I'm trying to do simple averaging in Simulink and need some help with it.

Assume AI'm acquiring a noisy signal at 1 mS interval and I want to get 1 point every 50 mS. I would like to collect the data for the 50 mS and spit out an average for logging. I'm not sure how to do it.

Any simple solutions would be appreciated.

Answer 1

[Alexander Bottema]

function y = my_average(u) %# codegen
persistent i buf bufSum;
if isempty(buf)
    buf = zeros(1,50);
    bufSum = 0;
    i = 1;
end
bufSum = bufSum - buf(i);
buf(i) = u;
bufSum = bufSum + u;
i = i + 1;
if i > numel(buf)
    i = 1;
end
y = bufSum / numel(buf);

[Guillermo]

Thank you, I will try to make this a sub-system and I hope it works in simulink...

[Guillermo]

It seems to have worked. It generates a warning at every iteration but I can live with that.

Thank you,

Answer 2

[Fangjun Jiang]

Use the "Weighted Moving Average" block from Simulink>Discrete, click help for details.

[Guillermo]

I don't have a Weighted Moving Average" block. I tried to create one with the following:

****************************************************

function Vector_out = fcn(Raw_In, Vector_in)

%#codegen

Vector_in = circshift(Vector_in, [1,1]);

Vector_in(1) = Raw_In;

Vector_out(:) = Vector_in(:);

******************************************************

Vector_in is initialized by a data store block

and then Vector_out would be summed with the Sum of elements block then divided by the dimension. In this way I would have a moving average which is more desirable but a block average would do too.

If this sparks ideas let me know. It didn't work for me because I could not use the data store block to define the vector_in or Out I don't recall which.

[Fangjun Jiang]

Hah, it is obsolete! http://www.mathworks.com/help/toolbox/simulink/ref_obsolete_blocks/weightedmovingaverageobsolete.html

Use the new Discrete FIR Filter, http://www.mathworks.com/help/toolbox/simulink/slref/discretefirfilter.html

Answer 3

[Walter Roberson]

Feed the signal in to an integration block that is reset to 0 every 50 samples; take the 50th output and divide it by 50 to get the average for that period.

There might be a cleaner way to only trigger the output collection every 50 outputs and tie that in to the integration block being cleared right afterwards.

[Guillermo]

I would like to try this, but I'm not yet very familiar with simulink. How do it set up the integration block?

The initial condition is set to 0, how about the saturation limits? and the Ignore limit and reset when linearizing check box?

Thanks again,

Answer 4

[Guillermo]

I got my code to work too, thanks to the sample by Alexander. The two codes give the same result and I'm not sure either which is more efficient. I tried passing the number of elements n to acquire so that I can pass a number of points to average but got an error...

function y = fcn(u)
%#codegen
persistent i n buf;

if isempty (buf)
    i = 0;
    n = 50;
    buf = NaN(50,1);
end
buf = circshift(buf, [1,1]);
buf(1) = u;
n = numel(~isnan(buf));
y = sum(buf)/n;

Answer 5

[Martin Murtagh]

I have replaced the moving average block in my models with the subsystem shown. By setting the delay length to the moving average value divided by the time-step and the gain value to one over the moving average value, this subsystem will calcuate the moving average in the same way as the moving average function with a sliding window.