diff --git a/Src/arm_conv_f32.c b/Src/arm_conv_f32.c new file mode 100644 index 0000000..1fdc6a1 --- /dev/null +++ b/Src/arm_conv_f32.c @@ -0,0 +1,647 @@ +/* ---------------------------------------------------------------------------- +* Copyright (C) 2010-2014 ARM Limited. All rights reserved. +* +* $Date: 19. March 2015 +* $Revision: V.1.4.5 +* +* Project: CMSIS DSP Library +* Title: arm_conv_f32.c +* +* Description: Convolution of floating-point sequences. +* +* Target Processor: Cortex-M4/Cortex-M3/Cortex-M0 +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions +* are met: +* - Redistributions of source code must retain the above copyright +* notice, this list of conditions and the following disclaimer. +* - Redistributions in binary form must reproduce the above copyright +* notice, this list of conditions and the following disclaimer in +* the documentation and/or other materials provided with the +* distribution. +* - Neither the name of ARM LIMITED nor the names of its contributors +* may be used to endorse or promote products derived from this +* software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS +* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE +* COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, +* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, +* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; +* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT +* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN +* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +* POSSIBILITY OF SUCH DAMAGE. +* -------------------------------------------------------------------------- */ + +#include "arm_math.h" + +/** + * @ingroup groupFilters + */ + +/** + * @defgroup Conv Convolution + * + * Convolution is a mathematical operation that operates on two finite length vectors to generate a finite length output vector. + * Convolution is similar to correlation and is frequently used in filtering and data analysis. + * The CMSIS DSP library contains functions for convolving Q7, Q15, Q31, and floating-point data types. + * The library also provides fast versions of the Q15 and Q31 functions on Cortex-M4 and Cortex-M3. + * + * \par Algorithm + * Let a[n] and b[n] be sequences of length srcALen and srcBLen samples respectively. + * Then the convolution + * + * 
    
+ *                   c[n] = a[n] * b[n]    
+ *
+ * + * \par + * is defined as + * \image html ConvolutionEquation.gif + * \par + * Note that c[n] is of length srcALen + srcBLen - 1 and is defined over the interval n=0, 1, 2, ..., srcALen + srcBLen - 2. + * pSrcA points to the first input vector of length srcALen and + * pSrcB points to the second input vector of length srcBLen. + * The output result is written to pDst and the calling function must allocate srcALen+srcBLen-1 words for the result. + * + * \par + * Conceptually, when two signals a[n] and b[n] are convolved, + * the signal b[n] slides over a[n]. + * For each offset \c n, the overlapping portions of a[n] and b[n] are multiplied and summed together. + * + * \par + * Note that convolution is a commutative operation: + * + * 
    
+ *                   a[n] * b[n] = b[n] * a[n].    
+ *
+ * + * \par + * This means that switching the A and B arguments to the convolution functions has no effect. + * + * Fixed-Point Behavior + * + * \par + * Convolution requires summing up a large number of intermediate products. + * As such, the Q7, Q15, and Q31 functions run a risk of overflow and saturation. + * Refer to the function specific documentation below for further details of the particular algorithm used. + * + * + * Fast Versions + * + * \par + * Fast versions are supported for Q31 and Q15. Cycles for Fast versions are less compared to Q31 and Q15 of conv and the design requires + * the input signals should be scaled down to avoid intermediate overflows. + * + * + * Opt Versions + * + * \par + * Opt versions are supported for Q15 and Q7. Design uses internal scratch buffer for getting good optimisation. + * These versions are optimised in cycles and consumes more memory(Scratch memory) compared to Q15 and Q7 versions + */ + +/** + * @addtogroup