I will try as best as I can to give an explanation of what happens when you send an SMS from your phone.
Disclaimer: Telco stuff is hard.
Also disclaimer: this blog post will also contain alot of ackronyms, after all, it is telco.
Aaand down the rabbit hole we go…
Where to even start
In the SS7 (telco/telecom/telecommunications) network there are many different nodes (servers), with different kinds of tasks.
The group of protocols that is used to send signals over IP between these nodes is called SIGTRAN (derived from “signaling transport”). Older networks that have not switched to IP do not use SIGTRAN.
SIGTRAN protocols are the lower layer protocols used for signaling, they range from SCTP (Stream Control Transmission Protocol) to M2PA (Message Transfer Part 2 User Peer-to-Peer Adaptation Layer) and M3UA (Message Transfer Part 3 User Adaptation Layer).
SCTP is like a mix between UDP and TCP. It is supposed to be quicker than TCP, but more reliable than UDP.
Both M2PA and M3UA support SCTP management, and the reporting of status changes of those, as well as providing transfer of MTP3 (Message Transfer Part 3) messages.
On top of SIGTRAN are the SS7 protocols.
What I’m going to talk about are the protocols on the very top of SS7, specifically the MAP (Mobile Application Part) as well as the TCAP (Transaction Capabilities Application Part). There are other protocols inbetween, for instance SCCP (Signalling Connection Control Part) which handles some handshaking, routing, and resilience.
The MAP layer is used when talking to some of the telco nodes such as HLR (Home location registry), VLR (Visitor location registry), MSC (Mobile switching centre), SGSN (Serving GPRS [ackronym in ackronyms; go telco!] support node) and the SMSC (Short message service centre).
MAP versions and TCAP dialogues
There are some iterations of MAP (v1, v2, v3, and v4) and messages almost always come in pairs, an acknowledgement (
ReturnError) for each sent message (
To determine which version to use between two nodes, the sending node tries to start the transaction (called a dialogue) by sending a TCAP
Begin message with the MAP message and it’s highest compatible version. If the receiving node cannot talk that version, it sends a TCAP
Abort message with it’s highest compatible version. In v1 there might not even be a reason, just an empty
Abort; the sending node might then try to send the MAP message as v1 anyway.
In my head it goes like this:
Node 1: "Hi, I want to talk version 3 to you about this" Node 2: "No I don't understand you, but we can talk version 2 about it instead" Node 1: "Ok, then I want to talk version 2 to you about this instead" Node 2: "Aah, now I see..."
Node 1: "Hi, I want to talk version 3 to you about this" Node 2: "No" Node 1: "Ok, then I want to talk to you about this in version 1 instead" Node 2: "Maybe I will talk to you, maybe I will not"
For TCAP dialogues there are (mainly) four message types.
Abort. Each of the types have an ID (or two, as I said, telco is complicated), a component and a dialogue part. The component contains the MAP message. The dialogue part contains the version and application to use (that is MAP Application; i.e. which type of message it contains), but it is only used in the first message from both nodes for the version negotiation.
I think this covers most of it, let’s get back to the fun part.
How does SMS work?
SMS was initally implemented because of the wish to send text messages to pagers using the phoneline when it was not in use for phonecalls. It was decided at a meeting in Oslo to be released to the public when some French and German company understood it’s value. (Don’t quote me on any of this).
When you send an SMS, the SMS is transfered to the MSC or the SGSN in your current (serving) network. The MSC or SGSN then sends an packet called
MO-Forward-SM towards the SMSC in your current network. It stands for “Mobile Originating Forward Short Message” meaning it started from your (mobile-)phone.
The SMSC then asks the recipients HLR about the routing details for the SMS. It does so by sending another MAP message of type
sendRoutingInfoForSM requesting the location of the recipients MSC or SGSN, or both.
The SMSC then sends another packet, this time a
MT-Forward-SM, towards the MSC in the recipients network. In this case MT stands for Mobile Terminated, meaning it goes towards the recipients phone.
