9

I know the publication Biosignal PI, an Affordable Open-Source ECG and Respiration Measurement System which uses Raspberry Pi A+ / B+ as a component isolated by ADums, since RP is not itself a medical device. The final system has been accepted to be used in some medical testing in Sweden where regulations are very strict. The specific health-related status of the system is TODO. I would like extend the project by doing FFT computation in the Raspberry's own GPU, BCM2835, as described in the blog post Accelerating Fourier Transforms Using the GPU in studying autonomic dysfunctions. However, I am unsure if the model Pi 1 A+ is enough. Raspberry homepage is about

We recommend the Raspberry Pi 2 Model B for use in schools: it offers more flexibility for learners than the leaner (Pi 1) Model A+, which is more useful for embedded projects and projects which require very low power.

There are strict isolation policies in ECG systems, which is why I am thinking the Raspberry 2 B model may not be suitable. I am especially interested in the power management of the different models in the GPU computation.

Basic Safety Characteristics

  • Power-off power is 20-30 mA (0.1W) (here) but 1.0W (here) when USB mouse and keyboard connected, until you physically disconnect the power.
  • Maximal power-off power in all devices? 10x difference between no-devices and devices is rather high.
  • Lowest idle power in A+, B+ and Zero.
  • GPU-Power stability in all models? Tests by shooting video and rendering video (here) where the video recording is done by computing FFT in the GPU.
  • GPU-power usage is different between RBi B+ and other models because of the different power circuitry (here).
  • At least two level isolation. 1st level ADAS1000, creepage air clearance and SP720. 2nd level [conjuncture] negative feedback to the change in the viscoelastic characteristic.
  • ADAS1000 power dissipation is 41 mW (here) which fluctuates as a function of CPU usage (0,1.0). How does it fluctuate over N clocks? Unknown. Power measurement error is normal distributed.

No RP is a medical device. RP must be isolated from the ECG front-end (etc power and SPI commiserations) which is done by ADums in Biosignal Pi design (Farhad).

Isolation Strategies of ECG front-end from RP

  1. Assume Pi B+ could behave like any other component. (used in the publication)
  2. Switching to Pi 2 B should not alter the situation but the maximum power of the circuit unknown and dependent probably on ADAS1000.
  3. When proving the Pi is sufficiently isolated by the ADAS1000, the assumption that Pi behave like any other component must hold.
  4. If the RPi suddenly decides to act as a 0 Ohm resistor between the power supply and the patient, ADAS1000BSTZ should ensure the isolation. (1-3) but the upper limit of the power is TODO in the circuit.
  5. If RPi catches fire, isolation of the system, Creepage air clearance and SP720.
  6. 0.5W extra power draw is safe so RPi zero and A+ accepted. How sufficient is 0.75W power? Limitations of RPi B+ in the power sense?
  7. ADuM4400 safely withstands 5000 Volt for 60 seconds. The power supply isn't shown, but it is reasonable to work on the assumption that it's a cheap 220V transformer. No risk when 380 V peak (<< 5000) which is well within safety margins. (Joan)
  8. Keep RPi in noninflammable enclosure to prevent burns. TODO I sent email about Raspberry Pi Case to the producers. (Joan)
  9. [conjuncture for Double verification of the isolation]. Viscoelastic material characteristic can be used to estimate continuously without altering the system in the runtime if the system's resistance changes by some FFT of the system. If zero resistance, the schema probably changes from the Kelvin-Voigt model to the Maxwell model. (here) This mechanism can be connected to the system as a negative feedback such that it automatically switches off the power if the event occurs. I think the first level mechanism of ADAS1000 can be problematic with high powers because the components can break in such cases.

Schema in the publication

enter image description here

where the change is an addition of a GUI for real-time visualization by FFT and multi-threading. This addition ensures the isolation of the ECG front-end from the RP because of ADums in the Biosignal Pi design.

Raspberry Pi models' power usage in a day

The thread How much energy does the raspberry pi consume in a day? is about the power usage in a day

B  with keyboard                              = 1.89 W -> daily 45   Wh
B+ with keyboard                              = 1.21 W -> daily 29   Wh  
B+ with LAN/USB chip off (no i/o except GPIO) = 0.76 W -> daily 18.2 Wh  
B+ shut down                                  = 0.26 W -> daily 6.2  Wh  
A  idle                                       = 0.7  W -> daily 17   Wh 
A+ idle                                       = 0.52 W -> daily 12.5 Wh 
Pi2 B at idle                                 = 1.15 W -> daily 28   Wh
Pi Zero at idle                               = 0.51 W -> daily 12.2 Wh

where A+, B+ and Zero offer much benefits in the power circuitry. The values are all about 10% greater than in the post Power consumption. Recall B+ is the chosen device in the application, but the publication is older than Pi 2 B. I asked already the author of the publication how he would improve the settings of the electronics if Pi 2 B in use.

