Sensys'12 (day2)

Here are my much anticipated (at least by Ted :-) notes from the second day of Sensys.

Kyun queue: a sensor network system to monitor road traffic queues

I missed this talk, but this sounds like an interesting paper. The idea seems to be to deploy a transmitter and a receiver node on the opposite sides of a road. Depending on the density and the speed of the cars passing between the transmitter and the receiver, the received radio signal has different signatures. The work studies these signals and figures out the road conditions based on these.

Low cost crowdcounting using audio tones

Crowd counting can be useful for event planning, learning popularity of aisles, etc. This work assumes that everyone in the crowd has a smartphone and this counting app running on their phone. (This is the biggest limitation of the work.) This work uses the audio tones (again ultrasound tones) emitted by the phones to count the people. The app employs the smartphone mic and speaker, no additional hardware is needed. The system uses an initiation tone to start. The phone that hears this tone picks a frequency using uniform or better geometric choosing method and forwards the tone. The frequencies are picked in the ultrasound region (that same 4khz band alluded to in yesterday's work.) Using uniform choosing method is not prone to collisions, the counts can be off due to the losses in collisions. The geometric choosing method employs geometric distributions and hence is more prone to collisions. (My work on thresholding counting is very applicable for this problem: Singlehop Collaborative Feedback Primitives for Threshold Querying in Wireless Sensor Networks.) They did evaluations with 25+ phones, but in pretty controlled environments (phones at hand). The phones don't work well at pockets or behind clothing. The audio tone method is said to be low-power, but it is actually similar cost to the wifi. The authors didn't use wifi because its range can be too long for the region to be counted. They said that, if ZigBee were available on the phones, they would use it.

AudioDAQ: turning the mobile phone's ubiquitious headset port into a universal data acquisition interface

How refreshing! The presenter and a lead author of this work is an undergraduate student. This work aims to build a platform (hardware and software components) to collect analog sensor (any sensor) data through the audio port of any phone. But the problem is there is almost no standard I/O connector, no standard programming interface, no third party app support on the smartphones. HiJack did a great job on the iPhone using the audio headset port (the Energy-efficient GPS and the MusicalHeart works at this Sensys used the HiJack approach). But HiJack is iPhone specific. This work extends to other phones using available power from "mic bias voltage" at the audio headset port. Powered this way, the analog data/voltage from the sensor attached to the audio port are encoded as the amplitude of a square wave. This work uses the phone's builtin voice recording app (which is already optimized to be low-energy and use hardware compression) to capture and store the attached sensor data and the phone's ability to send the voice recording via email/mms for exporting the captured data. The software to process the signal and interpret it is at the cloud. The platform is evalued with an EKG application. Now you can use your smartphone as a holter device! The authors estimate that the hardware board for this platform would cost $5 in mass production.

Prius: generic hybrid trace compression for WSNs

The goal of this work is to compress traces efficiently. Traces are important for debugging WSN deployments, but trace sizes can become large, so they need to be compressed otherwise whether you write to flash (most often the case) or exfiltrate to the basestation you are wasting a lot of precious energy. Traditional compression techniques (prediction-based, dictionary-based grammar-based) are not applicable for WSNs because they require more memory and computation than is available in WSNs. The insights behind Prius work are that traces are highly repetitive, traces evolve slightly, and program memory is not as scarce as data/var memory. The Prius approach is to mine trace patterns offline (static-analysis of the program code?) and to store them in program memory. In the offline learning phase, Prius creates a compression dictionary, this dictionary is then processed by data structure generator to obtain an efficient data structure data structure to be used in the headerfile of the deployed program.

DoubledipL leveraging thermoelectric harvesting for low power monitoring of sporadic water use

This work does waterflow monitoring. The motes they built include acceleration and temperature sensors and use wireless communication. The motes are harvesting their energy on the pipes using the thermal gradients of the hot/cold water in the pipe and the environment temperature.

