arm_biquad_cascade_df1_fast_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_biquad_cascade_df1_fast_q31.c  
00009 *  
00010 * Description:  Processing function for the  
00011 *               Q31 Fast Biquad cascade DirectFormI(DF1) filter.  
00012 *  
00013 * Target Processor: Cortex-M4/Cortex-M3
00014 *  
00015 * Version 1.0.3 2010/11/29 
00016 *    Re-organized the CMSIS folders and updated documentation.  
00017 *   
00018 * Version 1.0.2 2010/11/11  
00019 *    Documentation updated.   
00020 *  
00021 * Version 1.0.1 2010/10/05   
00022 *    Production release and review comments incorporated.  
00023 *  
00024 * Version 1.0.0 2010/09/20   
00025 *    Production release and review comments incorporated.  
00026 *  
00027 * Version 0.0.9  2010/08/27   
00028 *    Initial version  
00029 *  
00030 * -------------------------------------------------------------------- */ 
00031  
00032 #include "arm_math.h" 
00033  
00034 /**  
00035  * @ingroup groupFilters  
00036  */ 
00037  
00038 /**  
00039  * @addtogroup BiquadCascadeDF1  
00040  * @{  
00041  */ 
00042  
00043 /**  
00044  * @details  
00045  *  
00046  * @param[in]  *S        points to an instance of the Q31 Biquad cascade structure.  
00047  * @param[in]  *pSrc     points to the block of input data.  
00048  * @param[out] *pDst     points to the block of output data.  
00049  * @param[in]  blockSize number of samples to process per call.  
00050  * @return     none.  
00051  *  
00052  * <b>Scaling and Overflow Behavior:</b>  
00053  * \par  
00054  * This function is optimized for speed at the expense of fixed-point precision and overflow protection.  
00055  * The result of each 1.31 x 1.31 multiplication is truncated to 2.30 format.  
00056  * These intermediate results are added to a 2.30 accumulator.  
00057  * Finally, the accumulator is saturated and converted to a 1.31 result.  
00058  * The fast version has the same overflow behavior as the standard version and provides less precision since it discards the low 32 bits of each multiplication result.  
00059  * In order to avoid overflows completely the input signal must be scaled down by two bits and lie in the range [-0.25 +0.25). Use the intialization function  
00060  * arm_biquad_cascade_df1_init_q31() to initialize filter structure.  
00061  *  
00062  * \par  
00063  * Refer to the function <code>arm_biquad_cascade_df1_q31()</code> for a slower implementation of this function which uses 64-bit accumulation to provide higher precision.  Both the slow and the fast versions use the same instance structure.  
00064  * Use the function <code>arm_biquad_cascade_df1_init_q31()</code> to initialize the filter structure.  
00065  */ 
00066  
00067 void arm_biquad_cascade_df1_fast_q31( 
00068   const arm_biquad_casd_df1_inst_q31 * S, 
00069   q31_t * pSrc, 
00070   q31_t * pDst, 
00071   uint32_t blockSize) 
00072 { 
00073   q31_t *pIn = pSrc;                             /*  input pointer initialization  */ 
00074   q31_t *pOut = pDst;                            /*  output pointer initialization */ 
00075   q31_t *pState = S->pState;                     /*  pState pointer initialization */ 
00076   q31_t *pCoeffs = S->pCoeffs;                   /*  coeff pointer initialization  */ 
00077   q31_t acc;                                     /*  accumulator                   */ 
00078   q31_t Xn1, Xn2, Yn1, Yn2;                      /*  Filter state variables        */ 
00079   q31_t b0, b1, b2, a1, a2;                      /*  Filter coefficients           */ 
00080   q31_t Xn;                                      /*  temporary input               */ 
00081   int32_t shift = (int32_t) S->postShift + 1;    /*  Shift to be applied to the output */ 
00082   uint32_t sample, stage = S->numStages;         /*  loop counters                     */ 
00083  
00084  
00085   do 
00086   { 
00087     /* Reading the coefficients */ 
00088     b0 = *pCoeffs++; 
00089     b1 = *pCoeffs++; 
00090     b2 = *pCoeffs++; 
00091     a1 = *pCoeffs++; 
00092     a2 = *pCoeffs++; 
00093  
00094     /* Reading the state values */ 
00095     Xn1 = pState[0]; 
00096     Xn2 = pState[1]; 
00097     Yn1 = pState[2]; 
00098     Yn2 = pState[3]; 
00099  
00100     /* Apply loop unrolling and compute 4 output values simultaneously. */ 
00101     /*      The variables acc ... acc3 hold output values that are being computed:  
00102      *  
00103      *    acc =  b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2]  
00104      */ 
00105  
00106     sample = blockSize >> 2u; 
00107  
00108     /* First part of the processing with loop unrolling.  Compute 4 outputs at a time.  
