Slashdot Mirror
A Radiologist Has the Fastest Home Internet In the US (vice.com)
An anonymous reader writes: Jason Koebler via Motherboard has interviewed James Busch -- a radiologist and owner of "the first 10 Gbps residential connection in the United States" -- at a coffee shop in Chattanooga, Tennessee. Motherboard reports: "For reference, the Federal Communications Commission officially classifies 'broadband' as 25 Mbps. His connection is 400 times faster than that. Busch found a way to make good use of his 1 Gbps connection, and now he's found a use for 10 Gbps, too. 'An X-ray averages around 200 megabytes, then you have PET scans and mammograms -- 3D mammograms are 10 gig files, so they're enormous,' Busch said. 'We go through terabytes a year in storage. We've calculated out that we save about 7 seconds an exam, which might seem like, 'Who cares,' but when you read 20,000 or 30,000 exams every year, it turns out to be something like 10 days of productivity you're saving just from a bandwidth upgrade.' While 10 gig connections sound excessive at the moment, Busch says his family quickly started using all of its 1 gig bandwidth. 'We ballooned into that gig within eight or nine months. With my kids watching Netflix instead of TV, with me working, we did utilize that bandwidth,' he said. 'There were situations where my daughter would be FaceTiming and the others would be streaming on the 4K TVs and they'd start screaming at each other about hogging the bandwidth. We don't see that at 10 gigs.' So why does Busch have a 10 Gbps and the rest of us don't? For one, 10 Gbps offerings are rare and scattered in mostly rural communities that have decided to build their own internet networks. Most companies that have the technology offer gigabit connections (a still cutting-edge technology only available in a handful of cities) at affordable prices and 10 Gbps connections at comparatively exorbitant ones. In Chattanooga, 1 gig connections are $69.99 per month; 10 gig connections are $299. Thus far, 10 Gbps connections are available in Chattanooga; parts of southern Vermont; Salisbury, North Carolina; and parts of Detroit and Minneapolis. But besides Busch, I couldn't find any other people in the United States who have signed up for one. EPB, the Chattanooga government-owned power utility that runs the network, confirmed that Busch is the city's only 10 Gbps residential customer. Rocket Fiber, which recently began offering 10 Gbps in Detroit, told me that it has 'no customers set in stone,' but that it's in talks with prospective ones. Representatives for U.S. Internet in Minneapolis and Fibrant in Salisbury did not respond to my requests for comment. Michel Guite, president of the Vermont Telephone Company, told me his network has no 10 Gbps customers, either."
Well ... if you are too lazy to Google and use Wikipedia, I'll provide you some 'decompression' of GP's Jargon. -Anonymously-, because I'm using modpoints.
SDR - Software Defined Radio. A type of radio equipment that uses an Analog to Digital converter to digitize the radio signal (as close to the antenna as is practical) and use digital processing with software filters instead of hardware ones. Non-techie examples: All WiFi and bluetooth equipment, most TV and radio USB sticks
TDOA - Time Difference of Arrival. Apparently GP wants to locate radio sources by measuring the time difference between signals arriving at a couple of radios. It's for triangulation purposes, I guess. You should then have 3 radios at least to make this work failsafe (only 2 radios leave you with two planar points to guess from, or when also uncertain of height, a circle). All three must be close enough to the source to identifiably pick up the signal and you need synchronized timers with each of them unless your are absolutely certain of your connection lag.
Now for the bandwidth usage. He wants to sample at 200 mega-samples per second (Msps). What does he want to sample? 16 bit complex samples. Complex numbers have two components and are twice as large as real numbers. Analogously, complex samples take twice the space of normal samples.
200 Msps * 16 (bit) * 2 (complex) = 6400Mb/s = 6.4Gb/s raw data rate. When sent uncompressed and package switched over the internet, add some overhead and you'll get even closer to that 10Gb/s. But as I said earlier, he probably wants two remote stations for full triangulation, so 10Gb/s isn't going to cut it unless he can use some compression or compromises a bit on the bandwidth.
Expert question: Why complex samples? A common technique in Software Defined Radio is to sample the signal, straight and 45 degrees phase shifted, also called Quadrature Sampling. This has a long technical explanation why it's advantageous but to cut short, it's so you'll be able to differentiate between signals above and below the mixer frequency of your SDR. Else you wouldn't know if a signal 'Y' Hz away from the mixer frequency 'X', originated at frequency 'X'-'Y' or 'X'+'Y'.
A typical modern X-ray imager is about 20 megapixels at 16 bit (so 40 MB). A typical exam of a chest may use 2 full frames. Other types of examination may use more frames, particularly images of joints (although in that case, you don't need a full field). On that basis, 200 MB per exam seems completely unreasonable.
Typically, medical images are stored and transmitted as raw, uncompressed bitmaps. Lossy compression while it is supported has some issues, technical as well as legal. As an example of a technical issue the classic JPEG algorithm does not support 16-bit depth images. Compression when supported is typically with JPEG 2000, but some implementations may use JPEG LS, and older software lossless JPEG.
The normal networking protocol for medical image transfer is not very efficient as it does not allow pipelining of data transfers. For studies like a CT scan, which might have 10,000 images of 256 kB each, connection latency can be devastating. As a result, there are a profusion of proprietary proxy server packages, which basically wait for a study to arrive, package it into a tar file, and the ftp it to the other end, which then untars the files and retransmits them. This has the additional advantage that FTP is much better behaved when traversing NAT, dynamic IPs or firewalls than the industry standard data transfer protocol (DICOM).
That said, 10 Gb to the end-user seems excessive. I use a pair of 1 Gb links to the servers on our systems, and 1 Gb to the end-user which serve 20 radiologists. The network backbone is 10 Gb, but that includes all the other traffic. We have 5 Mb uplink from the hospital site, which is perfectly fine to allow multiple radiologists to work from home. It's all about the software finesse - a decent image viewer will seamlessly use lossless image compression, a proprietary transfer protocol using streaming/pipelining, and you supplement this with work queues. You allocate 50 studies to a radiologist, and while the radiologist is reading the first study, the 2nd study is downloading in the background, etc. Not all software is clever enough to do this, and not all supports image compression or techniques to deal with high latency networks.
The complex samples are quadrature sampling (I+jQ); just a different name. It's actually 90 degrees phase shift, and you do it for the local oscillator, which is where the j comes from. Yeah, I'd want two, so I'd have to decimate to 100 MHz BW. But as you said, now I need two 10G connections. I suppose my point was there are probably many uses for 10G connections; build it and they will come. Hell, maybe within the next year Google fiber will finally get to my house, and at least I can get 20 MHz real-time BW.
Municipally owned. Provided to citizens at, basically, cost.
I've got "gigabit" fiber at my house for about the same price, although it slows down to around 400-500 Mbps once you leave my provider's network. Provided by a private company. Yeah, it's awesome. Zero downtime since it was installed.