Smart Pi Radio Assistant

• Introduction to Smart Pi Radio Assistant
  • Features
  • Why
• Required Hardware
• Setting up Signal Processor
  • Install dependencies
  • Set up Spotify
  • Run main
• Setting up the Raspberry Pi
  • Install operating system on the Pi
  • Set up Internet Access to Pi via WiFi and SSH
  • Install rpitx
  • Get python script to transmit audio
  • Connect a wire to GPIO pin 4
• Instructions for Setting up Pi as Signal Processor
  • Modify USB buffer size
  • Install conda
  • Install onnx runtime
  • Install SDR stuff
  • Install pytorch
• Possible Improvements

Introduction to Smart Pi Radio Assistant

Smart Pi Radio Assistant is a smart home assistant that you can access via voice commands on a walkie talkie! Turn your Raspberry Pi into an IOT device that can control devices and reply to you via radio waves. It's fully customizable!

Features

Voice Command	Example	What it does	Demo
play	play castaways	Plays Spotify song	Link
weather	weather seattle	Pi replies with current weather forecast for city (currently hardcoded response) 3 times	Link
echo	echo how are you?	Pi echoes back message 3 times

Why

• Current smart home assistants are proprietary which means it is near impossible to customize your home assistant and add new commands.
  • The Smart Pi Radio Assistant could be used as an interface to Home Assistant, giving an infinite amount of possibilities due to huge community support.
  • For example, use Smart Pi Radio Assistant to open your garage door, turn off your lights in your house, or detect if someone is at home.
  • The customizability of Smart Pi Radio Assistant + Home Assistant beats Siri, Alexa, Cortona, and Google Assistant!
• Current smart home assistants require you to be in hearing distance of the home device. This home assistant works from much farther range (atleast a house away, potentially much farther).

Required Hardware

• Computer (fast enough to process SDR samples in realtime)
• SDR
• Walkie Talkie
• Raspberry Pi 4
• SD Card + SD Card Reader
• Ethernet cable or WiFi for LAN w/ Pi
• (Recommended) Temporary keyboard, mouse, monitor
• (Recommended) Antenna material (e.g. jumper wire)

An example final set-up

• Raspberry Pi 4 (w/ case + 7.8 inch female-to-male jumper wire antenna)
• Laptop (MacBook Pro) + USB Hub
• Walkie Talkie (Retevis RT-5R)
• SDR (Nooelec NESDR SMArt XTR SDR + SMA connector & Walkie Talkie Antenna)
• Long-haired gray cat (Gumbo)

Setting up Signal Processor

The 'Signal Processor' is a computer that is fast enough to process radio signals via an SDR (i.e. without losing samples + realtime processing).

Install dependencies

git clone https://github.com/dominickta/pi-radio-assistant.git
cd pi-radio-assistant

Optionally, create a Python virtual environment for this project:

python3 -m venv myprojenv
source myprojenv/bin/activate
# to deactivate the virtual environment:
deactivate

Install python dependencies

pip3 install -r requirements.txt

Set up Spotify

In Spotify Developer portal, create an app.

Then setup a .env environment file

CLIENT_ID = 'from_spotify_dev_portal'
CLIENT_SECRET = 'from_spotify_dev_portal'
REDIRECT_URI = 'http://localhost:8000'
USERNAME = 'your_spotify_username'

Trigger Spotify setup by running the following command. It should redirect you to a web browser and ask you to paste the redirection link into the terminal.

python3 myspotify.py

Run main

Start the program by running

python3 assistant.py

If you configured Spotify already, the spotify feature should be available via the voice command "play " on frequency of 433.5MHz via a walkie talkie. (Note: requires you to have Spotify open on a computer with the user in config file logged in).

Otherwise, you need to set-up the Raspberry Pi and have it turned on before you can use the other commands.