00109      ** a second loop below computes the remaining 1 to 3 samples. */ 
00110     while(sample > 0u) 
00111     { 
00112       /* Read the input */ 
00113       Xn = *pIn++; 
00114  
00115       /* acc =  b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] */ 
00116       /* acc =  b0 * x[n] */ 
00117       acc = (q31_t) (((q63_t) b0 * Xn) >> 32); 
00118       /* acc +=  b1 * x[n-1] */ 
00119       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b1 * (Xn1))) >> 32); 
00120       /* acc +=  b[2] * x[n-2] */ 
00121       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b2 * (Xn2))) >> 32); 
00122       /* acc +=  a1 * y[n-1] */ 
00123       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a1 * (Yn1))) >> 32); 
00124       /* acc +=  a2 * y[n-2] */ 
00125       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a2 * (Yn2))) >> 32); 
00126  
00127       /* The result is converted to 1.31 , Yn2 variable is reused */ 
00128       Yn2 = acc << shift; 
00129  
00130       /* Store the output in the destination buffer. */ 
00131       *pOut++ = Yn2; 
00132  
00133       /* Read the second input */ 
00134       Xn2 = *pIn++; 
00135  
00136       /* acc =  b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] */ 
00137       /* acc =  b0 * x[n] */ 
00138       acc = (q31_t) (((q63_t) b0 * (Xn2)) >> 32); 
00139       /* acc +=  b1 * x[n-1] */ 
00140       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b1 * (Xn))) >> 32); 
00141       /* acc +=  b[2] * x[n-2] */ 
00142       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b2 * (Xn1))) >> 32); 
00143       /* acc +=  a1 * y[n-1] */ 
00144       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a1 * (Yn2))) >> 32); 
00145       /* acc +=  a2 * y[n-2] */ 
00146       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a2 * (Yn1))) >> 32); 
00147  
00148       /* The result is converted to 1.31, Yn1 variable is reused  */ 
00149       Yn1 = acc << shift; 
00150  
00151       /* Store the output in the destination buffer. */ 
00152       *pOut++ = Yn1; 
00153  
00154       /* Read the third input  */ 
00155       Xn1 = *pIn++; 
00156  
00157       /* acc =  b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] */ 
00158       /* acc =  b0 * x[n] */ 
00159       acc = (q31_t) (((q63_t) b0 * (Xn1)) >> 32); 
00160       /* acc +=  b1 * x[n-1] */ 
00161       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b1 * (Xn2))) >> 32); 
00162       /* acc +=  b[2] * x[n-2] */ 
00163       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b2 * (Xn))) >> 32); 
00164       /* acc +=  a1 * y[n-1] */ 
00165       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a1 * (Yn1))) >> 32); 
00166       /* acc +=  a2 * y[n-2] */ 
00167       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a2 * (Yn2))) >> 32); 
00168  
00169       /* The result is converted to 1.31, Yn2 variable is reused  */ 
00170       Yn2 = acc << shift; 
00171  
00172       /* Store the output in the destination buffer. */ 
00173       *pOut++ = Yn2; 
00174  
00175       /* Read the forth input */ 
00176       Xn = *pIn++; 
00177  
00178       /* acc =  b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] */ 
00179       /* acc =  b0 * x[n] */ 
00180       acc = (q31_t) (((q63_t) b0 * (Xn)) >> 32); 
00181       /* acc +=  b1 * x[n-1] */ 
