High inter-channel coherence between signals emitted from multiple loudspeakers can cause undesirable acoustic and psychoacoustic effects. One example is commonly known as “comb-filtering”, which occurs when time delayed coherent signals sum. In sound reinforcement, comb-filtering is perceived by audience members as variations in magnitude response which depend upon their position in the audience area.
This situation is the antithesis of a key objective of the sound reinforcement system designer – that all audience members should, within reason, receive a signal that is faithful to the source material.
Nevertheless, the problem is present in all sound reinforcement scenarios that employ multiple loudspeakers emitting coherent signals to meet audience coverage and sound pressure level (SPL) needs.
Typically, low frequency sound reproduction is handled by a dedicated array of distributed
loudspeakers, each emitting an identical signal (with the possible exceptions of electronic time delay and polarity). Because of the size of audience areas in large-scale sound reinforcement scenarios, it is possible for audience position to loudspeaker path-length differences to be on the order of meters , leading to significant differences in the time of arrival of contributing signals at a given audience location. The result of this is positionally dependant comb-filtering that extends to the low frequency range.
This work seeks to develop a signal processing algorithm that addresses this issue by lowering the inter-channel coherence of distributed loudspeaker arrays so that the summation of their outputs is no longer dependent on their relative phases. This means that regardless of path-length differences, the summation of multiple coherent sound sources results in comparable magnitude responses across audience areas.
Initially, the work investigates the suitability of a method of signal decorrelation termed diffuse signal processing (DiSP), which utilises FIR decorrelation filters termed temporally diffuse impulses (TDIs). Several TDI optimisation methods are described, and the performance of the algorithm is examined using an image-source (IS) acoustic model. Whilst DiSP is found to be successful in reducing magnitude response variation across audience areas in non-enclosed acoustic spaces, in enclosed acoustic spaces the performance of the algorithm is reduced.
This reduction in performance is the result of early reflection/direct source summation. The non time-variant nature of DiSP means that whilst inter-channel coherence is successfully reduced, the output of a single discrete source still maintains high correlation with itself over short time-frames. Therefore, when early reflections sum with their direct source’s output, comb-filtering is produced.
A variation of DiSP termed dynamic DiSP is described, which can reduce the correlation of a discrete source’s output from one time instant to the next. This aspect of the processing mitigates comb-filtering that is a product of direct source/early reflection summation. It is also a potentially useful tool in reducing modal behaviour in acoustically small spaces. These claims are investigated in detail in both simulated and real-world scenarios.
A number of user-variables are described which enable easy control of the algorithm, allowing users to balance the level of decorrelation performance required versus the audibility of the processing. Suitable values for these variables are proposed and their perceptual transparency is investigated using a MUSHRA style subjective test. In this test, the algorithm scored 87.3/100 versus an unprocessed signal score of 91.1/100, showing that dynamic DiSP may be applied in a perceptually transparent manner.
Figures - uploaded by Jon Moore
All figure content in this area was uploaded by Jon Moore
Content may be subject to copyright.