Dia have amazing icons:
The similarities in MO and MT requests are that they both contain a origin and destination address as well as the user data (your actual text message), and a possibly a correlation id which is basically a mapping between your SIM-card id and a temporary id and was originally used for making sure that the sending network paid for SMSs towards the receiving network.
For MO the origin address is your MSISDN (read telephone number), and the destination is the GT address (Global title; a way to route stuff) of the SMSC. For MT messages the origin address is the GT of the SMSC and the destination address is either the recipients IMSI (read SIM-card) or the recipients correlation id. It could also be a LMSI which is a 4-byte network location identifier if the recipient is also within the same network as the sender.
Ever wondered why there is a limit to the size of the text message you are sending?
If you (god forbid) you would break the protocol and send a text message greater than 140 bytes, which translates to 160, 152, or 70 characters depending on locale , then your phone would break up the message into multiple text messages. This arbitrary size of 140 bytes is not really arbitrary at all. It was chosen because it would precisely fit into a single MTP3 SIF (Signalling Information Field) when routing label, SCCP, TCAP and MAP layers were taken into account.
 There is something called GSM7 bit-packing. Instead of using 1 byte (8 bits) per character, GSM7 uses 7 bits. This means that instead of 140 characters, you can get up to 160 characters per SMS. The drawback is that you will have a smaller subset of characters to use, only the most common is supported. If you include any non-GSM7 characters in your SMS then the SMS will automatically be converted to use either USC-2 or GSM7 with a different charset instead. USC-2 uses 2 bytes, or 16 bits, instead of GSM7s 7 bits. That leaves you with 70 characters per SMS. USC-2 is similar for the basic multilingual plane (BMP) to UTF-16. In fact UTF-16 is an extension of UCS-2. The main difference is that USC-2 is fixed width and does not allow for the extended characters in the private use area of BMP. UTF-16 is variable width of one or two 16-bits code points, and does allow the extended characters. For extended characters to work (for instance “praying/folded hands” 🙏), phones might try to fake UTF-16 by using two USC-2 characters. New phones can handle this fine, but older phones might receive two question marks as they cannot decode it properly.
If GSM7 have a modified charset (i.e. not the default BMP) then there will be a header in front that specifies that. That header will take up 7 bytes after packing (in other words 8 characters), making the maximum length of the SMS 152 characters.
However when MAP v2 started to use TCAP dialogues there was more information to put into the packet and 140 bytes might not be left for the SMS. The SMSC would then need to break up the message into chunks, and start the transaction an empty TCAP
Begin message and set a flag in the MT request called
moreMessagesToSend. It would then send the actual text inside
Continue messages. In the end the
End (hehe) message is transmitted as a response and the transaction is finished.
The response back to a MO request is, as previous stated, an acknowledgement if the SMS have been successfully submitted to the SMSC or not (again either
returnError). For MT requests the acknowledgement is if the SMS is successfully delivered or not.
If the MT request is not successful, the SMSC could ask the HLR (the Home Location Registry is basically a database containing user subscriptions and knowledge of which nodes the mobile talked to latest) to be notified when the user comes back online. A bunch of other MAP messages are then involved, such as
At least this is main idea I think…
Differences in MAP versions for SMS
There are three MAP versions defined for SMS. The latest version (v4) is not used in the context of SMS.
In version 1, the dialogue portion was not invented and all chunks are sent in new TCAP dialogues. The size of the user data could then be 140 bytes.
In version 1 and version 2 there is no difference between MT and MO. Everything is sent as another type of message
Forward-SM, which does not include any privacy correlation ids, and there are no fancy responses with delivery status. There is still an acknowledgement, but is in a form of an empty message.
Only way to see difference between an MO and a MT in version 1 is to look at the addresses and see if they are either coming from an SMSC or going to an SMSC.
moreMessagesToSend flag was implemented in version 2, so it exist only for version 2 and version 3.
Ok, to recap, what do we have now
- Dialogue handshake in the first request/response messages sent.
