Improvement 1: Moving sum to further reduce noise

To further reduce the noise, a moving average which is also a low pass filter should be a good way. In order to avoid divide operation which is very slow, a moving sum is done instead. The moving window is selected to be 7 points. This operation could induce a little delay to the signal but this doesn't influence the final measurement. The signal after the moving sum is shown in the following figure. The filtering is successful and a lot of the noise peaks are removed.

Noise Cancelled Signal

Improvement 2: Make the measurement immune to DC shift

To make the measurement immune to DC level shift (very low frequency noise or breath influence), local maximum and local minimum are used to measure the period of the pulse. Two criteria are used to identify true pressure wave pulses.

Criterion 1-- valid pulse in amplitude: After the moving sum, the noise peaks on the rising and falling edges are removed. According to the experiment result, there might be some noise peaks at the local maximum of a pulse. The noise peaks should be ignored for correct detection. The most important thing is that the second signal peak in a pulse period (point F in following figure) should not be identified as a separate pulse. The solution is: every local maximum point and minimum point is detected. Once a local maximum is detected, it is compared with the most recent local minimum in the past. In the following figure, point B is compared with point A, point D is compared with point C, and point F is compared with point E. If the amplitude difference between the local maximum and minimum is bigger than the threshold TH1, then this is identified as a valid pulse in amplitude. By choosing proper TH1, C-D and E-F will be identified as noise and ignored.

Criterion for valid pulse in amplitude

Criteria 2 -- stable pulse in time: After a pulse is identified as valid, then the time of its local maximum is recorded. The difference between this time and the time recorded for last valid pulse is calculated as the period of this pulse. If the calculated period is within the predefined range, T1-T2, then this pulse is identified as a true pressure wave pulse. If five pressure wave pulses are detected in a row, then the system enters stable state. A status LED will light up to indict the user that the device is working properly. Once a pulse with an abnormal period is detected, the system will exit stable state and the status LED is turned off.

Improvement 3: Dynamically adjusted TH1 and T1, T2

After the system enters stable state, TH1 is updated to one third of the average amplitude of previous five pressure wave pulses and T1 and T2 are updated to the average period of previous five pressure wave pulses -10 points and +10 points, respectively. Once the period of a valid pulse in amplitude is not within T1 to T2, then all of the thresholds, T1, T2 and TH1 are reset to original values and the system exits stable state.

Advantages and drawbacks

This approach successfully solved the problem of DC level shift caused by low-frequency noise because the amplitude difference between adjacent local minimum and local maximum is independent of the DC level. But another problem arises: when there is a flat top or bottom, the method could not identify local maximum and minimum correctly, thus giving wrong results. Flat top is very common because for some users, the signal is so strong that the output gets saturated. To solve this problem, when a local maximum or minimum is detected, the following 5 points are ignored. But this is not a good method because the width of the flat region is different depending on the signal and this also results in inaccuracy of detection because the first point of the flat region is regarded as the local minimum or maximum point. Since the location of local maximum point is used to calculate pulse period, this is not a very accurate algorithm.

Improvement 1: Higher sampling frequency to improve accuracy

First, the sampling frequency is changed from 50Hz to 200Hz. This further improves detection accuracy and makes the pressure wave signal smoother when this curve is plotted on the connected webpage. The moving sum is kept and the window is changed to 18 points because the sampling frequency is higher.

Improvement 2: Combination of version 1 and version 2

To solve the problem of flat top or bottom and further improve detection accuracy, the first two versions are combined together to form the final solution. The basic idea is that on the rising edge, once a data point minus the previous local minimum is bigger than the threshold TH1, the pulse can be identified as a valid pulse in amplitude. Therefore, the point on this rising edge can be used to calculate the period and this is more accurate because the rising edge is sharp. The dynamic adjustment of TH1 is still controlled by valid local maximum and minimum. To combine these two methods together, two states are introduced: state up which means the rising edge and state down which means the falling edge.The state machine is shown below:

State Diagram

In the down state, if the value of the previous local maximum minus the current ADC value is bigger than a predefined threshold, TH1, and this is the first time that this condition is satisfied since last local maximum is detected, a LED is turned off. If a local minimum is found, the value of this local minimum is recorded and then the state machine enters up state.

