Preface

Much progress has been made in recent years in the "eternal quest" for quieter passenger-carrying-aircraft interiors. Advances have been made both in the types of passive materials used for aircraft thermal and acoustic insulation and in highly complex, electronic noise cancellation systems currently under investigation (and even in limited use) by most large aircraft manufacturers. This progress has been driven by the need for low levels of noise in high-priced corporate aircraft. Such aircraft - more often than not - have highly individualized interiors fitted by "aircraft completion centers" specializing in just about every facet of aviation technology. This includes the design and fabrication of ultra light weight furniture, beautifully hand fashioned to customer specifications by highly skilled craftsmen. Such furniture utilizes costly laminates of rare hardwoods and is realized in every conceivable style from accurate Louis 14th. reproductions to "Ultra Modern". Of course these wonderful vehicles would be incomplete without the joys of an elaborate entertainment system and the added safety of the most up to date avionics package. Retrofitting these "high comfort" interiors may ultimately cost many millions of dollars - but then these aircraft are amongst the most costly and sophisticated flying machines in existence (Gulfstream V's, Boeing 737's, MD 80 series to name but a few). Owners of such enormously costly transportation are intolerant of anything other than "the very best" and this is especially true with regard to tolerance for discomforting noise levels in flight.

It has become the practice on the part of the aviation industry concerned with producing these "quieter" interiors to measure the effectiveness of noise treatment systems by calculating the SIL's or Speech Interference Levels for various locations inside the aircraft (in combination with simple dB"A" measurements) whilst the aircraft is in flight. SIL's are derived from "octave band" acoustical data produced by either analyzing directly; or more commonly by analyzing digital recordings of noise at various locations within the aircraft interior whilst the aircraft is at typical cruise altitude and speed. To be of most use, analyzed locations must represent the position of a passenger's or crew member's ears whilst seated in normal locations within the aircraft; however a thorough analysis will comprise data sets taken throughout the aircraft including the aircraft interior walls, ceiling, floor and the various bulkheads to determine possible noise entry paths and subsequent insulation and isolation needs.

At least one "trickle down" benefit of the above to the "average" flyer is evident in the gradual emergence of quieter airliners. Most, if not all commercial airline operators are stubbornly reluctant to add anything other than the minimum required thermal /acoustic insulation because of the cost of flying the added weight. However, advances noted above are yielding materials which are acoustically more effective for a given weight and which are far more resistant to moisture absorption and subsequent undesirable weight gain than traditional materials. Moisture absorption is a problem of enormous proportion for all airline operators (a '747 can gain thousands of pounds in added weight from moisture absorbed by the thermal/acoustic insulation materials after just a few months of passenger carrying duty). Once acquired, this moisture is difficult to remove and substantially increases the cost of operating the aircraft. A solution to this problem alone will probably be sufficient to speed a change over to more effective acoustic insulation systems for commercial airliners.

Brief history

Much of the work leading to our current understanding of speech intelligibility and corresponding interference factors began after World War II. The renowned acoustician Leo L. Beranek along with French, Steinberg and others, studied factors affecting speech intelligibility and devised a number of ways of quantifying these factors. One such quantity was named the Articulation Index. In calculating this, the speech band (Almost all speech information is contained in the frequency band from 300 Hz. to 7000 Hz.) was broken into a number of frequency bands - each of which appeared to contribute equally to the overall intelligibility of speech. The speech energy in each band was then related to the available dynamic range within that band. Experiments had been conducted that showed that for speech to be clearly intelligible, the noise content within each frequency band containing speech information had to be at least 30 dB below the speech energy in that band (i.e. speech arriving at the listener had to have at least a 30 dB signal to noise ratio within each relevant band). Further, the psycho-acoustic effects of masking (that is, the observation that one sound will tend to hide a sound of similar or slightly higher frequency which is a lower in level) were being recognized and compensated for in the calculations of Articulation Index. For the most part the Articulation Index was calculated by determining relative areas from a graphical chart with curves representing the dynamic range of speech, i.e. peak levels, average levels and minimum levels for each of the individual frequency bands. Other methods involved complex calculations which rendered the index impractical for most purposes. One consequence of this work was the development of a single number - Speech Interference Level - scale to quantify the ease of speech communication (telephonic or person-to-person) for individuals in a noise environment.

