Localized global variables, corrected comments

Added descriptive comments on data handling and also a couple of Todo notes.

examples/FT8Receive/FT8Receive.ino

@@ -123,6 +123,12 @@ char Station_Call[11];  // six character call sign + /0
 char home_Locator[11];  // four character locator  + /0
 char Locator[11]; // four character locator  + /0
 uint8_t ft8_hours, ft8_minutes, ft8_seconds;
+//From gen_ft8.cpp, i.e., also used for transmit:
+char Target_Call[7]; //six character call sign + /0
+char Target_Locator[5]; // four character locator  + /0
+int Target_RSL; // four character RSL  + /0
+float32_t Station_Latitude, Station_Longitude;
+float32_t Target_Latitude, Target_Longitude;
 
 // rcvFT8State
 #define FT8_RCV_IDLE 0
@@ -131,20 +137,15 @@ uint8_t ft8_hours, ft8_minutes, ft8_seconds;
 #define FT8_RCV_READY_DECODE 3
 #define FT8_RCV_DECODE 4
 int rcvFT8State = FT8_RCV_IDLE;
-int offset_step;
 
-//From gen_ft8.cpp, i.e., also used for transmit
+int offset_step;
-char Target_Call[7]; //six character call sign + /0
-char Target_Locator[5]; // four character locator  + /0
-int Target_RSL; // four character RSL  + /0
-float32_t Station_Latitude, Station_Longitude;
-float32_t Target_Latitude, Target_Longitude;
 int32_t current_time, start_time;
 int32_t ft8_time, ft8_mod_time, ft8_mod_time_last;
 int32_t tOffset = 0;  // Added Sept 22
 uint8_t secLast = 0;
 const int ledPin = 13;
 bool showPower = false;
+bool noiseMeasured = false;
 uint32_t tp = 0;
 uint32_t tu = 0;
 
@@ -283,6 +284,9 @@ void loop(void)  {
 
    if(rcvFT8State==FT8_RCV_DATA_COLLECT && demod1.available())
       {
+	  // Note: The RadioFT8Demodulator provides at least 2.7 milliseconds, after data is
+      // available, before pData2K is written over. The extract_power() here must stay
+      // on schedule to protect the data.  THis should be OK.
       // Here every 80 mSec for FFT
       rcvFT8State = FT8_RCV_FIND_POWERS;
       // 1472 * 92 = 135424
@@ -317,7 +321,7 @@ void loop(void)  {
       Serial.println(micros() - tu);
 #endif
       }
-   delay(1);
+   delay(1);   // Don't increase
    }   // End loop()
 
 time_t getTeensy3Time()  {

examples/FT8Receive/Process_DSP_R.ino

@@ -42,25 +42,14 @@ SOFTWARE.
  * to pass all frequencies up to, at least 2800 Hz.
  */
 
-// Following are used inside extract_power()
+float32_t   window[1024];  // Holds half of symmetrical curve
-float32_t fft_buffer[2048];
-float  fftOutput[2048];
-float   window[2048];  // Change to 1024 by symmetry  <<<<<<<<<<<<<<<<<<<
 arm_rfft_fast_instance_f32 Sfft;
 
-float32_t powerSum = 0.0f;    // Use these for snr estimate
-float32_t runningSum = 0.0f;
-float32_t powerMax = 0.0f;
-float32_t runningMax = 0.0f;
-float32_t noiseBuffer[8];     // Circular storage
-uint16_t  noiseBufferWrite = 0;   // Array index
-bool      noiseMeasured = false;     // <<<<<<GLOBAL
-uint8_t   noisePower8 = 0;         // half dB per noise estimate GLOBAL
-
 void init_DSP(void) {
    arm_rfft_fast_init_f32(&Sfft, 2048);
-   for (int i = 0; i < FFT_SIZE; ++i)
+   // The window is symmetric, so create half of it
-      window[i] = ft_blackman_i(i, FFT_SIZE);
+   for (int i = 0; i < 1024; ++i)
+     window[i] = ft_blackman_i(i, 2048);
    offset_step = 1472;    // (int) HIGH_FREQ_INDEX*4;
    }
 
@@ -78,25 +67,60 @@ float ft_blackman_i(int i, int N) {
 
 // Compute FFT magnitudes (log power) for each timeslot in the signal
 void extract_power( int offset) {
+   float32_t fft_buffer[2048];
+   float32_t fftOutput[2048];
+   float32_t powerSum = 0.0f;    // Use these for snr estimate
+   float32_t runningSum = 0.0f;
+   float32_t powerMax = 0.0f;
+   float32_t runningMax = 0.0f;
+   float32_t noiseBuffer[8];     // Circular storage
+   uint16_t  noiseBufferWrite = 0;   // Array index
+   uint8_t   noisePower8 = 0;         // half dB per noise estimate GLOBAL
+	
    float32_t y[8];
    float32_t noiseCoeff[3];
 
