The Impact of Frequency on Resistance and Compliance Settings in Lung Simulators

Introduction

A recent study in the Journal of Biomechanics (Ferencz et al., 2026) compares four lung simulators: the ASL5000, alveo, SmartLung 2000, and Training & Test Lung. It assessed their accuracy in mimicking a linear single-compartment lung model across frequencies from 0.25 to 15 Hz and flow rates from 5 to 40 L/min, using a CITREX H4 flow analyzer as a reference.

The alveo simulator demonstrated the most consistent accuracy across frequencies, being "the least affected by the frequency of the input stimulus" (Section 5).

This whitepaper summarizes the study's results, explains the implications of "frequency-independent accuracy" for device performance in bench tests, and discusses scenarios in which linearity doesn't significantly affect outcomes, particularly in standard adult breathing tests below 1 Hz. It also mentions a resistance offset in the pre-launch alveo unit under high-resistance settings, addressed in Section 9 alongside verification details from IMT Analytics.

1. Why accuracy matters

Bench testing has shifted from a training tool to a source of preclinical evidence. Submissions to regulators, comparative effectiveness studies, algorithm validation pipelines, and design-control verification increasingly rely on simulator-based data to substantiate device behavior before patient exposure. Ferencz et al. open their introduction by noting that bench testing "has become an essential pre-clinical step in verifying appropriate functioning of ventilatory devices" (citing Sayas Catalán and Patout, 2025).

This is a sensible direction. Patient studies are slow and expensive, while a well-characterized bench can run the same test many times. However, the simulator must also be well-characterized; otherwise, it becomes a confounding variable. For two decades, the ASL5000 has been the standard for active simulation, claiming a "small signal" bandwidth over 15 Hz. Recently, independent measurements have assessed how this bandwidth affects accuracy in actual bench tests. Ferencz, Drath, and Dömer's 2025 study (CDBME 11(1):377–380) started this work, and the 2026 Journal of Biomechanics paper expanded it to four devices.

We are reading that paper from the perspective of an engineer who must choose, configure, and trust a bench.

2. The bench-testing landscape: active vs. passive simulators

Lung simulators can be categorized into two types: passive and active.

Passive simulators (e.g., SmartLung 2000, TTL) use distensible bags and springs, reacting like a balloon to inflation. Their mechanical properties limit their behavior, making them inflexible.

Active simulators (e.g., ASL5000, alveo) employ real-time numerical lung models and actuators, enabling them to simulate complex parameters such as non-linear compliance and direction-dependent resistance. They can respond dynamically to changes in device pressure, unlike passive models.

Each type has its trade-offs: passive simulators are mechanically simple but fixed in behavior, while active simulators are flexible but rely on the quality of their control loops. The Ferencz study evaluates both against the linear single-compartment lung model, a common reference in bench testing, but alveo can simulate more than just this model.

3. What the study did

The methodology is worth understanding before reading the results:

1. Stimulate: A driver — an ASL5000 in flow-pump mode — generated sinusoidal flow patterns at 0.25 to 15 Hz, with amplitudes between 5 and 40 l/min.
2. Measure: A CITREX H4 flow analyzer (IMT Analytics) recorded flow and pressure independently at 200 Hz between the driver and the simulator under test. Independent measurement matters: it isolates the bench-under-test's behavior from any inaccuracy in the driver.
3. Compare: For each frequency and amplitude, the measured impedance magnitude |Z| and the phase between flow and pressure were calculated. These were compared to the ideal response of a linear single-compartment lung model parameterized to the same R and C, computed in Simulink.
4. Summarize: Root-mean-square error (RMSE) was reported for |Z| and for phase, across three lung parameter sets: healthy (R = 5 cmH₂O/l·s, C = 75 ml/cmH₂O), restrictive (R = 5, C = 25), and obstructive (R = 20, C = 60).

Figure 1. Bench-test setup. A driver (ASL5000 in flow-pump mode) generates sinusoidal flow; a CITREX H4 flow analyzer (IMT Analytics) records flow and pressure independently at 200 Hz between the driver and the simulator under test, isolating the bench-under-test from any inaccuracy in the driver. After Ferencz, Drath, Dömer (2026), Section 3.

4. What the study found

We reproduce the headline numbers below. RMSE is a single number summarizing average deviation across all amplitudes and the tested frequency band, so it captures consistency, not just best-case behavior.

Figure 2. Mean RMSE across the tested frequency band for impedance magnitude |Z| and phase, redrawn from Ferencz et al. (2026), Table 3 (lower is better). The mean averages over all amplitudes and frequencies; it understates both the amplitude dependence of the passive simulators (SmartLung, TTL) and the frequency-band concentration of the ASL5000 error — see Section 4. CC BY 4.0.

