janjongboom

Sidely related: using machine learning to detect different types of booze with a gas sensor: https://twitter.com/kartben/status/1258791793073815552

FYI, this runs a realtime custom keyword spotting model in 31K of RAM and 47K of ROM (including all DSP code and Neural Network) on the updated micro:bit three times a second. We can do a lot these days on embedded devices!

Thanks! We got approached by some of the people in micro:bit to see if we could get this to work on short notice, and they sent some samples out last week!

Amazing new little device, and with it's much more powerful MCU (nRF52-series Cortex-M4 rather than the nRF51 Cortex-M0+) with 8x the RAM you can now even run small Machine Learning algorithms on it. Here it's responding to my voice: https://www.youtube.com/watch?v=fNSKWdIxh8o&feature=emb_titl...

Edge Impulse | Fullstack Web Developer | Amsterdam, San Jose, REMOTE | Full-time

Edge Impulse enables developers to create the next generation of intelligent device solutions with embedded Machine Learning. Machine Learning at the very edge will enable valuable use of the 99% of sensor data that is discarded today due to cost, bandwidth or power constraints.

We were founded a year ago, have customers, and are now looking for our 8th musketeer. You’ll make complex features like transfer learning or computer vision available to every user, and work with our infrastructure engineers to build backends that can quickly transform terabytes of customer data.

We’re a remote first company (even me and my cofounder are in two continents) and value work-life balance! More info: https://angel.co/company/edge-impulse/jobs/887141-full-stack... or reach out via jan@edgeimpulse.com !

I really like my Autonomous chair (https://www.autonomous.ai/office-chairs/ergonomic-chair) - prefer it over Aeron and decently priced at under 400$.

Edge Impulse (https://edgeimpulse.com) | Technical marketeer | REMOTE (North America or Europe)

Edge Impulse enables developers to build the next generation of intelligent devices with embedded machine learning (also known as TinyML). We believe that our platform will unlock immense value from 99% of edge devices data that is currently being discarded due to cost, bandwidth or power constraints. Our SaaS platform provides new ways for data ingestion, labeling, training, and deployment capabilities to accelerate developers' success with ML on the edge. In short, Edge Impulse helps developers make hardware magic.

As our first technical marketing hire you'll work with our customers, our developer community, and partners to define the market, build compelling stories, educate the industry and have a big impact on our roadmap. We’re remote equal, with leadership split between San Jose and Amsterdam, and welcome candidates in North America and Europe.

https://angel.co/company/edge-impulse/jobs/761284-ml-marketi...

Author here. Our main focus right now is on vibration and non-voice audio. For vibration we use spectral analysis through butterworth filters, FFT over the result, looking at the peaks and locations of the FFT result, and then at spectral power buckets. For audio we look at MFCC (sometimes paired with a bandpass filter). The source for these are in https://github.com/edgeimpulse/processing-blocks (Python) and https://github.com/edgeimpulse/inferencing-sdk-cpp (optimised C++). What’s though on our end is that we need to run the DSP in real-time on very constrained devices.

Unfortunately I have very little good resources on learning more. I’ve done lots of experimentation in Matlab, just getting raw sensor data and playing with the signal processing toolkit that they sell.

The cheapest 8-channel LoRaWAN gateways can be found for 69$ (https://www.thethingsnetwork.org/docs/gateways/thethingsindo...). Prices have come down significantly. Outdoor gateways with IP67 rating can be found for about 400$.

Alternatively two LoRa radios can just talk point-to-point between each other.

No. It’s been defunct since Firefox OS died.

We did exactly that four years ago with JanOS (http://janos.io/), it turned (Firefox OS) mobile phone motherboards into IoT devices. You could also control some GPIO, see http://blog.telenor.io/gonzo/hardware/2015/02/10/gpio.html for an example of remapping the volume buttons to control a new LED. Unfortunately we couldn't find a single OEM that wanted to sell the motherboards if they weren't going in a mobile phone, and ODM's didn't want to work with us either if we wouldn't build phones.

See also uTensor (https://github.com/uTensor/uTensor) which scales down TensorFlow-trained models to something you can run on a microcontroller. Currently working on hardware acceleration on MCUs with DSP extensions.

I think the biggest problem of IoT right now is the perception of what's possible... We've been building internet connected devices for decades, but that's still what people perceive as IoT ('we put a chip'-movement). However, I think the true opportunity for IoT lays in the combination of the cost of sensors going down, the range of sensors going up, and the advent of machine learning.

With cheaper and longer-range sensors we can create vast sets of data on a scale that we haven't seen before, and with machine learning we can extract useful information from this data. With that premise it's also not hard to see why IoT in the consumer market is not such a big success, but you see big industrial players pour billions into IoT projects [1].

There's some other concerns of course too, like the impact that all this data might have on society and that the state of IoT security reflects the state of web security around 1999. If you're interested in that, I did a talk about the subject last week at the AMSxTech conference (20 minutes) [2].

