TESTS: Siglent SDG5082 Function/Arbitrary waveform generator. (+Inside images.)

[rf-loop]

New test images about SDG5082 Square wave period jitter.

Here some samples  frome new test after some test setup checking.
And it is period jitter, not cycle to cycle jitter. (ref  SiTime, SiT-AN10007 Rev. 1.0)

SDG5082 free run with its own TCXO reference.
HP oscilloscope free run with its own internal reference.

First two images with same test setup using extremely low jitter and low drift 10MHz signal using 9ms and 2ms delay from trig time position. (vertical adjusted for around sime rising angle what SDG5000 rising is nin trigger level. (reference signal is 10MHz  nearly sine)
All under 2ms delay time give same or better result than this 2ms.

Capture time in all images is 150s.  For long period times it capture less samples but in other hand, markable more than 150s capture time may lead too big amount of frequency references drifts and may give  false peak to peak.
Scope display infinite persistence.

Other images, SDG5000 CH1 (CH2  is equal) to oscilloscope using oscilloscope internal 50ohm impedance.

note about images 2 and 3. You can see in image 2 what result it give for extremely low jitter signal. (so part of jitter/drift is from HP oscilloscope with longest delay times after trigger)

Note about last image. It is one of "golden frequency". (what give extremely low jitter)

Image 4. 255us cycle time.
min-max period time 1.8ns  and  sdev 58ps.
capturing time 150s.
Cycle time is 255000ns. If time error distribution is symmetric it means 0.9ns + or - error from "ideal time".
it is around 3.5ppm from cycle time.  (0,00035%)

[rf-loop]

Small test about Arb jitter.

Signal is ARB where is 128points

high and rest of 16384 points low.
(512k memory give compatible results)

Capturing time in both pictures 10 minutes.
Scope persistence infinite.


Picture 1.

Period time is 9.999ms
max-min time (peak to peak) 6ns
stdev is 1.064ns
Stdev is high becouse distribution looks have two peaks around  +/- 1ns from ideal time.

Period time ~9999000ns
sdev is 0.11ppm
-peak to +peak is 0.6ppm (0.00006%)
If distribution is symmetric to ideal time,  peak max time error is 0.3ppm.


Picture 2.

Period time 51us
max-min time difference 2.4ns
sdev 91ps
As it can clearly see distribution is not symmetric.

[rf-loop]

More explanation for picture 1 about Arb jitter.
(other functions work different. Example if you do some pulse using Arb, it may need jumping for adjust example frequency.  It may need also +/- 2ns jumps in pulse width. 
But pulse mode works differently. You can se example pulse wide with 100ps resolution! Ands it keeps it independent of frequency. Also you can  adjust rising and falling time, agen with 100ps resolution. With low jitter and with this feature Siglents own EasyPulse technology is very nice.)


Here picture 3.

I told that jitter distribution have two peaks (with this period time!).
Here is proofment.
This picture is short time capturing so that more rare parts of distribution do not fill whope p-p jitter area. Here these two peaks can clearly see. (around double peak gaussian distribution)
Peaks are around -1ns and +1 ns from total mean value. (of course it fully explain sdev result)

It need remember this is 500MSa/s sampling  machine. And middle values between 2ns raster it need do by "hopping".  If I connect frequency counter with some reasonable gate time I get just right frequancy. 
This "distribution"  is very very different in different period times (frequencies) as can see also in picture 2. (but this same 2ns is also visible there, if capture short time, sometimes even example 30seconds, there can not see this 2ns hop for adjust average period time. If there is way for fine adjust clock instead of this hopping, it may give lot of less jitter. In this price class, I have not seen any with these all other features)
 
So it is good to understand that with this need live if use this equipment in ARB mode.

But if you overall (all waveforms) accept much more jitter, well there is also Rigol 4000 available. (ref. Rigol specs),


Picture 3. is just same all settings as picture 1. but capturing time is lot of more short.

[jpb]

I did some analysis to work out where the Rigol 500psecs or rms jitter comes from. Here I think that they carry over the surplus phase to the start of the next cycle - what I mean is if the sample step is dT and the wave length of the arb (say a square pulse) is 10.3*dT then it will do 10 samples and start the next cycle with a phase offset of 0.7*dT in the phase accumulator.

