From 817b33e2d67ae330d4e74c7ded81b5812de5f02e Mon Sep 17 00:00:00 2001 From: boblark Date: Sat, 13 Mar 2021 10:43:44 -0800 Subject: [PATCH] Add analyze_fft4096_iq_F32 and example ino for it --- analyze_fft4096_iq_F32.cpp | 382 +++++++++++++++++++++++ analyze_fft4096_iq_F32.h | 302 ++++++++++++++++++ examples/TestFFT4096iq/TestFFT4096iq.ino | 73 +++++ 3 files changed, 757 insertions(+) create mode 100644 analyze_fft4096_iq_F32.cpp create mode 100644 analyze_fft4096_iq_F32.h create mode 100644 examples/TestFFT4096iq/TestFFT4096iq.ino diff --git a/analyze_fft4096_iq_F32.cpp b/analyze_fft4096_iq_F32.cpp new file mode 100644 index 0000000..38760cb --- /dev/null +++ b/analyze_fft4096_iq_F32.cpp @@ -0,0 +1,382 @@ +/* + * analyze_fft4096_iq_F32.cpp Assembled by Bob Larkin 9 Mar 2021 + * + * This class is Teensy 4.x ONLY. + * F32 Bolocks are always 128 floats, and any data rate is OK. + * + * Converted to F32 floating point input and also extended + * for complex I and Q inputs + * * Adapted all I/O to be F32 floating point for OpenAudio_ArduinoLibrary + * * Future: Add outputs for I & Q FFT x2 for overlapped FFT + * * Windowing None, Hann, Kaiser and Blackman-Harris. + * + * Conversion Copyright (c) 2021 Bob Larkin + * Same MIT license as PJRC: + * + * Audio Library for Teensy 3.X + * Copyright (c) 2014, Paul Stoffregen, paul@pjrc.com + * + * Development of this audio library was funded by PJRC.COM, LLC by sales of + * Teensy and Audio Adaptor boards. Please support PJRC's efforts to develop + * open source software by purchasing Teensy or other PJRC products. + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice, development funding notice, and this permission + * notice shall be included in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + */ + +// *************** TEENSY 4.X ONLY **************** +#if defined(__IMXRT1062__) + +#include +#include "analyze_fft4096_iq_F32.h" + +// Note: Suppports block size of 128 only. Very "built in." + +// Move audio data from audio_block_f32_t to the interleaved FFT instance buffer. +static void copy_to_fft_buffer1(void *destination, const void *sourceI, const void *sourceQ) { + const float *srcI = (const float *)sourceI; + const float *srcQ = (const float *)sourceQ; + float *dst = (float *)destination; // part of fft_buffer array. 256 floats per call + for (int i=0; i < 128; i++) { + *dst++ = *srcI++; // real sample, interleave + *dst++ = *srcQ++; // imag + } + } + +static void apply_window_to_fft_buffer1(void *fft_buffer, const void *window) { + float *buf = (float *)fft_buffer; // 0th entry is real (do window) 1st is imag + const float *win = (float *)window; + for (int i=0; i < 4096; i++) { + buf[2*i] *= *win; // real + buf[2*i + 1] *= *win++; // imag + } + } + +void AudioAnalyzeFFT4096_IQ_F32::update(void) { + audio_block_f32_t *block_i,*block_q; + int ii; + + block_i = receiveReadOnly_f32(0); + if (!block_i) return; + block_q = receiveReadOnly_f32(1); + if (!block_q) { + release(block_i); + return; + } + // Here with two new blocks of data + + switch (state) { + case 0: + blocklist_i[0] = block_i; blocklist_q[0] = block_q; + state = 1; + break; + case 1: + blocklist_i[1] = block_i; blocklist_q[1] = block_q; + state = 2; + break; + case 2: + blocklist_i[2] = block_i; blocklist_q[2] = block_q; + state = 3; + break; + case 3: + blocklist_i[3] = block_i; blocklist_q[3] = block_q; + state = 4; + break; + case 4: + blocklist_i[4] = block_i; blocklist_q[4] = block_q; + state = 5; + break; + case 5: + blocklist_i[5] = block_i; blocklist_q[5] = block_q; + state = 6; + break; + case 6: + blocklist_i[6] = block_i; blocklist_q[6] = block_q; + state = 7; + break; + case 7: + blocklist_i[7] = block_i; blocklist_q[7] = block_q; + state = 8; + break; + case 8: + blocklist_i[8] = block_i; blocklist_q[8] = block_q; + state = 9; + break; + case 9: + blocklist_i[9] = block_i; blocklist_q[9] = block_q; + state = 10; + break; + case 10: + blocklist_i[10] = block_i; blocklist_q[10] = block_q; + state = 11; + break; + case 11: + blocklist_i[11] = block_i; blocklist_q[11] = block_q; + state = 12; + break; + case 12: + blocklist_i[12] = block_i; blocklist_q[12] = block_q; + state = 13; + break; + case 13: + blocklist_i[13] = block_i; blocklist_q[13] = block_q; + state = 14; + break; + case 14: + blocklist_i[14] = block_i; blocklist_q[14] = block_q; + state = 15; + break; + case 15: + blocklist_i[15] = block_i; blocklist_q[15] = block_q; + state = 16; + break; + case 16: + blocklist_i[16] = block_i; blocklist_q[16] = block_q; + + // This next forming of the sumsq[] takes 48 uSec + count++; + for (int i = 0; i < 2048; i++) { + // From complex FFT the "negative frequencies" are mirrors of the frequencies above fs/2. So, we get + // frequencies from 0 to fs by re-arranging the coefficients. These are powers (not Volts) + // See DD4WH SDR + float ss0 = fft_buffer[2 * i] * fft_buffer[2 * i] + + fft_buffer[2 * i + 1] * fft_buffer[2 * i + 1]; + float ss1 = fft_buffer[2 * (i + 2048)] * fft_buffer[2 * (i + 2048)] + + fft_buffer[2 * (i + 2048) + 1] * fft_buffer[2 * (i + 2048) + 1]; + + if(count==1) { // Starting new average + sumsq[i+2048] = ss0; + sumsq[i] = ss1; + } + else if (count <= nAverage) { // Adding on to average + sumsq[i+2048] += ss0; + sumsq[i] += ss1; + } + } + // sumsq[] is filled. Wait to state==17 to convert to dBFS, etc + state = 17; + break; + case 17: + blocklist_i[17] = block_i; blocklist_q[17] = block_q; + + // This state==17 block takes 710 uSec for DBFS, but + // only 65 for POWER. DB conversions do not need to be under + // this interrupt and POWER output should be used if time is short. + if (count >= nAverage) { // Average is finished + // count = 0; + outputflag = false; // Avoid starting read() during block 17 to 18 + float inAf = 1.0f/(float)nAverage; + for (int i=0; i < 2048; i++) { + // xAxis, bit 0 left/right; bit 1 low to high + if(xAxis & 0X02) + ii = i; + else + ii = i^2048; + if(xAxis & 0X01) + ii = (4095 - ii); + + if(outputType==FFT_RMS) + output[i] = sqrtf(inAf*sumsq[ii]); + else if(outputType==FFT_POWER) + output[i] = inAf*sumsq[ii]; + else if(outputType==FFT_DBFS) + output[i] = 10.0f*log10f(inAf*sumsq[ii])-66.23f; // Scaled to FS sine wave + else + output[i] = 0.0f; + } + // outputflag = true; Wait for next block + } // end of Average is Finished + state = 18; + break; + case 18: + blocklist_i[18] = block_i; blocklist_q[18] = block_q; + + // Second half of post-FFT processing. dBFS (log10f) is the big user of time. + if (count >= nAverage) { // Average is finished + count = 0; + float inAf = 1.0f/(float)nAverage; + for (int i=2048; i < 4096; i++) { + // xAxis, bit 0 left/right; bit 1 low to high + if(xAxis & 0X02) + ii = i; + else + ii = i^2048; + if(xAxis & 0X01) + ii = (4095 - ii); + + if(outputType==FFT_RMS) + output[i] = sqrtf(inAf*sumsq[ii]); + else if(outputType==FFT_POWER) + output[i] = inAf*sumsq[ii]; + else if(outputType==FFT_DBFS) + output[i] = 10.0f*log10f(inAf*sumsq[ii])-66.23f; // Scaled to FS sine wave + else + output[i] = 0.0f; + } + outputflag = true; + } // end of Average is Finished + state = 19; + break; + case 19: + blocklist_i[19] = block_i; blocklist_q[19] = block_q; + state = 20; + break; + case 20: + blocklist_i[20] = block_i; blocklist_q[20] = block_q; + state = 21; + break; + case 21: + blocklist_i[21] = block_i; blocklist_q[21] = block_q; + state = 22; + break; + case 22: + blocklist_i[22] = block_i; blocklist_q[22] = block_q; + state = 23; + break; + case 23: + blocklist_i[23] = block_i; blocklist_q[23] = block_q; + state = 24; + break; + case 24: + blocklist_i[24] = block_i; blocklist_q[24] = block_q; + state = 25; + break; + case 25: + blocklist_i[25] = block_i; blocklist_q[25] = block_q; + state = 26; + break; + case 26: + blocklist_i[26] = block_i; blocklist_q[26] = block_q; + state = 27; + break; + case 27: + blocklist_i[27] = block_i; blocklist_q[27] = block_q; + state = 28; + break; + case 28: + blocklist_i[28] = block_i; blocklist_q[28] = block_q; + state = 29; + break; + case 29: + blocklist_i[29] = block_i; blocklist_q[29] = block_q; + state = 30; + break; + case 30: + blocklist_i[30] = block_i; blocklist_q[30] = block_q; + state = 31; + break; + case 31: + blocklist_i[31] = block_i; blocklist_q[31] = block_q; + + // This state==31 takes about 500 uSec, including the FFT. + copy_to_fft_buffer1(fft_buffer+0x000, blocklist_i[0]->data, blocklist_q[0]->data); + copy_to_fft_buffer1(fft_buffer+0x100, blocklist_i[1]->data, blocklist_q[1]->data); + copy_to_fft_buffer1(fft_buffer+0x200, blocklist_i[2]->data, blocklist_q[2]->data); + copy_to_fft_buffer1(fft_buffer+0x300, blocklist_i[3]->data, blocklist_q[3]->data); + copy_to_fft_buffer1(fft_buffer+0x400, blocklist_i[4]->data, blocklist_q[4]->data); + copy_to_fft_buffer1(fft_buffer+0x500, blocklist_i[5]->data, blocklist_q[5]->data); + copy_to_fft_buffer1(fft_buffer+0x600, blocklist_i[6]->data, blocklist_q[6]->data); + copy_to_fft_buffer1(fft_buffer+0x700, blocklist_i[7]->data, blocklist_q[7]->data); + copy_to_fft_buffer1(fft_buffer+0x800, blocklist_i[8]->data, blocklist_q[8]->data); + copy_to_fft_buffer1(fft_buffer+0x900, blocklist_i[9]->data, blocklist_q[9]->data); + copy_to_fft_buffer1(fft_buffer+0xA00, blocklist_i[10]->data, blocklist_q[10]->data); + copy_to_fft_buffer1(fft_buffer+0xB00, blocklist_i[11]->data, blocklist_q[11]->data); + copy_to_fft_buffer1(fft_buffer+0xC00, blocklist_i[12]->data, blocklist_q[12]->data); + copy_to_fft_buffer1(fft_buffer+0xD00, blocklist_i[13]->data, blocklist_q[13]->data); + copy_to_fft_buffer1(fft_buffer+0xE00, blocklist_i[14]->data, blocklist_q[14]->data); + copy_to_fft_buffer1(fft_buffer+0xF00, blocklist_i[15]->data, blocklist_q[15]->data); + copy_to_fft_buffer1(fft_buffer+0x1000, blocklist_i[16]->data, blocklist_q[16]->data); + copy_to_fft_buffer1(fft_buffer+0x1100, blocklist_i[17]->data, blocklist_q[17]->data); + copy_to_fft_buffer1(fft_buffer+0x1200, blocklist_i[18]->data, blocklist_q[18]->data); + copy_to_fft_buffer1(fft_buffer+0x1300, blocklist_i[19]->data, blocklist_q[19]->data); + copy_to_fft_buffer1(fft_buffer+0x1400, blocklist_i[20]->data, blocklist_q[20]->data); + copy_to_fft_buffer1(fft_buffer+0x1500, blocklist_i[21]->data, blocklist_q[21]->data); + copy_to_fft_buffer1(fft_buffer+0x1600, blocklist_i[22]->data, blocklist_q[22]->data); + copy_to_fft_buffer1(fft_buffer+0x1700, blocklist_i[23]->data, blocklist_q[23]->data); + copy_to_fft_buffer1(fft_buffer+0x1800, blocklist_i[24]->data, blocklist_q[24]->data); + copy_to_fft_buffer1(fft_buffer+0x1900, blocklist_i[25]->data, blocklist_q[25]->data); + copy_to_fft_buffer1(fft_buffer+0x1A00, blocklist_i[26]->data, blocklist_q[26]->data); + copy_to_fft_buffer1(fft_buffer+0x1B00, blocklist_i[27]->data, blocklist_q[27]->data); + copy_to_fft_buffer1(fft_buffer+0x1C00, blocklist_i[28]->data, blocklist_q[28]->data); + copy_to_fft_buffer1(fft_buffer+0x1D00, blocklist_i[29]->data, blocklist_q[29]->data); + copy_to_fft_buffer1(fft_buffer+0x1E00, blocklist_i[30]->data, blocklist_q[30]->data); + copy_to_fft_buffer1(fft_buffer+0x1F00, blocklist_i[31]->data, blocklist_q[31]->data); + + if (pWin) + apply_window_to_fft_buffer1(fft_buffer, window); + + // Teensyduino core for T4.x supports arm_cfft_f32 + // arm_cfft_f32 (const arm_cfft_instance_f32 *S, float32_t *p1, uint8_t ifftFlag, uint8_t bitReverseFlag) + arm_cfft_f32(&Sfft, fft_buffer, 0, 1); + + release(blocklist_i[0]); release(blocklist_q[0]); + release(blocklist_i[1]); release(blocklist_q[1]); + release(blocklist_i[2]); release(blocklist_q[2]); + release(blocklist_i[3]); release(blocklist_q[3]); + release(blocklist_i[4]); release(blocklist_q[4]); + release(blocklist_i[5]); release(blocklist_q[5]); + release(blocklist_i[6]); release(blocklist_q[6]); + release(blocklist_i[7]); release(blocklist_q[7]); + release(blocklist_i[8]); release(blocklist_q[8]); + release(blocklist_i[9]); release(blocklist_q[9]); + release(blocklist_i[10]); release(blocklist_q[10]); + release(blocklist_i[11]); release(blocklist_q[11]); + release(blocklist_i[12]); release(blocklist_q[12]); + release(blocklist_i[13]); release(blocklist_q[13]); + release(blocklist_i[14]); release(blocklist_q[14]); + release(blocklist_i[15]); release(blocklist_q[15]); + + blocklist_i[0] = blocklist_i[16]; + blocklist_i[1] = blocklist_i[17]; + blocklist_i[2] = blocklist_i[18]; + blocklist_i[3] = blocklist_i[19]; + blocklist_i[4] = blocklist_i[20]; + blocklist_i[5] = blocklist_i[21]; + blocklist_i[6] = blocklist_i[22]; + blocklist_i[7] = blocklist_i[23]; + blocklist_i[8] = blocklist_i[24]; + blocklist_i[9] = blocklist_i[25]; + blocklist_i[10] = blocklist_i[26]; + blocklist_i[11] = blocklist_i[27]; + blocklist_i[12] = blocklist_i[28]; + blocklist_i[13] = blocklist_i[29]; + blocklist_i[14] = blocklist_i[30]; + blocklist_i[15] = blocklist_i[31]; + + blocklist_q[0] = blocklist_q[16]; + blocklist_q[1] = blocklist_q[17]; + blocklist_q[2] = blocklist_q[18]; + blocklist_q[3] = blocklist_q[19]; + blocklist_q[4] = blocklist_q[20]; + blocklist_q[5] = blocklist_q[21]; + blocklist_q[6] = blocklist_q[22]; + blocklist_q[7] = blocklist_q[23]; + blocklist_q[8] = blocklist_q[24]; + blocklist_q[9] = blocklist_q[25]; + blocklist_q[10] = blocklist_q[26]; + blocklist_q[11] = blocklist_q[27]; + blocklist_q[12] = blocklist_q[28]; + blocklist_q[13] = blocklist_q[29]; + blocklist_q[14] = blocklist_q[30]; + blocklist_q[15] = blocklist_q[31]; + + state = 16; + break; // From case 31 + } // End of switch & case 31 + } // End update() +#endif diff --git a/analyze_fft4096_iq_F32.h b/analyze_fft4096_iq_F32.h new file mode 100644 index 0000000..3609541 --- /dev/null +++ b/analyze_fft4096_iq_F32.h @@ -0,0 +1,302 @@ +/* + * Analyze_fft4096_iq_F32.h Assembled by Bob Larkin 9 Mar 2021 + * + * Note: Teensy 4.x Only, 3.x not supported + * + * Does Fast Fourier Transform of a 4096 point complex (I-Q) input. + * Output is one of three measures of the power in each of the 4096 + * output bins, Power, RMS level or dB relative to a full scale + * sine wave. Windowing of the input data is provided for to reduce + * spreading of the power in the output bins. All inputs are Teensy + * floating point extension (_F32) and all outputs are floating point. + * + * Features include: + * * I and Q inputs are OpenAudio_Arduino Library F32 compatible. + * * FFT output for every 512 inputs to overlapped FFTs to + * compensate for windowing. + * * Windowing None, Hann, Kaiser and Blackman-Harris. + * * Multiple bin-sum output to simulate wider