Reading these tables requires nuance. The mean RMSE values place SmartLung 2000 and TTL nominally at the top — but this number obscures strong flow-amplitude dependence, which the study documents in Section 5. The passive simulators' parabolic resistors produce a low average across the tested amplitudes, while the individual amplitude curves vary significantly. The alveo and ASL5000 — both active simulators — show much less amplitude variability.

Where active simulators differ from each other:

• The ASL5000 reproduces healthy and restrictive parameter sets accurately up to about 2 Hz. Beyond, the study documents "a growing |Z| error … translating to a higher simulated resistance than set", along with "rapidly growing phase error" (Section 4). For the specific set of obstructive parameters, the study reports resonant behavior from approximately 8 Hz, with gain exceeding the ideal curve.
• The alveo, per the same study, "provides a simulation accuracy that is largely independent of flow amplitude and remains the most uniform across frequencies among the examined simulators" (Section 4). Phase error stays "uniformly low across tested frequencies" (Section 5). The exception is the obstructive parameter set, where the study found a |Z| offset corresponding to a resistance higher than set — addressed in Section 9 below.

The "uniformity across frequencies" finding is the substantive engineering result. When the error concentrates on a specific frequency band — as it does for the ASL5000 above 2 Hz — the Mean RMSE understates the practical impact.

5. What "frequency-independent accuracy" means

The key point is that a simulator must accurately mimic the patient model it represents. When a device sends flow into the simulator, the output pressure should match that of a patient with the configured resistance (R) and compliance (C). If the simulator shows inconsistent resistance across frequencies, it leads to measurement inaccuracies that require corrections based on unknown signal content.

Think of a measuring stick that reads consistently at the center but varies at the ends; it can measure, but you can't trust the readings without adjustments. A frequency-independent simulator reads consistently across all frequencies and flow ranges, enabling clean measurements without additional corrections.

Figure 3. The measuring-stick metaphor. A frequency-dependent bench reports the same true value differently across a device's operating range (28 / 30 / 32 cm); a frequency-independent bench reads the true value at every frequency. Illustrative — not measurement data.

In practice, "frequency-independent" means:

• Consistent impedance at the lung opening, regardless of the speed of pressure change.
• Accurate amplitude response across all flow levels.
• Low phase error, ensuring proper timing alignment of pressure and flow events.
• No resonance in the simulator's control loop, so you measure the device alone, not a combined system.

These characteristics mitigate risks during bench testing, ensuring observations are accurately attributed to the device rather than the testing setup.

Figure 4. Schematic frequency response. Conceptual interpretation of how impedance error varies with frequency for alveo and the ASL5000 parameter sets, with the FOT band (4–10 Hz) shaded. Curves illustrate qualitative behavior after Ferencz et al. (2026), Sections 4–5; not extracted plot data. CC BY 4.0.

6. Where frequency-independent accuracy is important

We admit that the linearity argument is not equally important for every kind of bench work. It matters most where the device under test exercises the simulator outside the typical adult-breathing band of 0.17–0.33 Hz.

Figure 5. The use-case spectrum. Breathing rates and bench-test application bands on a log-frequency axis. Below ~1 Hz, all four simulators perform acceptably; above it, FOT, HFJV, HFOV, fast NIV ramps — frequency-independent accuracy becomes decisive. Application bands after Ferencz et al. (2026), Section 6; bands are indicative, implementation frequencies vary by manufacturer.

6.1 Sleep therapy — FOT-based apnea classification

The clearest case. CPAP and APAP devices that distinguish central from obstructive apnea use the Forced Oscillation Technique (FOT). During a suspected apnea, the device sends low-amplitude pressure oscillations and observes the flow response. Implementation frequencies vary by manufacturer between 4 Hz and 10 Hz at 1 cmH₂O peak-to-peak. An open airway returns a different impedance signature than a closed one across this band.

This bench test puts a small signal at exactly the frequency band where active piston-driven simulators have historically struggled. The Ferencz 2025 predecessor paper documented a perceived resistance of ~8.5 cmH₂O/l/s, instead of the set value of 5 cmH₂O/l/s. The 2026 paper confirms the same behavior in the wider comparison.

An algorithm developer testing a classifier on such a bench must disentangle two effects: what their algorithm did, and what the simulator did to the signal before it reached the device. The cleaner the simulator, the cleaner the algorithm validation.

6.2 NIV and bilevel — fast pressure transitions

Bilevel ventilation, pressure-support breath delivery, and leak-compensation responses all introduce steep pressure changes. A 100 ms rise time at IPAP onset carries significant spectral content above 5 Hz on top of the breathing fundamental.

