Analysis of transient signals requires extra thought and care. The main challenges are:
- Avoiding overloads
- Editing files correctly to prevent artifacts
- Understanding Filter settling times
- Properly scaling to make comparisons possible.
- 3D Measurements: Spectra as a Function of Time.
Fortunately, noiseLAB Capture and its companion Batch Processor provide significant capabilities for these applications.
Avoiding Overloads
- Make a trial recording and observe whether noiseLAB Flashes red when overloads occur. Adjust you input gains, if possible, or choose a less sensitive transducer.
- If you have Recordings with overload,
use the Preview graph of the A-Weighted Sound Level (Fast) on the left graph with Red markers indicating overload. The Level Waveform graph on the right shows the oscilloscope waveform of one second duration after the cursor (Blue) on the Preview graph. Waveform portions on the right graph greater than +-1 are overloaded.
- you should Edit the Recording to not include these.
Editing to exclude Overloads
The above clip is tricky, because we have to avoid the OVERLOAD.
TRICK 1: You can single step the GREEN (Start) Cursor with SHIFT <Right or Left Arrow>
TRICK 2: You can single step the RED (End) Cursor with CTRL <Right or Left Arrow>
TRICK 3: Step back and forth between peaks using the UP or Down arrow, with the same SHIFT and CTRL tricks.
TRICK 4: (Shown above) Right click on the Clip name in the Clips listbox, to update the Clip Start and End from the Current Cursor positions. This effectively re-does the edit.
Analysis of Rapid Series of Transients:
BAD TRICK: Do not create short Clips. This can create significant settling issues and errors where performing spectrum analysis.
Settling time of Octave Filters
GOOD TRICK: noiseLAB always shows the settling time of 1/3 octave (also 1/N octave) filters. These are shown in the Result listbox in the Filter Settled column. There are two cases:
- Normal analysis of a Recording or Clip: noiseLAB automatically resets the RMS averaging when the filters have settled. Thus the final averaged result is correct, but is of shorter duration than the actual clip. noiseLAB shows the actual settling time at the beginning which is not included in the analysis. This must be subtracted from the Duration to give the actual analysis time.
- Time Slice Analysis (in the Batch Processor): One or more of the first Time Slices may not be settled. Note that after 1 second, the Results list box shows all the following slices are settled:
Results with filters that are not settled are incorrect and must not be used.
The filter settling time is determined by the Bandwidth of the lowest frequency filter in the Octave Analysis.
The settling time is approximately as follows:
T = 5 / Bandwidth (Formula 1)
The Bandwidth of a Third Octave Filter is 23% of the Center Frequency.
So if the lowest center frequency in your third-octave analysis is 20 Hz, then the filter bandwidth is 4.6 Hz, and the Settling Time is about 1 second.
Broader filters, such as 1/1 Octave filters settle more quickly, and more narrow filters such as 1/6 to 1/24 Hz settle proportionately more slowly. In all cases, noiseLAB reports the settling time.
To optimize settling times, noiseLAB Batch provides several low frequency cut-off choices. noiseLAB first reports Settled as True, when all filters have settled.
To intuitively better understand the response and settling characteristics of octave filters, we can compare them to a stringed piano, where the spacing between the keys is 1/12 octave. The longer the string, the deeper the tone, and the more sluggish the response, and also, with a longer settling time. Bass strings can “sing” a long time, whereas high tones are very short duration and very “stacatto”.
Scaling of Octave Measurements
To make a valid measurement of transient signals, one or more transients must be fully contained in a single Clip or slice, and all filters must have settled.
To make comparisons possble you must the normalize the dB value to 1 second duration.
For example if you want to find the average transient level of 10 transients in a 100 second recording, you measure the 1/3 octave level over the 101 seconds (to make the math easy), and subtract 1 second for the settling time.
- You then correct the duration to a normalized 1 second, by adding 20 dB.
- And to get the average level of the individual 10 transients you subtract 10 log N, where N is the number of transients.
The formula for calculating the average, normalized energy of N transients is:
dB (n=1, t=1) = 10 (log (T/N)) (Formula 2)
Where T is the analysis time in seconds
N is the Number of transients.
The same formula applies for normalizing and averaging the Level of Transients, and corresponds to computing LE or SEL as it previously was called.
For FFT Measurements the above formula, and in addition, the normalization to 1 Hz bandwidth must also be included, and the measurement must use a Hanning weighting and an overlap of 66.6% or 75% to ensure a ripple free time domain weighting.
Sound Level Measurement
noiseLAB Analysis results show the Sound Level in 125 ms intervals. This applies for Fast, Slow, or Custom Exponential time constants. However, internally, noiseLAB samples the RMS detector 500 times per second, to ensure it captures the correct Maximum or Minimum values in each 125 ms interval.
For a rapid series of transients, this can give a plot which may not show the individual transients we know exist. Even though multiple transients may occur in a single125 ms window, only the highest value will be shown, even with a very short time constant of 2 ms.
However, on the Advanced Tab, you can select High Res Time Base :
to reveal the individual peaks:
To get the average energy of the peaks, use the Formula 2 above.
To export the Graph use the
Save (Graph icon) button.
- If Slice is Off, only the whole graph shown above will be exported.
- If Slice is On, graphs of all slices will be exported.
To save the overall values from the Table below the graph, use the Save (Table icon) button:
To save a table of the numeric values of the Graph first select the Report Tab and select Level vs. Time and the use the Save (table) button.
If Slice is Off, the time series of the Entire Clip will be exported as the selected Result.
If Slice is On, all Slices will be Exported, one slice per column:
Notes on “Excel” Tables:
- Data are exported as Tab delimited files.
- The decimal of the file follows the system settings of the computer.
- The extension of the file is .xls to make it easier to find with Excel.
- Because of the Tab delimited format of the file, Excel will come with a caution dialog box when opening it, but will still open it properly.
- From the tab-delimited data you can post-process for functions such as peak picking using Excel, MatLab or LabVIEW. Tab-delimited files are the most user-friendly for these tools. For MatLab you should normally export files only using the period . as the decimal symbol.
Psychoacoustic tools for Impulse Analysis
The Nordic method for Impulse Analysis (NT ACOU 112) is selected by using the Impulse function of the noiseLAB Batch Processor. This Nordic standard is currently on its way to becoming an International Standard.
3D Analysis of Spectra as a Function of Time
Using the Slice feature of noiseLAB Batch Processor you can view individual lines of frequency spectra as a function of time. The finest time resolution is 125 ms.
For the above signals, this is not sufficient resolution to resolve individual transients response, but the varying level of the 5910 Hz frequency selected by the Cursor on the upper graph, is still shown on the lower graph.
and likewise for the third octave spectrum.
A final example with two slowly decaying transients whose sound level plot is shown:
Now with FFT Analysis (3 Hz resolution) the first rapid decay of the 1029 Hz resonance:
and the slow dual slope decay at 1188 Hz:
And below the same transient analyzed with third octave analysis:
The initial impulse above is dominated by 1000 Hz which rapidly decays but is not as clearly differentiated as with the higher resolution of the FFT earlier shown.
The slow decay after 0.25 s is dominated by 1250 Hz as seen below where the leakage of the 1000 Hz component is not as strong.
You can freely slice in Time and Frequency by moving the respective cursors on the two plots.
These graphs as well as the corresponding tables can be exported as described above.
For more information contact:
Carsten Thomsen
ct@pxi.dk