The worse case jitter is when the remainder is 0.5*dT so it starts odd cycles with 0.5*dT phase and even cycles with 0 phase so there is a peak-peak jitter (for a vertical edge) of 0.5*dT every odd cycle. This gives an rms jitter of 0.25*dT which 0.25*2nsecs = 500psecs which is what Rigol quote.

From your measurements and Siglent's specifications it looks as if they do something different as their peak-to-peak is dT rather than 0.5*dT. Perhaps they start (or have to start) each cycle from zero and to get the correct average frequency they adjust the value of the phase step per dT on different cycles. So in my example of a wavelength of 10.5*dT they would do one cycle of exactly 10*dT and the next cycle of exactly 11*dT.

[rf-loop]

I did some analysis to work out where the Rigol 500psecs or rms jitter comes from. Here I think that they carry over the surplus phase to the start of the next cycle - what I mean is if the sample step is dT and the wave length of the arb (say a square pulse) is 10.3*dT then it will do 10 samples and start the next cycle with a phase offset of 0.7*dT in the phase accumulator.

The worse case jitter is when the remainder is 0.5*dT so it starts odd cycles with 0.5*dT phase and even cycles with 0 phase so there is a peak-peak jitter (for a vertical edge) of 0.5*dT every odd cycle. This gives an rms jitter of 0.25*dT which 0.25*2nsecs = 500psecs which is what Rigol quote.

From your measurements and Siglent's specifications it looks as if they do something different as their peak-to-peak is dT rather than 0.5*dT. Perhaps they start (or have to start) each cycle from zero and to get the correct average frequency they adjust the value of the phase step per dT on different cycles. So in my example of a wavelength of 10.5*dT they would do one cycle of exactly 10*dT and the next cycle of exactly 11*dT.

In rigol specifications there read about jitter:  rms   2ppm+500ps

from 10000000ns period,  2 ppm is 20ns. 20.5ns  rms jitter is lot of.

(of course if they want they can rewrite specifications)

Have you measured what is period time jitter, example over 2-3minutes follow up.
Say example 9.999373 ms or some random other decimals for avoid "golden numbers"
Using Arb and using Square and Pulse.

rms (sdev) alone do not tell much.  (as can see in image 2)

Peak to peak period jitter, in practice in real world.
I need live within peak values, not with rms values when I need timing.
(personally I do not admire this fashion to tell rms values only in specs as we talk pulse generator  time jitter. Who have birth this fashion... )

[jpb]

I did some analysis to work out where the Rigol 500psecs or rms jitter comes from. Here I think that they carry over the surplus phase to the start of the next cycle - what I mean is if the sample step is dT and the wave length of the arb (say a square pulse) is 10.3*dT then it will do 10 samples and start the next cycle with a phase offset of 0.7*dT in the phase accumulator.

The worse case jitter is when the remainder is 0.5*dT so it starts odd cycles with 0.5*dT phase and even cycles with 0 phase so there is a peak-peak jitter (for a vertical edge) of 0.5*dT every odd cycle. This gives an rms jitter of 0.25*dT which 0.25*2nsecs = 500psecs which is what Rigol quote.

From your measurements and Siglent's specifications it looks as if they do something different as their peak-to-peak is dT rather than 0.5*dT. Perhaps they start (or have to start) each cycle from zero and to get the correct average frequency they adjust the value of the phase step per dT on different cycles. So in my example of a wavelength of 10.5*dT they would do one cycle of exactly 10*dT and the next cycle of exactly 11*dT.

In rigol specifications there read about jitter:  rms   2ppm+500ps

Have you measured what is period peak to peak jitter, example over 2-3minutes follow up.
Say example 9.99ms period.
Using Arb and using Square and Pulse.
rms (sdev) alone do not tell much.
Peak to peak, in practice in real world.

I don't have a Rigol, I was merely trying to understand where their specs come from. The 2ppm presumably comes from the 2ppm timebase accuracy. The 500psec I think comes from the way they implement the phase accumulator as analysed above.

My own position is trying to decide on a AWG or AFG hence my interest in the Siglent, the Rigol and various other ones such as the TTi TG5011, a Tabor WW5062 which I could get cheap as it is 7 years old or perhaps an Agilent 33522A from their e-bay store.

As you say, the specs are very difficult to compare, for instance a lot of Rigol's figures are given for 0dBm and it is rather unfair to compare these with Agilent figures which are probably for the full amplitude range to 10Vpp.

What you are doing with the Siglent is very useful as it is giving real measurements, unfortunately without similar measurements from the Rigol say it is still hard to compare.

