Siglent SDG2000X having much worse rise times than cheaper SDG1000X?

[RBBVNL9]

Hello, I recently wanted to upgrade my Siglent SDG1032X function generator to a SDG2042X. Among other things, I was hoping for higher signal fidelity. After all, the sample rate goes up from 150MSa/s to 1.2GSa/s.

Doing some quick tests, however, it seems that a standard 10 MHz block wave output is much worse on the SDG2042X. Its rise time is 8.4 ns, whereas the SDG1032X has a 3.2 nS rise time. This is measured with identical settings on the signal generators (10MHz block, 4Vpp) and used 50 Ohms input termination on the scope. I synced the clocks of the devices to get them on the same screen, but the syncing does not affect rise times. It is measured on a 300 MHz scope, which I have certified to be able to measure rise times as short as 950ps.

Checked the SDG1000X series product specifications: Lists 3.8ns rise time for 10% ~ 90%, 2.5 Vpp, 50 Ω load. And checked the SDG2000X series product specifications (sorry for the unusual link, the one of the Siglent website itself is broken :~|.
Lists 9ns rise time for 10% ~ 90%, 1 Vpp, 50ΩLoad.

So, yes, the product specs confirm that the more expensive model has like 2-3 times worse rise times than the cheaper mode… Still feel puzzled and a bit disappointed.

[tautech]

Yes SDG2000X is an older product than SDG1000X that has better HW that also supports the same max frequency for both sine and square waves.
Unfortunately to push past the limitations of these models the dearer SDG6000X series are the only existing solution until SDG7000 is released.

As always the devil is in the data sheets.

[bdunham7]

The SDG1032X has a special, specific square wave circuit that goes to 60MHz.  The SDG2042X can be hacked to a 120MHz generator (and is a great value and very useful tool overall), but the square wave frequency is limited to 25MHz.  Also, it really is a 300MSa/s AWG with a 4X oversampling output interpolation filter of some sort.  I think it's a bit deceptive for Siglent to call it 1.2MSa/s unless I'm missing some aspect.

[tautech]

The SDG1032X has a special, specific square wave circuit that goes to 60MHz.  The SDG2042X can be hacked to a 120MHz generator (and is a great value and very useful tool overall), but the square wave frequency is limited to 25MHz.  Also, it really is a 300MSa/s AWG with a 4X oversampling output interpolation filter of some sort.  I think it's a bit deceptive for Siglent to call it 1.2MSa/s unless I'm missing some aspect.

You are.
TrueArb or DDS mode.
https://int.siglent.com/u_file/document/SDG2000X_DataSheet_DS0202X-E02E.pdf

[bdunham7]

Cryptic you are.  I've not see anything indicating that the 1.2GSa/s is anything more than a 2-bit digital low pass filter.  True Arb is 75MSa/s maximum and Easy Pulse jitter reduction is something different that doesn't depend on higher sample rates. 

From the datasheet:

Sampling Rate:   1.2 GSa/s (4X Interpolation)

The only thing we don't know is the exact interpolation strategy.

[rf-loop]

Signal is produced by 300MSa/s DDS. It goes to interpolator. Result goes to DAC what use 1.2GHz clock, derived using PLL from DDS 300MHz clock.






Explained more in this Siglent app note.

[RBBVNL9]

Dear Tautech,

You are.
TrueArb or DDS mode.

You are now raising a point of confusion. On the SDG1032X, when you go to "Waveform" and "Arb", you get a menu button where you can choose between DDS and TruArb. (See attached picture.). Moreover, on Siglent web page for the SDG1000X series, they write "The proprietary EasyPulse & TrueArb technique helps to solve the weaknesses inherent in traditional DDS generators when generating pulse waveforms".

So I must assume that both the SDG1000X series and the SDG2000X series.

[AaronLee]

Yes SDG2000X is an older product than SDG1000X that has better HW that also supports the same max frequency for both sine and square waves.
Unfortunately to push past the limitations of these models the dearer SDG6000X series are the only existing solution until SDG7000 is released.

As always the devil is in the data sheets.

Ah, thanks for that information. I was planning to buy a SDG2000X series, blindly thinking it would be able to handle up to 100MHz square wave. In checking the datasheet, yep, max 25MHz. But for the SDG6032X which can do 100MHz, the price of $3400 is way beyond what I'm willing to spend. Hmm, now I need to rethink my purchase plans.