Conv + * @{ + */ + +/** + * @brief Convolution of floating-point sequences. + * @param[in] *pSrcA points to the first input sequence. + * @param[in] srcALen length of the first input sequence. + * @param[in] *pSrcB points to the second input sequence. + * @param[in] srcBLen length of the second input sequence. + * @param[out] *pDst points to the location where the output result is written. Length srcALen+srcBLen-1. + * @return none. + */ + +void arm_conv_f32( + float32_t * pSrcA, + uint32_t srcALen, + float32_t * pSrcB, + uint32_t srcBLen, + float32_t * pDst) +{ + + +#ifndef ARM_MATH_CM0_FAMILY + + /* Run the below code for Cortex-M4 and Cortex-M3 */ + + float32_t *pIn1; /* inputA pointer */ + float32_t *pIn2; /* inputB pointer */ + float32_t *pOut = pDst; /* output pointer */ + float32_t *px; /* Intermediate inputA pointer */ + float32_t *py; /* Intermediate inputB pointer */ + float32_t *pSrc1, *pSrc2; /* Intermediate pointers */ + float32_t sum, acc0, acc1, acc2, acc3; /* Accumulator */ + float32_t x0, x1, x2, x3, c0; /* Temporary variables to hold state and coefficient values */ + uint32_t j, k, count, blkCnt, blockSize1, blockSize2, blockSize3; /* loop counters */ + + /* The algorithm implementation is based on the lengths of the inputs. */ + /* srcB is always made to slide across srcA. */ + /* So srcBLen is always considered as shorter or equal to srcALen */ + if(srcALen >= srcBLen) + { + /* Initialization of inputA pointer */ + pIn1 = pSrcA; + + /* Initialization of inputB pointer */ + pIn2 = pSrcB; + } + else + { + /* Initialization of inputA pointer */ + pIn1 = pSrcB; + + /* Initialization of inputB pointer */ + pIn2 = pSrcA; + + /* srcBLen is always considered as shorter or equal to srcALen */ + j = srcBLen; + srcBLen = srcALen; + srcALen = j; + } + + /* conv(x,y) at n = x[n] * y[0] + x[n-1] * y[1] + x[n-2] * y[2] + ...+ x[n-N+1] * y[N -1] */ + /* The function is internally + * divided into three stages according to the number of multiplications that has to be + * taken place between inputA samples and inputB samples. In the first stage of the + * algorithm, the multiplications increase by one for every iteration. + * In the second stage of the algorithm, srcBLen number of multiplications are done. + * In the third stage of the algorithm, the multiplications decrease by one + * for every iteration. */ + + /* The algorithm is implemented in three stages. + The loop counters of each stage is initiated here. */ + blockSize1 = srcBLen - 1u; + blockSize2 = srcALen - (srcBLen - 1u); + blockSize3 = blockSize1; + + /* -------------------------- + * initializations of stage1 + * -------------------------*/ + + /* sum = x[0] * y[0] + * sum = x[0] * y[1] + x[1] * y[0] + * .... + * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0] + */ + + /* In this stage the MAC operations are increased by 1 for every iteration. + The count variable holds the number of MAC operations performed */ + count = 1u; + + /* Working pointer of inputA */ + px = pIn1; + + /* Working pointer of inputB */ + py = pIn2; + + + /* ------------------------ + * Stage1 process + * ----------------------*/ + + /* The first stage starts here */ + while(blockSize1 > 0u) + { + /* Accumulator is made zero for every iteration */ + sum = 0.0f; + + /* Apply loop unrolling and compute 4 MACs simultaneously. */ + k = count >> 2u; + + /* First part of the processing with loop unrolling. Compute 4 MACs at a time. + ** a second loop below computes MACs for the remaining 1 to 3 samples. */ + while(k > 0u) + { + /* x[0] * y[srcBLen - 1] */ + sum += *px++ * *py--; + + /* x[1] * y[srcBLen - 2] */ + sum += *px++ * *py--; + + /* x[2] * y[srcBLen - 3] */ + sum += *px++ * *py--; + + /* x[3] * y[srcBLen - 4] */ + sum += *px++ * *py--; + + /* Decrement the loop counter */ + k--; + } + + /* If the count is not a multiple of 4, compute any remaining MACs here. + ** No loop unrolling is used. */ + k = count % 0x4u; + + while(k > 0u) + { + /* Perform the multiply-accumulate */ + sum += *px++ * *py--; + + /* Decrement the loop counter */ + k--; + } + + /* Store the result in the accumulator in the destination buffer. */ + *pOut++ = sum; + + /* Update