20180324
Purpose: The ESP-32 has Analog To Digital (ADC) and Digital To Analog (DAC) converters built-in. Could these on-Board devices be used in place of external analog interface peripherals (typically with i2s or spi interfaces) to reduce system cost/complexity in voice audio applications?
For the purpose of these demos, circuits external to the ESP32 were restricted to analog signal conditioning components only.
The scope of this demonstration is limited to voice audio (<4KHz). Applications in the voice audio range include: game audio, voice notifications, voice grade telephone service or radio communications circuits, walkie-talkie/intercom/baby monitor/surveillance type apps, voice compression and processing, vocoders, baseband signaling and audio test signals.
This ESP32 based audio framework uses the popular Arduino IDE and could be used to teach Digital Signal Processing concepts. At a university or in a workshop at a technical conference or maker class, it could be used as a low cost hands-on lab for students. Time wouldn't be wasted with setup or teaching the build tools.
Proof of concept demos programs were created and tested over a one month period.
- The 8-bit DAC limits the dynamic range to 48 dB and limits fidelity but may be adequate for some voice applications.
- The ADC is not driven directly by a hardware sample clock but is sampled in software on a timer interrupt creating sample clock skew. This results in some phase noise on the sampled audio.
- The sample service core processor is only loaded at 13% at the 8Ksps rate and could be increased to a higher sample rate.
- A frame rate processing technique used in the demos (v.s. sample-by-sample processing) introduces latency--audio input to output.
- The ring buffers in FreeRTOS were too slow to operate at the test Frame Rate (128 msec., needed 256 msec. )
- The demos were compatible with concurrent Bluetooth Low Energy operation. WiFi not checked.
Dual core concurrent multi-task processing of continuous real-time audio on the ESP-32.
Tested on Espressif ESP32 Dev board Rev. 1 silicon
Arduino IDE with the ESP32 core from espressif
ADC/DAC sample processing at 8Ksps for voice audio range (< 4KHz). Not suitable for music as is.
Core 1 -- Sample Service
Core 0 -- Application Processor
Cores use alternating ping-pong buffers to pass samples between them every N samples. Tasks synchronize on a "FrameAvailable" semaphore.
Frame processing runs every N samples on the Application Processor. The Frame processing technique introduces latency delay (audio input to output).
Concurrency examples include:
- Continuous sine wave
- Dual tone (frequencies re-create old telephone dial tone)
- Continuous phase frequency sweep test signal (300--3000 Hz in 5 sec.,repeated)
- Microphone talk-through
- Pink Noise example from portaudio.com
- Voice Alert Example
bobolink twitter: @wm6h
ref: esp32 arduino ide https://github.com/espressif/arduino-esp32
Sensorsslot's ESP32 Arduino IDE Dual core multi-tasking intro: YouTube video: https://www.youtube.com/watch?v=k_D_Qu0cgu8 http://esp32.info/docs/esp_idf/html/dd/d3c/group__xTaskCreate.html
some watchdog info: espressif/arduino-esp32#595
You can do Digital Filtering with a MCU YouTube https://www.youtube.com/watch?v=k0LHp6jtHA8
High-Voltage Signal Conditioning for Low-Voltage ADCs http://www.ti.com/lit/an/sboa097b/sboa097b.pdf
Audio Out Circuit.
The PAM 8403 uses 5VDC from the ESP-32 Dev Board. Output is on DAC1 GPIO pin 25 of the ESP32 No DAC re-construction filter is used. Some aliasing on the sweep app is heard at high volume. Maybe a re-construction at the DAC output would help.
Audio In circuit: SparkFun Electret Microphone Breakout BOB-12758 https://www.sparkfun.com/products/12758 It has a freq. response to 10KHz and we sample at 8Ksps so a pole is added to its response to turn it into a 1st order (20 dB/Decade power drop-off) anti-aliasing filter (LPF) with 4KHz cutoff Add a 47pF cap across R5 (820K). From JP1-3 to U1-4 is easier. We would like a 2nd order filter here (40 dB/Decade power drop-off). https://cdn.sparkfun.com/datasheets/BreakoutBoards/Electret_Microphone_Breakout_v20.pdf
note: this circuit could be avoided by sampling at 16Ksps
mic audio is on analogRead(34) ADC_WIDTH_9Bit to more or less match the DAC 8 bits. An ADC_ATTEN_11db pad is set in order to run the mic at 3.3 Volts
3/10/18 Pink Noise example from portaudio.com modified for ESP32 Audio Framework. Statistics haven't been verified with spectrum analyzer.
3/15/18 modified sine generator with nkolban's BLE library. Select freq of sine over BLE (30-3000 Hz UTF-8 unsigned int string < 5 chars). https://youtu.be/MGNU7COFlT4
3/21/18 Voice Alert example. Audio samples stored as an array in the source. File edited in Audacity (Audacity.com) and saved as mono, 8Ksps, signed 16-bit PCM Microsoft WAV file. Converted to C source with xxd -i fn.wav fn.c
