Real-time heart rate calculation

Question

bilal malik 2017-3-22

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/331333-real-time-heart-rate-calculation

评论： Iqra Chaudhary 2019-12-14

I have a real-time(phonocardiograph) plot which displays data from an electronic stethoscope via the COM port. I am trying to calculate the heart rate but having little success. I have created a uicontrol on the figure to display the BPM and have included the formulas to count the peak signals(beat_count). At the momement 'BPM:' displays on the plot figure but without a value so I am unable to get any BPM_avg reading.

 a = arduino('COM4','Due');
 y=0;
 hPlot = plot(NaN);
 intervalSize = 200;
 currentInterval = 200;
 t = 1; % number of samples
 atInterval = 1;
 beat_count = 0;
 quitbutton = uicontrol('style','pushbutton',...
   'string','Quit', ...
   'fontsize',12, ...
   'position',[10,2,50,20], ...
   'callback','quitit=1;close');
quitit = 0;
bpmtext = uicontrol('style', 'text',...
    'string', ['BPM: '],...
    'fontsize', 12,...
    'position', [80, 2, 100, 20]);
 while(1)
    k = 1;
    while(t<currentInterval)
        b=readVoltage(a, 'A0');
        y=[y,b];
        if ishandle(hPlot)
            set(hPlot, 'YData', y);
        else
            break;  % break out of the loop
        end
        xlabel('Samples')
        ylabel('Voltage')
        title('Phonocardiogram')
        axis([currentInterval - intervalSize,currentInterval,0,3]);
        %grid
        t=t+k;
        pause(0.002)
    end
    for m = 2 : length(b)-1
    if(b(m) > b(m-1) & b(m) > b(m+1) & b(m) > 2.4)
        %disp('Prominant peak found');
        beat_count = beat_count + 1;
    set(bpmtext, 'string', ['BPM: ',...
                            num2str(BPM_avg,4)]);
end
    end
   currentInterval = currentInterval + intervalSize;
   atInterval = atInterval + 1;
   if ~ishandle(hPlot)
       break;
   end
fs = 500;
N = length(b);
duration_in_seconds = N/fs;
duration_in_minutes = duration_in_seconds/60;
BPM_avg = beat_count/duration_in_minutes;
 end

2 个评论
显示无隐藏无

Star Strider 2017-3-22

I cannot follow your code. You likely need to be doing real-time signal processing of the sounds to isolate the ones you need from the ones you do not.

Note that the Phonocardiogram consists of as many as 4 usually distinct sounds, and varies over the respiratory cycle. It is normal in healthy people for the second heart sound, ‘S2’ (closure of the aortic and pulmonic valves), to be ‘physiologically split’, so that they do not occur at exactly the same time. I refer you to any textbook on cardiac physical diagnosis for the details and a discussion of the relevant physiology.

Iqra Chaudhary 2019-12-14

How to find heart beat rate of an audio wav file without using arduino?

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Geoff Hayes 2017-3-22

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/331333-real-time-heart-rate-calculation#answer_259898

bilal - it doesn't look like BPM_avg is defined before you try and use it, so I'm surprised that you are not getting an error when you call

set(bpmtext, 'string', ['BPM: ', num2str(BPM_avg,4)]);

Also, why is the above in a for loop that does not do anything to BPM_avg

 for m = 2 : length(b)-1
    if(b(m) > b(m-1) & b(m) > b(m+1) & b(m) > 2.4)
        %disp('Prominant peak found');
        beat_count = beat_count + 1;
        set(bpmtext, 'string', ['BPM: ',num2str(BPM_avg,4)]);
    end
 end

I think that that is part of the problem - BPM_avg is not being set or updated when it should be. Why not update it when you receive the data from the Arduino instead having this for loop outside of the while loop?

