January 2005, Issue 174

Light-to-Frequency Conversion (Part 2)
Pulse and Oxygen Content

---

PROJECT

Let’s take a closer look at how you can use a TSL230R to monitor a heart rate. The push buttons allow for manual changes to the system so you can determine the optimum settings even during program execution. I predefined PB1 as an LED toggle between red and IR LEDs. PB2 was predefined as a Sensitivity mode increment to the TSL230R. PB3 was predefined as a Divisor mode increment to the TSL230R. I used these inputs to manually change the TSL230R’s settings and observe the frequency output of the device for each of the modes. Because these are all based on the light source, it was convenient to choose between the two sources for light, the red LED and the IR LED. These LED colors were chosen because they’re within the TSL230R’s bandwidth and they have different absorption characteristics based on the amount of oxygen in the blood. I tried choosing LED currents that would allow the LEDs to have similar intensities based on having no objects between the LEDs and the TSL230R sensor.

The sensor’s frequency output is connected to the external interrupt input on a microprocessor. The frequency’s period is measured by counting timer ticks between any two consecutive rising edges (interrupts) of the input signal. Last month I showed you that the best mode to use would be that which produced the highest count without overflowing the 16-bit (1 µs/count) counter.

The maximum count of Timer1 is set at 31.25 ms (or 1/32 s) by reloading with a constant each overflow, and it becomes the sampling rate. However, if the tick count exceeds half that value, there is a good chance a count during a successive sample period will overflow because the count period of the sensor and the sample times are asynchronous. So you can trap any count over this value and assign the 16-bit count an 0xFFFF value, which indicates a count that’s over the maximum limit (or a timer overflow, which would also be over that limit).

Just as the count for a low-frequency output can be too long, a count can be too short for a high-frequency output. A short count is limited by how fast the count recovery routine (interrupt) can be serviced. For obvious reasons, counts never will be less than this value, but they could be illegal for counts up to twice this value. 

With the establishment of maximum and minimum count limits, a sample can be taken while the TSL230R cycles through each mode to automatically determine the best one to use (while the sensor is connected to a patient). Mode selection must be done separately for each LED because there is no guarantee the LEDs will provide acceptable counts using the same mode for each.

A sample output similar to Figure 1 can be recorded at this point. To help identify the maximum and minimum sample counts, positive and negative peak detectors are implemented in code according to certain rules (see Figure 2). Like a hardware peak detector in which a voltage passes through a diode to charge a capacitor, anytime a sample is greater than the last saved maximum or less than the last saved minimum, it’s saved as the new maximum or minimum peak (see Figure 3).

(Click here to enlarge)

Figure 2—Maximum and minimum samples are held for use when determining both the number of samples between maximum peaks (beats per minute) and the ratio of the AC count and the DC count.

 

(Click here to enalrge)

Figure 3—Code algorithms in this project act like a hardware peak detector. The diode allows the capacitor to charge and hold the highest voltage applied to VIN. The resister can be sized to slowly leak off the capacitor’s charge (VCAP).

To prevent cases in which there’s a shift downward (or upward in the case of minimums) because of a change in DC level or a change in the absorption constant, and a new maximum peak can’t occur, you must leak off some charge or allow the maximum to work its way down (like a resistor discharging the capacitor). To determine the rate of leakage, find the mean between the maximum and minimum peaks and then reduce the former (or increase the latter) by some factor of it.

I temporarily programmed PB4 to manually cycle this value by a factor of two each time (1, 2, 4, … 64, 128, 1, 2, etc.). This leakage adjustment recalculates the maximum and minimum peaks each sample time, except when a new peak is detected and held until the slope changes state. I ultimately let the program automatically select the leakage factor by looking at sample data. Too little leakage can cause a maximum or minimum peak to be missed (if the DC level shifts), while too much leakage can cause the recognition of false peaks. 

Referring to Figure 1, notice how a rising slope changes state when a sample falls below the leaking minimum value. The opposite is true for a falling slope: it changes state when a sample exceeds the leaking maximum value. Not only is the slope affected by this action, but a change-of-state (COS) flag is recorded for both a positive slope [SLOPE=1 (COS_SLOPE1)] and a negative slope [SLOPE=0 (COS_SLOPE0)]. As you can see in Figure 2, special events occur during the execution of the sampling loop. This is when the number of beats per minute (bpm) and O2 can be displayed.

The DoSomething flag is set each time the slope equals one and the COS_SLOPE0 equals one, which means the slope has just changed to one. A fortune telling circuit would be handy because there’s no way of knowing that any particular peak sample will be the maximum before the wave’s slope changes. Wait until some point when you can be sure that this has happened. When a sample value falls lower than the leaking minimum peak (a slope change), you can be sure you have the last maximum peak. The CNT (number of samples since the last determined peak) is adjusted to reflect the real CNT where the new maximum peak has occurred, such that it will reference that new maximum peak the next time the CNT is saved during a new maximum detection.

The new maximum peak value is saved. A ratio is calculated using this new value along with the last minimum peak value. The ratio is a number that represents the AC value (the difference between maximum and minimum peak samples) divided by the DC value (the mean sample). The ratio is calculated and stored separately during red and IR LED operations. Because this is the point in time when a positive peak has indeed occurred, enabling the piezo beeper will give an audible indication of such an event. By simply enabling Timer0, background interrupts from this timer overflowing will allow the interrupt routine to toggle the piezo output once each overflow a number of times before disabling itself, self-completing a short beep. Although some calculations are performed in this portion of the code, most display routines are handled at the end of the loop. All display routines are serial datastreams. Although this project requires connecting to a PC running HyperTerminal to see the data, it very easily could be formatted for a small LCD.