G-
T
W
-000
7444.2.2
Packet Loss
Concealment
Noise
Suppression
Adaptive
Equaliser
Equaliser
AGC
Clipper
Equaliser
Comfort Noise
Noise
Suppression
NDVC
Auxiliary Stream
Mix
Wind Noise
Reduction
Mic Gain
Side
Tone
DAC
B
lue
to
oth
R
ad
io
Acoustic Echo
Canceller
Dual-microphone
Signal
Separation
AGC
Figure 16.3: 2-mic CVC Block Diagram
Section 16.2.3 to Section 16.2.13 describe the audio processing functions provided within CVC.
16.2.1
Wind Noise Reduction
The wind noise algorithm achieves excellent wind noise reduction with very low power overhead, which has a
negligible impact on battery life. The wind noise capability operates in the noise suppression block in the transmit
path and dynamically detects and engages when wind noise is present. SNR improvements depend on wind
direction, speech and microphone placement. Improvements of up to 32dB are achievable using the DSP module.
CVC wind noise performance is further improved by suitable mechanical baffling of the microphone which is
optimised during the tuning process.
16.2.2 Dual-microphone Signal Separation
The dual-microphone signal separation is the major dynamic noise suppression block in 2-mic CVC. It separates
the speech from the competing noises. It achieves this by first applying a pre-stage algorithm using a blind source
separation processing technique. Blind source separation is a rules based filter which uses the 2 microphone's spatial
information, direction of arrival and power ratios assumptions etc.
Blind source separation results in speech (S1) and noise (S2) dominant outputs. These outputs are then processed
by a post stage adaptive noise canceller filter to further reduce the environmental noise, resulting in a single-channel
noise suppressed output. Depending on the acoustic arrangement of the microphone and the noise type, the dual-
microphone signal separation block provides up to 22dB SNR of dynamic noise suppression.
16.2.3 Noise Suppression
The noise suppression block is implemented in both signal paths. It is completely independent and is individually
tuned. Noise suppression is a sub-band stationary / quasi-stationary noise suppression algorithm that uses the
temporal characteristics of speech and noise to remove the noise from the composite signal while maximising speech
quality. The current implementation has the capability to improve the SNR by > 20dB.
16.2.4 Acoustic Echo Cancellation
The AEC includes:
■
A referenced sub-band adaptive linear filter that models the acoustic path from the receive reference point
to the microphone input
■
Non-linear echo cancellation. A non-linear processing function that adaptively applies additional attenuation
when excessive residual echo is detected after the linear filter
Advance Information
This material is subject to CSR's non-disclosure agreement
© Cambridge Silicon Radio Limited 2011
Page 97 of 110
CS-209182-DSP1
CSR8640 BGA
Data Sheet
