250 Mhz out of a Rigol DG4102 , a 100 Mhz waveform generator

[Wim13]

The Rigol waveform generator DG 4102 goes to 100 Mhz. for sinus waves
In ARB mode it is 25 Mhz, in ARB mode you can make free wave forms of 16384 points.

The DG4000 serie has a fixed 500 Mega samples/ sec and 16384 points for the waveform.
So if you want to use the full swing of 16384 points , the frequency is max 30,5 Khz.
At higher frequencies with 500 Msamp/s it has to use less points per cycle.
At 25 Mhz in ARB mode then only 20 points per cycle  ...how will this work out ?

So i made in the arb mode a waveform of 10 x sinus. and loaded it to the DG4102.
To see what will hapen..

And i got now a sinus 10 times the freq. displayed in the screen.
So 25 Mhz is then 250 Mhz. It works quite well.

If you want to try it, i attach a RAF an CSV file for the Rigol DG4000 serie
Load the RAF or CSV file in ARB mode.
 

[AndyC_772]

Funnily enough, I was trying to do something very similar on my DG4062 only a couple of weeks ago. I wanted to test out the frequency response of an amplifier up to about 150 MHz or so, and it's clear from playing with the 4062 that its frequency restrictions are purely artificial limits imposed in the UI. Even if the amplitude flatness isn't that great beyond the rated upper limit, I only needed to compare input with output to get an idea of the gain.

I was only using a few sample points, and I struggled with a discontinuity in the waveform when it wrapped round from the end of the buffer back to the beginning. You've clearly not had that problem

How did you generate a CSV file that the DG4000 will accept? There was a thread on here recently on that subject, it doesn't seem to be documented anywhere

[edit]: just seen that you've seen that thread and solved it too, nice one!

[jpb]

Rather than set the frequency to 25MHz where it will only look at a fraction 30.517578125kHz/25MHz of the points, use step-by-step mode and use 1,0,1,0,1,0 etc as the arbitrary waveform (i.e. a square wave). That way you should have more control and less jitter. (Generating 16k such points should be ok on a computer - it is not viable on the UI!)

(At 25MHz the period is 40 nsecs, with a sample rate of 500MS/s that represents only 20 points out of the 16k, i.e. 2 per sin wave but you have little control over which points are sampled.)

I should add that I don't have a Rigol DG4062 but I've studied it (alongside rival AWG/AFGs) as I'm still trying to make my mind up.

[Wim13]

How did i  make the file,.. with Rigols own software, Ultra Station.
You can edit all the 16384 points the easy way.
There are sinus waves standard in Ultra station.

You can save it as a RAF or CSV or TXT file. So i discoverd , how to make a CSV file.

@ jpb,  i can follow you calculation. but..

I looked at an analog scope, and 100 mhz looks like a nice steady sinus, there seems to be
more then just 5 bits. I can not explain yet.

[AndyC_772]

The sine waves look just fine to me, though I'd need to hook it up to a spectrum analyser to be sure. I think I know where I can borrow one.

Here's how the amplitude response of the DG4062 looks doing a linear sweep from 100kHz to 15 MHz of the 10x sine wave file. Scope is a Tek TDS754D (500 MHz bw) with 50 Ohm termination enabled and a reasonable cable:



The falling edge on ch 2 is coincident with 6 MHz, which equates to the 60 MHz that the 4062 is rated for.

[Wim13]

I have listen on a receiver for the signal at 150 Mhz, got a nice steady tone,
could not hear any jitter yet.

According to jpb calculation there should be, with so less bits.??
 
So is this correct, with 500 Mega samples, and 16384 points,
only at 30,5 Khz it can use the full swing.

At all higher frequenties the generator has to skip points, ( which ones ?? ),
So what is the use then of more space then 16k, as others make a problem of.??

For test i added RAF files of 3 , 4 and 5 times the sinus.

[jpb]

Yes, allowing the instrument to sample the 10x sine waves will also lead to sine waves as you have demonstrated and if you are doing a frequency sweep then this probably is the best way to do it. I just thought that using the step-by-step mode would give you more control, in particular on wrap around but obviously this will only work at one frequency.

Certainly at 250MHz you could only get a sine wave out as the image rejection filter after the DAC has a cut-off around 160 to 200MHz so the second harmonic at 500MHz will be very attenuated.

AndyC_772 it would be interesting to compare your sweep results using arb with the sine sweep results. This would show how much of the attenuation with frequency is compensated for with calibration when the official sine output is used.

