Project Pages

Hardware Design

Physical components and assembly of Hu- nodes

Components Overview

  • Main Board: XIAO ESP32-S3 Sense
  • Sensors: Integrated Digital Microphone (MSM261D3526H1CPM on ESP32-S3 Sense), Artificial Pinna
  • Display & Input: Adafruit 2.4" Colour TFT Touchscreen, Tactile Switch
  • Audio Output: TPA2016 Amplifier + Speaker
  • Power: LiPo Battery
  • Communication: WIO-SX1262 LoRa Module (for Meshtastic), ESP32-S3 Onboard Wi-Fi

Hardware Technology Investigation

A comprehensive analysis of various hardware options for each component of the Hu- platform.

Microcontroller Comparison

Analysis of potential microcontrollers for the Hu- platform, considering factors such as processing power, power consumption, and development ecosystem.

Criteria Teensy 4.1 Raspberry Pi CM4 ESP32-S3 STM32L4/L5 NVIDIA Jetson Orin Nano
Processor 600MHz ARM Cortex-M7 1.5GHz Quad-core ARM Cortex-A72 240MHz Dual-core LX7 120MHz ARM Cortex-M4/M33 1.5-2.0GHz 6-core ARM Cortex-A78AE
Memory 1MB RAM, 8MB Flash 1-8GB RAM, eMMC/SD 512KB RAM, 16MB Flash 128-640KB RAM, 1MB Flash 8GB LPDDR5, 64GB eMMC
ML Capabilities Good (TensorFlow Lite) Excellent (Full ML frameworks) Good (TensorFlow Lite Micro) Good (X-CUBE-AI) Excellent (1024 CUDA cores, 32 Tensor cores)
Power Usage 40-300mA @ 3.3V 400-800mA @ 5V 10-80mA @ 3.3V 5-100mA @ 3.3V 5-15W (configurable)
Audio Features I2S, ADC, DMA I2S, USB, HDMI I2S, PDM, ADC I2S, SAI, PDM I2S, DMIC, HDMI
Connectivity USB only WiFi, BT, Ethernet, USB WiFi, BLE None built-in Gigabit Ethernet, USB 3.2
Development Ease Arduino, PlatformIO Linux, Python Arduino, ESP-IDF STM32Cube, Arduino Linux, CUDA, Python
Cost (USD) $20-30 $25-100 $5-10 $10-20 $249
Community Support Good Excellent Excellent Good Excellent

Microcontroller Details

Teensy 4.1

The Teensy 4.1 is a powerful and versatile microcontroller board that excels in real-time audio processing applications. Built around the ARM Cortex-M7 processor, it offers exceptional performance for its size and price point. Its extensive I/O capabilities and robust hardware design make it particularly suitable for audio projects requiring precise timing and multiple interface options.

Advantages

  • 600MHz ARM Cortex-M7 processor
  • Abundant I/O pins
  • Excellent real-time capabilities
  • Arduino IDE compatibility
  • Large memory capacity
  • Robust build quality

Limitations

  • No built-in WiFi/BLE
  • Higher power consumption
  • Limited ML-specific resources
  • Steeper learning curve
  • Higher cost than alternatives
Raspberry Pi CM4

The Raspberry Pi Compute Module 4 represents a significant step up in processing power, offering a full Linux-capable system in a compact form factor. Its quad-core processor and flexible memory options make it ideal for complex applications requiring substantial computing resources. The CM4 particularly shines in scenarios requiring sophisticated data processing, multimedia handling, or running full-scale machine learning models.

Advantages

  • Quad-core ARM Cortex-A72 CPU
  • Rich connectivity options
  • Up to 8GB RAM
  • Flexible form factor
  • Comprehensive software ecosystem
  • Hardware acceleration support

Limitations

  • High power consumption
  • Complex integration
  • Thermal management needs
  • Higher cost with peripherals
  • Size constraints
ESP32-S3

The ESP32-S3 is a highly integrated, low-power system on chip that combines robust wireless connectivity with efficient processing capabilities. This dual-core microcontroller is specifically designed for IoT applications, offering an excellent balance between power consumption and performance. Its native support for WiFi and Bluetooth LE makes it particularly well-suited for connected devices requiring wireless communication.

