pCompiledNumChannels error in audio plugin generation

Hi

I´m trying to use crossoverFilter function as part of a more elaborated signal process plugin, but I find the following error message when I try to generate the audio plugin:

"Failed to compute constant value for nontunable property 'pCompiledNumChannels'. In code generation,nontunable properties can only be assigned constant values".

No issues at all when I run it through audioTestBench, but it does when I try to compile.

The code is really long, but as an example, I reproduce the same with a basic code attached. So, even when this one is not useful at all, I think help to reproduce the error easily.

classdef (StrictDefaults)CrossTest < matlab.System & audioPlugin
% CrossOver plugin test
    properties        

        FCross=100;

    end
    
    properties (Constant, Hidden)
        % Define the plugin interface
        PluginInterface = audioPluginInterface( ...
            'InputChannels',2,...
            'OutputChannels',2,...
            'PluginName','CrossOver Test',...
            audioPluginParameter('FCross', ...
            'DisplayName',  'Frequency Cutoff', ...
            'Mapping', { 'int', 20, 18000}, ...
            'Style', 'rotaryknob'))

    end    
    
    properties (Access = private)    
        xFilt;
        NumCrossovers = 1;
        hpf;
        pSR;
    end
    
    methods
        % Constructor
        function plugin = CrossTest 
            
            fs = getSampleRate(plugin);
            plugin.xFilt=crossoverFilter("SampleRate",fs,"NumCrossovers",1,"CrossoverFrequencies",100,"CrossoverSlopes",6);
            plugin.hpf=dsp.HighpassFilter('FilterType','IIR','PassbandFrequency',300,'StopbandFrequency',100,'StopbandAttenuation',40);                     % <<--- Agregado
            
        end
        
        function set.FCross (plugin,val)
            plugin.FCross=val;
            plugin.crossdesign;
        end
        

    end

    methods(Access = protected)
 %% OUT
        function out = stepImpl(plugin, in)
           
            infilt=plugin.hpf(in); % Filter the input signal
            [O1,O2]=plugin.xFilt(infilt); % CrossOver the filtered Signal
              
            out=O1+O2;
          end

        function resetImpl(plugin)
            reset(plugin.xFilt);
        end
     

    end
        
    methods (Access = private)

        function crossdesign(plugin)
            plugin.xFilt.CrossoverFrequencies=plugin.FCross;

        end        
       
    end
    
end

I would really appreciate if someone give me a hint about how to overcome this.

Thanks

Pablo

Answer 1

[jibrahim]

Hi Pablo,

The corssover filter object is complaining that the number of channels is unknown. This is because the code generation compiler interprets the input to the crossover as having a varying 2nd dimension.

Modify the call to your crossover in stepImpl to this:

[O1,O2]=plugin.xFilt(infilt(:,1:2)); % CrossOver the filtered Signal

[Pablo Panitta]

Hi Jilbrahim

Thanks a lot for your prompt an effective answer.

As you said, this solve the problem. However is still not clear to me why we need to do this maneuver, since infilt is well defined at all time (128,2). Is it something related with the way we pass the info to compiler? If I hadn´t applied hp filtered before (e.g and just make a gain change of in), I woudn´t have had the error.

Thanks again!

Pablo

[jibrahim]

Hi Pablo,

This is indeed related to the way inputs are defined in the plugin.

When the plugin is generated, the input is set to have variable-size frame length (the number of rows). This is to enable the genenrated plugin for any frame size selected by the user. The number of channels (columns) is fixed, but dsp.HighpassFilter's output will be variable-sized in both dimensions, even though only the first dimension of the input is varsize. This is a limitation of the object.

This can be illustrated by an example.

Consider this function:

function [O1,O2] = foo(x)

fs = 16e3;
xFilt=crossoverFilter("SampleRate",fs,"NumCrossovers",1,"CrossoverFrequencies",100,"CrossoverSlopes",6);
hpf=dsp.HighpassFilter('FilterType','IIR','PassbandFrequency',300,'StopbandFrequency',100,'StopbandAttenuation',40);                     % <<--- Agregado

infilt=hpf(x); % Filter the input signal
[O1,O2]=xFilt(infilt(:,1:2)); % CrossOver the filtered Signal

Generate code with a varsize input (frame length varies):

x = coder.typeof(0,[1024 2],[1 0]);
codegen foo -args x -report

Hover over the infilt variable, and you will see that it is now varsize in both dimensions (the ':' before the dimension means it is not constant-size):

We can look into seeing if the output of the highpass filter (and other objects) is :1024x2 instead of :1024x:2, but this is why you are seeing this behavior.

[Pablo Panitta]

Hi jibrahim

Outsanding explanation! Now I see why pre placing the hpf caused the issue. I didn´t know about the bahaviour of this object, but now is clear!

Thanks a lot!

br

Pablo