Document toolboxDocument toolbox

(8.D.2.7) SbMonoAECRNSOut

Owned by Sam Stuart
Aug 14, 2024
4 min read

Overview

Frequency domain subband adaptive filter

Discussion

Subband based mono echo canceler / adaptive filter. This module works in conjunction with the WOLA filterbank modules and implements an adaptive filter within each subband. Each subband operates independently which provides fast adaptation time and excellent cancellation.

The length of the effective echo tail equals the number of subband taps times the block size at the input to the WOLA analysis module. For example, assume that the WOLA analysis module has an input block size of 256 samples and an FFT size of 512 samples. Then if the subband echo canceler has 10 taps, this corresponds to an effective tail length of 10 x 256 = 2560 samples.

The module has two input pins. dataIn is the desired response (typically the microphone signal in an AEC). The second input is the mono reference signal.

At instantiation time you specify the length of each subband filter. Then at run time, you specify the number of taps currently used. This allows you to easily vary the filter length at run time and verify performance and CPU load. On some systems - especially those with caches - the computational load of the filter is a function of both the length of each subband and the number of taps currently in use. For most efficient operation, the length of the filter should equal the number of taps in use.

The smoothingTime variable controls the adaptation speed of the filter. smoothingTime is in units of seconds and usually in the range of 10 to 20 seconds. The higher the smoothingTime the slower the filter will adapt but it will achieve higher overall cancellation.

The module has optional residual noise suppression which is enabled via the enableRNS variable. By default, residual noise suppression is on and the output of the echo canceler (the error signal) is further reduced in level. The residual noise suppression is configured by a 7 element interpolation table. This table operates just like the Table X-Y module with configurable (X, Y) lookup points. The lookup table specifies values in dB for the input and output function

The residual noise calculation gains are stored internally in the W2 state array. W2 is updated every block. If the variable .enableRNS is set, then the output is multiplied by W2.

The module also has an optional output pin that is enabled by the module argument RNSPIN. When this pin exists, the module outputs the RNS weights on the second pin. The RNS output data is real and of dimension 1 x numBins, where numBins is the number of FFT bins.

Type Definition

typedef struct _ModuleSbMonoAECRNSOut { ModuleInstanceDescriptor instance; // Common Audio Weaver module instance structure INT32 maxTaps; // Maximum length of the filter INT32 numTaps; // Current length of the filter FLOAT32 V; // Adaptation constant FLOAT32 smoothingTime; // Time constant of the smoothing process. FLOAT32 A; // Smoothing factor FLOAT32 p0; // Starting value for covariance estimate INT32 numIter; // Number of iterations of adaptive loop INT32 enableRNS; // Enable residual noise suppression INT32 stateIndex; // Circular buffer index. [0 numTaps-1] INT32 RNS_numPoints; // Number of values in the lookup table. The total table size is [maxPoints 2]. INT32 RNS_order; // Order of the interpolation. This can be either 2 (for linear) or 4 (for pchip). FLOAT32* WW1; // Adaptive filter coefficients. FLOAT32* P; // Real estimation error covariance matrix. FLOAT32* CC; // Complex state variables (previous inputs). FLOAT32* Pplus; // Intermediate result (real scratch). FLOAT32* Wplus; // Intermediate result (complex scratch). FLOAT32* energyC; // Intermediate result (real scratch). FLOAT32* mu; // Intermediate result (real scratch). FLOAT32* W2; // Residual noise suppression weights. FLOAT32* RNS_table; // Lookup table. The first row is the X values and the second row is the Y values. FLOAT32* RNS_polyCoeffs; // Interpolation coefficients returned by the grid control. } ModuleSbMonoAECRNSOutClass;

Variables

Properties

Name

Type

Usage

isHidden

Default value

Range

Units

maxTaps

int

const

0

8

Unrestricted

samples

numTaps

int

parameter

0

8

1:1:8

samples

V

float

const

0

0.5

Unrestricted

 

smoothingTime

float

parameter

0

10

0.1:1000

sec

A

float

derived

0

1

0:1

 

p0

float

parameter

0

1

Unrestricted

 

numIter

int

parameter

0

1

1:20

 

enableRNS

int

parameter

0

1

0:1:1

 

stateIndex

int

state

1

0

Unrestricted

 

RNS_numPoints

int

const

1

7

4:1:1000

 

RNS_order

int

const

1

2

2:2:4

 

WW1

float*

state

0

[8 x 32]

Unrestricted

 

P

float*

state

0

[8 x 32]

Unrestricted

 

CC

float*

state

0

[8 x 32]

Unrestricted

 

Pplus

float*

state

0

[1 x 8]

Unrestricted

 

Wplus

float*

state

0

[1 x 8]

Unrestricted

 

energyC

float*

state

0

[8 x 1]

Unrestricted

 

mu

float*

state

0

[1 x 8]

Unrestricted

 

W2

float*

state

0

[32 x 1]

Unrestricted

 

RNS_table

float*

parameter

0

[2 x 7]

Unrestricted

 

RNS_polyCoeffs

float*

state

1

[4 x 6]

Unrestricted

 

Pins

Input Pins

Name: dataIn

Description: Micro input

Data type: float

Channel range: 1

Block size range: Unrestricted

Sample rate range: Unrestricted

Complex support: Complex

 

Name: reference

Description: Reference input

Data type: float

Channel range: 1

Block size range: Unrestricted

Sample rate range: Unrestricted

Complex support: Complex

Output Pins

Name: errorSignal

Description: Error signal = difference between filter output and dataIn

Data type: float

 

Name: RNSWeights

Description: Weights from the residual noise suppressor

Data type: float

MATLAB Usage

File Name: sb_mono_aec_rns_out_module.m

M = sb_mono_aec_rns_out_module(NAME, MAXTAPS) Frequency domain subband echo canceler using a Kalman filter based algorithm. Arguments: NAME - name of the module. MAXTAPS - maximum length of each frequency domain complex adaptive filter RNSPIN - Boolean which specifies that the pin will have a second output pin containing the RNS data. This wire holds real data of size 1 x numBins. By default RNSPIN = 0.

Â