Issue in HDL Coder

I get the following errors (shown in snapshots at the end of the post) when trying to do fixed point conversion. I'm not sure where to go from here.

I did some testing and it appears that the code that was generated doesn't even work on it's own:

>> confirming_behav

Error using movmean

Invalid data type. First input must be numeric or logical.

Error in confirming_behav>WaveletFilteredSection4 (line 54)

mean = movmean(abs((w)), fi(15, 0, 4, 0, fm));

Error in confirming_behav (line 12)

w2 = WaveletFilteredSection4(coeffs,sig);

>> confirming_behav

I think the issue is in the fixed point function but I don't know a lot about that function. I will be reading up on the documentation to see what could be done but I'd like to know how to fix this in the base HDL coder settings.

Answer 1

[Kiran Kintali]

Can you share the design, testbench and project files?

Feel free to reach out to MathWorks tech support or DM me with the reproduction steps.

[John Brooks]

Just sent you a message.

Answer 2

[Kiran Kintali]

        t = 1:10;
        x = [4 8 6 -1 -2 -3 -1 3 4 5];
        yc = movmean(x,5);
        plot(t,x,t,yc);

The movemean function does not support fixed-point inputs and hence the error input should be .

>> movmean(fi(x), 5)
Error using movmean
Invalid data type. First input must be numeric or logical.

I would consider writing a local function that replaces movemean with equivalent MATLAB code before float to fixed conversion.

In addition you have too much IO coming into the design:

Both inputs to the design 'coeffs' and 'sig' and output 'w' are "double(8192 x 1)" and eventually when converted to fixed point wordlength of 'N'; You would need 3xNx8192 pins which would violate the IO threshold on the hardware.

You might want to consider a streaming version of this algorithm if you are targeting HDL code generation.