Occupational Soundscapes – Part 10: Communication Systems

     When it comes to communication and communication systems, the definition of “effective” or “satisfactory” can vary significantly.  It depends on the level of compatibility of all components of the system and the potential consequences of a breakdown.  The need for higher performance when conducting safety-critical operations is implicit and, for many, held in the subconscious.  To develop a successful communication system, this requirement should be brought into conscious dialogue, made explicit.
     To support successful communication system design, this installment of the “Occupational Soundscapes” series provides a rapid-fire presentation of system components and recommendations.  Methods for determining the performance required, which affects how a system is designed and operated, are also discussed.

Signal Detection Theory
     For minimally-challenging soundscapes or simple communications, the concepts presented in Part 9 may provide all the information necessary to choose appropriate signals and design an effective system.  As complexity of communications increases or the consequences of miscommunication become more severe, an additional level of analysis may be needed.
     A framework for additional analysis is provided by Signal Detection Theory (SDT).  SDT expands the concept of signal-to-noise ratio (S/N) by assessing the impacts of variance and the application of a decision criterion on the rate of correct interpretation of signals.  A decision criterion establishes the level of certainty, or confidence, that a signal has been received needed to report its detection.
     The choice of criterion influences the perceived effectiveness of a communication system, as shown in Exhibit 1.  The curves correspond to three basic qualitative criteria that can be summarized as follows:
     (a) “Don’t miss.”        (b) “Do your best.”        (c) “Be sure.”
Progressing through these criteria, from left to right in the chart, reduces the proportion of signals reported at a given level because the “standard” for detection is raised.

     To choose the most-appropriate criterion, the consequences of failures and the probabilities of occurrence must be assessed and compared.  Two types of failure are possible:

1. A signal is not reported, though present – a “miss” (Type II error).
2. A signal is reported, though none is present – a “false alarm” (Type I error).

[Type I and Type II errors are discussed in “The War on Error – Vol. V:  Get Some R&R – Attributes” (26Aug2020); the signal in that context is a product defect.]
The “don’t miss” criterion results in the highest rate of “hits” (correct detection of signals) and the highest rate of false alarms.  The “be sure” criterion reduces false alarms to the minimum rate, but maximizes the miss rate.  The “do your best” criterion results in intermediate rates of all metrics – hits, misses, false alarms, and correct rejections.
     Data for correct and erroneous signal detection rates can be summarized in a “stimulus-response matrix,” such as that in Exhibit 2.  The matrix can act as a record of empirical data, represented in each cell by a letter and used to calculate system performance metrics.  It can also be used to present predicted or targeted hit rates, etc. in the design phase of a system.  These are referenced in the format P(x|y), read “the probability of ‘x’ response given ‘y’ condition.”  The probability of a hit (“yes” response in presence of signal+noise) is notated P(Y|S+N) and so on.

     When performance data are available from a system in use, the proportion of each stimulus-response pair can be calculated as follows:
     P(Y|N) = A/(A+C);                 P(Y|S+N) = B/(B+D);
     P(N|N) = C/(A+C);                 P(N|S+N) = D/(B+D).
Also, P(Y|N) + P(N|N) = 1.0 and P(Y|S+N) + P(N|S+N) = 1.0 irrespective of the decision criterion chosen.

     Thus far, consequences of failure – miss or false alarm – have been referenced generically, as they can take many forms.  Quality spill, machine breakdown, property damage, and personal injury, among others with a wide range of severities, are possible consequences of signal detection, or communication, failure.
     To facilitate design decisions, foreseeable consequences of each stimulus-response pair should be assigned monetary values (cost or benefit).  Placing this information in the corresponding cells of the stimulus-response matrix creates a payoff table that can be used to calculate the financial impact of a communication system’s performance.  This information could also be displayed in a tree format, such as that shown in Exhibit 3, where costs (or benefits) are represented by the notation C(x|y), read “the consequence of ‘x’ response’ given ‘y’ condition.”  Costs are represented by negative values and benefits by positive values.  [See “Making Decisions – Vol. IX:  Decision Trees” (23Feb2022) for a thorough presentation of payoff tables and decision trees.]

