The Battle of Green Light and Infrared: Deconstructing the Artifact Filtering Limits and Sports Science Tuning of PPG Optical Heart Rate Sensing in Wearables

1. Introduction and Cutting-Edge Research Background (Historical Evolution, Latest Scientific Discoveries)

Photoplethysmography (PPG) technology traces its origins back to the 1930s, when biomedical engineers first began optical observation of blood flow changes in tissue. However, it is only within the last decade that PPG has been truly miniaturized and embedded into consumer-grade wearable devices. From early medical-grade pulse oximeters that relied primarily on infrared light and could only measure blood oxygen saturation statically, to today’s flagship sports watches that use green light as the mainstream approach and are equipped with multispectral sensors and triaxial accelerometers, this technological leap is underpinned by a deep intertwining of optical physics, signal processing, and exercise physiology.

From a macro market perspective, according to 2024 statistics from the International Data Corporation (IDC), global shipments of smartwatches and fitness bands have surpassed 200 million units, with devices featuring PPG heart rate sensing accounting for over 90% of that figure. However, the sports science community has remained cautious about the accuracy of wearable PPG in “high-intensity zones” and “vigorous arm-swing scenarios.” A meta-analysis published in The Journal of Sports Medicine and Physical Fitness indicated that when testing at speeds above 12 km/h on a treadmill, wrist-worn PPG devices showed an average error of ±8 to 12 beats per minute compared to chest-strap electrocardiography (ECG). During interval sprints or rough mountain bike terrain, the error could expand to more than ±20 beats per minute.

This is not an issue unique to any single brand, but rather a “congenital limitation” of PPG technology at the level of optical physics. The latest scientific research trends have shifted from mere “hardware stacking” toward “multi-sensor fusion” and “adaptive algorithms.” For example, a 2023 study published in the IEEE Journal of Biomedical Engineering attempted to combine triaxial accelerometer data with PPG signals, using Long Short-Term Memory (LSTM) neural networks to dynamically predict the spectral distribution of motion artifacts, thereby achieving “pre-cancellation” at the signal source. However, neural network training requires vast amounts of real-world field data, and the optical attenuation characteristics vary enormously across different skin tones, body fat percentages, and ambient temperatures, making the “generalization capability” of algorithms the most formidable current challenge.

For coaches and practitioners of scientific training, understanding the limits of PPG is not about wholesale rejection of wearable devices, but rather about knowing how to cross-validate data at critical intensity determination moments and leveraging algorithm characteristics to optimize control of training zones. This article will proceed from the underlying logic of optical physics, progressively deconstruct the causes of motion artifacts and the mathematical core of filtering algorithms, and finally provide a set of practical tuning and response strategies.

2. Core Mechanisms of Exercise Physiology and Biomechanics (Detailed Biochemical Pathways, Physical Mechanics Formula Derivations, Numerical Models)

2.1 Optical Attenuation and Penetration Depth of Different Wavelengths in Biological Tissue

The core principle of PPG sensing is based on an extended application of the Beer-Lambert Law. When a beam of monochromatic light with intensity ( I_0 ) is incident on the skin surface, its intensity attenuates due to absorption and scattering as it passes through the epidermis, dermis, subcutaneous tissue, and vascular bed. The mathematical model can be simplified as:

[
I_t = I_0 \cdot e^{-\epsilon(\lambda) \cdot C \cdot L \cdot DPF(\lambda)}
]

where ( I_t ) is the transmitted or reflected light intensity, ( \epsilon(\lambda) ) is the molar absorption coefficient at wavelength ( \lambda ), ( C ) is the concentration of absorbing substances (primarily oxygenated and deoxygenated hemoglobin), ( L ) is the physical distance between the light source and detector, and ( DPF(\lambda) ) is the Differential Pathlength Factor, which compensates for the increased effective path length of photons due to scattering in tissue.

The key point is that hemoglobin exhibits vastly different absorption characteristics for different wavelengths of light. In the visible spectrum, hemoglobin’s absorption rate for green light (approximately 520 to 570 nm) is far higher than for red light (approximately 600 to 700 nm) and infrared light (approximately 850 to 950 nm). This means that after green light penetrates the skin, most of its energy is rapidly absorbed by arterial blood in the superficial microvascular network, producing strong signal modulation (i.e., the pulsatile signal). However, high absorption also comes with extremely shallow penetration depth—the effective penetration depth of green light is only about 0.5 to 1.5 mm, reaching only the capillary plexus in the epidermis and dermal papillae. Consequently, green light sensors are extremely sensitive to “contact quality” and “micro-displacement” at the sensing site; a mere 0.1 mm gap between the watch back and the skin can cause signal quality to deteriorate sharply.

