dsp | Mbed

Users » nprobably » Code » dsp

src/Cortex-M4-M3/FilteringFunctions/arm_lms_norm_f32.c@0:1014af42efd9, 2011-03-10 (annotated)

Committer:: simon
Date:: Thu Mar 10 15:07:50 2011 +0000
Revision:: 0:1014af42efd9

Who changed what in which revision?

User	Revision	Line number	New contents of line
simon	0:1014af42efd9	1	/* ----------------------------------------------------------------------
simon	0:1014af42efd9	2	* Copyright (C) 2010 ARM Limited. All rights reserved.
simon	0:1014af42efd9	3	*
simon	0:1014af42efd9	4	* $Date: 29. November 2010
simon	0:1014af42efd9	5	* $Revision: V1.0.3
simon	0:1014af42efd9	6	*
simon	0:1014af42efd9	7	* Project: CMSIS DSP Library
simon	0:1014af42efd9	8	* Title: arm_lms_norm_f32.c
simon	0:1014af42efd9	9	*
simon	0:1014af42efd9	10	* Description: Processing function for the floating-point Normalised LMS.
simon	0:1014af42efd9	11	*
simon	0:1014af42efd9	12	* Target Processor: Cortex-M4/Cortex-M3
simon	0:1014af42efd9	13	*
simon	0:1014af42efd9	14	* Version 1.0.3 2010/11/29
simon	0:1014af42efd9	15	* Re-organized the CMSIS folders and updated documentation.
simon	0:1014af42efd9	16	*
simon	0:1014af42efd9	17	* Version 1.0.2 2010/11/11
simon	0:1014af42efd9	18	* Documentation updated.
simon	0:1014af42efd9	19	*
simon	0:1014af42efd9	20	* Version 1.0.1 2010/10/05
simon	0:1014af42efd9	21	* Production release and review comments incorporated.
simon	0:1014af42efd9	22	*
simon	0:1014af42efd9	23	* Version 1.0.0 2010/09/20
simon	0:1014af42efd9	24	* Production release and review comments incorporated
simon	0:1014af42efd9	25	*
simon	0:1014af42efd9	26	* Version 0.0.7 2010/06/10
simon	0:1014af42efd9	27	* Misra-C changes done
simon	0:1014af42efd9	28	* -------------------------------------------------------------------- */
simon	0:1014af42efd9	29
simon	0:1014af42efd9	30	#include "arm_math.h"
simon	0:1014af42efd9	31
simon	0:1014af42efd9	32	/**
simon	0:1014af42efd9	33	* @ingroup groupFilters
simon	0:1014af42efd9	34	*/
simon	0:1014af42efd9	35
simon	0:1014af42efd9	36	/**
simon	0:1014af42efd9	37	* @defgroup LMS_NORM Normalized LMS Filters
simon	0:1014af42efd9	38	*
simon	0:1014af42efd9	39	* This set of functions implements a commonly used adaptive filter.
simon	0:1014af42efd9	40	* It is related to the Least Mean Square (LMS) adaptive filter and includes an additional normalization
simon	0:1014af42efd9	41	* factor which increases the adaptation rate of the filter.
simon	0:1014af42efd9	42	* The CMSIS DSP Library contains normalized LMS filter functions that operate on Q15, Q31, and floating-point data types.
simon	0:1014af42efd9	43	*
simon	0:1014af42efd9	44	* A normalized least mean square (NLMS) filter consists of two components as shown below.
simon	0:1014af42efd9	45	* The first component is a standard transversal or FIR filter.
simon	0:1014af42efd9	46	* The second component is a coefficient update mechanism.
simon	0:1014af42efd9	47	* The NLMS filter has two input signals.
simon	0:1014af42efd9	48	* The "input" feeds the FIR filter while the "reference input" corresponds to the desired output of the FIR filter.
simon	0:1014af42efd9	49	* That is, the FIR filter coefficients are updated so that the output of the FIR filter matches the reference input.
simon	0:1014af42efd9	50	* The filter coefficient update mechanism is based on the difference between the FIR filter output and the reference input.
