1. Exploring Indoor White Spaces in Metropolises
Xuhang Ying, Jincheng Zhang, Lichao Yan
Guanglin Zhang, Minghua Chen
Ranveer Chandra

2. Skyrocketing Wireless Data Demand
Source: Cisco VNI Global Mobile Data Traffic Forecast,
Our story starts with an alarming observation that the worldwide wireless data demand grows exponentially – according to CISCO, the demand triples almost every 2 years.
CAGR: Compound Annual Growth Rate
Hong Kong 2.5G-4G data demand: 23Gbps (double every year)

3. A Vision: Improve Spectrum Utilization to Satisfy the Growing Demand
15%
Spectrum Occupancy
How can we support the increasing demand? Well, it is observed in practice that although most spectrum are licensed, they are underutilized -– in a case study in Chicago, the average utilization is less than 15%.
Thus one way to satisfy the growing demand is to improve the spectrum utilization, by allowing unlicensed users to access the license band when the band is not in use.
Most spectrum are licensed but underutilized

4. A Trend: Explore TV White Spaces
“White Spaces” are unoccupied TV channels
FCC allows unlicensed devices to operate in white spaces (2008, 2010)
TV “White Space”
dbm
Frequency
-60
-100
“White spaces”
470 MHz
800 MHz
Primarily UHF ~ MHz (channel 21-62)
A trend in DSA is to explore the unoccupied TV channels for wireless communications. These unoccupied TV channels are called white spaces.
This trend is rather promising, esp since FCC passed the historic rulings and opened the door for unlicensed devices to operate in TV white spaces. (From the award ceremony this morning, we know victor is the person behind these moves).

5. TV White Space Networking Scenario
ISM (Wi-Fi)
MHz
54-90
470
700
2400
2500
5180
5300
7000
MHz
Signal Strength
Vacant Spectrum
up to 3x of g
Signal Strength
Conceptually, using TV white spaces for communication is similar to using the wifi band for communications. But the scenario also involves new challenges to be addressed. For example, how to identify the unused TV channels? How does two devices agree on which channels to communicate?
Frequency
Frequency

6. Prior Works and Our Observation
Measurement
Identification
Medium Access
Network Design
Outdoor
Chicago [1, 2], Singapore [3], Guangzhou [4],
UK [5], Europe [6], etc.
Cabric [7], Kim [8, 9],
Murty [10], etc.
Yuan [11]
Borth [12]
Bahl [13], etc.
Murty [10],
Borth [12],
Bahl [13],
Feng [14], etc.
Indoor ?
802.11af
The trend is promising, and the scenario incurs new challenges. Of course, there are existing works and many of them are really nice. These works focus on various aspects of white space networking. …
But outdoor is only one part of the story. We believe that indoor is another important part of the story, … and the indoor white space story is widely open for exploration.
More than 70% of data demand comes from indoors[15]
Most people are indoors 80% of the time[16]

7. WISER – White-space Indoor Spectrum EnhanceR
Our Contributions
Measurement
Identification
Medium Access
Network Design
Outdoor
Chicago[1, 2], etc.
Cabric[7], Murty[10], etc.
Yuan[11],
Bahl[13], etc.
Murty[10],
Indoor
This work
802.11af
Upcoming
First large scale measurement in metropolises
50% and 70% of the TV spectrum are white spaces in outdoors and indoors
WISER design and proto-typing
Data-driven design
WISER prototype identifies 30%~50% more indoor white spaces compared with alternative approaches
In this work, we take the first step in filling in the blanks. In particular,
WISER – White-space Indoor Spectrum EnhanceR

8. How much more white spaces are indoor? What are their characteristics?

9. White Space Availability in Hong Kong
A Large-scale measurement study in Hong Kong
Outdoor white space ratio: 50%
Indoor white space ratio: 70%
Urban area : occupancy rate from 57 to 69%
Suburban area: occupancy rate from 43 to 52%
Rural area: occupancy rate below 40%
FCC: 
DTV Threshold = signal strength of -114 dBm averaged over a 6 MHz bandwidth, and
ATV Threshold = signal strength of -114 dBm averaged over a 100 kHz bandwidth.
However, we are now using:
DTV Threshold = signal strength of dBm summed over a 8 MHz bandwidth, and
ATV Threshold = signal strength of dBm summed over a 100 kHz bandwidth
Since we are using 2048 bins for 8 MHz and 25 bins for 100 kHz, if we converted summed signal strength to averaged signal strength (i.e. dividing the summed signal strength by the total number of bins), thresholds are:
DTV Threshold = signal strength of dBm averaged over a 8 MHz bandwidth, and
ATV Threshold = signal strength of dBm averaged over a 100 kHz bandwidth.
Principle TV Station
Fill-in TV Station
Measurement Location
31 measurement locations
Hardware : USRP + Antenna + Laptop

10. Indoor White Space Measurement
Experiment Scenario:
7th floor of a 10-floor office building
65 measurement locations (cover all rooms and corridors)
Measurement
Across four months
One time profiling every day
Record the signal strengths
for all channels at all locations

11. Indoor White Space Characteristics
Indoor white spaces show spatial variation – single location sensing is not enough
Indoor white spaces are long-term unstable – one time profiling is not enough

12. Indoor White Space Correlation
We will use this characteristic later to design our system.
TV signal strengths show strong correlation across channels and locations

13. How to identify the indoor white spaces?

14. Design Space and Solution Comparison
Approach
False Alarm Rate
White Space Loss Rate
Total Cost
Geo-database
Low
High
Outdoor-Sensing-Only
One-Time-Profiling-Only
Sensor-All-Over-The-Place
WISER (This work)
Design space of indoor white space identification system
We want a low cost, low FA rate, low WS Loss rate approach.
The indoor white space characteristics give us some implications.
MC: Need cosmetic improvement.
Intuition: Exploiting indoor white space correlation to save sensor cost!

