V4.0.1 of the ARM CMSIS DSP libraries. Note that arm_bitreversal2.s, arm_cfft_f32.c and arm_rfft_fast_f32.c had to be removed. arm_bitreversal2.s will not assemble with the online tools. So, the fast f32 FFT functions are not yet available. All the other FFT functions are available.
Dependents: MPU9150_Example fir_f32 fir_f32 MPU9150_nucleo_noni2cdev ... more
FilteringFunctions/arm_lms_norm_f32.c@0:3d9c67d97d6f, 2014-07-28 (annotated)
- Committer:
- emh203
- Date:
- Mon Jul 28 15:03:15 2014 +0000
- Revision:
- 0:3d9c67d97d6f
1st working commit. Had to remove arm_bitreversal2.s arm_cfft_f32.c and arm_rfft_fast_f32.c. The .s will not assemble. For now I removed these functions so we could at least have a library for the other functions.
Who changed what in which revision?
User | Revision | Line number | New contents of line |
---|---|---|---|
emh203 | 0:3d9c67d97d6f | 1 | /* ---------------------------------------------------------------------- |
emh203 | 0:3d9c67d97d6f | 2 | * Copyright (C) 2010-2014 ARM Limited. All rights reserved. |
emh203 | 0:3d9c67d97d6f | 3 | * |
emh203 | 0:3d9c67d97d6f | 4 | * $Date: 12. March 2014 |
emh203 | 0:3d9c67d97d6f | 5 | * $Revision: V1.4.3 |
emh203 | 0:3d9c67d97d6f | 6 | * |
emh203 | 0:3d9c67d97d6f | 7 | * Project: CMSIS DSP Library |
emh203 | 0:3d9c67d97d6f | 8 | * Title: arm_lms_norm_f32.c |
emh203 | 0:3d9c67d97d6f | 9 | * |
emh203 | 0:3d9c67d97d6f | 10 | * Description: Processing function for the floating-point Normalised LMS. |
emh203 | 0:3d9c67d97d6f | 11 | * |
emh203 | 0:3d9c67d97d6f | 12 | * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0 |
emh203 | 0:3d9c67d97d6f | 13 | * |
emh203 | 0:3d9c67d97d6f | 14 | * Redistribution and use in source and binary forms, with or without |
emh203 | 0:3d9c67d97d6f | 15 | * modification, are permitted provided that the following conditions |
emh203 | 0:3d9c67d97d6f | 16 | * are met: |
emh203 | 0:3d9c67d97d6f | 17 | * - Redistributions of source code must retain the above copyright |
emh203 | 0:3d9c67d97d6f | 18 | * notice, this list of conditions and the following disclaimer. |
emh203 | 0:3d9c67d97d6f | 19 | * - Redistributions in binary form must reproduce the above copyright |
emh203 | 0:3d9c67d97d6f | 20 | * notice, this list of conditions and the following disclaimer in |
emh203 | 0:3d9c67d97d6f | 21 | * the documentation and/or other materials provided with the |
emh203 | 0:3d9c67d97d6f | 22 | * distribution. |
emh203 | 0:3d9c67d97d6f | 23 | * - Neither the name of ARM LIMITED nor the names of its contributors |
emh203 | 0:3d9c67d97d6f | 24 | * may be used to endorse or promote products derived from this |
emh203 | 0:3d9c67d97d6f | 25 | * software without specific prior written permission. |
emh203 | 0:3d9c67d97d6f | 26 | * |
emh203 | 0:3d9c67d97d6f | 27 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
emh203 | 0:3d9c67d97d6f | 28 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
emh203 | 0:3d9c67d97d6f | 29 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
emh203 | 0:3d9c67d97d6f | 30 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
emh203 | 0:3d9c67d97d6f | 31 | * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
emh203 | 0:3d9c67d97d6f | 32 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
emh203 | 0:3d9c67d97d6f | 33 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
emh203 | 0:3d9c67d97d6f | 34 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
emh203 | 0:3d9c67d97d6f | 35 | * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
emh203 | 0:3d9c67d97d6f | 36 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN |
emh203 | 0:3d9c67d97d6f | 37 | * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
emh203 | 0:3d9c67d97d6f | 38 | * POSSIBILITY OF SUCH DAMAGE. |
emh203 | 0:3d9c67d97d6f | 39 | * -------------------------------------------------------------------- */ |