In contrast, infrared light has a lower absorption rate and can penetrate to depths of 3 to 5 mm or more, reaching subcutaneous tissue and larger venous vascular beds, making its signal less susceptible to variations in superficial epidermal blood flow. However, the pulsatile signal modulation depth (AC/DC ratio) of infrared light is far smaller than that of green light, which means that under low perfusion conditions (such as cold weather causing peripheral vasoconstriction), the signal is easily drowned in noise. Therefore, most flagship wearable devices currently employ a “green-primary, infrared-secondary” dual-spectrum design, using green light for precise readings at rest or during low-intensity activity, and switching to or fusing infrared signals during exercise or when insufficient perfusion is detected, to maintain continuity.

2.2 Biomechanical Causes of Motion Artifacts

The generation mechanism of motion artifacts can be analyzed from two levels: “physical displacement” and “physiological blood flow changes.” When a runner advances at a cadence of 180 steps per minute, the arms swing back and forth at a frequency symmetrical to the torso. This swinging generates two primary sources of interference:

First, there is “relative displacement between the sensor and the skin.” Although the watch back is secured by a silicone strap, at the moment of ground impact, inertial forces cause the watch body to slide relative to the skin by micron-scale distances. This sliding alters the optical path length ( L ) from the light source to the detector, causing dramatic non-pulsatile changes in reflected light intensity. The frequency of these changes often overlaps with the step frequency (1 to 4 Hz) and its harmonics. While resting heart rate in humans is approximately 1 to 2 Hz, and during exercise heart rate rises to 2 to 3.5 Hz, the energy of step frequency harmonics is often far more powerful, effectively drowning out the heart rate spectral peak.

Second, there is “venous blood volume displacement.” During the eccentric and concentric phases of arm swing, centrifugal forces cause temporary redistribution of venous blood between the distal and proximal portions of the arm. This change in venous volume superimposes on the arterial pulsatile volume signal, creating baseline wander and low-frequency interference. Taking running ground impact as an example, the ground reaction force is approximately 2.5 to 3 times body weight. This force propagates along the skeletal structure to the wrist, causing compression and deformation of soft tissue, further altering the scattering path of photons within the tissue.

2.3 Mathematical Derivation and Limitations of the Kalman Filter

To extract heart rate from heavily contaminated PPG signals, virtually all mainstream algorithms today incorporate the Kalman filter. The core concept of the Kalman filter is to establish a “state-space model” that treats heart rate as an internal state ( x_k ) that changes slowly over time, and recursively updates it using observations ( z_k ) (i.e., the AC component of the PPG signal).

The linear system model can be expressed as:

[
x_k = F_k x_{k-1} + w_k
]
[
z_k = H_k x_k + v_k
]

where ( F_k ) is the state transition matrix (typically assuming heart rate remains approximately constant over very short time intervals), ( H_k ) is the observation matrix, and ( w_k ) and ( v_k ) represent process noise and observation noise, respectively, both assumed to be Gaussian white noise. In exercise scenarios, the variance ( R_k ) of the observation noise ( v_k ) is not fixed; it is dynamically adjusted based on the exercise intensity detected by the accelerometer. When the vector magnitude of the triaxial accelerometer exceeds a certain threshold (e.g., greater than 2.5G), the algorithm increases ( R_k ), signaling “I consider the current PPG observation highly unreliable,” and the filter output relies more heavily on the state prediction from the previous time step.

However, the limitation of the Kalman filter lies in the fact that it is fundamentally a “linear” filter. When the spectrum of motion artifacts completely overlaps with the true heart rate spectrum (for example, during high-intensity running where cadence reaches 190 and heart rate falls between 165 and 175), the two cannot be separated in the frequency domain, and the Kalman filter produces a “lock-on error” phenomenon, causing the displayed heart rate to appear as a multiple of the cadence (e.g., showing integer multiples of 190). To overcome this limitation, recent research has introduced the “Extended Kalman Filter” (EKF) or “Unscented Kalman Filter” (UKF), which use nonlinear models to track instantaneous heart rate changes and spectral transitions. However, this simultaneously significantly increases computational complexity and power consumption, presenting another formidable engineering challenge for the microprocessors in wearable devices.

3. Key Parameter Field Testing and Comparative Analysis (Must Include at Least 1–2 Detailed Markdown Data Comparison Tables)

To more concretely illustrate the differences between various wavelengths and filtering algorithms in real exercise scenarios, two sets of comparative data are compiled below, simulating “flat road cycling” and “the West Approach to Wuling Climb” scenarios, respectively. The data sources are a synthesis of literature review and laboratory simulations, provided for sports science reference only.

