Overview

Sparse FIR filter in which most coefficients equal zero

Discussion

This module implements a sparse FIR filter in which most coefficients are equal to zero. The filter is implemented using a tap delay line rather than a traditional FIR filter. The number of output taps (NUMTAPS) is specified at instantiation time. For each tap, you specify the delay time (tapDelay array) and the weighting coefficient (tapCoeffs array). Delays are specified in units of samples.

At construction time, you specify the maximum delay (maxDelay) of the module. maxDelay determines the size of the internal state buffer. To be precise, the size of the internal state buffer equals maxDelay+blockSize samples. The delay time per tap must be in the range [0 MAXDELAY]. Delay times and coefficients can be adjusted at run-time. However, the module is not smoothly updating and you may get audible clicks when adjusting the tapDelay at run-time. The module uses one scratch buffer whose size equals the size of the input pin.

The module gives you some control over where the state buffer is allocated in memory. By default, the memory heap AWE_FAST2SLOW is used and thus the state buffer is first allocated in internal DM memory. If this fails, then internal PM memory is used. And finally, if this fails, then external memory is used. The 4th input argument MEMHEAP provide control over the allocation process. See the fir_sparse_module.m function header for more details.

The tapCoeffs and tapDelay array are always allocated in internal DM memory.

Type Definition

typedef struct _ModuleFIRSparse
{
    ModuleInstanceDescriptor instance;            // Common Audio Weaver module instance structure
    INT32 numTaps;                                // Number of non-zero coefficients (taps) in the filter
    INT32 maxDelay;                               // Maximum delay that can be realized, in samples
    INT32 stateHeap;                              // Heap in which to allocate memory
    INT32 stateIndex;                             // Index of the oldest state variable in the array of state variables
    INT32* tapDelay;                              // Current delay per channel
    FLOAT32* tapCoeffs;                           // Coefficient applied to each tap output
    FLOAT32* state;                               // State variable array
} ModuleFIRSparseClass;

Variables

Properties

Name	Type	Usage	isHidden	Default value	Range	Units
numTaps	int	const	0	5	Unrestricted
maxDelay	int	const	0	100	1:100000	samples
stateHeap	int	const	1	2	Unrestricted
stateIndex	int	state	1	0	Unrestricted
tapDelay	int*	parameter	0	[5 x 1]	0:1:100	samples
tapCoeffs	float*	parameter	0	[5 x 1]	Unrestricted
state	float*	state	1	[132 x 1]	Unrestricted

Pins

Input Pins

Name: in

Description: audio input

Data type: float

Channel range: 1

Block size range: Unrestricted

Sample rate range: Unrestricted

Complex support: Real

Output Pins

Name: out

Description: audio output

Data type: float

Scratch Pins

Channel count: 1

Block size: 32

Sample rate: 48000

MATLAB Usage

File Name: fir_sparse_module.m

 M=fir_sparse_module(NAME, NUMTAPS, MAXDELAY, MEMHEAP)
 Creates a sparse FIR filter which is equivalent to a multi-tap time
 delay in which the outputs are weighted and summed.  The module
 has a single mono input pin and a single mono output pin.
 Arguments:
    NAME - name of the module.
    NUMTAPS - number of taps
    MAXDELAY - Maximum delay, in samples.  This is the largest delay
               which will be supported and determines the length of the
               internal state buffer.
    MEMHEAP - specifies the memory heap to use to allocate the main
              state buffer.  This is a string and follows the memory
              allocation enumeration in Framework.h.  Allowable values
              are:
              'AWE_HEAP_FAST' - always use internal DM memory.
              'AWE_HEAP_FASTB' - always use internal PM memory.
              'AWE_HEAP_SLOW' - always use external memory. (the default)
              'AWE_HEAP_FAST2SLOW' - use internal memory.  If this fails
                 then use external memory (the default).