Setting up the Raspberry Pi

Install operating system on the Pi

1. Install Raspberry Pi Imager on a computer
2. Plug in SD card into SD card reader
3. Insert SD card to be used to hold operating system and storage for the Raspberry Pi
4. Open Raspberry Pi Imager
5. Choose an Operating System -> Raspberry Pi OS (Other) -> Raspberry Pi OS Lite (32-bit) no desktop environment
6. Choose SD Card
7. Select Write
8. Once writing is done, insert SD card into Raspberry Pi.
9. Plug in monitor, keyboard, and mouse before plugging in power for Raspberry Pi
10. Once it successfully boots, you should be able to login with the following credentials user: pi password: raspberry

Good to know facts:

• Raspberry Pi 4 has a unique ARM cpu architecture (specifically armv7l) which means that you might have to build programs from source (like Pytorch which takes 6-12 hours) or find community prebuilt binaries.
• armv7l is a 32-bit processor (anything ARMv8 and above is 64-bit)
• rpitx requires our Raspberry Pi 4 to be run in headless mode (i.e. without monitor, keyboard, mouse connected; connected via SSH), which is why we use the Lite OS. We only use keyboard, mouse, monitor for ease of set-up. It's possible to set-up without these peripherals but its not worth figuring out.
• Raspberry Pi OS was previously called Raspbian because it is a Debian-based operating system, so you might hear it referred to as Raspbian (like in the rpitx repo)

Issues I've faced in the past:

• I used a TV monitor as a display which meant there was a lot of compatibility issues I needed to figure out. This involved editing config files on the SD card on my computer before plugging into the Pi.
• Trying to provide Internet access to Pi via ethernet cable connected to a Macbook involved turning on "Internet Sharing" over WiFi in the Sharing settings in MacOS
• Trying to install the Raspberry Pi OS desktop version will cause problems with rpitx. rpitx installation sets GPU to 250MHz which I think is underclocking, which is required for rpitx to be stable. This means that running rpitx + monitor causes desktop and/or rpitx to crash. I've had success installing the non-desktop Raspberry Pi OS and then manually installing a lightweight desktop afterwards if you insist on having a desktop interface.

Set up Internet Access to Pi via WiFi and SSH

Following directions in here

1. sudo raspi-config
2. Setup localisation in Localisation Options
3. Enable SSH via Interfacing Options -> SSH
4. Setup SSID & passphrase in System Options -> Wireless LAN
5. Exit
6. Test connection with ping google.com

If you don't have a temporary keyboard/mouse/monitor available to setup Raspberry Pi, there are 'headless' configuration instructions online. It involves editing network config files and messing around with the SD card before plugging it into the Pi.

After this, you should only connect to Raspberry Pi via SSH (over WiFi or Ethernet), aka run the Raspberry Pi in 'headless' configuration. Otherwise there might be issues with rpitx and monitor/GPU crashing (even if your monitor is only displaying a terminal).

Assuming your Pi is the only Pi on your local network, you can connect via

ssh pi@raspberrypi.local

Otherwise you can connect to your Pi by finding its IP address on your local network.

Install rpitx

Follow the up-to-date instructions in the README here.

As a convenience, the instructions as of June 5th, 2021 are listed in this section.

sudo apt-get update
sudo apt-get install git
git clone https://github.com/F5OEO/rpitx
cd rpitx
./install.sh
sudo reboot

Test out the installation by running this command in the rpitx folder:

sudo ./pifmrds -audio sampleaudio.wav -freq 434

This will transmit the included "sampleaudio.wav" file via the FM radio protocol on a frequency of 434MHz.

This should work without an antenna (~1 foot radius) but range will increase with a simple jumper wire antenna to up to 100 ft.

There are a bunch of other fun programs in this repo such as

• transmitting images via the waterfall in a SDR receiver
• replay signals with SDR + Pi
• transmitting data to pagers via pocsag

Major Issues: You must be running your Pi on headless mode while using rpitx (there are some workarounds though) F5OEO/rpitx#244 F5OEO/rpitx#231

Get python script to transmit audio

In the /home/pi/ directory run this command to download the script that will play a .wav file and terminate automatically.

wget https://raw.githubusercontent.com/dominickta/pi-radio-assistant/main/pifmrds.py

Connect a wire to GPIO pin 4

Connect a jumper wire to GPIO 4, pin 7(reference diagram here).