At 150MHz the number of points will be 33 1/3 over the 10 sin waves (if my calculation is correct) so it will sample differently each set but as the 10 sines wrap around ok maybe this doesn't matter and there is no significant jitter.

[jpb]

At all higher frequenties the generator has to skip points, ( which ones ?? ),
So what is the use then of more space then 16k, as others make a problem of.??

This is the difference between a AFG such as the Rigol with a fixed clock (of 500MHz in this case) and a more expensive AWG such as the Tabor WW5061 which has a variable clock rate and always samples all the points.

For just generating different sorts of waves the Rigol system works very well. Where a true AWG is useful is when you want to have more sophisticated wave forms where perhaps you want to introduce a glitch every 10th wave to test how your circuit reacts. With a variable clock the glitch is reproduced properly, with a fixed clock you might find the glitch (if it is very narrow) is sometimes one size and sometimes another and  perhaps may disappear altogether. Another example is if you want to skip a wave every hundred normal waves - you can't do that on the Rigol while it is very straight forward with the WW5061 which allows loops etc and has 1M of memory. But the Rigol goes to much higher frequency, has 500MS/s vs 50MS/s and generally represents good value for money.

[Wim13]

@ JpB, thanks for your explanation, i will make some other files to test it later
with some nice glitches in it.. so i can see how good ( or not ) it works.

[jpb]

The sine waves look just fine to me, though I'd need to hook it up to a spectrum analyser to be sure. I think I know where I can borrow one.

Here's how the amplitude response of the DG4062 looks doing a linear sweep from 100kHz to 15 MHz of the 10x sine wave file. Scope is a Tek TDS754D (500 MHz bw) with 50 Ohm termination enabled and a reasonable cable:



The falling edge on ch 2 is coincident with 6 MHz, which equates to the 60 MHz that the 4062 is rated for.

Doing a very rough measurement on the screen it looks like 150MHz is 3dB down which makes sense if the image reject filter after the DAC has a bandwidth of around 160MHz.

So if you use the official sin wave output the amplitude is adjusted to compensate for the droop to give a flatness of  -0.8dB (according to the specs). While using the arb output is uncompensated and it shows the frequency response of the raw instrument which is interesting.

Can you repeat the experiment at higher magnitude (being careful not to exceed your scope's limit at 50ohms which I guess is 5V). Doing so will show the performance of the output amplifier. The sweep you did was at 1Vpp (I think) which doesn't require amplification and comes straight from the DAC. If you could add 2Vpp and 4.1Vpp then this would test the frequency response of the amplifier set to 4x and to 10x (I think those are the gains set - above 1V and up to 4V the gain is set to 4x, above 4V it is set to 10x).

[alm]

This is the difference between a AFG such as the Rigol with a fixed clock (of 500MHz in this case) and a more expensive AWG such as the Tabor WW5061 which has a variable clock rate and always samples all the points.

I believe Tegam used to have a nice appnote about this (emphasizing all of the virtues of AWGs). Don't have a link handy, but I'm sure it's still out there.

One weak point of AWGs is that implementing sweeping is much harder, since you need to sweep the DAC clock. They also tend to be less like general purpose function generators. I imagine filtering the signal would also be more complex, since the filter would need to adjust to the DAC clock, but I'm sure the manufacturers have figured that out.

[jpb]

This is the difference between a AFG such as the Rigol with a fixed clock (of 500MHz in this case) and a more expensive AWG such as the Tabor WW5061 which has a variable clock rate and always samples all the points.

I believe Tegam used to have a nice appnote about this (emphasizing all of the virtues of AWGs). Don't have a link handy, but I'm sure it's still out there.

One weak point of AWGs is that implementing sweeping is much harder, since you need to sweep the DAC clock. They also tend to be less like general purpose function generators. I imagine filtering the signal would also be more complex, since the filter would need to adjust to the DAC clock, but I'm sure the manufacturers have figured that out.

The Tabor instruments (and I think the TTi ones) get around this by using DDS to generate the sample clock thus allowing very rapid sweeps.

