The complexity and high circuit density of cell phones
challenge the system designer wishing to establish a highquality,
audio record/replay path that meets provider specifications.
New models with additional multimedia features,
such as a still camera, ring-tone generators, MP3 replay,
and voice memo, usually require incremental product
changes. This necessitates not only new components, but
also PC-board layout changes that can give rise to nonideal
grounding and subsequent new noise problems.
Noise and interference issues in a cell phone analog-audio
signal path can usually be attributed to demodulation of
RF into the audio band or shared/nonideal grounding.
When subjected to high levels of RF energy from the
phone antenna, the comparatively low-bandwidth audio
circuitry in a phone can unintentionally demodulate the
complex RF-transmitted signal. This degrades the noise
floor of the audio path. While certain techniques and
topologies can be applied to minimize this degradation
within the audio amplifier circuitry, suppression components
placed adjacent to the input pins are a cost-effective cure.
Low-value capacitors-to-GND are often used, due to the designer selecting the minima of the capacitor's impedance to correspond with the carrier frequency of the radio.
One effective audio solution that minimizes shared/nonideal grounding integrates all the typically required
analog-audio I/O functions into a single IC. This design
transfers most of the problematic grounding issues from
the PC-board layout engineer to the IC manufacturer.
Besides including the necessary analog-audio I/O
functions, that same IC must provide digital-audio interfacing
sufficient to support voice band and any multimedia
(i.e., application processor) functions. The IC also
must provide partitioned shutdown control over the
various blocks to maximize battery life.
The following discourse highlights some of the analog-/digital-audio issues that arise in single-chip implementation.
The MAX9851 is used as an example of techniques
and features that simplify GSM/GPRS cell phone design.
Analog Audio—Minimizing Microphone Noise
High-gain audio circuits, such as microphone amplifiers
(mic amps), are subject to degradation from poor grounding.
This is particularly true for single-ended circuit topologies,
in which small voltage differences between the mic-amp
ground reference and the source ground reference (in this
case, the GND pin of the mic capsule) are amplified into the
signal path. In a complex product, like a cell phone, where
audio ground planes are often shared with other circuits,
degradation can be problematic, because the copper plane is
not "zero ohms" (as is frequently assumed). Consequently,
any current flowing through this finite resistance can cause
small potential differences across the plane.
The grounding problem can be addressed with a mic amp
that features a fully differential input. This approach,
which allows remote sensing of the mic's GND pin, is
incorporated into the MAX9851
. Remote sensing forces
any AC-voltage differences between the CODEC
reference and the mic GND to appear as a common-mode
signal to the mic amp. These differences are then reduced
by the common-mode rejection ratio (CMRR) of the amp,
thereby significantly attenuating the effective noise
contributed to the signal path. The only penalties with this
design are an extra PC-board trace to the mic from the
CODEC and an extra coupling capacitor.
The MAX9851 also allows stereo, external mic inputs to
be switched in to override the internal mic. These inputs
typically would be sourced from a car kit or other external
headset. In this case, using the amp input's CMRR, the
EXTMICGND pin acts as a 'Kelvin sense' for both L and
R channels, canceling ground differential noise in the
same way as described above. The EXTMICGND PC-board
trace should be extended to the GND connector of
the car kit jack or headset connector for best results
Figure 1. The use of differential input amplifiers allows remote sensing of the socket "GND" reference. Any AC voltages between local and socket grounds are largely rejected, and not amplified by the mic amp gain.
Mic bias circuits can introduce significant noise in the
signal path. A large percentage of any bias output-voltage
noise appears directly across the mic-amp input. More
proficiently designed mic amps, as in the MAX9851,
provide a regulated, low-noise bias voltage with output-noise
levels matched to the internal mic-amp input noise.
Analog Audio—Stereo DirectDrive™ Headphone and Receiver Outputs
The ability to replay compressed music files at near-CD
quality places high demands on headphone audio reproduction.