simon	0:1014af42efd9	51	* This "error signal" tends towards zero as the filter adapts.
simon	0:1014af42efd9	52	* The NLMS processing functions accept the input and reference input signals and generate the filter output and error signal.
simon	0:1014af42efd9	53	* \image html LMS.gif "Internal structure of the NLMS adaptive filter"
simon	0:1014af42efd9	54	*
simon	0:1014af42efd9	55	* The functions operate on blocks of data and each call to the function processes
simon	0:1014af42efd9	56	* <code>blockSize</code> samples through the filter.
simon	0:1014af42efd9	57	* <code>pSrc</code> points to input signal, <code>pRef</code> points to reference signal,
simon	0:1014af42efd9	58	* <code>pOut</code> points to output signal and <code>pErr</code> points to error signal.
simon	0:1014af42efd9	59	* All arrays contain <code>blockSize</code> values.
simon	0:1014af42efd9	60	*
simon	0:1014af42efd9	61	* The API functions operate on a block-by-block basis.
simon	0:1014af42efd9	62	* Internally, the filter coefficients <code>b[n]</code> are updated on a sample-by-sample basis.
simon	0:1014af42efd9	63	* The convergence of the LMS filter is slower compared to the normalized LMS algorithm.
simon	0:1014af42efd9	64	*
simon	0:1014af42efd9	65	* \par Algorithm:
simon	0:1014af42efd9	66	* The output signal <code>y[n]</code> is computed by a standard FIR filter:
simon	0:1014af42efd9	67	* <pre>
simon	0:1014af42efd9	68	* y[n] = b[0] * x[n] + b[1] * x[n-1] + b[2] * x[n-2] + ...+ b[numTaps-1] * x[n-numTaps+1]
simon	0:1014af42efd9	69	* </pre>
simon	0:1014af42efd9	70	*
simon	0:1014af42efd9	71	* \par
simon	0:1014af42efd9	72	* The error signal equals the difference between the reference signal <code>d[n]</code> and the filter output:
simon	0:1014af42efd9	73	* <pre>
simon	0:1014af42efd9	74	* e[n] = d[n] - y[n].
simon	0:1014af42efd9	75	* </pre>
simon	0:1014af42efd9	76	*
simon	0:1014af42efd9	77	* \par
simon	0:1014af42efd9	78	* After each sample of the error signal is computed the instanteous energy of the filter state variables is calculated:
simon	0:1014af42efd9	79	* <pre>
simon	0:1014af42efd9	80	* E = x[n]^2 + x[n-1]^2 + ... + x[n-numTaps+1]^2.
simon	0:1014af42efd9	81	* </pre>
simon	0:1014af42efd9	82	* The filter coefficients <code>b[k]</code> are then updated on a sample-by-sample basis:
simon	0:1014af42efd9	83	* <pre>
simon	0:1014af42efd9	84	* b[k] = b[k] + e[n] * (mu/E) * x[n-k], for k=0, 1, ..., numTaps-1
simon	0:1014af42efd9	85	* </pre>
simon	0:1014af42efd9	86	* where <code>mu</code> is the step size and controls the rate of coefficient convergence.
simon	0:1014af42efd9	87	*\par
simon	0:1014af42efd9	88	* In the APIs, <code>pCoeffs</code> points to a coefficient array of size <code>numTaps</code>.
simon	0:1014af42efd9	89	* Coefficients are stored in time reversed order.
simon	0:1014af42efd9	90	* \par
simon	0:1014af42efd9	91	* <pre>
simon	0:1014af42efd9	92	* {b[numTaps-1], b[numTaps-2], b[N-2], ..., b[1], b[0]}
simon	0:1014af42efd9	93	* </pre>
simon	0:1014af42efd9	94	* \par
simon	0:1014af42efd9	95	* <code>pState</code> points to a state array of size <code>numTaps + blockSize - 1</code>.
simon	0:1014af42efd9	96	* Samples in the state buffer are stored in the order:
simon	0:1014af42efd9	97	* \par
simon	0:1014af42efd9	98	* <pre>
simon	0:1014af42efd9	99	* {x[n-numTaps+1], x[n-numTaps], x[n-numTaps-1], x[n-numTaps-2]....x[0], x[1], ..., x[blockSize-1]}
