'Could not determine the size of this expression' in simulink function block

When running my Simulink simulation i am faced with an error with regards to the code in my function block. The error is:

Error:Could not determine the size of this expression.

Function 'Model Predictive Control ' (#52.3574.3583), line 94, column 49:

"1:i_value"

In the first snippet the phase in which MATLAB says is causing the error is the underlined part: 'H(i_value,1:i_value)'

The loop of code in which this error is highlighted is as follows:

i=1; %i is the level

while (i<=p)

minK = min(Kbest, 2^(i-1));

for j = 1:2:minK*2

for C = u_min:u_max

i_value = i;

Ucand(i,j+C) = C;

Ucand(p+1,j+C) = (H(i_value,1:i_value)*Ucand(1:i_value,j+C) - U_bar_unc(i_value))^2 + Ucand(p+1,j+C);

Niter=Niter+1;

end

end

Ucand = sortrows(Ucand',p+1)';

minK = min(Kbest, 2^i);

l = 1; j = 1;

while j <= minK*2

if Ucand(p+1,l) > 0

Ucand_aux(:,j+0) = Ucand(:,l);

Ucand_aux(:,j+1) = Ucand(:,l);

j = j + 2;

end

l = l + 1;

end

Ucand = Ucand_aux;

i=i+1;

end

The matrix H is of size (6,6) and the value of p is 6. This means the while loops is for values of i from 1:6 so the indexing doesnt exceed the limits of the matrices, so i am unsure as to the problem.

The rest of the code is as follows (to clarify the variables and matrices used within the loop):

function [U_op,Niter,Rs2_o]=MPCFunctionDraft(Vc,i_filter,vref,Vdc,Ts,Lfilter,Cfilter, u_min, u_max, p, Kbest, A, B, D, Np, lambda)

% In order to call "cont2dis" function

coder.extrinsic('cont2dis');

coder.extrinsic('MPC_Matrices_l');

persistent Aq Bq Bdq Vc1 Vc_old if_old UpsilonT Gamma lSTE W_inv H Rr

% Initialize variables and get the parameters of the Discrete model based on Ts

if isempty(Vc1), Vc1 = 0 + 1j*0; end

if isempty(Vc_old), Vc_old = 0 + 1j*0; end

if isempty(if_old), if_old = 0 + 1j*0; end

if isempty(Aq) && isempty(Bq) && isempty(Bdq)

Aq = zeros(2); Bq = zeros(2,1); Bdq = zeros(2,1);

% Call "cont2dis" function

[Aq, Bq, Bdq] = cont2dis(Ts,Lfilter,Cfilter);

end

if isempty(UpsilonT) && isempty(Gamma) && isempty(lSTE) && isempty(W_inv) && isempty(H) && isempty (Rr)

UpsilonT = zeros(6,4); Gamma = zeros(4,6); lSTE = zeros(6,3); W_inv = zeros(6,6); H = zeros(6,6); Rr = zeros(2,2);

% Call "MPC_Matrices_l" function

[UpsilonT,Gamma,lSTE,W_inv,H] = MPC_Matrices_l(A,B,D,Np,lambda);

end

% To generate the switching states matrix

states = [1 0 0;

1 1 0

0 1 0

0 1 1

0 0 1

1 0 1

1 1 1];

a = exp(1j*(2*pi/3));

% Read current measurements at sampling instant k

ifk = i_filter;

Vck = Vc;

% Equation of io(k-1) = if(k-1)-(c/Ts)(vc(k)-vc(k-1)).

io1 = if_old - (Cfilter/Ts)*(Vck - Vc_old);

% Store the measured current for the next iteration

if_old = ifk;

Vc_old = Vck;

%% Vo Reference

Vo_xy_ref = [vref; vref]; % Vo reference in alpha-beta

%% Output sequences

Yref=complex(zeros(4,1)); %(4,1)

Yref(1:2,1)=Rr*Vo_xy_ref; % Construct output reference sequence

for k=3:2:4

row_i=k;

row_f=k+1;

Yref(row_i:row_f,:)=Rr*Yref(row_i-2:row_f-2,:);

end

% FOR loop to find the best switch in the next instant and its corresponding V(i)

for k = 1:7

s = states(k,:);

sw = (2/3)*(s(1)+a*s(2)+a^2*s(3));

Vi = Vdc*sw;

% Output Voltage Prediction at instant k+1

Vc1 = Aq(2,1)*ifk + Aq(2,2)*Vck + Bq(2)*Vi + Bdq(2)*io1;

%% System state

Vo_xy = [Vc1; Vc1];

Io_xy = [ifk ifk]';

Ii_xy = [io1; io1]; % Approximate filter current

x_k = [Ii_xy ; Vo_xy; Io_xy]; % System state variables

%% Unconstrained solution calculation

% Initialize variables

F_k=zeros(p,1);

Niter=0.0; % Count nodes

uabc_k_1 = [0 0 0]'; % Assume previous state is 0 0 0

F_k=UpsilonT*(Gamma*x_k-Yref)-lSTE*uabc_k_1;

U_unc=-W_inv*F_k;

U_bar_unc = H*U_unc; % Center of the sphere in the transformed space

%Initialize SDA matrices

Niter=0;

Ucand=complex(zeros(p+1,2*Kbest));

Ucand_aux=complex(zeros(p+1,2*Kbest));

U_op=zeros(p,1);

i=1; %i is the level

while (i<=p)

minK = min(Kbest, 2^(i-1));

for j = 1:2:minK*2

for C = u_min:u_max

i_value = i;

Ucand(i,j+C) = C;

Ucand(p+1,j+C) = (H(i_value,1:i_value)*Ucand(1:i_value,j+C) - U_bar_unc(i_value))^2 + Ucand(p+1,j+C);

Niter=Niter+1;

end

end

Ucand = sortrows(Ucand',p+1)';

minK = min(Kbest, 2^i);

l = 1; j = 1;

while j <= minK*2

if Ucand(p+1,l) > 0

Ucand_aux(:,j+0) = Ucand(:,l);

Ucand_aux(:,j+1) = Ucand(:,l);

j = j + 2;

end

l = l + 1;

end

Ucand = Ucand_aux;

i=i+1;

end

end

Rs2_o = Ucand(p+1,1);

U_op = real(Ucand(1:p,1));

end

Any suggestions would be much appreciated.

Answer 1

[Fangjun Jiang]

In general, it is better to specify the size (number of rows and columns) of all the inputs, outputs and parameters that are used by the MATLAB Function block inside a Simulink model.

In MATLAB, it is not a problem. But in Simulink, it prefers any data to be a fixed size. Variable size data is allowed but it's very tricky to use.

You have pre-allocate the size of variable UpsilonT = zeros(6,4); Gamma = zeros(4,6); lSTE = zeros(6,3); W_inv = zeros(6,6. That is good. I wonder if you clicked "Edit Data" in the code editor and specify the size of inputs and outputs.

Also, you seem to have complex number. It's better to not to use i or j as a variable name then.