Signal-to-noise ratio (SNR), linearity, and
bandwidth must be improved over the basic 300Hz to
4kHz voice-path requirement. Low-frequency extension
can be problematic, as headphone drivers typically need a
series capacitor to prevent the headphone amp's DC bias
from appearing across the headphone transducer. The
typical impedance range of common stereo headphones
extends down to 16Ω, so that any series capacitor forms a
highpass filter to low-frequency content. For extension of
the listening response down to 100Hz, for example, two
100µF DC-blocking capacitors would be needed to
guarantee 16Ω stereo headphone operation.
Use of Maxim's DirectDrive technique allows headphone
operation without series capacitors, due to the amp
outputs being referenced to 0V. The low-frequency
content is then limited by either a DC-removal filter
(digital source), as designed into the MAX9851, or by the
input-coupling capacitors on the line or mic inputs
(analog source). A further advantage of the DirectDrive
design is the inherently low click/pop levels when
bringing the device into or out of shutdown. As there are
no series capacitors to charge or discharge, no net turn
on/off current flows through the headphones.
The stereo headphones outputs of the MAX9851 are also
capable of bridged-mono operation, (Figure 2
enables compatibility for different headsets and accessories.
The same socket could accommodate stereo headphones
or a mono (mic plus hook switch and speaker) headset. The outputs remain ground referenced in this mode, so no DC voltage appears on the headset cable. Therefore,
fault or short-circuit conditions are less problematic.
Figure 2. The DirectDrive headphone outputs are capable of bridged mono and stereo operation. The Maxim proprietary GND-referenced output means no series coupling capacitors are necessary, saving cost and PC-board area.
The receiver speaker output also uses the on-board charge
pump in the DirectDrive design, so the output is single
ended and the negative speaker connection is GND (0V).
The output still has nearly the same voltage swing of a
more typical BTL (differential) output, because the
inverting charge pump provides a negative rail almost equal
to that of the applied AVDD
. The resulting peak-to-peak
output across the receiver speaker is almost 2 x AVDD
Analog Audio—Class D Speaker Amplifiers
The MAX9851 incorporates Maxim's third-generation
class D technology to drive 8Ω (or 4Ω) speakers. The
main advantage of class D (switching) amps over class AB
(linear) amps is efficiency. Class AB amps dissipate
significant power in the output devices unless the amp is
driven into clipping. However, because class D topologies
have their output devices either on or off, their thermal
dissipation is less and battery life can be extended.
Extended battery life can be significant if a cell phone is
used in speakerphone mode frequently, or supports push-to-
talk (PTT) operation.
There are caveats to using class D topologies, however,
especially in a product whose core function is RF transmission/reception, such as a cell phone. The fast-switching
edges associated with efficient class D amp operation can
lead to RF-emission problems, especially with long PC-board
traces and speaker leads. To counter the RF-emission
problem, the MAX9851 stereo, class D speaker
amps incorporate a proprietary EMI-reduction topology
(active emission limiting) that suppresses high-frequency
RF-emissions from speaker leads/traces at the expense of
slight efficiency degradation. State-of-the-art IC construction
techniques also minimize any interaction between the
class D switching output stages and any of the sensitive,
low-noise analog circuitry on the CODEC.
With the ability to connect to an unregulated, single-cell
Li+ battery, the stereo amps are capable of 1W output
from a 4.2V supply into an 8Ω speaker (Figure 3
power is available if lower impedance speakers are used,
but 4Ω speakers are not generally found in the smaller
diameter drivers commonly used in cell phone designs.
Figure 3. The MAX9851 stereo class D speaker amplifier can operate from a raw battery supply, delivering 1W continuous (at 10% THD+N, 1kHz signal) from a 4.2V supply.
Digital Audio—General Architecture, Signal Flow
To allow conversations, the basic functionality of the
GSM/GPRS cell phone must be supported by an 8kHz (or
optional 16kHz) sampling ADC/DAC path, with a 16-bit
depth in either direction. In the MAX9851, this I/O
function is fully synchronous to the 13MHz (or 26MHz for
= 16kHz) MCLK input to ensure no dropped or repeated
samples. The S1 digital I/O lines in GSM voice mode
) are used to access this basic functionality; the S1
digital interface can operate in MASTER or SLAVE mode.