The up state is more complicated and the process is shown in the following figure. First, if the value of the current ADC value minus the value of the previous local minimum is bigger than the predefined threshold, TH1, and this is the first time that this condition is satisfied since last local minimum is detected, then the LED is turned on and the time since last similar situation is calculated as the period of the pulse. If the period is within a predefined range, T1-T2, this pulse is considered as a true pressure wave pulse. Otherwise it is considered as bad signal and system is in an unstable state. If three (definition of Good) consecutive pressure wave pulses are detected, then the system enters stable state and a status LED is lit up to inform the user that the system is working well. In the stable state, the average period of most recent 20 pulses is used to calculate the heart rate (if number of stable pulses is less than 20, then only these stable pulses are used to calculate). The average period of these 20 pulses is also used to update T1 and T2: T1 is updated as N-30 in points and T2 is updated as N+30 in points. N is the average period in points. In this way, the standard for true pressure wave pulses is automatically adjusted. Once the system exits stable state, everything is reset so that the system can quickly switch to new conditions. In the up state, only local maximum is searched for. Once a local maximum is detected, the state machine will go to down state and the value of this local maximum is record. If the difference between this local maximum and the previous local minimum is bigger than TH1, then this is considered as a valid pulse in amplitude and TH1 is updated to half of this amplitude.

Flow Diagram of Heart Rate Detection int he UP State

Improvement 3: Fast adjustment of TH1

In this algorithm, once a valid pulse in amplitude in detected, TH1 is immediately updated to one half of this amplitude. There is a tradeoff here: this method can make quick adjustment and adapt to new environment very fast. This also prevents false measurement, as shown in the following figure: even if the second maximum, like F, is identified as a valid pulse in amplitude at the beginning, the next valid pulse is still identified and the threshold quickly rises to one half of this true valid pulse. Then the second maximum of the second pulse is blocked. The drawback of this method is that TH1 is easily influenced by large-amplitude noises. The good thing observed during experiment is that large-amplitude noise is rare after the system enters stable state, before the user takes the device off. Therefore, this method is doable.

Detection Analysis

To make the device workable for every user, the initial value of TH1 is very small to accommodate users who have very small signals. Then TH1 is updated to half of the amplitude once a valid pulse in amplitude is detected. If no valid pulse is detected in the next 1.5 seconds after a valid pulse in amplitude is detected, then TH1 is decreased by delta TH1 every 5 ms until a valid pulse is found or TH1 has reached the initial value which is very small. In this way, the system is applicable to every user and could quickly respond to changes. For the signal shown and the change of TH1 are shown in the following two figures. These two figures prove that TH1 changes very fast and appropriately.

Signal After Moving Sum

Again, in the stable state, the average period of 20 stable pulses are used to calculate the final output.

Dynamically Adjusted Threshold

In this section, the detailed data processing is analyzed using the waveforms measured from the experiment. The moving sum will be ignored here because it doesn't influence the nature of this algorithm.

Actual Pulse Example 1

The above figure shows three typical pulses. Using the algorithm described above, the local minimum at point A and the local maximum at B point will be detected. Because the amplitude difference between A and B is big enough, this pulse will be identified as a valid pulse in amplitude. The threshold for identifying valid pulses, TH1, will be immediately updated to half of the amplitude difference between A and B. Therefore, TH1 is approximately 75 after B is detected. Then C and D will be identified as the local minimum and maximum, but the amplitude difference between C and D is smaller than TH1, so they will be ignored. Next, E is identified as the local minimum and the state machine is in the state of rising edge. At point G, the amplitude difference between G and E is bigger than TH1, and then the time at point G is recorded. This time is then used to calculate the period of the pulse: T=TG-TH, where point H is similarly identified as point G.

