Forum Replies Created
18 June 2020 at 11:35 am #180
Even though your measurement system might be doing specific time windowing, the IR still has both loudspeaker phase and high delay which manifests as high phase slope when exporting as freq/mag/phase TXT. (I’m guessing this high delay is the time-of-flight across your chamber, plus soundcard buffer delays.)
In FIR Designer there is no error to give. The TXT file HAS phase and so the loading process is doing the best it can to interpolate the TXT file to the finer resolution FFT spacing. The high phase slope is the problem, and in this case is causing spurious distortion in the calculated IR. (You can see this on the FIR Designer Import tab if you expand the time axis and zoom the vertical axis.) If you truely want to ignore phase, then you can set all the phase values to 0 deg in the TXT file. (FWIW SMAART spectrum measurements come across to FIR Designer with 0 deg phase.)
In this situation it’s always best to avoid using log-frequency-spaced freq/mag/phase data – no matter how fine in resolution – for transfer between programs. Conversion to this format looses information which then causes problems when trying to reconstructed the impulse response – which is what FIR Designer is doing. It’s always best to transfer using an impulse response or time format, where information doesn’t get lost. (E.g. AFMG ETM files, or the time impulse as TXT or WAV.
>> And since Auto Mag should be trying to work on matching measured magnitude to target only, I do not see the relation to phase.
Auto Mag works with complex input and calculates the magnitude using either power averaging or complex averaging across frequency. You’re correct that power averaging should avoid issues of high phase slope but the additional distorting (added by the import interpolation process due to the high phase) is also causing problems in the Auto Mag.
I guess we could add an option to completely ignore phase on import, but it’s not something people normally want. When doing any filtering on a loudspeaker, most folks want to see both the phase of the filtering and the effect – both magnitude and phase – of the filter on the loudspeaker measurement.
EA Support18 June 2020 at 10:27 am #176
I’ve looked over the project and measurement files. The measurement is a 3 column freq/mag/phase TXT file with very high phase slope. In other words the impulse response of this measurement wasn’t near time=0 in the measurement system, before the TXT file was exported. (In the excerpt below, the phase starts at -76 degrees and ends up < -40,000 degrees.) The file import process in FIR Designer does it’s best to interpolate this TXT file up to the internal FFT size, but very high phase slope causes errors in this interpolation. And both the interpolation errors and high phase slope are causing the Auto Mag problems that you are seeing.
Two suggestions. Either remove any bulk time offset from the measurement before exporting the freq/mag/phase TXT file, or export the impulse response itself as a time signal – TXT or WAV. With time signals, FIR Designer can remove bulk delay without any response error.
Note that freq/mag/phase files are great for plotting, but inherently loose information*, when compared to the time waveform IR.
* unless they are a linear frequency spaced FFT data.
Frequency Magnitude Phase
20 74.9498 -76.0055
20.3 75.0916 -78.8764
20.6 75.2281 -81.1653
20.9 75.2503 -82.5894
21.2 75.3959 -82.2123
18500 88.7457 -40505.1289
18750 88.765 -41041.2812
19000 88.7698 -41577.7695
19250 88.7277 -42114.3984
19500 88.6458 -42650.1914
19750 88.6929 -43186.1406
20000 88.7875 -43758.5703
5 June 2020 at 6:03 pm #169
- This reply was modified 2 years, 3 months ago by EA Support.
All our products can export FIR’s as TXT files that load into Powersoft Armonia. FIR Creator EX, FIR Designer and FIR Designer M also have a Powersoft DTP dialog where you can export an XML file, for Armonia, that includes gain, delay, polarity, IIR and FIR coefficients.
EA Support5 June 2020 at 8:56 am #167
FIR Designer is available as a subscription license (with continual feature updates) and a perpetual license (with only bug fixes).
With the subscription license, a credit card is kept on file and charged annually. Shortly before the annual charge, you will receive an email notice and can cancel the subscription. Once the subscription is cancelled, the software reverts to demo mode and you can’t save projects, systems and can’t export filters.
With the perpetual license, you can use the product indefinitely.
All our products contact our license server periodically to verify that the license is still valid. This helps us reset your license in the event of a computer failure or loss.
EA Support3 June 2020 at 7:35 am #165
Your FIR filter length of 400 at 48 kHz is way too short for the amount of LF magnitude and phase filtering you are trying to achieve. Filtering at lower frequencies requires longer filters, as does aggressive filtering – high Q and high slope regions in the magnitude and phase of the filter. Generally speaking a 400 tap filter (at 48 kHz) will struggle below approximately 500 Hz, and that can be seen in the “FIR Filter Frequency Response” plot on the Export tab. Powersoft amps can take at least 2048 length filters now and the modules might be similar. I’d recommend switching to just IIR filters at the point at which the FIR can’t go any lower.
Maybe take a look at the Audio FIR Filtering paper
It has examples of differing filter lengths.
One other suggestion. On the Import tab, using “Normalise magnitude to max” will result in much of the filtering being boosts. Maybe turn this off and use the manual “Magnitude off (dB)” field to place the smoothed response roughly either side of 0 dB. In this way the filtering will be a mix of boosts and cuts. Although I’ve not seen it yet, it’s possible some processors will run into headroom problems in the FIR filtering. And gain can always be made up later in a separate gain stage in the processor.
EA Support29 May 2020 at 6:57 pm #162
What file format are you using? With some formats such as WAV and single column TXT, the import process looks for the peak then loads from 1500 samples before the peak. Details are on https://eclipseaudio.com/loudspeaker-measurements/
If you know where the peak is, you can indicate it with a number at the end of the filename e.g. “measurement 423.wav”. Or to align time=0 with the start of the file, set the number at the end of the filename to 0. For example “measurement 0.wav”.
EA Support25 May 2020 at 12:53 pm #157
We mostly answer direct product questions but here are few things to try.
Regarding the measurements at different distances. You could design the filtering with the measurements you have then verify/adjust the relative driver levels later with measurements at a common distance outside (where there are no walls). Or take some new, gated measurements outside. I’d also suggest talking some off-axis measurements to help choose a crossover point where the drivers beam-width’s are roughly similar. Floyd Toole has published on the topic of subjective preference for loudspeakers with well-controlled directivity that varies smoothly over frequency.