Errors in MLS Measurements Caused by Time Variance in Acoustic Systems

A study is presented of the effects of time variance on maximum-length sequence (MLS) measurements. Both interperiodic and intraperiodic effects are studied, using time stretching and an analogy with delay modulation as models. It is shown that both types of time variance cause apparent level losses, which typically represent an error with a + 12-dB per octave characteristic relative to an ideal reference impulse response (IR). Furthermore this error is localized around the reference IR and cannot be decreased by averaging or truncation. Intraperiodic time variance might also cause an apparent random noise component, typically with a +6-dB per octave characteristic. This component is spread in time and can consequently be decreased by truncation. Averaging also decreases this noise if the time variance is fast so that the correlation of this apparent noise between MLS periods is low. The existence of such apparent noise indicates that the measured system response is distorted. The MLS measurement under the influence of intraperiodic time variance extracts the carrier transfer function if the system that is measured is viewed as a linear, time-varying system. Measurements were carried out in different rooms, and IR differences from these demonstrated the existence of apparent time-variance noise, increasing by approximately 6 dB per octave, in reverberant rooms. The existence of this apparent noise could also be detected in the initial part of a single IR, before the direct sound. In a large concert hall this apparent noise dominated over the random noise floor above 500-600 Hz and was decreased by averaging. Intraperiodic time variance was accordingly the dominant noise source, which explains the often observed fact that explosive sources might give decays with much better dynamics in large halls than MLS measurements do. Practical consequences of the findings in this study are that both intra-and interperiodical time variance can be detected in a measured IR, especially the type of time variance with a + 6-dB per octave characteristic, and the consequences of averaging can be quantified.