Reviewer #1

1, We concur with the reviewer's comment that a beam splitter cannot serve the role of a trusted random number generator in a device-independent QKD scenario. However this discussion is outside the scope of this paper.

2, A trace of the unfiltered spectrum is now included in Figure.1 (a) and caption modified accordingly.

3, An additional description is added to the second paragraph in the "Implementation" section, stating specifically that the signal photons are sent to Alice and the idler photons are sent to Bob. The respective bandwidths of signal and idler are now stated in the first paragraph in the "Implementation" section. We didn't observe an appreciable difference in QBER when the other part of photons are sent through the long fiber possibly because a 4 nm bandwidth difference is not significant compare to other QBER contributions (imperfect fiber compensation, etc).

4, An additional description is added to the caption of Figure.1 (b) to clarify the details of a visibility measurement.

5, We added a description in the first paragraph in the "Performance" section that details the estimated contributions of QBER from different sources of error. The contribution to QBER are from accidental coincidences (1.9%), imperfect state preparation at the source (0.4%), and any remaining optical errors due to polarization optics, imperfect fiber compensation, and possible fiber depolarization.

6, Reference 36 is added as a sidenote explaining why only one side of the fiber needs to be compensated.

7, The typo is corrected.

8, The 1.9 ns FWHM of the coincidence peak is mainly due to the chromatic dispersion of fiber. Polarization mode dispersion contributes negligible amount of timing uncertainty (< 1ps). The timing jitter of the single photon detectors is on the order of 100 ps, which is also small compare to 1.9 ns. More quantitative details about how the width of the coincidence peak scales with distance can be found in reference 27.

9, We do believe that a narrower photon bandwidth would reduce the effect of polarization mode dispersion and therefore result in a lower optical error contribution to QBER. Regrettably we are unable to include a detailed characterization in this paper, as it is still work in progress. We appreciate the reviewers' insight on this matter and hopefully this can be covered in a future paper.

10, We added Figure.3 (b) to address the issue of the polarization stability of the fiber. The caption of Figure.3 is also modified accordingly.

11, The singles count rate is rounded to the nearest thousands of counts.

12, Description added is the third paragraph in the "Implementation" section stating that the PMD is measured with a commercial device (PMD analyzer FTB-5500B housed in a FTB-500 platform from EXFO).

13, The pump power of the source (2.4 mW) is specified again in the first paragraph in "Performance" section. 

14, A brief description is added in the caption Figure.2 (a) to address the role of the YVO4 crystal in the source setup. Reference 35 is also cited in the description which includes more details. Also note that in Figure.2 (a) the text label of YVO4 is corrected (previously was "YvO4")

15, Notation in Figure.2 (b) is now changed to "D" and "A"


Reviewer #2

1, We appreciate the reviewer's interest in knowing more about our polarization scheme. Polarization compensation of a fiber is a well understood technique that has been used in many fiber based QKD implementations (reference 31 for example). We would like to include this content in a future paper that covers this aspect in a more systematic way. 

2, Following our response to the previous comment: 4 LCVRs are used instead of 3 because each of our LCVRs does not provide a full 2pi retardance. A 4th LCVR is added to ensure that we can apply an arbitrary polarization rotation. An estimation of QBER contribution due to fiber compensation is given in the first paragraph in the "Performance" section. However, we cannot break down this fraction of error and distinguish between imperfect compensation and depolarization.

3, A break down of QBER is included in the first paragraph in the "Performance" section. (see also response to comment 5 from reviewer 1)

4, Description is added in the second paragraph in the "Implementation" section, stating that the signal photons are sent to Alice and the idler photons are sent to Bob. The signal photons are sent through the 10km fiber because of their relatively small bandwidth compare to the idler. We did not observe an appreciable difference in QBER by sending in the idler instead. (see also response to comment 3 from reviewer 1)

5, We have added Figure.3 (b) to demonstrate the polarization stability of the fiber and modifiec the caption of Figure.3 accordingly.

6, The title of the paper is changed to "Stable Polarization Entanglement based Quantum Key Distribution over a deployed Metropolitan Fiber" for clarification.

7, The typo is corrected.

8, Reference 46 is added to address this issue.