     While S/N considers average or instantaneous levels of signal and noise, SDT considers the variability of signal and noise levels.  Variability and its effects on signal detection performance is visualized by plotting the probability distributions of signal and noise levels, as shown in Exhibit 4.  The distributions are assumed normal and equal, with that of the noise centered at μ1 and signal+noise at μ2.

     The decision criterion is represented by a vertical line, treated as a “slider.”  Sliding the line to the left approaches the “don’t miss” criterion and to the right approaches the “be sure” criterion.  A central position is akin to the “do your best” criterion (see Exhibit 1).  The probability of each stimulus-response pair is defined by the area under one of the curves relative to the criterion line:
     P(Y|N) = area under the noise curve to the right of the criterion line.
     P(Y|S+N) = area under the signal+noise curve to the right of the criterion line.
     P(N|N) = area under the noise curve to the left of the criterion line.
     P(N|S+N) = area under the signal+noise curve to the left of the criterion line.
These probabilities are represented by the shaded areas in Exhibit 5; the colors used correspond with those used in Exhibit 2, visually linking the two presentation formats.

     The representation of the discriminability index, d’ (sometimes called the sensitivity index) in Exhibit 4 could be misleading.  The distance between the means of the noise and signal+noise distributions is equivalent to the S/N, while d’ accounts for the variance of each:
     d’ = (μS+N - μN)/σ ,
where, for clarity, μS+N and μN have replaced μ2 and μ1 as the means of the signal+noise and noise distributions, respectively, and σ is the standard deviation of the distributions.
     When d’ = 0, the probability of a hit and that of a false alarm are equal:  P(Y|S+N) = P(Y|N).  As discriminability of signals improves (d’ increases), the hit rate increases relative to the false alarm rate.  This performance “shift” can be seen in the Receiver Operating Characteristic (ROC) curves in Exhibit 6.

     While the ROC curves of Exhibit 6 aid in comprehension, interpolation of intermediate values can be difficult.  An alternative method is provided by d’ tables, such as that compiled by P.B. Elliott (see references).  Modern technology provides a faster, more-precise determination of d’ with attractive visuals – and it’s freely available.  Sample output of an online tool can be seen in Exhibit 7; the upper panel presents an equal-variance example.  A crosshair identifies the point on the ROC curve corresponding to the specified criterion level.
     The tool can also be used to determine d’ when the variance of the distributions differ, as shown in the lower panel of Exhibit 7.  To do this manually, the denominator of the equal-distribution d’ calculation (σ) is replaced by the square root of the average variance (√[(σS+N^2 + σN^2)/2]).  The difference in variance is reflected in the shape of the associated ROC curve.

     System performance requirements can be defined using any of the metrics discussed – discriminability index (d’), articulation index (AI), signal-to-noise ratio (S/N), Speech Interference Level (SIL), etc.  Whenever feasible, use of multiple indices should be considered.  Favorable results on multiple assessments can increase confidence in system performance in a range of use cases that may be encountered or anticipated.

System Components
     When discussing systems, focusing on tangible items is a common trap.  However, a communication system is much more than a collection of sound system hardware; in fact, some communication systems utilize no such hardware whatsoever.  The following list is offered to expand one’s thinking about the constituent content of communication systems, though it should not be treated as comprehensive or limiting.