The Tegam application note you refer to is probably 401, it is quite clear :

http://www.tegam.com/HTML/Application_Notes/?AN401.html
 
A good description of the pros and cons of each (though a little dated) is given in :

http://www.edn.com/file/12978-Choosing_a_waveform_generator_The_devil_is_in_the_details.pdf

Tabor use fixed filters (25MHz and 12.5MHz) in the WW5061. Agilent in their AWGs have filters that are a fraction of the sweep rate, though the odd thing is that I think Tabor build Agilents instruments anyway :

http://www.agilent.co.uk/about/newsroom/presrel/2010/24may-em10066.html

Another example of where a proper AWG would be better is something like reproducing a 1pps signal where the AWG could have a lot of detail with very little jitter while something like the DG4062 would have a time resolution of 1sec/16k giving steps of something like 60 microseconds which wouldn't be much good for trying to reproduce say a 20 nsec rise time.

[jpb]

I have listen on a receiver for the signal at 150 Mhz, got a nice steady tone,
could not hear any jitter yet.

According to jpb calculation there should be, with so less bits.??
 
So is this correct, with 500 Mega samples, and 16384 points,
only at 30,5 Khz it can use the full swing.

At all higher frequenties the generator has to skip points, ( which ones ?? ),
So what is the use then of more space then 16k, as others make a problem of.??

For test i added RAF files of 3 , 4 and 5 times the sinus.

I think you may be confusing time resolution with amplitude resolution.

It skips points in time at higher frequencies but for sine waves the sampling rate, provided it is twice the Nyquist frequency, should be sufficient to reproduce the waveform.

Where it might be a problem is with more general, non-sinusoidal waveforms, for instance if you wanted to have a very narrow pulse represented by a single data spike then the sampling points may hit it or may miss it all together.

For smooth waveforms it shouldn't be a problem, the 150MHz filter in the instrument would filter out higher frequency components that are above the Nyquist frequency anyway - that is why the filter is there.

[Wim13]

For very low frequencies Rigol gives indeed lower rise times.

But in ARB mode you can get max rise time, i got around 3.5-3.8 nSec.
better then the specs.

To test the resolution i made an ARB file with 163 squares in 16384 point, so 10 points per square
I atttach the RAF file so others can test also 163 squares on the RG4000

On the pictures you can see the result of 163 sq on 100 Khz,( gives 16 Mhz signals )
 800 Khz( gives 130 Mhz ) and 1 Mhz ( gives 163 Mhz pulses )

Also on the first picture, you can see now clearly the jitter, which is no surprise with 163 sq.

you can see the lose of samples...

 

[jpb]

For very low frequencies Rigol gives indeed lower rise times.

But in ARB mode you can get max rise time, i got around 3.5-3.8 nSec.
better then the specs.

To test the resolution i made an ARB file with 163 squares in 16384 point, so 10 points per square
I atttach the RAF file so others can test also 163 squares on the RG4000

On the pictures you can see the result of 163 sq on 100 Khz,( gives 16 Mhz signals )
 800 Khz( gives 130 Mhz ) and 1 Mhz ( gives 163 Mhz pulses )

Also on the first picture, you can see now clearly the jitter, which is no surprise with 163 sq.

you can see the lose of samples...

Thanks for doing these experiments. I find it very interesting the way you can use the arb facility to probe the instrument.

As 163 doesn't go into 16384 exactly there will be a shift each complete arb wave so the jitter is not surprising. It would be better to divide it into
exactly 128 or 256 squares so that it would wrap around nicely.

The rise time of 3.5 to 3.8 nsecs is close to the rise time that might be associated with the 150MHz image reject filter (from AndyC_772's results there is a 3dB drop at 150MHz implying that this is the bandwidth of the filter).

0.35/150MHz = 2.3 nsecs

there must be some rise time associated with the slew of the DAC and also the output amplifier if it is used so 3.5 to 3.8 makes sense overall.

It would be really interesting to see AndyC_772's sweep experiment repeated at higher amplitude (2Vpp and 4.1Vpp) as this would give us a picture of the frequency response of the output amplifier. With a 1Vpp output I think that the amplifier is not used or is simply used as a buffer as the DAC itself can drive 20mA or so into 50 ohms.

[Wim13]

I have changed it to 128 squares in the ARB mode, see the picture,
128 sq in 16384 points, 200 Khz, gives  25,6 Mhz

You can clearly see the jitter of 2 nSec, i  think from the clock of 500 Mhz

500 Mhz is also 2 nSec, and in the picture it is perfect 2 nSec
if i change frequency the 2 nSec jitter stays the same.

The other picture shows Yellow, ARB 4 sinus in 25 Mhz = 100 Mhz
and the blue channel is 100 Mhz direct from the sinus.

