Why do old alarm signals have trouble with modern VOIP systems?

[e100]

As in alarm dialers designed in the 1980s and early 1990s that use simple in-band tones to send data to a monitoring station at the other end of a telephone line.
Are modern VOIP systems incapable of satisfying the low bandwidth requirements of these old systems?

[jhpadjustable]

Modern VoIP systems are designed to carry voice efficiently and minimize bandwidth used. Lossy compression using sophisticated prediction and psychoacoustic models is the modern norm in telephony. The days where you could expect a switched circuit from one end of the connection to the other are long gone. "If you have the IP network right there, why wouldn't you use it directly?" goes the modern thinking. Sorry!

Free advice follows, worth every penny you paid for it 

• disabling echo cancellers with an answer tone according to ITU-T recommendation G.168 (no purchase required) to fix dropouts that may be introduced by the VoIP switch, which if you're lucky may have the side effect of enabling less lossy compression
• on top of that, if you have university library access, you may also have full access to ITU-T standards. Look at the G and V series. They may explain what the standard protocols and modulation schemes are doing and provide insight into why they are doing it. Fax standards may work better than modem standards.
• choose a more complex modulation that is less compressible by the channel, maybe pack more bits per baud into fewer bauds, like a half-rate V.22bis QAM scheme
• experiment with a pair of COTS modem chips, observe any improvements in the channel fidelity, and do whatever they do to get it
• drop the requirement for a minimum-BOM discrete solution. You may be able to use some of the Signetics-originated PLL and other chips (NE5xx/NE6xx)

This drawing may be interesting as well. Point being, there is no substitute for investigating the prior art.

[SiliconWizard]

Yup. Problem is, the compression algorithms for VoIP are not lossless, and don't have a flat response in their rated bandwidth.
So simple tones are likely to get heavily compressed to the point of getting unrecognizable, or even filtered out radically, if considered by the algorithms as noise, which is rather likely.

[coppice]

As in alarm dialers designed in the 1980s and early 1990s that use simple in-band tones to send data to a monitoring station at the other end of a telephone line.
Are modern VOIP systems incapable of satisfying the low bandwidth requirements of these old systems?

A VoIP channel can have various issues which mess up tone based signalling. Even things like DTMF dialling can be quirky through a VoIP channel. The two main issues are 1) low bit rate codecs are frequently used, which are designed to carry a speaking voice well. They tend to make a mess of anything else, even multiple voices or a voice that's singing.  2) To allow for slightly different timings through the signal chain, samples may be inserted, dropped, or more subtly munged to adjust those timings. This can result in hiccups in tones, which tone decoders really don't like.

The bottom line is things like alarm tone signalling are hit or miss through a VoIP channel. There are some services which receive and decode the signalling tones, and pass on the information contained in those tones in a robust way, that crosses the internet reliably. However, those services don't seem to be widespread.

[Kleinstein]

The compression may filter out things like the phase relation between harmonics.

Another possible problem is a delay added by compression - this the main reason why fax transmissions usually don't over VOIP, as in normal mode they require a maximum response time for error correction. It may work without error correction (compatibility mode to work with very old units from the 1980 and before) or a special international / long distance mode.

[coppice]

The compression may filter out things like the phase relation between harmonics.

The main ways heavy compression screws up tone signalling is by quantising pitch fairly coarsely and only assigning one average pitch for each few milliseconds of the signal.

Another possible problem is a delay added by compression - this the main reason why fax transmissions usually don't over VOIP, as in normal mode they require a maximum response time for error correction. It may work without error correction (compatibility mode to work with very old units from the 1980 and before) or a special international / long distance mode.

Compression only delays audio by 10s of milliseconds. FAX carried as audio (rather than as T.38) mostly fails because either a highly compressed codec is being used, or the timing keeps hiccuping and the modems can't cope with the resulting phase hops. Using T.38 fixes most of that. The ECC parts of T.30 allow for round trip times of several seconds. If your VoIP channel gives you delays worse than that, how could you possibly hold a conversation during a voice call over it?