Gday all,
For context, I am building a simulink model that will communicate with the simulator X-plane through UDP to control an aircraft. The Simulink model inputs a UDP connection and a live video feed. Using an object detection system to identify key features, a command direction will begiven, and outputted to X-plane through UDP communication.
My issue is that the UDP communication input significantly decreases the performance of the model. I have tried every UDP commenction block avaialble in Simulink. I have tested the simulation with other blocks, and I’m confident that the UDP communication input block (either of them) are the root of the issue. I havent had this issue with other models (such as image segmentation model). Nor does the UDP output show performance issues.
Is there a way to fix the performance issues seen in the model? Is there some UDP secret I dont know about? Workarounds? Would the UDP comminication block be interfereing with any other block (deep learning, PID…)?
Any quidence or tips would be appreciatedGday all,
For context, I am building a simulink model that will communicate with the simulator X-plane through UDP to control an aircraft. The Simulink model inputs a UDP connection and a live video feed. Using an object detection system to identify key features, a command direction will begiven, and outputted to X-plane through UDP communication.
My issue is that the UDP communication input significantly decreases the performance of the model. I have tried every UDP commenction block avaialble in Simulink. I have tested the simulation with other blocks, and I’m confident that the UDP communication input block (either of them) are the root of the issue. I havent had this issue with other models (such as image segmentation model). Nor does the UDP output show performance issues.
Is there a way to fix the performance issues seen in the model? Is there some UDP secret I dont know about? Workarounds? Would the UDP comminication block be interfereing with any other block (deep learning, PID…)?
Any quidence or tips would be appreciated Gday all,
For context, I am building a simulink model that will communicate with the simulator X-plane through UDP to control an aircraft. The Simulink model inputs a UDP connection and a live video feed. Using an object detection system to identify key features, a command direction will begiven, and outputted to X-plane through UDP communication.
My issue is that the UDP communication input significantly decreases the performance of the model. I have tried every UDP commenction block avaialble in Simulink. I have tested the simulation with other blocks, and I’m confident that the UDP communication input block (either of them) are the root of the issue. I havent had this issue with other models (such as image segmentation model). Nor does the UDP output show performance issues.
Is there a way to fix the performance issues seen in the model? Is there some UDP secret I dont know about? Workarounds? Would the UDP comminication block be interfereing with any other block (deep learning, PID…)?
Any quidence or tips would be appreciated simulink, deep learning MATLAB Answers — New Questions

​

I do not want any dynamic memory allocation to be done in the C++ codegen. Here is the relevant Matlab code:
% Calculate cross-correlation matrix from a set of vectors (whose length
% can vary from one call to another). All 4 channels will have the length.
function [R] = xcorr_matrix(ch0,ch1,ch2,ch3,channelCount)

% channelCount is 2 or 4
% R is either 2×2 or 4×4 complex double matrix.
% ch0 … ch3 are complex single vectors.

N = int32(size(ch0,1));

% Convert ch0, …, ch3 to complex double
x_in = coder.nullcopy(complex(zeros(N,channelCount,’double’)));
switch channelCount
case 2
x_in(1:N,1) = ch0(1:N);
x_in(1:N,2) = ch1(1:N);
case 4
x_in(1:N,1) = ch0(1:N);
x_in(1:N,2) = ch1(1:N);
x_in(1:N,3) = ch2(1:N);
x_in(1:N,4) = ch3(1:N);
end

R = (x_in’ * x_in); % Compute cross-correlation matrix
Here is the C++ codegen result:
void xcorr_matrix(const creal32_T ch0_data[], const int ch0_size[1], % all the 4 ch_size will be the same value
const creal32_T ch1_data[], const int ch1_size[1],
const creal32_T ch2_data[], const int ch2_size[1],
const creal32_T ch3_data[], const int ch3_size[1],
int channelCount,
::coder::array<creal_T, 2U> &R)
{
::coder::array<creal_T, 2U> x_in;
R.set_size(channelCount, channelCount);
x_in.set_size(N, channelCount);
…
}
I think I can eliminate the x_in.set_size by not using x_in, and replace the matrix multiply with nested for-loops and using double casting; but am unsure how to define R (either 2×2 or 4×4) so as to remove the R.set_size allocations.
One idea I had was to try making R a fixed length 16 element vector, and just use 4 elements for the 2×2 R, and all 16 for the 4×4 R. Would be nicer to be able to use two indices for rows and columns.
Thanks in advance for your help.
PaulI do not want any dynamic memory allocation to be done in the C++ codegen. Here is the relevant Matlab code:
% Calculate cross-correlation matrix from a set of vectors (whose length
% can vary from one call to another). All 4 channels will have the length.
function [R] = xcorr_matrix(ch0,ch1,ch2,ch3,channelCount)