Saturated Pulse Signal

Sometimes, the signal is so big that it gets saturated, as shown in the above figure. In the flat region, the algorithm will detect local minimum and local maximum alternatively and amplitude difference between the local maximum and local minimum is 0. Therefore, these local maximums and local minimums will be automatically ignored. Take the third pulse in the above figure as an example, only point A and point B will be identified as the valid local minimum and local maximum.

Saturated Pulse Signal

Another extreme case is that the amplitude of the effective signal is very small, as shown in the above figure. In this scenario, because the initial value of TH1 is even smaller, these pulses can still be identified as valid and stable pulses. Due to the adjustment mechanism of TH1, TH1 will automatically be small for small-amplitude signal and quickly arises to large value for large signals. When changes back to small signals, TH1 will also quickly drop to small value.

From the above analysis, it can be seen that this algorithm very robust. It successfully deals with noise problem and the amplitude issue of different signals. The final version of algorithm works well even in noisy environment. The detection accuracy is high and the details are in the result session.

System Overview

On the system level, two sensing tasks, heart rate sensing and temperature sensing, needs to be done. Another task is needed for communication between the MCU and webserver. The temperature sensing task just needs ADC conversion so this can be done in interrupt handlers. The other two tasks need separate functions to perform. The behavior of the system should be predictable and the time when the measurements are performed needs to be recorded. In this scenario, a real time system is used to schedule the tasks.

Real-Time OS Configuration

A customized real-time OS UART library trtUart.c is used in the program with an creation of two semaphores, SEM_RX_ISR_SIGNAL and SEM_STRING_DONE. The variable of heart rate is a shared variable between and the task of communication (task 1) and the task of heart rate measurement (task 2). The semaphore HEART_OUTPUT is created to protect this variable. Also, the task of measuring heart rate is only allowed to run after a new data point is ready. Therefore, another semaphore, DATA_READY is created to control the execution of task2. Semaphore HEART_OUTPUT is initialized as unblocked and all the others are initialized as unblocked.

Tasks Description

Task 1 communicates with the webserver and it is desirable to guarantee that the information is received or transmitted. Therefore, task 1 is assigned higher priority over task 2. The priority is assigned by giving task 1 an earlier deadline than task 2 when the two tasks are created and their deadlines are kept fixed.

In the communication between the MCU and the webserver, all the communications are initiated by the webserver. Therefore, task 1 keeps receiving instruction from the server and it is blocked until a "\r\n" is received. Once task 1 receives a "\r\n", it will perform different things according to the instructions received. The details about these instructions and reactions can be found in chart 1.

Task 2 is the heart rate detection algorithm and it is executed after a data point of the heart pressure wave is ready and it must be done before the next data point of the heart pressure wave is ready. So at the end of the task 2, the process waits for the next data point to be ready. And when the next data point is ready in the ISR, the ISR will signal task 2 to go. Task 2 is also in charge of saving data for the most recent 5 stable pulses. Two buffers are used to do this task. Once one buffer successfully records the data for 5 successive stable pulses, this buffer becomes the ready buffer and the other buffer will be in charge of recording data for the next 5 stable pulses. This guarantees that 5 successive stable pulses are recorded and the recorded data are up to date.

The data sampling of the heart wave needs to be accurate. Data points must be sampled at 5ms intervals to ensure correct calculation. Since interrupts have highest priority in a real time system, the sampling of data is controlled by the ISR. Timer 0 is used for the interrupt and the ISR is called every 1ms. In the 5ms interval, 2 milliseconds are assigned to the ADC conversion of the heart wave and the rest 3 milliseconds are assigned to the ADC conversion of the temperature. So at the end of the first two milliseconds, the ADC result is taken as a data point for the heart wave and the ISR signals task 2 to run. Then the ADC channel is switched and the conversion of temperature starts. At the end of the last 3 milliseconds, the ADC result is taken as a data point for the temperature. Then ADC channel is switched back and another conversion of the heart wave starts. In this way, both heart rate sensing and temperature sensing go well and the data points for the heart wave are sampled at 5ms intervals accurately.

