|Project ESP audio sensor|
|ESP32-based audio sensor|
This project is about creating an audio level meter, for example as an environmental noise measuring node in a citizen science project or as a standalone "decibel meter".
Marcel Meek built a sound sensor based on an ESP32 board with built-in LoRa-chip. It samples audio from a digital I2S microphone, analyses it into sub-bands, then forwards it over LoRaWAN. His code can be found here: https://github.com/meekm/LoRaSoundkit
From TheThingsNetwork, the data is picked up by a kind of 'integration' that sends it to sensor.community. This is handled by my noise forwarder Java project.
- Bluetooth interface. The ESP32 supports bluetooth as an RF technology. It would be nice to be able to quickly see the current/most recently measured data. This can be shown on the display. Perhaps even more convenient, this can be exposed over bluetooth for display on a mobile phone.
- See GATT_Specification_Supplement_v5 paragraph 3.151, it has a noise characteristic with 1 dB resolution.
- 0x27C3 is the GATT unit for sound pressure, see https://btprodspecificationrefs.blob.core.windows.net/assigned-values/16-bit%20UUID%20Numbers%20Document.pdf
- 0x181A is the GATT environmental sensing service apparently
- 0x2A6E/F is temperature/humidity apparently, specified in https://btprodspecificationrefs.blob.core.windows.net/assigned-values/16-bit%20UUID%20Numbers%20Document.pdf
- See https://programmaticponderings.com/2020/08/04/getting-started-with-bluetooth-low-energy-ble-and-generic-attribute-profile-gatt-specification-for-iot/
Measuring audio as a citizen science project
The concept is the following:
- A small box containing a microphone outside your house measures environmental sounds (traffic, etc), for example it takes a 1 second audio recording every 10 seconds
- A spectral analysis is made of the audio (separating it in different frequency bands), calculating sound intensity for each individual band
- Every 5 minutes, the measured sound intensity is sent to a central server on the internet. Because we only communicate intensities, this does not reveal audio fragments (conversations for example).
- The central server takes the measured intensities and can do corrections, like a microphone specific correction, or a correction to apply A-weighting
- The central server forwards the measurements to sensor.community, from there they are automatically picked up by RIVM
External things to investigate:
- How is audio being evaluated in general, are there norms for it?
- how often should audio be sampled? 1 second/minute? 10 seconds/min? How long should the audio fragment be?
- is it a valid assumption to use full octaves? or should the spectrum be divided in smaller parts?
The audio spectrum is divided up into octaves and the total energy in each octave is calculated. We can then easily apply sensor/housing specific corrections, do A weighting, etc.
Subjective audio levels are generally calculated on a logarithmic scale in dB using "A-weighting". A-weighting calculates a subjective loudness level from a physical loudness, applying a correction factor for each (part of an) octave band.
|Octave start||Octave center||Octave end||Remark|
|-||-||22627 Hz||sample rate|
|5657 Hz||8000 Hz||11314 Hz|
|2828 Hz||4000 Hz||5657 Hz|
|1414 Hz||2000 Hz||2828 Hz|
|707 Hz||1000 Hz||1414 Hz|
|354 Hz||500 Hz||707 Hz|
|177 Hz||250 Hz||354 Hz|
|88 Hz||125 Hz||177 Hz|
|44 Hz||63 Hz||88 Hz|
|22 Hz||31 Hz||44 Hz|
The sampling rate is chosen such that the FFT bands align exactly with the bands as defined in the A-norm.
The energy in each octave is calculated by applying an FFT (fast fourier transform) on the audio data. The FFT takes in real values and outputs complex values. The intensity in each octave band is by summing the energy in a set of FFT 'bins'. The energy in each bin is calculated as the real part squared plus the imaginary part squared. The division of the audio spectrum for the FFT is chosen so it matches the octaves used in the A-weighting as described above. I plan to use the 'flat-top' window because it has good properties for measuring power levels.
Investigation into FFT libraries:
- arduinoFFT does all calculations in double, severely limiting the audio buffer size. For each sample, you need two doubles, so that's 16 bytes per sample. I can use about 6000 samples in the audio buffer. Using floats gives me twice the audio buffer size.
- Alternatively, esp32-fft looks nice, it uses floats instead of doubles and is optimised for esp32, but it uses malloc internally ... Also I can't use it as an Arduino library because it doesn't have the arduino library structure (with a src dir, examples dir, library.json file, library.properties file, etc) Maybe I can still use it by copying it in my sketch.
- Overview of octave bands https://www.engineeringtoolbox.com/octave-bands-frequency-limits-d_1602.html
-> I went with ArduinoFFT, modified it to use float instead of double
Commercially available meters:
Fairly typical specs:
- dynamic range: 30 - 130 dB
- accuracy: 1.5-2 dB
- frequency range: 31.5 Hz - 8 kHz (!)