Although the original studies used a greater number of narrower bands, Beranek eventually concluded that use of the 3 octave bands, 600 -1200 Hz., 1200 - 2400 Hz. and 2400 -4800 Hz, was sufficient for useful analysis under most circumstances and determined that a simple arithmetic average of the sound pressure levels within the three bands yielded a meaningful number. Beranek did, however, find it important to add information from the 300 Hz. to 600 Hz. band if the SPL (Sound Pressure Level) in that band was 10 dB or more higher than that in the 600 Hz. to 1200 Hz. band. (The 10 dB being derived from Fletcher-Munson "equal loudness" contours for typical speech levels and the contribution of associated masking affects).

In 1967 Webster extended Beranek's work with a slight modification to the method of determining SIL's so that the "preferred" octave bands of 500 Hz., 1 kHz. and 2 kHz. could be used. This method became known as the PSIL or the three-band Preferred-octave Speech Interference Level quantity. Although similar, calculated SIL's based on the redefined octave bands yield numbers typically about 3 dB less than with Beranek's original bands.

It is interesting to note Beranek's suggestion of the need to add information from the 300 to 600 Hz. octave band into SIL calculations when conditions so required. We should view this as an indication of the importance placed on the lower frequency components of speech in the formative work in this field!

Application

It is important to realize, particularly in view of the work upon which the SIL scale was based, that an erroneous understanding of the this quantity persists throughout the aircraft cabin noise suppression industry.

1. The SIL scale was not meant to measure perceived "noisiness". In fact recent work on perceived cabin noise levels would suggest that dB"A" weighted noise measurements are still more informative than the more widely (mis-) used and misunderstood SIL measurements particularly when based on 1000, 2000 and 4000 Hz. octave bands.
2. When used as a measure of "noisiness", the SIL scale breaks down in meaningfulness when the steps between the octave bands become large. This typically results in much lower numbers than the predominant octave band noise would justify! Again, by way of example: it is not unusual to see octave band numbers of about 70 dB for the 1 kHz band, 55 dB for the 2 kHz. band and 40 dB for the 4 kHz. band. This yields an SIL number of 55 dB even though we have a noise component of 70 dB in the 1 kHz band! This 15 dB difference will almost certainly lead to a questionable interpretation of the result! That is, the 55 dB SIL number suggests a "noisiness" level below that which is actually perceived . As a further illustration of the potential for misunderstanding; consider two different noise environments. In the first we measure octave band SPL's of 56 dB, 55 dB and 54 dB. In the second we measure levels as in the first example of 70 dB, 55 dB and 40 dB. Both environments yield 55 dB SIL measurements however the perceived "noisiness" of the two environments would be very quite different!
3. The omission of the 500 Hz. octave band - as has become the practice in aviation related SIL measurements - renders the use of SIL numbers further questionable, particularly when Sound Pressure Levels within that band are significant! (for example: in a typical jet aircraft SPL's in the 500 Hz. octave band (353.5 Hz. to 707 Hz.) are often higher than levels in the adjacent 1000 Hz. octave band. In this regard, Fletcher - Munson and masking compensation required for inclusion of these frequencies would be much less than the 10 dB required for the 300 Hz. to 600 Hz. band as determined by Beranek, particularly in respect of the higher overall noise levels experienced in aircraft cabins and the resultant "flattening" of the Fletcher - Munson contours at these levels.

Measurement induced error

Measurements of noise levels aboard aircraft are often conducted by persons not intimately familiar with acoustical science, the related techniques and practices and the need for an understanding of the total acoustical environment. Often measurements are conducted in less than ideal circumstances and with little understanding of the affect on the result of such variables as indicated air speed, pressure altitude, cabin pressure, outside air temperature, measurement locations, observer effect and - of extreme importance - instrument calibration, measurement duration and averaging method. This along with the "pressure" to produce "low numbers" and the unfortunate lack of standardized measurement practices generates a large unknown in any attempt to compare numbers from one source (Aircraft manufacturer or Completion Center) to another.

