10 gigabit Ethernet 80 GHz point-to-point bridges | Hacker News Reader

10 gigabit Ethernet 80 GHz point-to-point bridges

elva-1.com

166 pointswalrus0191 comments

It's interesting to see the huge channel bandwidths they're using to attain this. Advances like this are largely driven by higher sample rate ADCs and DACs becoming more viable in recent years.

Edit for clarity: channel bandwidths in the datasheet are up to 2 GHz. Need to close the link at 32 QAM to hit 10 gb before error correction overhead at that bandwidth, which is certainly doable. Also, it's interesting to note that they quote 7 gbps throughput @ 1 GHz bandwidth, and 10 gbps at 2 GHz. It implies that they can run at 256 QAM at 1 GHz, but only 32 QAM at 2 GHz, which also makes sense.

Having worked in the satellite communications industry, where channel sizes used to be limited to 72 MHz (because of hardware limitations of the spacecraft), getting modems designed to operate on larger channels is no small task. I would love to learn more about the internal architecture of these radios to understand exactly what's going on - are they using interleaved ADCs? Is the modulation/demodulation being done in an FPGA, or an ASIC?

If there are terrestrial microwave engineers on here, I'd love to hear your thoughts on this!

terrestrial microwave engineer here: "80 GHz" is actually 71-86 GHz FDD. The original FCC band plan allowed for 5000 MHz wide channels each direction and basically OOK or BPSK level modulation. Newer radios use 250, 500 or 1000 MHZ wide FDD channels and QPSK or better. Incredibly wide channels can be used because it falls off in the atmosphere so rapidly after a few km, and the antennas are all very narrow parabolic reflectors with less than 1 degre beamwidth.

Basically no colocation interference issues are possible unless two companies try to shoot from the same rooftop, to also the same rooftop, using the same channel AND the same linear polarity.

Thanks for jumping in! Sounds like it's really nice for urban small-cell deployments, campus networks, etc. Do you experience a lot of issues with keeping links aligned at those tiny beamwidths?

Not really, 80 GHz is definitely difficult to aim, the very center of the beam is quite small. The mounts for the most popular antennas accommodate this and have very fine adjustments for the azimuth and elevation. Keeping them aligned long term? Not a problem as long as the mounts the radios are on (whether non penetrating or bolted to a wall/structure on a building roof) are done properly.

I'm wondering how the natural motion of some taller buildings affects this kind of deployment.

Lots of buildings swing and twist in the normal course of the day. We found this out with some of the free space optical solutions available.

It can have some effect but at the distances aimed (max 3km for a high reliability, high Tx power 80 GHz link) it doesn't affect much. A link that is a nominal -34 RSL on a clear sunny day might drop to -36.5 as a result of building sway (such as if one end is on top of a 60 floor tower). Weather is the main issue not heat expansion/contraction or sway.

You can pack many bits into a constellation point. Think 1024 QAM.

Most new licensed band 6, 11, 18, 23 GHz radios these days are 1024QAM capable. For 80 GHz the hot new thing is radios capable of 16/64/256QAM at varying code rates.

What's the limiting factor when designing to a constellation? Otherwise, what's improved about the chip or firmware to be able to move from 256QAM to 512QAM?

signal/noise ratio is a big one and RSL level. There are even 4096QAM radios now but they need a signal like -51 to operate at their full modulation. Otherwise they ACM down to 1024 (around -62), 256 and then 64QAM.

That's a pretty narrow ledge to stand on @ 800 GHz. Wonder what the FEC regime is?

>higher sample rate ADCs and DACs becoming more viable

Question, on optical networks doesn't this make dispersion more of a problem?

80 GHz isn't quite optical yet. It's a couple octaves down from infrared. SFAIK, it's still very much radio.

a great deal lower, the optical ITU DWDM grid is around 191.7 THz to 196.1 THz

The more general dividing line between microwave and infrared ( outside comms ) is 300 GHz, which was what I meant.

That is true. One of the fun things illustrating first-hand how microwave/millimeter wave and fiber are the same thing is if you ever get to handle a 120 GHz band waveguide/feed. Some radar and military stuff operates in the 120 band.

