A portable WSPR decoder for microcontrollers!
We would like to decode WSPR traffic in real-time with less than 100mA of current consumption @ 5v.
-
Add two 8MB PSRAM chips to the Teensy 4.1 board.
-
(Optional) Run the
debug_code/teensy41_psram_memteston Teensy to check if PSRAM is working fine. -
Copy
samples/smaller_0918.wavto a FAT32 formatted microSD card. -
Insert this microSD card into the Teensy.
-
Write the 'wsprd demo firmware' to Teensy.
-
Monitor the Teensy's serial connection for successful (test) WSPR decodes.
sox samples/150426_0918.wav -r 6000 samples/smaller_0918.wav
$ make -f Makefile.debug; ./wsprd -qs samples/smaller_0918.wav
make: Nothing to be done for 'all'.
0918 -9 1.1 0.001446 1 ND6P DM04 30
0918 -15 0.1 0.001460 -1 W5BIT EL09 17
0918 -6 0.6 0.001489 0 WD4LHT EL89 30
0918 -21 0.5 0.001517 0 KI7CI DM09 37
0918 -18 -1.9 0.001530 0 DJ6OL JO52 37
0918 -11 0.8 0.001587 0 W3HH EL89 30
<DecodeFinished>
$ make -f Makefile; /usr/bin/time -v ./wsprd -qs samples/smaller_0918.wav
0918 -9 1.1 0.001446 0 ND6P DM04 30
0918 -15 0.1 0.001460 0 W5BIT EL09 17
0918 -6 0.6 0.001489 0 WD4LHT EL89 30
0918 -1 -0.8 0.001503 0 NM7J DM26 30
0918 -21 0.5 0.001517 0 KI7CI DM09 37
0918 -18 -1.9 0.001530 0 DJ6OL JO52 37
0918 -11 0.8 0.001587 0 W3HH EL89 30
<DecodeFinished>
Command being timed: "./wsprd -qs samples/smaller_0918.wav"
User time (seconds): 0.21
System time (seconds): 0.00
Percent of CPU this job got: 99%
Elapsed (wall clock) time (h:mm:ss or m:ss): 0:00.22
Average shared text size (kbytes): 0
Average unshared data size (kbytes): 0
Average stack size (kbytes): 0
Average total size (kbytes): 0
Maximum resident set size (kbytes): 12328
Average resident set size (kbytes): 0
Major (requiring I/O) page faults: 0
Minor (reclaiming a frame) page faults: 3501
Voluntary context switches: 3
Involuntary context switches: 2
Swaps: 0
File system inputs: 0
File system outputs: 24
Socket messages sent: 0
Socket messages received: 0
Signals delivered: 0
Page size (bytes): 4096
Exit status: 0
After writing the firmware and monitoring /dev/ttyACM0:
card initialized.
reads succeeded 1440044 bytes in 65880 us at 21.858591 MB/sec
002e03e4cc351c51cf17e2e3f621ec50
-9 1.1 0.001446 0 ND6P DM04 30
-15 0.1 0.001460 0 W5BIT EL09 17
-6 0.6 0.001489 0 WD4LHT EL89 30
-1 -0.8 0.001503 0 NM7J DM26 30
-21 0.5 0.001517 0 KI7CI DM09 37
-18 -1.9 0.001530 0 DJ6OL JO52 37
-11 0.8 0.001587 0 W3HH EL89 30
Current consumption: < 95mA @ 5V.
Can we optimize a bit more to fit the decoder in 6080KiB of PSRAM (attached
to a ESP32 TTGO T8 board)?
I (0) cpu_start: App cpu up.
I (1150) esp_psram: SPI SRAM memory test OK
I (1158) cpu_start: Pro cpu start user code
I (1158) cpu_start: cpu freq: 160000000 Hz
I (1158) cpu_start: Application information:
I (1161) cpu_start: Project name: himem_test
I (1166) cpu_start: App version: v5.1-dev-3213-gd29e53dc0c
I (1173) cpu_start: Compile time: Feb 7 2023 19:38:07
I (1179) cpu_start: ELF file SHA256: 23877ba8aa9f0f88...