With 7.8 inch jumper wire, radius of ~70-100ft with varying obstructions; radius of ~60ft should have strong signals.

• not a perfect circular radius, but close
• jumper wire was laying flat on a table on the 3rd floor of a house
• signal was captured at street level with walkie talkie
• rough measurements were from Google Maps and only captured horizontal distance (so actual distance could be more)

Without jumper wire, able to capture signal with a strong half-foot radius.

Instructions for Setting up Pi as Signal Processor

I spent a lot of time trying to get the Raspberry Pi to stream and process samples from an SDR with reasonable performance, but was unable to.

I believe that the Raspberry Pi 4 should be able to process these signals in real time based on what I've seen people do on the Internet. The following instructions were created when I was still trying to get it to work. So if you're trying to get your Pi to stream & process radio waves, these instructions are probably useful.

Modify USB buffer size

The SDR needs to stream samples via the USB port and buffers are allocated to handle that. We need to manually increase the buffer size and also ensure this change persists throughout reboots. The default on Linux is only 16MB which is not enough for our SDR streaming.

The following instructions are an offshoot of this guide.

1. Open startup command file in a text editor

Open /etc/rc.local. This file contains commands that will be run on boot. We'll add a command to this file that will change the usb buffer size to 1024MB everytime on boot.

sudo vim /etc/rc.local

2. Edit the file

/etc/rc.local should already exist on your Pi. If so, add this line right before exit 0 (the last line):

echo 1024 > /sys/module/usbcore/parameters/usbfs_memory_mb

If /etc/rc.local doesn't already exist, make this the content of your file:

#!/bin/sh -e
echo 1024 > /sys/module/usbcore/parameters/usbfs_memory_mb
exit 0

3. Save changes and reboot Save your changes to the file and then reboot the Pi. You can reboot with sudo reboot.

4. Check your work

If it worked then this command should spit out 1024.

cat /sys/module/usbcore/parameters/usbfs_memory_mb

Install conda

Just in case you need conda on the Raspberry Pi, this is useful. Currently anaconda doesn't support armv7l cpus.

Install conda following these Stack Overflow instructions

Note that this works because we are getting the armv7l version. You can confirm that you have the correct CPU architecture by running uname -m

wget http://repo.continuum.io/miniconda/Miniconda3-latest-Linux-armv7l.sh
sudo md5sum Miniconda3-latest-Linux-armv7l.sh
sudo /bin/bash Miniconda3-latest-Linux-armv7l.sh

Important: Only follow the first set of instructions (listed below); we already have Python3.7 via Raspberry Pi OS installation Important: follow these directions for the prompts carefully

Accept the license agreement with yes

When asked, change the install location: /home/pi/miniconda3

Do you wish the installer to prepend the Miniconda3 install location to PATH in your /root/.bashrc ? yes

Now add the install path to the PATH variable:

sudo nano /home/pi/.bashrc

Go to the end of the file .bashrc and add the following line:

export PATH="$PATH:/home/pi/miniconda3/bin NOTE: This is different from the Stack Overflow path. We want to append miniconda3 because we want to avoid using miniconda3's version of python3 (python3.4), and use the default python3.7 that came with Raspberry Pi OS

Save the file and exit.

To test if the installation was successful, open a new terminal and enter

conda

If you see a list with commands you are ready to go.