Advantages

  • Built-in WiFi/BLE
  • Dual-core processor
  • Rich peripheral set
  • Excellent low power modes
  • Arduino IDE support
  • Large community support

Limitations

  • Limited processing for complex ML
  • Memory constraints
  • Power management complexity
  • Potential thermal issues
  • Basic security features
STM32L4/L5

The STM32L4/L5 series represents STMicroelectronics' ultra-low-power microcontroller line, combining energy efficiency with advanced security features. These ARM Cortex-M4/M33-based processors are designed for applications where power consumption is critical. The L5 series, in particular, adds TrustZone security features, making it suitable for applications requiring enhanced data protection.

Advantages

  • High performance ARM cores
  • Low power consumption
  • Rich peripheral set
  • Robust development ecosystem
  • Enhanced security (L5)
  • Good scalability

Limitations

  • Steeper learning curve
  • Higher cost than ESP32
  • Limited ML examples
  • Complex for simple projects
NVIDIA Jetson Orin Nano

The NVIDIA Jetson Orin Nano is a compact yet powerful AI computing device that brings workstation-class performance to edge devices. Built on NVIDIA's latest architecture, it delivers exceptional machine learning capabilities through its 1024 CUDA cores and dedicated Tensor cores. This platform is specifically designed for advanced AI applications, computer vision, and complex data processing at the edge.

Advantages

  • Powerful 6-core ARM processor
  • 1024 CUDA cores for GPU acceleration
  • 32 Tensor cores for AI/ML
  • 8GB LPDDR5 RAM
  • Excellent for complex ML workloads
  • Full Linux support
  • Rich I/O options

Limitations

  • Higher power consumption (5-15W)
  • More expensive solution
  • Larger form factor
  • Requires active cooling
  • Complex development environment
  • Overkill for simple applications

Final Decision: ESP32-S3

After careful evaluation of all options, the XIAO ESP32-S3 Sense emerges as the most suitable choice for developing a simple ML audio service. Its integrated features, including a digital microphone and camera capabilities, provide a compact and efficient platform for this project. Here's a detailed breakdown of the decision:

Key Decision Factors

Primary Advantages
  • Cost-Effective ($5-10 for base S3, Sense slightly more): Budget-friendly solution for project development
  • Adequate Processing: Dual-core architecture sufficient for simple ML audio tasks
  • Integrated Peripherals: Built-in digital microphone (MSM261D3526H1CPM) and camera sensor (OV2640) simplify design and reduce component count.
  • Built-in Connectivity: Integrated WiFi/BLE for seamless IoT integration
  • Development Ease: Excellent IDE support and development tools
Additional Benefits
  • Community Support: Extensive documentation and libraries available
  • Power Efficiency: Low consumption (10-80mA) ideal for battery operation
  • Firmware Flexibility: Multiple development framework options
  • Quick Prototyping: Rapid development and testing capabilities

Important Considerations

While the ESP32-S3 is our chosen platform, it's important to note these limitations:

  • Processing power may limit complex ML model deployment
  • Memory constraints affect model size and complexity
  • Careful power management required for optimal performance
  • Thermal considerations under heavy loads


Possible future consideration of alternative microcontroller platforms and functionality:

  • Raspberry Pi CM4: Need for complex processing and larger datasets
  • Teensy 4.1: Require high real-time processing and extensive I/O
  • STM32L4/L5: Advanced security features and extreme power efficiency are critical


Audio Sensor Comparison

Evaluation of MEMS microphones and audio capture solutions suitable for environmental sound monitoring.

Criteria ICS-43434 SPH0645LM4H ICS-43432 MP34DT05
SNR 65 dB 65 dB 63 dB 64 dB
Power Usage 600 µA 600 µA 700 µA 650 µA
Interface I2S I2S I2S PDM
Frequency Range 20Hz-20kHz 20Hz-20kHz 20Hz-20kHz 20Hz-16kHz
Cost (USD) $4-6 $5-7 $4-6 $3-5

Audio Sensor Details

ICS-43434

Advantages

  • TBD

Limitations

  • TBD
SPH0645LM4H

Advantages

  • TBD

Limitations

  • TBD
ICS-43432

Advantages

  • TBD

Limitations

  • TBD
MP34DT05

Advantages

  • TBD

Limitations

  • TBD

Final Decision: Integrated Digital Microphone (MSM261D3526H1CPM)

The project will utilize the MSM261D3526H1CPM digital microphone that is integrated into the chosen XIAO ESP32-S3 Sense microcontroller. This approach simplifies the hardware design and leverages the capabilities of the Sense board.

