hannahm8
commented
3 years ago Section IV. The term “Markov process” may not be familiar to readers. [E]
- [x] Clarify / add explanation [added sentence to define]
While I applaud your efforts to explain the Viterbi algorithm (VA), I still think there’s room for improvement. I don’t understand how it improves the signal to noise ratio. [E] [Also linked to Figure 4 comments in issue #19]
- [x] Clarify Viterbi algorithm explanation
- [x] Clarify that it's strength is in finding wandering signals
daccordeon
The data analysis problems discussed in this paper are related to continuous wave searches. However, I did not find a mention of the key feature involved in these searches, namely, Doppler modulation of the signal frequency caused by the rotation of the Earth. This is the principal reason behind the extremely high computational cost of CW searches. A discussion of doppler modulation should be included. (It would be neat if the setup could be extended in the future to show the Doppler effect.)
Is the HMM defined properly in Sec. IV A? The observables here are the amplitudes of the spectrogram pixels at a given time index and the hidden markov process is the underlying drift of frequency and amplitude of the signal. I was expecting to see a conditional pdf that connects the state of the underlying (suitably discretized) process to the observables. I don't see it. [R3]
Is Eq. 1 correct? The LHS is a marginal probability while the RHS is a joint probability. Shouldn't there be a marginalization on the RHS? [R3]
The normalization used for F(t_i, f_j) should be defined explicitly. [R3]
Looking at Fig. 3, it appears that the noise is not white and rises in power at lower frequencies. How is the color of the noise taken into account, if at all, in the HMM or the Viterbi algorithm? A discussion of this point should be included although it is fine to assume white noise as a first approximation. [R3]
The Viterbi algorithm used in this paper falls along the general lines of finding the optimal path through a chirplet graph (which includes the spectrogram) that has been explored extensively in the GW literature. More broadly, the detection and estimation of chirp signals, which leave a track in the time-frequency plane, has been a topic of many papers. For the benefit of the readers, the authors should provide an expanded literature review in Sec. IV A and state clearly if their approach is new and, if so, how it differs from the existing ones. A partial list of papers is given below: [R3] E. Chassande-Mottin and A. Pai, Phys. Rev. D 73, 042003 (2006). E. J. Candes, P. R. Charlton, and H. Helgason, Classical Quantum Gravity 25, 184020 (2008). W. G. Anderson and R. Balasubramanian, Phys. Rev. D 60, 102001(1999). P. Addesso, M. Longo, S. Marano, V. Matta, I. Pinto, and M.Principe, in 2015 3rd International Workshop on Compressed Sensing Theory and its Applications to Radar, Sonar and Remote Sensing (CoSeRa) (IEEE, New York, 2015), p. 154. E. Thrane et al., Phys. Rev. D 83, 083004 (2011). E. Thrane and M. Coughlin, Phys. Rev. D 89, 063012 (2014). S. D. Mohanty, Phys. Rev. D D 96, 102008 (2017). Margaret Millhouse, Neil J. Cornish, and Tyson Littenberg, Phys. Rev. D 97, 104057 (2018).