So to get around this to some extent, I try to analyse what the hardware is intrinsically capable of as a starting point - or rather trying to reverse engineer what the hardware is doing from the specs. So Rigol's specs tell me they don't use a comparator whilst Siglents specs tell me that they do for both pulse and square waves.

[rf-loop]

Here is Rigol DG4000 specs about Jitter (Picture attacment).

I can not read it so that 2ppm come from frequency reference accuracy.

Firsst becouse this affect to all frequencies. Not difference if over or under 5MHz.
2ppm is 2ppm.

But it reads after Jitter(rms)

I think typically in this place we talk about cycle to cycle jitter or period jitter.

So, now I ask: What is time for get value for 2ppm. Is it period time?  If it is period time then for 10ms period it is 20ns. (still not bad for 10ms period - exept we do not know peak values)

If period is 1ms it is 2ns.  In this case 2ns + 500ps is Jitter (rms) 2.5ns and peak-peak unknown.


But then this also:
Agilent 33250A
Signal Characteristics
Squarewave
Jitter (rms)
< 2 MHz 0.01% + 525 ps
? 2 MHz 0.1% + 75 ps

And what are perhaps peak to peak values. (can we name this Agilent as jitter generator)

Last image from Siglent EasyPulse introduction "ad".

[jpb]

Here is Rigol DG4000 specs about Jitter (Picture attacment).

I can not read it so that 2ppm come from frequency reference accuracy.

Firsst becouse this affect to all frequencies. Not difference if over or under 5MHz.
2ppm is 2ppm.

But it reads after Jitter(rms)

I think typically in this place we talk about cycle to cycle jitter or period jitter.

So, now I ask: What is time for get value for 2ppm. Is it period time?  If it is period time then for 10ms period it is 20ns. (still not bad for 10ms period - exept we do not know peak values)

If period is 1ms it is 2ns.  In this case 2ns + 500ps is Jitter (rms) 2.5ns and peak-peak unknown.


But then this also:
Agilent 33250A
Signal Characteristics
Squarewave
Jitter (rms)
< 2 MHz 0.01% + 525 ps
? 2 MHz 0.1% + 75 ps

And what are perhaps peak to peak values. (can we name this Agilent as jitter generator)

That is quite an old Agilent design, though I'd guess it is what Siglent base their design on (in terms of using comparators for square and pulse - I know you don't like me suggesting Siglent do anything other than completely original desings ) which Agilent describe in quite a lot of detail in their manuals.

The Agilent jitter changes with frequency on that model because they change the waveshape used prior to the comparator at that frequency point (Agilent explain things quite well in their manual), they use sine wave above 2MHz and a different shape below (maybe nearer to square wave which then adds some jitter of its own).

With the Rigol I don't know, but at 5MHz 2ppm is only 0.4psecs so maybe they followed the Agilent lead (but without the same reason for it) and argued that above 5MHz the extra jitter would be less than 0.1% of the 500psecs so would be not noticeable. I'm pretty sure that Rigol don't use a comparitor as they have the same jitter specs for square, pulse and arb.

[rf-loop]

Yes, also SDG5000 do not use comparator method for produce Square from sinewave.
SDG1000 use this method, with known cons.

Agilent (destroyed HP), but specially original HP have published lot of details about technology and solutions. This is one reason why I admire (and own many) old HP gears. Full documents, even CLIP manuals. Also example HP journal  have been very nice.
I hope chinese manufaturer read these companies history and publications as lessons how to produce brand and trust. Now they just have started - I hope they learn. Even if equipments are cheap.  I do not know why they all afraid technical details. Copycats - no need afraid.  Perhaps sometimes but this must not be reason to total lack of documentation.  They need learn one thing, Circuits can always copy. But copying example quality - it is more difficult. Do high quality and no one can copy with reasonable price.

We need know, in test and measurement world, all errors and accuracy data - honestly.
If there is maximum peak jitter example 100ns it can write to specifications and not hide it behind some claim that jitter rms is under 200ps.
After people buy equipment and if he have wrong imagine what he buy, you have just made one more disappointed customer whos voice can heard every place in world. 
If buyer know reality, and he accept these before he buy, he is mostly satisfied. 
Chinese manufacturers, do not need shame facts. One unsatisfied customer may shut down 1000 satisfied customers voice. After then, all is loosed.
 Keep one satisfied old customer, it costs one euro. Get new customer may cost 10 euro. Get back loosed customer, it may cost 1000 or it is impossible. 