[tv84]

But for the SDG6032X which can do 100MHz, the price of $3400 is way beyond what I'm willing to spend. Hmm, now I need to rethink my purchase plans.

There are guys here that have upgraded their SDG6022X to SDG6052X...

[RBBVNL9]

Thanks for the various replies.

In their marketing communications, Siglent touted the SDG2000X series as much better as their 1000X series, but after looking at the signals it generates, and the comments above (commenting the 1000X has better hardware), it feels more like a downgrade to me. I did some further measurements, like on the harmonic distortion of a sine wave, but also there, the 2000X does not really improve upon the 1000X, at least not in the ways the plots in the marketing document show. The second and third largest harmonic tones are just as strong.

I will consider sending this unit back and reconsider my options.

Should I decide to buy an SDG6022X, is there anything I should know or be aware of before purchase?

[AaronLee]

But for the SDG6032X which can do 100MHz, the price of $3400 is way beyond what I'm willing to spend. Hmm, now I need to rethink my purchase plans.

There are guys here that have upgraded their SDG6022X to SDG6052X...

That's still $1,500 for that SDG6022X. I'll consider it, but maybe just settle for a much lower maximum square wave than what I was hoping for, and buy a SDG2000X series.

[RBBVNL9]

@AaronLee,

I don't know about your specific needs for square waves, but if the budget is lacking for an expensive general purpose frequency generator, then perhaps a separate specialist device for that square wave can meet your needs?

Maybe a far stretch, but Leo Bodnar sells very affordably 10MHz block wave generators with rise times as low as 30ps (!), reviewed in The Signal Path, episode 3138. It's based on the MAX3939, a chip often used as optical driver. I have one and its a handy device. There is another design that uses the ultra fast comparator ADCMP580 available from Analog Devices. Should create waves that are pretty squared. Of course, they do not have all the features and output settings of a regular function generator.

I've been considering using either of these two chips to build a versatile circuit that can be driven by a (lower rise time) function generator, or a very accurate time source. Or maybe modify the Leo Bodnar pulse generator I have.

[AaronLee]

@AaronLee,

I don't know about your specific needs for square waves, but if the budget is lacking for an expensive general purpose frequency generator, then perhaps a separate specialist device for that square wave can meet your needs?

Maybe a far stretch, but Leo Bodnar sells very affordably 10MHz block wave generators with rise times as low as 30ps (!), reviewed in The Signal Path, episode 3138. It's based on the MAX3939, a chip often used as optical driver. I have one and its a handy device. There is another design that uses the ultra fast comparator ADCMP580 available from Analog Devices. Should create waves that are pretty squared. Of course, they do not have all the features and output settings of a regular function generator.

I've been considering using either of these two chips to build a versatile circuit that can be driven by a (lower rise time) function generator, or a very accurate time source. Or maybe modify the Leo Bodnar pulse generator I have.

I'd already thought of that, and did some searching and found a cheap 160MHz square wave generator which will suit my purposes just fine. So I ordered that, and then I'll get a Siglent 2000X series for when I need non-square waves.

[gf]

Cryptic you are.  I've not see anything indicating that the 1.2GSa/s is anything more than a 2-bit digital low pass filter.  True Arb is 75MSa/s maximum and Easy Pulse jitter reduction is something different that doesn't depend on higher sample rates. 

From the datasheet:

Sampling Rate:   1.2 GSa/s (4X Interpolation)

The only thing we don't know is the exact interpolation strategy.

I wonder what EasyPulse actually does? Seems to me that it just reduces the bandwidth of the given waveform to < 150MHz (therefore the rise time degradation), in order that the waveform can be safely re-sampled by the DDS at 300MSa/s w/o violating the sampling theorem?

[ The traditional one-clock-period-jitter for square waves is actually an aliasing artifact. It does not happen when a DDS resamples properly band-limited signals from waveform memory. Unfortunately, a square wave is not band-limited, and a "band-limited sqaure wave" is not square any more. ]

The up-sampling to 1.2GSa/s (with a digital up-sampling filter) does of course not increase the DDS sampling rate, but it certainly makes life easier for the analog reconstruction filter, so that frequencies closer to Nyqist can be reproduced, eventually. Nyquist still remains 150MHz, though.

