Error Detection and Recovery For Reliable In-Band Signaling Over Error-Prone Links

3.1. MF Frequencies

This table shows the frequencies used for each MF tone.¶

3.2. DTMF Frequencies

This table shows the frequencies used for each DTMF tone.¶

3.3. Frequency Similarities

Several MF/DTMF tone pairs share similar frequencies.¶

Improperly tuned DTMF detectors are thus capable of falsely detecting MF digits as the frequency-similar DTMF digits.¶

Extensive testing has shown that the MF digits 7 and 8 are most likely to be falsely conflated as DTMF by out-of-band DTMF detectors. Typically, this results in the actual MF digit being squelched out of the audio and a spurious DTMF tone being added to the audio.¶

4.1. Signaled Values

It is assumed that for any signaling transmission, two numbers are sent, one of fixed length (or of a set of fixed lengths known in advance) and one of variable length that is unknown by the other side in advance.¶

On incoming calls, terminating to a subscriber system, it is assumed that the DNIS is of fixed length and the ANI is of variable length. On outgoing calls, the ANI is of fixed length while the called number may be of variable length.¶

4.2. MF Signaling with Error Detection

The class of error that occurs with MF signaling are loss of digits, typically the digits 7 and 8.¶

Normally, only KP, the numeric digits, and ST are needed to transmit numeric information using MF signaling. This leaves the digits STP, ST2P, and ST3P, which have been used over the years in various signaling schemes to convey specific types of informations. The use of these three additional start digits can thus be used to enrich the payload at no additional cost.¶

Because the typical failure mode of MF signaling is loss of digits, error detection can be performed by ues of a checksum over the number of digits in the variable-length number (errors can be detected trivially in the fixed-length number simply by checking its length). Because four start digits are available, the number of digits in the number modulo 4 can be mapped to one of the start digits, e.g. as follows:¶

This error detection scheme can successfully detect errors with no additional signaling digits as long as the number of errors is not a multiple of 4. Unfortunately, 4 dropped digits is possible if there are four 7s or four 8s in the number, which is not particularly unlikely. If the lack of a valid number (i.e. invalid dialed number or invalid ANI) can be detected in the presence of 4 dropped digits (i.e. the numbering length cannot differ by up to 4, for any valid numbers), error detection may still be possible by another method.¶

An extension of this scheme involves using KP2 (same as ST2P) instead of KP in certain calls, allowing both the key pulse and the start digit to be used for checksum information. Instead of 4 possible checksums, this would allow for 8 possible checksums. However, a number with 8 '7' and '8' digits combined could still result in this checksum passing.¶

A more robust checksum scheme involves the addition of an additional checksum digit, prefixed to the numeric digits, e.g. KP + C + XXX + ST, where C represents the dedicated checksum digit. Since C can have ten possible values (as opposed to only two possible key pulse digits), there are now 40 possible checksums possible. This should be more than adequate to cover any number of dropped digits in any reasonably plausible signaling scenario, particularly for ten-digit U.S. numbers.¶

4.3. SF Signaling with Error Detection

Unlike with MF signaling, because SF signaling involves individual pulses, which themselves are immune to the problems posed to MF signaling by out of band DTMF, SF signaling is more robust against errors. However, occasionally a pulse from a digit may be dropped, leading, for example, a sent 0 to be received as a 9 instead. Therefore, the class of error with SF signaling is not outright loss of digits but digit mutation - unlike MF signaling where the digits received are assumed to be correct, but digits may be missing, received SF digits themselves may be suspect¶

The checksum for SF signaling involves the use of the digit sum over all the digits in the number sent, both the fixed-length number and the variable-length number. (The digit sum of the number 9-8-7, for example, is 9+8+7 = 24.) The checksum digit is then simply the digit sum modulo M, where M is a suitably large modulus, but no greater than 10. M can be interpreted as the number of SF pulses that would need to be dropped from the entire number in order for a false negative to occur in error detection. 10 is a natural choice for M, but 5 MAY also be used since it evenly divides 10. The reason for preferring 5 is that it is still reasonably resistant to false negatives and ensures that the checksum digit can take at most a little over half a second of transmission time, as opposed to a little over a second.¶

As an example with M=5, if the digits 9+8 are sent, the checksum is 17, and 17 % 5 = 2. This digit is sent first. If the digits were received as 8+8, the receiver would calculate this value to be 1, and the mismatch indicates at least one digit has been corrupted.¶

An optimization that SHOULD be made is to use a single checksum calculated from the digit sum over both the fixed and variable-length numbers that are sent. For efficiency, only one digit is used for checksum purposes instead of two, which is more important with SF signaling due to how slow it is compared to MF signaling. As long as no more than five pulses are dropped from both numbers combined, this method will successfully detect SF signaling errors. In practice, it has been found that most SF signaling transmissions complete without error, or with only one error if errors occur.¶

4.4. Communicating Error Detection Results

Errors can occur in both MF signaling and SF signaling; however, the nature of the errors differs in a manner that usually allows one to complete successfully. A typical failure mode is for MF signaling to fail because the transmitted number includes 7s or 8s, while SF signaling failures tend to involve digits being off by one. If MF fails, both sides can agree to fall back to SF signaling to transmit the corrupted information. This section describes how the receiver can communicate to the sender when it has detected errors to request that a number (or multiple numbers) be retransmitted.¶

MF signaling SHOULD be attempted first, because if the receiver does not detect any errors, the transmission can be completed relatively quickly, in a couple seconds or so. Using the error detection scheme described for MF signaling, the receiver calculates whether the variable-length number was corrupted. Using a simple length check, the receiver calculates whether the fixed-length number was corrupted. The receiver SHOULD then indicate to the sender whether the fixed-length number, the variable-length number, both, or neither was corrupted. The receiver indicates this to the sender by the use of an MF digit, referred to here as the "error digit". If error recovery is being used, the sender MUST wait up to 1 second for the error digit after completing MF sending.¶

Importantly, if an MF digit is sent, the digit sent is not one that is likely to be dropped or corrupted due to out-of-band DTMF squelching.¶

In the case that no errors occur, no error digit is sent and the 1 second timeout expires at the sender.¶

If the sender receivers an error digit, it MUST retransmit the corrupted number(s) using SF signaling. It will first decode the digit to determine if one or two failures occured, and then retransmit this information using SF, according to the Error-Detecting SF signaling scheme described above. The receiver will then detect if errors have occured. If the information has been corrupted a second time, the receiver routes the call to an intercept messaging informing the caller that his call did not go through. The caller SHOULD then retry his call as it is possible the call will complete successfully in a subsequent attempt.¶

7.1. Normative References

[RFC2119]

Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, <https://www.rfc-editor.org/info/rfc2119>.

[RFC8174]

Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, <https://www.rfc-editor.org/info/rfc8174>.

7.2. Informative References