% channelCount is 2 or 4
% R is either 2×2 or 4×4 complex double matrix.
% ch0 … ch3 are complex single vectors.

N = int32(size(ch0,1));

% Convert ch0, …, ch3 to complex double
x_in = coder.nullcopy(complex(zeros(N,channelCount,’double’)));
switch channelCount
case 2
x_in(1:N,1) = ch0(1:N);
x_in(1:N,2) = ch1(1:N);
case 4
x_in(1:N,1) = ch0(1:N);
x_in(1:N,2) = ch1(1:N);
x_in(1:N,3) = ch2(1:N);
x_in(1:N,4) = ch3(1:N);
end

R = (x_in’ * x_in); % Compute cross-correlation matrix
Here is the C++ codegen result:
void xcorr_matrix(const creal32_T ch0_data[], const int ch0_size[1], % all the 4 ch_size will be the same value
const creal32_T ch1_data[], const int ch1_size[1],
const creal32_T ch2_data[], const int ch2_size[1],
const creal32_T ch3_data[], const int ch3_size[1],
int channelCount,
::coder::array<creal_T, 2U> &R)
{
::coder::array<creal_T, 2U> x_in;
R.set_size(channelCount, channelCount);
x_in.set_size(N, channelCount);
…
}
I think I can eliminate the x_in.set_size by not using x_in, and replace the matrix multiply with nested for-loops and using double casting; but am unsure how to define R (either 2×2 or 4×4) so as to remove the R.set_size allocations.
One idea I had was to try making R a fixed length 16 element vector, and just use 4 elements for the 2×2 R, and all 16 for the 4×4 R. Would be nicer to be able to use two indices for rows and columns.
Thanks in advance for your help.
Paul I do not want any dynamic memory allocation to be done in the C++ codegen. Here is the relevant Matlab code:
% Calculate cross-correlation matrix from a set of vectors (whose length
% can vary from one call to another). All 4 channels will have the length.
function [R] = xcorr_matrix(ch0,ch1,ch2,ch3,channelCount)

% channelCount is 2 or 4
% R is either 2×2 or 4×4 complex double matrix.
% ch0 … ch3 are complex single vectors.

N = int32(size(ch0,1));

% Convert ch0, …, ch3 to complex double
x_in = coder.nullcopy(complex(zeros(N,channelCount,’double’)));
switch channelCount
case 2
x_in(1:N,1) = ch0(1:N);
x_in(1:N,2) = ch1(1:N);
case 4
x_in(1:N,1) = ch0(1:N);
x_in(1:N,2) = ch1(1:N);
x_in(1:N,3) = ch2(1:N);
x_in(1:N,4) = ch3(1:N);
end

R = (x_in’ * x_in); % Compute cross-correlation matrix
Here is the C++ codegen result:
void xcorr_matrix(const creal32_T ch0_data[], const int ch0_size[1], % all the 4 ch_size will be the same value
const creal32_T ch1_data[], const int ch1_size[1],
const creal32_T ch2_data[], const int ch2_size[1],
const creal32_T ch3_data[], const int ch3_size[1],
int channelCount,
::coder::array<creal_T, 2U> &R)
{
::coder::array<creal_T, 2U> x_in;
R.set_size(channelCount, channelCount);
x_in.set_size(N, channelCount);
…
}
I think I can eliminate the x_in.set_size by not using x_in, and replace the matrix multiply with nested for-loops and using double casting; but am unsure how to define R (either 2×2 or 4×4) so as to remove the R.set_size allocations.
One idea I had was to try making R a fixed length 16 element vector, and just use 4 elements for the 2×2 R, and all 16 for the 4×4 R. Would be nicer to be able to use two indices for rows and columns.
Thanks in advance for your help.
Paul matlab, codegen, embedded coder, c++, dynamic memory allocation MATLAB Answers — New Questions