- norm: EN 61672-1
The physical device consists of:
- an ESP32, it has an I2S digital audio input for sampling data from a digital microphone and a WiFi interface to communicate things to the internet. It has more internal RAM than an ESP8266 (for example), this helps to take a larger audio sample and do analysis on it. It is still comparatively simple, cheap, easy to flash and powerful enough to do communication over WiFi and do audio analysis.
- a digital I2S microphone, like the INMP441 (datasheet)
My initial prototype uses an NodeMCU ESP32.
The microphone is connected to the microcontroller as follows (old):
- INMP441 GND to ESP32 GND
- INMP441 VDD to ESP32 3.3V
- INMP441 SD to ESP32 A4/32
- INMP441 SCK to ESP32 A16/14
- INMP441 WS to ESP32 15
- INMP441 L/R to ESP32 GND
The connection carries only digital signals (max 3 MHz or so). No sensitive analog electronics are needed, the microphone and the microcontroller are simply connected using "dupont" wire.
The I2S clock signal is 64 times higher than the sample clock, so at a sample rate of 44100 Hz, this means 2.8 MHz. This might be a bit high for a random wire, probably we should keep this connection short.
Heltec LoRa32 v2
Connection of an Heltec LoRa32 v2 will be described here.
Initial code for sampling audio from the digital microphone can be found on github.
What the software should do:
- Take an audio measurement from the microphone at a regular interval (say 1 second every 10 seconds)
- On the recorded audio, perform a 4096-point real->complex FFT with a windowing function (flat-top for example).
- Calculate power for each FFT-bin (Im-squared + Re-squared) and sum up bins per octave.
- Calculate statistics, e.g. minimum/average/maximum over a 5 minute interval and convert to a logarithmic scale (decibels)
- Every 5 minutes, send the statistics to the network using WiFi or LoRa
The network receives the raw decibel values and can apply corrections for specific microphones, do A-weighting, etc.
Backend code for processing and forwarding the data from the sensor of Marcel Meek can be found at https://github.com/bertrik/noiseforwarder.
Use as a citizen science sound/noise sensor
The sensor.community node has support for a "DNMS" sound sensor, see https://sensor.community/en/sensors/dnms/
We emulate the network protocol used by a sensor.community node for DNMS, sending a "LAeq", minimum and maximum value.
In my opinion, the DNMS module has the following disadvantages:
- the module produces a single value with a minimum and a maximum value (not a spectrum), no spectral corrections are possible
- the communication interface is I2C, where the sound sensor is I2C slave, which is more difficult to implement than (for example) a 9600 bps serial interface
- the communication protocol is a bit weird, e.g. sends an 8-bit CRC for every 16-bit data word
- it needs a custom PCB
It would be nice to be compatible with this sensor, so an ESP32 audio sensor could be a drop-in replacement for the "official" DNMS module.
- as reverse engineered from the sensor.community firmware:
- device is at i2c address 0x55 (7-bit or 8-bit address?)
- commands exchanged are 16-bit
- 1 DNMS_CMD_RESET
- 2 DNMS_CMD_READ_VERSION
- 3 DNMS_CMD_CALCULATE_LEQ
- 4 DNMS_CMD_READ_DATA_READY
- 5 DNMS_CMD_READ_LEQ
- a command looks like this:
- 16-bit (big endian) command id
- a variable number of 16-bit data words, each with an 8-bit CRC
- getting the data from the DNMS:
- send the DNMS_CMD_CALCULATE_LEQ command
- repeat DNMS_CMD_READ_DATA_READY command until it indicates that a measurement is available
- send the DNMS_CMD_READ_LEQ command, this results in three values: a, a_min and a_max
- if an error occurred, send the DNMS_CMD_RESET command
Conclusion: Use of the I2C slave protocol makes it very difficult to use an ESP32 to emulate a DNMS sensor.
The protocol is basically the same as the particulate matter concentration protocol, with different "value_type" fields in the JSON.
- REST with JSON
- HTTP POST request to "https://api.sensor.community" with path "/v1/push-sensor-data/"
- the POST contains the following HTTP headers:
- HTTP header "X-Sensor" contains the unique hardware id, for an ESP32-board, an id of type "esp32-<number>" would be appropriate
- HTTP header "X-Pin" with value 15
- HTTP header "Content-type" with value "application/json"
- the POST body contains the following fields:
- "software_version", description of the software version
- "sensordatavalues", list of structures:
- "value_type" = "noise_LAeq", "value" = value in dB
- "value_type" = "noise_LA_min", "value" = value in dB
- "value_type" = "noise_LA_max", "value" = value in dB
- https://github.com/hbitter/DNMS github page