emh203 | 0:3d9c67d97d6f | 40 | |
emh203 | 0:3d9c67d97d6f | 41 | #include "arm_math.h" |
emh203 | 0:3d9c67d97d6f | 42 | |
emh203 | 0:3d9c67d97d6f | 43 | /** |
emh203 | 0:3d9c67d97d6f | 44 | * @ingroup groupFilters |
emh203 | 0:3d9c67d97d6f | 45 | */ |
emh203 | 0:3d9c67d97d6f | 46 | |
emh203 | 0:3d9c67d97d6f | 47 | /** |
emh203 | 0:3d9c67d97d6f | 48 | * @defgroup LMS_NORM Normalized LMS Filters |
emh203 | 0:3d9c67d97d6f | 49 | * |
emh203 | 0:3d9c67d97d6f | 50 | * This set of functions implements a commonly used adaptive filter. |
emh203 | 0:3d9c67d97d6f | 51 | * It is related to the Least Mean Square (LMS) adaptive filter and includes an additional normalization |
emh203 | 0:3d9c67d97d6f | 52 | * factor which increases the adaptation rate of the filter. |
emh203 | 0:3d9c67d97d6f | 53 | * The CMSIS DSP Library contains normalized LMS filter functions that operate on Q15, Q31, and floating-point data types. |
emh203 | 0:3d9c67d97d6f | 54 | * |
emh203 | 0:3d9c67d97d6f | 55 | * A normalized least mean square (NLMS) filter consists of two components as shown below. |
emh203 | 0:3d9c67d97d6f | 56 | * The first component is a standard transversal or FIR filter. |
emh203 | 0:3d9c67d97d6f | 57 | * The second component is a coefficient update mechanism. |
emh203 | 0:3d9c67d97d6f | 58 | * The NLMS filter has two input signals. |
emh203 | 0:3d9c67d97d6f | 59 | * The "input" feeds the FIR filter while the "reference input" corresponds to the desired output of the FIR filter. |
emh203 | 0:3d9c67d97d6f | 60 | * That is, the FIR filter coefficients are updated so that the output of the FIR filter matches the reference input. |
emh203 | 0:3d9c67d97d6f | 61 | * The filter coefficient update mechanism is based on the difference between the FIR filter output and the reference input. |
emh203 | 0:3d9c67d97d6f | 62 | * This "error signal" tends towards zero as the filter adapts. |
emh203 | 0:3d9c67d97d6f | 63 | * The NLMS processing functions accept the input and reference input signals and generate the filter output and error signal. |
emh203 | 0:3d9c67d97d6f | 64 | * \image html LMS.gif "Internal structure of the NLMS adaptive filter" |
emh203 | 0:3d9c67d97d6f | 65 | * |
emh203 | 0:3d9c67d97d6f | 66 | * The functions operate on blocks of data and each call to the function processes |
emh203 | 0:3d9c67d97d6f | 67 | * <code>blockSize</code> samples through the filter. |
emh203 | 0:3d9c67d97d6f | 68 | * <code>pSrc</code> points to input signal, <code>pRef</code> points to reference signal, |
emh203 | 0:3d9c67d97d6f | 69 | * <code>pOut</code> points to output signal and <code>pErr</code> points to error signal. |
emh203 | 0:3d9c67d97d6f | 70 | * All arrays contain <code>blockSize</code> values. |
emh203 | 0:3d9c67d97d6f | 71 | * |
emh203 | 0:3d9c67d97d6f | 72 | * The functions operate on a block-by-block basis. |
emh203 | 0:3d9c67d97d6f | 73 | * Internally, the filter coefficients <code>b[n]</code> are updated on a sample-by-sample basis. |
emh203 | 0:3d9c67d97d6f | 74 | * The convergence of the LMS filter is slower compared to the normalized LMS algorithm. |
emh203 | 0:3d9c67d97d6f | 75 | * |
emh203 | 0:3d9c67d97d6f | 76 | * \par Algorithm: |
emh203 | 0:3d9c67d97d6f | 77 | * The output signal <code>y[n]</code> is computed by a standard FIR filter: |
emh203 | 0:3d9c67d97d6f | 78 | * <pre> |
emh203 | 0:3d9c67d97d6f | 79 | * y[n] = b[0] * x[n] + b[1] * x[n-1] + b[2] * x[n-2] + ...+ b[numTaps-1] * x[n-numTaps+1] |
emh203 | 0:3d9c67d97d6f | 80 | * </pre> |
emh203 | 0:3d9c67d97d6f | 81 | * |
emh203 | 0:3d9c67d97d6f | 82 | * \par |
emh203 | 0:3d9c67d97d6f | 83 | * The error signal equals the difference between the reference signal <code>d[n]</code> and the filter output: |
emh203 | 0:3d9c67d97d6f | 84 | * <pre> |