[jpb]

Yes, also SDG5000 do not use comparator method for produce Square from sinewave.
SDG1000 use this method, with known cons.

I'm a bit puzzled by that, as I thought that was how they were able to get jitter below 100psec vs 500psecs on the Rigol which uses arb for square waves.
Also their specs for square wave duty cycle look very similar to other AFGs that use comparator methods I think.

Agilent use comparators to get squares from sines on their 33250 (80MHz arb) though they change the waveform below 2MHz to get better results at lower frequency.
As you've pointed out the jitter is rather worse below 2MHz. See page 300 of the Agilent manual:

http://cp.literature.agilent.com/litweb/pdf/33250-90002.pdf
 
I know the earlier Siglent model SDG1000 had trouble doing this but they may just have improved the way they do it (e.g. not try and use sines at lower frequencies).

I realise that you've taken it apart so and have done a lot of analysis, so do you know how they generate square waves? They can't use the arb approach as the jitter
is a lot better than they get for arbitrary waves.

[rf-loop]

I know the earlier Siglent model SDG1000 had trouble doing this but they may just have improved the way they do it (e.g. not try and use sines at lower frequencies).

I realise that you've taken it apart so and have done a lot of analysis, so do you know how they generate square waves? They can't use the arb approach as the jitter
is a lot better than they get for arbitrary waves.

My comment was some amount loosy. I mean, they do not use comparator as they have done in SDG1000. This is sure.

9.999999ms period square and peak to peak around 1.5ns jitter where sdev is 66ps. Peak-peak 0.15ppm (0.000015%) and jitter rms <100ps. (0.0067ppm,  period jitter sdev (rms) 0.00000067% )
It do not need any other proofment that this can not do from this period sinewave with simple comparator. If someone can, I want see this world 8'ths miracle.

But, also how they generate now Square, I do not know enough.
It is also possible that with different time period ranges there is some changes how it produce sqr.

Due to many other busy things I have not find time for look deeper.
Of course I'm interest to do it later somehow.
Also there is one undefined IC (only this chip markings removed)  near FPGA and final DAC.
And becouse they have so much advertised this Siglent's  new EasyPulse technology this chip is perhaps related somehow to this?
 

But more than how they do it internally in circuits, I'm interest now how it works for user.  This is now priority.  Accuracy and other things related to signal quality and also some amount of UI. 

It is nice they have added push switch to this round control knob and now can jog between displayed signal parameter settings without need select with separate key..

[rf-loop]

I realise that you've taken it apart so and have done a lot of analysis, so do you know how they generate square waves? They can't use the arb approach as the jitter
is a lot better than they get for arbitrary waves.

Now it is clear.  Siglent SDG5000 produce all waveforms directly by system main DAC.
Including also Square. (I have followed signal with scope  in some different points in signal pathways, with SDG many different settings). This give very low jitter even with very long period time square waves.


What all is this Siglent "EasyPulse" technology doing, I do not know.
Also there can adjust pulse edges rise and fall time wide range (6ns - 6s with 100ps resolution) and same adjust resolution also for pulse width with independent of pulse period time (frequency) (and it do it, but I can not proof how accurate these 100ps steps are)

Main DAC is Analog Devices  AD9781, Dual 14 Bit, 500 MSa/s. (+Dual 10 bit aux dac's)
http://www.analog.com/static/imported-files/data_sheets/AD9780_9781_9783.pdf

DAC output signal go to Intersil EL5166 Current Feedback Amplifier.
http://www.intersil.com/content/dam/Intersil/documents/fn73/fn7365.pdf

Then couple of filters  (and there come some differencies between waveforms) and so on ...level control...
up to final output amplifiers (both channels 4 x Ti 3095)
http://www.ti.com/lit/ds/symlink/ths3095.pdf

Primary ("test") points where I have checked all waveforms.

[rf-loop]

Some tiny Siglent made video.

There is Rigol DG4000 and Siglent SDG5000

It demonstrate somehow what means Siglent  developed  EasyPulse technology, which has been used in SDG5000. I have not Rigol DG4000 so I can not say anything about it. But what all I have tested SDG5082 I can not see any false claim about SDG5000 itself. Rigol case, I can not take a stand.

Rigol looks gorgeous. If want decorate own lab it looks nice and "professional".  But what is inside, what is signal, it is other question.  But in other hand, Rigol have more many kind of features, so it can use also as entertainment machine and just surf and play with it.

