Aural Cue Technicalities
The Level D requirement in particular indicates that the sound system must produce accurate sound cues. An obvious inference from this statement is that the frequency response of the audio system must be adequate to achieve this goal. The current FAA requirement stops short of defining the frequency range specifically, but the JAA JAR STD 1A amendment 3 standard defines the range 50Hz to 16kHz to be the significant spectrum. The currently accepted Level D definition identifies the use of 1/3rd octave spectral comparison between aircraft data and simulation data; this data is required to match within the 50Hz to 16kHz band, with a tolerance of +/-5dB for each 1/3rd octave band.
The upper frequency required for reproduction has a direct impact on a digital audio system. Due to the mathematics behind the theory of digital sampling, the sampling rate of the audio system must be a minimum of twice that of the highest frequency of interest. Therefore, for a 16kHz upper frequency the sampling rate of the audio system must be at least 32kHz.
All ASTi audio processing is performed at a 48kHz sample rate using 16-bit audio, resulting in a 24kHz upper frequency bound. ASTi employs the studio-quality audio sample rate of 48kHz since this offers better high frequency phase response.
ASTi decided that the use of 16-bit is entirely adequate for communications and aural cue purposes within the simulation world. Other manufacturers may claim that 24-bit audio provides better quality, but consider the following:
- 16-bit audio = 96dB dynamic range
- 24-bit audio = 146dB dynamic range
In practice, real measures of 16-bit audio show a 90dB dynamic range versus 120dB for 24-bit. Although this is significant, the effect is only noticeable at the lowest audio signal levels (since bit depth increases the resolution for signals at the lowest signal levels… valuable for classical music perhaps, but not really for the cockpit environment of a typical aircraft). Factor in that the typical dynamic range of a very good quality power amplifier is around 100dB—at best—and it is easy to realize that the supposed benefit of the increased bit depth is meaningless.
Recent programs for modern wide-body commercial aircraft have shown that typical maximum and minimum 1/3rd octave band sound pressure values of 86dB (63Hz band) down to around 27dB (16kHz band) are found. This indicates that the operating range we need to consider is perhaps only 60dB, and that the noise floor of the simulator is typically defined by the on-board air-conditioning, and might present a noise floor of 25dB, so at best we have to cover a range of 61dB. Therefore, the 90dB we have to work with is more than adequate.
Aural Cue All Things Being Equal…
The current generation of ASTi Telestra product suite has some extremely powerful capabilities, that until now have been difficult (and expensive) to implement - per channel full range equalization, and per channel time alignment. In order to understand why this has any great significance we need to review how a sound system is installed on a simulator.
The data used as the basis for development of the simulation sound model is captured using a single measurement class microphone positioned at the Sim Std reference position in the cockpit (typically centered between the crew seats, over the center console). The frequency response of measurement class microphones is typically very flat (+/-0.5dB) and covers an extended frequency range (often 5Hz to 20kHz or more). The certification standards define that the sound system covers at least the range 50Hz through 16kHz so that is not a problem with modern loudspeakers.
However, we have many other considerations when installing the sound system into the simulator. From the inside of the cockpit the appearance must faithfully replicate the aircraft, so the loudspeakers for the sound system must be 'hidden', but equally the speakers must be distributed around the device to ensure adequate spatial directionality. The result is that the speakers are often installed through the external shell of the device, but the direct path is blocked by the cockpit lining, or some other obstruction. As a result the resultant frequency response of the speaker is no longer the original somewhat flat response.
You may still be asking why this would matter? Can't the model simply compensate for this distortion? Well the truth is it could, but consider that the model was probably developed off-line in the ASTi sound lab (for ASTi supplied models) or in the device manufacturer's engineering facilities (for customer developed models), and almost certainly not in the finished simulator. Therefore, the frequency response of the development environment will be quite different from the final device with the model installed in the simulator. It would be possible to modify every sound in the model (and there could easily be hundreds), but that will take a long time, and makes porting the model between different simulators difficult (often we want to use the same model in a FFS and then again in an FTD, but the cockpit structure and acoustics will be quite different). A better solution is to equalize the sound system to approximate as closely to the original development environment.
The current Telestra architecture includes the use of ACENet-compatible power amplifiers which includes the ability to apply complex equalization curves on a per channel basis under software control. This capability, combined with the realtime spectral analysis functions now included with the system, allows the installation engineer to equalize the installed system to produce a near flat frequency response. By positioning a measurement class microphone at the Sim Std reference position, we can analyze a standard pink noise signal (as used in loudspeaker testing) and tune the response using an off-line tool. This set-up only needs to be carried out once. Future enhancements of this capability will be to fully automate this functionality, and allow the system to automatically calibrate the entire speaker installation.
A secondary but related issue is one of time alignment of the speakers installed in the simulator. The sound from each speaker takes a finite time to reach any specific location. If each speaker is set-up to play the same sound, then we will experience significant nodal behavior as we move around the sound-field. The net result of this is the sound becomes smeared and 'filtered' due the sound fields from each speaker arriving at slightly different times due to the different distances each speaker is located from the reference location. Although this effect is quite subtle, this is something that can be tuned using the internal processing capability of the selected ACENet-compatible power amplifiers.
One important note about this section is that it is true that the capabilities described above could be implemented in the model processing platform, however the Telestra architecture off-loads the additional load to the amplifier itself, avoiding the use of what would otherwise be useful modeling processing.