emh203 | 0:3d9c67d97d6f | 85 | * e[n] = d[n] - y[n]. |
emh203 | 0:3d9c67d97d6f | 86 | * </pre> |
emh203 | 0:3d9c67d97d6f | 87 | * |
emh203 | 0:3d9c67d97d6f | 88 | * \par |
emh203 | 0:3d9c67d97d6f | 89 | * After each sample of the error signal is computed the instanteous energy of the filter state variables is calculated: |
emh203 | 0:3d9c67d97d6f | 90 | * <pre> |
emh203 | 0:3d9c67d97d6f | 91 | * E = x[n]^2 + x[n-1]^2 + ... + x[n-numTaps+1]^2. |
emh203 | 0:3d9c67d97d6f | 92 | * </pre> |
emh203 | 0:3d9c67d97d6f | 93 | * The filter coefficients <code>b[k]</code> are then updated on a sample-by-sample basis: |
emh203 | 0:3d9c67d97d6f | 94 | * <pre> |
emh203 | 0:3d9c67d97d6f | 95 | * b[k] = b[k] + e[n] * (mu/E) * x[n-k], for k=0, 1, ..., numTaps-1 |
emh203 | 0:3d9c67d97d6f | 96 | * </pre> |
emh203 | 0:3d9c67d97d6f | 97 | * where <code>mu</code> is the step size and controls the rate of coefficient convergence. |
emh203 | 0:3d9c67d97d6f | 98 | *\par |
emh203 | 0:3d9c67d97d6f | 99 | * In the APIs, <code>pCoeffs</code> points to a coefficient array of size <code>numTaps</code>. |
emh203 | 0:3d9c67d97d6f | 100 | * Coefficients are stored in time reversed order. |
emh203 | 0:3d9c67d97d6f | 101 | * \par |
emh203 | 0:3d9c67d97d6f | 102 | * <pre> |
emh203 | 0:3d9c67d97d6f | 103 | * {b[numTaps-1], b[numTaps-2], b[N-2], ..., b[1], b[0]} |
emh203 | 0:3d9c67d97d6f | 104 | * </pre> |
emh203 | 0:3d9c67d97d6f | 105 | * \par |
emh203 | 0:3d9c67d97d6f | 106 | * <code>pState</code> points to a state array of size <code>numTaps + blockSize - 1</code>. |
emh203 | 0:3d9c67d97d6f | 107 | * Samples in the state buffer are stored in the order: |
emh203 | 0:3d9c67d97d6f | 108 | * \par |
emh203 | 0:3d9c67d97d6f | 109 | * <pre> |
emh203 | 0:3d9c67d97d6f | 110 | * {x[n-numTaps+1], x[n-numTaps], x[n-numTaps-1], x[n-numTaps-2]....x[0], x[1], ..., x[blockSize-1]} |
emh203 | 0:3d9c67d97d6f | 111 | * </pre> |
emh203 | 0:3d9c67d97d6f | 112 | * \par |
emh203 | 0:3d9c67d97d6f | 113 | * Note that the length of the state buffer exceeds the length of the coefficient array by <code>blockSize-1</code> samples. |
emh203 | 0:3d9c67d97d6f | 114 | * The increased state buffer length allows circular addressing, which is traditionally used in FIR filters, |
emh203 | 0:3d9c67d97d6f | 115 | * to be avoided and yields a significant speed improvement. |
emh203 | 0:3d9c67d97d6f | 116 | * The state variables are updated after each block of data is processed. |
emh203 | 0:3d9c67d97d6f | 117 | * \par Instance Structure |
emh203 | 0:3d9c67d97d6f | 118 | * The coefficients and state variables for a filter are stored together in an instance data structure. |
emh203 | 0:3d9c67d97d6f | 119 | * A separate instance structure must be defined for each filter and |
emh203 | 0:3d9c67d97d6f | 120 | * coefficient and state arrays cannot be shared among instances. |
emh203 | 0:3d9c67d97d6f | 121 | * There are separate instance structure declarations for each of the 3 supported data types. |
emh203 | 0:3d9c67d97d6f | 122 | * |
emh203 | 0:3d9c67d97d6f | 123 | * \par Initialization Functions |
emh203 | 0:3d9c67d97d6f | 124 | * There is also an associated initialization function for each data type. |
emh203 | 0:3d9c67d97d6f | 125 | * The initialization function performs the following operations: |
emh203 | 0:3d9c67d97d6f | 126 | * - Sets the values of the internal structure fields. |
emh203 | 0:3d9c67d97d6f | 127 | * - Zeros out the values in the state buffer. |
emh203 | 0:3d9c67d97d6f | 128 | * To do this manually without calling the init function, assign the follow subfields of the instance structure: |
emh203 | 0:3d9c67d97d6f | 129 | * numTaps, pCoeffs, mu, energy, x0, pState. Also set all of the values in pState to zero. |
emh203 | 0:3d9c67d97d6f | 130 | * For Q7, Q15, and Q31 the following fields must also be initialized; |
