Did you know that the ear is the best place to measure your heart rate?

Fitness trackers can measure heart rate and other vital signs to help users set up daily exercise routines. Fitness trackers often have built-in motion sensors that can detect movement patterns to help differentiate between walking, running, and swimming, so they can also be used as pedometers. For comfort and convenience in daily life, measurements are often taken on the wrist, as sensors can be placed in accessories such as watches, jewelry, and wristbands. However, this location is not optimal for measurement quality. Heart rate detection is limited by motion artifacts and is difficult to perform, as the relatively large muscle mass limits contact with arteries.

In contrast, the ear is much more suitable for optical heart rate measurement. The earlobe has been used by medical professionals to measure blood oxygen levels. But until now, this has not been fully exploited at the consumer level because ear-based measurement devices are space-constrained and very power-hungry, requiring large batteries. But with the introduction of highly integrated, lower-power chips, ADI has developed a solution to these problems. It is now possible to integrate an effectively functioning vital sign measurement device into a typical in-ear headphone. The improved responsiveness opens up entirely new applications and possibilities. This paper describes and evaluates the system.

The basic measurement method is optical. The measurement uses short pulse signals from up to three LEDs. The LED current can be up to 370mA with a minimum pulse width of 1μs. The optimal wavelength of the LEDs is selected depending on the measurement location and the measurement method. Unlike the wrist, where only surface arteries can be measured and green light is chosen, the ear can use infrared light, which results in a greater penetration depth and a higher SNR. A photodiode, whose detection area is directly related to its responsivity, measures the reflected light. It therefore measures both the signal and the background noise. A downstream analog front end provides a higher SNR. It acts as a signal filter and converts the detected current into a voltage and then into digital form. In addition to the reflection measurement, the algorithm also includes corrections for filtering out motion artifacts from the accelerometer.

The components that make up the measurement system are described below. ADI's ADPD144RI chip is used as the analog front end, which also integrates a photodiode and LED. The measurement is supported by a three-axis accelerometer, which is used not only to identify gait and movement but also to remove artifacts. The ADXL362 is used in this example. The entire process is controlled by the ADuCM3029 microcontroller, which serves as an interface for the various sensors and contains the algorithms.

Figure 1 shows the test system, where both the optical sensor and accelerometer are housed in a conventional earbud. Care was taken to limit the ADC sampling rate to 100 aHz and minimize the LED intensity to keep power consumption as low as possible.

Figure 1. Test system integrating optical sensors and accelerometers. Scale bars are used for comparison.

To characterize the system behavior, five different scenarios were considered for different motion patterns. The evaluation used only optical signals, so it was possible to understand in which scenarios pulse measurement inaccuracies appear and when accelerometer data is needed to improve the accuracy of the pulse measurement.

Figure 2 is a plot of amplitude versus sampling rate showing the spectrum of the raw data. Over time, the pulse can be identified by the peaks. In the absence of motion, the signal is very clear and the heart rate can be determined from the peak locations and the known sampling rate.

Figure 2. Measuring the amplitude oversampling rate provides information about heart rate.

The optical sensor records the heart rate with two LED colors - infrared and red - and four channels for each color. This allows the measurement to be differentiated by the two differently colored channels and the more robust version to be selected. The signals of the various channels are shown in Figure 3A. With six channels a very clear signal can be identified, while two channels are saturated. To obtain a stronger and more robust signal, the algorithm adds the corresponding unsaturated channels and calculates the heart rate. Figure 3B shows the heart rate for the red channel (top) and the infrared channel (bottom), while the confidence of the measurement is indicated with the help of a color scale. The multiples of the heart rate are also given, which allows the original signal (dashed line) to be distinguished by the sampling rate and the confidence indication.

Figure 3. The red area (top) shows the four-channel measurement for the standing still condition, while the infrared area (bottom) shows the raw and summed data. Heart rate (black line) can be determined by the algorithm from the summed data, and the color scale indicates the confidence level.