I (1185) cpu_start: ESP-IDF: v5.1-dev-3213-gd29e53dc0c
I (1192) cpu_start: Min chip rev: v0.0
I (1197) cpu_start: Max chip rev: v3.99
I (1202) cpu_start: Chip rev: v3.0
I (1207) heap_init: Initializing. RAM available for dynamic allocation:
I (1214) heap_init: At 3FFAE6E0 len 00001920 (6 KiB): DRAM
I (1220) heap_init: At 3FFB2BA0 len 0002D460 (181 KiB): DRAM
I (1226) heap_init: At 3FFE0440 len 00003AE0 (14 KiB): D/IRAM
I (1233) heap_init: At 3FFE4350 len 0001BCB0 (111 KiB): D/IRAM
I (1239) heap_init: At 4008FEF4 len 0001010C (64 KiB): IRAM
I (1246) esp_psram: Adding pool of 2112K of PSRAM memory to heap allocator
I (1254) spi_flash: detected chip: winbond
I (1258) spi_flash: flash io: dio
W (1262) spi_flash: Detected size(16384k) larger than the size in the binary image header(2048k). Using the size in the binary image header.
I (1275) esp_himem: Initialized. Using last 62 32KB address blocks for bank switching on 6080 KB of physical memory.
I (1287) app_start: Starting scheduler on CPU0
I (1291) app_start: Starting scheduler on CPU1
I (1291) main_task: Started on CPU0
I (1301) esp_psram: Reserving pool of 32K of internal memory for DMA/internal allocations
I (1301) main_task: Calling app_main()
Himem has 6080KiB of memory, 6080KiB of which is free. Testing the free memory...
Done!
I (7541) main_task: Returned from app_main()
Further challenge: Can we make this work on Raspberry Pi Pico?
With the following hacky patch, we can reduce the memory usage at the cost of
decode performance:
diff --git a/wsprd.c b/wsprd.c
index 19536b0..ba0d9ee 100644
--- a/wsprd.c
+++ b/wsprd.c
@@ -153,9 +153,9 @@ unsigned long readwavfile(char *ptr_to_infile, int ntrmin, float *idat, float *q
nfft2=46080; //this is the number of downsampled points that will be returned
nh2=nfft2/2;
-#define SAMPLE_RATE 6000.0
+#define SAMPLE_RATE 3000.0
if( ntrmin == 2 ) {
- nfft1=nfft2*16; //need to downsample by a factor of 16
+ nfft1=nfft2*8; //need to downsample by a factor of 16
df=SAMPLE_RATE/nfft1;
i0=1500.0/df+0.5;
npoints=114*SAMPLE_RATE;
@@ -182,7 +182,7 @@ unsigned long readwavfile(char *ptr_to_infile, int ntrmin, float *idat, float *q
fclose(fp);
// realin=pffft_aligned_malloc(sizeof(float)*2*(nfft1/2+1));
- realin=malloc(sizeof(float)*2*(nfft1/2+1));
+ realin=malloc(sizeof(float)*4*(nfft1/2+1));
fftout=realin;
// fftsetup1=pffft_new_setup(nfft1, PFFFT_REAL);
kiss_fftr_cfg cfg = kiss_fftr_alloc(nfft1, 0, 0, 0);$ /usr/bin/time -v ./wsprd -qs smallest.wav
lest 54 1.1 0.001446 0 ND6P DM04 30
lest 36 0.0 0.001460 0 W5BIT EL09 17
<DecodeFinished>
Command being timed: "./wsprd -qs smallest.wav"
User time (seconds): 1.08
System time (seconds): 0.01
Percent of CPU this job got: 99%
Elapsed (wall clock) time (h:mm:ss or m:ss): 0:01.10
Average shared text size (kbytes): 0
Average unshared data size (kbytes): 0
Average stack size (kbytes): 0
Average total size (kbytes): 0
Maximum resident set size (kbytes): 6688
Average resident set size (kbytes): 0
Major (requiring I/O) page faults: 0
Minor (reclaiming a frame) page faults: 2112
Voluntary context switches: 3
Involuntary context switches: 7
Swaps: 0
File system inputs: 0
File system outputs: 24
Socket messages sent: 0
Socket messages received: 0
Signals delivered: 0
Page size (bytes): 4096
Exit status: 0
Finding 1: It seems Raspberry Pi Zero 2 W remains a decent option to decode
WSPR traffic.
Idea 1: Can we use an IQ HF receiver and the .c2 handling code path to reduce
the RAM usage by a lot?