Efficient decimation with Embedded Coder

I have the following code as part of a Matlab Function block in Simulink/Embedded Coder:

    [dcsDec, state.dcsDec1FiltState] = filter(state.dcsDec1Filt, 1, state.phaseDiffBuf(1:numSampsToProcess), state.dcsDec1FiltState);
    dcsDec = dcsDec(state.dcsDec1Factor:state.dcsDec1Factor:end);
    [dcsDec, state.dcsDec2FiltState] = filter(state.dcsDec2Filt, 1, dcsDec, state.dcsDec2FiltState);
    dcsDec = dcsDec(state.dcsDec2Factor:state.dcsDec2Factor:end);
    [dcsDec, state.dcsDec3FiltState] = filter(state.dcsDec3Filt, 1, dcsDec, state.dcsDec3FiltState);
    dcsDec = dcsDec(state.dcsDec3Factor:state.dcsDec3Factor:end);
    state.dcsPhaseDiffBuf(state.dcsPhaseDiffBufLen+1:state.dcsPhaseDiffBufLen+length(dcsDec)) = dcsDec;
    state.dcsPhaseDiffBufLen = state.dcsPhaseDiffBufLen + length(dcsDec);

I apologize for the ugly indexing, but the gist of it is that it is doing standard FIR filtering on a signal and then decimating it. The same buffer is reused to minimize the memory usage. The problem is that the D_Work global struct that Embedded Coder generates includes the following:

real_T dcsDec_data[1331];
real_T b_dcsDec_data[1328];
real_T c_dcsDec_data[1329];
real_T d_dcsDec_data[1330];
real_T c_dcsDec_data_m[1330];
real_T b_dcsDec_data_c[1329];
real_T state_data[1328];
real_T b_dcsDec_data_k[1331];

There are no other references to "dcsDec" in the Simulink model or Matlab code, so it appears to just be creating a new array for each iteration, without even recognizing that the arrays should be getting smaller. There has to be a better way to do this. What is the "right" way to do this with Embedded Coder?

Answer 1

[Fred Smith]

If you could make all occurrences of dcsDec the same size then you would only get one variable.

If that is not possible, you could use variable-sizing. Use coder.varsize('dcsDec',[1 1331]), and that should also force a single variable dcsDec but it will carry run-time size information.

Hope one of these two ideas helps.

-fred

[Jim Clay]

The coder.varsize method worked very nicely. Thank you.