00182       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b1 * (Xn1))) >> 32); 
00183       /* acc +=  b[2] * x[n-2] */ 
00184       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b2 * (Xn2))) >> 32); 
00185       /* acc +=  a1 * y[n-1] */ 
00186       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a1 * (Yn2))) >> 32); 
00187       /* acc +=  a2 * y[n-2] */ 
00188       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a2 * (Yn1))) >> 32); 
00189  
00190       /* The result is converted to 1.31, Yn1 variable is reused  */ 
00191       Yn1 = acc << shift; 
00192  
00193       /* Every time after the output is computed state should be updated. */ 
00194       /* The states should be updated as:  */ 
00195       /* Xn2 = Xn1    */ 
00196       /* Xn1 = Xn     */ 
00197       /* Yn2 = Yn1    */ 
00198       /* Yn1 = acc    */ 
00199       Xn2 = Xn1; 
00200       Xn1 = Xn; 
00201  
00202       /* Store the output in the destination buffer. */ 
00203       *pOut++ = Yn1; 
00204  
00205       /* decrement the loop counter */ 
00206       sample--; 
00207     } 
00208  
00209     /* If the blockSize is not a multiple of 4, compute any remaining output samples here.  
00210      ** No loop unrolling is used. */ 
00211     sample = (blockSize & 0x3u); 
00212  
00213     while(sample > 0u) 
00214     { 
00215       /* Read the input */ 
00216       Xn = *pIn++; 
00217  
00218       /* acc =  b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] */ 
00219       /* acc =  b0 * x[n] */ 
00220       acc = (q31_t) (((q63_t) b0 * (Xn)) >> 32); 
00221       /* acc +=  b1 * x[n-1] */ 
00222       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b1 * (Xn1))) >> 32); 
00223       /* acc +=  b[2] * x[n-2] */ 
00224       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b2 * (Xn2))) >> 32); 
00225       /* acc +=  a1 * y[n-1] */ 
00226       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a1 * (Yn1))) >> 32); 
00227       /* acc +=  a2 * y[n-2] */ 
00228       acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a2 * (Yn2))) >> 32); 
00229       /* The result is converted to 1.31  */ 
00230       acc = acc << shift; 
00231  
00232       /* Every time after the output is computed state should be updated. */ 
00233       /* The states should be updated as:  */ 
00234       /* Xn2 = Xn1    */ 
00235       /* Xn1 = Xn     */ 
00236       /* Yn2 = Yn1    */ 
00237       /* Yn1 = acc    */ 
00238       Xn2 = Xn1; 
00239       Xn1 = Xn; 
00240       Yn2 = Yn1; 
00241       Yn1 = acc; 
00242  
00243       /* Store the output in the destination buffer. */ 
00244       *pOut++ = acc; 
00245  
00246       /* decrement the loop counter */ 
00247       sample--; 
00248     } 
00249  
00250     /*  The first stage goes from the input buffer to the output buffer. */ 
00251     /*  Subsequent stages occur in-place in the output buffer */ 
00252     pIn = pDst; 
00253  
00254     /* Reset to destination pointer */ 
00255     pOut = pDst; 
00256  
00257     /*  Store the updated state variables back into the pState array */ 
00258     *pState++ = Xn1; 
00259     *pState++ = Xn2; 
00260     *pState++ = Yn1; 
00261     *pState++ = Yn2; 
00262  
00263   } while(--stage); 
00264 } 
00265  
00266 /**  
00267   * @} end of BiquadCascadeDF1 group  
00268   */