Key Decision Factors

Primary Advantages
  • Integrated directly onto the XIAO ESP32-S3 Sense board.
  • Reduces external component count and complexity.
  • PDM digital output suitable for direct connection to the ESP32-S3's I2S peripheral.
  • Specified by Seeed Studio for the Sense board, ensuring compatibility.
Additional Benefits
  • Simplifies PCB design if a custom board were to be developed later.
  • Leverages the TinyML capabilities of the Sense board for voice AI.

Important Considerations

The microphone's performance will be influenced by the enclosure design. The Artificial Pinna will still be used to aid in monaural sound localisation.

  • Sensitivity to mounting and enclosure acoustics.
  • The specific characteristics of the MSM261D3526H1CPM (SNR, frequency response) will define the audio input quality.


This decision supersedes the previous comparison of discrete MEMS microphones, as the integrated solution is now preferred.



Power System Comparison

Analysis of power solutions including batteries and charging systems.

Criteria LiPo Li-Ion LiFePO4 Primary Battery
Capacity 2500mAh 3000mAh 2000mAh 3400mAh
Voltage 3.7V 3.7V 3.2V 3.6V
Lifecycle 300-500 cycles 500-1000 cycles 2000+ cycles Single use
Temperature Range 0° to 45°C -20° to 60°C -20° to 60°C -40° to 85°C
Cost (USD) $15-25 $20-30 $25-35 $5-10

Power System Details

LiPo

Advantages

  • TBD

Limitations

  • TBD
Li-Ion

Advantages

  • TBD

Limitations

  • TBD
LiFePO4

Advantages

  • TBD

Limitations

  • TBD
Primary Battery

Advantages

  • TBD

Limitations

  • TBD

Final Decision: LiPo Battery with USB Charging

The project will use a LiPo battery as the primary power source, with USB charging and battery management. Solar charging is not included in this version.

Key Decision Factors

LiPo Battery Advantages:
  • High energy density for compact size.
  • Well-supported by battery management ICs.
  • Rechargeable via USB.
Additional Benefits:
  • Simple integration with existing hardware.
  • Cost-effective and widely available.

Important Considerations:

  • Battery management IC required for safe charging and protection.
  • USB charging circuit must be reliable and accessible.


Possible future consideration of alternative power system methods:

  • Li-Ion or LiFePO4 batteries for different deployment needs.
  • Primary (non-rechargeable) batteries for ultra-long deployments.


Communication Module Comparison

Evaluation of wireless communication options for data transmission.

Project Requirements

A) Long Range Communication (node to node)
  • Low bandwidth requirement
  • High Signal Penetration
  • Low Cost
  • High Reliability
  • Very Low Power
B) Local Communication (node to gateway)
  • Wide bandwidth
  • Medium Signal Penetration

Modules Under Investigation

  • SX1262 (LoRa) - A high-performance, low-power long range transceiver designed for IoT applications. Features LoRa modulation for extended range and excellent interference immunity, making it ideal for node-to-node communication in challenging environments.
  • WM1302 - A LoRaWAN gateway module that enables connection between LoRa nodes and IP networks. Features high sensitivity and supports multiple channels for concurrent communication.
  • ESP32-S3 WiFi - Integrated WiFi solution offering high-speed data transfer and extensive protocol support. Well-suited for local communication where bandwidth and existing network infrastructure are priorities.
  • nRF24L01+ - Ultra low-power 2.4GHz transceiver offering good balance between range and power consumption. Simple to implement but limited by frequency band constraints.
  • XBee-PRO - Professional-grade module with built-in mesh networking capabilities. Offers reliable communication and good range but at higher cost and power consumption.
Criteria SX1262 (LoRa) WM1302 ESP32-S3 WiFi nRF24L01+ XBee-PRO
Range 15+ km 5+ km 100m 1km 1.6km
Power Usage (TX) 22mA @ +22dBm 400mA 240mA 11.3mA 120mA
Data Rate 0.018-62.5 kbps Up to 62.5 kbps 150 Mbps 2 Mbps 250 kbps
Frequency 150-960 MHz 470-510 MHz 2.4 GHz 2.4 GHz 900 MHz
Cost (USD) $7-10 $30 $8 Datasheet Datasheet -
Files Datasheet Datasheet Datasheet Datasheet -

Communication Module Details

SX1262 (LoRa)

Advantages

  • Ultra-low power consumption (22mA TX @ +22dBm)
  • High sensitivity (-148 dBm in LoRa mode)
  • Long range capability (15+ km line of sight)
  • Integrated power amplifier (+22 dBm output)
  • Wide frequency range support (150-960 MHz)
  • Multiple modulation types (LoRa/FSK/GFSK)
  • Advanced interference mitigation