the inputA and inputB pointers for next MAC calculation */ + py = pIn2 + count; + px = pIn1; + + /* Increment the MAC count */ + count++; + + /* Decrement the loop counter */ + blockSize1--; + } + + /* -------------------------- + * Initializations of stage2 + * ------------------------*/ + + /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0] + * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0] + * .... + * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0] + */ + + /* Working pointer of inputA */ + px = pIn1; + + /* Working pointer of inputB */ + pSrc2 = pIn2 + (srcBLen - 1u); + py = pSrc2; + + /* count is index by which the pointer pIn1 to be incremented */ + count = 0u; + + /* ------------------- + * Stage2 process + * ------------------*/ + + /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed. + * So, to loop unroll over blockSize2, + * srcBLen should be greater than or equal to 4 */ + if(srcBLen >= 4u) + { + /* Loop unroll over blockSize2, by 4 */ + blkCnt = blockSize2 >> 2u; + + while(blkCnt > 0u) + { + /* Set all accumulators to zero */ + acc0 = 0.0f; + acc1 = 0.0f; + acc2 = 0.0f; + acc3 = 0.0f; + + /* read x[0], x[1], x[2] samples */ + x0 = *(px++); + x1 = *(px++); + x2 = *(px++); + + /* Apply loop unrolling and compute 4 MACs simultaneously. */ + k = srcBLen >> 2u; + + /* First part of the processing with loop unrolling. Compute 4 MACs at a time. + ** a second loop below computes MACs for the remaining 1 to 3 samples. */ + do + { + /* Read y[srcBLen - 1] sample */ + c0 = *(py--); + + /* Read x[3] sample */ + x3 = *(px); + + /* Perform the multiply-accumulate */ + /* acc0 += x[0] * y[srcBLen - 1] */ + acc0 += x0 * c0; + + /* acc1 += x[1] * y[srcBLen - 1] */ + acc1 += x1 * c0; + + /* acc2 += x[2] * y[srcBLen - 1] */ + acc2 += x2 * c0; + + /* acc3 += x[3] * y[srcBLen - 1] */ + acc3 += x3 * c0; + + /* Read y[srcBLen - 2] sample */ + c0 = *(py--); + + /* Read x[4] sample */ + x0 = *(px + 1u); + + /* Perform the multiply-accumulate */ + /* acc0 += x[1] * y[srcBLen - 2] */ + acc0 += x1 * c0; + /* acc1 += x[2] * y[srcBLen - 2] */ + acc1 += x2 * c0; + /* acc2 += x[3] * y[srcBLen - 2] */ + acc2 += x3 * c0; + /* acc3 += x[4] * y[srcBLen - 2] */ + acc3 += x0 * c0; + + /* Read y[srcBLen - 3] sample */ + c0 = *(py--); + + /* Read x[5] sample */ + x1 = *(px + 2u); + + /* Perform the multiply-accumulates */ + /* acc0 += x[2] * y[srcBLen - 3] */ + acc0 += x2 * c0; + /* acc1 += x[3] * y[srcBLen - 2] */ + acc1 += x3 * c0; + /* acc2 += x[4] * y[srcBLen - 2] */ + acc2 += x0 * c0; + /* acc3 += x[5] * y[srcBLen - 2] */ + acc3 += x1 * c0; + + /* Read y[srcBLen - 4] sample */ + c0 = *(py--); + + /* Read x[6] sample */ + x2 = *(px + 3u); + px += 4u; + + /* Perform the multiply-accumulates */ + /* acc0 += x[3] * y[srcBLen - 4] */ + acc0 += x3 * c0; + /* acc1 += x[4] * y[srcBLen - 4] */ + acc1 += x0 * c0; + /* acc2 += x[5] * y[srcBLen - 4] */ + acc2 += x1 * c0; + /* acc3 += x[6] * y[srcBLen - 4] */ + acc3 += x2 * c0; + + + } while(--k); + + /* If the srcBLen is not a multiple of 4, compute any remaining MACs here. + ** No loop unrolling is used. */ + k = srcBLen % 0x4u; + + while(k > 0u) + { + /* Read y[srcBLen - 5] sample */ + c0 = *(py--); + + /* Read x[7] sample */ + x3 = *(px++); + + /* Perform the multiply-accumulates */ + /* acc0 += x[4] * y[srcBLen - 5] */ + acc0 += x0 * c0; + /* acc1 += x[5] * y[srcBLen - 5] */ + acc1 += x1 * c0; + /* acc2 += x[6] * y[srcBLen - 5] */ + acc2 += x2 * c0; + /* acc3 += x[7] * y[srcBLen - 5] */ + acc3 += x3 * c0; + + /* Reuse the present samples for the next MAC */ + x0 = x1; + x1 = x2; + x2 = x3; + + /* Decrement the loop counter */ + k--; + } + + /* Store the result in the accumulator in the destination buffer. */ + *pOut++ = acc0; + *pOut++ = acc1; + *pOut++ = acc2; + *pOut++ = acc3; + + /* Increment the pointer pIn1 index, count by 4 */ + count += 4u; + + /* Update the inputA and inputB pointers for next MAC calculation */ + px = pIn1 + count; + py = pSrc2; + + + /* Decrement the loop counter */ + blkCnt--; + } + + + /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here. + ** No loop unrolling is used. */ + blkCnt = blockSize2 % 0x4u; + + while(blkCnt > 0u) + { + /* Accumulator is made zero for every iteration */ + sum = 0.0f; + + /* Apply loop unrolling and compute 4 MACs simultaneously. */ + k = srcBLen >> 2u; + + /* First part of the processing with loop unrolling. Compute 4 MACs at a time. + ** a second loop below computes MACs for the remaining 1 to 3 samples. */ + while(k > 0u) + { + /* Perform the multiply-accumulates */ + sum += *px++ * *py--; + sum += *px++ * *py--; + sum += *px++ * *py--; + sum += *px++ * *py--; + + /* Decrement the loop counter */ + k--; + } + + /* If the srcBLen is not a multiple of 4, compute any remaining MACs here. + ** No loop unrolling is used. */ + k = srcBLen % 0x4u; + + while(k > 0u) + { + /* Perform the multiply-accumulate */ + sum += *px++ * *py--; + + /* Decrement the loop counter */ + k--; + } + + /* Store the result in the accumulator in the destination buffer. */ + *pOut++ = sum; + + /* Increment the MAC count */ + count++; + + /* Update the inputA and inputB pointers for next MAC calculation */ + px = pIn1 + count; + py = pSrc2; + + /* Decrement the loop counter */ + blkCnt--; + } + } + else + { + /* If the srcBLen is not a multiple of 4, + * the blockSize2 loop cannot be unrolled by 4 */ + blkCnt = blockSize2; + + while(blkCnt > 0u) + { + /* Accumulator is made zero for every iteration */ + sum = 0.0f; + + /* srcBLen number of MACS should be performed */ + k = srcBLen; + + while(k > 0u) + { + /* Perform the multiply-accumulate */ + sum += *px++ * *py--; + + /* Decrement the loop counter */ + k--; + } + + /* Store the result in the accumulator in the destination buffer. */ + *pOut++ = sum; + + /* Increment the MAC count */ + count++; + + /* Update the inputA and inputB pointers for next MAC calculation */ + px = pIn1 + count; + py = pSrc2; + + /* Decrement the loop counter */ + blkCnt--; + } + } + + + /* -------------------------- + * Initializations of stage3 + * -------------------------*/ + + /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1] + * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2] + * .... + * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2] + * sum += x[srcALen-1] * y[srcBLen-1] + */ + + /* In this stage the MAC operations are decreased by 1 for every iteration. + The blockSize3 variable holds the number of MAC operations performed */ + + /* Working pointer of inputA */ + pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u); + px = pSrc1; + + /* Working pointer of inputB */ + pSrc2 = pIn2 + (srcBLen - 1u); + py = pSrc2; + + /* ------------------- + * Stage3 process + * ------------------*/ + + while(blockSize3 > 0u) + { + /* Accumulator is made zero for every iteration */ + sum = 0.0f; + + /* Apply loop unrolling and compute 4 MACs simultaneously. */ + k = blockSize3 >> 2u; + + /* First part of the processing with loop unrolling. Compute 4 MACs at a time. + ** a second loop below computes MACs for the remaining 1 to 3 samples. */ + while(k > 0u) + { + /* sum += x[srcALen - srcBLen + 1] * y[srcBLen - 1] */ + sum += *px++ * *py--; + + /* sum += x[srcALen - srcBLen + 2] * y[srcBLen - 2] */ + sum += *px++ * *py--; + + /* sum += x[srcALen - srcBLen + 3] * y[srcBLen - 3] */ + sum += *px++ * *py--; + + /* sum += x[srcALen - srcBLen + 4] * y[srcBLen - 4] */ + sum += *px++ * *py--; + + /* Decrement the loop counter */ + k--; + } + + /* If the blockSize3 is not a multiple of 4, compute any remaining MACs here. + ** No loop unrolling is used. */ + k = blockSize3 % 0x4u; + + while(k > 0u) + { + /* Perform the multiply-accumulates */ + /* sum += x[srcALen-1] * y[srcBLen-1] */ + sum += *px++ * *py--; + + /* Decrement the loop counter */ + k--; + } + + /* Store the result in the accumulator in the destination buffer. */ + *pOut++ = sum; + + /* Update the inputA and inputB pointers for next MAC calculation */ + px = ++pSrc1; + py = pSrc2; + + /* Decrement the loop counter */ + blockSize3--; + } + +#else + + /* Run the below code for Cortex-M0 */ + + float32_t *pIn1 = pSrcA; /* inputA pointer */ + float32_t *pIn2 = pSrcB; /* inputB pointer */ + float32_t sum; /* Accumulator */ + uint32_t i, j; /* loop counters */ + + /* Loop to calculate convolution for output