Comments on bolded items:
Yep, the harmonics are measurable to the limits of the Rigol's capacity. So at worse the Hantek is doing in the neighborhood of no more than 0.4%, about the 1/8 bit resolution of the Rigol. I looked at the distribution in dB too, but converted I can't go below~ -50dB. I also swept the timebase from end to end, looking for trouble. I checked views with all the various FFT windows: Hanning, Hamming, Blackman, to get the best view. The pics are just representative of the best and worst.
Although the jitter is there, I think its effect is a bit overemphasized, it only matters when circuits are sensitive or nearer the jitter's frequency, such as using it to clock a ADC. Since it appears fixed, it becomes more rate limiting at high frequencies, at 2.5 MHz the jitter represents 5ns/400ns a 1.25% variation in say, the duty cycle of a square wave clock. But at 25 MHz, its 12.5%. Can't hurt when used as a signal generator for input emulation knowing this restriction.
Formally, its effect can be thus:
![](http://www.analog-europe.com/images/01-edit-photo-uploads/2010/2010-12-december/kress_eq8.jpg)
It can drive an ADC ~ 100kHz before the noise begins to rise above the ENOB limits of 50dB.
http://www.analog-europe.com/en/understanding-ac-behaviors-of-high-speed-adcs.html?cmp_id=71&news_id=222901444&vID=35I think the jitter is a pickup of the maximum clock frequency, 5ns ~ 200MHz, as the manual states, its just an educated guess. Being the maximum DAC clock speed, the Nyquist frequency is 100 MHz, so that's the ceiling on waveform generation.
There does seem to be periodicity to the jitter, if you switch the acquisition mode to 'average' the jitter completely disappears, so its appearing at a fixed interval, like its frequency modulating the clock period.
Yes, the manual is very superficial. There is no access to adjusting the clock period directly. As you pointed out before, the software has a demo mode, why they didn't make it available for download for promotion purposes is curious, but methinks they were afraid of criticism of its dullness and thus, kill 3x25 sales.
Your AWG comments spurred me to examine it in detail. When selected, it takes the existing waveform on the screen, be it sine, sq, ramp etc., and enters editor mode. You can thus, mess up the waveform easily, as in the photo I left of a sine wave with spikes on it.
I'll report on uploading a CSV file later.
That would be around -50dB, much better than the specs, but nothing extraordinary for DDS. 0.36% is at the very limit of the 8-bit dynamic range of the Rigol, so I would be suspicious. Setting the vertical scale to dB (assuming the Rigol supports that) would make it easier to see, although it doesn't change the dynamic range obviously. I would also set the sweep speed to a higher setting to get more horizontal resolution on the FFT for looking at the first few harmonics. For more accurate measurements, something like a notch filter might help.
Glad that it appears to perform better than expected, except for the jitter issue. I wonder what kind of change takes place above 2.5MHz, different DAC frequency? 200MS/s and 4kS waveform memory means that the full memory depth can only be used up to 50kHz, so they have to be down sampled above that. Maybe there's some rounding/dithering error there? At 2.5MHz, there will only be 80 points per period, so even minor changes will be significant, especially for something like 2.6MHz, which is not a divisor of 200MHz. Any idea if there's some periodicity to the jitter? If they run the DAC at 200MHz, and alternate periods of 38 and 39, that would generate some jitter.
The manual is useless, as expected (good thing you don't need one), so it provides no clue how it changes the output frequency and what it does with the extra points.
After looking through the manual, I wonder if the arbitrary waveform capability is as bad as it looks? From what I see in the manual, you're forced to edit each individual point individually (if it has more features, it's not mentioned in the manual). I thought the Rigol software was bad, but this seems completely useless. Like Rigol, the compensate for this by leaching on the software that their competitors make available for free. Guess you can't expect them to write their own software for this kind of money.