hchenji

very interesting!! Do you have a source for this? Would like to read up on it

Yes. See Haar Basis on page 4 of http://www.fourierandwavelets.org/FSP_v1.1_2014.pdf

Lectures and video from http://www.cse.wustl.edu/~jain/cse574-16/index.html

Book: https://www.pearsonhighered.com/program/Beard-Wireless-Commu...

In my experience WiFi clients at 2.4GHz never end up using the high MCS indices whenever they transmit (whenever the channel is clear). This has something to do with the fact that the CCA thresholds are fairly high, and that high SNR (30-50dB) is required for activating the high MCS indices[1]. I don't think these SNRs are achievable in a typical setting.

It is true that MAC backoffs are also a contributing factor to the throughput being low. The CCA assessment procedure detects both wifi preambles and non-WiFi interference (pure energy detection), which is why I say interference is modeled as noise. CCA does not for example, have some intelligent coexistence algorithm for dealing with zigbee or LTE-U or other ISM traffic.

[1]: http://www.revolutionwifi.net/revolutionwifi/2014/09/wi-fi-s...

See my comment https://news.ycombinator.com/item?id=13793876 .

I suggest reading Andrea Goldsmith's or Pramod Vishwanath's book on Wireless Communication. If not, Intro to Communication Systems by Madhow is recommended.

Reducing transmit power reduces interference, but it also reduces capacity since the received power (and hence the SNR at the receiver) is reduced! The Shannon limit on channel capacity dictates exactly how much bit rate (C=B*log2(1+SNR)) is achievable over a channel. SNR is a fundamental ingredient of this equation.

802.11ax won't mean anything unless you reduce interference (see my reply below). Without high SNR, all these new efficiency techniques won't get you anything. Shannon capacity is still what limits performance.

The article misses out on explaining WHY we cannot get the promised bitrates. The answer is Shannon's limit on channel capacity, which mandates that you pay in either bandwidth or high power or low noise (SNR) to get higher capacity. Now these WiFi devices have internal rate adaptation algorithms that choose a particular modulation and coding scheme (MCS) index based on the measured SNR. A higher MCS index means better higher bits per symbol (modulation) and lower overhead code rates, and more antennas (spatial streams), which is how you get the xx Gbps bandwidth advertised on the box. List of MCS indices: http://mcsindex.com/

In today's devices, interference is considered as noise, which means that SNR simply drops to a point where the higher MCS indices are not chosen at all. So, even though the device is capable of the advertised XX Gbps bitrate, the SNR isn't high enough to switch to those higher rates.

No, since wired networks are full duplex and are their own collision domains. No need to do carrier sensing at all.

You mean like 3G/4G/LTE?

You can also do channel bonding on 5G and use 40MHz channels. Almost double the capacity!

Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway. — Tanenbaum, Andrew S. (1989). Computer Networks. New Jersey: Prentice-Hall. p. 57. ISBN 0-13-166836-6.

What would be more interesting is the internals of numpy. What is it that makes numpy 'blazing fast'?

Instead, there is only high school level math in the article, with dot products and matrices and so on.

Then why not Dijkstra + source routing?

Use a Pseudoinverse or a decomposition instead. A great book to read is Matrix Computations by Golub/van Loan.

Any idea why they chose DV based routing? Looks like they have ignored about 20 years of Wireless Sensor Networks/cyber-physical systems research.

Interesting read. Could you blog about the math/techniques/algorithms you used to determine the best fiber paths?

Consider writing a SIGCOMM paper about it!!

Why doesn't the article mention Southwest Airlines??

So can we use it in a macbook somehow with a usb air interface?

EE degrees yes definitely. I wish CS students learnt it too. Every fancy language du jour eventually has to generate machine code after all.

Yeah but if GNULinux is free in the first place, why spend so much govt. money on it?

Just because borders are open does not mean that people will immediately move.

That said, CA and RI have similar quality of life, but Africa and North American probably do not. So CA/RI is not a good example.

Yes I agree...instead of importing low skilled labor, we should import highly skilled immigrants. Increasing the H1B and employment based greencard quotas is absolutely necessary.

The Wire (season 2) gives an excellent insight into how docks are run (Baltimore). There's even a scene where some Dutch company wants to bring Rotterdam's automation to Baltimore.

Is this new or has it existed since some time?

Why should we care how the science was applied?

'"Once the rockets are up, who cares where they come down That's not my department," - says Wernher von Braun' - Tom Lehrer

Nice explanation. Is there any literature that explains how faces are stored in memory? Fascinating.

Why can't we just do it in hardware/FGPAs at that point?

Anyone know if the font rendering issue was fixed? The email reading pane in Outlook (on Windows) didn't do any subpixel smoothing, and so the fonts looked fugly as a result. Hopefully they fixed it.

Office for Mac 2016 doesn't seem to have this issue though.