[bdunham7]

I wonder what EasyPulse actually does? Seems to me that it just reduces the bandwidth of the given waveform to < 150MHz (therefore the rise time degradation), in order that the waveform can be safely re-sampled by the DDS at 300MSa/s w/o violating the sampling theorem?

You've got the idea.  DDS square wave jitter occurs when calculated transition falls between two DDS samples and the square wave algorithm only allows for high and low samples.  AFAIK, these schemes for reducing DDS jitter typically just adjust the amplitude of the closest sample, or perhaps two samples, according to how close they are to the transition--a sort of time-based interpolation if you will--and then blur the whole thing by reducing the bandwidth.  In any case, the overall effect is more or less as you state--they change the targeted output waveform to a bandwidth-limited square wave and then the DDS sample rate is sufficient to reconstruct that.

The up-sampling to 1.2GSa/s (with a digital up-sampling filter) does of course not increase the DDS sampling rate, but it certainly makes life easier for the analog reconstruction filter, so that frequencies closer to Nyqist can be reproduced, eventually. Nyquist still remains 150MHz, though.

Yes, the 4X thing is just an oversampling low pass filter that allows for a steeper cutoff.  It's a good idea, but using it to claim 1.2GSa/s is a bit sketchy IMO.  CD players used to use this method because the specified analog reconstruction filter for CD audio is hard to implement in pure analog.  They didn't call it hi-rez audio or claim 176.4kSa/s.

Making clean square waves requires a separate circuit.  The SDG1032X has that, the SDG2042X does not--it's that simple.

[RBBVNL9]

Yes, the 4X thing is just an oversampling low pass filter that allows for a steeper cutoff.  It's a good idea, but using it to claim 1.2GSa/s is a bit sketchy IMO.  CD players used to use this method because the specified analog reconstruction filter for CD audio is hard to implement in pure analog.  They didn't call it hi-rez audio or claim 176.4kSa/s.

Thanks for mentioning this. When I was reading the Siglent app note posted by rf-loop above, I had the same impression, that this is a technology we have seen in CD players for many years. While the very earliest CD players, introduced in 1982, used 44.1kHz DAC followed by an analog low passfilter, a Philips publication from 1984 proposed the use of oversampling. Not before long, all major brands implemented oversampling in their commercial products. So in nothing new under the sun...

[mawyatt]

Yes, the 4X thing is just an oversampling low pass filter that allows for a steeper cutoff.  It's a good idea, but using it to claim 1.2GSa/s is a bit sketchy IMO.  CD players used to use this method because the specified analog reconstruction filter for CD audio is hard to implement in pure analog.  They didn't call it hi-rez audio or claim 176.4kSa/s.

Thanks for mentioning this. When I was reading the Siglent app note posted by rf-loop above, I had the same impression, that this is a technology we have seen in CD players for many years. While the very earliest CD players, introduced in 1982, used 44.1kHz DAC followed by an analog low passfilter, a Philips publication from 1984 proposed the use of oversampling. Not before long, all major brands implemented oversampling in their commercial products. So in nothing new under the sun...

The 1st CD players from Sony had an elegant (and expensive) passive low pass filter, this was a very complex filter that achieved ~90dB attenuation from ~20KHz passband to ~22KHz stopband which was the alias frequency. Oversampling allowed the use of a much simpler and lower cost low pass filter, which significantly lowered the CD player cost. Remember studying that Sony filter in detail, and engineering work of art. Another elegant analog filter showed up in the 1st SRI function generator.

Best,

[richmit]

There is another design that uses the ultra fast comparator ADCMP580 available from Analog Devices. Should create waves that are pretty squared. Of course, they do not have all the features and output settings of a regular function generator.

I've been considering using either of these two chips to build a versatile circuit that can be driven by a (lower rise time) function generator, or a very accurate time source. Or maybe modify the Leo Bodnar pulse generator I have.

I needed something similar, but not as fast (anything under 1ns would have done the job).  I found a DC2767A board for the LTC6754 on ebay for $7US, and had very good luck getting ~700ps pulses out of them.  I think the boards are about $25US new.

