Frequency domain based beamformer
This module implements a frequency domain based beamformer. The module has a multichannel input pin "in" which correspond to N microphone signals. The microphone signals must already be in the frequency domain (e.g., after an FFT or WOLA filterbank). The module has an internal table of complex filter coefficients of size [numMics, numBins]. The module multiples each microphone signal by complex coefficients and then sums across all microphone channels. This corresponds to filtering each microphone signal and summing the results.
The module has the ability to store multiple coefficient sets. The argument NUMBEAMS specifies how many different beams to store. The first input pin index is an integer value and specifies which set of beamformer coefficients to use. The index is zero based and must be in the range [0 NUMBEAMS-1]. The function will automatically clip index so that it is in a valid range.
By default, the module outputs a single beam. The third argument OUTPUTALL allows you to output all beams simultaneously. In this case, the module will have a single multichannel output pin with NUMBEAMS channels. In the multichannel case, the input index pin is removed.
The coefficients are ordered as follows. The first 2 * numBins * numMics coefficients correspond to the first beamformer. The coefficient are ordered first based on input channel and then on frequency bin. The coefficient are interleaved real / imag values. The beamformer coefficients can be loaded using the function sb_beamformer_config.m
typedef struct _ModuleSbBeamformerV3
ModuleInstanceDescriptor instance; // Common Audio Weaver module instance structure
INT32 numBeams; // Number of beams allocated.
INT32 numCoeffs; // Number of coefficients to allocate
FLOAT32* coeffs; // Complex beamformer coefficients.
[2 x 33]
Description: Complex data
Data type: float
Channel range: 2
Block size range: Unrestricted
Sample rate range: Unrestricted
Complex support: Complex
Description: Beam output
Data type: float
File Name: sb_beamformer_v3_module.m
M = sb_beamformer_v3_module(NAME, NUMBEAMS, OUTPUTALL)
Creates a frequency domain based beamformer. The module operates
on complex subband data (complex in and complex out). The WOLA
filterbanks go outside of this module. Arguments:
NAME - name of the module.
NUMBEAMS - number of look directions [default = 6]
OUTPUTALL - Boolean which specifies that the module should
output all beams at once (default = 0).
NUMBINS - the default number of bins this subband beam former will
use. This is the initial value and the final NUMBINS is taken from
the input wire.
NUMMICS - value with specifies the number of microphones in this
FFTSIZE - size of the WOLA FFT which will be used with this module.
If empty (the default), then the module will try and infer the FFTSIZE
based on NUMBINS. For example, if NUMBINS = 65 then the FFTSIZE is
assumed to be 128.
The first input pin is an index which specifies which beam
to output. This is zero based in the range [0 NUMBEAMS-1]. The
second input is the complex data.
When OUTPUTALL==1, then the module will have NUMBEAMS output pins
each outputing a different beam direction.