- Involved parties: Mobile phones, MSC and SMSC
Wait we are missing something. I’ve only covered 2G,3G..
What about 4G/LTE and beyond (5G)?
LTE networks does not use any of the M3UA, SCCP, TCAP, MAP protocols. In LTE networks the main message type is Diameter which doesn’t contain fragmentation and can contain larger messages. Everything is sent in one request and every request is answered with a response. Diameter could use either TCP or SCTP as transport layer.
To make SMSes work on LTE networks a new interface SGs was invented which translates SS7 messages to Diameter messages. This interface is in most cases used by the MSC to translate the messages to Diameter and forward it to the MME (Mobility Management Entity, similar to SGSN but in the LTE network). The MME then forwards it to the UE (user equipment, same as mobile subscriber or MS in GSM/GPRS networks).
There is also the SM-over-IP that does not use Diameter. Instead it uses the SIP-protocol (Session Initiation Protocol) to transfer messages over IP and TCP or UDP to the IMS (IP Multimedia Subsystem). SIP is also used to enable VoLTE (Voice over LTE).
For 5G the SMSC is called SMSF; The Centre becomes a Function. The signalling will be based on HTTP2/JSON ontop of TCP. The SMSF will still need to support both MAP and Diameter.
Hopefully you did not get a (too severe) headache by reading this post.
I’ve spared you with a lot of details on the lower level of protocols. There are loads of implementation details that must match the specifications, otherwise you will get all kinds of aborts and possibly even dropped traffic. For instance we learned that we accidentally sent dialogue portions in more than the first response back, which seemed to work at first glance; at closer inspection we found out that some messages were dropped because the length of the packet became larger in size than an allowed value. We could still send them, but the other side was not able to receive them.
Remember: Telco is old and complex. However, it should still function with different setups and on different hardware, vendors and with environment.
Fun-fact: Sometimes a boolean value is not just encoded as a 1 or 0. To save bandwith telco decided that you could also just define it as a
NULL OPTIONAL meaning that if it is defined (but lacks a value), then it is considered true. if it is not defined then it is considered false. This is the case for the
Hope you enjoy the reading as much as I enjoy digging into these protocols!
Special thanks to Stein Eldar and Tobias for giving me feedback and answering all my stupid questions on this subject, and Atanas for making me realize there are yet other protocols to carry SMS. Also Bung for this amazing addition:
Addition from Bung
“I’ve spared you with a lot of details on the lower level of protocols” needs a lot of emphasis.
Some funny extra complexities that just came into my mind while reading:
The actual SMS text goes into a field called “user data”. There is a field called “user data length”. When the message is GSM7 encoded, the “user data length” is the number of characters in the text message, otherwise it’s the number of bytes in the user data.
Normally the user data only contains the (encoded) text of the message, but there is a field called “user data header” which indicates that there is a length prefixed TLV header in the “user data”. If the message is GSM7 encoded, then the “user data lenght” field needs to be filled as if the “user data header” was really GSM7 encoded, which it isn’t.
GSM7 is really a variably septet encoding, one character can consist of either 7 or 14 bits similar to how a UTF-8 code point can be 8, 16, 24, or 32 bits. Unlike UTF-8 however, there are not multiple byte ranges corresponding to the different locales (called code pages in Unicode) but a single 7 bit shift character that says that following 7 bits should be interpreted as a character from a translation table which is communicated out of band.
So all that is only the complexities of a single field (the user data) for a single encoding (GSM7).
Then the real kicker: The protocol for SMS is really called SM-TP (Short Message Transfer Protocol). SM-TP is the same for 2G/3G (on top of MAP), 4G (on top of SIP), 5G (on top of HTTP). So the very same stupid “length prefixed TLV encoded headers concatenated with encoded text with length either in characters or in bytes depending on encoding and actual meaning of the encoding communicated out of band but only sometimes” field exists no matter if your talking over old legacy MAP or the modern HTTP/XML based 5G.