Two external interrupts are set up to communicate with the server. One interrupt is triggered by button 1 push and the corresponding ISR resets the LEDs and vibration motor which might be set by the notification message from the server. Another interrupt is triggered by button 2 push and the correspond ISR set the emergency flag. As the server checks the emergency flag every 15 seconds, emergency emails and text messages will be sent out 15 seconds later after the user presses the emergency button in the worst case.

Communication Command Set

All the communications are originated from the web server side. As stated in the section above, the MCU respond according to the receiving command. A specific formatted string is sent out from the MCU to the server and being parsed on the web server.

As stated in the previous section, a data logger program was created to attempt to log the data periodically. However, due to the limitation that COM port can be accessed by one at one time, this approach was abandoned. Nevertherless, the software is created and the following explains how it has been created and the design theory behind it.

Program Overview

The date-logger is shown in the figure below. The data-logger has several fields. Both the temperature and the heart rate have to be saved in different files. In the text box where it says heart Rate File Name, one has to enter the appropriate location of the heart rate file. The same is true of the temperature file name. The Com port field allows one to choose from 9 COM ports. And the baud rate field allows one to choose from the following baud rates: 9600, 2400, 4800, 9600, 19200, 38400. For our projects purposes, we decided to stick with 9600. Once one has selected the baud rates, the com port and the file locations and names, one has to click communicate in order to activate the communication.

Date Logger Program

Code Description

A quick description about how windows operating system works is necessary in order to explain how the code was written. Windows operating system is very much based on the idea of an interrupt event driven operating system. It has these things called event handlers. For example when you click, the communicate check box, windows will call the following event handler in your visual basic code

Private Sub cbCommunicate_CheckedChanged(ByVal sender As System.Object, ByVal e As System.EventArgs) Handles cbCommunicate.CheckedChanged

So when the status of the check box changes ( which happens whenever the user clicks on the checkbox), the event handler will get called. Once this event handler gets called, the code will check to see if the box is checked or not. If the box is checked, it will disable the com port bbox, the baud rate box and the file name boxes, since we felt it unfeasable for the user to change the com port to which the receiver is connected to, or the baud rate while he or she was taking measurements. The code also enables the timer if the box is cchecked. However, if the box is unchecked,all of this is undone.

Once the timer is enabled, the event

Private Sub Timer1_Tick1(ByVal sender As Object, ByVal e As System.EventArgs) Handles Timer1.Tick
Dim heartRate As String

is called once every 15 milliseconds. Basically what happens in this events is that the Serial Port is opened. Next by using the write( String message) function in the serial port object that comes as one of the VB.NET libraries, we send the MCU a request for the heart rate using the string "H\r\n" or a request for the temperature using the string "T\r\n". When such a request is sent out, the program then waits until the a byte is received in the serial port buffer. Whenever this happens, another event called

Private Sub serialPort_DataReceived(ByVal sender As Object, ByVal e As System.IO.Ports.SerialDataReceivedEventArgs) Handles serialPort.DataReceived

is called. Whenver this gets called, we use the ReadExisting() function in the serial Port library, to store both the current heart rate and the current temperature in two global variables. Since we are expecting the heart rate and the temperature to be received from the datalogger, this event Handler also ends up closing the serial port after the temperature is received in order to allow the WebServer to access the com port whenever it needs to. In order to differentiate between the tempterature and the heart rate, the event handler has a static variable that keeps track of the number of times the event is called. Every even number of times the event is called, the software assumes that the heart rate is received from the MCU. Every odd number of times this evnet gets called, the data that is received will be the termperatrue. The reason why we can make this assumption is because the data logger is always sending out both rquests for the heart rate and the termperature one right after the other, and when the MCU sends the data, this event gets called once for the termperatrue and once for the heart rate. Next, this event handler will store the termparute and the heart rate in their corresponding files. But before it can do this, it will of course have to open it. Please note that since two programs cannot access the same file at once, the program will wait until the files are accessable in order to access them. Once the data has been written to the appropiate files, the data logger will then close the files so as to give access to the web-server.