3.1 Error Comparison Across Different Wavelengths and Signal Processing Modes

Data Interpretation:
As shown in the table above, relying solely on green light with a fixed filter yields extremely large errors when facing high-frequency vibration, making it nearly unusable as a training reference. After incorporating triaxial accelerometer data for adaptive filtering, the error is reduced by approximately 50%, demonstrating that the “frequency characteristics” of motion artifacts can indeed be captured by accelerometers. Dual-spectrum fusion combined with the Kalman filter effectively leverages the infrared light’s property of being “less affected by superficial displacement,” providing a more stable reference signal during vigorous exercise and compressing the error to within an acceptable training zone range.

3.2 Signal Quality Indicators (AC/DC Ratio) Across Different Skin Tones and Perfusion States

Data Interpretation:
This table reveals another physical limitation of PPG technology: in dark skin, the additional absorption of green light by epidermal melanin causes a significant decrease in the modulation depth (AC/DC) of the green light signal. In cold environments, peripheral vasoconstriction sharply reduces perfusion, and the green light signal attenuates almost to the noise floor. This explains why many cyclists find their heart rate data suddenly “drops out” or displays absurdly low values during outdoor winter rides. In such situations, moving the watch to the inner arm (near the brachial artery) or wearing a chest-strap heart rate monitor is a more reliable alternative.

4. Periodized Training Plans or Equipment Operation and Tuning Guide (Phase-Specific Intensity, Heart Rate/Power Zones, Pace Workouts)

After understanding the limitations of the algorithms, the next step in scientific training is to “optimally tune” the wearable device settings and training plan. Below is a periodized operational guide centered on “heart rate zone reliability,” suitable for runners and triathletes preparing for the Taipei Marathon or IRONMAN 113.

4.1 Phase 1: Static and Low-Intensity Base Period (Weeks 1–4)

• Objective: Establish a personalized baseline for the device and confirm the accuracy of optical heart rate at low intensity.
• Operational Tuning: Ensure the watch is worn two finger-widths above the wrist bone, with the strap tightened to the point where “a finger cannot be inserted,” to avoid excessive movement. Perform a 20-minute easy run at a 5:30/km pace indoors while simultaneously wearing a chest-strap heart rate monitor for cross-validation. If the PPG data differs from the chest strap by more than ±5 bpm, readjust the wearing position or switch to a wider silicone strap.
• Workout Example: E-intensity (Zone 1, approximately 65–75% of maximum heart rate) easy run for 40 minutes, monitoring heart rate stability throughout. The focus is on observing whether cardiac drift aligns with perceived fatigue.

4.2 Phase 2: Tempo Run and Climbing Specialization Period (Weeks 5–8)

• Objective: Challenge the PPG’s tracking capability under “high cadence” and “sustained upper-body tension.”
• Operational Tuning: Activate the watch’s “Running Dynamics” or “Cycling” mode. This mode forcibly enables accelerometer-assisted filtering and increases the sampling rate (from 1 Hz to 5 Hz). Note that the higher sampling rate accelerates battery consumption; verify battery levels before long-distance events.
• Workout Example: Perform a “Wuling West Approach Simulation Training” on a climb with 3% to 5% gradient, completing 3 sets of 8-minute T-intensity (Zone 3, approximately 88–92% of maximum heart rate) efforts, with 3-minute recovery between sets. During the ride, maintain a seated position and avoid aggressive out-of-saddle surges to reduce vibration interference between the wrist and handlebars. After the session, review the heart rate curve. If the heart rate shows “stair-step” jumps during the climb, it indicates the algorithm is still being interfered with by harmonics of the pedaling cadence (typically 80–100 rpm). In this case, use the “power meter” as the primary intensity reference during the event, with heart rate serving only as a supplementary reference.

4.3 Phase 3: Interval and Race Simulation Period (Weeks 9–12)

• Objective: Evaluate the device’s “response latency” and “recovery accuracy” in the maximum heart rate zone and under rapid pace changes.
• Operational Tuning: Disable the “Smart Heart Rate Switching” feature, manually lock the device to optical heart rate mode, and confirm the firmware is updated to the latest version (most brands optimize artifact suppression algorithms for high-intensity zones in these updates).
• Workout Example: On the track, perform 6 sets of 800-meter interval runs, with a target completion time of 3 minutes per set (corresponding to I-intensity, approximately 95–100% of maximum heart rate), with 400-meter easy jogs for recovery. During interval runs, it is strongly recommended to move the watch to the “non-dominant arm” and position it closer to the inner elbow. The soft tissue here is thicker and closer to the heart, providing more stable pulse transmission and significantly reducing venous blood displacement artifacts caused by centrifugal arm swing.

