Room effects

It is generally assumed in headphone testing that the room in which the measurements are conducted has no effect on the outcome. But with open-back headphones in particular, and 'floating' designs even more so, this is simply not true. Open-back headphones both emit significant levels of sound into the room and provide little attenuation of external sound. The inevitable result is that room reflections find their way back through the headphone to the measurement microphone, at levels which can affect frequency response traces and cumulative spectral decay waterfalls.

HTL's measurements are made with the artificial head atop a stand, positioned in the measurement space so as to achieve >10ms time delay between the start of the headphone's impulse response and arrival of the first strong room reflection. That's a good start but the issue remains.

It is addressed in HTL's uncorrected frequency response measurements of floating and open-back headphones not by applying frequency domain fractional-octave smoothing (as commonly used elsewhere, which smooths away relevant as well as room-induced response features) but by using an adaptive windowing technique to analyse the impulse responses obtained during testing.

Derivation of the frequency responses is performed sequentially using a rectangular time window whose length is inversely proportional to the centre frequency of each FFT bin in turn. So, for example, at 41.015625Hz the window length is 23,406 samples (equivalent to 244ms at the measurement sampling frequency of 96kHz) to provide a frequency resolution one-twentieth that of the bin centre frequency. As frequency increases, the measurement window progressively shortens until at frequencies above 2kHz it is fixed at 10ms (960 samples). The outcome is that the frequency response above 2kHz is truly anechoic (because the measurement is free of room reflections within the time window), while at lower frequencies the disruption caused by reflections is minimised consistent with maintaining response accuracy. At these frequencies the effect of the adaptive window is similar to, but not the same as, 1/14th-octave frequency-domain smoothing.

This adaptive windowing process significantly lengthens the measurement processing time. But the improvement in the results adds confidence that artefacts resulting from room reflections are negligible, and that the fine structure of the frequency responses is due to the headphone and not the room. As an example, shown below are uncorrected responses for the Warwick Acoustics Aperio calculated using the full impulse response (first graph) and with adaptive windowing (second graph). Using browser zoom will help you see the differences more clearly.

Adaptive windowing is only applied to calculation of the uncorrected frequency responses, not the responses that appear in the leakage or acoustical crosstalk graphs (if present). So these give an indication of the improvement that adaptive windowing achieves with each headphone.

For the CSD waterfalls a different approach is taken. The analysis window is fixed at 960 samples (10ms at 96kHz sampling rate) to eliminate all room reflections from the measurement, and 1/12th-octave frequency domain smoothing is applied. The combined effect is a limitation of frequency resolution which 'fattens' any ridges caused by low-frequency resonances but has minimal effect on those above 1kHz, while removing spurious features resulting from room reflections. The examples below, for the Abyss AB-1266 Phi TC, show the CSD waterfall if 960 samples of the impulse response are used to eliminate reflection effects (first graph) or 2000, 4000 or 8000 samples in which progressively more room reflectons are included (remaining graphs). Pay particular attention to the gap between 1.5kHz and 3kHz in the first graph, which becomes filled with room reflection artefacts in the other waterfalls, which worsen as the analysis window is lengthened. The downside of the 960-sample result is that details of resonances below 1kHz are lost because of the resticted frequency resolution, in particular here the high Q (sharpness) of the resonance at 350Hz.