Ferencz et al. report that the ASL5000, under the obstructive parameter set (R = 20 cmH₂O/l·s), shows resonance starting at approximately 8 Hz. The study notes that this "may further complicate the testing of ventilatory devices employing multiple pressure levels with steep ramps. … Rapid pressure transitions, that introduce high-frequency components, may cause resonance between the simulator and the ventilatory device" (Section 5).

A simulator that does not resonate at those frequencies removes the question of "is this the device or the bench?" investigations. For teams working on pressure-targeted modes, asynchrony detection, or leak compensation, this is operationally significant.

6.3 Neonatal and pediatric ventilation

Neonatal breathing rates range from 30 to 60 breaths per minute (0.5–1 Hz). Pediatric breathing extends similarly to adult ranges. Some modes — like High Frequency Oscillation Ventilation (HFOV) at 5–20 Hz, or High Frequency Jet Ventilation (HFJV) at 5-7 Hz — operate an order of magnitude above adult breathing.

This is the application domain where the linearity argument is not optional. A simulator that breaks above 2 Hz cannot validly evaluate a Babylog, a Sensormedics 3100A, or a Bunnell HFJV. The Ferencz study explicitly cites the predecessor finding by Stránská et al. (2014), who reported that the ASL5000 "fails to simulate preset mechanical parameters during HFJV" — work that predates the present study by over a decade.

For neonatal and HFOV device development, the practical implication is that a simulator with frequency-independent accuracy is a prerequisite for meaningful bench evidence.

6.4 Research reproducibility and cross-lab comparisons

A subtler but important case. Academic bench research, multi-site algorithm validation, and consortium pre-clinical studies depend on the comparability of results across instruments and labs. If two groups use the same simulator with the same configuration, they should produce the same results — and if a third group uses a different simulator, the differences should be ascribable to the device or the configuration, not to the simulator's frequency response.

7. Where the linearity argument does not materially help

Most adult NIV and CPAP bench testing is conducted at 10–20 breaths per minute, corresponding to 0.17–0.33 Hz. Below 1 Hz, all four simulators perform as intended in the study. The differences in |Z| and phase response between alveo and ASL5000 in this band are small enough not to drive a purchase decision.

If your application is squarely within the standard adult NIV/CPAP territory, the study does not provide a reason to switch simulators. What it does is provide independent peer-reviewed validation that alveo's underlying engineering meets a high bar. The actual differentiation at standard breathing rates lies in capabilities the study did not evaluate at all, because the linear single-compartment reference model it used cannot represent them. We turn to those next.

8. Beyond the linear single-compartment model

The Ferencz study evaluated alveo using the same reference model applied to all four devices: linear resistance, constant compliance, and a single compartment. This is bench-testing practice for a reason — it is well-specified and reproducible — but it is also a deliberate simplification of how real lungs behave.

Real airways resist differently on inspiration than on expiration. Real alveolar tissue recruits at low volumes and overdistends at high ones, producing the familiar sigmoidal pressure-volume curve. Real obstructive disease modifies airway behavior as a function of flow, not just of pressure. Real ARDS patients exhibit a recruitable lung that responds dynamically to PEEP. None of this can be expressed in the linear single-compartment model.

alveo's design includes configurations the study did not evaluate:

• Sigmoidal compliance — a parameterizable pressure-volume curve that captures recruitment and overdistention behavior. Useful for any device whose algorithm responds to volume-dependent lung mechanics (most modern volume-targeted modes, ARDS-aware PEEP search).
• Direction-dependent resistance — inspiratory and expiratory resistance can be configured independently. Useful for any device tuned to expiratory-flow patterns, including most modern pressure-cycled triggers.
• Configurable lung recruitability — simulates the recruitment-derecruitment dynamics that distinguish ARDS bench tests from healthy-lung bench tests.
• Both linear and parabolic resistance models — allow like-for-like comparison with passive simulators when needed, and clean linear behavior when the test setup requires it.

These are the features that determine whether a bench test resembles a real patient. The Ferencz study, by confirming frequency- and amplitude-independent accuracy under the simplest model, gives us a credible foundation: the underlying engineering meets independent peer-reviewed scrutiny. The richer models live on top of that foundation.

9. Honest disclosure: the resistance-accuracy specification and what v1.2 changes

The alveo unit used in this study was a prototype device supplied to the authors by IMT Analytics in November 2025. The paper transparently documents that, under the obstructive parameter set (R = 20 cmH₂O/l·s), the simulated impedance magnitude consistently exceeded the set value. From the conclusion (Section 6): "Using a pre-release device provided in November 2025, we observed a resistance offset over the frequency spectrum, particularly when using a high resistance setting. Consequently, actual resistance should be verified to ensure accurate simulation."