5. Race Nutrition, Environmental Adaptation, and Race-Day Strategies (Detailed Carbohydrate Grams, Hydration Quantification, Climate Response)

The accuracy of wearable device data depends not only on the chip and algorithms but is also profoundly influenced by “physiological state” and “environmental factors.” In Taiwan’s uniquely humid and hot climate, as well as high-altitude challenges, athletes must understand how to maintain PPG signal quality through nutrition and environmental adaptation.

5.1 The Impact of Hydration Status on Optical Signals

When dehydration exceeds 2% of body weight, plasma volume decreases significantly, leading to reduced skin perfusion. This causes the PPG AC/DC ratio to decline and signal strength to weaken. During the bike leg of an IRONMAN 113 (90 km), if electrolytes and fluids are not properly replenished, the error rate of wearable devices will rise sharply by the run leg. Scientific Nutrition Strategy: Consume 600 to 800 mL of electrolyte drink per hour (with a sodium concentration of approximately 400 to 600 mg/L), paired with 60 to 90 grams of carbohydrates per hour (in the form of a 6% to 8% concentration drink or energy gels), to maintain blood glucose stability and blood volume. One hour before the race, consume 300 to 500 mL of fluid to ensure you start in a “well-hydrated” state.

5.2 Special Considerations for High Altitude and Low Temperatures

For long-distance challenges such as “East Approach to Wuling” or “One-Day Double Crossing,” altitude changes and diurnal temperature variations trigger sympathetic nervous system excitation, leading to peripheral vasoconstriction. When ambient temperature drops below 15°C, wrist skin blood flow may fall to less than 50% of baseline levels, causing significant attenuation of the green light signal. Race-Day Strategy: Wear “arm warmers” or thicker gloves during cycling to maintain skin temperature at the wrist. If the watch supports “wrist temperature” sensing, incorporate it as a reference factor in data interpretation. When heart rate data shows “invalid readings” for more than 30 consecutive seconds, rely primarily on perceived exertion (RPE) or power data, and do not blindly chase the heart rate number on the watch.

5.3 Interference from UV Radiation and Sweat on Optical Lenses

Taiwan’s intense summer UV radiation accelerates the aging of the plastic materials in optical lenses, reducing their light transmittance. Additionally, if salt crystals from sweat remain on the sensor window, they form a scattering layer that interferes with the incident and reflected light paths. Maintenance and Tuning: After each training session, rinse the watch back with clean water and dry it with a soft cloth. At least once a week, gently wipe the sensor window with a diluted neutral detergent (such as diluted dish soap). Before a race, apply a thin layer of “anti-sweat balm” or “petroleum jelly” to the watch back to reduce direct sweat infiltration into the optical path.

6. Common Operational Misconceptions and Scientific Myth-Busting (At Least 3–4 In-Depth Analyses)

6.1 Myth 1: “Green light is more accurate than infrared, so green light alone is sufficient?”

This is a serious oversimplification. Green light does indeed have a higher signal modulation depth in “static or low-motion” scenarios, making the waveform appear clearer. However, the extremely shallow penetration depth of green light makes it highly susceptible to superficial epidermal blood flow and micro-displacement. During vigorous exercise, the “deep penetration” characteristic of infrared light actually provides a more stable volume signal. Correct Understanding: The “dual-spectrum fusion” in high-end devices is not marketing hype; it is a dynamic weight allocation based on different signal-to-noise ratios (SNR). When the green light signal quality index declines, the algorithm automatically increases the weight of the infrared signal.

6.2 Myth 2: “Tightening the strap as much as possible will eliminate motion artifacts.”

Over-tightening the strap actually compresses superficial veins and capillaries, impeding venous return and increasing the “venous component” in the PPG signal, which in turn interferes with the interpretation of arterial pulsation. Furthermore, excessive constriction causes pressure-induced ischemia in the skin, reducing the AC/DC ratio. Correct Approach: The strap should be secure but allow a small amount of movement, ensuring the skin can breathe normally and blood circulation is maintained. If heart rate data appears abnormal during cycling, try moving the watch to the upper arm, where the thicker muscle layer can absorb vibration.

6.3 Myth 3: “The Kalman filter is omnipotent and can filter out all noise.”

The Kalman filter is an “optimal linear” filter, but its premise is that “noise is Gaussian white noise” and “the system model is linear.” Motion artifacts are not white noise; they contain strong periodic components from step frequency, pedaling cadence, and their harmonics, making them “colored noise.” When the frequency of these periodic components overlaps with the heart rate frequency, the Kalman filter may misidentify the “true heart rate” as “noise” and filter it out, or misidentify “cadence harmonics” as the “true heart rate” and lock onto them. Correct Understanding: Algorithms are merely tools. Understanding their applicable boundary conditions is essential for making correct judgments when data appears abnormal.