Limitations

  • Lower data rates compared to WiFi (max 62.5 kbps)
  • More complex protocol implementation
  • Requires external antenna matching
  • Regional frequency regulations must be considered
  • Higher latency than traditional RF solutions
WM1302

Advantages

  • Multi-channel reception (8 channels)
  • High sensitivity (-139 dBm)
  • Gateway functionality built-in
  • Supports multiple modulation types
  • Concurrent channel monitoring
  • LoRaWAN protocol compatible

Limitations

  • Higher power consumption
  • More expensive solution
  • Requires additional IP connectivity
  • More complex setup
  • Larger form factor
ESP32-S3 WiFi

Advantages

  • High data transfer rates (150 Mbps)
  • Built-in to ESP32 (no additional hardware)
  • Widespread protocol compatibility
  • Easy integration with existing networks
  • Rich software ecosystem and libraries
  • Supports both 2.4GHz and 5GHz bands

Limitations

  • Higher power consumption
  • Limited range (100m typical)
  • Susceptible to interference
  • Requires WiFi infrastructure
  • Security considerations
nRF24L01+

Advantages

  • Very low power consumption
  • Low cost solution
  • High data rate (2 Mbps)
  • Simple protocol implementation
  • Small form factor
  • Auto-acknowledgment feature

Limitations

  • Limited range (1km max)
  • 2.4GHz band only
  • Susceptible to interference
  • No built-in encryption
  • Limited payload size
XBee-PRO

Advantages

  • Easy to configure and use
  • Built-in mesh networking
  • Good range (1.6km)
  • Reliable data transmission
  • Built-in security features
  • Multiple network topologies

Limitations

  • Higher cost
  • Moderate power consumption
  • Limited data rate (250 kbps)
  • Larger form factor
  • Proprietary protocol

Final Decision: SX1262 (LoRa for Meshtastic) & ESP32-S3 Onboard Wi-Fi

The project will utilize a dual-communication strategy:

  • LoRa (via WIO-SX1262 module): For robust, long-range, low-power mesh networking using Meshtastic between Hu- nodes.
  • Wi-Fi (ESP32-S3 onboard): For higher-bandwidth communication to a central web dashboard when in range of a Wi-Fi network.

Key Decision Factors for this Dual Approach:

SX1262 (LoRa for Meshtastic) Advantages:
  • Excellent range and penetration for node-to-node communication.
  • Very low power consumption suitable for battery operation.
  • Meshtastic provides an out-of-the-box mesh networking solution (with WIO-SX1262 or similar).
ESP32-S3 Onboard Wi-Fi Advantages:
  • High data rates for richer data transfer to a web dashboard.
  • Utilises existing Wi-Fi infrastructure.
  • Integrated into the main MCU, reducing component count for this feature.

Important Considerations:

  • Firmware will need to manage both communication interfaces and their respective power states.
  • Antenna design and placement will be important for both LoRa and Wi-Fi performance.


The WIO-SX1262 module is a specific product example that integrates the SX1262 chip and is suitable for Meshtastic.



Hardware Overview Diagram

Detailed hardware component layout and connections coming soon...

The diagram will showcase:

  • XIAO ESP32-S3 Sense and its peripheral connections (including integrated microphone and camera connector).
  • Power system layout (LiPo battery and management).
  • Connections for WIO-SX1262 LoRa module, TFT touchscreen, speaker, and switch.
  • Overall integration within the enclosure design.

Core Components

Processing Unit

  • XIAO ESP32-S3 Sense Microcontroller
    • Seeed Studio XIAO ESP32S3 Sense integrates a camera sensor, digital microphone, and SD card support. Combining embedded ML computing power and photography capability, this development board is a great tool to get started with TinyML (intelligent voice and vision AI).
    • Powerful MCU: 32-bit, dual-core, Xtensa processor up to 240 MHz.
    • Multiple development ports, Arduino / MicroPython supported.
    • Advanced Functionality: Detachable OV2640 camera sensor (1600*1200), compatible with OV5640; integrated MSM261D3526H1CPM digital microphone (PDM output, I2S: IO41-CLK, IO42-DATA).
    • Elaborate Power Design: Lithium battery charge management, 4 power consumption models (deep sleep as low as 14μA).
    • Great Memory: 8MB PSRAM and 8MB FLASH (16MB in Plus version), supports SD card slot for external 32GB FAT memory.
    • Outstanding RF Performance: 2.4GHz Wi-Fi and BLE dual wireless communication, 100m+ remote communication with U.FL antenna.
    • Thumb-sized Compact Design: 21 x 17.8mm, classic XIAO form factor, suitable for space-limited projects.
    • For more details, refer to the Seeed Studio XIAO ESP32S3 Getting Started Guide and XIAO ESP32S3 Sense Product Page.