The publication is based on Pi B+. The thread How Much Less Power does the Raspberry Pi B+ use than the old model B? is about

[T]he new Raspberry Pi B+ uses 1.21 Watts with just a keyboard dongle vs 1.89 Watts for the old model B. [I]t’s 36% less power usage. This great if you’re running on batteries, or have a barely adequate solar panel.

The table has similar experimental results. More about the stability of the power management is wanted.

Power consumption in a day in all models under GPU load by acc. FFT

The GPU-power usage is different between B+ and other models because of the different power circuitry (here). The accelerated FFT puts the chip under heavy load so the behavior is dependent on the power circuitry.

Selection = Raspberry Pi 2 B + SnickerDoodle + piSmasher SBC

The power circuit of RPi 2 B is not too different from RP 1 B+. Still, both are not medical devices so the ECG front-end has to be isolated from the RP (etc power and SPI commiserations) which is done by ADums in Biosignal Pi design. (Farhad)

I profiled my system and noticed that I need FPGA much in my prototyping phase and many GPIOs. I started to support the SnickerDoodle project here and piSmasher RBC such that I can integrate the existing RB blueprint into the SnickerDoodle. The SnickerDoodle is just going to work as a computational device, supporting RP2B, completely isolated from the ECG front-end. I will let you know when I understand the limitations of the project better after getting the chips for the development.

How are the Raspberry models different in the GPU computation in the power usage?

Improve this question
7
  • 1
    Good question BTW, well formed
    – webo80
    Commented Jan 11, 2016 at 14:54
  • 5
    "Power stability" is primarily the concern of your power supply, not the power consumer! That said, you are absolutely right to worry about strict isolation policies. The first publication you quote quite sensibly talks about the necessary precautions between patient and the RPi. Do not assume the Pi is safe for medical use. You will be liable!
    – MSalters
    Commented Jan 11, 2016 at 14:58
  • 5
    @Masi: I'm somewhat worried by that statement. I get the feeling that you don't understand the meaning of "safety", while you are working in a medical environment where mistakes can be lethal. I was very serious when I said you will be liable, and that includes liable for wrongful death. Your choice for RPi model should not matter to your safety analysis, because you must prove your contraption is safe regardless of any failure mode of the RPi.
    – MSalters
    Commented Jan 11, 2016 at 15:13
  • 4
    Let me give one concrete example: you must prove that your patient is safe, even if the RPi suddenly decides to act as a 0 Ohm resistor between the power supply and the patient, or if it catches fire. If you can give that level of assurance, it's pretty clear that 0.5W extra power draw is also safe.
    – MSalters
    Commented Jan 11, 2016 at 15:18
  • 3
    @Masi: I am working for a company that makes real medical devices, although not directly (electrically) connected to a patient. Even so, we produce the HW under far stricter QA procedures than the RPi folks. We have to prove the patient cannot be harmed, even if every system outside our QA control fails. SInce the RPi production is outside your QA control, you must assume it is totally unsafe and will fail in the most lethal way possible. IOW, all safety must be provided by the interface between patient and RPi. Yet you still talk about "Rpi must be stable". You may not assume that
    – MSalters
    Commented Jan 11, 2016 at 15:32

1 Answer 1

Reset to default
9

I believe the GPU is identical in all Pis and makes up 95% of the silicon. The remaining 5% is used by the relatively puny ARM core(s).

See Link

Improve this answer
8
  • 1
    Probably the Pi Zero. Google for Raspberry Pi power usage.
    – joan
    Commented Jan 11, 2016 at 10:31
  • 3
    @Masi: Assuming the original design was safe despite using a Pi B+, we can state with certainty that it still is safe. Remember, when proving the Pi was sufficiently isolated by the ADOS1000, we had to assume that the Pi B+ could behave like any other component. That happens to include the Pi B+ acting as a broken Pi 2 B.
    – MSalters
    Commented Jan 11, 2016 at 16:31
  • 1
    @Masi: In case you hadn't realized it, the "0 Ohm resistor" refers to an electrical short where the failing RPi suddenly draws a lot more power, becomes very hot, and may catch fire. If you think there's such a thing as the "maximum circuit power where the ADOS1000 is not sufficient", you lack essential electrical engineering skills to design medical devices.. Please, for the safety of the patients, pick another project to learn electrical engineering.
    – MSalters
    Commented Jan 11, 2016 at 17:14
  • 2
    @Masi: Of course it can change. In fact, it changes all the time as the CPU goes from 0% busy to 100% busy.
    – MSalters
    Commented Jan 11, 2016 at 20:20
  • 2
    @Masi: Ehm, normal distributions are associated with probabilities. Power usage as function of CPU usage is NOT a probability, and treating it as such is nonsensical. Power measurement error is almost certainly normally distributed, though, and how big that error is depends on your measurement setup.
    – MSalters
    Commented Jan 11, 2016 at 20:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.