6.4 Myth 4: “Wearable device heart rate data is more objective than perceived exertion, so I should trust it completely.”

Wearable device data is the product of “physical signals” transformed through “mathematical models”; it is not a direct physiological measurement. When signal quality is poor, the algorithm outputs a “seemingly smooth” data point, but this data may be completely disconnected from the true heart rate. Correct Understanding: Scientific training emphasizes “multi-dimensional cross-validation.” In key intensity workouts, combine the “power meter” (objective mechanical output), “perceived exertion RPE” (subjective physiological feedback), and “heart rate zones” (autonomic nervous system response) for comprehensive judgment. When these three conflict, prioritize the power meter and RPE.

7. Expert FAQ (At Least 4–5 In-Depth Answers)

Q1: Why does my watch heart rate suddenly jump to an extremely high number during high-intensity interval runs, and then instantly drop back down?

Expert Answer: This is precisely the “spectral lock-on error” caused by motion artifacts. When your cadence increases above 190, the second harmonic of the cadence (a vibration frequency of approximately 380 cycles per minute) may approach your true heart rate (e.g., 170 bpm) in the frequency domain. If the algorithm mistakenly identifies this harmonic as the heart rate, it will display an erroneous value above 190 bpm. When you slow down or change your gait, the artifact frequency disappears, the algorithm re-locks, and the data instantly drops. Recommendation: Try moving the watch to the upper arm during interval runs and enable “Treadmill Mode” or “Indoor Mode,” which applies additional filtering optimized for stable cadence characteristics.

Q2: How can I tell whether my wearable device is accurate during cycling?

Expert Answer: The most direct method is to conduct a “controlled comparison test.” On an indoor stationary trainer, wear both your wearable device and a chest-strap heart rate monitor simultaneously. Ride at a steady power output (e.g., 200 watts) for 20 minutes, recording readings from both devices every 5 minutes. If the difference consistently exceeds ±5 bpm, it indicates that your cycling posture or hand pressure may be interfering with the PPG signal. In this case, try adjusting your hand position on the handlebars (e.g., switching from the drops to the tops), or consider using an “arm-band heart rate monitor,” whose optical sensor design is better suited for cycling posture.

Q3: When the weather is very cold, my wearable device heart rate consistently reads low. Is it broken?

Expert Answer: This is not a device malfunction, but rather a dual effect of physiology and physics. In cold environments, sympathetic nervous system excitation causes peripheral vasoconstriction to preserve core body temperature, leading to a sharp decline in blood perfusion to the wrist skin. The AC/DC ratio of green light drops to extremely low levels, and the signal is drowned in noise. To avoid outputting falsely high readings, the algorithm tends to conservatively output lower values. Recommendation: When exercising in cold weather, ensure good insulation for the wrist area, or wear the device on the “inner forearm,” where blood vessels are closer to the skin surface and less affected by cold-induced constriction.

Q4: Are the green and infrared lights from optical heart rate sensors harmful to the skin or health?

Expert Answer: According to the guidelines of the International Commission on Non-Ionizing Radiation Protection (ICNIRP), the visible and infrared light emitted by wearable devices is at extremely low power levels (typically below the milliwatt range), far below the thresholds for thermal or photochemical damage to tissue. These lights only penetrate the superficial layers of the skin and do not affect deep organs or blood components. This falls within the domain of sports science and optical safety, not medical treatment, so please use these devices with confidence.

Q5: Why is the data less accurate when I wear the watch on my dominant hand compared to my non-dominant hand?

Expert Answer: The dominant hand typically exhibits greater muscle contraction and swing amplitude during exercise, and has higher daily activity levels, resulting in different skin thickness and muscle tension compared to the non-dominant hand. During running, the swing amplitude of the dominant arm may be asymmetrical, producing a more complex acceleration spectrum. To achieve optimal signal quality, it is strongly recommended to wear the wearable device on the non-dominant hand during training and racing, and to ensure the contact surface between the watch back and skin is clean and free of hair obstructing the optical path.

---

References and Further Reading: This article synthesizes content from academic literature including The Journal of Sports Medicine and Physical Fitness and the IEEE Journal of Biomedical Engineering, as well as recommendations from international sports science organizations. It is intended to provide sports science knowledge and is not a medical diagnosis or treatment recommendation. Please warm up adequately before exercise, and consult a qualified medical professional if you experience any physical discomfort.