Commit 257dd0e9 authored by Archan MISRA's avatar Archan MISRA

added explanatory text for user expts in experiment.tex

modified acmart.cls to refer to Sensys
parent 508e38af
......@@ -1061,8 +1061,7 @@ Computing Machinery]
\ifx\acmConference@shortname\@empty
\gdef\acmConference@shortname{#2}%
\fi}
\acmConference[Conference'17]{ACM Conference}{July 2017}{Washington,
DC, USA}
\acmConference[Sensys'18]{ACM Sensys}{November 2018}{Shenzen, China}
\def\acmBooktitle#1{\gdef\@acmBooktitle{#1}}
\acmBooktitle{Proceedings of \acmConference@name
\ifx\acmConference@name\acmConference@shortname\else
......
\section{Experimental Results}
\section{Constrained User Studies}
\label{sec:experiment}
In this section, we present the results of our user-study driven
evaluation.
We now present results on the evaluation of our wearable prototype, used in constrained user studies performed in an office environment. The goals of these studies are to: (1) validate that the AoA-based directional strategy is capable of tracking normal user movements in an office, over different observation periods, and (2) understand the operation of the system when multiple (specifically 2) users are present.
\subsection{Experiment Setup}
\begin{figure}
\centering
\begin{tabular}[c]{cc}
\begin{subfigure}[b]{.2\textwidth}
\centering
\includegraphics[scale=0.38]{setup.jpg}
\label{fig:exprsetup}
\end{subfigure}&
\begin{subfigure}[b]{.2\textwidth}
\centering
\includegraphics[scale=0.17]{setupideal.jpg}
\label{fig:exprsetupideal}
\end{subfigure}
\end{tabular}
\caption{a. The setup in reality. b. The ideal experiment setup.}
\label{fig:setup}
\includegraphics[height=1.9in,scale=0.38]{setup.jpg}
\label{fig:exprsetup}
\caption{Experimental setup for office room user study.}
\end{figure}
We conducted all our experiments in a meeting room setup to mimic a typical
office working environment. Figure~\ref{fig:setup} shows the setup. We
implement an AP with 8-antenna array to transfer energy to our custom
wearable device using beamformed WiFi band signal.
office working environment. Figure~\ref{fig:exprsetup} shows the setup--it it fairly similar to the WARP system setup in Figure~\ref{fig:antennarray}, except that the room also contains a table where one or more users can perform their usual desk-based office chores, while wearing the \name wearable device. Because users can move their arms in many different ways, the harvested power fluctuates as well (unlike the case of the static clients evaluated in the previous section).
Unless otherwise stated, experiments are performed using an 8-antenna array on the WiFi AP. In these studies, the WiFi AP performs AoA estimation and adjustment of the beam orientation whenever it receives `ping' packets from one or more wearable devices (i.e., whenever the wearable device undergoes ``significant movement"). Note that each antenna can transmit at a maximum power of 20 dBm (achieved when the antenna gain=63); to minimize interference, we limit the antenna gain to 35 (about half of the maximum power). Accordingly, the overall radiated power from the WARP-based AP is no more than approx. 400-450mW, which is well below the EIRP limit.
%\raj{need more details here. what was being powered wirelessly?
We setup our experiment in a meeting room to simulate a office environment.
The antenna sets are placed at the 2 corners of the table (Figure
\ref{fig:exprsetup}). Ideally, the antenna array should be placed as shown
in Figure \ref{fig:exprsetupideal}, so that the AoA at the receiver and
transmitter are the same. However, even we use widely separated channels,
the interference is too strong, and the receiver cannot receive packets. In
the setup shown in Figure \ref{fig:exprsetup}, there is still interference
and it increases the packet drop rate, so we have to transmit 10 ping
packets once triggered instead of only 1 packet. After estimation of the
AoA of ping packets, we then use triangulation to estimate the direction to
the transmitter antennas and then change the beam direction accordingly.
The distance between transmitter and receiver is 1m, the table width is
XXXm, and length is XXXm. Because the performance of AoA estimation
algorithm at extreme angle (0\degree or 180\degree) is poor, the receiver is
tilted about 35\degree to increase the overlapped coverage between
transmitter and receiver.
Because of interference between transmitter and receiver, we set the
transmission gain of 35 instead of a maximum of 63. Given that at maximum
power (gain of 63), each antenna can transmit 20dBm theoretically.
When we measure harvested power at fixed positions on the table, the energy
is much more than the power consumption of the wearable, so we want to test
evaluate how the wearing of the device affect the harvested energy.
\subsection{Office Working Environment}
In this experiment, we want to capture the amount of harvested energy when the user sit and work at the table. In this experiment only, the data packet transmission is disable to isolate the effect of RF transmission because it is currently the most energy consuming module. Two user join this experiment, each user sit at the table for 1 hour.
