eploring role of information and communication technologies in community radio stations in india
TRANSCRIPT
1
Exploring the Role of Information and Communication Technologies in Community Radio Stations in IndiaZahir Koradia
Supervisors
Bhaskaran Raman Aaditeshwar SethApril 2015
PhD Defense
2
Motivation: Role of Media in a Democracy
Checks and balances against those in powerInformation and knowledge sourceSocial Media– Can be producers of information– Can express opinion about relevant issues
3
Motivation: Media in rural India
Literacy– 26% illiterate
Connectivity– 80% disconnected
Devices
4
Motivation: Community Radio a Viable Alternative
Short range (10Km - 15Km) FM stations– Literacy not required– Connectivity available– Devices affordable
About 200 CR stations operational in India– Predominantly rural locations– Run by non-profit organizations– Local information, Community participation– Improve local governance– Promote local culture
Thesis theme: How can Information and Communication Technologies (ICTs) help community radio stations?
Mix of Ad-hoc Technologies for Daily
Tasks
Little usage of High Mobile Penetration
No Content Sharing
Challenges
● GRINS design• GRINS at 40 CR stations across 4 countrins in India and Africa • Usage and impact analysis of GRINS
• PhonePeti answering machine to increase participation
• Gurgaon Idol singing competition for IVR usability analysis
Cellular data measurements to inform design of a content sharing platform for CR
stations
Contributions
Telephonic interviews with 10 stationsPhysical visit at one station in Feb 2009
Thesis theme: How can Information and Communication Technologies (ICTs) help community radio stations?
Mix of Ad-hoc Technologies for Daily
Tasks
Little usage of High Mobile Penetration
No Content Sharing
Challenges
● GRINS design• GRINS at 40 CR stations across 4 countrins in India and Africa • Usage and impact analysis of GRINS
• PhonePeti answering machine to increase participation
• Gurgaon Idol singing competition for IVR usability analysis
Cellular data measurements to inform design of a content sharing platform for CR
stations
Contributions
Telephonic interviews with 10 stationsPhysical visit at one station
Summary Detail
Summary: Mix of Ad-hoc Technologies for Daily Tasks
Schematic of how GRINS fits into a CR station
• GRINS plays pre-recorded audio, records live speech, puts calls live on air and records them, manages contacts and content.
GRINSPC
To Transmitter
Pre-recorded audio for broadcast
Microphone Input
GSM/PSTN to SIP
gateway
Broadcast feed for recording
and streaming
To streaming server
Key Challenges
GRINS DesignGRINS Training and
Support
Challenges
• How to ensure audio quality in low end commodity hardware?
• What software design strategies are relevant for products built for developing regions? (Medhi '07, Parikh '03, Huenerfauth '02)
● How to provide support to remote technology users that are not technologically proficient?
● How to train users with no prior exposure to ICTs in using technology?
● How to evaluate GRINS usage? (Davies '05)
GRINS Evaluation
What software development strategies are relevant for building products for developing regions?
• Features built in the second iteration are used more by the stations.– Learning: Subjects in developing regions are often unable to conceive their
requirements. Iterative software development is necessary in such contexts.• Usage of different features of GRINS varies across stations because of difference in
ideologies and process.– Learning: Design of tools for developing regions must use larger sample of target users in
design phase, to avoid Overfitting.
11
How to provide support to remote CR stations?
• Support calls can come any time.– Standardize configuration to avoid remembering station wise
configuration
• Most faults can be identified on phone calls• Most faults can be fixed on phone or remote connection
Mode of Communication Identified Fixed
Phone Calls 41 16
Remote Desktop 0 9
Remote Desktop and Phone 3 6
Reverse SSH 1 5
Reverse SSH and Phone 0 2
Physical Visit 0 6
Summary: Little usage of High Mobile Penetration
Challenges and Related Work
IVR Learnability IVR Design
Challenges
● Focus on context of IVR usage. (Plauche '06, Sambasivan '11, Sherwani '07)
● Experiments on input modality: DTMF vs Voice (Patel '09, Plauche '06, Sherwani '09)
• Local dialect important (Lerer '10)• No learnability studies
● IVR Design for menu based systems (Agarwal '09, Grover '09, Sambasivan '11)
● IVR Design for peer-to-peer information sharing networks (Mudliar '12, Patel '10)
● No studies on IVR design for voting and recording audio
Related Work
IVR Experiments
• PhonePeti– 24x7 Ansering machine system– Over 1000 calls in 6 months– Studied self learnability with variations in prompts
• Gurgaon Idol– Singing competition over IVR and Radio– Formal usability studies with 130 subjects– ~350 calls over 2 months
IVR Learnability
● Only 40% of calls had useful content
● Across people who made at least three calls 36% could not learn to use PhonePeti even after three calls.
