Sulfuros de cobre y hierro

The downsides and limitations of the use of MLS in acoustic measurements as highlighted in Dunn & Rife (1994), Dunn & Hawksford (1993), and Vanderkooy (1994), are chiefly inherent to its susceptibility to non-linearity and time-variance. They are summarised in the following.

Time Variance Susceptibility

For RIR analysis, the condition of time invariance, which is one of the assumptions behind LTI systems, is unlikely to be guaranteed for several practical reasons. Air drifts, temperature changes, and people’s movements modify the physical states of enclosures. This results in RIRs that can be slightly different from measure to measure. The assumption of time invariance will therefore hold true only if it is slowly compared with the measurements duration. For example, the BS EN ISO 18233:2006 standard sets the amount of variance that can be accepted during measurements, and

also highlights that some measurement methods are more or less prone to incur in time-variance artefacts. For example, it states that MLS, and more generally noisy signals, are more susceptible to time variance than chirp signals.

As reported in Vorlander & Kob (1997), MLS based measurements can be affected by two types of time variance, “intraperiodic” and/or “interperiodic”. The intraperiodic relates to quick time variance effects occurring during a single MLS (i.e. air drift produced by the ventilation/air conditioned systems), and is particularly detrimental when long MLS sequences are used (e.g. when measuring large halls). On the other hand, interperiodic refers to slow time variance effects (like slow changes in temperature), which may affect the efficiency of a synchronous averaging.

Svensson & Nielsen (1999) estimated that intraperiodic and interperiodic time variance causes an “apparent” noise in the recovered RIR, of noise level by 6 dB /and 12 dB per octave respectively. Time variance affects in greater measure both higher frequency and the late part of the RIR, as it is more sensitive to small variations of time of flight path of acoustic waves, and as a consequence, the measured RT may be underestimated (Bradley, 1996).

Detecting environmental changes like temperature and airflow can also be used to monitor the degree of variance during measurements, or as suggested by Svensson & Nielsen (1999), post-averaging can help to control the amount of variance introduced in an average measurement by discarding the terms that exceeded a determinate threshold. It should be noted, however, that Satoh et al. (2002) reported that a

temperature change of 0.1 degrees already leads to gross errors in the MLS-based measurements, with the high frequencies being the most affected.

Noise Rejection and Averaging

An attractive feature of an MLS is its ability to deal with noise. Its pseudorandom nature phase-randomises every extraneous noise, both impulsive and steady, with the consequential effect of spreading the noises along the recovered RIR. Moreover, since MLS is also a deterministic signal it can be synchronously averaged leading to a +3 dB increment of the SNR for each doubling of the number of averages, if background noise is uncorrelated. This, however, holds only true when time variance is negligible, as repeated measurements will give exactly the same responses. The reason is because identical system outputs accumulate over a large number of repetitions, while the background noise is uncorrelated among these repetitions and will be averaged out Rife & Vanderkooy (1989). The length of MLS used for a measurement is commonly chosen 2 to 5 times longer than the (supposed) RT of the room (Stan et al., 2002).

The MLS noise immunity, and averaging techniques allow measurements to be performed under very low or even negative signal-to-noise ratios (Nielsen, 1996), which gives rise to the idea of using MLS for occupied measurements. However, it has been proven (Serafini et al., 2010) that time variance will eventually halt the SNR increment due to the averaging process, making it reach a maximum value before decreasing to lower values. Lengthy measurement sessions are therefore unlikely to produce useful results.

Satoh et al. (2002) demonstrated that MLS measurements using synchronous averaging could lead to gross errors in the estimation of RT from a temperature change of 0.1 degrees. Therefore, MLS averaging must be performed with extreme care, especially when measurements are performed in large spaces (i.e. an auditorium or a cathedral) where temperatures may change very quickly and air drifts may be quite substantial.

Artefacts and Crest Factor

Acoustic spaces like rooms, halls, etc. are perfect linear systems. Nonetheless, the devices used for acoustic measurements tend to show some degree of non-linearity especially when operating close to saturation levels, i.e. high amplitude/volume. Vanderkooy (1994) accurately explained the MLS sensitivity to non-linearity, and showed that peak-like artefacts will appear along the deconvolved RIR. He also noted that artefacts are related to the particular MLS used, and then published a list of sequences that had a better behaviour when applied to non-linear systems.

However, if MLS with different lengths or tap-configurations are averaged together artefacts will be reduced because artefacts appear in specific positions along the RIR, which is dependent on the particular MLS used (Greest & Hawksford, 1995). Commonly, a useful, but very time-consuming practice to reduce artefacts consists of carefully calibrating the amplifiers by selecting the maximum MLS level that produces the least amount of distortions. It is worth noting that, although the theoretical CF of an MSL is of 0 dB, the digital-analogic converter and the anti- aliasing filter of the measurement device can lead to a CF as high as 11 dB (Müller & Massarani, 2001).

Hybrid MLS

Paulo et al., (2009), proposed the use of a particular technique to increase the SNR in MLS measurements. Two approaches were used and combined to form what he called a hybrid MLS. A set of eight ML sequences (only a 15th and a 17th order MLS were tested), were fed into a room and the room’s response acquired. To simulate sing such method in occupied conditions, disturbances like speech or music (various genres) were used during the measurements.

The first part of the method consists of dividing each sequence into “n bits” long slices (for n = 1024 for speech and n = 2048 for music, the method gives the best results, as stated in Paulo’s work) and measuring the mean square energy (MSE) of each of them. Since the MLS power is constant in all sequences but disturbances vary, by comparing the MSE of each slice with their homologues in each sequence, a new ML sequence can be built using the slices with the lowest SNRs. In his work Paulo’s described the procedure as a “noise scrambler”, which gives an MLS with the lowest possible SNR.

The second part of Paulo’s method involves a multi-band filtering of each sequence (both third-octave and octave bands filter bank were experimented), and similarly to the first method each sub-band is compared to their homologues to find the ones with the lowest MSE. However, he also pointed out that in the case scenario where all homologous slices have low SNR, the method gives worse results than when all sequences are averaged together.

Paulo’s work concluded that the combination of both techniques (hybrid MLS) gives an overall improvement of the SNR, compared with the standard MLS technique, as high as 24 dB for speech disturbances and between 6 and 10 dB for music disturbances (depending on the music genre). In general, he stated that the method achieves a better SNR than conventional MLS techniques in presence of high non- stationary background noise; an advantage that may be exploited to reduce the measurement time in slightly time-variant systems.

Inverse Repeated Sequences (IRS)

To overcome some of the limitations of MLS, mainly its non-linearity susceptibility, a different version of the technique named ‘inverse repeated sequence’ (IRS) was proposed to partially solve the problem (Ream, 1970). The basic idea behind the IRS technique was to form an MLS sequence by joining two equal MLS sequences but with the second being an inverted copy of the first (i.e. inverting the signs of the MLS bits), as proposed by Dunn & Hawksford (1993). In such a way the odd-order distortions will be cancelled out, which is a real advantage only for those systems that exhibit such types of distortions. Another drawback is that the deconvolution of the system’s response cannot be done using the FHT, but it requires the classic FFT based deconvolution (Angelo Farina & Righini, 1997). Moreover, as proved by Stan et al. (2002), differences between MLS and IRS tend to disappear when a low-level MLS is used because of the minor introduced distortions.