Respiratory impedance spectral estimation for digitally created random noise

Area of Science:

• Pulmonary physiology
• Biomedical engineering
• Signal processing

Background:

• Respiratory input mechanical impedance (Zrs) measurement is a noninvasive method to assess lung mechanical function.
• Commonly, Zrs is estimated using random noise pressure signals and Fast-Fourier Transform (FFT) techniques.
• Digital generation of input signals introduces potential signal processing artifacts affecting Zrs estimation quality.

Purpose of the Study:

• To quantify the impact of signal processing parameters on the quality of Zrs spectral estimates.
• To evaluate factors including ensemble averaging, windowing, signal-to-noise ratio (SNR), and pressure spectral magnitude.
• To determine optimal conditions for accurate digital Zrs estimation.

Main Methods:

• Simulated random noise driven pressure and flow data using three models covering different frequency ranges (0-4 Hz, 4-32 Hz, 4-200 Hz).
• Evaluated Zrs estimate quality based on the number of averaged runs, window type (rectangular vs. Hanning), flow SNR, and pressure spectral magnitude shape.
• Assessed quality using coherence (gamma^2) as a primary metric.

Main Results:

• Poor Zrs estimates (coherence < 0.75) were observed with uniform pressure power and SNR < 100 for 10 averaged runs.
• High coherence (> 0.95) was achieved with SNR > 200 and 10 averaged runs.
• Matching pressure spectral magnitude to impedance yielded high coherence (> 0.91) even with 5 runs and SNR of 20.
• Rectangular windowing proved superior to Hanning for digitally generated data.
• Coherence alone may not be a definitive measure of Zrs quality.

Conclusions:

• Accurate Zrs estimation is best achieved by matching the input pressure signal's magnitude to the impedance magnitude (P(jω) to Zin).
• Utilizing rectangular windowing is recommended for digitally generated signals.
• Careful consideration of SNR, averaging, and spectral matching is crucial for reliable Zrs measurements.