120 likes | 275 Views
VoIP/PSTN Call Separator. By Arpan Ghosh and Anand Joshi. VoIP Spam : The Problem. VoIP is geographically agnostic No concept of ‘long distance’ PSTN Receiver is ‘dumb’ Cannot make a Spam filter Easier to automate Higher volume of Spam generation per device Cheaper.
E N D
VoIP/PSTN Call Separator By Arpan Ghosh and Anand Joshi
VoIP Spam : The Problem • VoIP is geographically agnostic • No concept of ‘long distance’ • PSTN Receiver is ‘dumb’ • Cannot make a Spam filter • Easier to automate • Higher volume of Spam generation per device • Cheaper
How VoIP to PSTN Works? IP Network VoIP Client Speex VoIP -> PSTN Gateway Speex -> G.711 PSTN phone G.711 PSTN backbone network
The Challenge • Calls at receiver end always have PSTN characteristics. • Most properties of original codecs lost in last-hop encoding • Only have received call to work with • No way to perform a comparative analysis b/w sender and receiver. • Need to extract ‘artifacts’ of original encoding embedded in the signal received at the PSTN side.
Call Separation Ideologies : Past Work • Frequency Cutoffs • Different compression technologies -> different sampling rates • Power components not observed after half of sampling frequency Power spectrum of a recorded VoIP call
Call Separation Ideologies : Past Work • Packet Loss • VoIP suffers from packet loss as compared to PSTN • Gateway might compensate for loss in a characteristic way, generating patterns. • Detect patterns of silence, interpolation or repetition.
Call Quality Based Classifier • Codecs reduce call quality by compression. • PESQ measures call quality by providing a quality value from 0 to 4.5 • Perceptual Evaluation of Speech Quality • Different codecs compress differently • It is possible to correlate quality value to a codec.
Call Quality Based Classifier • Codecs applied in sequence -> stepwise drop in quality • Possible mapping between final PESQ value and a sequence of codecs applied • First codec in a sequence will tell whether call is VoIP or PSTN
PESQ value for G.711 encoding PESQ value for sequence of Speex -> G.711 -> G.711 encoding
Frequency Cutoffs Revisited • VoIP clients use variable bit rate codecs which sample at up to 16KHz • Original high frequency components might remain in signal after final PSTN encoding • Aliasing • Skype and Speex use wideband VBR
Future Work Implement one of the classifying ideologies in code and run it on a ‘Honeypot’ at GT