Precision Issue in adding very small fractions of seconds to datenum

Hello All:

I am trying to plot a .wav file audio signature with 'true time of recording' on the x-axis. My ideas was to use datenum, and the datetick to achieve this. But I ran into problems pretty early.

The file is coded at 14kHz, and is exactly 1 minute long. So audioread gives me 14000*60=840000 points. So the time 'relative to the wav file' runs from t=0 to t=60 seconds in steps of 1/14000 second.

I am then trying to use datenum to add 'true' time to the above. The actual time of recording is 05-March-2013 14:00:00 hrs. For simplicity, lets say it was 05-March-2013 00:00:00 hrs.

05-March-2013 00:00:00 is 735298. Call it 'a'. 1 second on top of this is 1/86400. The fraction of second I am dealing with is 1/14000. So fraction in 'datenum' terms is 1/14000/86400. Call it 'b'.

If I add 1/14000th of a second to the date, I already have 735298.0000000008267195767195768 which runs out of precision.

What can I do?

My ultimate aim is to achieve the audio signature of 'y' axis and 'actual date & time' on x-axis.

Thanks all for your attention. Any help will be highly appreciated.

[José-Luis]

How do you expect your time to be displayed?

[Amit]

Hello Jose-Luis.

I want the x-axis time label to be actual (true) 'HH:MM'.

Additionally, I want to put a label (an annotation or a text box) on the top right of the graph (or in the title) where I give the actual (true) date.

Overall, I am plotting 12-13 hours of data. I have 750+ 1-minute wav files. Holding on to the plot, I plot 180 such 1-minute wav files, then my RAM runs out. Then I export it to a jpeg for a report. At this stage, I 'hold off', and plot the next 3 hours.

Thanks for your attention. Any tips will be precious.

[Amit]

This is my code -

load FileNameList.mat;

%starting plot

format compact; hold off; figure; plot(0); hold on; set(gcf,'units','normalized','position',[0,0.82,1,0.2]);

t_end=0; t_unit=180; %3 hours at a time recommended on Amit's 4GB RAM computer

for i=1:1:length(FListM1)

    [yM1Temp fsM1Temp]=audioread(FListM1(i,:));
    t=[t_end:1/fsM1Temp:t_end+((length(yM1Temp)-1)/fsM1Temp)]';
    t_end=t(end);

%fsM1Temp is 14000. length(yM1Temp)=14000*60=840000

    %plot(VDR_start_time + (t/86400), yM1Temp); datetick('x', 'HH:MM'); hold on; %%%%%%%%%%leads to precision issues
%%%THIS IS WHAT I WANT

    plot(t/60, yM1Temp); hold on; %temporary measure leading to relative time in minutes
%%INSTEAD THIS IS WHAT I HAVE TO LIVE WITH AT THE MOMENT

    i %to track progress on screen

    if (mod(i,t_unit))==0
        title('Full Audio Signature Channel 1'); xlabel('Time in minutes'); axis tight; grid on; grid minor; %datetick('x', 'HH:MM');
        screen2jpeg(strcat('FullAudioSignature_',num2str((i-t_unit)/60), '-', num2str(i/60), 'hrs.jpg'));
        hold off;
    end
end

if mod(i, t_unit)~=0

    title('Full Audio Signature Channel 1'); xlabel('Time in minutes'); axis tight; grid on; grid minor; %datetick('x', 'HH:MM');

    screen2jpeg(strcat('FullAudioSignature_',num2str(uint8((i-t_unit)/60)), 'hrs+.jpg'));

end

hold off; close; clear t; clear yM1Temp; clear i; clear fsM1Temp;

[James Tursa]

Can you work with datevec formats instead?

[Amit]

Hello James:

Won't that amount to the same thing? Unless you mean that I somehow forget about the date-month-year and focus only on time of the day, as that's what I want to be shown on the xlabels.

Good idea to try. Though I am currently not sure how to do it. Thanks indeed. If you have some quick illustration for me, that'll be amazing.