bins. + * * Power averaging of multiple FFT + * * Soon: F32 audio outputs for I & Q + * + * Conversion Copyright (c) 2021 Bob Larkin + * Same MIT license as PJRC: + * + * From original real FFT: + * Audio Library for Teensy 3.X + * Copyright (c) 2014, Paul Stoffregen, paul@pjrc.com + * + * Development of this audio library was funded by PJRC.COM, LLC by sales of + * Teensy and Audio Adaptor boards. Please support PJRC's efforts to develop + * open source software by purchasing Teensy or other PJRC products. + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice, development funding notice, and this permission + * notice shall be included in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + */ + +/* Does complex input FFT of 4096 points. Multiple non-audio (via functions) + * output formats of RMS (same as I16 version, and default), + * Power or dBFS (full scale). Output can be bin by bin or a pointer to + * the output array is available. Several window functions are provided by + * in-class design, or a custom window can be provided from the INO. + * + * Functions (See comments below and #defines above: + * bool available() + * float read(unsigned int binNumber) + * float read(unsigned int binFirst, unsigned int binLast) + * int windowFunction(int wNum) + * int windowFunction(int wNum, float _kdb) // Kaiser only + * float* getData(void) + * float* getWindow(void) + * void putWindow(float *pwin) + * void setNAverage(int NAve) // >=1 + * void setOutputType(int _type) + * void setXAxis(uint8_t _xAxis) // 0, 1, 2, 3 + * + * x-Axis direction and offset per setXAxis(xAxis) for sine to I + * and cosine to Q. + * If xAxis=0 f=fs/2 in middle, f=0 on right edge + * If xAxis=1 f=fs/2 in middle, f=0 on left edge + * If xAxis=2 f=fs/2 on left edge, f=0 in middle + * If xAxis=3 f=fs/2 on right edgr, f=0 in middle + * If there is 180 degree phase shift to I or Q these all get reversed. + * + * Timing, max is longest update() time: + * T4.0 Windowed, dBFS Out, 987 uSec + * + * Scaling: + * Full scale for floating point DSP is a nebulous concept. Normally the + * full scale is -1.0 to +1.0. This is an unscaled FFT and for a sine + * wave centered in frequency on a bin and of FS amplitude, the power + * at that center bin will grow by 4096^2/4 = about 4 million without windowing. + * Windowing loss cuts this down. The RMS level can growwithout windowing to + * 4096. The dBFS has been scaled to make this max value 0 dBFS by + * removing 66.2 dB. With floating point, the dynamic range is maintained + * no matter how it is scaled, but this factor needs to be considered + * when building the INO. + */ + +#ifndef analyze_fft4096iq_h_ +#define analyze_fft4096iq_h_ + +// *************** TEENSY 4.X ONLY **************** +#if defined(__IMXRT1062__) + +#include "Arduino.h" +#include "AudioStream_F32.h" +#include "arm_math.h" +#include "mathDSP_F32.h" +#include "arm_const_structs.h" + +#define FFT_RMS 0 +#define FFT_POWER 1 +#define FFT_DBFS 2 + +#define NO_WINDOW 0 +#define AudioWindowNone 0 +#define AudioWindowHanning4096 1 +#define AudioWindowKaiser4096 2 +#define AudioWindowBlackmanHarris4096 3 + +class AudioAnalyzeFFT4096_IQ_F32 : public AudioStream_F32 { +//GUI: inputs:2, outputs:4 //this line used for automatic generation of GUI node +//GUI: shortName:FFT4096IQ +public: + AudioAnalyzeFFT4096_IQ_F32() : AudioStream_F32(2, inputQueueArray) { + // __MK20DX128__ T_LC; __MKL26Z64__ T3.0; __MK20DX256__T3.1 and T3.2 + // __MK64FX512__) T3.5; __MK66FX1M0__ T3.6; __IMXRT1062__ T4.0 and T4.1 + + // Teensy4 core library has the right files for new FFT + // arm CMSIS library has