Hello, I am writing some manufacture tests for digital microphones (an array of) attached to a PCB. I want to verify that they are 1) producing valid audio without interference and 2) within +-X db of one another. The PDM output of each microphone is connected to a micro-controller capable to decimating the PDM to PCM.

I know little about the test environment, i.e. they will not be tested in a EM shielded anechoic environment, more likely a factory floor, i.e. there will be a fair bit of noise. The quicker the test the better, I am aiming for 10-30 seconds per array(which is 8 PDM microphones).

My questions are: 

• are there any specific failure modes I should be looking out for? 
• how/can I use the unknown environment to verify the relative gain of each microphone in the array?

Thanks for any ideas.

Back in the day, I was involved with the factory testing of telephones and other equipment that included microphones and "speakers" somehow plugged into the PSTN network.

The various manufacturing facilities had many and varied arrangement for the test environment and the technical methods to be used.

Environments - ranged from :-

• a sound-insulated room
• a sound-insulated "box"
• Close-coupled Mouth speaker and Ear Microphone with flexible housing to isolate from noise
• Completely open acoustic environment

Acoustic Test Signals were typically :-

• A sequence of tones across the frequency band of interest
• Noise

Daily Calibration was typically :-

• A sound level calibrator provides a known pressure at a know frequency
• A high quality flat-response microphone whose o/p voltage is calibrated using the sound level calibrator
• Sound sources that are then calibrated using the reference microphone
• A "golden" unit under test which has a full set of high-quality test results which is used both as an overall sanity test that the test rig is working and to check test results remain within set limits.

Sound sources :-

• B+K Head - expensive
• BT copy of B+K head - still expensive and in-house only
• Decent quality miniature speaker with housing - reasonable cost, but required careful calibration

Note - in the telephone test world there was no concept of directional microphone sensitivity, it was simply an artificial mouth speaker placed at one position.

Technical Test aspects :-

• Tone sequence testing would include tracking filters to measure only the signal being generated
• Noise tests would use FFTs to generate an entire spectral response after the prescribed length of the random noise signal.

Economics/complexity of a solution should be appropriate to :-

• Cost of the product
• Quality culture of the product
• Is the detailed product performance critical ?
• Volume of production

Putting it all together

  At one end I have seem high-value, safety-critical products have the luxury of an acoustic room and comprehensive testing and capture of detailed performance. At the other end, in another case perhaps just a few tones were tested.

   In summary, It is quite possible to make good-quality acoustic tests in a noisy factory environment. You may set various degrees of coverage for say a pre-production run compared to main production. You might also test every nth unit to a greater degree than standard.

  The use of a calibrated "golden" unit run through the test solution can provide great confidence in the test rig and the products that run through it.

  As a final comment, the ability to provide a higher test coverage mode with logged results can be of great help to the product design team and is also very impressive to show the customer. Both the Engineers and the Sales guys will be impressed.

Thanks, I like the idea of the optional increased test coverage.

This particular device is able to produce it's own sound(with an attached speaker). Maybe using that I could emit and detect test tones.

Alternatively, a setup can be constructed such that a constant sequence of tones are played by a calibrated external device. The DUT would then detected the presence/absence of these tones and there magnitudes. Calibration cost is negligible compared to the time required to run the whole batch of tests.

Hi, 

I have been also interested in microphone array, but I am not much familiar. This is just quick my opinion, so note it lightly.

1) As I know, the major purpose of microphone arrays is beamforming. So I recommand you to test and check the bandwidth (or cutoff frequency) of the microphone array when applied to a simple beamforming technique. Your mic array may not support some higher frequency audio.

2) I am not sure that checking the relative gain of each mic in the array is meaningful. This is because, as I mention above, mic array is mostly used for beamingforming. I think checking beamforming gain is more important. 

And I have a question. Can you tell me why you record 10-30 sec for the test. Is there any reason ? 

I want share the information on your mic array and it test. 

Thank you and cheer up! :-)

Jaewook

The purpose of a 10-30 seconds test limit is to facilitate testing a lot of PCBs in a short amount of time, as time costs money. Ideally, the microphone testing would run in zero time so the bulk of the test procedure is the physical connecting/power checks/etc.