● Learning to use IVR is a challenge
● Repeated use does not help learn IVR use
● One-to-one training over phone call and in person handholding are significantly better training methods than pre-recorded instructions and repeated use.
Learned – 1 CallLearned – 3 CallsDid not learn
PromosMultiple IVR use
Training on phoneIn-person handholding
0
20
40
60
80
100
120
% s
ubje
cts
com
plet
ed t
ask
IVR Design
• All voting methods – Thumbs-Up-Thumbs-Down, Best-of-Two, and Best-of-Four – were easy to use– 100% task completion rate for all methods
• Interesting social preferences for the methods that deserve further study– Calling a song bad is impolite– Easier to compare two songs that rating a single song without
knowing all the songs in the playing field
Detail: No Content Sharing
18
Motivation
Content sustainability is a serious challengeVariety across stations in created content and its treatmentStations have developed various techniques to optimize staff efficiencyExisting approaches of email, content sharing websites have failed
19
Research Questions
What type of connection is available? (2G/3G)What is the throughput achieved?Are there diurnal patterns?What is the latency to different landmark nodes?What are the DNS look up times observed?What is the network architecture like?
Existence of HTTP Proxy, NAT, Firewall
20
Related Work – 1
Early GPRS, 1xRTT, WCDMA measurementsConducted in late 1990s – early 2000s
• Chakravorty '02, '03, Alcaraz '06, Inamura '03
Small initial CWND + high latency = inefficient short flowsLarge buffers impact interactive flows
Combined with lack of TCP SACK delays recovery
Most of these studies are at least a decade oldInitial CWND, Commonly used TCP MSS, receive window size have increased since thenTCP SACK is common
21
Related Work – 2
Recent 3G measurements– Mehlfuhrer '10, Tan '08, Elmokashfi '12, Jurvansuu '07
Norway: Latency influenced by configurationUS: 4-6 gateways for the whole country
Residential broadband network measurementsKreibich '10, Dischinger '07Netalyzr used to test existence of web cache and NAT.
Measurements to study Internet usage in developing regions
Du '06, Chen '14 Complimentary to our work.
22
Measurement Design Considerations
Choice of measurement typeActive vs Passive
We chose active measurements for better control
23
Measurement Design Considerations
Choice of measurement typeActive vs Passive
Choice of locations7 locations – 5 rural, 1 semi-urban, 1 urban
Choice of locations were dependent on logistics partners: Gram Vaani, PRADAN, Air Jaldi, Vikas Samvad.
24
Measurement Design Considerations
Choice of measurement typeActive vs Passive
Choice of locations7 locations – 5 rural, 1 semi-urban, 1 urban
Choice of service providers3 GSM based, 1 CDMA based
Provider G1 G2 G3 C1
# of locations 6 6 5 3
We chose ‘best’ available connections at the locations.
25
Measurement Design Considerations
Choice of measurement typeActive vs Passive
Choice of locations7 locations – 5 rural, 1 semi-urban, 1 urban
Choice of service providers3 GSM based, 1 CDMA based
Choice of measurement node device
Provider G1 G2 G3 C1
# of locations 6 6 5 3
We chose netbook with dongles for faster deployment and battery backup
Measurement Design Considerations
• Choice of tests
Test Decription
Throughput Iperf uplink and downlink for 5 minutes
Latency Ping (30 packets per test) & Traceroute to 20 landmark nodes
DNS Lookup Two consecutive DNS lookups. Use second as the lookup time.
New Connection Tests Log IP address, DNS sever, and default gateway assigned.
One Time Tests Netalyzr to note existence of NATs, HTTP proxies, and web caches
27
Measurement Suite – Flow Diagram
Control ServerMeasurementNode 1.
Req
uest
Test
s
2. T
est l
ist
28
Measurement Suite – Flow Diagram
Control ServerMeasurementNode
MeasurementServer
3. Conduct tests
1. R
eque
st Te
sts
2. Te
st lis
t
29
Measurement Suite – Flow Diagram
Control ServerMeasurementNode
MeasurementServerData Server
3. Conduct tests
4. Send Data4. Send Data
1. R
eque
st Te
sts
2. T
est l
ist
30
Measurement Suite – Flow Diagram
Control ServerMeasurementNode
MeasurementServerData Server
3. Conduct tests
4. Send Data4. Send Data
Email Daily Test Report
1. R
eque
st Te
sts
2. Te
st lis
t
31
Measurement Suite – Flow Diagram
Control ServerMeasurementNode
MeasurementServerData Server
3. Conduct tests
4. Send Data4. Send Data
Email Heartbeat Failure
Heartbeat
Email Daily Test Report
1. R
eque
st Te
sts
2. Te
st lis
t
32
Practical Challenges
Symptom Root Cause Resolution/Workaround
All 3 modems not connecting to the netbook simultanously.
