ECG remains the diagnostic gold standard for cardiac rhythm and high-frequency heart rate variability metrics. PPG reliably tracks pulse rate and longitudinal trends but needs careful signal-quality filtering and ECG confirmation before any diagnostic claim. Pulse rate variability approximates HRV reasonably well in some domains, yet diverges under motion and for high-frequency components, and optimised multimodal pipelines narrow that gap without replacing clinical ECG.
TL;DR:
- At rest, PPG tracks pulse rate well; optimized atrial fibrillation screening has exceeded 95% sensitivity and 99% specificity, but ECG must confirm diagnoses.
- Reconstruction can bring time domain and low frequency measures to correlations of 0.6 to 0.8, while RMSSD and high frequency power remain unreliable during movement.
- Signal quality filtering can improve agreement but sacrifices data: one pipeline retained about 75% of epochs, while a breathing study discarded 63% of PPG windows.
- Use PPG for continuous wellness trends and screening, especially at rest; movement, posture, sensor fit, skin tone, and vascular disease can alter signal quality.
Table of Contents
- Technical origin and timing: ECG vs PPG and implications of pulse transit time
- Measurement accuracy: HR, HRV, arrhythmia detection and respiratory rate
- What degrades PPG and ECG signals, and how to limit the damage
- Preprocessing and advanced methods for improving PPG-derived estimates
- Choosing the right modality: screening, monitoring and where ECG is required
- Where the evidence is thin and what would move the field forward
- A researcher’s checklist for PPG-ECG comparison studies
- Voltra devices for longitudinal pulse and trend monitoring
- FAQ
Technical origin and timing: ECG vs PPG and implications of pulse transit time
The two modalities measure fundamentally different physiological events, and that distinction explains almost every downstream accuracy difference researchers encounter. Electrocardiography records the electrical depolarisation of cardiac muscle, producing a waveform with clearly defined fiducial points: the P wave, the QRS complex and the T wave. The R peak within the QRS complex gives a sharp, unambiguous timing marker for each heartbeat, which is why ECG-derived interbeat intervals are treated as the reference standard for heart rate variability research.
Photoplethysmography instead measures peripheral blood-volume changes using light absorption at the skin surface, typically at the wrist, finger or ear. The resulting waveform shows a systolic peak, a dicrotic notch and a diastolic phase, none of which align in time with the electrical event that triggered them. Between ventricular depolarisation and the arrival of the corresponding pulse wave at a peripheral site lies a measurable delay: pulse transit time (PTT), sometimes expressed as pulse arrival time (PAT) when measured from the ECG R peak to the PPG waveform foot.
This delay is not fixed. It varies with vascular tone, arterial stiffness, blood pressure, posture and sensor site, which means interval-based metrics calculated from PPG carry physiological noise that ECG does not. A beat-to-beat PPG interval reflects both the true cardiac cycle length and fluctuations in how fast the pulse wave travelled through the vasculature, as detailed in a review of PPG and ECG differences, which also notes that agreement improves in supine or seated positions where vascular dynamics are most stable.
The practical consequences for signal processing are significant:
- Fiducial extraction differs in difficulty: the ECG R peak is sharp and algorithmically simple to detect, while PPG systolic peaks can be rounded, split by notches, or distorted by vasomotor activity.
- Interbeat interval calculation inherits vascular noise: PRV sequences carry both cardiac timing and arterial transit variability, which inflates certain variance measures relative to true HRV.
- Latency-sensitive analyses are affected: any application timing PPG against external electrical events (such as cross-modal validation) must account for PTT rather than assuming simultaneity.
- Posture and vascular disease shift the error: populations with stiffer arteries or peripheral vascular disease show larger and more variable PTT, which is a known confound in PRV-HRV comparison studies.
Understanding this timing offset is the starting point for interpreting every accuracy figure that follows, because a correlation coefficient between HRV and PRV is, in effect, also a measure of how stable a person’s vascular transit time was during recording.
Measurement accuracy: HR, HRV, arrhythmia detection and respiratory rate
Accuracy figures for PPG vary enormously depending on what is being measured, the motion state of the subject and how aggressively poor-quality data was excluded before analysis. Reporting a single “accuracy” number for PPG without these qualifiers is close to meaningless, so the evidence is best read metric by metric.
For resting pulse rate and heart rate, PPG performs well. Under low-motion, seated or supine conditions, correlation with ECG-derived heart rate is typically strong, with small average bias, consistent with the posture-dependent agreement described in the PPG and ECG comparison review. This is the use case PPG was originally built for, and it remains its strongest.
PPG-based atrial fibrillation screening can reach sensitivity above 95% and specificity above 99% in optimised systems once poor-quality segments are removed, according to a study reporting AF detection performance from photoplethysmography. Confirmatory ECG remains mandatory for any AF diagnosis regardless of how strong the screening signal looks.
Heart rate variability presents the most nuanced picture:
- Time-domain and low-frequency metrics often show moderate-to-strong agreement between PRV and HRV, with reconstruction-assisted pipelines reaching correlations in the range of r = 0.6 to 0.8 for several metrics in dynamic settings, per the signal quality assessment and reconstruction study.