[1] http://tech.newstatesman.com/iot/general-electric-billion-do... [2] https://www.youtube.com/watch?v=XxwWmgbRxwU&t=7s

mbed applications can actually be debugged by any IDE that supports GDB, including Eclipse and Visual Studio Code. More info here [1].

[1] https://docs.mbed.com/docs/debugging-on-mbed/en/latest/

The mbed online compiler runs ARMCC - which is a commercial compiler. But mbed applications can be compiled locally using GCC, ARMCC and IAR.

The app is running Android, the device is OpenWRT as far as I understand.

Apple does something similar, uses a ARM M-series microcontroller to do sensor readings (activity detection etc.) when the device is sleeping [1].

[1] https://en.wikipedia.org/wiki/Apple_motion_coprocessors

For one, advertised components are not always in the actual phone. Two, licensing fees are probably not paid, given that they are ~7$ for a 3G handset.

Still, $4 is undoable. Cheapest Spreadtrum 2G SOC with WiFi/BT/Sim holder/camera will go over that amount.

On JanOS it's possible to repurpose some of the GPIO pins, as I did here: http://blog.telenor.io/gonzo/hardware/2015/02/10/gpio.html

That is one blatant ripoff of the Nordic semiconductor logo... http://www.nordicsemi.com/extension/nordic/design/bootnordic...

Depends on how they communicate between each other. I'm doing some stuff around LoRa [1] for long range communication, and my base station is a Rpi, the end nodes can be any type of microcontroller. E.g. [2] which comes with ARM's mbed, which even allows me to program them from the browser. For Bluetooth pick anything based on nrf58122, e.g. [3]. For WiFi pick ESP 8266 [4].

[1] http://www.link-labs.com/what-is-lora/ [2] http://www.multitech.com/brands/multiconnect-mdot [3] https://www.nordicsemi.com/eng/Products/Bluetooth-Smart-Blue... [4] http://www.espruino.com/ESP8266

Kinda unrelated to the whole Rpi discussion, but the SOC of any phone that runs Android would do. If that is Linux'y enough for you then it could be a solution. The problem is sourcing the SOCs. Spreadtrum makes some nice chips that include app processor, cellular, WiFi, Bluetooth, etc. for ~7$ but they're not for sale if you're not an ODM.

This is a great article [1] about how Chinese SOC manufacturers and Chinese IP works. I ended up just ripping out the mainboard of phones to get access to the SOC [2].

[1] http://www.bunniestudios.com/blog/?p=4297 [2] http://janos.io/

Why are we even using a Raspberry Pi for battery powered projects... 30 mA draw is so high you'll be changing batteries every 2 days, and that is without doing anything. Compare that to f.e. a normal phone chipset with deep sleep, which draws about 9 mA with a cellular connection, or .1-.2 mA in flight mode.

Or, if you just use the Pi for sensor readings, get a proper microcontroller, where current is measured in micro (or even nano-)amps...

> Also, since you already wrote the analysis code in Python, why not write the front-end in Python (rather than Node) as well?

I had the node code laying around from a conference presentation I did, so it was the quickest way to gather some data while the others were looking into the machine learning part.

Then we just made the processes talk to eachother through filesystem, as you do at a hackathon..

When you're typing and want to correct with BACKSPACE, but at the same time the timer runs out you'll be returned to previous page and can't see your score...

But this applies to software telco team, right. Other teams can use different stacks.

jeanine@telenordigital.com is the hiring manager for these listings. Questions might be more suited for her. Otherwise [username]@gmail.com

We have some remotes (me included), it depends a bit on the team, but as long as you're OK with ~20% onsite it should be discussable.

Telenor Digital - Oslo, Norway or Trondheim, Norway or Amsterdam, Netherlands - Full-time

We're the R&D Tech company within a big telco with branches in the Nordics, Eastern Europe and Asia; with ~150 million customers and counting. We're rethinking about how a telco should look like in the 21th century and roll out projects around the globe, so if you'd like to travel there's definitely an opportunity. We're looking to hire a bunch of great software engineers (C++, Java, NodeJS or frontend):

Team Software telco (https://www.linkedin.com/jobs2/view/48915513?trk=biz-overvie...) - Re-inventing the telco by building a modern communications infrastructure for voice and data communications fully at home in the Internet and modern distributed/virtualized/containerised/etc. architecture.

No specific team (https://www.linkedin.com/jobs2/view/48347266?trk=biz-overvie...) - Focused on a variety of our consumer products, which can change pretty quickly. Some focus areas that are in the pipeline: health solutions for South East Asia, advertisement supported phones for Bangladesh, consumer cloud storage (rolled out in Nordics & Malaysia), fashion, IoT, etc.

There is an apkjs IRC channel on mozilla [1] so they might cross-compile Android apps via emscripten to asm.js. Couldn't get any info from moz when I asked around in that channel.

[1] http://search.mibbit.com/networks/Mozilla/%23apkjs