How to update app designer GUI in real time at high refresh rate without stutter/lag?

I'm currently working on an application that simulates a medical monitor. I designed a GUI in matlab's app designer which 1) shows the 6 graphs/signals and 2) provides the controls to adjust those signals. A few simple functions have been implemented to simulate e.g. an ECG signal. I use a timer object within the app to call a function that 1) refreshes the current signal values and 2) plots the new signals in the GUI axes. It is currently working at a refresh rate of 20 Hz (way too low) and a physiological model to be used to replace the current 'simple functions' is yet to be implemented.

However, I'm already running into a problem that the application becomes too slow, shortly after starting up. The timer seems unable to maintain the desired refresh rate, resulting in stuttering or 'lag' in the displaying of the signals. The controls become slow to respond as well or even stop responding altogether.

The problem is so bad I'm currently debating rewriting everything in another language (maybe Java or C++), even though I have virtually no experience with those.

If anyone has an idea of how to solve this, I'd be very grateful to hear it.

To give an idea of what the (simplified) code for the primary function looks like:

function Timer = timed_dash(app)
      % Function to create and start the main timer needed for real-time simulation

        Timer = timer('ExecutionMode','fixedRate','Period',app.period);

        Timer.TimerFcn      = {@app.myfun};

       % <snip>...lots of one time variable creation...<snip>

        start(Timer);

 end

 function myfun(app, ~, ~)
 % Main Function called by timer to run / plot everything

       % Is ECG connected?
        if app.ConnectECGButton.Value % Yes
            app.ECGplot(app.shortcounter)=func_ECG(app.t(app.longcounter));
            app.HRplot(app.longcounter)=app.HR+random('Normal', 0,.3);%+2*sin(app.t(app.longcounter)*pi);
            app.RFplot(app.longcounter)=app.RF+2*random('Normal', 0,.1);
            app.Respplot(app.medcounter)=.6*sin(2*pi*app.RF/60*app.t(app.longcounter)); % temporary demo func
           % Adjust current value for numeric display of monitor data
            app.RFText.Value = sprintf('%d',round(app.RFplot(app.longcounter)));
            app.HRText.Value = sprintf('%d',round(app.HRplot(app.longcounter)));
        else % No
            app.ECGplot(app.shortcounter)=0;
            app.Respplot(app.medcounter)=0;
            app.HRplot(app.longcounter)=0;
            app.RFplot(app.longcounter)=0;
           % Adjust current value for numeric display of monitor data
            app.RFText.Value = '0';
            app.HRText.Value = '0';
        end

        if app.loopshortcounter
                plot(app.ECGAxes,app.t(1:app.shortcounter),app.ECGplot(1:app.shortcounter),'-g','LineWidth',2)
                plot(app.PlethAxes,app.t(1:app.shortcounter),app.Plethplot(1:app.shortcounter),'-c','LineWidth',2)
        else
                plot(app.ECGAxes,app.t(1:app.shortcounter),app.ECGplot(1:app.shortcounter), '-g',app.t(app.shortcounter+3:app.endshortcounter+3),app.ECGplot(app.shortcounter+3:end),'-g','LineWidth',2)
                plot(app.PlethAxes,app.t(1:app.shortcounter),app.Plethplot(1:app.shortcounter),'-c',app.t(app.shortcounter+3:app.endshortcounter+3),app.Plethplot(app.shortcounter+3:end),'-c','LineWidth',2)
        end

        if app.loopmedcounter % Resp loop
                plot(app.RespAxes,app.t(1:app.medcounter),app.Respplot(1:app.medcounter),'-y','LineWidth',2)
        else
                plot(app.RespAxes,app.t(1:app.medcounter),app.Respplot(1:app.medcounter),'-y',app.t(app.medcounter+3:app.endmedcounter+3),app.Respplot(app.medcounter+3:end),'-y','LineWidth',2)
        end

        if app.looplongcounter
                plot(app.RFAxes,app.t(1:app.longcounter)./60-10,app.RFplot(1:app.longcounter),'-y','LineWidth', 2)                
                plot(app.SpO2Axes,app.t(1:app.longcounter)./60-10,app.SpO2plot(1:app.longcounter),'-c','LineWidth', 2)
                plot(app.HRAxes,app.t(1:app.longcounter)./60-10,app.HRplot(1:app.longcounter),'-g','LineWidth', 2)
        else
                plot(app.RFAxes,app.t(1:app.longcounter)./60-10,app.RFplot(1:app.longcounter), '-y',app.t(app.longcounter+3:end),app.RFplot(app.longcounter+3:end),'-c', 'LineWidth',2)
                plot(app.SpO2Axes,app.t(1:app.longcounter)./60-10,app.SpO2plot(1:app.longcounter), '-c',app.t(app.longcounter+3:end),app.SpO2plot(app.longcounter+3:end),'-y', 'LineWidth',2)
                plot(app.HRAxes,app.t(1:app.longcounter)./60-10,app.HRplot(1:app.longcounter), '-g',app.t(app.longcounter+3:end),app.HRplot(app.longcounter+3:end),'-c', 'LineWidth',2)
        end

       % Update counters & reset when necessary. 
        if app.shortcounter==app.endshortcounter
            app.shortcounter=1;
            if app.loopshortcounter
                app.loopshortcounter=false;
            end
        else
            app.shortcounter=app.shortcounter+1;
        end

        if app.medcounter==app.endmedcounter
            app.medcounter=1;
            if app.loopmedcounter
                app.loopmedcounter=false;
            end
        else
            app.medcounter=app.medcounter+1;
        end

        if app.longcounter==app.endlongcounter
            app.longcounter=1;
            if app.looplongcounter
                app.looplongcounter=false;
            end
        else
            app.longcounter=app.longcounter+1;
        end

 end

Answer 1

[Jan]

Do I see correctly that you create a new set of plot's in each timer callback? This is rather inefficient. Prefer to create the line object once and update the XData and YData afterwards.

Search for "XData" or "animation" in this forum or in the net to see many examples.

See also: http://undocumentedmatlab.com/blog/plot-performance

[Camile van der Heijden]

Thank you for your quick answer Jan. I am indeed plotting each timer callback. It indeed seems very inefficient, but I didn't know of a way to avoid it. I had already implemented the tricks listed in the undocumentedmatlab link you provided, but unfortunately that didn't speed things up enough. I would like to try your suggested solution using XData and YData, however I'm confused as to how this would work with more than 1 line/set of data per axes couple. XData/YData don't seem to accept matrix input? -- nevermind that last bit, I found out already. I'll get back here with my results.

*I'm using 2 lines to create a (moving) 'gap' between the new and old part of the displayed data.

[Camile van der Heijden]

Reporting back after converting my code. I now have the app create all the necessary line objects once and then update the XData/YData using the 'set' function on every timer callback.

Although this does indeed to lead to improvement, it is still not enough to ensure fluent animation. Do you have any other suggestions I could try? Thanks again.

[Camile van der Heijden]

Some other optimizations I'm using: - Instead of using 2 line objects per plot to create a gap, I'm instead using a (moving) NaN value to get the same effect. - I'm using interp1 tot resample to fewer points as much as possible. - I've also found standard figure/GUIDE GUI's to be much more efficient. Updating a figure is much faster that updating a uifigure due to the latter employing a HTML/Javascript layer (Matlab support confirmed this to me).

As for myself, I'm going to try implementing the 'animatedline' function to see if that will solve my problems. I thought I would be able to keep my 'old' data on screen until overwritten, but it seems this happens automatically using 'animatedline'.