It's so incredibly tiny and narrow compared to a 11 or 18 GHz microwave waveguide. Then imagine continuing to make it narrower while increasing the frequency, follow that to its logical conclusion and you have imagined a 9/125 fiber.

But it's something of an apparently completely different... phenotype from a small waveguide.

The seemingly arbitrary 300Ghz line was probably more cultural than essential; the instrumentation was just very different. What's the analog for a prism for microwave?

Apparently a block of plastics like Delrin works:

http://www.physics.rutgers.edu/ugrad/205/manuals/micro.pdf

The key part in physics is it's still a wave... And a singlemode fiber is a type of waveguide. A 120GHz signal is a wave, a 171 THz signal is a wave, the latter just has a much shorter wavelength.

300Ghz ends up with a wavelength of 1mm, so it more follows the pattern of bandwidth assignment, which likes nice round numbers given in wavelength.

Back in 60's AT&T make lots and lots of circular waveguide, intending to bury it across the country. One of the circular modes is dispersionless; then fiber was invented; lot's of money lost on circular waveguide.

Interesting that mmWavguide is still rectangular. Wonder if one could do a mmWave fiber? Don't see why not.

millimeter wave waveguide on the rear of a parabolic antenna (30, 60cm dish) is actually circular (cylindrical), supporting dual polarity operation (opposite H and V). Individual radio heads like this russian radio or other 80 GHz radios are still single polarity. In certain scenarios you can mount two radio heads on a T-shaped waveguide on the rear of one dish antenna, one in H, one in V.

Thanks for the info. I got to mess with W-band stuff years ago, but I'm mostly Ku and below now. Would love to tear down some 80 GHz hardware.

Have you seen anything come of the "angular momentum" polarization discovered a few years ago? Seemed specific for P2P microwave.

I've yet to see a product of any sort I can buy with a credit card or purchase order. Would also like to know if anything comes of it.

I would love to hear more about that - have any links? From a quick google, all I can dig up makes it sound like it's normal circular polarization, which is used quite often in satellite. Not sure why it wouldn't make it into terrestrial, since it has some interesting advantages (no x-pol alignment issues!)

Google "angular momentum antenna". IEEE had a write up a few years ago, but details still lacking. It was simple as twisting the dish surface into a helix, but they didn't say how many modes could be generated.

Is it a Russian company? I only found Estonian and Swedish contact details on their website. Both countries are very much not Russia.

Yes elva-1 is a Russian company, most likely they're trying to obfuscate that on their English language web presence. It is sort of a shame because they are just as competent as many of the California based microwave engineering firms, and have been around for 25+ years. Their roots are in military radar systems much like some of the Israeli RF companies (Ceragon, etc).

Sorry, I work in a photonics lab so in casual conversation more of the spectrum is "optical" to me. I just meant to ask if you just increase your sampling rate doesn't that mean you have to worry more about things like dispersion happening along the path of the signal?

Optical and microwave are sort of an arbitrary distinction, so you're also right :) There's even the "millimeter wave" designation, or E(xtremely)HF.

Different frequencies scatter differently. This isn't bad: https://transition.fcc.gov/Bureaus/Engineering_Technology/Do...

What sort of latency would you expect with a system like this (I mean the link in this submission, not satellites)?

Point to point wireless links like these usually have low latency, some under a millisecond.

FWIW, I've got several similar point-to-point microwave links in my network, including the following setup:

  h1 - sw1 - ptp1a -/- ptp1z - sw2 - ptp2a -/- ptp2z - sw3 - r1 - h2

  h1    : Sun Fire server
  sw1   : Cisco 3560G switch
  ptp1a : 6 GHz point-to-point microwave transceiver ("A" end)
  ptp1z : 6 GHz point-to-point microwave transceiver ("Z" end)
  sw2   : Cisco 3560G switch
  ptp2a : 6 GHz point-to-point microwave transceiver ("A" end)
  ptp2z : 6 GHz point-to-point microwave transceiver ("Z" end)
  sw3   : Cisco 3560G switch
  r1    : Cisco 3945
  h2    : Sun Fire server

All connections are 1 GbE.

  ptp1a to ptp1z : 12.80 miles (20.60 km)
  ptp2a to ptp2z : 05.01 miles (08.06 km)

From h1 to h2:

  --- x.x.x.x ping statistics ---
  20 packets transmitted, 20 received, 0% packet loss, time 19305ms
  rtt min/avg/max/mdev = 0.955/0.995/1.157/0.049 ms

Each of the two microwave links are advertised as being capable of 761 Mbps full-duplex.