Business meeting

There was a short business meeting in the afternoon. We learned that next year Sensys will be in Rome, Italy at the Sapienza University, Nov 11-15. The rest of the business meeting was spent for some soul-searching about Sensys. Is the excitement for Sensys diminishing as maybe indicated by the Sensys submission and attendance numbers are going down. With the smartphone going big, and also getting a lot of focus at Sensys, how can Sensys distinguish itself from Ubicomp or Ubisys? Andrew Campbell (Darthmouth) said he is nervous and somewhat pessimistic about the future of Sensys. He called on to the PhD students, Assistant professors, to get up and be more engaged/central in Sensys community. Also several people voiced interest for having a workshop on big data / machine learning as part of Sensys.

Demo session

A lot of the papers in the conference got demoed in the demo and poster session. Probably there were about 10 demos, and 20 more posters. The most interesting demo must have been the proximity social interaction monitoring sensors demo from University of Utah. They had brought in 300 motes. Their large programming board, can program upto 100 motes at the same time. They produced their own hardware, and said each mote cost about $30. They distributed upto 100 motes for monitoring, about half of the people in attendance got a mote from them. They also had stationary motes under each poster table. Although the reception area was quite large, is L shaped, and a lot of humans were in the way (all very detrimental to wireless communication), the stationary motes talk to singlehop to their basestation connected to the laptop. The mobile nodes had much smaller communication radius, they captured signals they heard from nearby motes, and broadcasted these so that stationary motes can pick these up and forward them to the basestation.

The muscle fatigue monitoring demo was also very interesting. Yes, this was the guy with the yoga pants. I got to take a closer look at the Cleon mote from the best paper (energy efficient GPS). I tried the MusicalHeart demo. I was told that music would not be able to slow my heart rate at high activity level (jogging) as I suspected. I was told that I had to set my tempo to the music, and then my heart rate will slow. So the music only becomes the cue to speed up or slow down my activity. Sort of a bummer.

General comments

You could understand if a talk/paper is well received depending on the number of people asking questions after the talk. The talks that receive no questions mean something was not quite right. (On related news, if I do not receive any questions after my presentations, I will probably have a breakdown and start crying.)

The tea (especially the vanilla orchard variety) was very good and there was a good supply of tea. I observed that more people were drinking tea relatively to previous years. I like this trend as a tea person.

The lunch arrangement was the weak point in the otherwise great organization. Lunch is picked up, and since there wasn't a lunch ballroom, we had to eat lunch in the conference room... facing the screen. This made for a lonely and awkward lunch. Having a lunch room which forces people to sit together helps create chance acquaintances, which adds a lot to the conference experience. I still find it hard to meet new people at the conferences. I think it would be really nice if conferences organized some nerdy ice-breaking events. The demo and poster session in Sensys helped a lot for this. Conferences should try more playful things for ice-breaking. Maybe, instead of writing conference name on nametags (duh, everybody in attendance knows conference name), write a twitter bio-like self-describing line about the person, or what that person wants to talk about: "ask me about crowdsourcing coffee shop line waiting times". Another idea could be to get people collaborating on an acitivity, game, or challenge. For example get people in attendance collaboratively write/edit/produce a dozen different conference summaries! Editors and webservices already exists for collaborative editing, and maybe the conference should randomly or purposefully cluster people into these editing groups.

Hmm, maybe I should start a conference critiquing business. Just pay my registration and travel, and I am set :-) I will write conference critics and reviews for your conference.

Finally, this is my motivation level after this conference.


Unknown said…
I still think the current standard for conference ambience is best illustrated by this video.

Popular posts from this blog

Foundational distributed systems papers

Your attitude determines your success

Progress beats perfect

Cores that don't count

Silent data corruptions at scale

Learning about distributed systems: where to start?

Read papers, Not too much, Mostly foundational ones

Sundial: Fault-tolerant Clock Synchronization for Datacenters


Using Lightweight Formal Methods to Validate a Key-Value Storage Node in Amazon S3 (SOSP21)