[TurboTom]

Provided the output level suits your needs, you can use any AWG and feed its square output through a bunch of LVC gates to achieve sub-nanosecond rise & fall times. The tiny 74LVC2G04 (inverter) or 2G014 (Schmitt trigger inverter) are good candidates. I used them for pulse shaping in my LPRO101 reference clock with very good results.

[RBBVNL9]

Provided the output level suits your needs, you can use any AWG and feed its square output through a bunch of LVC gates to achieve sub-nanosecond rise & fall times. The tiny 74LVC2G04 (inverter) or 2G014 (Schmitt trigger inverter) are good candidates. I used them for pulse shaping in my LPRO101 reference clock with very good results.

TurboTom, that is a nice design with inspiring parts. The output stages remind me of the Frequency Divider Design for a GPSDO by Dave Partridge (schematics). I build one of these as well (with thanks to Dave to share the Gerber files). It uses the 74AC541DW in the output stage and I measure a rise time of 1.9ns. Not as short as the sub-1ns times your 74LVC2G04 is providing, but still pretty good.

PS. Should anyone be interested in building the Frequency Divider mentioned above, I still have a number of PCBs that I am happy to send you. Having a single PCB produced did not make economic sense...
 

[RBBVNL9]

Above in this threat, people have commented that the Siglent SDG1032X has special hardware for better square waves. Great, even if it is just a sufficiently fast port after the wave generator, if it works, it works.

I thought I may validate that by comparing its square wave output with the output of the arbitrary (Arb) patterns with also have something resembling a square wave – assuming the Arb mode does not engage that special hardware. Indeed, my measurements (table below) confirm that the regular square wave produces results much shorter rise times. All good.

A surprise comes, however, when measuring the Siglent SDG2042X. Here, I expected to find similar performance for square wave and Arb modes, as there is apparently no such special circuit. But the Arb pattern output is much better than the square wave output.

This seems to suggest that of you need square waves from the SDG2042X, then not to use the square wave function, but ARB Square Duty50 instead. (At least up to the 6MHz that is supported for Arb).

[blurpy]

A surprise comes, however, when measuring the Siglent SDG2042X. Here, I expected to find similar performance for square wave and Arb modes, as there is apparently no such special circuit. But the Arb pattern output is much better than the square wave output.

This seems to suggest that of you need square waves from the SDG2042X, then not to use the square wave function, but ARB Square Duty50 instead. (At least up to the 6MHz that is supported for Arb).

That's very interesting. I was able to reproduce. I set my SDG2042X to produce Duty50 Arb on channel 1, connected to my SDS1204X-E channel 2 (pink), and channel 2 on the SDG to produce a square wave, into channel 4 (green) on the SDS. Both terminated into 50ohm.

It's clear that the arb square wave is much faster than the regular square wave.

1MHz, showing the whole wave form:



1MHz, showing just the rise:



20MHz, showing the whole wave form:



20MHz, showing just the rise:



Wondering if it's different chips being used for generating arbitrary waveforms that makes this possible? Or could it be possible for Siglent to tweak their square wave code in the firmware to get similar rise times for the regular square wave too?

[gf]

Or could it be possible for Siglent to tweak their square wave code in the firmware to get similar rise times for the regular square wave too?

I guess so, but then the resulting waveform would no longer meet the specified maximum overshoot.
It's a trade-off between rise time and ringing.

[RBBVNL9]

@blurpy, thanks for validating the measurements. Your post made me realize I made a typo in my earlier post:

This seems to suggest that if you need square waves from the SDG2042X, then not to use the square wave function, but ARB Square Duty50 instead. (At least up to the 6MHz 20 MHz that is supported for Arb). From 20 MHz to 25 MHz, regular square waves are the only option. 


@GF:

Quote from: blurpy on Today at 11:22:13 am
Or could it be possible for Siglent to tweak their square wave code in the firmware to get similar rise times for the regular square wave too?

I guess so, but then the resulting waveform would no longer meet the specified maximum overshoot.
It's a trade-off between rise time and ringing.

Yes, that trade-off is a fair point. But they may actually use the opportunities offered by Arb patterns to find the best balance between both rise time and overshoot! Such an optimum might be frequency dependent, though...

[mawyatt]

Thanks a lot

This is very useful indeed

Best,