emh203 | 0:3d9c67d97d6f | 131 | * recipTable, postShift |
emh203 | 0:3d9c67d97d6f | 132 | * |
emh203 | 0:3d9c67d97d6f | 133 | * \par |
emh203 | 0:3d9c67d97d6f | 134 | * Instance structure cannot be placed into a const data section and it is recommended to use the initialization function. |
emh203 | 0:3d9c67d97d6f | 135 | * \par Fixed-Point Behavior: |
emh203 | 0:3d9c67d97d6f | 136 | * Care must be taken when using the Q15 and Q31 versions of the normalised LMS filter. |
emh203 | 0:3d9c67d97d6f | 137 | * The following issues must be considered: |
emh203 | 0:3d9c67d97d6f | 138 | * - Scaling of coefficients |
emh203 | 0:3d9c67d97d6f | 139 | * - Overflow and saturation |
emh203 | 0:3d9c67d97d6f | 140 | * |
emh203 | 0:3d9c67d97d6f | 141 | * \par Scaling of Coefficients: |
emh203 | 0:3d9c67d97d6f | 142 | * Filter coefficients are represented as fractional values and |
emh203 | 0:3d9c67d97d6f | 143 | * coefficients are restricted to lie in the range <code>[-1 +1)</code>. |
emh203 | 0:3d9c67d97d6f | 144 | * The fixed-point functions have an additional scaling parameter <code>postShift</code>. |
emh203 | 0:3d9c67d97d6f | 145 | * At the output of the filter's accumulator is a shift register which shifts the result by <code>postShift</code> bits. |
emh203 | 0:3d9c67d97d6f | 146 | * This essentially scales the filter coefficients by <code>2^postShift</code> and |
emh203 | 0:3d9c67d97d6f | 147 | * allows the filter coefficients to exceed the range <code>[+1 -1)</code>. |
emh203 | 0:3d9c67d97d6f | 148 | * The value of <code>postShift</code> is set by the user based on the expected gain through the system being modeled. |
emh203 | 0:3d9c67d97d6f | 149 | * |
emh203 | 0:3d9c67d97d6f | 150 | * \par Overflow and Saturation: |
emh203 | 0:3d9c67d97d6f | 151 | * Overflow and saturation behavior of the fixed-point Q15 and Q31 versions are |
emh203 | 0:3d9c67d97d6f | 152 | * described separately as part of the function specific documentation below. |
emh203 | 0:3d9c67d97d6f | 153 | */ |
emh203 | 0:3d9c67d97d6f | 154 | |
emh203 | 0:3d9c67d97d6f | 155 | |
emh203 | 0:3d9c67d97d6f | 156 | /** |
emh203 | 0:3d9c67d97d6f | 157 | * @addtogroup LMS_NORM |
emh203 | 0:3d9c67d97d6f | 158 | * @{ |
emh203 | 0:3d9c67d97d6f | 159 | */ |
emh203 | 0:3d9c67d97d6f | 160 | |
emh203 | 0:3d9c67d97d6f | 161 | |
emh203 | 0:3d9c67d97d6f | 162 | /** |
emh203 | 0:3d9c67d97d6f | 163 | * @brief Processing function for floating-point normalized LMS filter. |
emh203 | 0:3d9c67d97d6f | 164 | * @param[in] *S points to an instance of the floating-point normalized LMS filter structure. |
emh203 | 0:3d9c67d97d6f | 165 | * @param[in] *pSrc points to the block of input data. |
emh203 | 0:3d9c67d97d6f | 166 | * @param[in] *pRef points to the block of reference data. |
emh203 | 0:3d9c67d97d6f | 167 | * @param[out] *pOut points to the block of output data. |
emh203 | 0:3d9c67d97d6f | 168 | * @param[out] *pErr points to the block of error data. |
emh203 | 0:3d9c67d97d6f | 169 | * @param[in] blockSize number of samples to process. |
emh203 | 0:3d9c67d97d6f | 170 | * @return none. |
emh203 | 0:3d9c67d97d6f | 171 | */ |
emh203 | 0:3d9c67d97d6f | 172 | |
emh203 | 0:3d9c67d97d6f | 173 | void arm_lms_norm_f32( |
emh203 | 0:3d9c67d97d6f | 174 | arm_lms_norm_instance_f32 * S, |
emh203 | 0:3d9c67d97d6f | 175 | float32_t * pSrc, |
emh203 | 0:3d9c67d97d6f | 176 | float32_t * pRef, |
emh203 | 0:3d9c67d97d6f | 177 | float32_t * pOut, |
emh203 | 0:3d9c67d97d6f | 178 | float32_t * pErr, |
emh203 | 0:3d9c67d97d6f | 179 | uint32_t blockSize) |
emh203 | 0:3d9c67d97d6f | 180 | { |
emh203 | 0:3d9c67d97d6f | 181 | float32_t *pState = S->pState; /* State pointer */ |
emh203 | 0:3d9c67d97d6f | 182 | float32_t *pCoeffs = S->pCoeffs; /* Coefficient pointer */ |
emh203 | 0:3d9c67d97d6f | 183 | float32_t *pStateCurnt; /* Points to the current sample of the state */ |
