T&M: To Ears With Love

Despite the focus on data applications in cell phones, there is no doubt that
audio is still the killer application. The audio quality of a mobile phone is
one of its most important features. Three elements control the audio quality:
The choice of the Voice Coder (Vocoder), the mechanical design of the phone, and
the choice of the audio transducers, the microphone and the speaker.

The choice of Vocoder is made during the standards process, and is not
something that can be changed in a particular design. The mechanical design
includes audio isolation, coupling to the ear, and noise cancellation. The
choice of speaker and microphone are important elements, also made as part of
the development process. Qualification of a design is lengthy and difficult. The
hard part is simulation of the human voice and ear, and how the human physiology
must be taken into account when performing audio testing. Equipment for this is
commercially available, but is expensive. The acoustic qualification of a phone
is very time consuming.

Manufacturing Considerations

Once a phone enters production, all of these elements have been qualified
and are fixed. But are they? Will the test process catch a defective speaker or
microphone? Agilent Technologies has introduced audio test capability to the
Agilent 8960 wireless communications test set for use with either GSM or
CDMA2000 testing.

The
idea is to excite the microphone with noise, and excite the speaker with
electrical sounds that may be analyzed. Rather than duplicate the expensive,
difficult testing used for qualification, a simple fixture may be used. The
fixture is calibrated by using a known good copy of the design. The response is
learned for this MS, and the testing of each device is compared to this.

Audio Source

One choice of acoustic test signal is an artificial voice called P-50, which
is the name of the ITU standard that describes the generation process. The
statistics of this matches real speech quite well. Areas that match include
amplitude variations, frequency response, as well as probabilities over time of
signals matching the amplitude and frequency of real speech. Matching these
probabilities, though, takes a long time. That is not compatible with the
manufacturing process.

A
better alternative is to use a collection of discrete tones, which can later be
analyzed using Fourier techniques. These tones cover the range of 300 to 3000
Hz. This signal is not ideal for modern Vocoders, but will detect defective
transducers because of the comparison to a known good device. In addition, the
signal used to test the speaker can be improved in the test set by turning off
one element of the vocoder, the noise cancellation. This allows much better
acoustic generation of sine waves.

The speaker is tested as shown in figure 1 and using the following steps: The
acoustic file is generated in the test set, and sent to an internal vocoder, one
that has noise cancellation turned off. The resulting audio is transmitted to
the MS in a real radio link. A test fixture must have good acoustic coupling to
the speaker of the MS. Typically this is a ring of foam rubber that compresses
against the body of the MS, with a microphone suspended above the speaker. The
volume of the speaker is actually quite low; an air seal is necessary to hear
its output. The signal from the microphone is connected to the audio input of
the test set, where the frequency response is computed.

The test of the microphone does the same steps, but in the opposite
direction, as shown in figure 2. The audio is generated in the test set, and
sent out the audio out port. This drives a speaker that is fixed near the
microphone. The output of the vocoder in the MS is then transmitted to the test
set via the RF link, where the audio decode is performed, and the frequency
response computed.

By comparing both the shape and the amplitude of the analyzed signals against
the reference test, it can be assured that the components in the MS being tested
are a faithful copy of the qualified design. Shown in figure 3 is the result of
a speaker test on a GSM phone.

Conclusions

These tests are very fast, typically less than one second each. During the
manufacturing process phones are already tested in a real link with the test
set, so addition of these tests does not require any additional setup time.
These tests do require a specialized fixture, but this is not a difficult
design. The use of audio testing as allowed in the 8960 wireless communications
test set will ensure better quality in production phones.

Dave Whipple,
production engineer manager, Agilent Technologies