​

I am new to Matlab but am well aware of the bad practice notion associated with dynamically creating workspace variables. Unfortunately, Matlab’s volumeSegmenter only allows saving of segmentations as either MAT-files or workspace variables, and the former creates far too many individual files for the amount I require.
In the next step after creating them, I need to run all the segmentations (workspace vars seg1, seg2, seg3 …) through a for loop. I am currently using who() to try and find all the needed workspace variables, but this doesn’t work as only the names are stored in cells seg_options and cannot be called as variables:
vars = who();
find = contains(vars, ‘seg’);
seg_options = vars(find);

This is part of the for loop I need to call the segmentation variables for:
for i = 1:length(seg_options);
A = double(seg_options(i));
end
which obviously doesn’t work properly as I need to be calling the actual variable and not just its name.

The code also needs to work for a flexible number of segmentations (ie I cannot initialize an array as a specific size). Is there a way to:
1) load the workspace variable into array, overwrite it, load the next one into the next array cell, etc. (ie saves the first segmentation as seg, loads into seg_array cell 1, saves the next segmentation as seg, load that into seg_array cell 2, and so on)
2) load all the created variables (seg1, seg2…) into an array seg_array
or
3) call and loop through all the workspace variables in the for loop itself – I know this is not ideal

Thanks in advance!I am new to Matlab but am well aware of the bad practice notion associated with dynamically creating workspace variables. Unfortunately, Matlab’s volumeSegmenter only allows saving of segmentations as either MAT-files or workspace variables, and the former creates far too many individual files for the amount I require.
In the next step after creating them, I need to run all the segmentations (workspace vars seg1, seg2, seg3 …) through a for loop. I am currently using who() to try and find all the needed workspace variables, but this doesn’t work as only the names are stored in cells seg_options and cannot be called as variables:
vars = who();
find = contains(vars, ‘seg’);
seg_options = vars(find);

This is part of the for loop I need to call the segmentation variables for:
for i = 1:length(seg_options);
A = double(seg_options(i));
end
which obviously doesn’t work properly as I need to be calling the actual variable and not just its name.

The code also needs to work for a flexible number of segmentations (ie I cannot initialize an array as a specific size). Is there a way to:
1) load the workspace variable into array, overwrite it, load the next one into the next array cell, etc. (ie saves the first segmentation as seg, loads into seg_array cell 1, saves the next segmentation as seg, load that into seg_array cell 2, and so on)
2) load all the created variables (seg1, seg2…) into an array seg_array
or
3) call and loop through all the workspace variables in the for loop itself – I know this is not ideal

Thanks in advance! I am new to Matlab but am well aware of the bad practice notion associated with dynamically creating workspace variables. Unfortunately, Matlab’s volumeSegmenter only allows saving of segmentations as either MAT-files or workspace variables, and the former creates far too many individual files for the amount I require.
In the next step after creating them, I need to run all the segmentations (workspace vars seg1, seg2, seg3 …) through a for loop. I am currently using who() to try and find all the needed workspace variables, but this doesn’t work as only the names are stored in cells seg_options and cannot be called as variables:
vars = who();
find = contains(vars, ‘seg’);
seg_options = vars(find);

This is part of the for loop I need to call the segmentation variables for:
for i = 1:length(seg_options);
A = double(seg_options(i));
end
which obviously doesn’t work properly as I need to be calling the actual variable and not just its name.

The code also needs to work for a flexible number of segmentations (ie I cannot initialize an array as a specific size). Is there a way to:
1) load the workspace variable into array, overwrite it, load the next one into the next array cell, etc. (ie saves the first segmentation as seg, loads into seg_array cell 1, saves the next segmentation as seg, load that into seg_array cell 2, and so on)
2) load all the created variables (seg1, seg2…) into an array seg_array
or
3) call and loop through all the workspace variables in the for loop itself – I know this is not ideal

Thanks in advance! array, cell array, variable, variables, arrays, cell arrays, image segmentation, for loop MATLAB Answers — New Questions

​