Install onnx runtime

ONNX Runtime is required to run Pytorch models, because currently Pytorch uses Intel Math Kernel Library (mkl) which doesn't work on an ARM CPU.

https://github.com/nknytk/built-onnxruntime-for-raspberrypi-linux/tree/master/wheels/buster

wget https://github.com/nknytk/built-onnxruntime-for-raspberrypi-linux/raw/master/wheels/buster/onnxruntime-1.8.0-cp37-cp37m-linux_armv7l.whl

pip3 install onnxruntime-1.8.0-cp37-cp37m-linux_armv7l.whl

Install SDR stuff

Install with this command:

sudo apt-get install gr-osmosdr

Plug in your SDR into your Pi and test if it worked via this command:

rtl_test -t

Install pytorch

Pytorch with onnx runtime is NOT a bottleneck. For example, the silero models run surprisingly fast.

Pytorch does not provide builds for ARM cpus, so we either have to build it ourselves or find someone who pre-built it for the ARM cpu.

I attempted building it myself (before knowing someone prebuilt it) and it was compiling for ~6 hours before it eventually didn't work for some reason. I tried this twice, wasted 12 hours, and then found this slightly outdated guide that works with heavy modifications.

1. Check your ARM processor configuration via uname -a. It should contain "armv7l GNU/Linux" at the end.

2. Check we have the correct python3 version via python3 --version. It should say Python3.7

3. Download recent prebuilt Pytorch wheels

Download torch 1.7 and torchaudio 0.7 (more recent version) wheels. The guide I linked above that we are following has a link to an older version of torch. This repo has more recent versions uploaded.

We need to download the most recent available torch and torchaudio file that ends in "linux_armv7l.whl"

As of June 5th, 2021, these are the correct download links:

https://github.com/isakbosman/pytorch_arm_builds/raw/main/torch-1.7.0a0-cp37-cp37m-linux_armv7l.whl
https://github.com/isakbosman/pytorch_arm_builds/raw/main/torchaudio-0.7.0a0%2Bac17b64-cp37-cp37m-linux_armv7l.whl

Inside home directory, run

wget https://github.com/isakbosman/pytorch_arm_builds/raw/main/torch-1.7.0a0-cp37-cp37m-linux_armv7l.whl

wget https://github.com/isakbosman/pytorch_arm_builds/raw/main/torchaudio-0.7.0a0%2Bac17b64-cp37-cp37m-linux_armv7l.whl

4. Install the files

sudo apt install libatlas3-base
sudo apt-get install -y libopenblas-base
pip3 install INSERT_TORCH_WHEEL_FILE_DOWNLOADED_IN_STEP_3
pip3 install INSERT_TORCH_AUDIO_WHEEL_FILE_DOWNLOADED_IN_STEP_3

5. Check if installed properly Start a python3 session by typing python3 Run these commands (exit session by entering quit()):

import torch
import torchaudio
import numpy

Other useful Pytorch links

• pytorch/pytorch#41592
• https://stackoverflow.com/questions/66693151/how-can-i-emulate-run-pytorch-model-that-uses-aten-stft-implementation-on-arm-ba
• https://mathinf.eu/pytorch/arm64/2021-01/

Possible Improvements

• Get everything working on the Raspberry Pi as a standalone-package
  • 
• Improve python scripts to follow best coding practices
  • no spaghetti code
  • error handling for edge cases
  • don't hard-code file paths
  • implement way to cleanup and exit program
• Improve code to make it easier to add new features to the assistant
• Figure out how to ensure Tx & Rx don't interfere with each other
  • Currently audio-replies transmitted by the Pi can be perceived as commands from a user by the signal processor. This could lead to an infinite loop/unintended commands to be executed.
  • (easy) Make Tx & Rx on diff frequencies? (assuming walkie talkie has dual band & turns on 'listening mode')
  • (hard) Implement self-interference cancelling: https://kumunetworks.com/
• Connect Raspberry Pi to Home Assistant + IOT devices to showcase more cool stuff
• Measure effects of different antenna materials + orientations for the Raspberry Pi
• Create a lowpass filter because apparently the Pi transmits a lot of harmonics
• Improve speech-to-text and text-to-speech
  • Could involve improving machine learning models
  • Could involve improving radio waves audio quality