Audio Capture

  • MSM261D3526H1CPM (Integrated Digital Microphone)
    • PDM digital output MEMS microphone with Multi-modes.
    • Integrated into the XIAO ESP32-S3 Sense.
    • Connected to ESP32S3 via I2S bus (IO41 - Clock, IO42 - Data).
  • Artificial Pinna
    • Aids monaural sound localisation

Display, Input & Output

  • Adafruit 2.4" Colour TFT Touchscreen
  • Tactile Switch (Auxiliary input)
  • TPA2016 Amplifier + Speaker (Audio output)

Communication

  • WIO-SX1262 LoRa Module (for Meshtastic)
  • ESP32-S3 Onboard Wi-Fi (for Web Dashboard)

Power System Design

Battery Management

  • Primary Power Source: 3.7V 2500mAh LiPo Battery.
  • Battery protection circuit (planned).
  • Low voltage cutoff (via BMS or software).
  • Charge level monitoring (implementation TBD).

XIAO ESP32-S3 Battery Usage Notes (from Seeed Studio)

The XIAO ESP32S3 series has a built-in power management chip that allows the XIAO ESP32S3 to be powered independently by using a battery or to charge the battery through the XIAO ESP32S3's USB port.

If you want to connect the battery for XIAO, Seeed Studio recommends purchasing a qualified rechargeable 3.7V lithium battery. When soldering the battery, please be careful to distinguish between the positive and negative terminals. The negative terminal of the power supply should be the side closest to the USB port, and the positive terminal of the power supply is the side away from the USB port.

XIAO ESP32-S3 with LiPo Battery
XIAO ESP32-S3 connected to a LiPo battery. (Image credit: Seeed Studio Wiki)

Note from Seeed Studio:
Since all GPIO pins of the XIAO ESP32S3 are assigned their own functions, we do not have a GPIO configured for the battery pin. This means that we cannot get the battery voltage at the software level by reading the analog value of one of the GPIOs. If necessary, you can consider connecting the positive and negative terminals of the battery to two of the pins to measure the battery voltage.

XIAO ESP32-S3 with LiPo Battery
XIAO ESP32-S3 connected to a LiPo battery. (Image credit: Seeed Studio Wiki)
Charging Indicator Light

Seeed Studio also designed a red indicator light for battery charging, which informs the user of the current battery charging state:

  • When XIAO ESP32S3 is not connected to the battery, the red light comes on when the Type-C cable is connected and goes off after 30 seconds.
  • The red light flashes when the battery is connected and the Type-C cable is connected for charging.
  • When connecting Type-C to charge the battery fully, the red light turns off.
XIAO ESP32-S3 Charging Indicator LED
XIAO ESP32-S3 charging indicator LED. (Image credit: Seeed Studio Wiki)

Enclosure Design

Specifications

  • IP65 rated weatherproof enclosure
  • UV-resistant PLA/PETG material
  • Ventilated design with moisture protection
  • Easy access for maintenance
  • Mounting points for solar panel

Assembly Instructions

Step-by-Step Guide

  1. Prepare the main board
    • Flash bootloader
    • Test basic functionality
    • Attach headers and connectors
  2. Install sensors
    • Mount MEMS microphone
    • Connect I2S interface
    • Test sensor readings
  3. Power system setup
    • Connect solar controller
    • Wire battery management
    • Test charging circuit
  4. Final assembly
    • Mount in enclosure
    • Seal connections
    • Perform system test

Bill of Materials

Component Quantity Notes
XIAO ESP32-S3 Sense 1 Main controller board with integrated camera and microphone
Artificial Pinna 1 Custom (e.g., 3D Printed)
WIO-SX1262 LoRa Module 1 For Meshtastic communication
Adafruit 2.4" Colour TFT Touchscreen 1 Display and primary input (ILI9341 controller)
TPA2016 Amplifier 1 For speaker output
Small Speaker (e.g., 1W, 8 Ohm) 1 Audio feedback
Tactile Switch 1 Auxiliary input
LiPo Battery (e.g., 3.7V, >=1200mAh) 1 With protection circuit