For Larger Image
Figure 4. Basic GSM speech conversion functionality is supported by a generic, GSM voice-mode protocol on the S1 outputs. This can be operated as a MASTER or SLAVE; SLAVE mode expects both BCLK and LRCLK to be provided by the host.
Many mid- and high-tier phones are commonly required
to provide additional DAC functions at higher bit depths
and sample rates. Examples of these functions are
WMA/MP3 replay or generating WAV file ring tones.
Combining digital-to-analog conversion for these
functions with existing voice converters allows high integration,
and gives one 'point source' for all data conversion.
These features can be useful in product design, in
which ground loops and audio level differences can prove
problematic when trying to sum the two functions in the
Therefore, having one converter to combine voice and
multimedia data seems an ideal solution. The main difficulty
with this approach is that any voice conversion must
remain synchronous to the GSM/GPRS rate dictated by
the MCLK input. The multimedia replay, moreover, often
demands an unrelated sample rate: 44.1kHz or 48kHz, for
example. The MAX9851 solves these challenges by
implementing an algorithm similar to sample-rate conversion
(SRC) on the digital-input data, thus allowing a
single DAC to transform the summed voice and multimedia
data in a synchronous manner.
In SLAVE mode, the incoming sample rate for the GSM
voice data is necessarily sample-rate accurate (as dictated
by MCLK). However, an internal digital PLL locks to the
incoming LRCLK on the S2 digital input, allowing a
precise (averaged over many samples) replication of
nonsynchronous multimedia audio data. In MASTER
mode, again the voice data is correctly aligned to the
desired integer division of MCLK, but the S2 LRCLK
data rate is approximated with a slight fS
error, which is
usually insignificant. Sample rates from 8kHz to 48kHz
are supported on either S1 or S2 inputs.
The MAX9851 S2 digital I/Os have an interface that
supports I²S and popular minor variations thereof. When
not operating in GSM voice mode, the S1 interface can be
programmed to support I²S, maximizing the interface
flexibility often needed in feature-rich, high-end phones.
Digital Audio—GSM Filters
As can be seen in Figure 5
, the S1 digital I/Os have extra
filters that can be enabled in GSM voice mode. These
digital blocks are an efficient implementation of tightly
specified lowpass and highpass filters. This implementation
suppresses energy near the Nyquist band edge and at
low frequencies. The filters can prove beneficial to meet
noise and signal-leakage envelopes when a phone
undergoes testing and qualification. Figure 6
frequency response of enabled filters.
Figure 5. The MAX9851 integrates two independent sets (S1 and S2) of digital-audio interface I/Os. For DAC replay, each interface can run at differing and noninteger-related sample rates in either MASTER or SLAVE modes.
Figure 6. The frequency response of the GSM replay path has enabled GSM filters. At fS = 8kHz, note the steep roll-off just before the Nyquist frequency (4kHz). The highpass filter (HPF) can be optionally defeated.
The previous examples highlight only a few issues that
cell phone system designers/architects must address.
Design cycles are notoriously short for this end application
and feature sets continue to mature and change,
almost on a per-model basis. Therefore, investing time in
a well-engineered, flexible, comprehensive core chipset
architecture is a worthwhile decision.
Controlling low-noise analog circuitry, which interfaces
with multiple replay/record systems at different sample
rates, is only part of the overall design task. It is also
important to integrate the following features in one solution:
- Analog functionality and high performance
- A one-point, digital-/analog-audio interface
- Digital interfacing flexibility
- Comprehensive power management and partitioned
These features address a significant number of system-design,
architecture, and topology issues. The MAX9851
is a 48-pin, 7mm x 7mm single-chip solution that resolves
these issues, and forms the basis of either mid- or high-end
GSM/GPRS cell phone audio designs.