19 个评论
显示 17更早的评论隐藏 17更早的评论

bilal malik 2017-3-22

编辑：Geoff Hayes 2017-3-23

Thank you for your response Geoff. I have made the changes that you have mentioned but the only difference i notice is that in the first interval (from 0 to 200 samples) it displays 'BPM:' on the plot figure whereas in the rest of the intervals(200 - 400, 400-600 samples etc) it displays 'BPM: 0'. So basically it's made minimal difference and hasn't solved the problem. Just for reference, the changes applied are demonstrated in the code below:

a = arduino('COM4','Due');
y=0;
hPlot = plot(NaN);
intervalSize = 200;
currentInterval = 200;
t = 1; % number of samples
atInterval = 1;
beat_count = 0;
quitbutton = uicontrol('style','pushbutton',...
    'string','Quit', ...
    'fontsize',12, ...
    'position',[10,2,50,20], ...
    'callback','quitit=1;close');
quitit = 0;
bpmtext = uicontrol('style', 'text',...
    'string', ['BPM: '],...
    'fontsize', 12,...
    'position', [80, 2, 100, 20]);
while(1)
    k = 1;
    while(t<currentInterval)
        b=readVoltage(a, 'A0');
        y=[y,b];
        if ishandle(hPlot)
            set(hPlot, 'YData', y);
        else
            break;  % break out of the loop
        end
        xlabel('Samples')
        ylabel('Voltage')
        title('Phonocardiogram')
        axis([currentInterval - intervalSize,currentInterval,0,3]);
        %grid
        t=t+k;
        pause(0.002)
    end
    currentInterval = currentInterval + intervalSize;
    atInterval = atInterval + 1;
    if ~ishandle(hPlot)
        break;
    end 
    fs = 500;
    N = length(b);
    duration_in_seconds = N/fs;
    duration_in_minutes = duration_in_seconds/60;
    BPM_avg = beat_count/duration_in_minutes;
    set(bpmtext, 'string', ['BPM: ',...
                        num2str(BPM_avg,4)]);
end 
for m = 2 : length(b)-1
    if(b(m) > b(m-1) & b(m) > b(m+1) & b(m) > 0.4)
        %disp('Prominant peak found');
        beat_count = beat_count + 1;
    end
end

bilal malik 2017-3-23

Thank you for mentioning the debugger, it is a very useful tool. I have changed the positioning of the calling of the beat_count however it is always displaying 0 (even in the first interval). Whatever changes I make I just can't seem to get a reading for beat_count.

 a = arduino('COM4','Due');
 y=0;
 hPlot = plot(NaN);
 intervalSize = 200;
 currentInterval = 200;
 t = 1; % number of samples
 atInterval = 1;
 beat_count = 0;
 fs = 500;
 quitbutton = uicontrol('style','pushbutton',...
   'string','Quit', ...
   'fontsize',12, ...
   'position',[10,2,50,20], ...
   'callback','quitit=1;close');
 quitit = 0;
bpmtext = uicontrol('style', 'text',...
    'string', ('BPM: '),...
    'fontsize', 12,...
    'position', [80, 2, 100, 20]);
 while(1)
    k = 1;
    while(t<currentInterval)
        b=readVoltage(a, 'A0');
        y=[y,b];
        if ishandle(hPlot)
            set(hPlot, 'YData', y);
        else
            break;  % break out of the loop
        end
        xlabel('Samples')
        ylabel('Voltage')
        title('Phonocardiogram')
                for m = 2 : length(b)-1
            if(b(m) > b(m-1) && b(m) > b(m+1) && b(m) > 1.0)
        beat_count = beat_count + 1;
            end
                end
        N = intervalSize;
        duration_in_seconds = N/fs;
        duration_in_minutes = duration_in_seconds/60;
        BPM_avg = beat_count/duration_in_minutes;
        set(bpmtext, 'string', ['BPM: ',...
                            num2str(BPM_avg,4)]);
        axis([currentInterval - intervalSize,currentInterval,0,3]);
        t=t+k;
        pause(0.002)
    end
   currentInterval = currentInterval + intervalSize;
   atInterval = atInterval + 1;
   if ~ishandle(hPlot)
       break;
   end                        
 end

Geoff Hayes 2017-3-25

That means that every time that you read from the arduino, you are only getting one value (a scalar) as opposed to the assumed array of data. So your code to count the beats never executes because your loop is really

for m=2:0

since length(b) == 1. You will instead need to collect enough elements into y before you can start counting the beats. For example, you could try

 y = 0;
 atIndex = 2;
 % etc.
 while(1)
    k = 1;
    while(t<currentInterval)
        b=readVoltage(a, 'A0');
        y=[y,b];
        if length(y) >= 3
            if y(atIndex-1) < y(atIndex) && y(atIndex) > y(atIndex+1) && y(atIndex) > 1.0
                beat_count = beat_count + 1;
            end
            atIndex = atIndex + 1;
        end
        % etc.
    end
 end

We use atIndex to remember where we have last done our comparison within y. We increment the beat count if the conditions (like before) are satisfied. Once we have done the comparison, we increment atIndex so that we can move on to the next index on the subsequent iteration (after we have read in another new value).