length number of times */ + for (i = 0u; i < ((srcALen + srcBLen) - 1u); i++) + { + /* Initialize sum with zero to carry out MAC operations */ + sum = 0.0f; + + /* Loop to perform MAC operations according to convolution equation */ + for (j = 0u; j <= i; j++) + { + /* Check the array limitations */ + if((((i - j) < srcBLen) && (j < srcALen))) + { + /* z[i] += x[i-j] * y[j] */ + sum += pIn1[j] * pIn2[i - j]; + } + } + /* Store the output in the destination buffer */ + pDst[i] = sum; + } + +#endif /* #ifndef ARM_MATH_CM0_FAMILY */ + +} + +/** + * @} end of Conv group + */ diff --git a/TNC/SymbolSlopeIntegrator.h b/TNC/SymbolSlopeIntegrator.h new file mode 100644 index 0000000..ba3f954 --- /dev/null +++ b/TNC/SymbolSlopeIntegrator.h @@ -0,0 +1,124 @@ +// Copyright 2021 Mobilinkd LLC. + +#pragma once + +#include "Log.h" + +#include + +#include +#include +#include +#include +#include +#include +#include + +namespace mobilinkd { namespace m17 { + +/** + * Calculate the phase estimates for each sample position. + * + * This performs a running calculation of the phase of each bit position. + * It is very noisy for individual samples, but quite accurate when + * averaged over an entire M17 frame. + * + * It is designed to be used to calculate the best bit position for each + * frame of data. Samples are collected and averaged. When update() is + * called, the best sample index and clock are estimated, and the counters + * reset for the next frame. + * + * It starts counting bit 0 as the first bit received after a reset. + * + * This is very efficient as it only uses addition and subtraction for + * each bit sample. And uses one multiply and divide per update (per + * frame). + * + * This will permit a clock error of up to 500ppm. This allows up to + * 250ppm error for both transmitter and receiver clocks. This is + * less than one sample per frame when the sample rate is 48000 SPS. + * + * @inv current_index_ is in the interval [0, SAMPLES_PER_SYMBOL). + * @inv sample_index_ is in the interval [0, SAMPLES_PER_SYMBOL). + * @inv clock_ is in the interval [0.9995, 1.0005] + */ + +template +struct SymbolSlopeIntegrator +{ + static constexpr std::array IMPULSE = { + -1.0, -0.70710678, 0.0, 0.70710678, 1.0 + }; + + static constexpr size_t BEGIN = SamplesPerSymbol / 2 + IMPULSE.size() / 2; + static constexpr size_t END = BEGIN + SamplesPerSymbol; + + std::array integrator_; + std::array signal; + std::array output; + + FloatType prev_ = 0.0; + size_t index_ = 0; + + void reset() + { + integrator_.fill(0.0); + prev_ = 0.0; + index_ = 0; + } + + void operator()(FloatType value) + { + auto dy = value - prev_; + + // Invert the phase estimate when sample midpoint is less than 0. + integrator_[index_] += value < 0 ? -dy : dy; + index_ = (index_ == (SamplesPerSymbol - 1)) ? 0 : index_ + 1; + prev_ = value; + } + + [[gnu::noinline]] + int8_t update() + { + // Circular convolution, so we concatenate the signal back-to-back. + // This is effectively "wrap" convolution, just offset by + // (SamplesPerSignal + IMPULSE.size()) / 2. + auto it = std::copy(integrator_.begin(), integrator_.end(), signal.begin()); + std::copy(integrator_.begin(), integrator_.end(), it); + + arm_conv_f32(signal.data(), signal.size(), + (float*)IMPULSE.data(), IMPULSE.size(), + output.data()); + + auto argmax = std::max_element(&output[BEGIN], &output[END]); + int8_t index = std::distance(&output[BEGIN], argmax); + + // Normalize index from wrapped position. + index -= SamplesPerSymbol / 2; + if (index < 0) index += SamplesPerSymbol; + +#if 1 + INFO("output: %5d, %5d, %5d, %5d, %5d, %5d, %5d, %5d, %5d, %5d", + int(output[7]*1000), int(output[8]*1000), int(output[9]*1000), int(output[10]*1000), int(output[11]*1000), + int(output[12]*1000), int(output[13]*1000), int(output[14]*1000), int(output[15]*1000), int(output[16]*1000)); + + INFO("Index = %d", index); +#endif + + // Reset integrator for next frame. + integrator_.fill(0.0); + + return index; + } + + /** + * Return the current sample index. This will always be in the range of + * [0..SAMPLES_PER_SYMBOL). + */ + uint8_t current_index() const + { + return index_; + } +}; + +}} // mobilinkd::m17