1. Microsoft Research Asia
BeepBeep:
A High Accuracy Acoustic Ranging System using COTS Mobile Devices
Chunyi Peng
Guobin(Jacky) Shen, Yongguang Zhang, Yanlin Li, Kun Tan
Microsoft Research Asia

2. Ranging, basic to localization
Many excellent systems
Cricket (Mobicom’00)
RIPS (Sensys’05)
ENSBox (Sensys’06) …
Hardware and algorithmic innovation

3. Our motivation Proximity detection between devices
Among portable devices
MobiUS (Mobisys’07)
Between portable and non-portable devices
Phone to PC, Xbox, Printers, Projectors …

4. The requirement A widely applicable solution High accuracy!
Work on COTS devices
No additional hardware (e.g., ultrasound)
Pure user space software (no change to OS/driver)
Not dependent on infrastructure
Applicable in spontaneous, ad hoc situations
Minimum set of sensors
High accuracy!

5. A matter of time measurement
Mostly, based on time-of-flight measurement
Distance = speed x time
Sound often chosen
Slower speed => less demanding on time accuracy
Still, a challenging task
1 ms error in time = 34 cm error in distance
1 cm ranging accuracy requires 30us timing accuracy
Extremely challenging on COTS/software

6. The root cause of inaccuracy – three uncertainties
Clock synchronization uncertainty
Sending uncertainty
Receiving uncertainty
software issuing command
software aware of arrival
...
t0 = wall_clock();
write(sound_dev, signal);
...
read(sound_dev, signal);
t1 = wall_clock();
unknown delays (software, system, driver, hardware, …) ?
unknown delays (hardware, interrupt, driver, scheduling, …) ?
sound leaves
speaker
sound
reaches mic
time

7. Effects of the sending and receiving uncertainties
Example measurement of the lower bound on COTS mobile devices (HP iPAQ rw6828)
Highly fluctuating, appears unpredictable
Easily adds up to 1-2 ms (=> a few feet error)
CPU idle
CPU heavily loaded

8. Our approach – BeepBeep
A simple and effective solution
Each device just needs to emit a sound signal and record them simultaneously
Only require a speaker, a mic, and some way of communicating with the other device
Achieving 1cm accuracy while satisfying all the requirements

9. Beepbeep’s basic procedure
Device A
Device B
Device A emits a beep while both recording
Device B emits another beep while both continue recording
Both devices detect TOA of the two beeps and obtain respective ETOAs
Exchange ETOAs and calculate the distance
A’s recording
B’s recording
ETOAA
ETOAB
DAB=|ETOAA-ETOAB|/2

10. Mathematical derivation
1st Beep
tB1
tA0
tA1
dB,A-dA,A = c·(tB1-tA1)
ETOAA
dA,B-dB,B = c·(tA3-tB3)
ETOAB
tA3
2nd Beep
tB3
tB2
dB,A+dA,B= c·[(tA3-tA1)-(tB3-tB1)] +dA,A+dB,B
= c·(ETOAA-ETOAB)+dA,A +dB,B

11. Key techniques, effects and rationale (I)
Self-recording
Record signals from both the other party and itself
Establish the starting reference point of the whole ranging process
Duplex audio channel
Two-way sensing
Avoid clock synchronization uncertainty
To capture the ending reference point of the whole ranging process
not attempt to capture any system time info

12. Key techniques, effects and rationale (II)
Sample counting
Avoid referring to system clocks for timing info
Dedicated A/D converter, w/ fixed sampling rate
Achievable precision is determined by the sampling frequency: 0.8cm at 44.1kHz sampling rate
Putting together:
Bypass all the three uncertainties by making time measurement irrelevant to system clocks

13. Engineering Challenges (I) Signal Design
Good signal design helps detection
Easily detectable in digital recording
Robust against ambient noise
Robust against acoustic distortion
Low-fidelity speaker & mic in COTS mobile device
Within hardware capability
Most COTS devices have limited voice frequency range
Our empirical design (“chirp” sound)
50ms long, shifting frequency from 2 to 6 kHz

14. Engineering Challenges (II) Signal Detection Algorithm Design
Efficient and fast signal detection algorithm
Quickly locate possible signal regions
Robust against low SNR
Utilize noise floor to boost SNR
Combat multipath effect
Multipath: big issue indoor environment
We derived special algorithm to detect first “sharp peak” signal correlation

15. Engineering Challenges (III)
Protocol design
Coordinate two (or more) devices in entire ranging process
Minimize ranging time duration
Device form factor
Speaker/mic’s placement affect ranging results
Vary from one device model to another
Need calibration to adjust ranging calculation

16. System Implementation
Platform: Windows Mobile 5.0
Sound: “wave” API
Communication: “WinSock” (WiFi or Bluetooth)
Software architecture
User-mode dynamic linkable library
As service for other applications
Test devices
Dopod838
HP iPAQ 6828

17. Evaluation Case-A: Indoor, quiet Case-B: Indoor, noisy
Case-C: Outdoor, car park entrance
Case-D: Outdoor, subway station
50 runs each setting
Expr
Setting
Operation
Range
Conf.
Level
AA(|Err|) cm
MA(|Err|)
A(Std)
M(Std)
Case-A
4.0m
94%
0.9
1.4
1.2
1.9
Case-B
1.1
1.7
1.0
1.3
Case-C
12m
98%
2.7
3.8
2.1
Case-D
10m
92%
2.2
1.6

18. Summary
Identified three uncertainties and mitigated them with three key technologies
Two-way sensing
Self-recording
Sample counting
BeepBeep provides a simple ranging solution
Achieves 1cm accuracy
On very basic hardware set
Purely in software (user-space)

19. Thanks! Demo session. Welcome to try BeepBeep!
Software downloads:

20. Backup

21. Lower bound of sending and receiving uncertainties