The above code is untested but I think it should work. It assumes though that b is always 1x1. If more than one element, you probably will want a combination of the above with your original code.

Geoff Hayes 2017-3-26

I'm not sure what you mean by " If more than one element..."? When would that be the case?

You've written code assuming that b is an array of at least three elements when in fact (by your previous comment) it is exactly one element and so you have to write code to adjust for that (which I've tried to provide in my previous comment).

As for the heart beat count not being correct, you set your sampling frequency to 500

fs = 500;

but you pause for 0.002 seconds which, since your currentInterval is 200, means you are collecting 200 samples every 0.4 seconds. As you are counting the number of beats for this interval, then if you translate this to beats per minute then

 beat_count per 0.4 seconds
 beat_count/0.4 per second
 (beat_count/0.4)*60 per minute

So if there is 1 beat every 0.4 seconds then

beats per 0.4 seconds
5 beats per second
beats per minute

Which matches the 150 that you are observing. This doesn't seem to be correct so I think that you have to verify the sampling frequency or your threshold that determines whether the signal is a beat (or not).

As for the fact that your beats per minute grows over time, this is because you are accumulating all your beats from one interval to the next. You will want to reset the beat count prior to any interval i.e. before you enter the inner while loop

 while(1)
    k = 1;
    beat_count = 0;
    while(t<currentInterval)
        % etc.
    end
 end

Geoff Hayes 2017-3-30

bilal's answer moved here

Hi Geoff, I have been working on my code and a problem I came across was the line:

y=[y;b];

It causes a reallocation of y through each loop and every time it is executed the time required grows.I have tried to pre-allocate the array but when the waveform plots it just has a line going across the x-axis during the first interval, and it doesn't pick up the signal or move onto the next interval.

Could you please advice on the issue.

 a = arduino('COM4','Due');
 y=0;
 atIndex = 2;
 hPlot = plot(NaN);
 intervalSize = 500;
 currentInterval = 500;
 t = 1; % number of samples
 atInterval = 1;
 beat_count = 0;
 fs = 50;
 duration_in_minutes = 0;
 Nsamples = 100000;
 quitbutton = uicontrol('style','pushbutton',...
   'string','Quit', ...
   'fontsize',12, ...
   'position',[10,2,50,20], ...
   'callback','quitit=1;close');
 quitit = 0;
bpmtext = uicontrol('style', 'text',...
    'string', ('BPM: '),...
    'fontsize', 12,...
    'position', [80, 2, 100, 20]);
 while(1)
    tic
    k = 1;
    while(t<currentInterval)
        b=readVoltage(a, 'A0');
        %length(b)
        y=zeros(Nsamples,1);  % preallocate array 
      for i=1:Nsamples
       y(i)= b;    
        if ishandle(hPlot)
            set(hPlot, 'YData', y);
        else
            break;  % break out of the loop
        end
        xlabel('Samples')
        ylabel('Voltage')
        title('Phonocardiogram')
        %Calculate Heart Rate
        if length(y) >= 3
            if y(atIndex-1) < y(atIndex) && y(atIndex) > y(atIndex+1) && y(atIndex) > 1.5
                beat_count = beat_count + 1;
            end
            atIndex = atIndex + 1;
        end
        % Divide the beats counted by the signal duration (in minutes)        
        N = currentInterval;
        duration_in_seconds = N/fs;
        duration_in_minutes = duration_in_seconds/60;
        BPM_avg = beat_count/duration_in_minutes;
        set(bpmtext, 'string', ['BPM: ',...
                            num2str(BPM_avg,4)]);
        axis([currentInterval - intervalSize,currentInterval,0,3]);
        t=t+k;
        pause(0.002)
        duration_in_minutes = duration_in_minutes  + duration_in_seconds/60;
    toc
      end
    end
   currentInterval = currentInterval + intervalSize;
   atInterval = atInterval + 1;
   if ~ishandle(hPlot)
       break;
   end                        
 end

Geoff Hayes 2017-3-30

hi bilal - the reason for the straight line is because of

 y=zeros(Nsamples,1);  % preallocate array 
 for i=1:Nsamples
    y(i)= b;  
    % etc.
 end

You are fixing all Nsamples of y to be b!