- High-frequency components, including RMSSD and HF power, are frequently unreliable, particularly under movement, and a study of post-cardiac-surgery patients found insufficient agreement between PRV and HRV for RMSSD, HF-norm and LF/HF ratio despite strong correlations (up to r = 0.94 to 1) on other indices, as shown in the HRV and PRV comparison in CABG patients.
- Respiratory rate can be estimated from both ECG and PPG by exploiting baseline wander, amplitude modulation and frequency modulation, but accuracy improvements from fusion and quality-assessment steps come at the cost of discarding a sizeable share of recording windows, as documented in a review of breathing rate estimation from ECG and PPG, which reports one study discarding 44% of ECG windows and 63% of PPG windows to achieve its reported mean absolute error.
The pattern across all four outcomes is consistent: PPG accuracy climbs toward ECG-level performance exactly when data-inclusion criteria tighten, which means every accuracy figure needs its corresponding data-loss rate reported alongside it to be interpretable.
What degrades PPG and ECG signals, and how to limit the damage
Both modalities are vulnerable to artefacts, but the sources differ and so do the mitigation strategies.
Motion is PPG’s dominant failure mode because limb and finger movement produces frequency content that overlaps directly with the pulse waveform itself, so standard band-pass filtering cannot separate the two cleanly. This is why accelerometer-informed adaptive filtering, rather than filtering alone, is needed to recover usable signal during activity.
- Sensor site changes the error profile: finger-based PPG generally gives the cleanest waveform at rest, wrist-worn PPG trades signal quality for wearability and is more motion-sensitive, and ear-based sensors sit closer to central circulation but are prone to their own movement artefacts.
- Contact pressure matters as much as site: too little pressure weakens the optical signal, too much restricts blood flow and distorts the waveform shape, so device fit is a genuine source of measurement variance rather than a cosmetic detail.
- Skin tone and ambient light affect optical absorption: melanin concentration alters light absorption characteristics, and poorly shielded sensors pick up ambient light as noise. Multi-wavelength LED arrays and better optical shielding are the hardware-level mitigations manufacturers use to reduce this bias.
- ECG has its own artefact profile: electrode contact noise, sweat-related impedance changes, muscle (EMG) artefact and incorrect lead placement all degrade ECG quality, though ECG is generally less susceptible to motion-induced signal loss than optical PPG because it measures electrical rather than mechanical events.
Pro Tip: Report the percentage of recording time excluded by your signal-quality assessment alongside every accuracy figure, since the same pipeline can look excellent on quiet data and mediocre once motion is included.
Preprocessing and advanced methods for improving PPG-derived estimates
Raw PPG rarely produces research-grade HRV data; the gap is closed, when it can be closed, through a deliberate signal-processing pipeline rather than a single filter.
Signal quality assessment (SQA) is the first gate. Typical SQA metrics include waveform morphology checks, skewness and kurtosis of the pulse shape, and perfusion index thresholds, and raising the SQA threshold predictably improves agreement with ECG while reducing the number of usable epochs, a direct trade-off documented in the reconstruction pipeline study, which reports around 75% usable epochs after applying SQA and reconstruction together in a dynamic test setting.
Beyond filtering, several reconstruction and fusion strategies are now standard in serious validation work:
- Narrow-band spectral reconstruction rebuilds the pulse signal around its dominant frequency component, discarding broadband motion noise outside that band.
- Beat-by-beat reconstruction interpolates or corrects individual pulses flagged as low quality rather than discarding the whole epoch, preserving more data than blanket exclusion.
- Sensor fusion with accelerometer data allows adaptive filters to subtract motion-correlated noise components directly, rather than relying on fixed-frequency filtering.
- Multi-site PPG (combining two or more sensor locations) can recover epochs where one site is corrupted but another remains usable.
Machine learning approaches are the newest addition, particularly cross-modal methods that attempt to align PPG and ECG representations or even synthesise ECG-like waveforms directly from PPG input.
Experts emphasise that PPG is not merely a cheaper ECG surrogate but a complementary modality, and that multimodal sensors combined with careful signal quality assessment improve trustworthiness.
That framing, drawn from expert perspectives on the complementary roles of PPG and ECG, is a useful corrective against treating PPG as a simple ECG substitute. Cross-modal learning methods show promise for cardiovascular screening, but interpretability remains a caveat: a model that outputs an ECG-like waveform from PPG input still needs independent validation before any clinical weight is placed on it, and these methods should be reported with the same transparency about failure modes as any other screening tool.
Choosing the right modality: screening, monitoring and where ECG is required
The choice between PPG and ECG is less about which is “better” overall and more about matching the modality to the question being asked.
- PPG is appropriate for continuous trend detection: sleep monitoring, general wellness tracking, and population-level screening programmes that funnel flagged individuals toward ECG confirmation rather than diagnosing from PPG alone.
- ECG is required for diagnosis: arrhythmia classification, ST-segment and T-wave morphology analysis, and any decision that depends on the electrical waveform shape rather than pulse timing.