simon	0:1014af42efd9	100	* </pre>
simon	0:1014af42efd9	101	* \par
simon	0:1014af42efd9	102	* Note that the length of the state buffer exceeds the length of the coefficient array by <code>blockSize-1</code> samples.
simon	0:1014af42efd9	103	* The increased state buffer length allows circular addressing, which is traditionally used in FIR filters,
simon	0:1014af42efd9	104	* to be avoided and yields a significant speed improvement.
simon	0:1014af42efd9	105	* The state variables are updated after each block of data is processed.
simon	0:1014af42efd9	106	* \par Instance Structure
simon	0:1014af42efd9	107	* The coefficients and state variables for a filter are stored together in an instance data structure.
simon	0:1014af42efd9	108	* A separate instance structure must be defined for each filter and
simon	0:1014af42efd9	109	* coefficient and state arrays cannot be shared among instances.
simon	0:1014af42efd9	110	* There are separate instance structure declarations for each of the 3 supported data types.
simon	0:1014af42efd9	111	*
simon	0:1014af42efd9	112	* \par Initialization Functions
simon	0:1014af42efd9	113	* There is also an associated initialization function for each data type.
simon	0:1014af42efd9	114	* The initialization function performs the following operations:
simon	0:1014af42efd9	115	* - Sets the values of the internal structure fields.
simon	0:1014af42efd9	116	* - Zeros out the values in the state buffer.
simon	0:1014af42efd9	117	* \par
simon	0:1014af42efd9	118	* Instance structure cannot be placed into a const data section and it is recommended to use the initialization function.
simon	0:1014af42efd9	119	* \par Fixed-Point Behavior:
simon	0:1014af42efd9	120	* Care must be taken when using the Q15 and Q31 versions of the normalised LMS filter.
simon	0:1014af42efd9	121	* The following issues must be considered:
simon	0:1014af42efd9	122	* - Scaling of coefficients
simon	0:1014af42efd9	123	* - Overflow and saturation
simon	0:1014af42efd9	124	*
simon	0:1014af42efd9	125	* \par Scaling of Coefficients:
simon	0:1014af42efd9	126	* Filter coefficients are represented as fractional values and
simon	0:1014af42efd9	127	* coefficients are restricted to lie in the range <code>[-1 +1)</code>.
simon	0:1014af42efd9	128	* The fixed-point functions have an additional scaling parameter <code>postShift</code>.
simon	0:1014af42efd9	129	* At the output of the filter's accumulator is a shift register which shifts the result by <code>postShift</code> bits.
simon	0:1014af42efd9	130	* This essentially scales the filter coefficients by <code>2^postShift</code> and
simon	0:1014af42efd9	131	* allows the filter coefficients to exceed the range <code>[+1 -1)</code>.
simon	0:1014af42efd9	132	* The value of <code>postShift</code> is set by the user based on the expected gain through the system being modeled.
simon	0:1014af42efd9	133	*
simon	0:1014af42efd9	134	* \par Overflow and Saturation:
simon	0:1014af42efd9	135	* Overflow and saturation behavior of the fixed-point Q15 and Q31 versions are
simon	0:1014af42efd9	136	* described separately as part of the function specific documentation below.
simon	0:1014af42efd9	137	*/
simon	0:1014af42efd9	138
simon	0:1014af42efd9	139
simon	0:1014af42efd9	140	/**
simon	0:1014af42efd9	141	* @addtogroup LMS_NORM
simon	0:1014af42efd9	142	* @{
simon	0:1014af42efd9	143	*/