You need to pre-allocate the y array outside of the outer while loop and then update the appropriate element whenever you read in b.

 a = arduino('COM4','Due');
 Nsamples = 100000;
 y=zeros(Nsamples,1);
 atIndex = 2;
 hPlot = plot(NaN);
 intervalSize = 500;
 currentInterval = 500;
 t = 1; % number of samples
 while(1)
    tic
    k = 1;
    while(t<currentInterval)
        y(t)=readVoltage(a, 'A0');
        if ishandle(hPlot)
            set(hPlot, 'YData', y(1:t));
        else
            break;  % break out of the loop
        end
        % etc.
    end
 end

The above assumes that you will only read Nsamples (you may need to make this a larger number) and that only one sample is read from the arduino when we call readVoltage(a, 'A0'); (and that t is incremented by one on each iteration of the inner while loop. We then plot only up to the elements of y that we have populated

set(hPlot, 'YData', y(1:t));

I haven't tested the above but I think it may work...

dpb 2017-3-30

...
while(t<currentInterval)
  b=readVoltage(a, 'A0');
  %length(b)
  y=zeros(Nsamples,1);  % preallocate array 
  for i=1:Nsamples
    y(i)= b;    
    ...

You've convoluted up the stuff I gave you in the sidebar email without again thinking much about what you're really doing it seems...

Above you start the collection, read a sample and only THEN preallocate an array. You then go thru a loop over a number of samples to collect and simply stuff the one reading into that array...hardly useful.

I gave you the outline as

initializeArduinohere
nSamples = HoweverManyYouWant;
y=zeros(nSamples,1);   % preallocate
t=zeros(size(y));      % to collect the time of the conversion
tic                    % start the time
for i=1:nSamples       % start the collection
  y(i)=readVoltage(...   % read the A/D
  t(i)=toc;              % save time
end

Now post-process the time history.

Your timing experiments have shown that as set up without other ways to figure out where all the CPU cycles are going that the

a = arduino('COM4','Due');
while(1)
    tic
        b=readVoltage(a, 'A0');
    toc
    end
gives:
Elapsed time is 0.146963 seconds.
Elapsed time is 0.083865 seconds.
Elapsed time is 0.046447 seconds.
Elapsed time is 0.045178 seconds.
Elapsed time is 0.050683 seconds.
Elapsed time is 0.019395 seconds.
Elapsed time is 0.019464 seconds.
Elapsed time is 0.019031 seconds.
Elapsed time is 0.017789 seconds.

that the best you're doing without any other processing is only about 50 Hz and some cycles are as much as 4-7X that.

As we've spoken off-forum, until you can figure out what's going on there, you're wasting time worrying about other stuffs.

As I've said elsewhere, I've never seen an Arduino so only know what I can find online, but the doc's at their site seem to indicate with the C interface readDigital can run at about 10 kHz; don't know how TMW can manage to have slowed it down to this extent, but there's where your biggest problem is. It would seem you should download the direct access IDE and try a "sketch" as they seem to call a program directly to see how that performs--you may be able to collect the data that way and then use Matlab to post-process it, but at this point it seems acquisition under Matlab is going to be problematical at best.

bilal malik 2017-3-30

Geoff - The code does solve the pre-allocation issue however the problem is that the BPM count isn't counting anymore and it is just stuck on 0.

请先登录，再进行评论。

Answer 2

dpb 2017-3-30

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/331333-real-time-heart-rate-calculation#answer_261044

编辑：dpb 2017-3-30

I'm going to put this as an Answer to emphasize the point again...your timing code of just a single read in a tight loop shows that whatever is going on in the Matlab interface to the Arduino readVoltage routine at least on your system it's just killing performance. I again STRONGLY suggest you read the blog of another experimenter with the Due for some very informative background on using the Due and on data collection in general.