> Advances like this are largely driven by higher sample rate ADCs and DACs becoming more viable in recent years.

I'm looking forward to what these will do for SDR. I salivate over the thought of an SDR using something like the TI ADC12J4000 [1] which has a 4 GSPS sampling rate.

[1] http://www.ti.com/product/ADC12J4000/description

The digital hardware to handle the output of a 4 GSPS ADC is beefy as hell. This part has integrated DDC's if you're happy with downsampling, but if you want full nyquist data coming out of this, you need to handle 6 Gbyte/Sec.

The specs sound very similar to what the BCM8500 can do. Add an analog frontend and an antenna, and that's basically it.

https://www.broadcom.com/products/Microwave-%26-Mobile-Backh...

• Secure communication due inability to intercept the laser-like beam transmission at free air

I would not trust that as a security layer.

It's definitely not something to trust in, but intercepting (at a layer 1 level) a PTP 80 GHz link is actually harder than tapping fiber. You'd have to have Rx equipment either directly in the path or directly behind both ends of the radio link.

As compared to the effort required to cut an aerial or underground singlemode cable and fusion splice in place a passive prism split tap (basically the same thing as inserting a split in a GPON FTTH network). A practiced outside plant fiber crew of 2 persons and a bucket truck could do this with less than 5 minutes of downtime on a router-to-router optical interface, short enough time to clear any NMS alerts and prevent a repair team truck roll. Assuming we're talking about only two strands.

Either way actual security is accomplished through standard based crypto, not obfuscation or preventing people from messing with the layer-1.

On a fiber you don't need downtime at all. If you manage to carefully clean the fiber from isolation and buffer and then bend it just enough to leak some light into yours.

So, assume that everything might be tapped and encrypt your data.

Yes, or if your fast, cheap new metro ethernet 10GbE circuit between two buildings has been provisioned by your ISP as some form of transport (handoff into a WDM system, EoMPLS tunnel), etc, it's not difficult for them to 'mirror' your port. It could be pre-tapped before it was ever turned up for customer service.

Or the splitter could be installed in your riser cable at one end, before you ever started moving packets across it.

Or a myriad of other things.

I recall reading about Soviet spies buying a cabin close to the AT&T long lines microwave path to intercept calls & fax transmissions in the 60s or 70s.

http://www.lamont.me.uk/capenhurst/original.html

http://www.independent.co.uk/news/how-britain-eavesdropped-o...

I wonder if introducing a semi-translucent obstacle, like a cloud of smoke, would be enough to capture the signal while not disrupting the original connection too much?

or a drone with a reflector

I haven't read the datasheet on this particular radio, but a lot of them do optionally support AES encryption (sometimes as an extra, paid feature).

This would get really bad rain fade at 80GHz or even on a humid day the speed would back off a lot. You need to run a lower frequency backup link in parallel.

That's the nature of 80 GHz, design the links for your climate and don't try to go more than 2-3km. You can achieve five nines. And yes, run a 5.x GHz backup path in parallel.

> You can achieve five nines.

As long as nobody pilots a drone in the path.

It'd have to be a big ass drone, I have links that stayed up when a Canada goose was sitting 2 ft directly in front of the 60cm antenna on one end.

Crap, now I need to go buy a couple of these and lobby the nearest data center for some rooftop space.

That is the short way of saying I had no idea you could get antenna this effective for wireless data transmission. I'd seen the 5mbps ones but nothing close to a gigabit much less 10 gigabits. Time to draw a 10km radius circle around my home address :-)

I've seen up to about 400 Mbps on my Webpass [1] connection for a couple years, and I think that's pretty run-of-mill tech at this point [2]

1. https://webpass.net/

2. https://www.ubnt.com/broadband/

Webpass uses a fairly large number of Siklu 80 GHz 1 Gbps radios on rooftops for backbone links in SF.

There are solutions that are lower end than this one, but still gigabit, at relatively affordable prices.

