Keystroke interference

For the IEEE 802.11 MAC Protocol. Wirel. Netw. 2001, 7, 159–171. [Google Scholar] [CrossRef]Ziouva, E.; Antonakopoulos, T. CSMA/CA performance under high traffic conditions: Throughput and delay analysis. Comput. Commun. 2002, 25, 313–321. [Google Scholar] [CrossRef]Chen, Y.; Agrawal, D.P. Effect of Contention Window on the performance of IEEE 802.11 WLANs. In Proceedings of the 3rd Annual Mediterranean Ad Hoc Networking Workshop, Bodrum, Turkey, 27–30 June 2004; Citeseer: University Park, PA, USA, 2004; pp. 27–30. [Google Scholar]Manshaei, M.H.; Hubaux, J.P. Performance Analysis of the IEEE 802.11 Distributed Coordination Function: Bianchi Model. Mobile Networks. 2007. Available online: (accessed on 15 July 2023). Figure 1. Interference topology. Figure 1. Interference topology. Figure 2. Wi-Fi sensing system overview. Figure 2. Wi-Fi sensing system overview. Figure 3. WiFi-based keystroke detection model. Figure 3. WiFi-based keystroke detection model. Figure 4. Keystroke recognition experimental setup. Figure 4. Keystroke recognition experimental setup. Figure 5. Classification accuracy. Figure 5. Classification accuracy. Figure 6. Impact of CSI Rate. Figure 6. Impact of CSI Rate. Figure 7. Collision probability of 802.11n networks with increasing users. Figure 7. Collision probability of 802.11n networks with increasing users. Figure 8. Throughput of 802.11n networks with increasing users and different packet lengths. Figure 8. Throughput of 802.11n networks with increasing users and different packet lengths. Figure 9. CSI snapshot rate under varying packet lengths. Figure 9. CSI snapshot rate under varying packet lengths. Figure 10. Experimental setup for validating analytical model. Figure 10. Experimental setup for validating analytical model. Figure 11. CSI snapshot rate with varying number of stations. Figure 11. CSI snapshot rate with varying number of stations. Figure 12. Setup for interference experiment. Figure 12. Setup for interference experiment. Figure 13. CSI time series with no interference. Figure 13. CSI time series with no interference. Figure 14. CSI time series with 50 Hz interference. Figure 14. CSI time series with 50 Hz interference. Figure 15. CSI time series with 250 Hz interference. Figure 15. CSI time series with 250 Hz interference. Figure 16. CSI time series with 500 Hz interference. Figure 16. CSI time series with 500 Hz interference. Figure 17. CSI time series with 1 kHz interference. Figure 17. CSI time series with 1 kHz interference. Figure 18. Effect of interference on Wi-Fi sensing accuracy. Figure 18. Effect of interference on Wi-Fi sensing accuracy. Figure 19. Security rate for each key as interference increases. Figure 19. Security rate for each key as interference increases. Figure 20. BW vs. interference. Figure 20. BW vs. interference. Figure 21. Privacy gains vs. BW. Figure 21. Privacy gains vs. BW. Table 1. Approximated data rates for various sensing applications. Table 1. Approximated data rates for various sensing applications. ActivitySpeed (m/s) T coherence (s)Req. CSI Rate (Hz)Walking1.50.01663Running2.70.008129Typing40.005189Occupancy0.70.03133 Table 2. 802.11n CSMA/CA parameters. Table 2. 802.11n CSMA/CA parameters. ParameterValueSlot Time20 μsDIFS 50 μ sShort Inter-Frame Space (SIFS) 10 μ sAcknowledgement (ACK) Time 304 μ s T Header 400 μ s T Payload 3.4 msMinimum Contention Window (W)16 slots Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications

Accuracy, we proceed with evaluating the sensing output of the keystroke recognition threat model using this dataset.In Figure 18, we present the sensing accuracy against the interference packet rate. The x-axis represents the packet rate from the interfering laptop. The green left y-axis then represents the accuracy of the keystroke recognition system as it operates in the presence of said interference, and the blue right y-axis represents the CSI data rate under each level of interference. For each interference rate, the blue point thus denotes the achieved CSI data rate (Hz) and the green diamond denotes the achieved sensing accuracy. There is a clear negative correlation between the interference frequency and the sensing accuracy. With an initial interference rate of 50 Hz, the sensing accuracy drops from 85 % to 70 % . Beyond this, the sensing accuracy crosses below 50 % at an interference rate of 500 Hz, corresponding to a CSI data rate of 700 Hz. This corroborates our observations from Figure 13, Figure 14, Figure 15, Figure 16 and Figure 17, where the keystrokes could no longer be clearly identified with interference rates of 500 Hz. Beyond this, the accuracy appears to decay in a decreasing manner asymptotically. The poor performance of the sensing system with a CSI Data Rate of 700 Hz due to the interference injected at 500 Hz is an interesting result. We recall the earlier result in Figure 6, where a CSI sensing rate of 700 Hz was able to yield a keystroke detection rate of 100 % . Once more we reiterate that detection refers to the ability to pick up the keystroke event in the CSI time series, not the classification accuracy. Despite being able to pick up all the keystrokes, the sensing classification accuracy is below 50 % . This result corroborates the ideas proposed in Section 3, where we postulated that the dependence of Wi-Fi sensing systems on frequency-based feature engineering techniques creates a susceptibility to changes in the CSI data rate. Even though all the keystrokes are picked up within CSI, the change in CSI sampling rate distorts the DWT features and hence reduces the accuracy of the sensing system. Overall, this experiment clearly shows that minimal interference can still be effectively used to defend against Wi-Fi sensing. To further visualise the improvement in privacy, we define the Privacy Preserving Probability (PPP) metric as the percentage probability of safeguarding a key from being detected by Wi-Fi sensing. This is calculated as the complement of the true positive detection rate for each key. We plot the PPP under several rates of defensive interference in Figure 19. The x-axis represents the five keys that the Wi-Fi sensing system is attempting to detect. Then, the height of each column represents the PPP with different levels of defensive interference. We observe that the PPP generally increases as a function of defensive interference. Generally, the PPP is higher for all keys when comparing 500 Hz interference with 0 Hz interference, and higher again

The analytical result implies that a CSI data rate of 4400 Hz is achievable with a channel bit-rate of 72 Mbps, the actual bandwidth is limited by the computation power of the wireless devices and their Network Interface Cards (NICs). Overall, this result supports the theoretical model for contention-based defence against undesired Wi-Fi sensing attacks. 6. Application of Defence Model Against Keystroke AttacksThe previous section demonstrated analytically and empirically that channel contention can reduce the available CSI data rate. We utilise these ideas to develop a defence mechanism to thwart the efficacy of Wi-Fi-based keystroke recognition. 6.1. Experimental SetupWe utilise the same experimental apparatus as before, with the addition of an interference device as depicted in Figure 12.The interference device is a Macbook Pro laptop on the same communications channel (Channel 2 of the 2.4 GHz Wi-Fi spectrum). It is configured to propagate ping packets at varying rates between 1300 Hz and 0 Hz, whilst the keyboard sensing apparatus operates with Tx propagating ping packets at 1300 Hz. As demonstrated by the ellipses in Figure 12, the proximity of these devices causes the range of their communications to overlap, resulting in competition for control over the channel resources. As illustrated by the wavy red signal in Figure 12, movement of fingers on the keyboard causes time varying changes in the Wi-Fi signal which are classified by the keystroke sensing system. To investigate how the varying interference packet rate influences the accuracy of the keystroke sensing system, we once more collected CSI data during 100 keystroke events for each of the letters Q, Z, H, P and M and used the model in Section 4 to train a sensing system. The system was then evaluated with a dataset of CSI collected during 50 keystrokes for each of the same keys. We collected 6 such sets of testing data, under varying rates of contention from the laptop. 6.2. ResultsAs the laptop begins to transmit ping packets, the captured CSI waveforms degrade, as demonstrated by Figure 13, Figure 14, Figure 15, Figure 16 and Figure 17.We observe firstly from Figure 13 that the keypress events are clearly visible in the sampled CSI as time-varying ripples, in line with our expectations as discussed in Section 3 and in line with our earlier results in Section 4. In each Figure, a single keystroke is outlined by the dotted box, and the arrows point towards the other observed keystrokes. Then, in Figure 14, Figure 15, Figure 16 and Figure 17, there is an incremental degradation in the quality of the CSI and the clarity of the keystroke events. In Figure 14 and Figure 15, we are still able to discern three clear keypress events; however, in Figure 16 and Figure 17, this is no longer possible. This gradual degradation of the CSI time series further supports our idea that maximum interference rates are not necessary, and lower interference can be applied to sufficiently degrade sensing. To understand the impact of the degraded CSI on the sensing

4B, we were unable to investigate whether the linear trend in Figure 20 approaches an asymptote with further increases in the transmission rate. We motivate future work to investigate this, when commodity devices are able to propagate more dense packet streams. To visualise the trade-off between the security and channel utility, Figure 21 plots the PPP gain against the measured throughput loss. Here, the PPP gain is the percentage increase in the PPP over the baseline scenario with no interference. We observe a clear trade-off between the PPP gain and the available throughput. To achieve a PPP gain of 60 % , we sacrificed nearly 25 % of the available Wi-Fi throughput. Initially, we observe that a 20 % gain in the PPP can be achieved with only 4 % loss of throughput. This corresponds to the case of interference at a rate of 200 Hz. Referring back to Figure 18, this conservative interference rate is sufficient to reduce the keystroke sensing accuracy below 60 % . The trade-off becomes more severe beyond this, as shown by the increasing slope of the curve in Figure 21. A further 20 % gain in the PPP (from 20 % to 40 % ) entails a much larger throughput loss of 11 % . These results demonstrate the effectiveness of conservative interference injection in significantly mitigating sensing threats, while having minimal impact on Wi-Fi throughput. 8. Concluding RemarksThis paper introduces a conservative interference injection strategy to counteract privacy and security threats from Wi-Fi sensing. Through a survey of prior work in Wi-Fi sensing as well as the construction of our own testbed, we demonstrate that fine-grained human activities are under threat of being monitored with high accuracy. By experimental analysis using commodity devices, we demonstrate that this high accuracy of Wi-Fi sensing is highly dependent on the CSI data rate. We subsequently showed analytically and experimentally that this CSI data rate can be intentionally disrupted by other devices in the wireless channel. This principle is the basis of prior work in the state-of-the-art that utilises wireless interference to reduce the CSI rate and diminish sensing accuracy, thereby protecting user privacy. This paper further investigates this, revealing a positive correlation between the degree of interference introduced and the degradation in sensing accuracy. This establishes a trade-off between achieving increased privacy and maintaining acceptable channel throughput for other Wi-Fi users. Compared to state-of-the-art approaches, we can thwart sensing applications with comparable efficacy at a fraction of the cost to channel bandwidth by injecting interference conservatively. Although our approach is applied specifically to keystroke recognition, the results are generalizable to other Wi-Fi-based gesture recognition systems, where interference can degrade the resolution of environmental movement. We suggest that future work should investigate the suitability of this conservative interference injection scheme in commercial settings under interference from additional devices. In such environments, the consumption of channel resources by other devices will create uncertainty around our ability to finely control the CSI rate of a sensing system. Furthermore, the

O Logs the Mac's IP address o Automatically runs at startup stealthily o Enables you to apply settings to all users with one click o Sends logs to Email/FTP at preset intervals o Password protects keylogger access.The installed spyware can also help the attacker perform the following on target computers:o Steals users' personal information and sends it to a remote server or hijacker o Monitors users' online activity o Displays annoying pop-ups o Redirects a web browser to advertising sites o Changes the browser's default setting and prevents the user from restoring o Adds several bookmarks to the browser's favorites list o Decreases overall system security level o Reduces system performance and causes software instability o Connects to remote pornography sites o Places desktop shortcuts to malicious spyware sites o Steals your passwords o Sends you targeted email o Changes the home page and prevents the user from restoring o Modifies the dynamically linked libraries (DLLs) and slows down the browser o Changes firewall settings o Monitors and reports websites you visit.Source: Spytech SpyAgent is a computer spy software that allows you to monitor everything users do on your computer—in total secrecy. SpyAgent provides a large array of essential computer monitoring features, as well as website, application, and chat client blocking, logging scheduling, and remote delivery of logs via email or FTP.Source: Spy is a PC-user activity monitoring software. It runs and performs monitoring secretly in the background of computer system. It logs all users on the system and users will not know its existence. After you install the software on the PC you want to monitor, you can receive log reports via emails or FTP from a remote location, for example, every hour. Therefore, you can read these reports anywhere, on any device at any time as long as you have Internet access. Power Spy lets you know exactly what others do on the PC while you are away.The following are some more ways to defend against keyloggers: Use pop-up blockers and avoid opening junk emails Install anti-spyware/antivirus programs and keep the signatures up to date Install professional firewall software and anti-keylogging software Recognize phishing emails and delete them Update and patch system software regularly to defend against keyloggers Do not click on links in unwanted or doubtful emails that may point to malicious sites Use keystroke interference software, which inserts randomized characters into every keystroke. Antivirus and antispyware software is able to detect any installed software, but it is better to detect these programs before installation. Scan the files thoroughly before installing them onto the computer and use a registry editor or process explorer to check for keystroke loggers. Use the Windows on-screen keyboard accessibility utility to enter