Ola

It could be good to have a system datasheet available with all the most relevant measured data included. Like the normal datasheet from Analog Devices or TI with included typical waveform's like INL/DNL HD2/HD3, IM2/IM3 measurement's.

When switching range to 1V/div the input noise is 12.0 mV ACRMS. This is of course normal when you have to have an attenuator in the signal path.

Normally you do this with an open. When I do this with an open I have 1.55mV ACRMS. To see some problem's with on board component's in the chain you normally that with a spectrum analyzer and a known clean and filtered sine wave that have a frequency that is not an multiple of the sampling frequency (Prime frequency). By doing this you could see any unlinear or DC-DC coupling behavior in the system.

Thanks a lot for this measurement's. I have similar result's in my AD3 over the same region's that you measured. I know that the system could look a lot different than a pure DAC/ADC chip. That's why I wanted to do this measurement's. I am a chip designer myself with a loot of experience in this area. I really like the idea with the AD3 and the software and look's forward to the next generation. Hope you could increase the the sampling frequency and BW to support the next generation baseband (350MHz). Maybe you also could built in some support for different RF-BB modulation/demodulation in the software.

Sorry, in the M3 equation it should be C1 not W1. and the offset compensation is "-1.5e-3". (I did some other calculation in-between when I send the equation to you.)

In the datasheet from TI on the 14 bit DAC and ADC used in AD3 the INL and DNL is very good (< 1LSB).

You could read more about this at : https://www.analog.com/en/technical-articles/histogram-testing-determines-dnl-and-inl-errors.html

This equation for INL is an good approximation. If you want a more precise result you have to take all sample's accurate position and make a histogram calculation. The best method is to use the histogram method to a sinewave input (this is a separate discussion).

M3 is for INL and M4 is the derivate of M3 => DNL. W1 is just generated as a triangle wave with 10Hz and amp=5V. M3 is offset compensated with "-1.5e3" M3: 5*asin(sin(2*PI*10*Time))*2/PI - W1-1.5e-3 M4: if(isNaN(prev)) prev = M3; var dif = M3-prev; prev = M3; return dif;

I am not talking about discontinuity? I am doing 10 averaging and calculate the INL (Read plot) and DNL (Green plot). There are a loot of variation in the INL curve that is related to SFDR. The DNL is related to SQNR. There sesames to be something wrong also in the SFDR calculation in the measurement tool?

Could you also try try INL/DNL measurement's. As you could see the HD2 and HD3 is quite low. This is my measurement's:

I understand, Thanks again for your good answer's.

The "iterations" sesames to just repeat loop() function with a number and the iteration rate is defined in "Rate". Is there a variable that could be used to change data inside the loop() function? It sesames to be a limit on how many iteration's you could do until it fail's. In my case it is 10 with N=512 and frequency=15.625MHz and Rate=1kHz.

I just saw that the memory is 16*32k. Is it possible to decrease the number of data lines and increase the memory to the rest 3-data lines? Is this possible in software?

I increased the buffer memory to 32k and maximized the SPI frequency to 62.5MHz. In this case I could transfer 512 command's with 24 bit's without problem's. I could not understand the "Rate" variable. I guess it is related to "Iterations". I also tried this iteration's function but it failed at 5? Could you explain how many iteration's is possible depending on memory size.