15. Indoor Positioning System
WISER Architecture
Indoor Positioning System
Server
Outdoor Sensor
Function of the three modules
Profiled Location
Indoor Sensor

16. Key Challenge: Indoor Sensor Placement
Get the signal strengths
One-time spectrum profiling
Channel-Location clustering
Indoor sensor placement
Given k sensors to be placed, where are the best locations to place them?
Compute Channel- Location clusters
The number of indoor sensors is important. There is a tradeoff between the number of indoor sensors and the system performance.
The locations of indoor sensors are also very important.
3. What we answer: Given the number of sensors, how to obtain the best locations for these sensors?
The procedures are as follows:
First, OTP
Second, Channel-Location Clustering
Third, Place indoor sensors according to the clustering results
In the following, we introduce the detailed steps to do these procedures.
Place one sensor per cluster

17. Channel-Location Clustering
Simple Case:
One channel, 𝑀 locations
What we want: 𝑘 channel-location clusters
Compute the proximity matrix
Merge two “closest” clusters
Until k clusters
Proximity matrix: euclidean distance
Merge method: Ward’s minimum variance

18. Channel-Location Clustering
General Case:
𝑁 channels, 𝑀 locations
𝑘 channel clusters, 𝑘 𝑖 channel-location clusters for channel cluster 𝑖
Compute the proximity matrix
Merge two “closest” channel clusters
Channel 3,4
Channel 1,2
Proximity matrix: euclidean distance
Merge method: Ward’s minimum variance
General case:
First, we do channel clustering to get channel clusters.
Then for each channel cluster, we repeat the procedure for the simple case.
Repeat procedure for simple case

19. How well does WISER work?

20. WISER Experimentation
Implement a WISER prototype on the 7th floor of a campus building
20 indoor sensors and 1 outdoor sensor
11 experiments across 4 months
Compare WISER, Outdoor Sensing (OS-only), and One-Time-Profiling (OTP-Only)
WISER identifies 30%-50% more indoor white space as compared to baseline approaches.

21. How Many Indoor Sensors is Enough?
Balance between system performance and the total sensor cost
Why we study this problem? Because it is important to balance between WISER performance and the total sensor cost. To study the relationship between performance and cost, it is necessary to conduct this evaluation.

22. WISER – White-space Indoor Spectrum EnhanceR
Conclusions
Measurement
Identification
Medium Access
Network Design
Outdoor
Chicago[1, 2], etc.
Cabric[7], Murty[10], etc.
Yuan[11],
Bahl[13], etc.
Murty[10],
Indoor
This work
802.11af
Upcoming
First large scale measurement in metropolises
50% and 70% of the TV spectrum are white spaces in outdoors and indoors
WISER design and proto-typing
Data-driven design
WISER prototype identifies 30%~50% more indoor white spaces compared with alternative approaches
WISER – White-space Indoor Spectrum EnhanceR

23. Future Works More measurements at different buildings
Extending the single-floor design to multi-floor design
Building indoor white space network to utilize the white spaces
Extend the solution/idea to other spectrum bands
Limited understanding on how TV signal propagates into indoor environments

24. References
[1] M. McHenry et al., “Chicago Spectrum Occupancy Measurements & Analysis and A Long-term Studies Proposal”, ACM TAPAS, 2006.
[2] T. Taher et al., “Long-term Spectral Occupancy Findings in Chicago”, IEEE DySPAN, 2011.
[3] M. Islam et al., “Spectrum Survey in Singapore: Occupancy Measurements and Analyses”, IEEE CrownCom, 2008.
[4] D. Chen et al., “Mining Spectrum Usage Data: A Large-scale Spectrum Measurement Study”, ACM MobiCom, 2009.
[5] M. Nekovee et al., “Quantifying the Availability of TV White Spaces for Cognitive Radio Operation in the UK”, IEEE ICC joint workshop on cognitive wireless networks and systems, 2009.
[6] V. Jaap et al., “UHF White Space in Europe: A Quantitative Study into the Potential of the MHz band”, IEEE DySPAN, 2011.
[7] D. Cabric et al., “Experimental Study of Spectrum Sensing Based on Energy Detection and Network Cooperation”, ACM TAPAS, 2006.
[8] H. Kim et al., “Fast Discovery of Spectrum Opportunities in Cognitive Radio Networks”, IEEE DySPAN, 2008.
[9] H. Kim et al., “In-band Spectrum Sensing in Cognitive Radio Networks: Energy Detection or Feature Dection?”, ACM MobiCom, 2008.
[10] R. Murty et al., “Senseless: A Database-Driven White Space Network”, IEEE Transactions on Mobile Computing, 2012.
[11] Y. Yuan et al., “KNOWS: Kognitiv Networking Over White Spaces”, IEEE DySPAN, 2007.
[12] R. Borth et al., “Considerations for Successful Cognitive Radio Systems in US TV White Space”, IEEE DySPAN, 2008.
[13] P. Bahl et al., “White Space Networking with Wi-Fi Like Connectivity”, ACM Sigcomm, 2009.
[14] X. Feng et al., “Database-Assisted Multi-AP Network on TV White Spaces: Architecture, Spectrum Allocation and AP Discovery”, IEEE DySPAN, 2011.
[15] V. Chandrasekhar et al., “Femtocell networks: a survey”, IEEE Communications Magazine, 2008.
[16] N. Klepeis et al., “The national human activity pattern survey”, Journal of Exposure Analysis and Environmental Epidemiology, 2001.

25. Jincheng Zhang (zj012@ie.cuhk.edu.hk)
Thank you!
Jincheng Zhang