Originally Posted by Roger Dressler
The hardware and method is not proprietary. Checked out the documentation for the systems used for these measurements:
• SMAART, Multi-Channel Sound System Measurement, Optimization and Control Software, Rational Acoustics
• SIM-3, Audio Analyzer System, Meyer Sound Laboratories, Inc
• AFMG Systune, Live Sound Measurements in Real Time! Software, Ahnert Feistel Media Group
• D2, Acoustical Measurement System, AcoustX
These systems are proprietary in the sense that the source code is not available and their computations are not documented in enough detail to either: (1) understand what these programs are actually doing; or (2) reliably reproduce their output.
I found the manual for the first system on the list: SMAART. I want to make a few notes: First, the manual actually distinguishes between two different smoothing methods it offers.
The first is called "fractional octave (logarithmic) smoothing". The word "logarithmic" could be providing a clue as to what information is being smoothed, but they may simply be remarking on the fact that the resolution is constant when viewed on a *logarithmic* frequency scale, as is usually the case when looking at spectral data. They do not clarify this elsewhere in the manual, nor do they indicate what type of smoothing kernel is used for this.
The second is called "linear complex smoothing (LCS)", which is less ambiguous to me. This method is mathematically the same as "frequency dependent windowing (FDW)" except that the FDW window is the inverse Fourier Transform of the LCS smoothing kernel. Unfortunately, they don't specify which window or kernel is used for this operation either. (A good "educated guess" is Gaussian.)
My other note about SMAART is that it provides a feature called "multi-time-window" in the context of "transfer function measurement", which appears to be what SMPTE RP 2096-1 is referring to. This feature is described as follows:
MTW stands for multi-time-window. This is the default FFT selection for transfer function measurements and for a large majority of system tuning applications, there may rarely be any real need to change it. Rather than taking a single FFT for each input signal (reference and measurement) at a single sample rate, MTW uses a series of sample rate decimations and varying FFT sizes to produce a measurement with different time and frequency resolutions in different frequency ranges. There are a several benefits associated with this approach.
One benefit of MTW is that it sidesteps some of the time vs. frequency resolution trade-offs inherent in FFT analysis. Having only about 840 frequency data points makes MTW measurements much easier to
read than a single-size FFT measure with comparable low-frequency resolution. Another is that the use of shorter time windows at higher frequencies makes the coherence function a much more useful tool
for detecting timing mismatches between the reference and measurement than any single-size FFT based measurement with comparable low-frequency resolution. At 48k sampling rate, MTW produces better than 1 Hz resolution at the lowest frequencies, compared to ~1.5 Hz for a 32K FFT, with higher computational efficiency, and with nearly 20 times fewer frequency data points than a 32K FFT, it is much less work for your graphics hardware to plot.
The software really shows its age here. Even a cheap laptop these days can crunch through large FFTs very quickly. This hasn't always been the case though.
As a side-effect of this method, there is a kind of FDW applied at the time of the measurement, except it's not necessarily smoothly varying. In fact, it's not exactly clear how the data from the FFT at each scale is combined. Another thing is that late arriving high frequency energy will be truncated by the measurement method, which may or not be important depending on what type of analysis one wishes to perform after the fact.
I'm not going to go through the trouble to dig into the other measurement systems listed above because I'm fairly certain that the situation with them is similar. How many users of these programs pay attention to such details anyway? I would also point out that it's pretty unlikely that anyone implementing a "multi time window" feature in their measurement software would do so in the same way as SMAART did. How would they know what to do? It's not documented. Hence, this data can't be reliably compared.
Originally Posted by Roger Dressler
SMPTE published this report in 2014: TC-25CSS B-Chain Frequency and Temporal Response Analysis of Theatres and Dubbing Stages
, which is listed here
. It says one can register for free to download it. It states:
This suggests the measurement method would not affect the curve.
I have that report and have read it practically cover to cover. It has a lot of interesting information, and also lacks a lot of critical information for me to answer questions I have about the behavior of cinemas vs. small rooms. Unfortunately, according to the report their raw data suffered from a high noise floor because of the type of signal used to measure the transfer function. This would make it hard to assess the contribution of late arriving energy.
That specific paragraph refers to their bench testing group. Is there a separate report for their work? Two critical questions I would ask are: (1) Which specific measurement systems and methods were actually tested? and (2) What exactly is meant by "essentially identical"? Does this assume a certain "margin of error" to be negligible? If so, what?
In my opinion, what their statement suggests is not correct. I do expect many different measurement methods to give results that appear similar, but often significant differences will arise at the broader scales, which are the aspect of response that are usually being optimized to the "target", whatever it is.
To give the example of my room. Even up at 8 kHz I see about 1 dB difference between RTA pink noise and 1/48th octave FDW (which is as close as I can get to whatever SMAART is doing). In terms of total impulse response energy, about 25% of it around 8 kHz is "late arriving", being outside the 1/48th octave window. This is despite the following: (a) I use horn speakers with ~90 x 25ish degree pattern at 8 kHz; (b) I sit relatively close compared to the room size (both of these contribute to higher direct-to-reflected sound ratio); (c) the room has a fair amount of fabric; and broadband absorption; and (d) the air itself absorbs sound fairly rapidly at that frequency (-0.5 dB-ish per meter, depending a lot on temperature and humidity).
From a theoretical standpoint, I expect cinemas will exhibit substantially larger differences than I observe in my "close to near-field" situation. Discrepancies could be 3-6 dB or more in the mid frequencies for cinemas. I don't know for sure without being able to see the right data. And to emphasize again, some of these discrepancies may be very sensitive to room acoustic conditions. I'm reluctant to generalize until I see the data.