simon	0:1014af42efd9	144
simon	0:1014af42efd9	145
simon	0:1014af42efd9	146	/**
simon	0:1014af42efd9	147	* @brief Processing function for floating-point normalized LMS filter.
simon	0:1014af42efd9	148	* @param[in] *S points to an instance of the floating-point normalized LMS filter structure.
simon	0:1014af42efd9	149	* @param[in] *pSrc points to the block of input data.
simon	0:1014af42efd9	150	* @param[in] *pRef points to the block of reference data.
simon	0:1014af42efd9	151	* @param[out] *pOut points to the block of output data.
simon	0:1014af42efd9	152	* @param[out] *pErr points to the block of error data.
simon	0:1014af42efd9	153	* @param[in] blockSize number of samples to process.
simon	0:1014af42efd9	154	* @return none.
simon	0:1014af42efd9	155	*/
simon	0:1014af42efd9	156
simon	0:1014af42efd9	157	void arm_lms_norm_f32(
simon	0:1014af42efd9	158	arm_lms_norm_instance_f32 * S,
simon	0:1014af42efd9	159	float32_t * pSrc,
simon	0:1014af42efd9	160	float32_t * pRef,
simon	0:1014af42efd9	161	float32_t * pOut,
simon	0:1014af42efd9	162	float32_t * pErr,
simon	0:1014af42efd9	163	uint32_t blockSize)
simon	0:1014af42efd9	164	{
simon	0:1014af42efd9	165	float32_t pState = S->pState; / State pointer */
simon	0:1014af42efd9	166	float32_t pCoeffs = S->pCoeffs; / Coefficient pointer */
simon	0:1014af42efd9	167	float32_t pStateCurnt; / Points to the current sample of the state */
simon	0:1014af42efd9	168	float32_t px, pb; /* Temporary pointers for state and coefficient buffers */
simon	0:1014af42efd9	169	float32_t mu = S->mu; /* Adaptive factor */
simon	0:1014af42efd9	170	uint32_t numTaps = S->numTaps; /* Number of filter coefficients in the filter */
simon	0:1014af42efd9	171	uint32_t tapCnt, blkCnt; /* Loop counters */
simon	0:1014af42efd9	172	float32_t energy; /* Energy of the input */
simon	0:1014af42efd9	173	float32_t sum, e, d; /* accumulator, error, reference data sample */
simon	0:1014af42efd9	174	float32_t w, x0, in; /* weight factor, temporary variable to hold input sample and state */
simon	0:1014af42efd9	175
simon	0:1014af42efd9	176	/* Initializations of error, difference, Coefficient update */
simon	0:1014af42efd9	177	e = 0.0f;
simon	0:1014af42efd9	178	d = 0.0f;
simon	0:1014af42efd9	179	w = 0.0f;
simon	0:1014af42efd9	180
simon	0:1014af42efd9	181	energy = S->energy;
simon	0:1014af42efd9	182	x0 = S->x0;
simon	0:1014af42efd9	183
simon	0:1014af42efd9	184	/* S->pState points to buffer which contains previous frame (numTaps - 1) samples */
simon	0:1014af42efd9	185	/* pStateCurnt points to the location where the new input data should be written */
simon	0:1014af42efd9	186	pStateCurnt = &(S->pState[(numTaps - 1u)]);
simon	0:1014af42efd9	187
simon	0:1014af42efd9	188	blkCnt = blockSize;
simon	0:1014af42efd9	189
simon	0:1014af42efd9	190	while(blkCnt > 0u)
simon	0:1014af42efd9	191	{
simon	0:1014af42efd9	192	/* Copy the new input sample into the state buffer */
simon	0:1014af42efd9	193	pStateCurnt++ = pSrc;
simon	0:1014af42efd9	194
simon	0:1014af42efd9	195	/* Initialize pState pointer */
simon	0:1014af42efd9	196	px = pState;
simon	0:1014af42efd9	197
simon	0:1014af42efd9	198	/* Initialize coeff pointer */
simon	0:1014af42efd9	199	pb = (pCoeffs);
simon	0:1014af42efd9	200
simon	0:1014af42efd9	201	/* Read the sample from input buffer */
simon	0:1014af42efd9	202	in = *pSrc++;
simon	0:1014af42efd9	203