[AndyC_772]

Can you repeat the experiment at higher magnitude (being careful not to exceed your scope's limit at 50ohms which I guess is 5V). Doing so will show the performance of the output amplifier. The sweep you did was at 1Vpp (I think) which doesn't require amplification and comes straight from the DAC.

That sweep was with the Rigol set at 2.0 Vpp, so with a 50 Ohm output and a 50 Ohm load, we get 1.0 Vpp at the scope input.

Here's a more complete set of measurements. First the behaviour of the built-in sine function, swept from 1 MHz to 60 MHz at a few different amplitude settings.

1.0Vpp:


4.1Vpp:



8.0Vpp:



I tried smaller amplitudes too and they're just as flat, nothing really to add by posting them.

Now the 10x sine ARB waveform swept from 100kHz to 6MHz at the same levels:

1.0Vpp:



4.1Vpp:



8Vpp:



There's a relay click from the unit when switching from 1.0Vpp down to 0.9Vpp, but no visible difference in the frequency response. The whole trace just gets a bit smaller, as expected. Zooming in shows that the peak amplitude at 60 MHz is down by about 37mV compared to 1 MHz @ 0.9Vpp, and by 45mV @ 1.0Vpp.

HF rolloff @ 0.9Vpp:



HF rolloff @ 1.0Vpp:



Now the interesting bit, what's the response like above the unit's rated maximum output frequency? Sweeps of the 10x ARB waveform from 100kHz to 15 MHz, which is the fastest it'll allow, with the falling edge on ch2 coincident with 6 MHz:

0.9Vpp:



1Vpp:



4Vpp:



4.0001Vpp:



8Vpp:

[jpb]

Can you repeat the experiment at higher magnitude (being careful not to exceed your scope's limit at 50ohms which I guess is 5V). Doing so will show the performance of the output amplifier. The sweep you did was at 1Vpp (I think) which doesn't require amplification and comes straight from the DAC.

That sweep was with the Rigol set at 2.0 Vpp, so with a 50 Ohm output and a 50 Ohm load, we get 1.0 Vpp at the scope input.

Here's a more complete set of measurements. First the behaviour of the built-in sine function, swept from 1 MHz to 60 MHz at a few different amplitude settings.

1.0Vpp:


4.1Vpp:



8.0Vpp:



I tried smaller amplitudes too and they're just as flat, nothing really to add by posting them.

Now the 10x sine ARB waveform swept from 100kHz to 6MHz at the same levels:

1.0Vpp:



4.1Vpp:



8Vpp:



There's a relay click from the unit when switching from 1.0Vpp down to 0.9Vpp, but no visible difference in the frequency response. The whole trace just gets a bit smaller, as expected. Zooming in shows that the peak amplitude at 60 MHz is down by about 37mV compared to 1 MHz @ 0.9Vpp, and by 45mV @ 1.0Vpp.

HF rolloff @ 0.9Vpp:



HF rolloff @ 1.0Vpp:



Now the interesting bit, what's the response like above the unit's rated maximum output frequency? Sweeps of the 10x ARB waveform from 100kHz to 15 MHz, which is the fastest it'll allow, with the falling edge on ch2 coincident with 6 MHz:

0.9Vpp:



1Vpp:



4Vpp:



4.0001Vpp:



8Vpp:

Thanks for doing all those - it clearly shows (on the arb version of the waveforms) where the different amplifier gains are switched in with the corresponding increase in roll off.

For sine waves the instrument does a good job at calibrating out the drop, and I presume this extends to the harmonics which can be separately specified. It would be interesting to know if the built-in arb functions (I mean the 150 menu ones) are all defined in terms of their harmonics to minimize distortion.

[KedasProbe]

Hi,

I see this in the manual for duty cycle:
<=10 MHz 20.0% to 80.0%
10 MHz-40 MHz?40.0% to 60.0%
>40 MHz?50.0% ?fixed?

Does this mean you can't generate a square wave of 1kHz with a duty cycle of 10% without 'tricks'? (I find that a bit strange)

You could also make a 'pulse' with such settings.
and PWM is possible from 0-100% for 2mHz-50kHz

Thanks

(maybe buying the DG4102)

[Wim13]

Hi,

I see this in the manual for duty cycle:
<=10 MHz 20.0% to 80.0%
10 MHz-40 MHz?40.0% to 60.0%
>40 MHz?50.0% ?fixed?