It's clearly possible when running code directly on the board to run at much higher rates; as said above I don't understand how there can be as much latency using the Matlab interface, but until you can find some resolution there what you're trying to do just isn't going to work.

PS. I wonder if since there are 16 channels on the Due if they're all active the sample rate gets divided down by polling them all and just returning the one asked for, maybe? I didn't see where it says whether the A/D sampling is simultaneous or sequential, only that is 1MHz capable. As have emphasized off the board in our sidebar conversation, you need to dig into the guts of the board and the Matlab interface much more deeply to figure out what's going on here.

I don't suppose it could be some sort of communication issue slowing it down, is it? Can you test that somehow?

Mayhaps ask on one of the demo blogs from TMW what you should expect in a tight loop or, if you or your university has support, submit a support request and see if TMW will respond.

ADDENDUM

Btw, the last timing you sent is most curious...I pasted into a variable tacq and one gets the following interesting figure--

For some reason, the acquire time goes down with each loop; dramatically at the beginning and then looks like is approaching a steady-state rate, maybe. I don't know what would cause this behavior; seems very odd, indeed.

Why don't you try the above loop with nSamples at some sizable number and collect a waveform or two and see what you get? If you approach 60 hz as it appears this is, how long a time series do you want? Multiply N seconds by 60 and collect that many samples. Of course, do the preallocation and collect both the reading and the transpired time as shown and

plot(t,y)

to see what you get. Post those results back here. It would, of course also be informative to just collect the timing information for several trials of a large number to see if the above is consistent and what steady-state rate is. You might want to "pretrigger" by some number of samples before recording "real" data...

16 个评论
显示 14更早的评论隐藏 14更早的评论

dpb 2017-3-30

编辑：dpb 2017-3-30

TMW--The Mathworks, owners/publishers of Matlab.

How could taking a set of data with a loop similar to your above to see what the result is on timing be out of the scope of what you're trying to do?

If you're talking about the link, the importance there is the background insight into what the DUE board is doing and at least partially why speed isn't what you think it might/should be--altho there's still a big difference from what he gets directly from the device vis a vis what you're showing through the Matlab gateway (another level of distance from the board).

IF your objective is as was stated, the scope has to include figuring out a way to retrieve data fast enough and on a repeatable enough timing to be able to analyze it. So far it doesn't seem that either of those objectives may be achievable this way; does that change the scope or does it mean you've got to find another way to get to the end objective? Have you had the discussion with your advisor I suggested on the problems you're encountering yet?

Specifically, again, what I'm suggesting is put the following in a script file and run it...

a=arduino('COM4','Due');
n=input('How many seconds to run roughly)? ');
nSamples=60*n;   % assume roughly 60 Hz from past experience for now
t=zeros(nSamples,1); y=zeros(size(t));  % preallocate for time, data
tic                           % start the timer
for i=1:nSamples              % loop over the number set
  y(i)=readVoltage(a, 'A0');  % read the A/D, save
  t(i)=toc;                   % save elapsed time since tic accumulative
end
plot(t,y)               % see what you got
save('Timing','t','y')  % save to .mat file for further looking at

dpb 2017-3-30

It's waiting for you...answer the question. :)

>> n=input('How many seconds to run roughly)? ')
How many seconds to run roughly)? 5
n =
   5
>>

If it isn't showing the question, you've got a coding error somewhere...

Well, normally you create a serial port object via serial and then connect the device with fopen using that object handle. You can then set the properties of that port object. That somehow must all be inside the arduino code that creates that object that you address. I see there's then a read-only port property but don't see anywhere that is allows setting baud rate so I presume it must just be default. A gargle search finds references to 115200 baud, though, so if it's that high it's not up to USB speeds but wouldn't seem like it should be limiting to as slow a cycle rate as you're seeing. So, "I dunno"; seemed like a good question.

I did see here that you can load user-written libraries which apparently are way to package code to run on the Arduino; looks like that is the way you get to more in-depth control with Matlab-only interface.

I thought there was an optional USB terminal port; doesn't appear as it is set up that way on looking more; the USB port seems to be available to stream data over but guess it would be writing one of these libraries to be able to load that to the Due to execute.

But, try

a.port

after you execute the arduino command and see what it shows you...

