Skip to content

Performance

pschatzmann edited this page Oct 29, 2026 · 3 revisions

The speed of the modulators and demodulators, as multiples of real time: everything above 1 is fast enough, the rest is CPU time which is left for your application. The memory is documented in Memory usage.

Microcontrollers

Measured with the benchmark example at 8000 Hz: the values are multiples of real time, so everything above 1 is fast enough. All modes decoded all characters on all boards, except one additional (wrong) character of qpsk125 on the RP2350.

Demodulation

Mode ESP32-S3 ESP32-P4 RP2040 RP2350
FSK 1200 41 136 9.1 20
mfsk16 27 73 3.6 16
mfsk64 10 28 1.2 6.2
psk125 57 129 7.6 62
qpsk125 52 115 7.3 53
8psk250 40 86 6.2 44
ofdm32 10 29 2.0 7.0
aprs 25 60 5.0 23
fx25 29 110 6.6 41
il2p 22 60 4.9 22
olivia-8-250 4.1 8.1 0.7 3.2
olivia-32-1000 2.6 5.1 0.5 2.0
contestia-8-250 5.8 12 0.9 4.5
css7 5.5 7.3 1.0 5.2
css10 5.4 7.4 1.0 5.1
dtmf 42 88 2.6 27

Modulation

Mode ESP32-S3 ESP32-P4 RP2040 RP2350
FSK 1200 61 137 19.6 75
mfsk16 66 152 20 74
mfsk64 58 157 21 87
psk125 55 163 9.4 69
qpsk125 55 163 9.4 69
8psk250 56 150 9.2 63
ofdm32 6.5 14 3.3 1.5
aprs 55 117 18 62
fx25 30 111 17 57
il2p 43 115 18 60
olivia-8-250 5.4 10 5.5 23
olivia-32-1000 5.5 11 5.8 24
contestia-8-250 5.8 11 6.0 26
css7 5.3 8.4 4.3 17
css10 5.3 8.3 4.1 17
dtmf 15 32 18 83
  • ESP32-S3 at 240 MHz with 8 MB octal PSRAM (FQBN esp32:esp32:esp32s3:PSRAM=opi), arduino-esp32 3.3.12.
  • ESP32-P4 (Waveshare ESP32-P4-WIFI6) with PSRAM (FQBN esp32:esp32:esp32p4:ChipVariant=postv3,PSRAM=enabled).
  • RP2040 (Raspberry Pi Pico, 133 MHz, no floating point unit, FQBN rp2040:rp2040:rpipico), Arduino-Pico 6.3.0.
  • RP2350 (Raspberry Pi Pico 2 W with PSRAM, 150 MHz, single precision floating point unit, FQBN rp2040:rp2040:rpipico2w), Arduino-Pico 6.3.0. The benchmark doesn't use the PSRAM.
  • The RP2040 is too slow for the Olivia and Contestia decoders (values below 1, in bold): they can only decode recorded audio, not in real time. CSS (1.0) has no headroom left: use the ESP32 for CSS. mfsk64 (1.2) and ofdm32 (2.0) leave little time for the application. All other modes run fine. The RP2350 is fast enough for all modes: thanks to its floating point unit it decodes 2 to 6 times faster than the RP2040. Only the OFDM encoder is slow on it (1.5).
  • Olivia needs the most CPU: olivia-32-1000 has the least headroom. CSS needs about the same for all spreading factors, because the FFT size grows with the symbol length. The CSS encoder calculates the chirp phase with double precision, which is slow on these chips (no double precision FPU). At higher sample rates the decoder also converts the input to 8000 Hz, which costs additional time.
  • The benchmark passes the generated audio directly to the decoder, so it doesn't need memory for the audio and also runs on boards without PSRAM. The ESP32 values were measured with an earlier version, which kept the audio in memory (PSRAM): their modulation values include the time for storing the audio and are therefore too low, especially for the slow modes with a lot of audio (Olivia, CSS). The demodulation values are hardly affected (the decoder only read its input from there). Repeating the ESP32 measurements with the current version gives correct modulation values. On an ESP32-S3 with octal PSRAM, select PSRAM=opi: with quad PSRAM the PSRAM isn't found.

Use a recent version of the arduino-audio-tools: older versions compute double precision sin()/cos() in the Goertzel filter for each block, which is slow on MCUs without a double precision FPU (FSK, MFSK), and their Vector::push_back() grows the buffer by one element at a time (the modulation in the benchmark was then slower than real time).

To measure your board, run the benchmark example: it needs no audio hardware and prints the real time factor of each mode. The decoding speed depends mostly on the sample rate, so use the lowest sample rate your hardware supports (see the comment on Olivia and CSS above).

PC

Decoding speed on a PC at 8000 Hz (× real time): FSK 1200 1000, mfsk16 1100, mfsk64 300, psk125 5500, ofdm32 1600, aprs 5600. ggwave is much slower: ggwave-fast 79 at 24000 Hz. Olivia (olivia-32-1000) ran 270, CSS (css7) 780, QPSK (qpsk125) 6000 and 8PSK (8psk125) 2000 times faster than real time.

The tinymodem --benchmarks option measures the FSK and MFSK modems on the desktop.

Clone this wiki locally