- Hybrid workflows get the best of both: use PPG for always-on screening, define and report explicit SQA inclusion criteria, and route any flagged event through a confirmatory ECG pathway rather than acting on the PPG signal in isolation.
- Data-loss reporting is not optional: any study or product claim about PPG performance should state what fraction of recordings were excluded to reach that performance, since the two numbers together are what make the claim interpretable.
A regulatory caveat belongs in every research protocol and product description: PPG-only measurements should not be presented as a clinical diagnosis unless the device carries the appropriate regulatory clearance and labelling for that specific use. Consumer-grade PPG wearables are overwhelmingly cleared and marketed for wellness and trend monitoring, not for standalone arrhythmia diagnosis, and that distinction should stay visible in both study design and public communication.
Where the evidence is thin and what would move the field forward
Several gaps limit how confidently PPG performance can be generalised across populations and conditions. Closing them would materially strengthen the evidence base researchers rely on.
- Standardised acquisition protocols and public benchmark datasets covering diverse skin tones, ages and vascular disease states are still scarce, which limits how far any single study’s accuracy figures can be generalised.
- Ecologically valid movement conditions are under-represented; many validation studies still lean on seated or supine recordings, which overstate real-world performance during daily activity.
- Transparent SQA reporting is inconsistent across the literature, making it hard to compare studies that quote similar correlation coefficients but applied very different inclusion thresholds.
- Cross-modal learning methods need clearer, objective definitions of an “unrecoverable” signal segment, so that performance claims are not quietly built on datasets that have already discarded their hardest cases.
Progress on any of these fronts would do more to close the PPG-ECG gap than another incremental filtering algorithm.
A researcher’s checklist for PPG-ECG comparison studies
Anyone designing a validation study should record simultaneous PPG and ECG with matched, documented sampling rates, apply a pre-registered SQA threshold rather than one chosen after seeing the results, and report both Bland-Altman limits of agreement and correlation coefficients rather than either alone. Per-subject data-loss percentages belong in every results table, not just the aggregate.
When communicating results to mixed clinical and technical audiences, state uncertainty plainly and describe the ECG confirmation pathway rather than implying PPG alone settled the question. Our own practical write-ups on wrist versus chest heart-rate accuracy and HRV behaviour during sleep apply this same discipline when interpreting trend data from wearables.
— Sam
Voltra devices for longitudinal pulse and trend monitoring
We build Voltra devices for exactly the use case the evidence supports: continuous, low-friction trend monitoring rather than diagnosis. Every Voltra device tracks pulse rate, HRV trends and sleep quality without a subscription, and the companion app stays free for life, so long recording periods needed for meaningful trend data never run into a paywall.
- For general wellness and sleep trend tracking, VOLTRA AIR at £99.99 covers the core metrics at the lowest entry price.
- For researchers running longer pilot recordings, VOLTRA PRO at £129.99 adds extended battery life for multi-day data collection.
- For a balance of features and cost, VOLTRA CORE at £149.99 suits both individual tracking and small pilot studies.
None of these replace ECG when a diagnostic question is on the table. We position our devices as the screening and trend layer that sits before, not in place of, an electrocardiographic confirmation pathway. Clinicians working through arrhythmia recognition as part of their own training may find the arrhythmia flowchart from Zero Deficit a useful companion reference on the ECG side of that pathway.
This article is general information, not a substitute for advice from a qualified doctor. Consult a qualified healthcare professional about your own circumstances before acting on anything here.
FAQ
Is ECG more accurate than PPG?
For diagnostic questions such as arrhythmia classification or waveform morphology, yes: ECG is the reference standard because it records the heart’s electrical activity directly. For resting pulse rate and long-term trend tracking, well-processed PPG can approach ECG-level agreement, though high-frequency HRV metrics remain a persistent weak point for PPG.
Is PPG accurate for heart rate?
PPG is generally reliable for heart rate at rest or during low motion, with strong agreement to ECG-derived heart rate under those conditions. Accuracy drops during movement unless the device applies motion-compensation techniques such as accelerometer-based filtering.
How accurate is PPG for blood pressure?
PPG alone does not measure blood pressure directly; some research systems estimate it indirectly using pulse transit time or waveform features, but these approaches are still an active research area rather than a validated clinical measurement method. Any blood pressure figure derived from PPG should be treated as experimental unless backed by device-specific regulatory validation for that purpose.
What is a normal PPG?
A “normal” PPG waveform shows a clear systolic peak followed by a dicrotic notch and a smooth diastolic decline, repeating consistently with each heartbeat. There is no single numeric reference range for PPG itself, since it is a waveform shape and amplitude signal rather than a diagnostic value like heart rate or blood pressure.
Which is better, PPG or ECG, for arrhythmia screening?
PPG-based screening in optimised systems has reported sensitivity around 95.6% and specificity around 99.2% for atrial fibrillation detection after excluding low-quality segments, making it a strong screening tool. ECG remains required to confirm any arrhythmia diagnosis flagged by a PPG-based screen.
Share:
UK: Measure and Improve Metabolic Age in 2–3 Months With NHS Guidance