind you, the narrower RBW of the SA makes for a cleaner trace at the expense of much longer sweep times.



 
Note the horizontal scaling: 24.00, 24.00 (CF), 24.02 MHz....shaking head. 
School boy errors.

Of course SA make more clean trace.
Think about what is difference between oscilloscope and SA.

Some SA may have something like 16 bit  ADC and scope just  8 bit. If take example 30uVrms signal...  what you see with scope...yes perhaps you can pick up it out from noise but...  then how about same signal with SA..  just easy bit of bisquit 

So with Bode Plot there need be enough signal level... much more than with SA what have lot of better sensitivity.

Look these images about SDS1000XE  BodePlotII  (can not be very big difference dynamic)



First, This two channel setuo is more suitable in some situations. Not explain more ehere.





Looking this picture can give some imagine about dynamic range. Not SDS5000 or SDS2000 but still in sama ballpark in this thing.
20MHz.
If Signal highest level to DUT (and DUT can not have amplification) is 0dBm@50ohm (!),  max single measurement dynamic range is 78dB for good accuracy. For poor accuracy (also noisy) roughly 85dB and very poor game over limit 90dB.
But if signal level is more low, all is then away from available dynamic range and vice versa if signal level is higher this can add to dynamic range. Bottom do not move!

So example if you drive xtal as DUT with very low level this fp dip goes under noise and noise can see also bit higher there... it need remember scale is Log. Zero is zero. Then 00000001 is +6dB  and 00000010 again 6dB more and 00000100 again 6dB more.  When we are near bottom...near zero... signal and noise is bit challenging... after then we have 6 bit left. 36dB.
So with 8 bit ADC it is small miracle we have nearly 100dB dynamic between noise and 0dBm. (with better accuracy 90dB dynamic in one sweep @1MHz @ 0dBm )