Averaging dB levels

I will make special note here of one common problem that persists. This is in the use of simple arithmetic averaging to average dB SPL's. This can lead to very deceptive results. In order to properly average decibel Sound Pressure Levels within an environment the dB levels must first be converted to a power number (divide the dB SPL by 10 and take the anti-log. base-10 for each level); the power numbers are then averaged in the normal arithmetic way. The resultant average is then converted back into dB SPL (take the log. base 10 of the average power number and multiply by 10). This technique gives a much more meaningful representation of the true SPL average in a space, particularly when measurements are taken at equally spaced intervals through the environment of interest and at sufficient locations to properly represent the sound field.

Remedies and Recommendations

In an attempt to standardize measurement practices and techniques we would strongly recommend the adoption of - at a minimum - the following:

1. All sound measuring or recording instrumentation should be calibrated with corrections applied for "sea level" normal pressure and temperature calibration levels. When using recording apparatus: A Standard Calibration tone of precisely known level should be recorded just prior to and at the completion of a survey. The barometric pressure and temperature at the time of calibration should be noted. Calibration devices should be checked and verified regularly and observers must be instructed in the care and proper techniques required for proper calibration! Calibrators (and couplers if used) should only be used with microphones for which they were intended!
2. Flight variables (pressure altitude, indicated air speed, cabin pressure and weather conditions, etc.) should be standardized for particular types of aircraft to the greatest degree practical.
3. All measurement microphone locations must be carefully described and logged. Measurements along the cabin center axis should be used as a reference for all other measurements. Measurements at seat locations should include at least one measurement at a non-present, theoretically seated person's ear level. (Non-present, theoretically seated persons are available from Flight Environments Inc. at a nominal cost).
4. Measurements or recordings must be of sufficient time and of an averaging type to yield data with acceptable confidence limits. For most modern spectrum analyzers operating in the octave or third-octave mode and assuming the nature of most jet aircraft noise to be random, this means that recordings or measurements of at least 15 seconds (sufficient time to give a minimum of 254 non-overlapped samples at about 10 kHz overall bandwidth) are required to give 99% confidence limits of less than .25 dB at frequencies down to 125 Hz . Averaging should be rms summation power averaging. This gives equal weight to each sample or time period and results in a true "power average" of SPL's over the measurement period.
   
   NOTE: In particular it should be recognized that eyeball averaging of moving meter or digital displays is extremely susceptible to error. Correct averaging of decibel quantities requires logarithmic conversion to power, averaging of resultant power levels and then anti-log conversion back to decibels (see above section on dB averaging). This is not easily accomplished by guesswork and can lead to significant errors even if reading fluctuations are as little as plus or minus 2 dB!
5. Only high quality measurement type microphones and recorders should be employed. Condenser ( Electret or powered) microphones by Bruel & Kjaer, ACO and Ono Sokki are generally the most acceptable. The chosen microphone should be of the random-incidence type. Professional DAT type audio recorders are generally of high quality and so long as the above outlined calibration techniques are employed and methods of preventing inadvertent level changes are adopted, recorders of this type should be more than adequate.
6. Many excellent spectrum analyzers are available from a wide variety of companies including a number of PC based systems that offer powerful analytical tools. Of most importance is calibration of such devices; further, manufacturers procedures should be followed carefully. Care must be exercised with pure tone calibration of any spectrum analyzer as errors can result as a consequence of the inherent "ripple" or unevenness of the "flat top" portion of the digital filters employed to create the octave or fraction-octave bands and care must be taken to ensure that this factor is corrected for. Analyzers should be set up so as to give results according to the above guidelines, i.e averaging numbers set to at least 254, averaging mode set to summation rms. power average. (sometimes called linear averaging) and Time window set to Hanning (most available and acceptable).
7. SIL numbers must be calculated from unweighted octave bands. Preferably PSIL (500 Hz. 1000 Hz. and 2000 Hz.), SIL (100 Hz. 2000 Hz. and 4000 Hz. ) and dB"A" quantities should be given for each measured location.

Conclusion

Hopefully, with the adoption of these or similar guidelines we can eliminate some of the confusion and deception that has prevailed in our industry for too long. This can only lead to quieter aircraft and ultimately improved customer trust of our industry's claims.