predefined structures of type arm_cfft_instance_f32 + Sfft = arm_cfft_sR_f32_len4096; // This is one of the structures + useHanningWindow(); + } + // There is no varient for "settings," as blocks other than 128 are + // not supported and, nothing depends on sample rate so we don't need that. + + // Returns true when output data is available. + bool available() { +#if defined(__IMXRT1062__) + if (outputflag == true) { + outputflag = false; // No double returns + return true; + } + return false; +#else + // Don't know how you got this far, but.... + Serial.println("Teensy 3.x NOT SUPPORTED"); + return false; +#endif + } + + // Returns a single bin output + float read(unsigned int binNumber) { + if (binNumber>4095 || binNumber<0) return 0.0; + return output[binNumber]; + } + + // Return sum of several bins. Normally use with power output. + // This produces the equivalent of bigger bins. + float read(unsigned int binFirst, unsigned int binLast) { + if (binFirst > binLast) { + unsigned int tmp = binLast; + binLast = binFirst; + binFirst = tmp; + } + if (binFirst > 4095) return 0.0; + if (binLast > 4095) binLast = 4095; + float sum = 0; + do { + sum += output[binFirst++]; + } while (binFirst <= binLast); + return sum; + } + + // Sets None, Hann, or Blackman-Harris window with no parameter + int windowFunction(int wNum) { + if(wNum == AudioWindowKaiser4096) + return -1; // Kaiser needs the kdb + windowFunction(wNum, 0.0f); + return 0; + } + + int windowFunction(int wNum, float _kdb) { + float kd; + pWin = window; + if(wNum == NO_WINDOW) + pWin = NULL; + else if (wNum == AudioWindowKaiser4096) { + if(_kdb<20.0f) + kd = 20.0f; + else + kd = _kdb; + useKaiserWindow(kd); + } + else if (wNum == AudioWindowBlackmanHarris4096) + useBHWindow(); + else + useHanningWindow(); // Default + return 0; + } + + // Fast pointer transfer. Be aware that the data will go away + // after the next 256 data points occur. + float* getData(void) { + // available() sets outputflag false + return output; + } + + // You can use this to design windows + float* getWindow(void) { + return window; + } + + // Bring custom window from the INO + void putWindow(float *pwin) { + float *p = window; + for(int i=0; i<4096; i++) + *p++ = *pwin++; // Copy for the FFT + } + + // Number of FFT averaged in the output + void setNAverage(int _nAverage) { + nAverage = _nAverage; + } + + // Output RMS (default), power or dBFS (FFT_RMS, FFT_POWER, FFT_DBFS) + void setOutputType(int _type) { + outputType = _type; + } + + // xAxis, bit 0 left/right; bit 1 low to high; default 0X03 + void setXAxis(uint8_t _xAxis) { + xAxis = _xAxis; + } + + virtual void update(void); + +private: + float output[4096]; + float window[4096]; + float *pWin = window; + float fft_buffer[8192]; + float sumsq[8192]; // Avoid re-use of output[] + uint8_t state = 0; + bool outputflag = false; + audio_block_f32_t *inputQueueArray[2]; + audio_block_f32_t *blocklist_i[32]; + audio_block_f32_t *blocklist_q[32]; + // For T4.x + // const static arm_cfft_instance_f32 arm_cfft_sR_f32_len1024; + arm_cfft_instance_f32 Sfft; + + int outputType = FFT_RMS; //Same type as I16 version init + int count = 0; + int nAverage = 1; + uint8_t xAxis = 0x03; + + // The Hann window is a good all-around window + void useHanningWindow(void) { + for (int i=0; i < 4096; i++) { + // 2*PI/4095 = 0.00153435538 + window[i] = 0.5*(1.0 - cosf(0.00153435538f*(float)i)); + } + } + + // Blackman-Harris produces a first sidelobe more than 90 dB down. + // The price is a bandwidth of about 2 bins. Very useful at times. + void useBHWindow(void) { + for (int i=0; i < 