Netbook USB bus cannot supply enough power to all the modems
Use USB adaptors with external power supply.
All 3 modems not connecting to the Internet
NetworkManager not designed to handle multiple modems.
Use wavdial dialer utility and write scripts to manage connections.
Disconnection in rural locations followed by repeated reconnection failures.
Suspected firmware bug.
Requires manual disconnection of the modem to cut power supply and restart the firmware. Monitoring script built.
Signal strength measurement failure.
Standard AT command not accepted by all modems
Use device specific AT command
33
Results Summary
450 hours of measurement data3 Months4 Experiment sets per day7 locations4 service providers
Results – Availability
- Availability in rural areas is at 75%, about 15% less than urban regions.- Content sharing platform needs to be disruption tolerant
35
Results – Throughput
- Diurnal patterns observed in throughputs in G1 and G2- Content sharing platform can leverage higher night throughputs
36
Results – Throughput vs Signal Strength
Little correlation between measured signal strength and throughput.
37
Results – Connection Stall
Frequency of Connection Stalls
High frequency of connection stalls across service providers and locations
39
Results – Impact of Connection Stalls
Strong correlation between connectionstalls an reduction in throughput.
40
Connection Stall Causes
Transport protocolLocationService providerAccess technologySignal strength variationsOut of order bufferingModem hardware artifacts2G/3G switching (Balachandram '14)
41
Results – Content Placement
Four landmark nodes served by Akamaidig www.timesofindia.com => a1521.g.akamai.net
Akamai uses service providers to host contentReturns IP from same service provider as requester
Service Provider Percentage of In-Network servers returned by DNS lookup
G1 2.15
G3 72.86
C1 87.75
~50% reduction in latency in G1 and G3No reduction in C1
Results – Content Placement
• Download 5MB file 25 times from one in-network server and one outside-network server– Using G1 & C1 3G: 98% and 17% increase in throughput – Using G1 and G3 EDGE: Little increase in throughput
• Very few gateways serve large areas– G1 – 5, G2 – 3, G3 – 4, C1 – 2– Content placement ‘near’ these gateways can provide
performance benefits
Future Work
Community participation continues to be the primary goal of CR stations– Explore other forms of IVR systems to enable community
participation– Explore use of social media and the Internet
Ground work for understanding networks at CR stations has been done– Build a robust and useful content sharing platform for CR
stations– Explore mechanisms for stations to learn from each other
44
Thank you …
Acknowledgements: Kapil Dadheech, Balachandran C., Rohit Jain, members of Gram Vaani team, Piyush Aggarwal, Amitsingh Chandele, Gaurav Aggarwal, Goutham Mannava, Aravindh Raman, Vinay Ribeiro, PRADAN.
Extra Slides
How to ensure audio quality on low end commodity hardware?
• GRINS can have up to 5 parallel audio streams• To ensure that audio did not suffer from audio clicks (gaps in audio due
to processing delay) or high latency on low end commodity hardware we designed novel experiment setups.
Machine 1 Machine 2GRINS
H/WLine out
Line out
Line in
Line in
A sine wave audio is sent from machine 1 to machine 2 via GRINS h/w. Fourier analysis of the received audio shows audio clicks.
Q: What strategies help provide technical support effectively in developing regions?
• Technical support was provided most often over phone calls– We standardized the cabling, IP address, usernames and passwords (in spite
of security risks) to avoid remembering station specific configuration.
• Identifying faults over phone calls was possible, but resolving them often required remote access.
Phone Calls
Remote Desktop
Remote Desktop and Phone
Reverse SSH
Reverse SSH and Phone
Physical Visit
Faults Identified 41 0 3 1 0 0
Faults Resolved 16 9 6 5 2 6
Q: What was the qualitative impact of GRINS?
• Telephony is used by stations to connect communities with authorities – traffic commissioners, candidates for local elections, mayor, etc.
• SMS is used to run competitions, conduct polls, receive feedback and song requests
• Search and contacts management has been used to better “connect with callers”
Background – Impact of Community Radio
Soni Rana
Listener, Henvalvani.
