Comments on: How does VoIP connect to POTS lines?

Question: How does VoIP connect to POTS lines?

Brian James — Tue, 11 Jul 2006 23:10:18 -0800

How exactly do VoIP services place calls to POTS lines?

From my googling, I know that something called an analog terminal adapter is involved to connect the Internet to a POTS line.

But, who owns & maintains these devices, and how many does a VoIP company need to run its service? Did Skype or Vonage go around and set up a bank of these things in every area code in North America?

By: datacenter refugee

datacenter refugee — Wed, 12 Jul 2006 00:03:04 -0800

The concept is the same but the form factor is different. There are standardized 'pieces' to any VoIP network and the piece you're referring to is a gateway. A single-line ATA can be a standalone box like a Cisco 186 or 8x8 DTA310, or it can be a PCI card in a standard PC - in which case it's usually referred to as an FXO or FXS card (definitions).

In larger settings, a SIP or H.323 gatekeeper (call router) will connect to or itself contain a specialized hardware gateway to land t1 or t3 lines (~23 lines/t1, t3=t1*30)). Examples include routers with voice interface cards and digium equipment. ISPs use similar setups, the equipment simply gets larger and is able to handle more analog lines and IP voice sessions. A fully-loaded Cisco MGX series multi-service switch, for example, can translate thousands of calls between VoIP endpoints and the PSTN.

Please excuse the heavy Cisco references, it's the only equipment I've ever set such things up.

By: bystander

bystander — Wed, 12 Jul 2006 00:27:37 -0800

I work for a phone company.
The thing you mean is a gateway. On one side it is an ethernet port, on the other it accepts primary rate ISDN circuits. These are trunk lines with 30 channels (20 in the US?).
These are the same circuits that plug into the back of a corporate PABX, maybe you have seen an MDF in an office?
Anyway, this is functionally equivalent to 30 POTS lines, in that it can handle a total of 30 inbound or outbound concurrent calls. The circuit is terminated at the phone company switch at the exchange, in those big cement buildings with no windows. Then on to the rest of the world's phone system.
VoIP providers typically buy some rack space inside one of the major exchanges in each area code and plug in their gateway. This allows them to carry a VoIP call via IP across the net from say Sydney to Melbourne, then place a local call from the Sydney exchange to the destination (non-IP) number.
Obviously, they have big volumes, so they get cheaper call rates than you or I, and in the US and NZ local calls can be free (well, included in the rental of the primary rate service). This gives them the economic advantage over traditional telcos who time charge as the VoIP costs are a flat charge per call.
It is pretty unlikely that a VoIP operator would install a gateway in *every* area code. In a place like Australia you can cover half the population with just five gateways, and the last 10% of the population would take about 200 more, so once you have most of the population covered you make a toll call from the nearest gateway to the remote destination, and eat the loss on that call.
Obviously there will be many more calls to and from your coverage areas to offset this, and nobody would buy a service where the coverage was not universal.
The other thing to know is that primary rate circuits can map many phone numbers across them, usually a minimum of 100 (why you sometimes see a business switchboard as 555-8100 and the fax as 555-8199 even when you know its only 20 people) but potentially several hundred more.
These are the inbound numbers a VoIP supplier allocates to you, and the gateway sees the destination number of the incoming call and passes it on to your IP address.
If a VoIP provider doesn't have inbound numbers for an area code it is a good indication they don't have a gateway in that code.
If you are interested in this sort of thing, you might want to know how the number of lines needed is calculated and take a look at an explanation of erlangs (self link)

By: baylink

baylink — Wed, 12 Jul 2006 08:19:39 -0800

Bystander pretty much nailed it.

ISDN PRI lines (the trunk of choice for this sort of work) are 23B+D in the US (though some central office switches allow you to group trunks so only the first one has a D channel).

Short version: the VoIP provider has a Big Magic Box that speaks their particular flavor of VoIP and authentication on one side, and IP over Ethernet on the other, and they buy trunks from the local phone company to push the calls out over.

By: kindall

kindall — Wed, 12 Jul 2006 08:20:41 -0800

I imagine it's also possible that a lot of VOIP services are renting space on other companies' gateways, at least for outbound calls and possibly for inbound as well, much as a lot of "long distance companies" are actually MCI or Sprint resellers.

By: kindall

kindall — Wed, 12 Jul 2006 08:23:26 -0800

The Ethernet that goes into a gateway, by the way, is probably also connected to an ISDN PRI line through a CSU/DSU (or whatever they call them for ISDN, that's what you use for T1s). That goes to their Internet provider, probably a nearby backbone access point.

By: baylink

baylink — Wed, 12 Jul 2006 08:49:52 -0800

Well, kindall, actually, not much data moves over ISDN these days; that's a channelization of a T-1 mostly reserved for circuit-switched voice. Most data moves over raw T-1, T-3, OC-3, and up into STS- fiber data channels. A VoIP gateway would likely be being supplied 100BaseT, being in a data center, and what would be on the other side of the router it was plugged into depends pretty much on the datacenter in question.

Whether the back side of that gateway would be terminated *directly* into a T-span depends, I suppose, on where thy box was physically located: it would be a lot more likely if it was in a CO (with the data backhauled to a datacenter somewhere else) than if it was actually *in* that datacenter (as I assume above; with the *voice* circuits backhauled in)...