These are two distinct findings, and it is important not to conflate them:

1. The frequency- and amplitude-independence finding is architectural. It reflects how alveo's control loop responds across the operating space. It does not change across units, calibration cycles, or parameter sets — although, as the obstructive |Z| row shows, an offset can ride atop the architectural response without changing its shape.
2. The resistance offset finding is a specification question. The unit tested by Ferencz et al. met IMT Analytics' pre-launch linear-accuracy specification of ±20%. The 8.33 cmH₂O/l·s |Z| RMSE on R = 20 cmH₂O/l·s obstructive is consistent with that spec. It does not reflect a structural limit of the simulator's frequency response.

What changes with v1.2?

Release v1.2 of alveo, scheduled for early June 2026, tightens the linear-accuracy specification to ±10%, a 50% improvement and the engineering limit of the current architecture. This enhancement applies to both linear and parabolic resistance modes. For tighter absolute resistance than ±10%, fine-tuning the setpoint against measured behavior, as advised by Ferencz et al., remains the recommended workflow.

Independent re-test commitment.

Ferencz, Drath, and Dömer have committed to re-testing the v1.2 production unit using the same methodology described in the Journal of Biomechanics paper. A follow-up peer-reviewed publication will report those results.

In customer-facing communications about the study, we ask readers to keep both findings in mind. The architectural property confirmed by the study stands on its own; the resistance-accuracy specification has improved; an independent re-test is on the way.

10. Practical recommendations for bench-test engineers

For the engineer evaluating a lung simulator purchase or upgrade, here is what we read from the Ferencz study, condensed:

1. Characterize your operating frequency band first. If your device operates the simulator at or below 1 Hz almost exclusively, the linearity argument is not decisive. Look at patient-model fidelity, software ergonomics, parameter range, and integration with your existing test infrastructure.
2. FOT, HFOV, HFJV, Sleep Apnea. Fast bilevel transitions, or neonatal/pediatric breathing rates, are within scope — even occasionally — frequency response matters. Today and increasingly tomorrow.
3. Look at the figures, not just the RMSE table. Mean RMSE is a one-dimensional summary. The figures in this study show which frequencies drive the error, which determines whether the simulator is fit for your specific test.

11. Conclusion

The Ferencz, Drath, and Dömer study is the first peer-reviewed comparison of commercial lung simulators. It found that the alveo simulator's accuracy is least affected by input frequency, supporting the design principles of IMT Analytics. This is crucial for tests above the typical adult breathing band, such as FOT-based apnea classification and high-frequency ventilation.

For standard adult NIV and CPAP bench work below 1 Hz, the findings emphasize engineering rigor rather than influencing purchase decisions. alveo's broader functionalities, like sigmoidal compliance and configurable recruitability, were not evaluated in the study.

The resistance issue noted in the pre-launch unit is being addressed in the upcoming alveo release v1.2 in June 2026, which will improve linear accuracy from ±20% to ±10%. An application note detailing verification data will accompany the release, and the authors plan a peer-reviewed follow-up.

We are sharing this white paper now to highlight key findings and provide insights for the engineering community. Bench-test engineers and clinical researchers are encouraged to reach out for an alveo evaluation and to read the full Ferencz et al. paper, which is Open Access under CC BY 4.0.

References

Ferencz, M., Drath, R., Dömer, B. (2026). Dynamic accuracy of lung simulation hardware: A comparative evaluation. Journal of Biomechanics 204:113366. DOI: https://doi.org/10.1016/j.jbiomech.2026.113366. Open Access, CC BY 4.0.

Ferencz, M., Drath, R., Dömer, B. (2025). Analysis of Performance Limitations of a Widely Used Lung Simulator. Current Directions in Biomedical Engineering 11(1):377–380. DOI: https://doi.org/10.1515/cdbme-2025-0196. Open Access, CC BY 4.0.

Sayas Catalán, P., Patout, M. (2025). On the role of bench testing in NIV device evaluation. — cited as Ferencz 2026 ref.

Stránská, K., Roubík, K., Rožánek, M. (2014). ASL 5000 lung model fails to simulate preset mechanical parameters during HFJV and volume-control ventilation with a decelerating flow waveform in some ventilators.

IngMar Medical. ASL 5000 User Manual v3.6.

Johnson, K.G. (2022). APAP, BPAP, CPAP, and new modes of positive airway pressure therapy.

About the author/sponsor

This whitepaper was prepared by IMT Analytics AG, manufacturer of gas flow analyzers, lung simulators, and respiratory therapy test equipment. IMT Analytics was acknowledged in the Ferencz et al. paper for supplying the pre-launch alveo unit. The paper itself was authored independently of IMT Analytics by researchers affiliated with Pforzheim University.

For an alveo evaluation, production-unit verification data, when available, or further discussion of the topics raised in this paper:

Christian Remus — Product Manager
IMT Analytics AG, Gewerbestrasse 8, 9470 Buchs (SG), Switzerland
www.imtanalytics.com | remus@imtanalytics.com