Comment by sathackr | Hacker News Reader

Comment by sathackr | Hacker News Reader

Parent Full thread

Both cable and FTTH(single-strand fiber to the home, such as Fios) are full-duplex via frequency division.

Cable operates in RF bands, and frequencies are denoted in MHZ and sometimes GHZ(included in the linked answer)

Fiber operates in optical bands and frequencies are denoted in nanometers(wavelength) -- downstream data, downstream video, and upstream data are all operated on different wavelengths on the same piece of glass, and they are received at both the transmitting end and the receiving end.

They are typically separated and filtered by optical prisms to ensure the correct wavelength hits the correct receiver(most receivers are wide-band and will receive a wide range of optical wavelengths)

The medium is typically shared at a neighborhood level, in that a single optical strand may be split 16 or more times to serve multiple houses. Bandwidth is shared on that strand by the houses served by it, typically via TDM(time division multiplexing)

End users don't typically notice this because the line rates are higher than the maximum bandwidth package sold. IE downstream rate 2.5Gb/s, upstream 1.5Gb/s, max package available is 1Gb/s.

Edit: removed incorrect statement regarding gepon/csma

Full duplex DOCSIS 3.1 has theoretical maximums of 10Gbps down and up. It relies on precise timing and echo cancellation to allow the cable modem and CMTS to transmit simultaneously on the same frequency bands. Kind of like how gigabit Ethernet allows both ends to transmit simultaneously on the same four pairs.

The question is how practical that mode of operation is on real network, or in other words whether it actually makes sense to run DOCSIS on network that can reliably support that mode (think various xDSL technologies that can push >1Gbps over <100m of what essentially is Cat5 cable)

DOCSIS is somewhat inherently asymmetric due to nature of the CaTV network. Even if the upstream and downstream channels would have same bandwidth, the upstream direction tends to have significantly worse SNR. Historically this was caused by splitters along the way, but also the cable modem is relatively cheap device installed on customer premises and thus has inherently lower transmit power and worse SNR than CMTS. This is the main reason why CMTS tend(ed) to have more upstream receivers than downstream transmitters.

This is the generally correct answer for most time divided (tdm) services. I worked on dsl back in the 90s and we experimented with pretty much all the flavors (SDSL, ADSL,etc) and while some of the plant that was built to recent modern standards could handle a maxed-out capacity, a lot of the plant (50-80+ years old) simply wasn’t suitable for it. Telcos had a choice - give everyone the lowest common denominator, deploy multiple solutions to different neighborhoods or overbuild the plant with new plant. Verizon largely chose to do the later and you got fios as a result. Other telcos did overbuilds as demand materialized. Regardless of the technical later you still had the commercial offering to consider. One way to preserve “business” line economics was to create an asymmetric residential offering. For the most part it also reflected the underlying technical capabilities. WDM fiber I don’t see happening as a residential offering anytime soon. With current technologies 200G per wavelength is capable but the card costs alone for that level of service implies many thousands in monthly recurring to recover them. 10g cards obviously are lower but gpon is much better suited as a residential architecture both economically and technically.

How come fiber can't use the same frequency for uplink and downlink? Coherent and focused laser light only travels in one direction, so I'd expect the transmission not to interfere with the receiver.

Duplex fiber where a separate strand for TX/RX can use the same frequency.

Simplex fiber, that is point to point, may in fact be able to use the same frequency, but it's not done in practice. Possibly due to the difficulty in directing the outgoing and incoming signals to different pieces of hardware(receiver and transmitter)

GPON networks(typically used by Fios, etc...) have multiple passive splitters on the strand. For example, a common PON splitter would have 1 strand in from the OLT(optical line terminal, essentially the head-end), 32 strands out. The light received from the OLT is split x32 and sent to each downstream port. The devices on the downstream ports(ONU/ONT, Optical Network Terminal) receive this signal and also transmit it on the same strand. That transmitted signal must be coupled back into the single upstream port, from every attached ONT. I imagine there is significant difficulty getting the light to traverse the splitter and exit only via the upstream port. So a different wavelength is used so that it doesn't matter if the light is received by the other units, and also it can be more easily filtered (by wavelength instead of direction).

My knowledge of optics is not sufficient to fully explain why it's easier to filter by wavelength instead of direction.

> My knowledge of optics is not sufficient to fully explain why it's easier to filter by wavelength instead of direction.

Reciprocity. For most optical devices, if x% of light coming in port A comes out port B, then x% of light coming in port B comes out port A. Non-reciprocal materials exist, but they’re unusual and may be rather inefficient and complicated to work with. Wavelength splitters are reciprocal and are relatively easy to construct. In fact they’re cheap enough that people use them for art. (The usual splitter is a piece of dichroic glass.)

The Wikipedia article is so-so:

https://en.m.wikipedia.org/wiki/Reciprocity_(electromagnetis...

The reason why it is easier to filter by wavelength is that almost any optical path reflects significant amount of light back, primarily from various discontinuities along the fiber (connectors, splices, and also the actual receiver on the other end) but also from imperfections in the fiber itself. On copper interconnects there are similar issues, but for reasonably controlled cabling (eg. 1000-base-T) they can be minimized enough, that what remains is either insignificant or can be compensated for by DSP magic.

Reply on news.ycombinator.com