One is sure and clear. Rigol DG4000 series do not have anything like what Siglents EasyPulse technology make possible.  Tiny example. 100ms period pulses. Pulse wide example 20ns and around 6ns rise and fall time minimum. Pulse width adjustable up to 100ps resolution and rise and fall time also adjustable up to 100ps resolution from 6ns to 6s (under 6s pulse width is limiting factor. Risetime can not be so slow that pulse top level come clipped) 





More Siglent and example Rigol videos can find here
http://www.soku.com/search_video/q_Siglent
http://www.soku.com/search_video/q_Rigol

(becouse youtube is 97.5% "trash can" (it is pity becouse there is also good valued things under this all trash.)  is blocked in China, there is others like Youku/Soku etc. )

[jpb]

From what you write it appears that Siglent do signal processing of some sort prior to the DAC, in a similar (though presumably simpler as it only applies to square and pulse) way to Agilent's TrueForm architecture. (Agilent digitally low pass filter, interpolate via a filter and then re-decimate for the DAC.)

Looking at the Siglent description :

http://www.siglent.com/en/news/detail.aspx?id=100000041467187&nodecode=119002004

they must detect when sample points fall on an edge and perform accurate edge interpolation (essentially linear interpolation). For short edges they could just do this in memory but since they allow up to 6 seconds with 100 psec resolution they couldn't simply store points in memory. Agilent I think do interpolation using a low pass filter - the sample points are padded out with zeros and then the series of pulses digitally low pass filtered to provide linear interpolation. So it is possible.

[KedasProbe]

The Rigol does have its limitation like shown in the video the min fall/rise is 1/512 of the period.
But the Rigol can generate that very low duty cycle by using the burst mode of 1 cycle.
The higher pulse frequency obviously also limits the max rise time. (min. is 7.0ns)

Example or Rigol below 10Hz (like the video) with a small spike with set fall/rise time:

[rf-loop]

Here near same with SDG5082.
Around same settings.
Pulse adjusted 100ms period, 50ns wide, risetime 30ns, fall time 8ns.
(Owon accuracy for rise/fall measurement is ?)
Acquire mode just normal.

[grego]

I'm looking forward to Dave's review of this thing as it appears to be better than the Rigol DG4000 series - for a fair amount less (e.g. I can get one shipped from China to the US for about $650).  Hopefully Dave gets to it soon as soon as he recovers from the crud he caught.

[rf-loop]

Rise times  and some fall times with all waveforms.


A  100ms period Square
B  100ns  period Square
C  500ns period Ramp with Duty 0.0%
D  500ns period Ramp rise (same dyty as before)
E  100ms period Pulse
F  50ns period Pulse
G  2.5us period Arb (16k) where is 128points high and 16256 points down.
H  500us period Arb (16k) where is build in Stair Down.
J   same as H but one step down zoomed in. (fall time there)
K  same as H but first (full) rising edge.

Jitter have handled before but here one more example mesured now agen but after some selfcal for scope.

M  Pulse 10us period. Picture and measured next rising edge after trigger (period time jitter show 25ps sdev for jitter) (infinite persistence on also for show p-p jitter)

[jpb]

There seems to be an odd asymmetry in some of the waveforms. Particularly noticeable in F and G. The rise time is 6 nsecs as specced from 10% to 90% but above 90% the curve takes a relatively long time to go flat but then drops rapidly at the end of the top flat region.

I don't have enough other comparable curves to compare. The Agilent curves shown here :

https://www.eevblog.com/forum/blog/new-agilent-scopes/645/

are hard to compare because they are thicker and to a smaller scale. They show (different) asymmetry but it is less pronounced.

[rf-loop]

There seems to be an odd asymmetry in some of the waveforms. Particularly noticeable in F and G. The rise time is 6 nsecs as specced from 10% to 90% but above 90% the curve takes a relatively long time to go flat but then drops rapidly at the end of the top flat region.

I don't have enough other comparable curves to compare. The Agilent curves shown here :

https://www.eevblog.com/forum/blog/new-agilent-scopes/645/

are hard to compare because they are thicker and to a smaller scale. They show (different) asymmetry but it is less pronounced.

I do not have comparable test picture from SDG5082.

Later, after I'm back in normal working and have time, I will make some test using same reduced rise and fall speed as Agilent have (specified nominal 8.4ns). After then we can compare agen.