emh203 | 0:3d9c67d97d6f | 184 | float32_t *px, *pb; /* Temporary pointers for state and coefficient buffers */ |
emh203 | 0:3d9c67d97d6f | 185 | float32_t mu = S->mu; /* Adaptive factor */ |
emh203 | 0:3d9c67d97d6f | 186 | uint32_t numTaps = S->numTaps; /* Number of filter coefficients in the filter */ |
emh203 | 0:3d9c67d97d6f | 187 | uint32_t tapCnt, blkCnt; /* Loop counters */ |
emh203 | 0:3d9c67d97d6f | 188 | float32_t energy; /* Energy of the input */ |
emh203 | 0:3d9c67d97d6f | 189 | float32_t sum, e, d; /* accumulator, error, reference data sample */ |
emh203 | 0:3d9c67d97d6f | 190 | float32_t w, x0, in; /* weight factor, temporary variable to hold input sample and state */ |
emh203 | 0:3d9c67d97d6f | 191 | |
emh203 | 0:3d9c67d97d6f | 192 | /* Initializations of error, difference, Coefficient update */ |
emh203 | 0:3d9c67d97d6f | 193 | e = 0.0f; |
emh203 | 0:3d9c67d97d6f | 194 | d = 0.0f; |
emh203 | 0:3d9c67d97d6f | 195 | w = 0.0f; |
emh203 | 0:3d9c67d97d6f | 196 | |
emh203 | 0:3d9c67d97d6f | 197 | energy = S->energy; |
emh203 | 0:3d9c67d97d6f | 198 | x0 = S->x0; |
emh203 | 0:3d9c67d97d6f | 199 | |
emh203 | 0:3d9c67d97d6f | 200 | /* S->pState points to buffer which contains previous frame (numTaps - 1) samples */ |
emh203 | 0:3d9c67d97d6f | 201 | /* pStateCurnt points to the location where the new input data should be written */ |
emh203 | 0:3d9c67d97d6f | 202 | pStateCurnt = &(S->pState[(numTaps - 1u)]); |
emh203 | 0:3d9c67d97d6f | 203 | |
emh203 | 0:3d9c67d97d6f | 204 | /* Loop over blockSize number of values */ |
emh203 | 0:3d9c67d97d6f | 205 | blkCnt = blockSize; |
emh203 | 0:3d9c67d97d6f | 206 | |
emh203 | 0:3d9c67d97d6f | 207 | |
emh203 | 0:3d9c67d97d6f | 208 | #ifndef ARM_MATH_CM0_FAMILY |
emh203 | 0:3d9c67d97d6f | 209 | |
emh203 | 0:3d9c67d97d6f | 210 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
emh203 | 0:3d9c67d97d6f | 211 | |
emh203 | 0:3d9c67d97d6f | 212 | while(blkCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 213 | { |
emh203 | 0:3d9c67d97d6f | 214 | /* Copy the new input sample into the state buffer */ |
emh203 | 0:3d9c67d97d6f | 215 | *pStateCurnt++ = *pSrc; |
emh203 | 0:3d9c67d97d6f | 216 | |
emh203 | 0:3d9c67d97d6f | 217 | /* Initialize pState pointer */ |
emh203 | 0:3d9c67d97d6f | 218 | px = pState; |
emh203 | 0:3d9c67d97d6f | 219 | |
emh203 | 0:3d9c67d97d6f | 220 | /* Initialize coeff pointer */ |
emh203 | 0:3d9c67d97d6f | 221 | pb = (pCoeffs); |
emh203 | 0:3d9c67d97d6f | 222 | |
emh203 | 0:3d9c67d97d6f | 223 | /* Read the sample from input buffer */ |
emh203 | 0:3d9c67d97d6f | 224 | in = *pSrc++; |
emh203 | 0:3d9c67d97d6f | 225 | |
emh203 | 0:3d9c67d97d6f | 226 | /* Update the energy calculation */ |
emh203 | 0:3d9c67d97d6f | 227 | energy -= x0 * x0; |
emh203 | 0:3d9c67d97d6f | 228 | energy += in * in; |
emh203 | 0:3d9c67d97d6f | 229 | |
emh203 | 0:3d9c67d97d6f | 230 | /* Set the accumulator to zero */ |
emh203 | 0:3d9c67d97d6f | 231 | sum = 0.0f; |
emh203 | 0:3d9c67d97d6f | 232 | |
emh203 | 0:3d9c67d97d6f | 233 | /* Loop unrolling. Process 4 taps at a time. */ |
emh203 | 0:3d9c67d97d6f | 234 | tapCnt = numTaps >> 2; |
emh203 | 0:3d9c67d97d6f | 235 | |
emh203 | 0:3d9c67d97d6f | 236 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 237 | { |
emh203 | 0:3d9c67d97d6f | 238 | /* Perform the multiply-accumulate */ |
emh203 | 0:3d9c67d97d6f | 239 | sum += (*px++) * (*pb++); |
emh203 | 0:3d9c67d97d6f | 240 | sum += (*px++) * (*pb++); |
emh203 | 0:3d9c67d97d6f | 241 | sum += (*px++) * (*pb++); |
emh203 | 0:3d9c67d97d6f | 242 | sum += (*px++) * (*pb++); |
emh203 | 0:3d9c67d97d6f | 243 | |
emh203 | 0:3d9c67d97d6f | 244 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 245 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 246 | } |
emh203 | 0:3d9c67d97d6f | 247 | |
