How to code a moving average in a matlab function block using persistent variables

Hi all,

I'm trying to code a moving average in a MATLAB function block (I don't have access to the DSP toolbox). The block needs to take 1 input signal and then release a single output. However, the block is within a subsystem that is part of a closed loop system that will receive both new inputs, as well as 'recycled' inputs.

I've tried to declare a persistent variable that stores the inputs and then calculates a moving average to output a single signal every time a new 'slug' of material passes through the subsystem - BUT I'm not getting anywhere fast (I'm new to MATLAB/Simulink) and would greatly appreciate someone's help!

I've tried something along the lines of:

function Aprime = fcn(A)

persistent A_array
if isempty(A_array)
    A_array = zeros([],1);  % this trying to create an empty array to append/concatenate each new value of A \
    %  that passes through the block
end

A_array = [A_array, A];

Aprime = movmean(A_array, 3)  % the '3' is arbitrary at this point.

end

This block of code clearly does not function, and I'm getting errors with persistent variable assignment orders, left side array does not equal right side, etc., etc. depending on what combination of things I try.

While I've tried to use the 'movmean' built-in function here just to get me started, it doesn't really capture what I need either, as I need just one mean as an output signal each time a slug of material flows through the block.

I think maybe something along the lines of declaring an persistent index ('idx') variable and 'sliding' the window by a tunable size (user-defined) would be best, but I'm really not sure how to go about this.

Anybody's help would be UBER appreciated!

Thanks in advance.

Cheers

[Rik]

Did you complete a basic Matlab tutorial like the Onramp tutorial (which is provided for free by Mathworks)? It will get you most of the way there. The idea you have is sound (if your constraints are described correctly).

[Ryan Wilson]

Hi Rik,

Yes, I have gone through many of those tutorials, in addition to a 2-day course put on by Mathworks - while interesting for many applications and a good starting point, unfortunately I did not find these helpful with what I'm trying to do at the moment.

Thanks for your positive feedback on the underlying concept I'm trying to work out however. If you have any further details for my code that could help out, I'd appreciate it!

Cheers

[Rik]

Which tutorial taught you to pre-allocate an array like this?

What I would suggest is to create an array with NaN. Then you can use circshift to move the vector 1 element. Then replace the last value with the input scalar before calling movmean.

Note that this assumes the input is scalar and the output should be a vector of a given size.

[Walter Roberson]

Make the vector the size of the window.

Vector = [Vector(2:end),New];

emit mean(Vector)

However you need to decide on your boundary conditions when a full window has not been seen yet.

[Ryan Wilson]

I was finally able to come up with something that works using a combination of both of your comments.

Thanks

[Rik]

Feel free to explain what you did in an answer. That way, future readers will be able to use your solution.

Answer 1

[Ryan Wilson]

Using a combination of advice from Rik and Walter Roberson, here's what I ultimately ended up doing to code a simple moving average (sliding window) for the purposes of my model (which functions in a closed loop circuit):

function A_out = fcn(A, flowParams) % movAvgWindow comes from\
% myParams structure in base workspace.

% First the variables are declared.
persistent array n

% During the first call of the function, variables are initialized.
if isempty(n)
    array = zeros(1, myParams.movAvgWindow);
    n = 1;
end

% The following loop maintains the values needed by moving window.
% Notice that if n < window then nothing needs to be done.
% This for loop also sums up the values of the array.
sum = 0;
for i = 1:(n-1)
    if n >= myParams.movAvgWindow
        array(i) = array(i+1);
    end
    sum = sum + array(i+1);
end

% The last position is updated based on the most recent input.
% A running total is also captured for averaging purposes.
array(n) = A;
sum = sum + A;

% This is the value to be returned.
if n >= myParams.movAvgWindow
    A_out = sum / myParams.movAvgWindow;
else
    A_out = sum / n;
end

% Finally, variable 'n' is updated.
if n >= myParams.movAvgWindow
    n = myParams.movAvgWindow - 1;
end
n = n + 1;
end

I'd like to note that I tested it out against the 'Moving Average' block from the DSP toolbox and it appears to function pretty well identically. A possible difference is in the handling of the current index: my code always sets the window backward from the current index; I believe the movmean function in matlab does the same for even numbered windows, but straddles the present index for odd numbered windows.

Let me know if anyone else has any thoughts on how to improve.

Cheers