pause function gives inconsistent timing when used inside a loop.

Hello,

I am trying to create an animated plot. The pause command inside an infinite loop is supposed to created the timing for me:

%%
figure; clf;
t = 0:1/16000:1;
y = sin(2*pi*4*t);
h = plot(t,y);
xlim([ 0 t(end)]);
ylim([ -1 1]);
h.XDataSource = 't';
h.YDataSource = 'y';

rate = 10;
while 1
   tic
   pause(1/rate);
   %java.lang.Thread.sleep(1000 * (1/rate))
   toc
   y = sin(2*pi*freq*t);
   refreshdata(h);
   drawnow % needed?
   freq = freq + 0.1;
   if freq > 40
       freq = 4;
   end
end

When I try this tic toc shows some inconsistent timing:

Elapsed time is 0.100041 seconds.

Elapsed time is 0.599817 seconds.

Elapsed time is 0.100195 seconds.

Elapsed time is 0.100100 seconds.

Elapsed time is 0.499797 seconds.

Elapsed time is 0.100046 seconds.

If I remove everything from the loop and leave just the pause command it behaves the same.

I tried using java's sleep command but that does not allow the plots to be updated until the loop is exited which is useless.

Update:

The results shown above happens whn I run this on Linux. I tried running the same code on Windows and it works fine. Not sure what could be the reason.

Answer 1

[Adam Danz]

Careful using tic/toc. As summarized in the documentation, "If your code is faster than 1/10 second, consider measuring it running in a loop, and then average to find the time for a single run". Even with larger durations, there is unavoidable variability (shown in this comment within this thread).

Alternatively, use timeit to determine how long a process takes (not to control real-time processes). Unlike tic and toc, the timeit function calls your code multiple times, and, therefore, considers first-time costs.

Check out this comparison of timing pause(0.1) using toc/toc and timeit. Notice the difference in variability. For a bit more info on tic/toc timing within a loop, see this answer.

Additionally, the pause() function is susceptible to variations due to your OS and concurrent system activity. The finer the temporal resolution, the higher the relative error. For more info on limitations to the pause function, scroll down to the tips section in this link.

f = @()waitasec;
t = zeros(1,100); 
for i = 1:100
    t(i) = timeit(f);
end

t2 = zeros(1,100); 
for i = 1:100
    tic
    pause(0.1)
    t2(i) = toc; 
end

clf
hold on
h = plot(t, 'DisplayName', 'timeit')
h2 = plot(t2, 'DisplayName', 'tictoc')
legend()

function waitasec
pause(0.1)
end

If you're trying to control the speed of animation, consider using one of the animation techniques, some of which provide examples of controlling animation speed. This blog post uses a timer object to control animation speed. I have not read or vetted the article but keep in mind that Matlab recommends not using a timer for real-time applications.

Answer 2

[jmore]

I have encountered a similar issue on my Linux operating system. Mr. Adam Danz has suggested using timeit, the the following approach is a little easier to understand for me:

rate = 10
tic
while toc < 1/rate
end
toc

This will replace to faulty pause command that varies from 0.1 to 0.5 seconds:

tic
pause(1/rate)
toc

[Bruno Luong]

+1. I do the same.

TIC/TOC is the most lean and reliable in my book.

TIMEIT do something different (benchmark some piece of code) and I don't like the overhead it requires by having interface with a function.

PAUSE is has great variability since it might trigger all kind of events beside the main task of PAUSING execution.

[Adam Danz]

+1

To clarify, my answer was in response to a comment that was removed and recommends timeit() to benchmark the process for comparisons, not to control real-time applications.

jmore's method does a good job of avoiding the problems with pause(). However, users who are looking for very precise timing control should be aware of the limitations with tic/toc or any method of time control in Matlab.

As indicated in the documentation, tic/toc becomes unreliable for durations less than 0.1 sec. But even for larger durations there is unavoidable variability in the results. The plots below show the timing of 16 different durations repeated 10 times using the method in jmore's answer (with some slight changes). Results are jittered along the x axis to prevent overlap.

Performed on Windows 10 Pro Vs 10.0 (Matlab Online) using Matlab version 9.7.0.1434023 (R2019b) Update 6

To reproduce this analysis:

nReps = 10;
rate = [.001,.005,.01,.02,.05,.1,.11,.12,.15,.2,.5,1,1.5,2,5];  % seconds

nMinutes = sum(nReps.*rate)/60;
wb = msgbox(sprintf('This will take approximately %.1f minutes.', nMinutes), 'Confirm');
uiwait(wb)

dur = zeros(numel(rate),nReps); 
for j = 1:numel(rate)
    for i = 1:nReps
        s = tic;
        while toc(s) < rate(j)
        end
        dur(j,i) = toc(s); 
    end
end

mu = mean(dur,2); 
sd = std(dur,[],2); 
err = dur - rate(:); 

clf()
tiledlayout(2,1)
t(1) = nexttile;
hold on
errorbar(mu,sd,'Marker', 'none', 'DisplayName', '+/-1 std')
xJitter = (1:numel(rate))' + (rand(numel(rate),nReps)-.5)/2;
plot(xJitter, dur, 'bo', 'DisplayName', 'all iterations')
plot(1:numel(rate), mu, 'rd', 'DisplayName', 'mean')
set(gca, 'XTick', 1:numel(rate), 'XTickLabel', rate)
xlabel('Wait time requested (sec)')
ylabel(['Mean clocked wait time (sec)',newline,'(plus some overhead)']); 
title(['Timing results for various durations',newline,'across ',num2str(nReps),' reps'])
legend()

t(2) = nexttile;
plot(xJitter, err, 'bo', 'DisplayName', 'all iterations')
set(gca, 'XTick', 1:numel(rate), 'XTickLabel', rate, 'YScale', 'log')
xlabel('Wait time requested (sec)')
ylabel('Error (measured minus requested, sec)'); 
title(['Tic|Toc error for various durations',newline,'across ',num2str(nReps),' reps'])

[Bruno Luong]

@Adam: Why there is an horizontal dispersion in the plot?I mean in your code the xaxis is plotted with 1:numel(rate) yet I see the points are dispersed horizontally. Odd.

[Adam Danz]

I must have added that after copying the code to the comment. The comment states, "Results are jittered along the x axis to prevent overlap ", otherwise many of the points would lay on top of eachother.

To jitter the points,

xJitter = (1:numel(rate))' + (rand(numel(rate),nReps)-.5)/2;
plot(xJitter, dur, 'bo', 'DisplayName', 'all iterations')

I'll fix that in my comment. Thanks for pointing that out.