emh203 | 0:3d9c67d97d6f | 248 | /* If the filter length is not a multiple of 4, compute the remaining filter taps */ |
emh203 | 0:3d9c67d97d6f | 249 | tapCnt = numTaps % 0x4u; |
emh203 | 0:3d9c67d97d6f | 250 | |
emh203 | 0:3d9c67d97d6f | 251 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 252 | { |
emh203 | 0:3d9c67d97d6f | 253 | /* Perform the multiply-accumulate */ |
emh203 | 0:3d9c67d97d6f | 254 | sum += (*px++) * (*pb++); |
emh203 | 0:3d9c67d97d6f | 255 | |
emh203 | 0:3d9c67d97d6f | 256 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 257 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 258 | } |
emh203 | 0:3d9c67d97d6f | 259 | |
emh203 | 0:3d9c67d97d6f | 260 | /* The result in the accumulator, store in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 261 | *pOut++ = sum; |
emh203 | 0:3d9c67d97d6f | 262 | |
emh203 | 0:3d9c67d97d6f | 263 | /* Compute and store error */ |
emh203 | 0:3d9c67d97d6f | 264 | d = (float32_t) (*pRef++); |
emh203 | 0:3d9c67d97d6f | 265 | e = d - sum; |
emh203 | 0:3d9c67d97d6f | 266 | *pErr++ = e; |
emh203 | 0:3d9c67d97d6f | 267 | |
emh203 | 0:3d9c67d97d6f | 268 | /* Calculation of Weighting factor for updating filter coefficients */ |
emh203 | 0:3d9c67d97d6f | 269 | /* epsilon value 0.000000119209289f */ |
emh203 | 0:3d9c67d97d6f | 270 | w = (e * mu) / (energy + 0.000000119209289f); |
emh203 | 0:3d9c67d97d6f | 271 | |
emh203 | 0:3d9c67d97d6f | 272 | /* Initialize pState pointer */ |
emh203 | 0:3d9c67d97d6f | 273 | px = pState; |
emh203 | 0:3d9c67d97d6f | 274 | |
emh203 | 0:3d9c67d97d6f | 275 | /* Initialize coeff pointer */ |
emh203 | 0:3d9c67d97d6f | 276 | pb = (pCoeffs); |
emh203 | 0:3d9c67d97d6f | 277 | |
emh203 | 0:3d9c67d97d6f | 278 | /* Loop unrolling. Process 4 taps at a time. */ |
emh203 | 0:3d9c67d97d6f | 279 | tapCnt = numTaps >> 2; |
emh203 | 0:3d9c67d97d6f | 280 | |
emh203 | 0:3d9c67d97d6f | 281 | /* Update filter coefficients */ |
emh203 | 0:3d9c67d97d6f | 282 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 283 | { |
emh203 | 0:3d9c67d97d6f | 284 | /* Perform the multiply-accumulate */ |
emh203 | 0:3d9c67d97d6f | 285 | *pb += w * (*px++); |
emh203 | 0:3d9c67d97d6f | 286 | pb++; |
emh203 | 0:3d9c67d97d6f | 287 | |
emh203 | 0:3d9c67d97d6f | 288 | *pb += w * (*px++); |
emh203 | 0:3d9c67d97d6f | 289 | pb++; |
emh203 | 0:3d9c67d97d6f | 290 | |
emh203 | 0:3d9c67d97d6f | 291 | *pb += w * (*px++); |
emh203 | 0:3d9c67d97d6f | 292 | pb++; |
emh203 | 0:3d9c67d97d6f | 293 | |
emh203 | 0:3d9c67d97d6f | 294 | *pb += w * (*px++); |
emh203 | 0:3d9c67d97d6f | 295 | pb++; |
emh203 | 0:3d9c67d97d6f | 296 | |
emh203 | 0:3d9c67d97d6f | 297 | |
emh203 | 0:3d9c67d97d6f | 298 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 299 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 300 | } |
emh203 | 0:3d9c67d97d6f | 301 | |
emh203 | 0:3d9c67d97d6f | 302 | /* If the filter length is not a multiple of 4, compute the remaining filter taps */ |
emh203 | 0:3d9c67d97d6f | 303 | tapCnt = numTaps % 0x4u; |
emh203 | 0:3d9c67d97d6f | 304 | |
emh203 | 0:3d9c67d97d6f | 305 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 306 | { |
emh203 | 0:3d9c67d97d6f | 307 | /* Perform the multiply-accumulate */ |
emh203 | 0:3d9c67d97d6f | 308 | *pb += w * (*px++); |
emh203 | 0:3d9c67d97d6f | 309 | pb++; |
emh203 | 0:3d9c67d97d6f | 310 | |
emh203 | 0:3d9c67d97d6f | 311 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 312 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 313 | } |
emh203 | 0:3d9c67d97d6f | 314 | |
emh203 | 0:3d9c67d97d6f | 315 | x0 = *pState; |
emh203 | 0:3d9c67d97d6f | 316 | |
emh203 | 0:3d9c67d97d6f | 317 | /* Advance state pointer by 1 for the next sample */ |
emh203 | 0:3d9c67d97d6f | 318 | pState = pState + 1; |
emh203 | 0:3d9c67d97d6f | 319 | |
emh203 | 0:3d9c67d97d6f | 320 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 321 | blkCnt--; |
