Thursday, August 25, 2016

Eliminating Feedback in Next-Generation Cellular Networks

Best paper award winner.
Authors: Deepak Vasisht (presenting remotely), Swarun Kumar, Hariharan Rahul, Dina Katabi
This paper pioneers solving an important challenge for the deployment of next-generation cellular networks (5G MIMO and others): inferring channel parameters without explicit feedback from the clients. In the U.S., such task is greatly complicated by the widespread use of Frequency Division Duplexing, which means that the uplink and downlink connections (from and to the cell tower) use different frequency bands.

The authors’ solution to this problem is a transform that uses estimated path information to infer channel parameters, using a phase-spatial power profile measured at the AP, following a simple observation: "while the channels change with frequencies, the underlying physical paths traversed by the signal stay the same” (p. 399). Therefore, the system is capable of estimating wireless channels on a different frequency band to that where the measurements are collected, all on a band that’s only 20 MHz from commercial providers in the U.S.

The paper’s algorithm (sections 4 and 5) accounts for multiple complications in the estimation of channel parameters: phase variations related to the different wavelengths of each channel; windowing effect, related to the spatially-limited location of antennas on cell towers; and superposition, following multi-path propagation. They solve each of these issues using a combination of mathematical techniques (including modified fourier transforms).

Q: Were the transmitters and receivers on the same elevation? Can you comment on three-dimensional aspects.
A: Yes. For testing the angular analysis, the two-dimensional testing is all that matters.

Q: When you estimate the antenna parameters, how many measurements do you take to minimize channel error?
A: No need for multiple measurements.

Q: You need uplink packets being sent before, which might not work for UDP?
A: When the link is established an uplink packet is sent, before any protocol kicks in.

Q: Can you comment on the analysis inaccuracies introduced?

A: As far as we can estimate it, the analysis works quite nicely.