(In Siglent "overshoot" is perhaps reduced too much so that situation is "undershoot" but I think this is, perhaps,  some kind of compromize with some other things)
----------


It is also good to note that Agilent (33522A discontinued model and current B models) are  totally in different price class.
----------

Siglent have 3 year warranty, Agilent only 1 year (why?). Of course there can buy 3 year option from Agilent.  There can also buy ES4 or ES5 option from Siglent (3+1y and 3+2y) and of course calibration services.

[rf-loop]

Here is fast made (holiday visit in workshop) test and adjustments so that somehow comparable with test images here:

https://www.eevblog.com/forum/blog/new-agilent-scopes/645/

Note that this scope is old museum scope and max is 1GSa/s and image is njot as fine as today scopes, but it do still works after 20 years.

Siglent Agilent difference is very clearly seen in 30MHz. I think Agilent is better reject this max frequency under 30MHz.


10MHz  here Siglent rise and fall time adjusted to around same what Agilent can do.
20MHz  here Siglent rise and fall time adjusted to around same what Agilent can do.
30MHz  Siglent
30MHz sample from  Agilent 33522Atested by "Tooms" here  (note  horizontal scale 9.5ns/div)
30MHz (same picture but stretched horizontally so that visually looks nearly same)

[grego]

I just got one of these (SDG5082) and was playing with it for a few minutes last night (had kid issues so didn't get to really poke at it yet). So far it's not bad. The square wave really starts to break down pretty quickly which was disappointing but as I mentioned I didn't get enough time yet to thoroughly test.

I will try to put something together over the coming days. Rf-loop has already been putting this thing through the ringer so I will go through some usage displays and some additional testing.

[jpb]

I just got one of these (SDG5082) and was playing with it for a few minutes last night (had kid issues so didn't get to really poke at it yet). So far it's not bad. The square wave really starts to break down pretty quickly which was disappointing but as I mentioned I didn't get enough time yet to thoroughly test.

I will try to put something together over the coming days. Rf-loop has already been putting this thing through the ringer so I will go through some usage displays and some additional testing.

That would be great. RF-Loop's testing has been pretty thorough but the more different views the better.

Oscilloscopes get a lot of attention whilst function generators much less so. It is hard to much feedback on any of the models.

[jpb]

The square wave really starts to break down pretty quickly which was disappointing but as I mentioned I didn't get enough time yet to thoroughly test.

Can you clarify this a bit? I think all function generators allow square waves up to frequencies well beyond where they look square. For instance the Agilent 33522 with 8.4nsecs rise time will only look square-ish at a period of around 6 times this (to give a bit of flat region) which corresponds to 20MHz whilst square waves are allowed up to 30MHz which is a rather pointy 4 rise times per period.
The Siglent has a rise time of 6 nsecs so things should still look square at around 28MHz but on the higher models they are allowed up to 50MHz.

I guess, since square waves consist of odd harmonics the criteria is the maximum frequency for which the third harmonic is within the pass band of the DACs image reject filter which is set at around 1/3 to  40% of the sample rate. So square wave absolute max is 11 to 13% of the sample rate. Typically though the analogue band width is much less than the image reject filter pass band.

I would say that assuming a Gaussian filter response the flat region is 0.35/rise-time and you want at least the next highest harmonic (the third) to be in the pass-band. On this basis the maximum square-squarewave frequency should be at around 0.12/rise-time which is 20MHz for the Siglent and only about 15MHz for the Agilent but this is probably a bit too conservative.

[rf-loop]

The square wave really starts to break down pretty quickly which was disappointing but as I mentioned I didn't get enough time yet to thoroughly test.

What break down and disappoint?
There is specifications. Max is 30MHz
Risetime  specified <8ns  (and as very common practice 10% to 90% of amplitude)
If risetime is in specs, it means that one cycle (10%-90%) risetimes take <16ns  and with 30MHz whole period time is 33.3ns.  Now there is left 7.65ns for both, bottom and top of square. Including these four corners 0%-10%, 90%-100%, 100%-90% and 10%-0%.  How much there is then time for "flat" bottom and top.  These corners are some amount soft but in practice where exactly it is problem exept in human eye?
It is compromise of many things. And it meets it specifications well even with these max limits settings.
 (perhaps these corners shape can quite easy get more sharp with this same rise and fall times but good question is - why,  and what are possible bad side effects)
 
If need more fast risetimes than specified <8ns then it need buy different equipment.