• Microphones
• Loudspeakers
• Headsets
• Amplifiers
• Filters
• Radios
• Speakers (“talkers”)
• Listeners
• Vocabulary (verbal and nonverbal)
• Noise sources
• Signal characteristics
• Context of message
• Decision criterion
• Facility characteristics
• Room characteristics
• Hand & body motions

     Several components listed could be expanded into multiple entries reflecting specific characteristics.  Relevant hardware specifications, for example, could be listed separately.  Vocabulary could be split into entries defining the language used or specific words and phrases used in speech communication.  Nonverbal communication also uses a vocabulary of specified signals, each with well-defined meanings.  Signal characteristics could be replaced by frequency, intensity, duration, and so on.

Recommendations
     What follows is a series of recommendations for communication system design.  Little explanation is offered here; the background information needed can be found elsewhere in this series and linked references.  No set of recommendations can address all possible scenarios; the objective here is to provide a reasonable starting point for development of systems to be customized for their intended applications.

• Employ auditory signals when the message is simple, short, and “disposable” or a warning; when the listener is visually overloaded or must remain mobile.
• Use tones instead of speech when sufficient information can be conveyed by such a signal.
• For warning signals, use frequencies of 150 – 1000 Hz.
• For other (e.g. “routine”) signals, use frequencies of 1000 – 4000 Hz with prominent harmonics; avoid high frequencies.
• Include prominent frequency components <1500 Hz when hearing loss or HPD use is a factor.
• When localization ability is critical, include prominent frequencies >3000 Hz for intensity-difference detection and <1500 Hz for phase-difference detection.
• For listeners at long distances (>1000 ft), use frequencies <1000 Hz at high intensities.
• Use frequencies <500 Hz when line of sight between speaker and listener is obstructed.
• Frequency-discriminated signals should use a maximum of five frequencies; four or fewer is preferred.
• Match the type of alarm to signal requirements.  See Exhibit 8 for a comparison of alarm types.
  

• Consider use of communication aids or modifications to the system in ambient noise >50 dB.
• Set signal level halfway between masked threshold and 110 dB.
• Target S/Ns >15 dB for reliable detection and <30 dB to minimize annoyance and startle.
• At levels above 110 dB, substitute visual for auditory signals, or add redundant visual signals.  When designing visual signals, be sure to account for visual-field narrowing and changes in color perception that high-intensity noise can induce.
• Intensity-discriminated signals should use a maximum of four levels.

• Use signals of duration >300 ms; if shorter duration is necessary, increased intensity is required to maintain detectability.  For equal detectability, the product of duration and intensity is constant.
• Limit signal duration to a few seconds, unless explicit acknowledgement is required, for which a manual reset is required.
• Duration-discriminated signals should use only two durations.

• Modulated signals are preferred to steady-state signals; temporal variation of 1 – 3 times/s or 1 – 8 beeps/s is recommended.  Frequency shifting can also be used.
• Complex tones are preferred to pure tones.
• Use multichannel presentation to increase detectability/intelligibility (e.g. tones, speech, visual, tactile).
• Reduce S/N in lower-intensity noise to minimize startle and annoyance, particularly when performing high-concentration tasks.
• Limit duration of high-concentration and critical vigilance tasks to <30 min with rest periods or alternate assignments between.
• Limit the complexity of tasks or number of input channels to monitor.
• Match signal presentation to its urgency and context (intensity, repeat rate, tone vs. speech, etc.)
• Signals differentiated by combinations of frequency and intensity should be limited to eight combinations; fewer is better.

• In-person (“face-to-face”) communication is preferred to reproduced speech.
• Listener should face speaker when practicable.
• Begin messages with the context to increase clarity.
• Limit vocabulary as much as possible.
• Use common words and phrases.
• Use multisyllabic words.
• Provide feedback to speaker; e.g. repeat the message received as confirmation.
• Use reply tones, in lieu of speech, in high-intensity noise.
• Limit speech rate to 30 phonemes/s; 15 – 20 phonemes/s is preferable (phonemes are basic speech sounds, or “building blocks” of speech).
• Use digits to “spell out” numbers.
• Use a phonetic alphabet (“word spelling”) to increase intelligibility.  See Exhibit 9 for the standard phonetic alphabet used in military, aviation, and other sensitive applications.
  