I'm still interested in seeing what you get if you run the timing 'spearmint above a couple of times (note you'll have to change filename or you'll overwrite previous .mat file so be sure you're done with the first set of data or make that change before doing it again).

dpb 2017-3-31

编辑：dpb 2017-3-31

<arduinoio setup-and-configuration>. See link to <propertylist> at that page. Looks like it's 'Port' not 'port'...need to be more aggressive in looking for minor errors such as typos, it seems. (Apologize if seems "preachy", but ... :) )

What happens if you just open a connection at the command line and type the object name used? Will it not echo all the properties for you? You're not taking much advantage of the interactive nature of Matlab if you don't use command line extensively for such exploration (or use the debugger inside a script/function the same way when the object/variable is in scope).

For the data, well, looks smoother than what was before; if as we presume it's only approximately 60 Hz sample rate and that's the beat rate, you'd expect to asynchronously hit/miss the peaks and that's pretty-much what it looks like it did.(*) Did you look at the timing information to see what the timing actually is/was? Look at diff(t) to see (since we captured the running time from trigger, that's the dt); a plot of those against observation number and 1/dt for instantaneous sampling frequency would be interesting to see how it behaved with time; does the same initial delay occur and it reaches a near steady state? If so, what's that final sample rate? How uniform is that sample rate? So many questions, so few answers (so far)...attach the .mat file.

Again, think about what that experiment can tell you and try to get the most out of it you can...it's probably going to confirm what we think we already know that unless you figure out where the overhead is that polling isn't going to work, but it would surely seem you'd want to know just what it is that it can and is doing...

It seems hard to have to coax such information/explorations out...

(*) Thought 'spearmint--What would happen if you had a pure sine at 60 Hz and sampled at 60Hz? Now, change the sample rate to 60Hz +/-few Hz. How does the sampled waveform change? Now repeat both of those with a changing phase shift of where the initial sample is taken and you should begin to understand what you're looking at.

dpb 2017-3-31

编辑：dpb 2017-3-31

"Not sure it makes much sense to me."

Well, besides the comments above, it's clear directly from the plot that the previous estimate of somewhere in the 60 Hz polling rate is maintained; the overall end time is 6+ sec in first; interestingly however it's 5+ sec in the second (presuming, of course, you asked for 5 sec in running both and entered that value without a typo); the internal assumption of needing 60X that number of samples to get the time span seemed to hold up moderately well. The question remains then why there still seems if both were 5x60-->300 samples, the 1 sec overall time difference. Also, there appears to be a sizable lag for the first acquisition; granted I put the initial tic before the for loop to avoid the need for a branch around it inside the loop so there should be a minor delay, but wouldn't expect what it appears is the length of that one. Ergo, one must presume there's something going on regarding the acquisition about timing for it. And, of course, what happens if you subsequently sample for longer time periods; how stable will if finally get and what is the final true steady-state rate as one would presume must be?

The second trace almost looks like lost connection to the device or somesuch; looks mostly like noise but still the timing is useful as that was the prime thing looking for here. ADDENDUM: Actually the above thought 'spearmint well may play into the answer here as well...

Oh! The other thing need to do...rearrange the test code so the Arduino connection is made first, before the timing. Then, pass the object to the test function. This way if you re run multiple times you won't have whatever initialization effects there are on the board impacting the beginning of the test; as it's written now you're going through that every time; only need to do that once a day or after a major crash or something, not routinely.

I didn't look to see; is there some cleanup code to close a connection; is there, by any chance, any problems arising with having issued a zillion arduino commands to create a connection without cleaning up after prior ones before making a new one even though you reused the Matlab variable? Don't know, just a thought...

dpb 2017-4-1

编辑：dpb 2017-4-2

By "device" I mean the instrument you're reading with the Arduino.

You did read the comments of the previous link about stability of the reference power and all that if just use the USB port for the 3V reference, right? Have you tried a known input voltage from a reference and then from a signal generator to ensure you can actually measure a known input?

