Help request with Time Series Analysis: TSA ARMA Modelling, Prediction, and length of the input time series

[)]

Ladies and/or Gentlemen,

I would like to ask your help and advice with the following: I am trying to use the Time Series Analysis toolkit TSA ARMA Modelling, and TSA ARMA Prediction VIs (LabVIEW 2010 Pro SP1 x86 on a W7 x64 system) as a simple, initial exploration version of my prediction algorithm. I have removed the decoupled order estimation, for now, that is intended to estimate optimal (?) AR orders for individual vectors from 2 dimensional data. These vectors are identical in length, which length will be controlled programmatically later, alongside the predictive (positive) horizon (within mathematically constrained bounds of course).

Having checked (hoping I won't need to post a question as embarrasingly novice as it still is) the input data at each subVI I have designed, including the predictor one of course, I consistently recieve the following error message:

<error number, error propagator subVI chain>

Possible reason(s):

Time Series Analysis Toolkit: The input time series is empty or the length of the input time series must be greater than the required length.

I have double checked everything again, and I am 99.99% sure that the input time series is neither empty, nor short. For now, I am testing with data generated in real-time (wel, as real as it gets), using a 30 step negative, with a 5 step ahead (positive) horizon, using a 30 by 3 vertical array that is DEFINITELY ensured to be consistent no matter what. As for initialisation, I am feeding zeros of the same dimensions. Dimensions are of course mathematically definitely feasible, as well as I have done this in Matlab/Simulink before, but with this one, I have to admit I seem to be stuck. Data comes from sensors, fed to MathScript RT, where processing is done using matrices, then results are used in another subVI with feedback nodes to generate consistent time series data (several error correcting measures are used, actual sensor data is only propagated to System Identification if confirmed to be fully consistent, otherwise simulated - for now, by just padding with zeros, until the key issue is fixed, this also makes certain resultant phenomena much easier to observe).

I have tried, so far, partially to explore LabVIEW features along these lines:

- a range or AR / MA order pairs according to application and docs

- transposing the input array, just in case if time series are wrongly interpreted (found no documentation on MIMO TSA array alignment requirements)

- generating estimated optimal orders automatically using TSA AR Modelling Order (incl. for each vector, the averaging and rounding them to be used as global AR orders)

- using waveform instead of arrays (the source VI has been modified to be capable of generating both)

- moving on to System Identification Toolkit (problem is the same, so I must be missing something very basic and embarrassing)

I could code an AR/ARMA estimation algorithm analytically from scratch, either within LabVIEW or using the MathScript RT Node, which probably both are good exercises anyway, but at the same time, I definitely would like to explore the use of LabVIEW Time Series Analysis and System Identification in a practical real-time context. For that reason, I don't want to switch to manual, especially if that would probably result in a less efficient subsystem at the expense of much more work that has been done before me countless times anyway, and hence offers nothing new.

If you have the time and insight, I would greatly appreciate your help or suggesion with this input time series length error issue, but in any case, thanks for your interest and reading the above.

[)]

Sorted, should have checked algorithm implementation first. Sorry. Stupid mistake.