Summary of the Insertomatic 6000
- One Raspberry Pi 3 + three Pi Zero 2 Ws fully networked using USB Gadget
- 6 analogue RF channels in total
- ...4 of which are connected directly to the composite video outputs of the Raspberry Pis
- ...2 of which are connected to the two VBIT-Pis
- Each VBIT-Pi takes an external composite video source (such as that from a PVR) and adds a teletext service to the vertical blanking interval of the picture; the teletext service can be set independently from the service which is on the corresponding Pi's own output
- A controller board which switches the audio channel of the two VBIT-Pi channels to an auxiliary input when the external video input is switched off, and which allows the three Pi Zero channels to be switched between their own audio and the audio of the Pi 3 under control of each Pi Zero
- Audio selector switches, which select media playback on each of the four Raspberry Pis
- LCD, which displays information about the playing media, and allows the modulator channels to be conveniently changed
Safe shutdown
I decided it would be useful to incorporate a shutdown function into the Pi Pico-based controller so that all the Pis could easily be safely shut down without having to log into each Pi in turn and run sudo poweroff as I had been doing so far. I used this Sparkfun example to make the Pis shut down or reboot depending on the length of the pulse received on an input pin, and added some menu options to the controller which enable the user to issue a shutdown or reboot command and holds an output pin low for the correct amount of time.
A DuPont wire harness was made which connected pin 6 on each of the four Pis together and to a single I/O pin on the Pi Pico.
I found a flaw in the Sparkfun code which was that when attempting to issue a shutdown command, execution of the script does not terminate immediately, so if the pin goes high before the script has been terminated by the system shutdown, the restart command will be executed as well, and this overrides the earlier shutdown command, resuting in some of the Pis turning back on. I added a quit command immediately below the call to the shutdown function to resolve this problem.
Noise
My fear with this problem was that the wired construction simply resulted in poor grounds and that there would be no way around it except for creating a massive PCB with everything on it to provide a good ground plane. The fact that adding a single ground wire between a Pi and the modulators made little difference supported this view, but fortunately it wasn't necessary. All the Pis needed individual ground wires to the modulators, and once this was done, the picture quality was improved dramatically to the extent that it was now almost perfect.
Thermals
The Insertomatic 6000 now has a lot of stuff inside a case with very poor airflow, so I was concerned that the Pis would get too hot, especially when used for something more than just internet radio.
A slot was cut in the base acrylic sheet to hold a fan and some holes were drilled in the bottom panel to the left of the Pi 3 to encourage air to flow across the small heatsink on the Pi 3.
The 35mm fan runs at high speed directly from a 5V power supply and makes a lot of noise. Whilst very effective, the noise and the 4W of power consumption were unacceptable, so diodes were added in series until a low level of noise was achieved without affecting its effectiveness very much, and the 2W power consumption was better if still a little disappointing.
Temperature of the Pi 3 measured using vcgencmd measure_temp with all Pis powered up and running the Slideomatic software and internet radio:
- Case closed, no fan: 46.2°C
- Case open, no fan: 44.0°C
- Case closed, fan at full speed: 33.2°C
- Case closed, fan at low speed: 36.5°C
With the Pi 3 running stress for 15 minutes (using the command shown in Part 3) but conditions otherwise the same:
- Case closed, no fan: 72.0°C, soft temperature throttling activated
- Case closed, fan at full speed: 49.9°C
- Case closed, fan at low speed: 49.9°C
vcgencmd get_throttled was added to the for loop from where stress was run to allow monitoring of the throttling status too.
With the Pi 3 under load, the difference between the fan running fast or slow is eliminated which is surprising, but the results at least show that having the fan lowers the temperatures significantly. However, even at the lower speed, the fan is still audible when the room is silent.
I considered adding a simple MOSFET circuit which allows the Raspberry Pi to control the fan via PWM; a Python script could then automatically turn the fan on when a certain temperature is reached or even use PWM to vary the fan speed and possibly drive it at an even lower speed when idle. However, I didn't want to risk introducing another place where noise could get into the pictures, so I decided to keep things relatively simple and add an extra AliExpress buck converter board which would only power the fan; this reduced the fan's impact on mains power consumption to just 0.5W, and the voltage adjustment potentiometer could be used to adjust the voltage to a point where the fan was very quiet but would still reliably start, and this happened to be at 1.82V. The temperature of the Pi 3 running stress was then 55.8°C and idle 38.6°C, a little warmer than when the fan was spinning faster but still much better than no fan at all, so I think it's sufficient to just have the fan running at this speed all the time and not bother adding a speed control.
Keeping all services up to date
The vbit2 package includes a script which keeps the currently enabled service up to date. In the case of the two Pi Zeros which each have a VBIT-Pi, it would be useful if when a different service is inserted using the VBIT-Pi to that inserted in the Pi's own composite video output, it is also kept up to date. The script was modified to update every installed service rather than just the enabled one, which works well. Unused services should be uninstalled to avoid updating unused services.
And that wraps up all the work on the hardware of the Insertomatic 6000. However, it's a project that will never be completely done, because there's always room for future improvements. One idea I'd like to implement at some point is visualisers for the audio playback instead of static images in the slideshow.