% We setup our experiment in a meeting room to simulate a office environment.
% The antenna sets are placed at the 2 corners of the table (Figure
% \ref{fig:exprsetup}). Ideally, the antenna array should be placed as shown
% in Figure \ref{fig:exprsetupideal}, so that the AoA at the receiver and
% transmitter are the same. However, even we use widely separated channels,
% the interference is too strong, and the receiver cannot receive packets. In
% the setup shown in Figure \ref{fig:exprsetup}, there is still interference
% and it increases the packet drop rate, so we have to transmit 10 ping
% packets once triggered instead of only 1 packet.
% After estimation of the
% AoA of ping packets, we then use triangulation to estimate the direction to
% the transmitter antennas and then change the beam direction accordingly.
% The distance between transmitter and receiver is 1m, the table width is
% XXXm, and length is XXXm. Because the performance of AoA estimation
% algorithm at extreme angle (0\degree or 180\degree) is poor, the receiver is
% tilted about 35\degree to increase the overlapped coverage between
% transmitter and receiver.
% Because of interference between transmitter and receiver, we set the
% transmission gain of 35 instead of a maximum of 63. Given that at maximum
% power (gain of 63), each antenna can transmit 20dBm theoretically.
% When we measure harvested power at fixed positions on the table, the energy
% is much more than the power consumption of the wearable, so we want to test
% evaluate how the wearing of the device affect the harvested energy.
\subsection{Office--Short Term}
In outr first study experiment, we investigate the amount of harvested power that is likely to be realized under normal working conditions, when a user works at his or her office desk. To perform this study, two users each sat at a work desk (located within the office room) at different times, performing their usual desktop-based tasks for an hour. Each user was free to get up and move around the room, but did not leave the office room. To isolate the harvesting behavior and compare it with the static case, we disable the RF frontend of the wearable--i.e., each wearable actively collects the accelerometer data, but does not perform a wireless transfer to the backend.
\begin{figure}
\centering
\includegraphics[scale=0.25]{energy1hour.pdf}
\caption{Residual energy after 1 hour at an office desk.}
\label{fig:energy1hour}
\end{figure}
\subsection{Average Working Day}
In this experiment, the user work at the table for longer time, so it includes periods when the user go out for a while, for example, go to the restroom. The figure shows the power it consumes from the super capacitor. \textcolor{blue}{This shows power consumption, not the residual because both columns are negative values} \textcolor{blue}{}
\begin{figure}
Figure~\ref{fig:energy1hour} plots the average differential power (measured over the whole 1 hour) for the two users, and compares to a baseline where the wearable was left stationary. We see that one of the users has a surplus differential power, whereas the other user has a slight deficit ($\sim -10\mu$W). These differences are due to both the differing harvesting efficiency of the two receivers, as well as potential errors in the AoA estimation (which would lead to inaccuracies in the directed RF beams). Note, however, that in all these cases, the sensor+ microprocessor sub-systems are always on--the motion harvester is used only to trigger the `ping' packets (i.e., the AoA recalibration), but not the sensing pipeline. Accordingly, these results represent a \emph{pessimistic} case, where the sensor is always on. Broadly, we see that the overall differential power surplus is significantly reduced (only around 30$\mu$W for user 1, compared to over 300$\mu$W for static clients (Figure~\ref{fig:residual1}), due to variations in the harvesting gain cause by both (i) random movements of the user's hand, and (ii) AoA estimation errors. However, the result does suggest that a WiFi harvesting-based wearable is likely to be usable in office environments.
\subsection{Office--Longer Term}
This next study is identical to the previous one, except that it involves only 1 user and is conducted over a longer observation period (4 hours), This larger observation period helps captures the user's natural movement dynamics over effectively one half of a typical working day: in this study, the user occasionally left the office room (e.g., to visit the restroom). For these study, we enabled both the continuous sensing and the periodic (once every 1.5 minutes) data transfer components. Accordingly, this study is meant to monitor the worst-case energy drain; in practice, a user will exhibit \emph{significant motion} only intermittently, and the sensing and data transfer overheads will thus be dramatically lower.
