Sample Rate of a Custom Wavegen Waveform on a ADP3450

[JimR2]

I'm curious how a Custom waveform which, in this case I generate in Excel and copy-paste into the Sample area of the Custom-Edit window is then handled when I use it to generate a waveform, such as with the below two-tone table with 3 and 7 cycles over the 4096 table entries.

If I set the Wavegen to 1MHz, it would appear that the rate the 4096 sample table is repeating is 1MHz, which would produce a 3MHz and 7MHz tone as seen in the 3rd image. But what I don't understand is how the sample-rate could be 4.096GHz? I thought the Wavegen DAC is limited to 125MSPS? 

What am I missing here? 

 

Thanks, JimR2

 

[zygot]

It's probably reasonable to suspect that the ADC and DAC channels of the AD3PXXXX instruments use circuitry similar to the ZMODs that Digilent sells. Assuming that this is true, the analog bandwidth of the ADC channels is 70 MHz +/- 3 db, 30 MHz +/- 0.5 dB, 20 MHz +/- 0.1 dB; and the DAC analog bandwidth is 40 MHz +/- 3 dB, 20 MHz +/- 0.5 dB, 14 MHz +/- 0.1 dB. Regardless of the sampling rate presented to you these figures never change. Analog bandwidth is determined by the analog circuitry before a signal gets to an ADC and the analog circuitry between a DAC and the physical connector. Likely, unless you are using an external sampling clock, the actual sampling rate is usually going to be 125 MHz, which is the maximum rate that the converters are capable of. Usually, it pays to decimate high Fs data rather than change the rate at the sampler ( either ADC or DAC ). It should be noted that not all samplers are the same. It's a certainty that not all bits coming out of a 14-bit signed ADC are accurate representations of the actual input signal.

So what's with sampling rates that exceed 125 MHz? Likely this is equivalent time sampling. This is a mode where user gets to pretend that the instrument can perform at a higher sampling rate than possible. Equivalent time sampling can be done in a number of ways. While not necessarily useless, it needs to be understood for what it is. The user is generally responsible for determining exactly what "what it is" means. So what does a signal that's band-width limited to well below the sampling rate mean? It does bring up a good point for thought though.

Whether your test equipment is a $100K oscilloscope with 25 GHz samplers or a low end instrument with 125 MHz samplers, what you see is not necessarily what you get. On, most oscilloscopes you have the choice of how the data is rendered on a display. Usually this is a sin(x)/x representation. It's often possible to view data samples as dots, which represents one sample. Even this is a rendering as the display is likely not capable of showing 1 lsb in a single pixel. The same is true on the horizontal (time) scale. So what you are looking at in your 3rd picture is not exactly the data but a representative rendering of data. Sometimes a sin(x)/x rendering of data is appropriate for interpreting data renderings as in the case where you want o know if the resultant two-toned waveform more or less lokks like what you are expecting. Sometimes a non-interpolated rendering , like the dot display is more useful. It all depends on what you are looking for. Edited December 27, 2022 by zygot

[attila]

Hi @JimR2

Yes. The system (Scope-ADC, Wavegen-DAC, Digital In/Out) frequency is fixed, by default 100MHz but it is adjustable (50-125MHz) under Settings/Options/Device.
The Wavegen can use fractional sample increments, and hold samples for multiple DAC conversion (for low freq signals) or skip samples (for high freq signals). So, the 4.096GHz sample rate is kind of virtual, which for 100MHz system frequency it means the step increment is 40.96

[JimR2]

Thanks zygot and attila, that is what I suspected, but had to ask.

attila, from your explanation this implies the data table provided (say in my case I chose 4096K configuration of a two-tone signal), the data is resampled to scale to whatever frequency is selected in the wavegen? This may explain why I can achieve lower harmonic content using a Custom waveform generated in excel, and copy pasted in, vs building it using the "alter with" and sine functions?

 

Thanks, JimR2

[zygot]

It might be informative to connect your DAC outputs to a really good oscilloscope or network analyzer rather than loop it through the ADC ZMOD and analyze the results that way. A good digital scope these days has a pretty deep single-shot sample buffer.

[zygot]

Whatever a vendor calls it, when data is presented as sampled at a rate higher than is possible with the hardware, you need to keep one thing in mind. The adjacent samples are not related in time or phase. The idea of time equivalent sampling is that the ADC signal being digitized repeats itself over some time period. So, if you have a two tone signal that repeats at some some rate there will be the assumption of 0 offset drift or scale change over many of these periods and a lot more Fs periods. How reasonable this is depends on what you are doing. There are a lot of ways to draw inappropriate inferences from digitized sample data if you aren't careful, especially isolated blocks of contiguous data samples captured relative to some trigger event that might not be quite what you think that it is. Sometimes you can get away with misinterpreting data and sometimes you can't. Edited January 2, 2023 by zygot

[attila]

Hi @JimR2

Yes, the data is sampled, step adjusted based on the set frequency.
The imported, pasted and created data with the editor (fucn, math, alter...) are handled the same way, should generate similar output.

[JimR2]

Thanks for the detailed explanations zygot and attila, they made it much clearer for me what's going on, and where the sample-rate relationship comes from!

 

Regards, JimR2