4096; i++) { + float kx = 0.00153435538f; // 2*PI/4095 + int ix = (float) i; + window[i] = 0.35875 - + 0.48829*cosf( kx*ix) + + 0.14128*cosf(2.0f*kx*ix) - + 0.01168*cosf(3.0f*kx*ix); + } + } + + /* The windowing function here is that of James Kaiser. This has a number + * of desirable features. The sidelobes drop off as the frequency away from a transition. + * Also, the tradeoff of sidelobe level versus cutoff rate is variable. + * Here we specify it in terms of kdb, the highest sidelobe, in dB, next to a sharp cutoff. For + * calculating the windowing vector, we need a parameter beta, found as follows: + */ + void useKaiserWindow(float kdb) { + float32_t beta, kbes, xn2; + mathDSP_F32 mathEqualizer; // For Bessel function + + if (kdb < 20.0f) + beta = 0.0; + else + beta = -2.17+0.17153*kdb-0.0002841*kdb*kdb; // Within a dB or so + + // Note: i0f is the fp zero'th order modified Bessel function (see mathDSP_F32.h) + kbes = 1.0f / mathEqualizer.i0f(beta); // An additional derived parameter used in loop + for (int n=0; n<512; n++) { + xn2 = 0.5f+(float32_t)n; + // 4/(1023^2)=0.00000382215877f + // 4/(4095^2) = 2.3853504E-7 + xn2 = 2.3853504E-7*xn2*xn2; + window[511 - n]=kbes*(mathEqualizer.i0f(beta*sqrtf(1.0-xn2))); + window[512 + n] = window[511 - n]; + } + } + }; +#endif +#endif diff --git a/examples/TestFFT4096iq/TestFFT4096iq.ino b/examples/TestFFT4096iq/TestFFT4096iq.ino new file mode 100644 index 0000000..c11d615 --- /dev/null +++ b/examples/TestFFT4096iq/TestFFT4096iq.ino @@ -0,0 +1,73 @@ + +// TestFFT2048iq.ino for Teensy 4.x +// Bob Larkin 9 March 2021 + +// Generate Sin and Cosine pair and input to IQ FFT. +// Serial Print out powers of all 4096 bins in +// dB relative to Sine Wave Full Scale + +// Public Domain + +#include "OpenAudio_ArduinoLibrary.h" +#include "AudioStream_F32.h" + +// GUItool: begin automatically generated code +AudioSynthSineCosine_F32 sine_cos1; //xy=76,532 +AudioAnalyzeFFT4096_IQ_F32 FFT4096iq1; //xy=243,532 +AudioOutputI2S_F32 audioOutI2S1; //xy=246,591 +AudioConnection_F32 patchCord1(sine_cos1, 0, FFT4096iq1, 0); +AudioConnection_F32 patchCord2(sine_cos1, 1, FFT4096iq1, 1); +// GUItool: end automatically generated code + +void setup(void) { + + Serial.begin(9600); + delay(1000); + + // The 4096 complex FFT needs 32 F32 memory for real and 32 for imag. + // Set memory to more than 64, depending on other useage. + AudioMemory_F32(100); + Serial.println("FFT4096IQ Test"); + + sine_cos1.amplitude(1.0f); // Initialize Waveform Generator + + // Pick T4.x bin center + //sine_cos1.frequency(689.0625f); + + // or pick any old frequency + sine_cos1.frequency(1000.0f); + + // elect the output format + FFT4096iq1.setOutputType(FFT_DBFS); + + // Select the wndow function + //FFT4096iq1.windowFunction(AudioWindowNone); + //FFT4096iq1.windowFunction(AudioWindowHanning4096); + //FFT4096iq1.windowFunction(AudioWindowKaiser4096, 55.0f); + FFT4096iq1.windowFunction(AudioWindowBlackmanHarris4096); + + // Uncomment to Serial print window function + // float* pw = FFT4096iq1.getWindow(); // Print window + // for (int i=0; i<4096; i++) Serial.println(pw[i], 7); + + // xAxis, bit 0 left/right; bit 1 low to high; default 0X03 + FFT4096iq1.setXAxis(0X03); + + FFT4096iq1.setNAverage(1); + delay(100); + } + +void loop(void) { + static bool doPrint=true; + float *pPwr; + // Print output, once + if( FFT4096iq1.available() && doPrint ) { + pPwr = FFT4096iq1.getData(); + for(int i=0; i<4096; i++) + Serial.println(*(pPwr + i), 8 ); + doPrint = false; + } + Serial.print(" Audio MEM Float32 Peak: "); + Serial.println(AudioMemoryUsageMax_F32()); + delay(500); + }