simon	0:1014af42efd9	204	/* Update the energy calculation */
simon	0:1014af42efd9	205	energy -= x0 * x0;
simon	0:1014af42efd9	206	energy += in * in;
simon	0:1014af42efd9	207
simon	0:1014af42efd9	208	/* Set the accumulator to zero */
simon	0:1014af42efd9	209	sum = 0.0f;
simon	0:1014af42efd9	210
simon	0:1014af42efd9	211	/* Loop unrolling. Process 4 taps at a time. */
simon	0:1014af42efd9	212	tapCnt = numTaps >> 2;
simon	0:1014af42efd9	213
simon	0:1014af42efd9	214	while(tapCnt > 0u)
simon	0:1014af42efd9	215	{
simon	0:1014af42efd9	216	/* Perform the multiply-accumulate */
simon	0:1014af42efd9	217	sum += (px++) (*pb++);
simon	0:1014af42efd9	218	sum += (px++) (*pb++);
simon	0:1014af42efd9	219	sum += (px++) (*pb++);
simon	0:1014af42efd9	220	sum += (px++) (*pb++);
simon	0:1014af42efd9	221
simon	0:1014af42efd9	222	/* Decrement the loop counter */
simon	0:1014af42efd9	223	tapCnt--;
simon	0:1014af42efd9	224	}
simon	0:1014af42efd9	225
simon	0:1014af42efd9	226	/* If the filter length is not a multiple of 4, compute the remaining filter taps */
simon	0:1014af42efd9	227	tapCnt = numTaps % 0x4u;
simon	0:1014af42efd9	228
simon	0:1014af42efd9	229	while(tapCnt > 0u)
simon	0:1014af42efd9	230	{
simon	0:1014af42efd9	231	/* Perform the multiply-accumulate */
simon	0:1014af42efd9	232	sum += (px++) (*pb++);
simon	0:1014af42efd9	233
simon	0:1014af42efd9	234	/* Decrement the loop counter */
simon	0:1014af42efd9	235	tapCnt--;
simon	0:1014af42efd9	236	}
simon	0:1014af42efd9	237
simon	0:1014af42efd9	238	/* The result in the accumulator, store in the destination buffer. */
simon	0:1014af42efd9	239	*pOut++ = sum;
simon	0:1014af42efd9	240
simon	0:1014af42efd9	241	/* Compute and store error */
simon	0:1014af42efd9	242	d = (float32_t) (*pRef++);
simon	0:1014af42efd9	243	e = d - sum;
simon	0:1014af42efd9	244	*pErr++ = e;
simon	0:1014af42efd9	245
simon	0:1014af42efd9	246	/* Calculation of Weighting factor for updating filter coefficients */
simon	0:1014af42efd9	247	/* epsilon value 0.000000119209289f */
simon	0:1014af42efd9	248	w = (e * mu) / (energy + 0.000000119209289f);
simon	0:1014af42efd9	249
simon	0:1014af42efd9	250	/* Initialize pState pointer */
simon	0:1014af42efd9	251	px = pState;
simon	0:1014af42efd9	252
simon	0:1014af42efd9	253	/* Initialize coeff pointer */
simon	0:1014af42efd9	254	pb = (pCoeffs);
simon	0:1014af42efd9	255
simon	0:1014af42efd9	256	/* Loop unrolling. Process 4 taps at a time. */
simon	0:1014af42efd9	257	tapCnt = numTaps >> 2;
simon	0:1014af42efd9	258
simon	0:1014af42efd9	259	/* Update filter coefficients */
simon	0:1014af42efd9	260	while(tapCnt > 0u)
simon	0:1014af42efd9	261	{
simon	0:1014af42efd9	262	/* Perform the multiply-accumulate */
simon	0:1014af42efd9	263	pb += w (*px++);
simon	0:1014af42efd9	264	pb++;
simon	0:1014af42efd9	265
simon	0:1014af42efd9	266	pb += w (*px++);
simon	0:1014af42efd9	267	pb++;
simon	0:1014af42efd9	268
simon	0:1014af42efd9	269	pb += w (*px++);
simon	0:1014af42efd9	270	pb++;
simon	0:1014af42efd9	271
simon	0:1014af42efd9	272	pb += w (*px++);
simon	0:1014af42efd9	273	pb++;
simon	0:1014af42efd9	274
simon	0:1014af42efd9	275
simon	0:1014af42efd9	276	/* Decrement the loop counter */
simon	0:1014af42efd9	277	tapCnt--;
simon	0:1014af42efd9	278	}
simon	0:1014af42efd9	279
simon	0:1014af42efd9	280	/* If the filter length is not a multiple of 4, compute the remaining filter taps */
