fromnndata help is misleading + how to build a multi timesteps for narxnet prediction

Hi,

I have trouble understanding the help from "fromnndata".

"Here data is defined in standard neural network data cell form. Converting this data does not change it. The data consists of three time series samples of 2-element signals over 3 timesteps."

x = {rands(2,3);rands(2,3);rands(2,3)}
columnSamples = true;
cellTime = true;
[y,wasMatrix] = tonndata(x)
x2 = fromnndata(y,wasMatrix,columnSamples)

unfortunate that the rand also contains a 3 as there are 3-elements in the cell. What is the time series and timesteps? I dont quite understand the difference of time series and timesteps here?

I wanted to build a nndata object that contains 8-element signals with 3 timesteps. (ie 8x3) and 500 samples. Its a sensor that captures within a minute 3-times a signal and does that 500 times. The important part is the 3-times a signal. The 500 times should not depend on each other. so the only time dependency here is the 3 timesteps.

How would that look like?

thx