Figure~\ref{energy4hour} plots the differential average power of the user, wearing the energy harvesting wearable. FOr a baseline comparison, we utilize a setting where the WiFi RF transmission are turned off--i.e., all energy harvesting is disabled. The baseline thus indicates the total average power drain on the wearable, in the absence of any energy harvesting. The figure shows that, in the absence of any harvesting, the wearable device drains approx. 65$\mu$W; in comparison, the power drain on the supercapacitor with harvesting enabled is only 20$\mu$W. The average power drain from the sensing+ data transfer components is XXX$\mu$W \am{Vu:pls. replace XXX by the average power drain that you had calculated--I believe it was around 120 $\mu$W??}. Accordingly, as the use of motion triggering on these components will dramatically cut down this power drain, the overall harvesting power will become positive, allowing the wearable to operate continually.
\begin{figure}[!htb]
\centering
\includegraphics[scale=0.25]{energy4hour_neg.pdf}
\caption{Consumed energy after 4 hour at an office desk.}
\label{fig:energy4hour}
\end{figure}
\begin{figure}
\centering
\includegraphics[scale=0.25]{accelerometer.pdf}
\caption{Accelerometer data recorded by our wearable device.}
\label{fig:accel4hour}
\end{figure}
\subsection{Overnight Charging}
\begin{figure}
\centering
\includegraphics[scale=0.5]{placeholder.pdf}
\caption{Charging overnight. Use it the entire daytime.}
\label{fig:overnight}
\end{figure}
\subsection{Multi-user Scenarios}
\begin{figure}
\centering
\includegraphics[scale=0.25]{2user_neg.pdf}
\caption{Beam adaptation for multi-user scenarios with separate beams.}
\label{fig:2userseparated}
\end{figure}
% AM: next figure doesn't say anything useful--so it is commented
% \begin{figure}
% \centering
% \includegraphics[scale=0.25]{accelerometer.pdf}
% \caption{Accelerometer data recorded by our wearable device.}
% \label{fig:accel4hour}
% \end{figure}
\begin{figure}
\centering
\includegraphics[scale=0.25]{2user_time.pdf}
\caption{Beam adaptation for multi-user scenarios with time multiplexing.}
\label{fig:2usertime}
\end{figure}
% AM: Overnight charging experiment could not be conducted--hence, commented for now.
% \subsection{Overnight Charging}
% \begin{figure}
% \centering
% \includegraphics[scale=0.5]{placeholder.pdf}
% \caption{Charging overnight. Use it the entire daytime.}
% \label{fig:overnight}
% \end{figure}
\subsection{Office: Multi-user}
We finally experimented with the case where two users occupied the office room concurrently. The two users performed their task under two different AP operatinal modes: (a) the time-multiplexed mode where the entire 8-antenna beam was directed at each wearable in round robin fashion, and (b) the concurrent mode, where each user was continuously targetd by a 4-antenna beam. For each of these two modes of operation, the two users were collocated for a total duration of two hours (i.e., the overall study duration was 4 hours), with 1 hour of being in close proximity (working side by side, with a separation of XXX m), followed by 1 hour where they worked farther apart (separated by a distance of XXX m). \am{Vu:pls. replace the two XXX in the previous line}.
Figures~\ref{fig:2usertime} and~\ref{fig:2userseparated} plot the case for the time-multiplexed and concurrent mode of AP operation, respectively. We see that, as expected, the differential power is net negative: this is expected, as the wearable has its sensing and data transfer components enabled continuously, without any motion-based triggering. However, the differential power deficit is only around 40 $\mu$W (for either user) in the concurrent mode, whereas one of the users experiences a higher deficit (close to 85$\mu$W) when working further away from the other user, in the multiplexed mode. These findings corroborate our earlier observation (in Section~\ref{sec:multiuser}) that the mutliplexed mode is preferable only when the wearables are closer to each other.
\begin{figure*}[!tbh]
\centering
\begin{minipage}{.48\textwidth}
\centering
\includegraphics[height=1.6in, width=3in]{2user_time.pdf}
\caption{Power drain (2 users) in Multiplexed mode.}
\label{fig:2usertime}
\end{minipage}%
\begin{minipage}{.48\textwidth}
\centering
\includegraphics[height=1.6in, width=3in]{2user_neg.pdf}
\caption{Power drain (2 users)in Concurrent mode}
\label{fig:2userseparated}
\end{minipage}
\end{figure*}
% \begin{figure}
% \centering
% \includegraphics[scale=0.25]{2user_neg.pdf}
% \caption{Beam adaptation for multi-user scenarios with separate beams.}
% \label{fig:2userseparated}
% \end{figure}
% \begin{figure}
% \centering
% \includegraphics[scale=0.25]{2user_time.pdf}
% \caption{Beam adaptation for multi-user scenarios with time multiplexing.}
% \label{fig:2usertime}
% \end{figure}
Overall, the experimental studies suggest that WiFi-based energy harvesting is likely to prove sufficient for wearable-based inertial sensing monitoring, at least in office environments, provided (i) the wearable device activates its sensing and data transfer pipeline only \emph{on-demand}, when a significant motion event is detected, and (ii) the WiFi AP changes mode dynamically, based on the relative distance/separation between the wearable devices. Of course, larger-scale deployments, involving dozens of office workers observed over several days, are needed to establish the real-world operating characteristics more precisely. This will, however, require campus-wide deployment of our `power packet' transmitting APs.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment