Results and Discussion

In this section, we discuss the results of our proposed algorithm in simulation - and discuss the changes made to make the simulation as close as possible, to real life scenarios.

Comparison Baselines

  • PN (Pseudonoise): Uses non-sinusoid modulation and demodulation functions. Proposed for early ToF cameras in 2007.

  • ACO (AC Orthogonal): Using orthogonal frequencies for the interfering cameras.

  • SEC (Stochastic Exposure Coding): Makes use of a stochastic TDMA protocol

  • CMB: Combined ACO and SEC

  • CSMA: Intervention to use a-priori carrier sensing

In the figures below, we compare the performance of the proposed multiple access channel sensing (CSMA) protocol, compared to other baselines.

The X-axis shows the number of interfering cameras, while the Y-axis is the plot of the Inverse Standard Deviation. These graphs show how precise a particular protocol is. We evaluate these protocols for the systematic errors, and hence the standard deviation measurements are good for this purpose. Further, a higher plot in these graphs implicates a better performance.

CSMA with the same power amplification as the prior works of SEC and CSMA shows a huge improvement in the inverse standard deviation measurements.

CSMA with 50% of the power amplification as the prior works of SEC and CSMA shows a huge improvement in the inverse standard deviation measurements.

We evaluate the proposed algorithm in a simulation, and to make the simulations as close as possible to real-life scenarios, we must also consider the time spent to read out sensor values after channel sensing, and the time spent to make the decision as to whether or not there is an interfering source in the current slot.

For this purpose, we use another parameter 'frac' which denotes the fraction of the slot used for channel sensing. The outcome of such a fractional slot is that the time available for image capture will be less, thereby degrading the performance of the protocol. The results are as follows:

(A) CSMA with the fraction 0.1, and 50% of power amplification as the prior works of SEC and CSMA shows a small degradation in the performance compared to the simulation results of a simple protocol.

(B) CSMA with the fraction 0.3, and 50% of power amplification as the prior works of SEC and CSMA shows a further degradation compared to figure (A) alongside.

(C) CSMA with the fraction 0.5, and 50% of power amplification as the prior works of SEC and CSMA - the performance drops even below the current methods.

(D) CSMA with the fraction 0.5, and 75% of power amplification as the prior works of SEC and CSMA shows animprovement in performance owing to the increased power.

Observations:

  1. From the above figures, we can conclude that as the fraction of slot used for carrier sensing increases, there is a drop in the performance of the channel sensing protocol (as evident from figures A-C) - which is intuitive because of the reduction in the time available for capturing scene light.

  2. However, such a drop can be compensated by having a higher source peak power amplification. (from figures C-D)

  3. Even while having a fraction of 0.5 for the channel sensing, the channel sensing protocol outperforms exising methods, while using only 75% of the peak power amplification as prior works.

Further, our proposed algorithm also faces the problem of "Early Start Preference" - where in the presence of multiple cameras following the CSMA protocol, the camera which starts earliest shall gain ON access to all of its slots while all other cameras following CSMA protocol shall only get access to the medium after the earliest camera has completed its operation.

To tackle this, we introduce stochasticity in the CSMA protocol - even if the channel is found to be idle in a particular slot, light shall be transmitted with a probability pON. The impact of this is shown in figures below.

CSMA with the same power amplification as the prior works of SEC and CSMA, 0.5 pON shows a degradation in performance - owing to smaller time for light capture.

CSMA with 50% of the power amplification as the prior works of SEC and CSMA, 0.8 pON - the degrade in stochasticity can be compensated by the increase in the pON probability.

Conclusion

  1. CSMA approach gives a better performance over SEC under the same experimental setting (keeping all parameters same).

  2. There is a trade-off between time spent in collision detection and the peak power amplification of source.

  3. In all of our results, we could obtain a significant performance improvement over the existing works, with a lower power requirement.