• A noise-cancelling microphone at the lips is preferred to a contact (“throat”) microphone when both speaker and listener are in noise.
• Headsets are preferred to loudspeakers when high ambient noise levels require HPD use, different messages must be delivered to different listeners, or reverberation limits intelligibility of loudspeakers.
• Use distributed loudspeakers of lower power to limit reverberation.
• Operate earphones out of phase, or delay transmission to one earphone by ~500 μs.
• When the speaker is in quiet and the listener in noise, peak-clipping provides higher intelligibility.
• Digital systems require a sampling rate greater than twice the highest frequency in the signal to maintain intelligibility.
• Use automatic gain control (AGC) when signals’ dynamic range falls below 20 – 30 dB.  Dynamic range describes the difference between the strongest clear signal and weakest discernible signal.  Exhibit 10 shows how masking effects the dynamic range of a signal.
  

• Use sound insulation and sound-absorbing materials to reduce reverberation and minimize the interference of noise on communication.

     The recommendations provided above are generalizations.  Every environment is unique and may require deviation from a “standard” setup or a compromise solution to address conflicting recommendations (e.g. minimum S/N vs. maximum SPL).  If this list allows system design to begin or inspires relevant questions that lead to improved system performance, it has served its purpose.
     Recommendations for equipment, in particular, have been purposefully limited.  The range of options available is too broad to generalize coherently.  Review of equipment specifications and predictions of the performance of various combinations of components must take place during system development for a specific application environment to have merit.
     Upcoming installments of the “Occupational Soundscapes” series will provide additional information that can be used to improve communication system performance.  Noise-control techniques and use of hearing protection are relevant to both major themes of this series – hearing conservation and communication.  To these topics, the series now turns.

     For additional guidance or assistance with Safety, Health, and Environmental (SHE) issues, or other Operations challenges, feel free to leave a comment, contact JayWink Solutions, or schedule an appointment.

     For a directory of “Occupational Soundscapes” volumes on “The Third Degree,” see Part 1: An Introduction to Noise-Induced Hearing Loss (26Jul2023).

References
[Link] The Noise Manual, 6ed.  D.K. Meinke, E.H. Berger, R.L. Neitzel, D.P. Driscoll, and K. Bright, eds.  The American Industrial Hygiene Association (AIHA); 2022.
[Link] The Effects of Noise on Man.  Karl D. Kryter.  Academic Press; 1970.
[Link] Human Engineering Guide to Equipment Design (Revised Edition).  Harold P. Van Cott and Robert G. Kinkade (Eds).  American Institutes for Research; 1972.
[Link] Kodak's Ergonomic Design for People at Work.  The Eastman Kodak Company (ed).  John Wiley & Sons, Inc.; 2004.
[Link] Fundamentals of Industrial Ergonomics, 2ed.  B. Mustafa Pulat.  Waveland Press; 1997.
[Link] Engineering Noise Control – Theory and Practice, 4ed.  David A. Bies and Colin H. Hansen.  Taylor & Francis; 2009.
[Link] “Protection and Enhancement of Hearing in Noise.”  John G. Casali and Samir N. Y. Gerges.  Reviews of Human Factors and Ergonomics; April 2006.
[Link] “A d' Primer.”  Eshed Margalit.
[Link] “Signal-to-noise ratio.”  Wikipedia.
[Link] “Extra-Auditory Effects of Noise as a Health Hazard.”  Joseph R. Anticaglia and Alexander Cohen.  American Industrial Hygiene Association Journal; May-June 1970.
[Link] “Tables of d'.”  P.B. Elliott.  The University of Michigan Research Institute; October 1959.
[Link]  “Signal Detection Theory.”  David Heeger.  New York University; 1997.

Jody W. Phelps, MSc, PMP®, MBA
Principal Consultant
JayWink Solutions, LLC
[email protected]