Does this mean you can't generate a square wave of 1kHz with a duty cycle of 10% without 'tricks'? (I find that a bit strange)

You could also make a 'pulse' with such settings.
and PWM is possible from 0-100% for 2mHz-50kHz

Thanks

(maybe buying the DG4102)

That is in pulse mode, but in free ARB mode you can create what you want
you have 16384 point you can free edit, at 30000 times a second, 500 Msamples/sec
So you can put 1 pixel  up and you have a duty clycle of 1/16384

[jpb]

Hi,

I see this in the manual for duty cycle:
<=10 MHz 20.0% to 80.0%
10 MHz-40 MHz?40.0% to 60.0%
>40 MHz?50.0% ?fixed?

Does this mean you can't generate a square wave of 1kHz with a duty cycle of 10% without 'tricks'? (I find that a bit strange)

You could also make a 'pulse' with such settings.
and PWM is possible from 0-100% for 2mHz-50kHz

Thanks

(maybe buying the DG4102)

That is in pulse mode, but in free ARB mode you can create what you want
you have 16384 point you can free edit, at 30000 times a second, 500 Msamples/sec
So you can put 1 pixel  up and you have a duty clycle of 1/16384

The sample rate is fixed at 500MS/s and the memory is 16384 points. If every point is sampled (point-by-point mode) then the sample memory lasts 32.768 microsecs. A 10kHz cycle corresponds to 100 micro seconds so each point would need to be repeated 3 times on most cycles with the occasional 4 times repeat (which introduces some jitter). For 10% duty cycle, as Wim13 says, you can just set 164 points at maximum and the rest at zero and set the arb frequency to 10kHz.

I don't have the Rigol but from what people who have it have posted you get faster rise times (if the amplitude is kept low) but more ringing using the arb mode. (The bandwidth/rise times depend on the amplitude as the results above demonstrate).

[KedasProbe]

Have there been firmware updates?
Are they still improving it or is it not longer updated?

about the ARB mode:
I'm not really sure yet what Rigol is doing with the given settings.
I assume I can interpret it as resampling on a different frequency?
For freq X, points Y[] gives you all the points in time, then it gets resampled at 500Ms/s
(I can reproduce that with matlab)
And in step-by-step mode a fixed freq of 500MS/s divided by 16384 is used to have an exact match with the resampling hence no jitter or skipped points.
But there are more frequencies without jitter, all multiple periods of  it.
Correct?

It's itching to place the order

edit:
Placed my order for a DG4102 (in stock) although there are things I don't like, there is nothing better out there for this price now. (968 Euro incl. VAT)

[Wim13]

Have there been firmware updates?
Are they still improving it or is it not longer updated?

about the ARB mode:
I'm not really sure yet what Rigol is doing with the given settings.
I assume I can interpret it as resampling on a different frequency?
For freq X, points Y[] gives you all the points in time, then it gets resampled at 500Ms/s
(I can reproduce that with matlab)
And in step-by-step mode a fixed freq of 500MS/s divided by 16384 is used to have an exact match with the resampling hence no jitter or skipped points.
But there are more frequencies without jitter, all multiple periods of  it.
Correct?

It's itching to place the order

edit:
Placed my order for a DG4102 (in stock) although there are things I don't like, there is nothing better out there for this price now. (968 Euro incl. VAT)

No, i have not seen any updates ....

I think your are correct, 16384 point with a sampling rate of 500 M
In arb mode you can do Piont by Point, then it selects a frequency of 30517 Hz
In higher frequencies it has to skip points of course.

There is a utility delivered called Ultra station where you can make al kinds
of waveforms, that matches the 16384 points, to avoid jitter.

But indeed you can get 2 nsec jitter, if not proper aligned

[jpb]

It's itching to place the order

edit:
Placed my order for a DG4102 (in stock) although there are things I don't like, there is nothing better out there for this price now. (968 Euro incl. VAT)

It does represent good value for money for a two channel instrument. It is good for standard waveforms and other smooth repeating waveforms at low amplitudes to high frequency and large signal to 20MHz. It is perhaps less good for arbitrary waveforms and can't do large signal waveforms above 20MHz. The specifications are good at 0dBm and a bit unknown at larger amplitudes.

I'm still trying to make up my mind on it.

[KedasProbe]

Received the package today, I will 'play' with it this evening.
Date of calibration 19-Oct-2012 according to the document, not really recent/new. (6 months)

Edit:
Great 'toy'  rise and fall time is as expected.
I like it that you can set your number format to whatever number format you prefer.
FYI version numbers DG4102:
Software: 1.04
FPGA: 1.07
Hardware: 1.01
Keyboard: 4.01