Reason for the last is apparent from attached which is from the last of the three files above--

The top of those is

plot(t,-y)

The middle is same except expanded by

xlim([15 30])

to observe the first peak/disturbance area more closely. The third is from

dt=diff(t);        % the sampling interval
i1=find(t>15,1);   % the index corresponding to 15 sec
i2=find(t<30,1,'last');  % the 30 sec mark
plot(t(i1:i2),dt(i1:i2)  % and the sampling interval corresponding

What is most disturbing is that all the peaks in the sampled waveform are exactly correlated in time with an increase in sampling time as I had somewhat postulated might be going on earlier.

I think this shows the measurement is mostly noise induced from the PC and/or the Arduino board itself; NOT the monitor device. Is it, in fact, even hooked up or operating at all? Doesn't look like it to me from here...

Looks like time to break out your EE training... :)

ADDENDUM

Finishing up on some looking at the sampling rate...

First of those is just the instantaneous frequency (1/dt) from each sampling interval vs time which shows it is basically a random value about an approximately constant mean (the solid line is a the fitted linear regression; altho the slope isn't zero; the line is

fhat=0.0177*t+58.6
>> [fhat(2) fhat(end) fhat(end)-fhat(2)]
ans =
 58.6928   60.8584    2.1656
>>

(The reason for f(2) for first is there's one less element in f=./diff(t); than in t so for convenience I augmented it with the predicted fhat(0)). Colors weren't easy but there's a barely discernible dotted white line above/below the predicted line that's the 1-sigma bound. Those values were

>> sd=std([fhat(1);f]-fhat);  % sd from regression
>> [mean(f) sd sd/mean(f)*100]
ans =
 59.7833    6.7612   11.3094
>>

The second is just histogram that does show essentially a normally distributed value around the mean with a small cluster of longer time intervals (ergo, lower sampling rate). These would be expected latencies inherent in the occasional task-switching, etc., under Windows when polling.

While if it were sufficient to see the information you're looking to capture you might be able to live with the jitter to simply display waveforms, it would be difficult to do any very precise timing. Since the actual beat rate is roughly the same rate as this sampling rate, it's pretty clear that this won't work as a data collection scheme as we had pretty much already concluded; just needed to really make sure that is so without other distractions.

As noted before, perhaps there are some ways to make the polling run faster and it does really seem incredible that there is such a delay going on, but unless you can figure out where all the clocks are going, this is "just the way it is"--you've pared the loop down to nothing but the request for a new sample and that's as fast as it ran and it didn't do anything different over a period of time. That indicates there isn't some initialization going on that the first couple of very short samples looked like might be and that the achievable rate might possibly be better. I can't for the life of me imagine why it should be so slow, but can't deny the results.

dpb 2017-4-1

编辑：dpb 2017-4-2

And, since the top two plots are -y, what are peaks there are actually dips; I first (as mentioned in passing above) thought you had likely inadvertently switched polarity on the device and so thought should be the inverse of the readings.

Then when comparing the sampling rate, observed the coincident nature of the peaks illustrated. But, what may actually be happening is that the board is loading the input during sampling and pulling down the input signal and the values are actually as recorded. That would indicate the USB power supply isn't able to supply the device with all the current it needs so the voltage drops.

Fixing that with external reference/power supply wouldn't solve your timing problems, but at least you might then get clean data. As mentioned before, looks like it's time to go back to ground zero and follow a very precise sequence of steps to uncover just what is actually happening on the board. The outline in that previous link of what that individual did to confirm his results is, as noted before, a pretty good tutorial on how to approach a new piece of hardware; particularly one such as this that is open source and while a neat piece of gear, NOT guaranteed by the manufacturer to do much of anything as, say, one can rely on a Tektronix or Agilent lab instrument to actually measure what it says it measures; you really don't know that here without confirming it independently.

PCs (and computers in general) are just so rife with noise issues that without careful attention to grounding, shielding, etc., etc., it's easy to measure something but not necessarily anything much related to what it is one is trying to measure. I'm sorry to say it looks a lot like that may be what's happening here.

dpb 2017-4-2

I was asking about how it is physically connected to read the device? How the data transfer occurs is immaterial to trying to solve your instrumentation issues.

What was in the arduino.m file out of curiosity? Is it m-code too or did it drop down into a mex file or does it show an Arduino "sketch" or what?

Again, I'm still curious about whether there was anything attached to the Arduino on those timing runs that was powered up???

bilal malik 2017-4-2