emh203 | 0:3d9c67d97d6f | 322 | } |
emh203 | 0:3d9c67d97d6f | 323 | |
emh203 | 0:3d9c67d97d6f | 324 | S->energy = energy; |
emh203 | 0:3d9c67d97d6f | 325 | S->x0 = x0; |
emh203 | 0:3d9c67d97d6f | 326 | |
emh203 | 0:3d9c67d97d6f | 327 | /* Processing is complete. Now copy the last numTaps - 1 samples to the |
emh203 | 0:3d9c67d97d6f | 328 | satrt of the state buffer. This prepares the state buffer for the |
emh203 | 0:3d9c67d97d6f | 329 | next function call. */ |
emh203 | 0:3d9c67d97d6f | 330 | |
emh203 | 0:3d9c67d97d6f | 331 | /* Points to the start of the pState buffer */ |
emh203 | 0:3d9c67d97d6f | 332 | pStateCurnt = S->pState; |
emh203 | 0:3d9c67d97d6f | 333 | |
emh203 | 0:3d9c67d97d6f | 334 | /* Loop unrolling for (numTaps - 1u)/4 samples copy */ |
emh203 | 0:3d9c67d97d6f | 335 | tapCnt = (numTaps - 1u) >> 2u; |
emh203 | 0:3d9c67d97d6f | 336 | |
emh203 | 0:3d9c67d97d6f | 337 | /* copy data */ |
emh203 | 0:3d9c67d97d6f | 338 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 339 | { |
emh203 | 0:3d9c67d97d6f | 340 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 341 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 342 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 343 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 344 | |
emh203 | 0:3d9c67d97d6f | 345 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 346 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 347 | } |
emh203 | 0:3d9c67d97d6f | 348 | |
emh203 | 0:3d9c67d97d6f | 349 | /* Calculate remaining number of copies */ |
emh203 | 0:3d9c67d97d6f | 350 | tapCnt = (numTaps - 1u) % 0x4u; |
emh203 | 0:3d9c67d97d6f | 351 | |
emh203 | 0:3d9c67d97d6f | 352 | /* Copy the remaining q31_t data */ |
emh203 | 0:3d9c67d97d6f | 353 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 354 | { |
emh203 | 0:3d9c67d97d6f | 355 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 356 | |
emh203 | 0:3d9c67d97d6f | 357 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 358 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 359 | } |
emh203 | 0:3d9c67d97d6f | 360 | |
emh203 | 0:3d9c67d97d6f | 361 | #else |
emh203 | 0:3d9c67d97d6f | 362 | |
emh203 | 0:3d9c67d97d6f | 363 | /* Run the below code for Cortex-M0 */ |
emh203 | 0:3d9c67d97d6f | 364 | |
emh203 | 0:3d9c67d97d6f | 365 | while(blkCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 366 | { |
emh203 | 0:3d9c67d97d6f | 367 | /* Copy the new input sample into the state buffer */ |
emh203 | 0:3d9c67d97d6f | 368 | *pStateCurnt++ = *pSrc; |
emh203 | 0:3d9c67d97d6f | 369 | |
emh203 | 0:3d9c67d97d6f | 370 | /* Initialize pState pointer */ |
emh203 | 0:3d9c67d97d6f | 371 | px = pState; |
emh203 | 0:3d9c67d97d6f | 372 | |
emh203 | 0:3d9c67d97d6f | 373 | /* Initialize pCoeffs pointer */ |
emh203 | 0:3d9c67d97d6f | 374 | pb = pCoeffs; |
emh203 | 0:3d9c67d97d6f | 375 | |
emh203 | 0:3d9c67d97d6f | 376 | /* Read the sample from input buffer */ |
emh203 | 0:3d9c67d97d6f | 377 | in = *pSrc++; |
emh203 | 0:3d9c67d97d6f | 378 | |
emh203 | 0:3d9c67d97d6f | 379 | /* Update the energy calculation */ |
emh203 | 0:3d9c67d97d6f | 380 | energy -= x0 * x0; |
emh203 | 0:3d9c67d97d6f | 381 | energy += in * in; |
emh203 | 0:3d9c67d97d6f | 382 | |
emh203 | 0:3d9c67d97d6f | 383 | /* Set the accumulator to zero */ |
emh203 | 0:3d9c67d97d6f | 384 | sum = 0.0f; |
emh203 | 0:3d9c67d97d6f | 385 | |
emh203 | 0:3d9c67d97d6f | 386 | /* Loop over numTaps number of values */ |
emh203 | 0:3d9c67d97d6f | 387 | tapCnt = numTaps; |
emh203 | 0:3d9c67d97d6f | 388 | |
emh203 | 0:3d9c67d97d6f | 389 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 390 | { |
emh203 | 0:3d9c67d97d6f | 391 | /* Perform the multiply-accumulate */ |
emh203 | 0:3d9c67d97d6f | 392 | sum += (*px++) * (*pb++); |
emh203 | 0:3d9c67d97d6f | 393 | |
