upper C"/> for MNIST recognition.Figure 5.22 Structure of
for MNIST recognition.Figure 5.23 Test accuracy of InfNets and ObfNet–InfNet concatenations for MN...Figure 5.24 Obfuscation results of ObfNet
on MNIST.Figure 5.25 Structure of
for the ASL dataset.Figure 5.26 Test accuracy of InfNet and ObfNet–InfNet concatenations for ASL...Figure 5.27 Obfuscation results of ObfNet on ASL.Figure 5.28 InfNet's per‐sample execution time on Jetson versus batch size. ...Figure 5.29 Data sample transmission time versus network connection data rat...
6 Chapter 6Figure 6.1 Per‐second ENF measured at two locations
apart in a city. Mean ...Figure 6.2 Ac voltage and EMR data.Figure 6.3 Illustration of an EMR natural timestamping system.Figure 6.4 EMR capture devices.Figure 6.5 Challenges of zero crossing detection in a noisy EMR signal, and ...Figure 6.6 Comparison between the dead‐zone approach and our BPF‐based appro...Figure 6.7 Mean decoding error versus natural timestamp length under differe...Figure 6.8 Impact of sampling rate and timestamp length on the RPi‐based EMR...Figure 6.9 ENF trace computed based on EMR measurements captured by the Z1 m...Figure 6.10 Impact of sampling rate and timestamp length on the timestamping...Figure 6.11 ENF trace estimated by the Z1 mote based on EMR signals sampled ...Figure 6.12 Distribution of decoding errors of RPi‐based EMR sensor at sites...Figure 6.13 Decoding errors in one month. Each error bar is computed based o...Figure 6.14 Deployment spots at site B. An RPi‐based EMR sensor is deployed ...Figure 6.15 Z1 decoding error versus EMR signal strength.Figure 6.16 Decoding errors of Z1 under the sliding and neighboring approach...Figure 6.17 Factory floor plan and EMR sensors' locations.Figure 6.18 ENFs sensed by the RPi‐based EMR sensor and the Z1 on the factor...Figure 6.19 Natural timestamp decoding errors in the factory environment. Th...Figure 6.20 Dividing a natural timestamp to sub‐natural timestamps to verify...Figure 6.21 PDF of decoding offsets when the decoding condition is respectiv...Figure 6.22 A schematic of the EMR spoofer.Figure 6.23 EMR spoofer (left) and spoofing experiment (middle and right).Figure 6.24 The ENF sensed by the victim EMR sensor when the EMR spoofer is ...Figure 6.25 Packet delay attack and secure clock synchronization.Figure 6.26 NTP principle and packet timestamping.Figure 6.27 Performance of NTP over a BLE connection.Figure 6.28 Flora.Figure 6.29 TouchSync prototypes.Figure 6.30 No human body contact.Figure 6.31 iSEPs on the same wearer (shared ground).Figure 6.32 iSEPs on different wearers (shared ground).Figure 6.33 Absolute time displacement
between the EMR signals captured by...Figure 6.34 iSEPs on the same wearer (independent grounds).Figure 6.35 iSEPs on different wearers (independent grounds).Figure 6.36 Normalized range and time displacement
of iSEPs under differen...Figure 6.37 A synchronization process of TouchSync.Figure 6.38 iSEP signal processing pipeline.Figure 6.39 A synchronization session of TouchSync. The vertical arrows repr...Figure 6.40 An example of solving the integer ambiguity. The transmissions o...Figure 6.41 Convergence speed of IAS.Figure 6.42 Distribution of absolute clock offset estimation errors.Figure 6.43 An example of IPS‐based TouchSync (unit: ms).Figure 6.44 Convergence speed under various settings of IPS period.Figure 6.45 Impact of signal strength.Figure 6.46 Wearing position.Figure 6.47 EMR signals near an electric oven and a microwave oven.Figure 6.48 Errors introduced into time displacements of the EMR signals.Figure 6.49 Laboratory floor plan with test points marked.Figure 6.50 Home floor plan with test points marked.Figure 6.51 One‐way delays over a ngrok tunnel.Figure 6.52 Accuracy of TouchSync‐over‐Internet.
upper C"/> for MNIST recognition.Figure 5.22 Structure of 