In summary, in the absence of motion, the signal is strong and free of obstructing noise, so the algorithm can determine heart rate with high confidence. The signal from the infrared channel is stronger than the signal from the red channel.

Scenario 2 introduces additional chewing movements. The recorded spectrum is shown in Figure 4. Unlike test scenario 1, motion artifacts can be clearly seen here, which appear as jumps in the signal. They also become clear in the sum of the channels, which no longer show such clearly different rates. However, the algorithm is still able to correctly determine the heart rate with high confidence without the additional help of the motion sensor. Interestingly, the infrared signal strength is again greater than the signal strength of the red channel.

Figure 4. The red area (top) shows four-channel measurements for standing still and chewing, while the infrared area (bottom) shows the raw and summed data. Heart rate (black line) can be determined by the algorithm from the summed data, with the color scale indicating the confidence level. Heart rate can be determined without an accelerometer.

Another everyday situation was tested in Scenario 3. The test person sat at a table and performed some normal work and related movements. Similar to Scenario 2, motion artifacts can be detected, thanks to which the algorithm can identify the heart rate in both channels. As can be seen in Figure 5, the infrared signal is also dominant here.

Figure 5. The red area (top) shows four-channel measurements of a desk worker, while the infrared area (bottom) shows the raw and summed data. Heart rate (black line) can be determined by an algorithm from the summed data, with the color scale indicating the confidence level. Heart rate can be determined without an accelerometer.

The previous scenarios focused on stationary measurement situations, but in this scenario the test person moves evenly in one direction at a low speed (about 50 steps per minute). As shown in Figure 6, the heart rate and steps are mixed in the PPG signal, and the sum of the various channels shows a very blurred signal. Although no clear heart rate can be calculated in the red signal field, the algorithm finds a fitting heart rate in the infrared signal. However, due to the large fluctuations and the low confidence of the matrix, additional motion data from the accelerometer would be very useful, especially because the measurements have been performed only at low walking speeds so far.

Figure 6. The red area (top) shows four-channel measurements for the walking condition, while the infrared area (bottom) shows the raw and summed data. Heart rate (black line) can be determined by the algorithm from the summed data, with the color scale indicating the confidence level. For the infrared condition, heart rate can be determined without an accelerometer.

Instead of measuring uniform motion, Scenario 5 alternates between sprinting and jumping at regular intervals. The motion artifacts can now be identified very clearly, and the algorithm has a hard time isolating the correct heart rate, as shown in Figure 7. The need for motion sensor support seems inevitable.

Figure 7. The red area (top) shows four-channel measurements of running and jumping, while the infrared area (bottom) shows the raw and summed data. Heart rate (black line) can be determined by the algorithm from the summed data, and the color scale indicates the confidence level. It is difficult to determine heart rate without an accelerometer.

To better assess the need for motion sensors, scenario 5 tests the measurement technique with and without the use of an accelerometer. Figure 8 shows a comparison of the additive spectra of the accelerometer data without correction (left) and with correction (right). A clear improvement in the signal can be seen when identifying the heart rate, which would not be possible without the support of an accelerometer.

Figure 8. Comparison of additive spectra without accelerometer data (left) and with accelerometer data (right). The accelerometer can be used to reconstruct the user’s heart rate.

From the test cases it can be concluded that in most cases the heart rate can be determined very accurately using the sensors integrated in the earbuds. In the case of local or slow translational movements the heart rate can even be determined without using accelerometer data. However, in the extreme case of sudden and fast movements a comparison with the motion correction data also allows interpretation. In all cases the infrared signal is stronger than the red light signal.

The signal in the ear is stronger than that in the wrist, so the measurement accuracy can be higher. In addition, blood oxygen levels can be measured using red or infrared light.

In conclusion, ear measurements have also proven to be very promising in functional test systems. The measurement setup can also be improved with better mechanical integration and expanded to include additional measurements. In this way, accelerometers can also be used for fall detection and gait recognition, creating more value for customers.