simon	0:1014af42efd9	281	tapCnt = numTaps % 0x4u;
simon	0:1014af42efd9	282
simon	0:1014af42efd9	283	while(tapCnt > 0u)
simon	0:1014af42efd9	284	{
simon	0:1014af42efd9	285	/* Perform the multiply-accumulate */
simon	0:1014af42efd9	286	pb += w (*px++);
simon	0:1014af42efd9	287	pb++;
simon	0:1014af42efd9	288
simon	0:1014af42efd9	289	/* Decrement the loop counter */
simon	0:1014af42efd9	290	tapCnt--;
simon	0:1014af42efd9	291	}
simon	0:1014af42efd9	292
simon	0:1014af42efd9	293	x0 = *pState;
simon	0:1014af42efd9	294
simon	0:1014af42efd9	295	/* Advance state pointer by 1 for the next sample */
simon	0:1014af42efd9	296	pState = pState + 1;
simon	0:1014af42efd9	297
simon	0:1014af42efd9	298	/* Decrement the loop counter */
simon	0:1014af42efd9	299	blkCnt--;
simon	0:1014af42efd9	300	}
simon	0:1014af42efd9	301
simon	0:1014af42efd9	302	S->energy = energy;
simon	0:1014af42efd9	303	S->x0 = x0;
simon	0:1014af42efd9	304
simon	0:1014af42efd9	305	/* Processing is complete. Now copy the last numTaps - 1 samples to the
simon	0:1014af42efd9	306	satrt of the state buffer. This prepares the state buffer for the
simon	0:1014af42efd9	307	next function call. */
simon	0:1014af42efd9	308
simon	0:1014af42efd9	309	/* Points to the start of the pState buffer */
simon	0:1014af42efd9	310	pStateCurnt = S->pState;
simon	0:1014af42efd9	311
simon	0:1014af42efd9	312	/* Loop unrolling for (numTaps - 1u)/4 samples copy */
simon	0:1014af42efd9	313	tapCnt = (numTaps - 1u) >> 2u;
simon	0:1014af42efd9	314
simon	0:1014af42efd9	315	/* copy data */
simon	0:1014af42efd9	316	while(tapCnt > 0u)
simon	0:1014af42efd9	317	{
simon	0:1014af42efd9	318	pStateCurnt++ = pState++;
simon	0:1014af42efd9	319	pStateCurnt++ = pState++;
simon	0:1014af42efd9	320	pStateCurnt++ = pState++;
simon	0:1014af42efd9	321	pStateCurnt++ = pState++;
simon	0:1014af42efd9	322
simon	0:1014af42efd9	323	/* Decrement the loop counter */
simon	0:1014af42efd9	324	tapCnt--;
simon	0:1014af42efd9	325	}
simon	0:1014af42efd9	326
simon	0:1014af42efd9	327	/* Calculate remaining number of copies */
simon	0:1014af42efd9	328	tapCnt = (numTaps - 1u) % 0x4u;
simon	0:1014af42efd9	329
simon	0:1014af42efd9	330	/* Copy the remaining q31_t data */
simon	0:1014af42efd9	331	while(tapCnt > 0u)
simon	0:1014af42efd9	332	{
simon	0:1014af42efd9	333	pStateCurnt++ = pState++;
simon	0:1014af42efd9	334
simon	0:1014af42efd9	335	/* Decrement the loop counter */
simon	0:1014af42efd9	336	tapCnt--;
simon	0:1014af42efd9	337	}
simon	0:1014af42efd9	338
simon	0:1014af42efd9	339
simon	0:1014af42efd9	340	}
simon	0:1014af42efd9	341
simon	0:1014af42efd9	342	/**
simon	0:1014af42efd9	343	* @} end of LMS_NORM group
simon	0:1014af42efd9	344	*/

Repository toolbox

Repository details

Type:	Library
Created:	29 Jun 2017
Imports:	0
Forks:	0
Commits:	4
Dependents:	0
Dependencies:	0
Followers:	1

Important changes to repositories hosted on mbed.com

src/Cortex-M4-M3/FilteringFunctions/arm_lms_norm_f32.c@0:1014af42efd9, 2011-03-10 (annotated)

Who changed what in which revision?

Repository toolbox

Repository details

Important Information for this Arm website

Important changes to repositories hosted on mbed.com

src/Cortex-M4-M3/FilteringFunctions/arm_lms_norm_f32.c@0:1014af42efd9, 2011-03-10 (annotated)

Who changed what in which revision?

Repository toolbox

Repository details

Important Information for this Arm website

Access Warning