emh203 | 0:3d9c67d97d6f | 394 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 395 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 396 | } |
emh203 | 0:3d9c67d97d6f | 397 | |
emh203 | 0:3d9c67d97d6f | 398 | /* The result in the accumulator is stored in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 399 | *pOut++ = sum; |
emh203 | 0:3d9c67d97d6f | 400 | |
emh203 | 0:3d9c67d97d6f | 401 | /* Compute and store error */ |
emh203 | 0:3d9c67d97d6f | 402 | d = (float32_t) (*pRef++); |
emh203 | 0:3d9c67d97d6f | 403 | e = d - sum; |
emh203 | 0:3d9c67d97d6f | 404 | *pErr++ = e; |
emh203 | 0:3d9c67d97d6f | 405 | |
emh203 | 0:3d9c67d97d6f | 406 | /* Calculation of Weighting factor for updating filter coefficients */ |
emh203 | 0:3d9c67d97d6f | 407 | /* epsilon value 0.000000119209289f */ |
emh203 | 0:3d9c67d97d6f | 408 | w = (e * mu) / (energy + 0.000000119209289f); |
emh203 | 0:3d9c67d97d6f | 409 | |
emh203 | 0:3d9c67d97d6f | 410 | /* Initialize pState pointer */ |
emh203 | 0:3d9c67d97d6f | 411 | px = pState; |
emh203 | 0:3d9c67d97d6f | 412 | |
emh203 | 0:3d9c67d97d6f | 413 | /* Initialize pCcoeffs pointer */ |
emh203 | 0:3d9c67d97d6f | 414 | pb = pCoeffs; |
emh203 | 0:3d9c67d97d6f | 415 | |
emh203 | 0:3d9c67d97d6f | 416 | /* Loop over numTaps number of values */ |
emh203 | 0:3d9c67d97d6f | 417 | tapCnt = numTaps; |
emh203 | 0:3d9c67d97d6f | 418 | |
emh203 | 0:3d9c67d97d6f | 419 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 420 | { |
emh203 | 0:3d9c67d97d6f | 421 | /* Perform the multiply-accumulate */ |
emh203 | 0:3d9c67d97d6f | 422 | *pb += w * (*px++); |
emh203 | 0:3d9c67d97d6f | 423 | pb++; |
emh203 | 0:3d9c67d97d6f | 424 | |
emh203 | 0:3d9c67d97d6f | 425 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 426 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 427 | } |
emh203 | 0:3d9c67d97d6f | 428 | |
emh203 | 0:3d9c67d97d6f | 429 | x0 = *pState; |
emh203 | 0:3d9c67d97d6f | 430 | |
emh203 | 0:3d9c67d97d6f | 431 | /* Advance state pointer by 1 for the next sample */ |
emh203 | 0:3d9c67d97d6f | 432 | pState = pState + 1; |
emh203 | 0:3d9c67d97d6f | 433 | |
emh203 | 0:3d9c67d97d6f | 434 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 435 | blkCnt--; |
emh203 | 0:3d9c67d97d6f | 436 | } |
emh203 | 0:3d9c67d97d6f | 437 | |
emh203 | 0:3d9c67d97d6f | 438 | S->energy = energy; |
emh203 | 0:3d9c67d97d6f | 439 | S->x0 = x0; |
emh203 | 0:3d9c67d97d6f | 440 | |
emh203 | 0:3d9c67d97d6f | 441 | /* Processing is complete. Now copy the last numTaps - 1 samples to the |
emh203 | 0:3d9c67d97d6f | 442 | satrt of the state buffer. This prepares the state buffer for the |
emh203 | 0:3d9c67d97d6f | 443 | next function call. */ |
emh203 | 0:3d9c67d97d6f | 444 | |
emh203 | 0:3d9c67d97d6f | 445 | /* Points to the start of the pState buffer */ |
emh203 | 0:3d9c67d97d6f | 446 | pStateCurnt = S->pState; |
emh203 | 0:3d9c67d97d6f | 447 | |
emh203 | 0:3d9c67d97d6f | 448 | /* Copy (numTaps - 1u) samples */ |
emh203 | 0:3d9c67d97d6f | 449 | tapCnt = (numTaps - 1u); |
emh203 | 0:3d9c67d97d6f | 450 | |
emh203 | 0:3d9c67d97d6f | 451 | /* Copy the remaining q31_t data */ |
emh203 | 0:3d9c67d97d6f | 452 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 453 | { |
emh203 | 0:3d9c67d97d6f | 454 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 455 | |
emh203 | 0:3d9c67d97d6f | 456 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 457 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 458 | } |
emh203 | 0:3d9c67d97d6f | 459 | |
emh203 | 0:3d9c67d97d6f | 460 | #endif /* #ifndef ARM_MATH_CM0_FAMILY */ |
emh203 | 0:3d9c67d97d6f | 461 | |
emh203 | 0:3d9c67d97d6f | 462 | } |
emh203 | 0:3d9c67d97d6f | 463 | |
emh203 | 0:3d9c67d97d6f | 464 | /** |
emh203 | 0:3d9c67d97d6f | 465 | * @} end of LMS_NORM group |
emh203 | 0:3d9c67d97d6f | 466 | */ |