arm_lms_q31.c

00001 /* ----------------------------------------------------------------------  
00002 * Copyright (C) 2010 ARM Limited. All rights reserved.  
00003 *  
00004 * $Date:        29. November 2010  
00005 * $Revision:    V1.0.3  
00006 *  
00007 * Project:      CMSIS DSP Library  
00008 * Title:        arm_lms_q31.c  
00009 *  
00010 * Description:  Processing function for the Q31 LMS filter.  
00011 *  
00012 * Target Processor: Cortex-M4/Cortex-M3
00013 *  
00014 * Version 1.0.3 2010/11/29 
00015 *    Re-organized the CMSIS folders and updated documentation.  
00016 *   
00017 * Version 1.0.2 2010/11/11  
00018 *    Documentation updated.   
00019 *  
00020 * Version 1.0.1 2010/10/05   
00021 *    Production release and review comments incorporated.  
00022 *  
00023 * Version 1.0.0 2010/09/20   
00024 *    Production release and review comments incorporated  
00025 *  
00026 * Version 0.0.7  2010/06/10   
00027 *    Misra-C changes done  
00028 * -------------------------------------------------------------------- */ 
00029  
00030 #include "arm_math.h" 
00031 /**  
00032  * @ingroup groupFilters  
00033  */ 
00034  
00035 /**  
00036  * @addtogroup LMS  
00037  * @{  
00038  */ 
00039  
00040  /**  
00041  * @brief Processing function for Q31 LMS filter.  
00042  * @param[in]  *S points to an instance of the Q15 LMS filter structure.  
00043  * @param[in]  *pSrc points to the block of input data.  
00044  * @param[in]  *pRef points to the block of reference data.  
00045  * @param[out] *pOut points to the block of output data.  
00046  * @param[out] *pErr points to the block of error data.  
00047  * @param[in]  blockSize number of samples to process.  
00048  * @return     none.  
00049  *  
00050  * \par Scaling and Overflow Behavior:  
00051  * The function is implemented using an internal 64-bit accumulator.  
00052  * The accumulator has a 2.62 format and maintains full precision of the intermediate multiplication results but provides only a single guard bit.  
00053  * Thus, if the accumulator result overflows it wraps around rather than clips.  
00054  * In order to avoid overflows completely the input signal must be scaled down by log2(numTaps) bits.  
00055  * The reference signal should not be scaled down.  
00056  * After all multiply-accumulates are performed, the 2.62 accumulator is truncated to 1.32 format and then saturated to 1.31 format.  
00057  * The output signal and error signal are in 1.31 format.  
00058  * 
00059  * \par 
00060  *  In this filter, filter coefficients are updated for each sample and the updation of filter cofficients are saturted. 
00061  */ 
00062  
00063 void arm_lms_q31( 
00064   const arm_lms_instance_q31 * S, 
00065   q31_t * pSrc, 
00066   q31_t * pRef, 
00067   q31_t * pOut, 
00068   q31_t * pErr, 
00069   uint32_t blockSize) 
00070 { 
00071   q31_t *pState = S->pState;                     /* State pointer */ 
00072   uint32_t numTaps = S->numTaps;                 /* Number of filter coefficients in the filter */ 
00073   q31_t *pCoeffs = S->pCoeffs;                   /* Coefficient pointer */ 
00074   q31_t *pStateCurnt;                            /* Points to the current sample of the state */ 
00075   q31_t mu = S->mu;                              /* Adaptive factor */ 
00076   q31_t *px;                                     /* Temporary pointer for state */ 
00077   q31_t *pb;                                     /* Temporary pointer for coefficient buffer */ 
00078   uint32_t tapCnt, blkCnt;                       /* Loop counters */ 
00079   q63_t acc;                                     /* Accumulator */ 
00080   q31_t e = 0;                                   /* error of data sample */ 
00081   q31_t alpha;                                   /* Intermediate constant for taps update */ 
00082   uint8_t shift = (uint8_t) (32u - (S->postShift + 1u));        /* Shift to be applied to the output */ 
00083   q31_t coef;                                    /* Temporary variable for coef */ 
00084  
00085   /* S->pState points to buffer which contains previous frame (numTaps - 1) samples */ 
00086   /* pStateCurnt points to the location where the new input data should be written */ 
00087   pStateCurnt = &(S->pState[(numTaps - 1u)]); 
00088  
00089   /* Initializing blkCnt with blockSize */ 
00090   blkCnt = blockSize; 
00091  
00092   while(blkCnt > 0u) 
00093   { 
00094     /* Copy the new input sample into the state buffer */ 
00095     *pStateCurnt++ = *pSrc++; 
00096  
00097     /* Initialize state pointer */ 
00098     px = pState; 
00099  
00100     /* Initialize coefficient pointer */ 
