Is there a way to reduce Timer period to less than 1 ms?

I am trying to reduce timer(‘period’) to less than 0.001 if possible. In this forum link https://www.mathworks.com/matlabcentral/answers/37716-pause-function-in-matlab-for-1-millisecond, few people have suggested using java pause or writing your own pause function but how can I implement these alternatives inside the predefined timer? Or can I redefine timers ‘period’? I need the timer for its start finish and check functionality.The other question I have is why does one need wait? The fact that timer’s callback is running no further Matlab commands are getting executed. The wait is adding another significant time to my process. Thanks

[Steven Lord]

What is your application where you need a timer with periods that small? Are you trying to do something related to real-time processing?

[gujax]

Actually I realized I 'perhaps' do not have to shorten the period. I think my problem is something else.

I am wondering if the timer call back function can be updated with incremented variable without re-creating a new timer?

I am running a device (say a piezo mirror) and for every position I am generating data using another object (e.g., tetronix scope or a daq).

So the sequeunce is -Generate a for-loop:

move the mirror to first position(ii)

Run tetronix and record data

Check if the data record is complete

If complete go for the next mirror position(ii+1) and repeat for 10 positions

Data gets taken pretty fast like in <1ms

There are few issues.

If the mirror is not made to move at the correct interval it damages my sample so I got to stick to 1ms of data record and then move the mirror. If mirror move happens in 2 ms then I damage the sample (let us say I do not have other recourse other than moving the mirror).

So I ran the timer as such

t = timer('StartDelay', 0.000, 'Period', 0.001, 'TasksToExecute', 100, 'BusyMode', 'drop', 'ExecutionMode', 'fixedRate');

But each for-loop took 100ms. I found that was because in every for-loop I had to generate a new timer because old timer had to be deleted. Why? Because the timer callback function runs tetronix and provides data file names to save data from tetronix i.e., it needs to be updated for every for-loop of the mirror. And that completely damages my experiment.

So instead I made a timer callback function where I could move the mirror. So now I generate all 10 positions filenames once, no for-loop.

So generate timer function, start, check if first data is written. So period=0.001.

If it is then next callback function is in queue and that next call increments mirror position.No need to generate a new timer.

But then I found I couldn't update the mirror position .

For example in the function below aarg2 moves the mirror using DAQ and I thought I could update the mirror position so that in the 'next' callback a new mirror position will be a new input. That is after a period of 1ms, the next callback is in the queue. And in that next queue if I could affect the mirror position without re-creating a timer, my issue will be resolved.

But alas that is not the case.

function my_callback_fcn(obj, event, argtet,aarg1,aarg2, status)

if strcmp(status, 'START')

command = 'Tetronix_START';

runtetx(command, argtet);

elseif strcmp(status, 'FINISH')

command = 'Tetronix_STOP';

runtetx( command,argtet);

else

command = 'Tetronix_check';

runtetx( command, argtet,aarg1);

disp(aarg2);

write(aarg1,[aarg2 0 0 1]);%aarg1 is DAQ object and aarg2 is the voltage corresponding to position of the mirror

pause(0.04);

aarg2=aarg2+1;

end

So if I started with aarg2=-10Volts

The script ran and showed a series of aarg2=-9 and it wrote all 10 files pretty fast but only in one mirror position

Answer 1

[Walter Roberson]

You would need to reimplement timers.

[gujax]

How can this be done? By creating an identical timer function and then including changes to it? Do I have access to that original code?

[Walter Roberson]

You can edit (view) timer.m . However it invokes the built-in method matlab.internal.timer.TimerInfo to do all of the real work, and that is not something you have access to.

You would, in other words, need to completely re-implement the internals for timers .

See the discussion at https://www.mathworks.com/matlabcentral/answers/1887467-why-is-pause-function-so-inaccurate-on-windows#answer_1139927 for why you will not be able to use pause() for this work, at least not on Windows 10.

There are two different uses for timers.

One use of timers to execute something periodically while the rest of the MATLAB session sits idle. You have some hope of being able to implement that yourself, by using java sleep, having a queue of wake-up times for timers, and sleeping the amount of time needed for the next object in the queue. But java sleep does not allow MATLAB to proceed while it is going on.

The other use of timers is to execute something periodically while the rest of MATLAB is doing something, interrupting the regular MATLAB code . Internal mechanisms are used to set flags indicating to the interpreter that an interrupt needs to be serviced at the start of the next line of MATLAB code. Multiple statements on the same line are not interrupted in the middle -- but if they call MATLAB-coded functions then the interrupt can happen at lines inside them. There is no documented interface to tell MATLAB to service an interrupt as soon as practical, so you will not be able to implement this behaviour (at least not without a struggle.)

Historically there have been people who used java to create new java threads that could operate in parallell with MATLAB's execution engine, but those threads could only use interfaces such as tcp or java queues to communicate with the MATLAB thread, and historically any modification of a MATLAB variable inside one of those java threads could make MATLAB crash. These days now that backgroundPool exist there is reason to believe that MATLAB is now considerably more thread-safe then was historically the case, so you just might be able to get away with using Java to create additional threads that used the MATLAB Engine API to interface with the rest of MATLAB. Maybe .

You need to face the fundamental fact that MATLAB itself is not designed for real-time work. MATLAB deals with real-time work by having the Simulink Real-Time product that generates code for use with (very specific) real-time kernals (QNX these days).

You also need to face the fundamental fact that Windows is not designed for real-time work -- not until you get into Windows Enterprise IoT https://learn.microsoft.com/en-us/windows/iot/iot-enterprise/soft-real-time/soft-real-time (previously known as Windows Embedded)

MacOS is slightly more suited for Real-Time work... but not really.

Most Linux are not designed for real-time, but there are specialized Linux versions that are designed for real-time work.

If you need scheduling 1000 times per second... then you are pretty much doing real-time work.

[gujax]

Thank you.

I realized that my issue was not really the 'real time' operation but the fact that my implementation had to generate a timer function for every for-loop of a variable that got incremented and that the timer function was using it. Generating timer function is taking 100ms long. I do not see any other way out other than synchronizing several devices using hardware timing rather than using Matlab

[Walter Roberson]

I am not sure why a new timer function was being generated? Could you have used shared variables instead?

https://www.mathworks.com/help/matlab/matlab_prog/share-data-between-workspaces.html#f0-50984

[gujax]

It’s an ADC data acquisition card. It streams input data to a computer memory. So it runs by pre-assigning number of files, their names and trigger and handshake times for every corresponding position of a scanning mirror. These variables are then input to a timer callback. Then from the Matlab’s timer callback function it calls a mex function to start data record for a mirror position. So if I have to synchronize every mirror step to set of data files, I need to call this Mex function with new updated variables every time I change mirror position. I don’t know how can I create a timer function that starts data recording for every mirror position without generating a new timer for each loop!

Here is an example

for I=1:20 % for each mirror pos

Move mirror to pos(I);%DAQ command

P.numberoffiles =100

P.filesize =30K

P.filenames= {code generates 100 filenames}

P.triggertime=100e-3

P.triggertimesforeachfile = vector of times; etc..

t=timer(start,period,etc);

t.start= somefunction(object,event,P):

t.stop= same as above;

And check for status function if the data recording is complete..

start(t);

wait(t);

Some code here….

delete(t);

end

So that ‘somefunction’ calls the ADC’s mex function which uses P variable structure and records data of certain length for certain triggers and store to a file structure. It returns back on completion and script goes through a second loop.

I am not sure how nesting functions would help though it is a good suggestion and I will reflect on it. Thank you