This Ubiquiti setup is $999 per side: https://www.ubnt.com/airfiber/airfiber5/

An AF5 or AF5X PTP link isn't a transparent 1Gb full duplex bridge, realistically you'll achieve 250-400 Mbps max. They're great radios but physics is still physics, 40 or 80 MHz wide dual polarity channel at 256QAM 5/6 code rate is not 1GbE.

Given the context that the parent was saying "I'd seen the 5mbps ones but nothing close to a gigabit", and was talking about a non-commercial setup, it still seems relevant.

Also curious...they show 5470-5950 MHz...from my layperson standpoint, that seems to allow for more than 80MHz wide. What's the actual story? The UBNT forums seems to indicate 2 separate 50Mhz channels, one for TX...one for RX.

That span is the total allowable space in the FCC part 15 rules for 5 GHz gear, where the AF5/5X has been tested and certified. Other radios can be used from 5200 to 5850 MHz (DFS frequencies permitting).

But the size of the channels it can operate in are 5, 10, 15, 20, 30, 40, 60 or 80 MHz wide. Software configurable on the radio.

The NSA Bluffdale facility might just squeak in that line for me. Think they'd hook me up?

If they do, let me know ;) I'm just about there too.

Metrolinq has a 1GE product using unlicensed 60 Ghz which is like $500 a pop.

Though you need to be 500-650 meters max from your source of bandwidth for it to be reliable, 60 is super short range.

What kind of licensing (FCC or otherwise) would one need to operate such a radio bridge in the US? or is it public spectrum?

80 GHz is "light licensed" in the US, the paperwork requirements are not onerous. It's less costly and complicated than a regular part 101 licensed microwave link.

The word "Ethernet" in the title is misleading. That word does not appear in TFA, which appears to discuss a radio transceiver.

[EDIT:] I stand corrected; thanks!

It has a 10GbE SFP+ optical interface and functions as a layer 2 ethernet bridge. From an ethernet port perspective same as 99% of the other PTP microwave and millimeter wave radios on the market. What's new is the 10Gb (vs existing radios with 1Gb SFP).

Read the datasheet linked at the bottom of the page.

They're transparent Ethernet bridges. They have Ethernet interfaces, you push an Ethernet frame in one side and the same Ethernet frame comes out the interface on the other side.

I have line of sight to the building where I work, I would love something like this so I could get gigabit internet at home w/o paying through the nose :)

Check out Mimosa's offerings.. For a few hundred dollars you can get ~750mbps via unlicensed spectrum up to ~5km distance.

https://www.mimosa.co/Products/Backhaul/backhaul-specs/B5-Li...

I know Ubiquti has similar products as well but I'm more familiar with Mimosa's.

Most employers would never let you hook that up - too much risk for zero gain (for them).

Hell mine won't even let me bring my own screen to work. (It lacks some enterprise cert apparently).

it's worth noting that this is just the first one publicly announced and not under NDA.. It's from a russian radar/microwave/millimeter wave manufacturer. All of the other much larger players such as Bridgewave also have 10 Gbps radios coming.

Those BW 80 GHz links are pricy$$$, but like anything the cost will come down in a few years.

Its actually bit curious, they don't mention being Russian anywhere on their site and even their "Contacts" page refers the company being registered in Sweden.

Ooh, I really want two of these. I'm stuck on an island with poor last mile service providers. But, I have line of sight to a number of spots with good quality fiber.

If you've got line of sight then current (cheap) tech should do just fine. Sure not 10x gigabit but you can def get some decent internet...

Do you have any recommendations for current products for line of sight?

for what dollar figure, distance and speeds? anywhere from $300/link (a pair of ubiquiti powerbeam AC 5 GHz) up to $20,000 for 1 to 10GbE licensed.

I don't have any first hand experience, but I know you can do long range (couple kilometers), reliable (for consumer level) decent speed links for dirt cheap (<150USD for each side with weather proofing etc) based on the WUGs in my home country (ZA).

Detour paragraph - background - Internet has historically been pretty sht in South Africa so a enthusiast spider network of point to point links sprung up much like the ham radio movement. Government decided to OK it as long as its not commercial & doesn't deliver internet services.