-   /* Format of export_fft_power[] array:
+   /* The FFT's are arranged to have overlap in both time and frequency.
-    368 bytes of power for even time for 0.32 sec sample  DESCRIBE BETTER <<<<<<<<<<<<<<<<<<<<<<
+    * This allows good decoding sensitivity.  It remains difficult to decode a weak
-    368 bytes of power for odd  time for 0.32 sec sample
+	* signal that is close in frequency to a strong one.  It also causes multiple
-                         ...
+	* combinations of time and frequency to show the same signal.
-    Repeated about  14.7/(0.08 sec) = 184 times. (Transmitted message length is 12.96 sec)
+	* 
-    Total bytes 4 * 368 * 92 = 135424
+	* The 2048 point real FFT yields 1024 power outputs, corresponding to 0 to 3200 Hz.
-
+	* Only the frequencies up to 2300 Hz are inspected in this implementation.  That
-    The power byte is log encoded with a half dB MSB.  This can handle a
+	* corresponds with outputs 0 to 736.  The windowing of the FFT results in frequency
-    dynamic range of 256/2 = 128 dB.
+	* resolution that is close to double the bin spacing.  So, the power data is grouped
-   */
+	* up in 736/2=368 power data points.  It is done twice, using every other bins.
+    * Format of export_fft_power[] array:
+    * 368 bytes of power for even frequencies, 0, 2, 4, ... 366
+    * 368 bytes of power for odd frequencies, 1, 3, 5, ... 367
+    * Repeated 14.72/(0.08 sec) = 184 times.
+	* The transmitted message length is 12.96 sec so the difference allows timing errors
+	* along with the decoding allowing missing symbols.
+    * Total bytes saved for decoding is 2 * 368 * 194 = 135424 for the 14.72 seconds.
+    *
+    * The power byte is log encoded with a half dB MSB.  This can handle a
+    * dynamic range of 256/2 = 128 dB.
+    */
 	
-   for(int i=0; i<2048; i++)
+	// TODO: It would seem easy to offset the frequencies being examined and, without
+	// increasing the data burden, make the range more productive.  For instance, move the
+	// 0 to 2300 up to 300 to 2600 Hz.  Needs investigation.
+	
+	// TODO: The 135424 byte array is using more than a fourth of Teensy 4.x main RAM1.
+	// Might the array be put in RAM2 (and perhaps enlarged somewhat) and re-arranged
+    // in frequency order.  This would allow transfers to RAM1 in, say, 300 Hz overlapping
+	// segments.   This might be 300 to 600 Hz, 500 to 800, 700 to 1000 and so forth.  That
+	// would cut way back on RAM 1 array size.  Problems with this, except complexity??
+
+   // Note: The RadioFT8Demodulator provides at least 2.7 milliseconds, after data is
+   // available, before pData2K is written over.  This needs to be thought of in the design
+   // of the loop() in the main INO.  Long delays are trouble.  The following transfer
+   // of data to fft_buffer is very fast.  After that, pData2K[] is available.
+   for(int i=0; i<1024; i++)
       {
       fft_buffer[i] = window[i]*pData2K[i];  // Protect pData2K from in-place FFT (17 uSec)
+	  fft_buffer[2047 - i] = window[i]*pData2K[2047 - i]; // Symmetrical window
       }
-
    //           (float32_t* pIn, float32_t* pOut, uint8_t ifftFlag)
    arm_rfft_fast_f32(&Sfft, fft_buffer, fftOutput, 0);
    arm_cmplx_mag_squared_f32(fftOutput, fftOutput, 1024);

radioFT8Demodulator_F32.cpp

@@ -1,9 +1,11 @@
 /*
- *   radioFT8Demodulator_F32.cpp       Assembled by Bob Larkin   8 Sept 2022
+ *         radioFT8Demodulator_F32.cpp 
+ * Assembled by Bob Larkin   8 Sept 2022
+ * Comments corrected 16 Nov 2022
  *
  *  This class is Teensy 4.x ONLY.
  *
- * Copyright (c) 2021 Bob Larkin
+ * Copyright (c) 2022 Bob Larkin
  *
  * Permission is hereby granted, free of charge, to any person obtaining a copy
  * of this software and associated documentation files (the "Software"), to deal
@@ -28,11 +30,6 @@
  *   Every 5 audio samples at 96 kHz produce 1 19.2 kHz sample
  *   Every 3 audio samples at 19.2   produce 1  6.4 kHz sample
  *   Every 15*128=1920 audio samples at 96 kHz produce 128 samples at 6.4 kHz
- *
- * Every 120 (for 48 kHz) or 60 (for 96 kHz) blocks produces 1024 data points
- * at 6.4 kHz reduced sample rate.  This 1024 size is sent to the .INO as it
- * is *half* of a 2048 data block, and thus supports 50% overlap FFT's.
- *
  */
 
  // NOTE Changed class name to start with capital "R"  RSL 7 Nov 2022