00101     pb = pCoeffs; 
00102  
00103     /* Set the accumulator to zero */ 
00104     acc = 0; 
00105  
00106     /* Loop unrolling.  Process 4 taps at a time. */ 
00107     tapCnt = numTaps >> 2; 
00108  
00109     while(tapCnt > 0u) 
00110     { 
00111       /* Perform the multiply-accumulate */ 
00112       /* acc +=  b[N] * x[n-N] */ 
00113       acc += ((q63_t) (*px++)) * (*pb++); 
00114  
00115       /* acc +=  b[N-1] * x[n-N-1] */ 
00116       acc += ((q63_t) (*px++)) * (*pb++); 
00117  
00118       /* acc +=  b[N-2] * x[n-N-2] */ 
00119       acc += ((q63_t) (*px++)) * (*pb++); 
00120  
00121       /* acc +=  b[N-3] * x[n-N-3] */ 
00122       acc += ((q63_t) (*px++)) * (*pb++); 
00123  
00124       /* Decrement the loop counter */ 
00125       tapCnt--; 
00126     } 
00127  
00128     /* If the filter length is not a multiple of 4, compute the remaining filter taps */ 
00129     tapCnt = numTaps % 0x4u; 
00130  
00131     while(tapCnt > 0u) 
00132     { 
00133       /* Perform the multiply-accumulate */ 
00134       acc += ((q63_t) (*px++)) * (*pb++); 
00135  
00136       /* Decrement the loop counter */ 
00137       tapCnt--; 
00138     } 
00139  
00140     /* Converting the result to 1.31 format */ 
00141     /* Store the result from accumulator into the destination buffer. */ 
00142     acc = (q31_t) (acc >> shift); 
00143  
00144     *pOut++ = (q31_t) acc; 
00145  
00146     /* Compute and store error */ 
00147     e = *pRef++ - (q31_t) acc; 
00148  
00149     *pErr++ = (q31_t) e; 
00150  
00151     /* Compute alpha i.e. intermediate constant for taps update */ 
00152     alpha = (q31_t) (((q63_t) e * mu) >> 31); 
00153  
00154     /* Initialize state pointer */ 
00155     /* Advance state pointer by 1 for the next sample */ 
00156     px = pState++; 
00157  
00158     /* Initialize coefficient pointer */ 
00159     pb = pCoeffs; 
00160  
00161     /* Loop unrolling.  Process 4 taps at a time. */ 
00162     tapCnt = numTaps >> 2; 
00163  
00164     /* Update filter coefficients */ 
00165     while(tapCnt > 0u) 
00166     { 
00167       /* coef is in 2.30 format */ 
00168       coef = (q31_t) (((q63_t) alpha * (*px++)) >> (32)); 
00169       /* get coef in 1.31 format by left shifting */ 
00170       *pb = clip_q63_to_q31((q63_t) *pb + (coef << 1u)); 
00171       /* update coefficient buffer to next coefficient */ 
00172       pb++; 
00173  
00174       coef = (q31_t) (((q63_t) alpha * (*px++)) >> (32)); 
00175       *pb = clip_q63_to_q31((q63_t) *pb + (coef << 1u)); 
00176       pb++; 
00177  
00178       coef = (q31_t) (((q63_t) alpha * (*px++)) >> (32)); 
00179       *pb = clip_q63_to_q31((q63_t) *pb + (coef << 1u)); 
00180       pb++; 
00181  
00182       coef = (q31_t) (((q63_t) alpha * (*px++)) >> (32)); 
00183       *pb = clip_q63_to_q31((q63_t) *pb + (coef << 1u)); 
00184       pb++; 
00185  
00186       /* Decrement the loop counter */ 
00187       tapCnt--; 
00188     } 
00189  
00190     /* If the filter length is not a multiple of 4, compute the remaining filter taps */ 
00191     tapCnt = numTaps % 0x4u; 
00192  
00193     while(tapCnt > 0u) 
00194     { 
00195       /* Perform the multiply-accumulate */ 
00196       coef = (q31_t) (((q63_t) alpha * (*px++)) >> (32)); 
00197       *pb = clip_q63_to_q31((q63_t) *pb + (coef << 1u)); 
00198       pb++; 
00199  
00200       /* Decrement the loop counter */ 
00201       tapCnt--; 
00202     } 
00203  
00204     /* Decrement the loop counter */ 
00205     blkCnt--; 
00206   } 
00207  
00208   /* Processing is complete. Now copy the last numTaps - 1 samples to the  
00209      satrt of the state buffer. This prepares the state buffer for the  
00210      next function call. */ 
00211  
00212   /* Points to the start of the pState buffer */ 
00213   pStateCurnt = S->pState; 
00214  
00215   /* Loop unrolling for (numTaps - 1u) samples copy */ 
00216   tapCnt = (numTaps - 1u) >> 2u; 
00217  
00218   /* copy data */ 
00219   while(tapCnt > 0u) 
00220   { 
00221     *pStateCurnt++ = *pState++; 
00222     *pStateCurnt++ = *pState++; 
00223     *pStateCurnt++ = *pState++; 
00224     *pStateCurnt++ = *pState++; 
00225  
00226     /* Decrement the loop counter */ 
00227     tapCnt--; 
00228   } 
00229  
00230   /* Calculate remaining number of copies */ 
00231   tapCnt = (numTaps - 1u) % 0x4u; 
00232  
00233   /* Copy the remaining q31_t data */ 
00234   while(tapCnt > 0u) 
00235   { 
00236     *pStateCurnt++ = *pState++; 
00237  
00238     /* Decrement the loop counter */ 
00239     tapCnt--; 
00240   } 
00241  
00242 } 
00243  
00244 /**  
00245    * @} end of LMS group  
00246    */