Anyway...the WUG websites (JAWUG, PTAWUG, CPTWUG - being Johannesburg/Pretoria/CapeTown) have lots of beginner style info on it (will trees and leaves in block it etc)...plus info on cheap kits...plus there are loads of people there that know this stuff in wild detail (as I said - 100% community driven so even the central high-sites are community run).

I know some people who have successfully used Ubiquiti's airFiber for point-to-point links. I think those run on the order of $5 grand for a complete installation (two radios, mounts, etc) of a faster-than-gigabit link.

We run some between offices across the street from each other. We were lucky as we didn't understand the minimum distance was probably more important than the maximum distance for this type of setup! They have run flawlessly at about 800Mbps for the last year.

I love Ubiquiti's products. They're easy to setup and cheap too. From experience, they're also quite reliable.

The company's community forums can be quite helpful too.

What kind of link are you after? (Speed, distance, how far over water, are both sites powered, etc)

Interesting.

There has been a huge effort in the UK for mobile carriers to add fiber to as many cell towers as possible.

Do people think this would undo this trend? I'm sure that 10gbit would be more than enough to carry the backhaul of 3G+4G with plenty of room to spare?

not really, fiber is still greatly preferred (For example: it's impossible to do 40Gb by microwave/millimeter wave, but a CWDM 4-channel 4x10GbE passive mux/demux on two strands of singlemode is trivial and very very cheap), or just a pair of 40Gb QSFP 10km reach optics using a single 1550nm wavelength between two routers or metro-E switches. But wireless can reach small cells, rooftops and towers that might be a very costly underground fiber build at $400-900/meter total construction cost to dig up streets in urban cores. It very much depends on the location.

Plus there's the reliability factor. With fiber you only have to worry about backhoe fade. Wireless has all kinds of fun failure modes.

Though with small monopole and cellular rooftop sites, MW can frequently be built in a ring or redundant traffic path configuration. Fiber to many endpoint sites like a cell tower is frequently a star topology network with one cable in a linear path along a ROW either aerial or underground.

What is the price for this?

Do these products require a clear line of sight to work?

Yes.

According to Nielsen's law of bandwidth, consumer gigabit last mile should be common by now and 10G would be a logical next step development, becoming ubiquitous in 2020 or so.

There's a huge number of houses with idle fiber installed 10+ years ago. Gigabit ethernet was introduced 17 years ago and carries 5 km over fiber. Cable is just waiting for providers to switch on 10 Gbps since years ago. Phone line copper has similar story..

Maybe something like this could jump-start the stalled development of last-mile consumer internet.

You can put the savings into better bandwidth for consumers or lower cost for ISPs. With no competitive pressure for the former in the majority of the country, providers have happily pocketed the 3+ order of magnitude improvements while providing the same old 1990 speed.

Some of those savings are needed in order to be able to provide 100M/1G to the consumer: - 10G in the distribution - 100G in the core

So, even with competitive pressure, consumers can't expect to receive the latest and greatest for peanuts.

> There's a huge number of houses with idle fiber installed 10+ years ago.

Do you have a source for that? Verizon may have passed 18 million homes, but only the homes which have placed an order for Fios were ever connected. So, no or very little idle fiber.

> Gigabit ethernet was introduced 17 years ago and carries 5 km over fiber.

Wrong. You can buy off the shelf SFPs with 200 km reach. Use amps if you want longer reach.

> Cable is just waiting for providers to switch on 10 Gbps since years ago.

You don't just switch on 10 Gbps on cable. First your vendor needs to release DOCSIS 3.1 equipment and you need to test it. Then you need to upgrade your CMTS to DOCSIS 3.1 and swap out any cable modems that don't support 3.1 and dedicate spectrum to DOCSIS 3.1 downstream channels.

And it's not since years ago. DOCSIS 3.1 was released a bit over two years ago. Comcast will start rolling out commercial DOCSIS 3.1 service this year.

> Phone line copper has similar story...

What?! G.fast can theoretically give you 1 Gbps, but only if you already have fiber to your driveway. There aren't even any commercial deployments yet and all vendors don't even have products yet.

And here I am with 12MbS down/800KbS up

Reply on news.ycombinator.com