49
The problems of the community people would not have been forwarded and resolved if Henvalvani was not established. Henvalvani Community Radio station … has provided us a platform wherein we can raise our voice.
During the municipal election in 2011, Gurgaon Ki Aawaz broadcasted live for 14 hours. From voting to result declaration people called and participated … and gave updates by calling from their booths.
Soumya Jha
Station Manager
Gurgaon Ki Aawaz
CR Improves Local Governance
Background – Impact of Community Radio
Sharmila
Staff, Gurgaon Ki Aawaz
50
Once we got a call from a person on the occasion of his brother’s marriage, saying they have not booked a DJ because they want to listen to ‘Raghini’. They requested us to play Raghini on air.
I like the songs played in our local language – Gadhwali. In the way a Punjabi speaks in Punjabi, a Marathi speaks in Marathi, but a Gadhwali shied from speaking their language but with Henvalvani, people feel proud of their language and speak in Gadhwali
Jabbar Singh
Listener, Henvalvani.
CR Promotes Local Culture
Background – Impact of Community Radio
Dr Sudesh
Listener, Gurgaon Ki Aawaz
51
I used to go late to clinic, but after listening to Gurgaon Ki Aawaz programs I have started reaching clinic on time.
My family and I listen to Gali Gali Sim Sim, which has changed my children in terms of cleanliness and studies. Earlier we used to say to our kids to go to school, but now they get ready and go themselves.
Hari Kishan
Listener, Alfaz-e-Mewat
CR Leads to Behavior Change
Protocol Stack for GPRS
Background – Impact of Community Radio
Soni Rana
Listener, Henvalvani.
53
The problems of the community people would not have been forwarded and resolved if Henvalvani was not established. Henvalvani Community Radio station … has provided us a platform wherein we can raise our voice.
CR Improves Local Governance
Background – Impact of Community Radio
54
I like the songs played in our local language – Gadhwali. In the way a Punjabi speaks in Punjabi, a Marathi speaks in Marathi, but a Gadhwali shied from speaking their language but with Henvalvani, people feel proud of their language and speak in Gadhwali
Jabbar Singh
Listener, Henvalvani.CR Promotes Local Culture
Background – Impact of Community Radio
55
My family and I listen to Gali Gali Sim Sim, which has changed my children in terms of cleanliness and studies. Earlier we used to say to our kids to go to school, but now they get ready and go themselves.
Hari Kishan
Listener, Alfaz-e-Mewat
CR Leads to Behavior Change
Voting over IVRS: Thumbs-Up-Thumbs-Down vs Best-of-Two
• Thumbs-Up-Thumbs-Down (TUTD)– Listen to a song and then vote it up or down.– Quite similar to the Facebook 'like' or Google '+1'
• Best-of-Two (B2)– Listen to two songs and choose the best one– Can use existing methods to get global order from pairwise
preferences
Experiment Design
• 9 subjects used TUTD to vote on 5 songs followed by B2 to compare 5 pairs of songs
• 8 subjects used the VUIs in reverse order• Radio program provided explicit instructions on how to use the VUI
TUTD vs B2: Results
• All but one subject were able to use both VUIs. The lone exception could not use either VUI
• Several cultural factors impacted VUI preference– “It is impolite to call a song bad. It is easier to state one
song better than another”– “I can't rate a song good or bad without knowing all songs
in the field. Comparing two songs is better”– “I feel uncomfortable rating one song better than another” -
She voted all songs UP in TUTD, while B2 gave more information
B2 vs Best-of-Four VUI (B4)
• Common ranking algorithms require lots of pairwise preferences to obtain global ranking– Glicko requires at least five preferences per song
• Using Best-of-n instead of Best-of-Two– Gives n-1 preferences per vote– But usability may drop a n increases
• How usable is B4 compared to B2?• How do the rankings obtained using the two VUIs
compare?
B2 vs B4
• Usability– 10 subjects used B2 followed by B4– Another 10 subjects used the VUIs in reverse order– All voters complained about the difficulty of using B4 due to
difficulty in remembering earlier songs
• Voting reliability– 10 university students used B2 to rate 4 well known Bollywood
songs, and another 10 used B4– Glicko was used to obtain ratings for the 4 songs in both cases– Ratings were similar, rankings were same
• Impact of order of songs in B4– Yet another 10 students used B4 with song order reversed– Ratings were similar, rankings were same
61
Measurement Suite – Architecture Requirements
Allow changing tests and test parameters post deployment
Deploy new scriptsInstall new software
Monitor and notify failuresPowerUSB modem detachment from netbookInternet disconnectionsNumber of successful/failed tests
Provide recovery mechanisms