50
50
Guenter Villnow
50
50
Guenter Villnow
50
50
50
50

This app note is part of a series on car audio sound. The top level app note and list of all parts is:

Contents

Microphone techniques and their caveats [Top]

First, there is no right nor wrong here. Every setup has its pros and cons, which I will explain from a hands-on perspective. During calibration and decision making it is important to know the potential and possible limitations of your workflow.

Every measurement is an estimation and is therefore never the reality, just a representative slice of it. How much time you want to spend on measurements, your focus, desire for accuracy and precision, and whether you want to or can sit in the car during calibration affects your choice.

Fortunately, miniDSP offers a lot of choices and options to find the right fit for you!

Variant A: Single mic fixed in place (with sine sweeps for sensitivity, polarity, timing/delays. E.g., with UMIK-1 or UMIK-2) [Top]

This is the simplest measurement technique available: the microphone is put in one place and stays there for the duration of the measurement. The UMIK-1 and UMIK-2 are ideal for this.

In the tuning process, data from this single point can be used for sensitivity checks of the speakers, polarity checks of the playback chain and log sweep analysis.

Note that log sweeps analyzed with DFT (discrete Fourier transform) from fixed single point mics are correct in its levels, but the acoustic resolution in the frequency domain is limited! Especially in the near field as in car cabins. This can be better understood in a filter reconstruction measurement:

This filter reconstruction measurement to verify an acoustic measurement chain shows a noisy result which is visible as ripple in the frequency domain. The measurement works like this: Baseline measurement – measurement with defined change in an EQ band = reconstruction of applied filter (blue). The applied filter electric signal is additionally displayed (green) in the graph. The further the reconstructed filter is "away" from the applied filter, the less resolution your measurement chain has.

Due to the 20 dBr errors from only a single data point, graphs obtained this way can be misleading and lack reproducibility. Errors occur mostly in the mid and high frequencies and then lead to mediocre DSP settings in your car. The missing acoustic frequency resolution also applies to measurements that use noise as a stimulus, which are also influenced by the settings of the FFT.

The missing resolution in the frequency domain does not affect other conclusions that can be drawn from log sweeps such as group delay, waterfall, etc.

The resolution can be increased through spatial averaging of several consecutive fixed-in-place measurements at slightly varied positions around likely listening positions. But this is quite time consuming for even simple automotive sound systems if a moderate number of speakers is involved and several tuning iterations are needed. Also, matching the measurement positions in space for each repeated measurement can be challenging.

What speaks for the "single mic fixed in place" method, though, is the fact that you do not have to be in the sound field yourself. Think here of very loud sound systems and long sessions.

For improved resolution in the frequency domain and simplified workflow the moving mic method (below) is a good choice.

Verdict

Single mic fixed in place with noise

Speed 4/5
Reproducibility 2/5
Accuracy: frequency domain 3/5
Accuracy: SPL 2/5
Tuning Engineer in car necessary No

Single mic fixed in place with log sweeps

Speed 4/5
Reproducibility 2/5
Accuracy: frequency domain 2/5
Accuracy: SPL 4/5
Tuning Engineer in car necessary No

Variant B: Moving mic (with noise for RTA frequency response) [Top]

For this measurement technique (sometimes also called "flaming torch") the mic is moved slowly around the listening space for 10 to 30 seconds. The test stimulus is a noise signal (pink noise or Meyer Sound's m-noise). An FFT real-time analyzer (RTA) is used for analysis. REW's RTA is very capable of performing this task either with signals directly fed into the application via an audio interface or with recorded files drag-dropped into the RTA window.

In the frequency domain, the result of the moving mic method is more accurate and also more precise when repeated than the fixed in place single measurement technique (several measurements of the same acoustic event will give you the same results). A negative aspect of this method is that the measurement requires you to be in the car, which can be problematic with very loud and long measurement sessions. Also, you could alter the measurement through your breathing, wind noise (created by moving the mic) and cable noise. These must be kept in mind as potential problems.

If you are new to FFT RTAs, use these settings in REW as a starting point with pink noise as stimulus:

With the "RTA" mode enabled, REW makes a level-correct measurement and compensates for FFT gain and the 3 dB/octave loss from the pink noise.

So, you can aim for your defined target curve in your tuning. It takes approximately 11 seconds to finish the measurement with these settings. Within this time, move the mic around your head where your listening positions are.

Note that the moving mic method cannot be used with sweeps!

This video shows a great real-world application of this process: https://youtu.be/6RiuwqzjqlQ.

Verdict

Moving mic with noise

Speed 4/5
Reproducibility 4/5
Accuracy: frequency domain 4.5/5
Accuracy: SPL 1/5 (with FFT)
4/5 (with RTA)
Tuning Engineer in car necessary Yes

Variant C: Array setup (for noise and sweep stimuli. E.g., with UMIK-X) [Top]

The gold standard. Array measurements in cars go back to the work of Earl Geddes who proposed the "localized sound power method" in the early 1980's when he worked for Ford (original AES paper here: https://www.aes.org/e-lib/browse.cfm?elib=11627). Geddes found out that six equally spaced microphones placed where the ears of potential listeners are (from small to very big persons, in the so-called "ear ellipsoid", see figure below) with their data points averaged, gives the best measurement result in terms of resolution, accuracy, reproducibility and coherence with personal perception. This measurement technique is often found at car manufacturers and car audio brand suppliers.

It is said that the six-microphone array gives a reliable resolution equivalent to 1/3rd of an octave.

The miniDSP UMIK-X is made for measurements with car cabins in mind. One group of 8 MEMS mics form an array for one seat, thus increasing the resolution over the original six-point setup proposed by Geddes:

If you want to further increase resolution, you can use up to 16 mics in the UMIK-X for one seat position. For more information on higher-order arrays in automotive sound, see: https://pub.dega-akustik.de/DAGA_2010/data/articles/000235.pdf.

Working with arrays gives greatest confidence and reproducibility in the tuning process. It can be used with log sweeps and noise. With a little practice, array results are in general acquired more quickly and are easier to read and to work with than single mic or binaural head results (especially for beginners and with REW as analysis tool).

Since arrays are taking the seat of the listener, you do not have to be in the sound field yourself. Note that when you are using an array, you must use only a single mic in the array for delay measurements! With arrays, a high acoustical resolution can be achieved.

The measurement results from arrays are also a great starting point for mono bass optimization with Multi-Sub Optimizer (MSO).

Verdict

Mic array with noise

Speed 5/5
Reproducibility 5/5
Accuracy: frequency domain 5/5
Accuracy: SPL 5/5 (in RTA mode)
Tuning Engineer in car necessary No

Mic array with log sweeps

Speed 5/5
Reproducibility 5/5
Accuracy: frequency domain 5/5
Accuracy: SPL 5/5
Tuning Engineer in car necessary No

Variant D: Binaural head setup (e.g., with miniDSP EARS or Head Acoustics HII.3 or KEMAR with log sine sweeps and noise) [Top]

Automotive "sound" consists not only of sounds that are created through the speakers, but also sound events that are determined by physical properties of car components or are unwanted in general. These are, for example, intake noise, engine vibrations or rattling interior trim pieces. These sound events are measured and influenced by NVH engineers (noise, vibration, harshness).

Traditionally, one important tool for these engineers are binaural heads. A binaural head simulates a human head, torso, pinna and ear canals. At the end of these ear canals microphones are placed. The idea is capturing the sound field as "a human would do" (at least from a physical standpoint).

From an acoustic resolution standpoint, the results from these heads are better than the single mic fixed in place approach, since there are two microphone signals with a (well-defined) head related transfer function (HRTF) through delay (distance between the mics) and acoustic shadowing (through the form of the torso) added. The HRTF and the limited resolution of the binaural head in the mid and high frequency make decision making on binaural head recordings hard for beginners in electro-acoustic applications, since a suitable correction function must be applied to the measurement.

For an accurate measurement in a car, several head positions must be averaged in a lengthy process, which slows the approach of having a "replacement ear" in the car.

Signals from binaural heads can be used for later reference and storing of sound impressions through impulse responses (sweep as stimulus), noise and reference track recording.

Verdict

Binaural head setup with noise

Speed 5/5
Reproducibility 3/5
Accuracy: frequency domain 2/5
Accuracy: SPL 2/5
Tuning Engineer in car necessary No

Binaural head setup with log sweeps

Speed 5/5
Reproducibility 3/5
Accuracy: frequency domain 2/5
Accuracy: SPL 2/5 (4/5 if cal files are used)
Tuning Engineer in car necessary No

Variant E: Yourself as a binaural head (for noise and sweep stimuli) [Top]

When your budget is too constrained for a professional looking fully-fledged binaural head or a mic array, you can use yourself as a binaural head.

As with torso-based binaural head recordings, the same principles apply here: the number of data points is relatively small and a HRTF is added. But, in contrast to torso binaural head recordings, you can move and point and reach different positions quicker and easier. I use the following pattern: looking straight ahead, left mirror, right mirror, sun visor and bottom of steering wheel as positional points to get a better understanding of the sound field and cover as much as possible of the ear ellipsoid.

This a super interesting method when you want to travel with light equipment or want to use as little equipment as possible and can be in the sound field during the measurement yourself.

When used with noise and you are moving your head during the measurement, this method very quickly gives a lot of data points.

It is also advantageous to use the binaural head method for later reference and storing of sound impressions through impulse responses (sweep as stimulus), noise and reference track recording.

As a product, you can use these:

Verdict

Yourself as binaural head setup with noise

Speed 5/5
Reproducibility 4/5
Accuracy: frequency domain 4/5 (in RTA mode)
Accuracy: SPL 3/5
Tuning Engineer in car necessary Yes

Yourself as binaural head setup with log sweeps

Speed 5/5
Reproducibility 4/5
Accuracy: frequency domain 4/5
Accuracy: SPL 3/5
Tuning Engineer in car necessary Yes

Variant F: Mixed setup (for noise and sweep stimuli) [Top]

There is nothing wrong with combining approaches. Depending on your special requirements or goals this can make sense.

In the photograph below, you can see my personal car measurement setup. I combine several techniques in my workflow. I am using a miniDSP UMA-16 as base array and helper in mid and high frequencies, supported by two Sonarworks Ref mics, miniDSP EARS and a UMIK-2. All these devices work together as one audio device (currently only possible under macOS).

The UMIK-2 is used for timing measurements and can easily be removed for moving mic measurements. The miniDSP EARS alone is used for noise and sound quality measurements (e.g. with MOSQITO as explained in this tutorial: https://www.minidsp.com/applications/acoustic-measurements/psychoacoustic-measurements-with-mosqito).

Another good working combination is shown in the next photograph. This is a Head Acoustics artificial recording head combined with a UMIK-X that provides four microphones on either side of the head. The signals of the artificial head and the UMIK-X are matched through calibration files, averaged and analyzed together in REW. If you want to 3D print a binaural head yourself, you can use this great project: https://www.thingiverse.com/thing:4691843.

Closing [Top]

That concludes this part! Jump back to the top:

Related products - Car Audio DSP platforms