Measurement Noise Floor

Measurement noise floor estimation is the second most important feature of Chirplab, after the actual measurement results. Noise impacts all measurements, and can be particularly difficult to separate from distortion products when trying to measure distortion close to the noise floor of the measurement system. Audio measurement software will often include RMS noise level with different weightings as a dedicated measurement, or will occasionally include noise floor in context alongside measurement results, but will almost never indicate the effective noise floor of the measurement itself.

Chirplab uses a unique approach of taking a sample of the system idle channel noise and passing it through the same calculations as the chirp response signal to estimate the minimum output level that you could expect to be able to measure with that measurement system or setup.

This can help you quickly asses whether the measurements that you take are producing meaninful data. This page provides a couple examples to show how measurement noise floor estimates are calculated and how they can be interpreted in the context of different measurements and test setups.

Noise floor estimation

The Frequency Response measurement process is straightforward, and a measurement noise floor estimate shows how effective the Farina method can be in reducing the impact of noise on the measurement. The measurement starts by calculating the raw impulse response of the system, then applies a window to the impulse response, and transforms the impulse response into the frequency domain to produce the measurement output.

The frequency response noise floor is estimated by instead taking a section of silence preceeding the chirp in the response signal, and using that in place of the chirp response signal

The resulting measurement noise floor estimation shows the noise in the same context as the actual frequency response result

This method allows you to quickly assess the impact that noise has on measurement results and determine whether the data that a measurement produces is valid. The frequency reponse measurement has less than 20dB margin from the noise floor at 20Hz, so the data may not be reliable at the lowest frequencies.

Estimating the measurement noise floor can also help show the effectiveness of different measurement parameters or processing. Skipping the windowing step in the frequency response measurement results in ripples below 50Hz, which correlates with a 10-15dB broadband degradation in the noise floor and a measurement to noise floor margin of less than 20dB

One area where measurement noise floor estimation can be very illuminating is in distortion measurements. Microphones intended for speaker and room measurment use cases usually are calibrated to accurately measure SPL and usually have a flat frequency response and but poor noise performance. When measuring speaker distortion with a noisy mic or with the microphone too far from the speaker, the distortion products can be buried in the noise floor of the system, and most measurement software will present the distortion results without any indication that the data is invalid.

Here is an example of the measurement output from a High Order Harmonic Distortion test, a measure of THD for harmonics 10-20 that is commonly used for Rub and Buzz analysis:

• One speaker driven at the same level for all tests
• Two different microphones at the same distance from the speaker - a Dayton EMM-6 with 29dBA noise floor, and an OpenRefMic with 18dBA noise floor
• Two different analysis programs with the measurement parameters set the same - Audio Precision and Chirplab

The measurements are quite similar, but Audio Precision and Chirplab both measure lower HOHD in some frequency bands with the OpenRefMic than the EMM-6. Looking at Chirplab's measurement noise floor estimates for both microphones clearly shows the difference between the microphones and gives much more confidence that the data from the OpenRefMic is accuracte accross the whole frequency range, while the EMM-6 data below 40Hz and between 150Hz and 500Hz is simply the noise floor of the measurement and should be discarded or given extra margin when testing against and HOHD limit

Limitations of Measurement Noise Floor Estimates

Measurement noise floor estimation can be very useful and is a good sanity check for your test setup when measuring signals that are close to the system noise floor, but it does not capture all sources of noise that could contaminate a measurement, and does not capture all sources of error that could skew measurement results. Some factors that limit the usefulness or accuracy of measurement noise floor estimation are:

• The measurement noise floor cannot be estimated or the estimate would not be meaningful for some measurements, like Phase Response
• Idle channel noise that is non-stationary, like traffic passing by on the street when taking measurements
• Dynamics processing, multi-band compressors, or idle muting that reduces the noise level when there is no stimulus signal present
• Nonlinearities in the system under test, like speaker distortion
• Reverberation, particularly for measurements that do not use any impulse response windowing or time-gating